Implementations of TCP and IPC transports separated

Signed-off-by: Martin Sustrik <sustrik@250bpm.com>

11 files changed, 1008 insertions, 45 deletions

diff --git a/src/Makefile.am b/src/Makefile.am
index 1da5be9..9f5ea65 100644
--- a/src/Makefile.am
+++ b/src/Makefile.am
@@ -24,6 +24,8 @@ libzmq_la_SOURCES = \
     io_object.hpp \
     io_thread.hpp \
     ip.hpp \
+    ipc_connecter.hpp \
+    ipc_listener.hpp \
     i_engine.hpp \
     i_poll_events.hpp \
     kqueue.hpp \
@@ -86,6 +88,8 @@ libzmq_la_SOURCES = \
     io_object.cpp \
     io_thread.cpp \
     ip.cpp \
+    ipc_connecter.cpp \
+    ipc_listener.cpp \
     kqueue.cpp \
     lb.cpp \
     mailbox.cpp \
diff --git a/src/ipc_connecter.cpp b/src/ipc_connecter.cpp
new file mode 100644
index 0000000..942cd49
--- /dev/null
+++ b/src/ipc_connecter.cpp
@@ -0,0 +1,381 @@
+/*
+    Copyright (c) 2007-2011 iMatix Corporation
+    Copyright (c) 2007-2011 Other contributors as noted in the AUTHORS file
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the GNU Lesser General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU Lesser General Public License for more details.
+
+    You should have received a copy of the GNU Lesser General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include <new>
+#include <string>
+
+#include "ipc_connecter.hpp"
+#include "tcp_engine.hpp"
+#include "io_thread.hpp"
+#include "platform.hpp"
+#include "random.hpp"
+#include "ip.hpp"
+#include "err.hpp"
+
+#if defined ZMQ_HAVE_WINDOWS
+#include "windows.hpp"
+#else
+#include <unistd.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <arpa/inet.h>
+#include <netinet/tcp.h>
+#include <netinet/in.h>
+#include <netdb.h>
+#include <fcntl.h>
+#ifdef ZMQ_HAVE_OPENVMS
+#include <ioctl.h>
+#endif
+#endif
+
+zmq::ipc_connecter_t::ipc_connecter_t (class io_thread_t *io_thread_,
+      class session_t *session_, const options_t &options_,
+      const char *address_, bool wait_) :
+    own_t (io_thread_, options_),
+    io_object_t (io_thread_),
+    s (retired_fd),
+    handle_valid (false),
+    wait (wait_),
+    session (session_),
+    current_reconnect_ivl(options.reconnect_ivl)
+{
+    memset (&addr, 0, sizeof (addr));
+    addr_len = 0;
+
+    //  TODO: set_addess should be called separately, so that the error
+    //  can be propagated.
+    int rc = set_address (address_);
+    zmq_assert (rc == 0);
+}
+
+zmq::ipc_connecter_t::~ipc_connecter_t ()
+{
+    if (wait)
+        cancel_timer (reconnect_timer_id);
+    if (handle_valid)
+        rm_fd (handle);
+
+    if (s != retired_fd)
+        close ();
+}
+
+void zmq::ipc_connecter_t::process_plug ()
+{
+    if (wait)
+        add_reconnect_timer();
+    else
+        start_connecting ();
+}
+
+void zmq::ipc_connecter_t::in_event ()
+{
+    //  We are not polling for incomming data, so we are actually called
+    //  because of error here. However, we can get error on out event as well
+    //  on some platforms, so we'll simply handle both events in the same way.
+    out_event ();
+}
+
+void zmq::ipc_connecter_t::out_event ()
+{
+    fd_t fd = connect ();
+    rm_fd (handle);
+    handle_valid = false;
+
+    //  Handle the error condition by attempt to reconnect.
+    if (fd == retired_fd) {
+        close ();
+        wait = true;
+        add_reconnect_timer();
+        return;
+    }
+
+    //  Create the engine object for this connection.
+    tcp_engine_t *engine = new (std::nothrow) tcp_engine_t (fd, options);
+    alloc_assert (engine);
+
+    //  Attach the engine to the corresponding session object.
+    send_attach (session, engine);
+
+    //  Shut the connecter down.
+    terminate ();
+}
+
+void zmq::ipc_connecter_t::timer_event (int id_)
+{
+    zmq_assert (id_ == reconnect_timer_id);
+    wait = false;
+    start_connecting ();
+}
+
+void zmq::ipc_connecter_t::start_connecting ()
+{
+    //  Open the connecting socket.
+    int rc = open ();
+
+    //  Connect may succeed in synchronous manner.
+    if (rc == 0) {
+        handle = add_fd (s);
+        handle_valid = true;
+        out_event ();
+        return;
+    }
+
+    //  Connection establishment may be dealyed. Poll for its completion.
+    else if (rc == -1 && errno == EAGAIN) {
+        handle = add_fd (s);
+        handle_valid = true;
+        set_pollout (handle);
+        return;
+    }
+
+    //  Handle any other error condition by eventual reconnect.
+    wait = true;
+    add_reconnect_timer();
+}
+
+void zmq::ipc_connecter_t::add_reconnect_timer()
+{
+    add_timer (get_new_reconnect_ivl(), reconnect_timer_id);
+}
+
+int zmq::ipc_connecter_t::get_new_reconnect_ivl ()
+{
+    //  The new interval is the current interval + random value.
+    int this_interval = current_reconnect_ivl +
+        (generate_random () % options.reconnect_ivl);
+
+    //  Only change the current reconnect interval  if the maximum reconnect
+    //  interval was set and if it's larger than the reconnect interval.
+    if (options.reconnect_ivl_max > 0 && 
+        options.reconnect_ivl_max > options.reconnect_ivl) {
+
+        //  Calculate the next interval
+        current_reconnect_ivl = current_reconnect_ivl * 2;
+        if(current_reconnect_ivl >= options.reconnect_ivl_max) {
+            current_reconnect_ivl = options.reconnect_ivl_max;
+        }   
+    }
+    return this_interval;
+}
+
+#ifdef ZMQ_HAVE_WINDOWS
+
+int zmq::ipc_connecter_t::set_address (const char *protocol_, const char *addr_)
+{
+    errno = EPROTONOSUPPORT;
+    return -1;    
+}
+
+int zmq::ipc_connecter_t::open ()
+{
+    zmq_assert (s == retired_fd);
+
+    //  Create the socket.
+    s = socket (addr.ss_family, SOCK_STREAM, IPPROTO_TCP);
+    if (s == INVALID_SOCKET) {
+        wsa_error_to_errno ();
+        return -1;
+    }
+
+    // Set to non-blocking mode.
+    unsigned long argp = 1;
+    int rc = ioctlsocket (s, FIONBIO, &argp);
+    wsa_assert (rc != SOCKET_ERROR);
+
+    //  Connect to the remote peer.
+    rc = ::connect (s, (sockaddr*) &addr, addr_len);
+
+    //  Connect was successfull immediately.
+    if (rc == 0)
+        return 0;
+
+    //  Asynchronous connect was launched.
+    if (rc == SOCKET_ERROR && (WSAGetLastError () == WSAEINPROGRESS ||
+          WSAGetLastError () == WSAEWOULDBLOCK)) {
+        errno = EAGAIN;
+        return -1;
+    }
+    
+    wsa_error_to_errno ();
+    return -1;
+}
+
+int zmq::ipc_connecter_t::close ()
+{
+    zmq_assert (s != retired_fd);
+    int rc = closesocket (s);
+    wsa_assert (rc != SOCKET_ERROR);
+    s = retired_fd;
+    return 0;
+}
+
+zmq::fd_t zmq::ipc_connecter_t::connect ()
+{
+    //  Nonblocking connect have finished. Check whether an error occured.
+    int err = 0;
+    socklen_t len = sizeof err;
+    int rc = getsockopt (s, SOL_SOCKET, SO_ERROR, (char*) &err, &len);
+    zmq_assert (rc == 0);
+    if (err != 0) {
+
+        //  Assert that the error was caused by the networking problems
+        //  rather than 0MQ bug.
+        if (err == WSAECONNREFUSED || err == WSAETIMEDOUT ||
+              err == WSAECONNABORTED || err == WSAEHOSTUNREACH ||
+              err == WSAENETUNREACH || err == WSAENETDOWN)
+            return retired_fd;
+
+        wsa_assert_no (err);
+    }
+
+    //  Return the newly connected socket.
+    fd_t result = s;
+    s = retired_fd;
+    return result;
+}
+
+#else
+
+int zmq::ipc_connecter_t::set_address (const char *addr_)
+{
+    return resolve_local_path (&addr, &addr_len, addr_);
+}
+
+int zmq::ipc_connecter_t::open ()
+{
+    zmq_assert (s == retired_fd);
+    struct sockaddr *sa = (struct sockaddr*) &addr;
+
+    if (AF_UNIX != sa->sa_family) {
+
+        //  Create the socket.
+        s = socket (sa->sa_family, SOCK_STREAM, IPPROTO_TCP);
+        if (s == -1)
+            return -1;
+
+        // Set to non-blocking mode.
+#ifdef ZMQ_HAVE_OPENVMS
+    	int flags = 1;
+    	int rc = ioctl (s, FIONBIO, &flags);
+        errno_assert (rc != -1);
+#else
+    	int flags = fcntl (s, F_GETFL, 0);
+    	if (flags == -1)
+            flags = 0;
+    	int rc = fcntl (s, F_SETFL, flags | O_NONBLOCK);
+        errno_assert (rc != -1);
+#endif
+
+        //  Connect to the remote peer.
+        rc = ::connect (s, (struct sockaddr*) &addr, addr_len);
+
+        //  Connect was successfull immediately.
+        if (rc == 0)
+            return 0;
+
+        //  Asynchronous connect was launched.
+        if (rc == -1 && errno == EINPROGRESS) {
+            errno = EAGAIN;
+            return -1;
+        }
+
+        //  Error occured.
+        int err = errno;
+        close ();
+        errno = err;
+        return -1;
+    }
+
+#ifndef ZMQ_HAVE_OPENVMS
+    else {
+
+        //  Create the socket.
+        zmq_assert (AF_UNIX == sa->sa_family);
+        s = socket (AF_UNIX, SOCK_STREAM, 0);
+        if (s == -1)
+            return -1;
+
+        //  Set the non-blocking flag.
+        int flag = fcntl (s, F_GETFL, 0);
+        if (flag == -1) 
+            flag = 0;
+        int rc = fcntl (s, F_SETFL, flag | O_NONBLOCK);
+        errno_assert (rc != -1);
+
+        //  Connect to the remote peer.
+        rc = ::connect (s, (struct sockaddr*) &addr, sizeof (sockaddr_un));
+
+        //  Connect was successfull immediately.
+        if (rc == 0)
+            return 0;
+
+        //  Error occured.
+        int err = errno;
+        close ();
+        errno = err;
+        return -1;
+    }
+#endif
+
+    zmq_assert (false);
+    return -1;
+}
+
+int zmq::ipc_connecter_t::close ()
+{
+    zmq_assert (s != retired_fd);
+    int rc = ::close (s);
+    if (rc != 0)
+        return -1;
+    s = retired_fd;
+    return 0;
+}
+
+zmq::fd_t zmq::ipc_connecter_t::connect ()
+{
+    //  Following code should handle both Berkeley-derived socket
+    //  implementations and Solaris.
+    int err = 0;
+#if defined ZMQ_HAVE_HPUX
+    int len = sizeof (err);
+#else
+    socklen_t len = sizeof (err);
+#endif
+    int rc = getsockopt (s, SOL_SOCKET, SO_ERROR, (char*) &err, &len);
+    if (rc == -1)
+        err = errno;
+    if (err != 0) {
+
+        //  Assert if the error was caused by 0MQ bug.
+        //  Networking problems are OK. No need to assert.
+        errno = err;
+        errno_assert (errno == ECONNREFUSED || errno == ECONNRESET ||
+            errno == ETIMEDOUT || errno == EHOSTUNREACH ||
+            errno == ENETUNREACH || errno == ENETDOWN);
+
+        return retired_fd;
+    }
+
+    fd_t result = s;
+    s = retired_fd;
+    return result;
+}
+
+#endif
diff --git a/src/ipc_connecter.hpp b/src/ipc_connecter.hpp
new file mode 100644
index 0000000..272e09c
--- /dev/null
+++ b/src/ipc_connecter.hpp
@@ -0,0 +1,112 @@
+/*
+    Copyright (c) 2007-2011 iMatix Corporation
+    Copyright (c) 2007-2011 Other contributors as noted in the AUTHORS file
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the GNU Lesser General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU Lesser General Public License for more details.
+
+    You should have received a copy of the GNU Lesser General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __IPC_CONNECTER_HPP_INCLUDED__
+#define __IPC_CONNECTER_HPP_INCLUDED__
+
+#include "fd.hpp"
+#include "ip.hpp"
+#include "own.hpp"
+#include "io_object.hpp"
+#include "stdint.hpp"
+
+namespace zmq
+{
+
+    class ipc_connecter_t : public own_t, public io_object_t
+    {
+    public:
+
+        //  If 'delay' is true connecter first waits for a while, then starts
+        //  connection process.
+        ipc_connecter_t (class io_thread_t *io_thread_,
+            class session_t *session_, const options_t &options_,
+            const char *address_, bool delay_);
+        ~ipc_connecter_t ();
+
+    private:
+
+        //  ID of the timer used to delay the reconnection.
+        enum {reconnect_timer_id = 1};
+
+        //  Handlers for incoming commands.
+        void process_plug ();
+
+        //  Handlers for I/O events.
+        void in_event ();
+        void out_event ();
+        void timer_event (int id_);
+
+        //  Internal function to start the actual connection establishment.
+        void start_connecting ();
+
+        //  Internal function to add a reconnect timer
+        void add_reconnect_timer();
+
+        //  Internal function to return a reconnect backoff delay.
+        //  Will modify the current_reconnect_ivl used for next call
+        //  Returns the currently used interval
+        int get_new_reconnect_ivl ();
+
+        //  Set address to connect to.
+        int set_address (const char *addr_);
+
+        //  Open TCP connecting socket. Returns -1 in case of error,
+        //  0 if connect was successfull immediately and 1 if async connect
+        //  was launched.
+        int open ();
+
+        //  Close the connecting socket.
+        int close ();
+
+        //  Get the file descriptor of newly created connection. Returns
+        //  retired_fd if the connection was unsuccessfull.
+        fd_t connect ();
+
+        //  Address to connect to.
+        sockaddr_storage addr;
+        socklen_t addr_len;
+
+        //  Underlying socket.
+        fd_t s;
+
+        //  Handle corresponding to the listening socket.
+        handle_t handle;
+
+        //  If true file descriptor is registered with the poller and 'handle'
+        //  contains valid value.
+        bool handle_valid;
+
+        //  If true, connecter is waiting a while before trying to connect.
+        bool wait;
+
+        //  Reference to the session we belong to.
+        class session_t *session;
+
+        //  Current reconnect ivl, updated for backoff strategy
+        int current_reconnect_ivl;
+
+        ipc_connecter_t (const ipc_connecter_t&);
+        const ipc_connecter_t &operator = (const ipc_connecter_t&);
+    };
+
+}
+
+#endif
diff --git a/src/ipc_listener.cpp b/src/ipc_listener.cpp
new file mode 100644
index 0000000..e11d522
--- /dev/null
+++ b/src/ipc_listener.cpp
@@ -0,0 +1,351 @@
+/*
+    Copyright (c) 2007-2011 iMatix Corporation
+    Copyright (c) 2007-2011 Other contributors as noted in the AUTHORS file
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the GNU Lesser General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU Lesser General Public License for more details.
+
+    You should have received a copy of the GNU Lesser General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include <new>
+
+#include <string.h>
+
+#include "ipc_listener.hpp"
+#include "platform.hpp"
+#include "tcp_engine.hpp"
+#include "io_thread.hpp"
+#include "session.hpp"
+#include "config.hpp"
+#include "err.hpp"
+
+#ifdef ZMQ_HAVE_WINDOWS
+#include "windows.hpp"
+#else
+#include <unistd.h>
+#include <sys/socket.h>
+#include <arpa/inet.h>
+#include <netinet/tcp.h>
+#include <netinet/in.h>
+#include <netdb.h>
+#include <fcntl.h>
+#ifndef ZMQ_HAVE_OPENVMS
+#include <sys/un.h>
+#else
+#include <ioctl.h>
+#endif
+#endif
+
+zmq::ipc_listener_t::ipc_listener_t (io_thread_t *io_thread_,
+      socket_base_t *socket_, const options_t &options_) :
+    own_t (io_thread_, options_),
+    io_object_t (io_thread_),
+    has_file (false),
+    s (retired_fd),
+    socket (socket_)
+{
+    memset (&addr, 0, sizeof (addr));
+    addr_len = 0;
+}
+
+zmq::ipc_listener_t::~ipc_listener_t ()
+{
+    if (s != retired_fd)
+        close ();
+}
+
+void zmq::ipc_listener_t::process_plug ()
+{
+    //  Start polling for incoming connections.
+    handle = add_fd (s);
+    set_pollin (handle);
+}
+
+void zmq::ipc_listener_t::process_term (int linger_)
+{
+    rm_fd (handle);
+    own_t::process_term (linger_);
+}
+
+void zmq::ipc_listener_t::in_event ()
+{
+    fd_t fd = accept ();
+
+    //  If connection was reset by the peer in the meantime, just ignore it.
+    //  TODO: Handle specific errors like ENFILE/EMFILE etc.
+    if (fd == retired_fd)
+        return;
+
+    //  Create the engine object for this connection.
+    tcp_engine_t *engine = new (std::nothrow) tcp_engine_t (fd, options);
+    alloc_assert (engine);
+
+    //  Choose I/O thread to run connecter in. Given that we are already
+    //  running in an I/O thread, there must be at least one available.
+    io_thread_t *io_thread = choose_io_thread (options.affinity);
+    zmq_assert (io_thread);
+
+    //  Create and launch a session object. 
+    session_t *session = new (std::nothrow)
+        session_t (io_thread, false, socket, options, NULL, NULL);
+    alloc_assert (session);
+    session->inc_seqnum ();
+    launch_child (session);
+    send_attach (session, engine, false);
+}
+
+#ifdef ZMQ_HAVE_WINDOWS
+
+int zmq::ipc_listener_t::set_address (const char *protocol_, const char *addr_)
+{
+    //  IPC protocol is not supported on Windows platform.
+    if (strcmp (protocol_, "tcp") != 0 ) {
+        errno = EPROTONOSUPPORT;
+        return -1;
+    }
+
+    //  Convert the interface into sockaddr_in structure.
+    int rc = resolve_ip_interface (&addr, &addr_len, addr_);
+    if (rc != 0)
+        return rc;
+
+    //  Create a listening socket.
+    s = ::socket (addr.ss_family, SOCK_STREAM, IPPROTO_TCP);
+    if (s == INVALID_SOCKET) {
+        wsa_error_to_errno ();
+        return -1;
+    }
+
+    //  Allow reusing of the address.
+    int flag = 1;
+    rc = setsockopt (s, SOL_SOCKET, SO_EXCLUSIVEADDRUSE,
+        (const char*) &flag, sizeof (int));
+    wsa_assert (rc != SOCKET_ERROR);
+
+    //  Bind the socket to the network interface and port.
+    rc = bind (s, (struct sockaddr*) &addr, addr_len);
+    if (rc == SOCKET_ERROR) {
+        wsa_error_to_errno ();
+        return -1;
+    }
+
+    //  Listen for incomming connections.
+    rc = listen (s, options.backlog);
+    if (rc == SOCKET_ERROR) {
+        wsa_error_to_errno ();
+        return -1;
+    }
+
+    return 0;
+}
+
+int zmq::ipc_listener_t::close ()
+{
+    zmq_assert (s != retired_fd);
+    int rc = closesocket (s);
+    wsa_assert (rc != SOCKET_ERROR);
+    s = retired_fd;
+    return 0;
+}
+
+zmq::fd_t zmq::ipc_listener_t::accept ()
+{
+    zmq_assert (s != retired_fd);
+
+    //  Accept one incoming connection.
+    fd_t sock = ::accept (s, NULL, NULL);
+    if (sock == INVALID_SOCKET && 
+          (WSAGetLastError () == WSAEWOULDBLOCK ||
+          WSAGetLastError () == WSAECONNRESET))
+        return retired_fd;
+
+    zmq_assert (sock != INVALID_SOCKET);
+
+    // Set to non-blocking mode.
+    unsigned long argp = 1;
+    int rc = ioctlsocket (sock, FIONBIO, &argp);
+    wsa_assert (rc != SOCKET_ERROR);
+
+    return sock;
+}
+
+#else
+
+int zmq::ipc_listener_t::set_address (const char *protocol_, const char *addr_)
+{
+    if (strcmp (protocol_, "tcp") == 0 ) {
+
+        //  Resolve the sockaddr to bind to.
+        int rc = resolve_ip_interface (&addr, &addr_len, addr_);
+        if (rc != 0)
+            return -1;
+
+        //  Create a listening socket.
+        s = ::socket (addr.ss_family, SOCK_STREAM, IPPROTO_TCP);
+        if (s == -1)
+            return -1;
+
+        //  Allow reusing of the address.
+        int flag = 1;
+        rc = setsockopt (s, SOL_SOCKET, SO_REUSEADDR, &flag, sizeof (int));
+        errno_assert (rc == 0);
+
+        //  Bind the socket to the network interface and port.
+        rc = bind (s, (struct sockaddr*) &addr, addr_len);
+        if (rc != 0) {
+            int err = errno;
+            if (close () != 0)
+                return -1;
+            errno = err;
+            return -1;
+        }
+
+        //  Listen for incomming connections.
+        rc = listen (s, options.backlog);
+        if (rc != 0) {
+            int err = errno;
+            if (close () != 0)
+                return -1;
+            errno = err;
+            return -1;
+        }
+
+        return 0;
+    }
+#ifndef ZMQ_HAVE_OPENVMS
+    else if (strcmp (protocol_, "ipc") == 0) {
+
+        //  Get rid of the file associated with the UNIX domain socket that
+        //  may have been left behind by the previous run of the application.
+        ::unlink (addr_);
+
+        //  Convert the address into sockaddr_un structure.
+        int rc = resolve_local_path (&addr, &addr_len, addr_);
+        if (rc != 0)
+            return -1;
+
+        //  Create a listening socket.
+        s = ::socket (AF_UNIX, SOCK_STREAM, 0);
+        if (s == -1)
+            return -1;
+
+        //  Set the non-blocking flag.
+        int flag = fcntl (s, F_GETFL, 0);
+        if (flag == -1) 
+            flag = 0;
+        rc = fcntl (s, F_SETFL, flag | O_NONBLOCK);
+        errno_assert (rc != -1);
+
+        //  Bind the socket to the file path.
+        rc = bind (s, (struct sockaddr*) &addr, addr_len);
+        if (rc != 0) {
+            int err = errno;
+            if (close () != 0)
+                return -1;
+            errno = err;
+            return -1;
+        }
+        has_file = true;
+
+        //  Listen for incomming connections.
+        rc = listen (s, options.backlog);
+        if (rc != 0) {
+            int err = errno;
+            if (close () != 0)
+                return -1;
+            errno = err;
+            return -1;
+        }
+
+        return 0;
+    }
+#endif
+    else {
+        errno = EPROTONOSUPPORT;
+        return -1;
+    }    
+}
+
+int zmq::ipc_listener_t::close ()
+{
+    zmq_assert (s != retired_fd);
+    int rc = ::close (s);
+    if (rc != 0)
+        return -1;
+    s = retired_fd;
+
+#ifndef ZMQ_HAVE_OPENVMS
+    //  If there's an underlying UNIX domain socket, get rid of the file it
+    //  is associated with.
+    struct sockaddr_un *su = (struct sockaddr_un*) &addr;
+    if (AF_UNIX == su->sun_family && has_file) {
+        rc = ::unlink(su->sun_path);
+        if (rc != 0)
+            return -1;
+    }
+#endif
+
+    return 0;
+}
+
+zmq::fd_t zmq::ipc_listener_t::accept ()
+{
+    zmq_assert (s != retired_fd);
+
+    //  Accept one incoming connection.
+    fd_t sock = ::accept (s, NULL, NULL);
+
+#if (defined ZMQ_HAVE_LINUX || defined ZMQ_HAVE_FREEBSD || \
+     defined ZMQ_HAVE_OPENBSD || defined ZMQ_HAVE_OSX || \
+     defined ZMQ_HAVE_OPENVMS || defined ZMQ_HAVE_NETBSD || \
+     defined ZMQ_HAVE_CYGWIN)
+    if (sock == -1 && 
+        (errno == EAGAIN || errno == EWOULDBLOCK || 
+         errno == EINTR || errno == ECONNABORTED))
+        return retired_fd;
+#elif (defined ZMQ_HAVE_SOLARIS || defined ZMQ_HAVE_AIX)
+    if (sock == -1 && 
+        (errno == EWOULDBLOCK || errno == EINTR || 
+         errno == ECONNABORTED || errno == EPROTO))
+        return retired_fd;
+#elif defined ZMQ_HAVE_HPUX
+    if (sock == -1 && 
+        (errno == EAGAIN || errno == EWOULDBLOCK || 
+         errno == EINTR || errno == ECONNABORTED || errno == ENOBUFS))
+        return retired_fd;
+#elif defined ZMQ_HAVE_QNXNTO 
+    if (sock == -1 && 
+        (errno == EWOULDBLOCK || errno == EINTR || errno == ECONNABORTED))
+        return retired_fd;
+#endif
+
+    errno_assert (sock != -1); 
+
+    // Set to non-blocking mode.
+#ifdef ZMQ_HAVE_OPENVMS
+    int flags = 1;
+    int rc = ioctl (sock, FIONBIO, &flags);
+    errno_assert (rc != -1);
+#else
+    int flags = fcntl (s, F_GETFL, 0);
+    if (flags == -1)
+        flags = 0;
+    int rc = fcntl (sock, F_SETFL, flags | O_NONBLOCK);
+    errno_assert (rc != -1);
+#endif
+
+    return sock;
+}
+
+#endif
diff --git a/src/ipc_listener.hpp b/src/ipc_listener.hpp
new file mode 100644
index 0000000..5d0e1d1
--- /dev/null
+++ b/src/ipc_listener.hpp
@@ -0,0 +1,83 @@
+/*
+    Copyright (c) 2007-2011 iMatix Corporation
+    Copyright (c) 2007-2011 Other contributors as noted in the AUTHORS file
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the GNU Lesser General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU Lesser General Public License for more details.
+
+    You should have received a copy of the GNU Lesser General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZMQ_IPC_LISTENER_HPP_INCLUDED__
+#define __ZMQ_IPC_LISTENER_HPP_INCLUDED__
+
+#include "fd.hpp"
+#include "ip.hpp"
+#include "own.hpp"
+#include "io_object.hpp"
+#include "stdint.hpp"
+
+namespace zmq
+{
+
+    class ipc_listener_t : public own_t, public io_object_t
+    {
+    public:
+
+        ipc_listener_t (class io_thread_t *io_thread_,
+            class socket_base_t *socket_, const options_t &options_);
+        ~ipc_listener_t ();
+
+        //  Set address to listen on.
+        int set_address (const char* protocol_, const char *addr_);
+
+    private:
+
+        //  Handlers for incoming commands.
+        void process_plug ();
+        void process_term (int linger_);
+
+        //  Handlers for I/O events.
+        void in_event ();
+
+        //  Close the listening socket.
+        int close ();
+
+        //  Accept the new connection. Returns the file descriptor of the
+        //  newly created connection. The function may return retired_fd
+        //  if the connection was dropped while waiting in the listen backlog.
+        fd_t accept ();
+
+        //  Address to listen on.
+        sockaddr_storage addr;
+        socklen_t addr_len;
+
+        //  True, if the undelying file for UNIX domain socket exists.
+        bool has_file;
+
+        //  Underlying socket.
+        fd_t s;
+
+        //  Handle corresponding to the listening socket.
+        handle_t handle;
+
+        //  Socket the listerner belongs to.
+        class socket_base_t *socket;
+
+        ipc_listener_t (const ipc_listener_t&);
+        const ipc_listener_t &operator = (const ipc_listener_t&);
+    };
+
+}
+
+#endif
diff --git a/src/session.cpp b/src/session.cpp
index b857724..5db080f 100644
--- a/src/session.cpp
+++ b/src/session.cpp
@@ -25,6 +25,7 @@
 #include "pipe.hpp"
 #include "likely.hpp"
 #include "tcp_connecter.hpp"
+#include "ipc_connecter.hpp"
 #include "vtcp_connecter.hpp"
 #include "pgm_sender.hpp"
 #include "pgm_receiver.hpp"
@@ -305,12 +306,17 @@ void zmq::session_t::start_connecting (bool wait_)
 
     //  Create the connecter object.
 
-    //  Both TCP and IPC transports are using the same infrastructure.
-    if (protocol == "tcp" || protocol == "ipc") {
-
+    if (protocol == "tcp") {
         tcp_connecter_t *connecter = new (std::nothrow) tcp_connecter_t (
-            io_thread, this, options, protocol.c_str (), address.c_str (),
-            wait_);
+            io_thread, this, options, address.c_str (), wait_);
+        alloc_assert (connecter);
+        launch_child (connecter);
+        return;
+    }
+
+    if (protocol == "ipc") {
+        ipc_connecter_t *connecter = new (std::nothrow) ipc_connecter_t (
+            io_thread, this, options, address.c_str (), wait_);
         alloc_assert (connecter);
         launch_child (connecter);
         return;
diff --git a/src/socket_base.cpp b/src/socket_base.cpp
index fb0f5fd..975934f 100644
--- a/src/socket_base.cpp
+++ b/src/socket_base.cpp
@@ -35,6 +35,7 @@
 
 #include "socket_base.hpp"
 #include "tcp_listener.hpp"
+#include "ipc_listener.hpp"
 #include "vtcp_listener.hpp"
 #include "tcp_connecter.hpp"
 #include "io_thread.hpp"
@@ -363,7 +364,7 @@ int zmq::socket_base_t::bind (const char *addr_)
         return -1;
     }
 
-    if (protocol == "tcp" || protocol == "ipc") {
+    if (protocol == "tcp") {
         tcp_listener_t *listener = new (std::nothrow) tcp_listener_t (
             io_thread, this, options);
         alloc_assert (listener);
@@ -376,6 +377,19 @@ int zmq::socket_base_t::bind (const char *addr_)
         return 0;
     }
 
+    if (protocol == "ipc") {
+        ipc_listener_t *listener = new (std::nothrow) ipc_listener_t (
+            io_thread, this, options);
+        alloc_assert (listener);
+        int rc = listener->set_address (protocol.c_str(), address.c_str ());
+        if (rc != 0) {
+            delete listener;
+            return -1;
+        }
+        launch_child (listener);
+        return 0;
+    }
+
 #if defined ZMQ_HAVE_VTCP
     if (protocol == "vtcp") {
         vtcp_listener_t *listener = new (std::nothrow) vtcp_listener_t (
diff --git a/src/tcp_connecter.cpp b/src/tcp_connecter.cpp
index 6deb145..76ec0bc 100644
--- a/src/tcp_connecter.cpp
+++ b/src/tcp_connecter.cpp
@@ -47,7 +47,7 @@
 
 zmq::tcp_connecter_t::tcp_connecter_t (class io_thread_t *io_thread_,
       class session_t *session_, const options_t &options_,
-      const char *protocol_, const char *address_, bool wait_) :
+      const char *address_, bool wait_) :
     own_t (io_thread_, options_),
     io_object_t (io_thread_),
     s (retired_fd),
@@ -59,8 +59,10 @@ zmq::tcp_connecter_t::tcp_connecter_t (class io_thread_t *io_thread_,
     memset (&addr, 0, sizeof (addr));
     addr_len = 0;
 
-    int rc = set_address (protocol_, address_);
-    zmq_assert (rc == 0); //TODO: take care ENOMEM, EINVAL
+    //  TODO: set_addess should be called separately, so that the error
+    //  can be propagated.
+    int rc = set_address (address_);
+    zmq_assert (rc == 0);
 }
 
 zmq::tcp_connecter_t::~tcp_connecter_t ()
@@ -104,6 +106,26 @@ void zmq::tcp_connecter_t::out_event ()
         return;
     }
 
+    //  Disable Nagle's algorithm. We are doing data batching on 0MQ level,
+    //  so using Nagle wouldn't improve throughput in anyway, but it would
+    //  hurt latency.
+    int nodelay = 1;
+    int rc = setsockopt (fd, IPPROTO_TCP, TCP_NODELAY, (char*) &nodelay,
+        sizeof (int));
+#ifdef ZMQ_HAVE_WINDOWS
+    wsa_assert (rc != SOCKET_ERROR);
+#else
+    errno_assert (rc == 0);
+#endif
+
+#ifdef ZMQ_HAVE_OPENVMS
+    //  Disable delayed acknowledgements as they hurt latency is serious manner.
+    int nodelack = 1;
+    rc = setsockopt (fd, IPPROTO_TCP, TCP_NODELACK, (char*) &nodelack,
+        sizeof (int));
+    errno_assert (rc != SOCKET_ERROR);
+#endif
+
     //  Create the engine object for this connection.
     tcp_engine_t *engine = new (std::nothrow) tcp_engine_t (fd, options);
     alloc_assert (engine);
@@ -175,13 +197,9 @@ int zmq::tcp_connecter_t::get_new_reconnect_ivl ()
 
 #ifdef ZMQ_HAVE_WINDOWS
 
-int zmq::tcp_connecter_t::set_address (const char *protocol_, const char *addr_)
+int zmq::tcp_connecter_t::set_address (const char *addr_)
 {
-    if (strcmp (protocol_, "tcp") == 0)
-        return resolve_ip_hostname (&addr, &addr_len, addr_);
-
-    errno = EPROTONOSUPPORT;
-    return -1;    
+    return resolve_ip_hostname (&addr, &addr_len, addr_);
 }
 
 int zmq::tcp_connecter_t::open ()
@@ -254,15 +272,9 @@ zmq::fd_t zmq::tcp_connecter_t::connect ()
 
 #else
 
-int zmq::tcp_connecter_t::set_address (const char *protocol_, const char *addr_)
+int zmq::tcp_connecter_t::set_address (const char *addr_)
 {
-    if (strcmp (protocol_, "tcp") == 0)
-        return resolve_ip_hostname (&addr, &addr_len, addr_);
-    else if (strcmp (protocol_, "ipc") == 0)
-        return resolve_local_path (&addr, &addr_len, addr_);
-
-    errno = EPROTONOSUPPORT;
-    return -1;
+    return resolve_ip_hostname (&addr, &addr_len, addr_);
 }
 
 int zmq::tcp_connecter_t::open ()
diff --git a/src/tcp_connecter.hpp b/src/tcp_connecter.hpp
index 41f2c93..6c7aa1b 100644
--- a/src/tcp_connecter.hpp
+++ b/src/tcp_connecter.hpp
@@ -38,7 +38,7 @@ namespace zmq
         //  connection process.
         tcp_connecter_t (class io_thread_t *io_thread_,
             class session_t *session_, const options_t &options_,
-            const char *protocol_, const char *address_, bool delay_);
+            const char *address_, bool delay_);
         ~tcp_connecter_t ();
 
     private:
@@ -66,7 +66,7 @@ namespace zmq
         int get_new_reconnect_ivl ();
 
         //  Set address to connect to.
-        int set_address (const char *protocol, const char *addr_);
+        int set_address (const char *addr_);
 
         //  Open TCP connecting socket. Returns -1 in case of error,
         //  0 if connect was successfull immediately and 1 if async connect
diff --git a/src/tcp_engine.cpp b/src/tcp_engine.cpp
index 4cb535a..f940b84 100644
--- a/src/tcp_engine.cpp
+++ b/src/tcp_engine.cpp
@@ -99,26 +99,6 @@ zmq::tcp_engine_t::tcp_engine_t (fd_t fd_, const options_t &options_) :
     rc = setsockopt (s, SOL_SOCKET, SO_NOSIGPIPE, &set, sizeof (int));
     errno_assert (rc == 0);
 #endif
-
-    //  Disable Nagle's algorithm. We are doing data batching on 0MQ level,
-    //  so using Nagle wouldn't improve throughput in anyway, but it would
-    //  hurt latency.
-    int nodelay = 1;
-    rc = setsockopt (s, IPPROTO_TCP, TCP_NODELAY, (char*) &nodelay,
-        sizeof (int));
-#ifdef ZMQ_HAVE_WINDOWS
-    wsa_assert (rc != SOCKET_ERROR);
-#else
-    errno_assert (rc == 0);
-#endif
-
-#ifdef ZMQ_HAVE_OPENVMS
-    //  Disable delayed acknowledgements as they hurt latency is serious manner.
-    int nodelack = 1;
-    rc = setsockopt (s, IPPROTO_TCP, TCP_NODELACK, (char*) &nodelack,
-        sizeof (int));
-    errno_assert (rc != SOCKET_ERROR);
-#endif
 }
 
 zmq::tcp_engine_t::~tcp_engine_t ()
diff --git a/src/tcp_listener.cpp b/src/tcp_listener.cpp
index d334768..360be8c 100644
--- a/src/tcp_listener.cpp
+++ b/src/tcp_listener.cpp
@@ -87,6 +87,26 @@ void zmq::tcp_listener_t::in_event ()
     if (fd == retired_fd)
         return;
 
+    //  Disable Nagle's algorithm. We are doing data batching on 0MQ level,
+    //  so using Nagle wouldn't improve throughput in anyway, but it would
+    //  hurt latency.
+    int nodelay = 1;
+    int rc = setsockopt (fd, IPPROTO_TCP, TCP_NODELAY, (char*) &nodelay,
+        sizeof (int));
+#ifdef ZMQ_HAVE_WINDOWS
+    wsa_assert (rc != SOCKET_ERROR);
+#else
+    errno_assert (rc == 0);
+#endif
+
+#ifdef ZMQ_HAVE_OPENVMS
+    //  Disable delayed acknowledgements as they hurt latency is serious manner.
+    int nodelack = 1;
+    rc = setsockopt (fd, IPPROTO_TCP, TCP_NODELACK, (char*) &nodelack,
+        sizeof (int));
+    errno_assert (rc != SOCKET_ERROR);
+#endif
+
     //  Create the engine object for this connection.
     tcp_engine_t *engine = new (std::nothrow) tcp_engine_t (fd, options);
     alloc_assert (engine);