diff options
author | malosek <malosek@fastmq.com> | 2009-09-11 17:58:37 +0200 |
---|---|---|
committer | malosek <malosek@fastmq.com> | 2009-09-11 17:58:37 +0200 |
commit | 1a4d6f91194c52795808baa07dcd61a20ff599be (patch) | |
tree | 9e5a59750b32f90a862a47c88d79653c4a75f12f | |
parent | e940878b3f8ff332df534c4df463cfdf085cc787 (diff) |
added OpenPGM sender - ZMQ_PUB
-rw-r--r-- | Makefile.am | 4 | ||||
-rw-r--r-- | c/zmq.h | 2 | ||||
-rw-r--r-- | src/Makefile.am | 6 | ||||
-rw-r--r-- | src/app_thread.cpp | 2 | ||||
-rw-r--r-- | src/config.hpp | 4 | ||||
-rw-r--r-- | src/options.cpp | 4 | ||||
-rw-r--r-- | src/options.hpp | 6 | ||||
-rw-r--r-- | src/pgm_sender.cpp | 224 | ||||
-rw-r--r-- | src/pgm_sender.hpp | 109 | ||||
-rw-r--r-- | src/pgm_socket.cpp | 754 | ||||
-rw-r--r-- | src/pgm_socket.hpp | 153 | ||||
-rw-r--r-- | src/socket_base.cpp | 105 | ||||
-rw-r--r-- | src/socket_base.hpp | 5 | ||||
-rw-r--r-- | src/sub.cpp | 2 |
14 files changed, 1360 insertions, 20 deletions
diff --git a/Makefile.am b/Makefile.am index 335d988..18636c0 100644 --- a/Makefile.am +++ b/Makefile.am @@ -17,5 +17,7 @@ endif SUBDIRS = src $(DIR_P) $(DIR_R) $(DIR_J) $(DIR_PERF) DIST_SUBDIRS = src python ruby java perf +EXTRA_DIST = $(top_srcdir)/foreign/openpgm/@pgm_basename@.tar.bz2 + dist-hook: - -rm -rf $(distdir)/third-party/openpgm/$pgm_basename + -rm -rf $(distdir)/foreign/openpgm/@pgm_basename@ @@ -52,6 +52,8 @@ extern "C" { #define ZMQ_IDENTITY 6 #define ZMQ_SUBSCRIBE 7 #define ZMQ_UNSUBSCRIBE 8 +#define ZMQ_RATE 9 +#define ZMQ_RECOVERY_IVL 10 // The operation should be performed in non-blocking mode. I.e. if it cannot // be processed immediately, error should be returned with errno set to EAGAIN. diff --git a/src/Makefile.am b/src/Makefile.am index 68b34fa..f4f338e 100644 --- a/src/Makefile.am +++ b/src/Makefile.am @@ -62,6 +62,8 @@ libzmq_la_SOURCES = $(pgm_sources) \ object.hpp \ options.hpp \ owned.hpp \ + pgm_sender.hpp \ + pgm_socket.hpp \ pipe.hpp \ platform.hpp \ poll.hpp \ @@ -101,6 +103,8 @@ libzmq_la_SOURCES = $(pgm_sources) \ object.cpp \ options.cpp \ owned.cpp \ + pgm_sender.cpp \ + pgm_socket.cpp \ pipe.cpp \ poll.cpp \ select.cpp \ @@ -122,7 +126,7 @@ libzmq_la_SOURCES = $(pgm_sources) \ zmq_listener.cpp \ zmq_listener_init.cpp -libzmq_la_LDFLAGS = -version-info @LTVER@ +libzmq_la_LDFLAGS = -version-info @LTVER@ @LIBZMQ_EXTRA_LDFLAFS@ if BUILD_PGM libzmq_la_CXXFLAGS = -I$(top_srcdir)/foreign/openpgm/@pgm_basename@/openpgm/pgm/include/ -Wall @LIBZMQ_EXTRA_CXXFLAGS@ diff --git a/src/app_thread.cpp b/src/app_thread.cpp index 2bcc724..517b721 100644 --- a/src/app_thread.cpp +++ b/src/app_thread.cpp @@ -145,7 +145,7 @@ zmq::socket_base_t *zmq::app_thread_t::create_socket (int type_) case ZMQ_PUB: case ZMQ_REQ: case ZMQ_REP: - s = new socket_base_t (this); + s = new socket_base_t (this, type_); break; default: // TODO: This should be EINVAL. diff --git a/src/config.hpp b/src/config.hpp index 17e67b9..43a4513 100644 --- a/src/config.hpp +++ b/src/config.hpp @@ -70,8 +70,10 @@ namespace zmq // Maximal number of non-accepted connections that can be held by // TCP listener object. - tcp_connection_backlog = 10 + tcp_connection_backlog = 10, + // Maximum transport data unit size for PGM (TPDU). + pgm_max_tpdu = 1500 }; } diff --git a/src/options.cpp b/src/options.cpp index cd07c44..804cb4f 100644 --- a/src/options.cpp +++ b/src/options.cpp @@ -24,6 +24,8 @@ zmq::options_t::options_t () : lwm (0), swap (0), mask (0), - affinity (0) + affinity (0), + rate (0), + recovery_ivl (0) { } diff --git a/src/options.hpp b/src/options.hpp index faf21b8..9f4a264 100644 --- a/src/options.hpp +++ b/src/options.hpp @@ -37,6 +37,12 @@ namespace zmq uint64_t mask; uint64_t affinity; std::string identity; + + // Maximum tranfer rate [kb/s]. + uint32_t rate; + + // Reliability time interval [s]. + uint32_t recovery_ivl; }; } diff --git a/src/pgm_sender.cpp b/src/pgm_sender.cpp new file mode 100644 index 0000000..60b4c3a --- /dev/null +++ b/src/pgm_sender.cpp @@ -0,0 +1,224 @@ +/* + Copyright (c) 2007-2009 FastMQ Inc. + + This file is part of 0MQ. + + 0MQ is free software; you can redistribute it and/or modify it under + the terms of the Lesser GNU 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 + Lesser GNU General Public License for more details. + + You should have received a copy of the Lesser GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#include "platform.hpp" + +#if defined ZMQ_HAVE_OPENPGM + +#include <iostream> + +#include "io_thread.hpp" +#include "pgm_sender.hpp" +#include "err.hpp" +#include "wire.hpp" + +//#define PGM_SENDER_DEBUG +//#define PGM_SENDER_DEBUG_LEVEL 1 + +// level 1 = key behaviour +// level 2 = processing flow +// level 4 = infos + +#ifndef PGM_SENDER_DEBUG +# define zmq_log(n, ...) while (0) +#else +# define zmq_log(n, ...) do { if ((n) <= PGM_SENDER_DEBUG_LEVEL) \ + { printf (__VA_ARGS__);}} while (0) +#endif + +#ifdef ZMQ_HAVE_LINUX +zmq::pgm_sender_t::pgm_sender_t (io_thread_t *parent_, + const options_t &options_, const char *session_name_) : + io_object_t (parent_), + pgm_socket (false, options_), + options (options_), + session_name (session_name_), + inout (NULL), + out_buffer (NULL), + out_buffer_size (0), + write_size (0), + write_pos (0), + first_message_offset (-1) +{ + +} + +int zmq::pgm_sender_t::init (const char *network_) +{ + return pgm_socket.init (network_); +} + +void zmq::pgm_sender_t::plug (i_inout *inout_) +{ + + // Alocate 2 fds for PGM socket. + int downlink_socket_fd; + int uplink_socket_fd; + + encoder.set_inout (inout_); + + // Fill fds from PGM transport. + pgm_socket.get_sender_fds (&downlink_socket_fd, &uplink_socket_fd); + + // Add downlink_socket_fd into poller. + handle = add_fd (downlink_socket_fd); + + // Add uplink_socket_fd into the poller. + uplink_handle = add_fd (uplink_socket_fd); + + // Set POLLIN. We wont never want to stop polling for uplink = we never + // want to stop porocess NAKs. + set_pollin (uplink_handle); + + // Set POLLOUT for downlink_socket_handle. + set_pollout (handle); + + inout = inout_; + + zmq_log (1, "plug: downlink_socket_fd %i, uplink_socket_fd %i, %s(%i)", + downlink_socket_fd, uplink_socket_fd, __FILE__, __LINE__); + + std::cout << std::flush; +} + +void zmq::pgm_sender_t::unplug () +{ + rm_fd (handle); + rm_fd (uplink_handle); + encoder.set_inout (NULL); + inout = NULL; +} + +void zmq::pgm_sender_t::revive () +{ + set_pollout (handle); +} + +zmq::pgm_sender_t::~pgm_sender_t () +{ + if (out_buffer) { + pgm_socket.free_buffer (out_buffer); + } +} + +// In event on sender side means NAK or SPMR receiving from some peer. +void zmq::pgm_sender_t::in_event () +{ + pgm_socket.process_upstream (); +} + +void zmq::pgm_sender_t::out_event () +{ + + // POLLOUT event from send socket. If write buffer is empty, + // try to read new data from the encoder. + if (write_pos == write_size) { + + // Get buffer if we do not have already one. + if (!out_buffer) { + out_buffer = (unsigned char*) + pgm_socket.get_buffer (&out_buffer_size); + } + + assert (out_buffer_size > 0); + + // First two bytes /sizeof (uint16_t)/ are used to store message + // offset in following steps. + write_size = encoder.read (out_buffer + sizeof (uint16_t), + out_buffer_size - sizeof (uint16_t), &first_message_offset); + write_pos = 0; + + // If there are no data to write stop polling for output. + if (!write_size) { + reset_pollout (handle); + } else { + // Addning uint16_t for offset in a case when encoder returned > 0B. + write_size += sizeof (uint16_t); + } + } + + // If there are any data to write, write them into the socket. + // Note that all data has to written in one write_one_pkt_with_offset call. + if (write_pos < write_size) { + size_t nbytes = write_one_pkt_with_offset (out_buffer + write_pos, + write_size - write_pos, (uint16_t) first_message_offset); + + // We can write all data or 0 which means rate limit reached. + if (write_size - write_pos != nbytes && nbytes != 0) { + zmq_log (1, "write_size - write_pos %i, nbytes %i, %s(%i)", + (int)(write_size - write_pos), (int)nbytes, __FILE__, __LINE__); + assert (false); + } + + // PGM rate limit reached nbytes is 0. + if (!nbytes) { + zmq_log (1, "pgm rate limit reached, %s(%i)\n", __FILE__, __LINE__); + } + + // After sending data slice is owned by tx window. + if (nbytes) { + out_buffer = NULL; + } + + write_pos += nbytes; + } + +} + +/* +void zmq::bp_pgm_sender_t::revive (pipe_t *pipe_) +{ + // We have some messages in encoder. + if (!shutting_down) { + + // Forward the revive command to the pipe. + engine_base_t <false, true>::revive (pipe_); + + // There is at least one engine (that one which sent revive) that + // has messages ready. Try to write data to the socket, thus + // eliminating one polling for POLLOUT event. + // Note that if write_size is zero it means that buffer is empty and + // we can read data from encoder. + if (!write_size) { + poller->set_pollout (handle); + out_event (handle); + } + } +} + +*/ +size_t zmq::pgm_sender_t::write_one_pkt_with_offset (unsigned char *data_, + size_t size_, uint16_t offset_) +{ + zmq_log (1, "data_size %i, first message offset %i, %s(%i)", + (int) size_, offset_, __FILE__, __LINE__); + + std::cout << std::flush; + + // Put offset information in the buffer. + put_uint16 (data_, offset_); + + // Send data. + size_t nbytes = pgm_socket.send (data_, size_); + + return nbytes; +} +#endif + +#endif diff --git a/src/pgm_sender.hpp b/src/pgm_sender.hpp new file mode 100644 index 0000000..2ad2a15 --- /dev/null +++ b/src/pgm_sender.hpp @@ -0,0 +1,109 @@ +/* + Copyright (c) 2007-2009 FastMQ Inc. + + This file is part of 0MQ. + + 0MQ is free software; you can redistribute it and/or modify it under + the terms of the Lesser GNU 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 + Lesser GNU General Public License for more details. + + You should have received a copy of the Lesser GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#ifndef __ZMQ_BP_PGM_SENDER_HPP_INCLUDED__ +#define __ZMQ_BP_PGM_SENDER_HPP_INCLUDED__ + +#include "platform.hpp" + +#if defined ZMQ_HAVE_OPENPGM + +#include <vector> + +#include "stdint.hpp" +#include "io_object.hpp" +#include "i_engine.hpp" +#include "options.hpp" +#include "pgm_socket.hpp" +#include "zmq_encoder.hpp" + +namespace zmq +{ + + class pgm_sender_t : public io_object_t, public i_engine + { + + public: + pgm_sender_t (class io_thread_t *parent_, const options_t &options_, + const char *session_name_); + ~pgm_sender_t (); + + int init (const char *network_); + + // i_engine interface implementation. + void plug (struct i_inout *inout_); + void unplug (); + void revive (); + + // i_poll_events interface implementation. + void in_event (); + void out_event (); + + private: + + // Send one APDU with first message offset information. + // Note that first 2 bytes in data_ are used to store the offset_ + // and thus user data has to start at data_ + sizeof (uint16_t). + size_t write_one_pkt_with_offset (unsigned char *data_, size_t size_, + uint16_t offset_); + + // Message encoder. + zmq_encoder_t encoder; + + // PGM socket. + pgm_socket_t pgm_socket; + + // Socket options. + options_t options; + + // Name of the session associated with the connecter. + std::string session_name; + + // Poll handle associated with PGM socket. + handle_t handle; + handle_t uplink_handle; + + // ? + i_inout *inout; + + // Output buffer from pgm_socket. +#ifdef ZMQ_HAVE_WINDOWS + unsigned char out_buffer [pgm_win_max_apdu]; +#else + unsigned char *out_buffer; + + // Output buffer size. + size_t out_buffer_size; +#endif + + size_t write_size; + size_t write_pos; + + // Offset of the first mesage in data chunk taken from encoder. + int first_message_offset; + + pgm_sender_t (const pgm_sender_t&); + void operator = (const pgm_sender_t&); + }; + +} + +#endif + +#endif diff --git a/src/pgm_socket.cpp b/src/pgm_socket.cpp new file mode 100644 index 0000000..5e2e764 --- /dev/null +++ b/src/pgm_socket.cpp @@ -0,0 +1,754 @@ +/* + Copyright (c) 2007-2009 FastMQ Inc. + + This file is part of 0MQ. + + 0MQ is free software; you can redistribute it and/or modify it under + the terms of the Lesser GNU 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 + Lesser GNU General Public License for more details. + + You should have received a copy of the Lesser GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#include "platform.hpp" + +#if defined ZMQ_HAVE_OPENPGM + +#ifdef ZMQ_HAVE_LINUX +#include <pgm/pgm.h> +#else +#include <Winsock2.h> +#include <Wsrm.h> +#include <ws2spi.h> +#endif + +#include <string> +#include <iostream> + +#include "options.hpp" +#include "pgm_socket.hpp" +#include "config.hpp" +#include "err.hpp" + +//#define PGM_SOCKET_DEBUG +//#define PGM_SOCKET_DEBUG_LEVEL 1 + +// level 1 = key behaviour +// level 2 = processing flow +// level 4 = infos + +#ifndef PGM_SOCKET_DEBUG +# define zmq_log(n, ...) while (0) +#else +# define zmq_log(n, ...) do { if ((n) <= PGM_SOCKET_DEBUG_LEVEL) \ + { printf (__VA_ARGS__);}} while (0) +#endif + +#ifdef ZMQ_HAVE_LINUX + +zmq::pgm_socket_t::pgm_socket_t (bool receiver_, const options_t &options_) : + g_transport (NULL), + options (options_), + receiver (receiver_), + port_number (0), + udp_encapsulation (false), + pgm_msgv (NULL), + nbytes_rec (0), + nbytes_processed (0), + pgm_msgv_processed (0), + pgm_msgv_len (0) +{ + +} + +int zmq::pgm_socket_t::init (const char *network_) +{ + // Check if we are encapsulating into UDP, natwork string has to + // start with udp:. + const char *network_ptr = network_; + + if (strlen (network_) >= 4 && network_ [0] == 'u' && + network_ [1] == 'd' && network_ [2] == 'p' && + network_ [3] == ':') { + + // Shift interface_ptr after ':'. + network_ptr += 4; + + udp_encapsulation = true; + } + + // Parse port number. + const char *port_delim = strchr (network_ptr, ':'); + if (!port_delim) { + errno = EINVAL; + return -1; + } + + port_number = atoi (port_delim + 1); + + // Store interface string. + if (port_delim <= network_ptr) { + errno = EINVAL; + return -1; + } + + if (port_delim - network_ptr >= (int) sizeof (network) - 1) { + errno = EINVAL; + return -1; + } + + memset (network, '\0', sizeof (network)); + memcpy (network, network_ptr, port_delim - network_ptr); + + + zmq_log (1, "parsed: network %s, port %i, udp encaps. %s, %s(%i)\n", + network, port_number, udp_encapsulation ? "yes" : "no", + __FILE__, __LINE__); + + // Open PGM transport. + int rc = open_transport (); + if (rc != 0) + return -1; + + // For receiver transport preallocate pgm_msgv array. + // in_batch_size configured in confing.hpp + if (receiver) { + pgm_msgv_len = get_max_apdu_at_once (in_batch_size); + pgm_msgv = new pgm_msgv_t [pgm_msgv_len]; + } + + return 0; +} + +int zmq::pgm_socket_t::open_transport (void) +{ + + zmq_log (1, "Opening PGM: network %s, port %i, udp encaps. %s, %s(%i)\n", + network, port_number, udp_encapsulation ? "yes" : "no", + __FILE__, __LINE__); + + // Can not open transport before destroying old one. + zmq_assert (g_transport == NULL); + + // Set actual_tsi and prev_tsi to zeros. + memset (&tsi, '\0', sizeof (pgm_tsi_t)); + memset (&retired_tsi, '\0', sizeof (pgm_tsi_t)); + + // Zero counter used in msgrecv. + nbytes_rec = 0; + nbytes_processed = 0; + pgm_msgv_processed = 0; + + // Init PGM transport. + // Ensure threading enabled, ensure timer enabled and find PGM protocol id. + // + // Note that if you want to use gettimeofday and sleep for openPGM timing, + // set environment variables PGM_TIMER to "GTOD" + // and PGM_SLEEP to "USLEEP". + int rc = pgm_init (); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // PGM transport GSI. + pgm_gsi_t gsi; + + // PGM transport GSRs. + struct group_source_req recv_gsr, send_gsr; + size_t recv_gsr_len = 1; + + rc = pgm_create_md5_gsi (&gsi); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // On success, 0 is returned. On invalid arguments, -EINVAL is returned. + // If more multicast groups are found than the recv_len parameter, + // -ENOMEM is returned. + rc = pgm_if_parse_transport (network, AF_INET, &recv_gsr, + &recv_gsr_len, &send_gsr); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + if (recv_gsr_len != 1) { + errno = ENOMEM; + return -1; + } + + // If we are using UDP encapsulation update send_gsr & recv_gsr + // structures. Note that send_gsr & recv_gsr has to be updated after + // pgm_if_parse_transport call. + if (udp_encapsulation) { + + // Use the same port for UDP encapsulation. + ((struct sockaddr_in*)&send_gsr.gsr_group)->sin_port = + g_htons (port_number); + ((struct sockaddr_in*)&recv_gsr.gsr_group)->sin_port = + g_htons (port_number); + } + + rc = pgm_transport_create (&g_transport, &gsi, 0, port_number, &recv_gsr, + 1, &send_gsr); + if (rc != 0) { + return -1; + } + + // Common parameters for receiver and sender. + + // Set maximum transport protocol data unit size (TPDU). + rc = pgm_transport_set_max_tpdu (g_transport, pgm_max_tpdu); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set maximum number of network hops to cross. + rc = pgm_transport_set_hops (g_transport, 16); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Receiver transport. + if (receiver) { + + // Set transport->may_close_on_failure to true, + // after data los recvmsgv returns -1 errno set to ECONNRESET. + rc = pgm_transport_set_close_on_failure (g_transport, TRUE); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set transport->can_send_data = FALSE. + // Note that NAKs are still generated by the transport. + rc = pgm_transport_set_recv_only (g_transport, false); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set NAK transmit back-off interval [us]. + rc = pgm_transport_set_nak_bo_ivl (g_transport, 50*1000); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set timeout before repeating NAK [us]. + rc = pgm_transport_set_nak_rpt_ivl (g_transport, 200*1000); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set timeout for receiving RDATA. + rc = pgm_transport_set_nak_rdata_ivl (g_transport, 200*1000); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set retries for NAK without NCF/DATA (NAK_DATA_RETRIES). + rc = pgm_transport_set_nak_data_retries (g_transport, 5); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set retries for NCF after NAK (NAK_NCF_RETRIES). + rc = pgm_transport_set_nak_ncf_retries (g_transport, 2); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set timeout for removing a dead peer [us]. + rc = pgm_transport_set_peer_expiry (g_transport, 5*8192*1000); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set expiration time of SPM Requests [us]. + rc = pgm_transport_set_spmr_expiry (g_transport, 25*1000); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set the size of the receive window. + // + // data rate [B/s] (options.rate is kb/s). + if (options.rate <= 0) { + errno = EINVAL; + return -1; + } + + rc = pgm_transport_set_rxw_max_rte (g_transport, + options.rate * 1000 / 8); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Recovery interval [s]. + if (options.recovery_ivl <= 0) { + errno = EINVAL; + return -1; + } + + rc = pgm_transport_set_rxw_secs (g_transport, options.recovery_ivl); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Sender transport. + } else { + + // Set transport->can_recv = FALSE, waiting_pipe wont not be read. + rc = pgm_transport_set_send_only (g_transport, TRUE); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set the size of the send window. + // + // data rate [B/s] (options.rate is kb/s). + if (options.rate <= 0) { + errno = EINVAL; + return -1; + } + + rc = pgm_transport_set_txw_max_rte (g_transport, + options.rate * 1000 / 8); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Recovery interval [s]. + if (options.recovery_ivl <= 0) { + errno = EINVAL; + return -1; + } + + rc = pgm_transport_set_txw_secs (g_transport, options.recovery_ivl); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Preallocate full transmit window. For simplification always + // worst case is used (40 bytes ipv6 header and 20 bytes UDP + // encapsulation). + int to_preallocate = options.recovery_ivl * (options.rate * 1000 / 8) + / (pgm_max_tpdu - 40 - 20); + + rc = pgm_transport_set_txw_preallocate (g_transport, to_preallocate); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + zmq_log (1, "Preallocated %i slices in TX window. %s(%i)\n", + to_preallocate, __FILE__, __LINE__); + + // Set interval of background SPM packets [us]. + rc = pgm_transport_set_ambient_spm (g_transport, 8192 * 1000); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Set intervals of data flushing SPM packets [us]. + guint spm_heartbeat[] = {4 * 1000, 4 * 1000, 8 * 1000, 16 * 1000, + 32 * 1000, 64 * 1000, 128 * 1000, 256 * 1000, 512 * 1000, + 1024 * 1000, 2048 * 1000, 4096 * 1000, 8192 * 1000}; + + rc = pgm_transport_set_heartbeat_spm (g_transport, spm_heartbeat, + G_N_ELEMENTS(spm_heartbeat)); + if (rc != 0) { + errno = EINVAL; + return -1; + } + } + + // Enable multicast loopback. + rc = pgm_transport_set_multicast_loop (g_transport, true); + if (rc != 0) { + errno = EINVAL; + return -1; + } + + // Bind a transport to the specified network devices. + rc = pgm_transport_bind (g_transport); + if (rc != 0) { + return -1; + } + + return 0; +} + +zmq::pgm_socket_t::~pgm_socket_t () +{ + // Celanup. + if (pgm_msgv) { + delete [] pgm_msgv; + } + + if (g_transport) + close_transport (); +} + +void zmq::pgm_socket_t::close_transport (void) +{ + // g_transport has to be valid. + zmq_assert (g_transport); + + pgm_transport_destroy (g_transport, TRUE); + + g_transport = NULL; +} + +// Get receiver fds. recv_fd is from transport->recv_sock +// waiting_pipe_fd is from transport->waiting_pipe [0] +int zmq::pgm_socket_t::get_receiver_fds (int *recv_fd_, + int *waiting_pipe_fd_) +{ + + // For POLLIN there are 2 pollfds in pgm_transport. + int fds_array_size = pgm_receiver_fd_count; + pollfd *fds = new pollfd [fds_array_size]; + memset (fds, '\0', fds_array_size * sizeof (fds)); + + // Retrieve pollfds from pgm_transport. + int rc = pgm_transport_poll_info (g_transport, fds, &fds_array_size, + POLLIN); + + // pgm_transport_poll_info has to return 2 pollfds for POLLIN. + // Note that fds_array_size parameter can be + // changed inside pgm_transport_poll_info call. + zmq_assert (rc == pgm_receiver_fd_count); + + // Store pfds into user allocated space. + *recv_fd_ = fds [0].fd; + *waiting_pipe_fd_ = fds [1].fd; + + delete [] fds; + + return pgm_receiver_fd_count; +} + +// Get fds and store them into user allocated memory. +// sender_fd is from pgm_transport->send_sock. +// receive_fd_ is from transport->recv_sock. +int zmq::pgm_socket_t::get_sender_fds (int *send_fd_, int *receive_fd_) +{ + + // Preallocate pollfds array. + int fds_array_size = pgm_sender_fd_count; + pollfd *fds = new pollfd [fds_array_size]; + memset (fds, '\0', fds_array_size * sizeof (fds)); + + // Retrieve pollfds from pgm_transport + int rc = pgm_transport_poll_info (g_transport, fds, &fds_array_size, + POLLOUT | POLLIN); + + // pgm_transport_poll_info has to return one pollfds for POLLOUT and + // second for POLLIN. + // Note that fds_array_size parameter can be + // changed inside pgm_transport_poll_info call. + zmq_assert (rc == pgm_sender_fd_count); + + // Store pfds into user allocated space. + *receive_fd_ = fds [0].fd; + *send_fd_ = fds [1].fd; + + delete [] fds; + + return pgm_sender_fd_count; +} + +// Send one APDU, transmit window owned memory. +size_t zmq::pgm_socket_t::send (unsigned char *data_, size_t data_len_) +{ + iovec iov = {data_,data_len_}; + + ssize_t nbytes = pgm_transport_send_packetv (g_transport, &iov, 1, + MSG_DONTWAIT | MSG_WAITALL, true); + + zmq_assert (nbytes != -EINVAL); + + if (nbytes == -1 && errno != EAGAIN) { + errno_assert (false); + } + + // If nbytes is -1 and errno is EAGAIN means that we can not send data + // now. We have to call write_one_pkt again. + nbytes = nbytes == -1 ? 0 : nbytes; + + zmq_log (4, "wrote %iB, %s(%i)\n", (int)nbytes, __FILE__, __LINE__); + + // We have to write all data as one packet. + if (nbytes > 0) { + zmq_assert (nbytes == (ssize_t)data_len_); + } + + return nbytes; +} + +// Return max TSDU size without fragmentation from current PGM transport. +size_t zmq::pgm_socket_t::get_max_tsdu_size (void) +{ + return (size_t)pgm_transport_max_tsdu (g_transport, false); +} + +// Returns how many APDUs are needed to fill reading buffer. +size_t zmq::pgm_socket_t::get_max_apdu_at_once (size_t readbuf_size_) +{ + zmq_assert (readbuf_size_ > 0); + + // Read max TSDU size without fragmentation. + size_t max_tsdu_size = get_max_tsdu_size (); + + // Calculate number of APDUs needed to fill the reading buffer. + size_t apdu_count = (int)readbuf_size_ / max_tsdu_size; + + if ((int) readbuf_size_ % max_tsdu_size) + apdu_count ++; + + // Have to have at least one APDU. + zmq_assert (apdu_count); + + return apdu_count; +} + +// Allocate buffer for one packet from the transmit window, The memory buffer +// is owned by the transmit window and so must be returned to the window with +// content via pgm_transport_send() calls or unused with pgm_packetv_free1(). +void *zmq::pgm_socket_t::get_buffer (size_t *size_) +{ + // Store size. + *size_ = get_max_tsdu_size (); + + // Allocate one packet. + return pgm_packetv_alloc (g_transport, false); +} + +// Return an unused packet allocated from the transmit window +// via pgm_packetv_alloc(). +void zmq::pgm_socket_t::free_buffer (void *data_) +{ + pgm_packetv_free1 (g_transport, data_, false); +} + +// pgm_transport_recvmsgv is called to fill the pgm_msgv array up to +// pgm_msgv_len. In subsequent calls data from pgm_msgv structure are +// returned. +ssize_t zmq::pgm_socket_t::receive (void **raw_data_) +{ + + // We just sent all data from pgm_transport_recvmsgv up + // and have to return 0 that another engine in this thread is scheduled. + if (nbytes_rec == nbytes_processed && nbytes_rec > 0) { + + // Reset all the counters. + nbytes_rec = 0; + nbytes_processed = 0; + pgm_msgv_processed = 0; + + return 0; + } + + // If we have are going first time or if we have processed all pgm_msgv_t + // structure previaously read from the pgm socket. + if (nbytes_rec == nbytes_processed) { + + // Check program flow. + zmq_assert (pgm_msgv_processed == 0); + zmq_assert (nbytes_processed == 0); + zmq_assert (nbytes_rec == 0); + + // Receive a vector of Application Protocol Domain Unit's (APDUs) + // from the transport. + nbytes_rec = pgm_transport_recvmsgv (g_transport, pgm_msgv, + pgm_msgv_len, MSG_DONTWAIT); + + // In a case when no ODATA/RDATA fired POLLIN event (SPM...) + // pgm_transport_recvmsg returns -1 with errno == EAGAIN. + if (nbytes_rec == -1 && errno == EAGAIN) { + + // In case if no RDATA/ODATA caused POLLIN 0 is + // returned. + nbytes_rec = 0; + return 0; + } + + // For data loss nbytes_rec == -1 errno == ECONNRESET. + if (nbytes_rec == -1 && errno == ECONNRESET) { + + // In case of dala loss -1 is returned. + zmq_log (1, "Data loss detected, %s(%i)\n", __FILE__, __LINE__); + nbytes_rec = 0; + return -1; + } + + // Catch the rest of the errors. + if (nbytes_rec <= 0) { + zmq_log (1, "received %i B, errno %i, %s(%i)", (int)nbytes_rec, + errno, __FILE__, __LINE__); + errno_assert (nbytes_rec > 0); + } + + zmq_log (4, "received %i bytes\n", (int)nbytes_rec); + } + + zmq_assert (nbytes_rec > 0); + + // Only one APDU per pgm_msgv_t structure is allowed. + zmq_assert (pgm_msgv [pgm_msgv_processed].msgv_iovlen == 1); + + // Take pointers from pgm_msgv_t structure. + *raw_data_ = pgm_msgv[pgm_msgv_processed].msgv_iov->iov_base; + size_t raw_data_len = pgm_msgv[pgm_msgv_processed].msgv_iov->iov_len; + + // Check if peer TSI did not change, this is detection of peer restart. + const pgm_tsi_t *current_tsi = pgm_msgv [pgm_msgv_processed].msgv_tsi; + + // If empty store new TSI. + if (tsi_empty (&tsi)) { + // Store current peer TSI. + memcpy (&tsi, current_tsi, sizeof (pgm_tsi_t)); +#ifdef PGM_SOCKET_DEBUG + uint8_t *gsi = (uint8_t*)(&tsi)->gsi.identifier; +#endif + + zmq_log (1, "First peer TSI: %i.%i.%i.%i.%i.%i.%i, %s(%i)\n", + gsi [0], gsi [1], gsi [2], gsi [3], gsi [4], gsi [5], + ntohs (tsi.sport), __FILE__, __LINE__); + } + + // Compare stored TSI with actual. + if (!tsi_equal (&tsi, current_tsi)) { + // Peer change detected. + zmq_log (1, "Peer change detected, %s(%i)\n", __FILE__, __LINE__); + + // Compare with retired TSI, in case of match ignore APDU. + if (tsi_equal (&retired_tsi, current_tsi)) { + zmq_log (1, "Retired TSI - ignoring APDU, %s(%i)\n", + __FILE__, __LINE__); + + // Move the the next pgm_msgv_t structure. + pgm_msgv_processed++; + nbytes_processed +=raw_data_len; + + return 0; + + } else { + zmq_log (1, "New TSI, %s(%i)\n", __FILE__, __LINE__); + + // Store new TSI and move last valid to retired_tsi + memcpy (&retired_tsi, &tsi, sizeof (pgm_tsi_t)); + memcpy (&tsi, current_tsi, sizeof (pgm_tsi_t)); + +#ifdef PGM_SOCKET_DEBUG + uint8_t *gsi = (uint8_t*)(&retired_tsi)->gsi.identifier; +#endif + zmq_log (1, "retired TSI: %i.%i.%i.%i.%i.%i.%i, %s(%i)\n", + gsi [0], gsi [1], gsi [2], gsi [3], gsi [4], gsi [5], + ntohs (retired_tsi.sport), __FILE__, __LINE__); + +#ifdef PGM_SOCKET_DEBUG + gsi = (uint8_t*)(&tsi)->gsi.identifier; +#endif + zmq_log (1, " TSI: %i.%i.%i.%i.%i.%i.%i, %s(%i)\n", + gsi [0], gsi [1], gsi [2], gsi [3], gsi [4], gsi [5], + ntohs (tsi.sport), __FILE__, __LINE__); + + // Peers change is recognized as a GAP. + return -1; + } + + } + + // Move the the next pgm_msgv_t structure. + pgm_msgv_processed++; + nbytes_processed +=raw_data_len; + + zmq_log (4, "sendig up %i bytes\n", (int)raw_data_len); + + return raw_data_len; +} + +void zmq::pgm_socket_t::process_upstream (void) +{ + zmq_log (1, "On upstream packet, %s(%i)\n", __FILE__, __LINE__); + // We acctually do not want to read any data here we are going to + // process NAK. + pgm_msgv_t dummy_msg; + + ssize_t dummy_bytes = pgm_transport_recvmsgv (g_transport, &dummy_msg, + 1, MSG_DONTWAIT); + + // No data should be returned. + zmq_assert (dummy_bytes == -1 && errno == EAGAIN); +} + +bool zmq::pgm_socket_t::tsi_equal (const pgm_tsi_t *tsi_a_, + const pgm_tsi_t *tsi_b_) +{ + // Compare 6B GSI. + const uint8_t *gsi_a = tsi_a_->gsi.identifier; + const uint8_t *gsi_b = tsi_b_->gsi.identifier; + + if (gsi_a [0] != gsi_b [0] || gsi_a [1] != gsi_b [1] || + gsi_a [2] != gsi_b [2] || gsi_a [3] != gsi_b [3] || + gsi_a [4] != gsi_b [4] || gsi_a [5] != gsi_b [5]) { + + return false; + } + + // Compare source port. + if (tsi_a_->sport != tsi_b_->sport) { + return false; + } + + return true; +} + +bool zmq::pgm_socket_t::tsi_empty (const pgm_tsi_t *tsi_) +{ + + uint8_t *gsi = (uint8_t*)tsi_->gsi.identifier; + + // GSI. + if (gsi [0] != 0 || gsi [1] != 0 || gsi [2] != 0 || + gsi [3] != 0 || gsi [4] != 0 || gsi [5] != 0) { + return false; + } + + // Source port. + if (tsi_->sport != 0) { + return false; + } + + return true; +} + +#endif + +#endif diff --git a/src/pgm_socket.hpp b/src/pgm_socket.hpp new file mode 100644 index 0000000..632288d --- /dev/null +++ b/src/pgm_socket.hpp @@ -0,0 +1,153 @@ +/* + Copyright (c) 2007-2009 FastMQ Inc. + + This file is part of 0MQ. + + 0MQ is free software; you can redistribute it and/or modify it under + the terms of the Lesser GNU 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 + Lesser GNU General Public License for more details. + + You should have received a copy of the Lesser GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#ifndef __PGM_SOCKET_HPP_INCLUDED__ +#define __PGM_SOCKET_HPP_INCLUDED__ + +#include "platform.hpp" + +#if defined ZMQ_HAVE_OPENPGM + +#ifdef ZMQ_HAVE_LINUX +#include <glib.h> +#include <pgm/pgm.h> +#else +#include <Winsock2.h> +#endif + +#include "stdint.hpp" +#include "options.hpp" + +namespace zmq +{ + // Encapsulates PGM socket. + class pgm_socket_t + { + +#ifdef ZMQ_HAVE_LINUX + + public: + // If receiver_ is true PGM transport is not generating SPM packets. + // interface format: iface;mcast_group:port for raw PGM socket + // udp:iface;mcast_goup:port for UDP encapsulacion + pgm_socket_t (bool receiver_, const options_t &options_); + + // Closes the transport. + ~pgm_socket_t (); + + // Initialize PGM network structures (GSI, GSRs). + int init (const char *network_); + + // Open PGM transport. Parameters are the same as in constructor. + int open_transport (void); + + // Close transport. + void close_transport (void); + + // Get receiver fds and store them into user allocated memory. + int get_receiver_fds (int *recv_fd_, int *waiting_pipe_fd_); + + // Get sender and receiver fds and store it to user allocated + // memory. Receive fd is used to process NAKs from peers. + int get_sender_fds (int *send_fd_, int *receive_fd_); + + // Send data as one APDU, transmit window owned memory. + size_t send (unsigned char *data_, size_t data_len_); + + // Allocates one slice for packet in tx window. + void *get_buffer (size_t *size_); + + // Fees memory allocated by get_buffer. + void free_buffer (void *data_); + + // Receive data from pgm socket. + ssize_t receive (void **data_); + + // POLLIN on sender side should mean NAK or SPMR receiving. + // process_upstream function is used to handle such a situation. + void process_upstream (void); + + protected: + + // OpenPGM transport + pgm_transport_t* g_transport; + + private: + + // Associated socket options. + options_t options; + + // Returns max tsdu size without fragmentation. + size_t get_max_tsdu_size (void); + + // Returns maximum count of apdus which fills readbuf_size_ + size_t get_max_apdu_at_once (size_t readbuf_size_); + + // Return true if TSI has empty GSI ('\0') and sport 0. + bool tsi_empty (const pgm_tsi_t *tsi_); + + // Compare TSIs, return true if equal. + bool tsi_equal (const pgm_tsi_t *tsi_a_, const pgm_tsi_t *tsi_b_); + + // true when pgm_socket should create receiving side. + bool receiver; + + // TIBCO Rendezvous format network info. + char network [256]; + + // PGM transport port number. + uint16_t port_number; + + // If we are using UDP encapsulation. + bool udp_encapsulation; + + // Array of pgm_msgv_t structures to store received data + // from the socket (pgm_transport_recvmsgv). + pgm_msgv_t *pgm_msgv; + + // How many bytes were read from pgm socket. + ssize_t nbytes_rec; + + // How many bytes were processed from last pgm socket read. + ssize_t nbytes_processed; + + // How many messages from pgm_msgv were already sent up. + ssize_t pgm_msgv_processed; + + // Size of pgm_msgv array. + ssize_t pgm_msgv_len; + + // Sender transport uses 2 fd. + enum {pgm_sender_fd_count = 2}; + + // Receiver transport uses 2 fd. + enum {pgm_receiver_fd_count = 2}; + + // TSI of the actual peer. + pgm_tsi_t tsi; + + // Previous peer TSI. + pgm_tsi_t retired_tsi; + +#endif + }; +} +#endif + +#endif diff --git a/src/socket_base.cpp b/src/socket_base.cpp index b4f7d6b..9412d67 100644 --- a/src/socket_base.cpp +++ b/src/socket_base.cpp @@ -17,6 +17,8 @@ along with this program. If not, see <http://www.gnu.org/licenses/>. */ +#include <iostream> + #include <string> #include <algorithm> @@ -35,9 +37,12 @@ #include "uuid.hpp" #include "pipe.hpp" #include "err.hpp" +#include "platform.hpp" +#include "pgm_sender.hpp" -zmq::socket_base_t::socket_base_t (app_thread_t *parent_) : +zmq::socket_base_t::socket_base_t (app_thread_t *parent_, int type_) : object_t (parent_), + type (type_), current (0), active (0), pending_term_acks (0), @@ -145,6 +150,22 @@ int zmq::socket_base_t::setsockopt (int option_, const void *optval_, errno = EFAULT; return -1; + case ZMQ_RATE: + if (optvallen_ != sizeof (uint32_t)) { + errno = EINVAL; + return -1; + } + options.rate = *((int32_t*) optval_); + return 0; + + case ZMQ_RECOVERY_IVL: + if (optvallen_ != sizeof (uint32_t)) { + errno = EINVAL; + return -1; + } + options.recovery_ivl = *((int32_t*) optval_); + return 0; + default: errno = EINVAL; return -1; @@ -170,6 +191,21 @@ int zmq::socket_base_t::connect (const char *addr_) std::string session_name ("#"); session_name += uuid_t ().to_string (); + // Parse addr_ string. + std::string addr_type; + std::string addr_args; + + std::string addr (addr_); + std::string::size_type pos = addr.find ("://"); + + if (pos == std::string::npos) { + errno = EINVAL; + return -1; + } + + addr_type = addr.substr (0, pos); + addr_args = addr.substr (pos + 3); + // Create the session. io_thread_t *io_thread = choose_io_thread (options.affinity); session_t *session = new session_t (io_thread, this, session_name.c_str (), @@ -198,20 +234,63 @@ int zmq::socket_base_t::connect (const char *addr_) send_plug (session); send_own (this, session); - // Create the connecter object. Supply it with the session name so that - // it can bind the new connection to the session once it is established. - zmq_connecter_t *connecter = new zmq_connecter_t ( - choose_io_thread (options.affinity), this, options, - session_name.c_str ()); - int rc = connecter->set_address (addr_); - if (rc != 0) { - delete connecter; - return -1; + if (addr_type == "tcp") { + + // Create the connecter object. Supply it with the session name so that + // it can bind the new connection to the session once it is established. + zmq_connecter_t *connecter = new zmq_connecter_t ( + choose_io_thread (options.affinity), this, options, + session_name.c_str ()); + int rc = connecter->set_address (addr_args.c_str ()); + if (rc != 0) { + delete connecter; + return -1; + } + send_plug (connecter); + send_own (this, connecter); + + return 0; } - send_plug (connecter); - send_own (this, connecter); - return 0; +#if defined ZMQ_HAVE_OPENPGM + if (addr_type == "pgm") { + + switch (type) { + case ZMQ_PUB: + { + pgm_sender_t *pgm_sender = + new pgm_sender_t (choose_io_thread (options.affinity), options, + session_name.c_str ()); + + int rc = pgm_sender->init (addr_args.c_str ()); + if (rc != 0) { + delete pgm_sender; + return -1; + } + + // Reserve a sequence number for following 'attach' command. + session->inc_seqnum (); + send_attach (session, pgm_sender); + + pgm_sender = NULL; + + break; + } + case ZMQ_SUB: + zmq_assert (false); + break; + default: + errno = EINVAL; + return -1; + } + + return 0; + } +#endif + + // Unknown address type. + errno = ENOTSUP; + return -1; } int zmq::socket_base_t::send (::zmq_msg_t *msg_, int flags_) diff --git a/src/socket_base.hpp b/src/socket_base.hpp index 5711dbe..398cd32 100644 --- a/src/socket_base.hpp +++ b/src/socket_base.hpp @@ -38,7 +38,7 @@ namespace zmq { public: - socket_base_t (class app_thread_t *parent_); + socket_base_t (class app_thread_t *parent_, int type_); virtual ~socket_base_t (); // Interface for communication with the API layer. @@ -87,6 +87,9 @@ namespace zmq // fair queueing. bool fetch (struct zmq_msg_t *msg_); + // Type of the socket. + int type; + // List of all I/O objects owned by this socket. The socket is // responsible for deallocating them before it quits. typedef std::set <class owned_t*> io_objects_t; diff --git a/src/sub.cpp b/src/sub.cpp index 8c1ef9b..1503fe2 100644 --- a/src/sub.cpp +++ b/src/sub.cpp @@ -23,7 +23,7 @@ #include "err.hpp" zmq::sub_t::sub_t (class app_thread_t *parent_) : - socket_base_t (parent_), + socket_base_t (parent_, ZMQ_SUB), all_count (0) { } |