/* Hanoh Haim Cisco Systems, Inc. */ /* Copyright (c) 2015-2015 Cisco Systems, Inc. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. */ #include "bp_sim.h" #include "latency.h" #include "utl_json.h" #include "utl_yaml.h" #include "msg_manager.h" #include "trex_watchdog.h" #include #include #include #undef VALG #ifdef VALG #include #endif CPluginCallback * CPluginCallback::callback; uint32_t getDualPortId(uint32_t thread_id){ return ( thread_id % (CGlobalInfo::m_options.get_expected_dual_ports()) ); } CRteMemPool CGlobalInfo::m_mem_pool[MAX_SOCKETS_SUPPORTED]; uint32_t CGlobalInfo::m_nodes_pool_size = 10*1024; CParserOption CGlobalInfo::m_options; CGlobalMemory CGlobalInfo::m_memory_cfg; CPlatformSocketInfo CGlobalInfo::m_socket; void CGlobalMemory::Dump(FILE *fd){ fprintf(fd," Total Memory : \n"); const std::string * names =get_mbuf_names(); uint32_t c_size=64; uint32_t c_total=0; int i=0; for (i=0; iMBUF_9k) && (im_is_exists); reset(); return (true); } bool CPlatformSocketInfoConfig::init(){ /* iterate the sockets */ uint32_t num_threads=0; uint32_t num_dual_if = m_platform->m_dual_if.size(); if ( m_num_dual_if > num_dual_if ){ printf("ERROR number of dual if %d is higher than defined in configuration file %d\n", (int)m_num_dual_if, (int)num_dual_if); } int i; for (i=0; im_dual_if[i]; if ( lp->m_socket>=MAX_SOCKETS_SUPPORTED ){ printf("ERROR socket %d is bigger than max %d \n",lp->m_socket,MAX_SOCKETS_SUPPORTED); exit(1); } if (!m_sockets_enable[lp->m_socket] ) { m_sockets_enable[lp->m_socket]=true; m_sockets_enabled++; } m_socket_per_dual_if[i]=lp->m_socket; /* learn how many threads per dual-if */ if (i==0) { num_threads = lp->m_threads.size(); m_max_threads_per_dual_if = num_threads; }else{ if (lp->m_threads.size() != num_threads) { printf("ERROR, the number of threads per dual ports should be the same for all dual ports\n"); exit(1); } } int j; for (j=0; jm_threads[j]; if (phy_thread>MAX_THREADS_SUPPORTED) { printf("ERROR, physical thread id is %d higher than max %d \n",phy_thread,MAX_THREADS_SUPPORTED); exit(1); } if (virt_thread>MAX_THREADS_SUPPORTED) { printf("ERROR virtual thread id is %d higher than max %d \n",virt_thread,MAX_THREADS_SUPPORTED); exit(1); } if ( m_thread_phy_to_virtual[phy_thread] ){ printf("ERROR physical thread %d defined twice\n",phy_thread); exit(1); } m_thread_phy_to_virtual[phy_thread]=virt_thread; m_thread_virt_to_phy[virt_thread] =phy_thread; } } if ( m_thread_phy_to_virtual[m_platform->m_master_thread] ){ printf("ERROR physical master thread %d already defined \n",m_platform->m_master_thread); exit(1); } if ( m_thread_phy_to_virtual[m_platform->m_rx_thread] ){ printf("ERROR physical latency thread %d already defined \n",m_platform->m_rx_thread); exit(1); } if (m_max_threads_per_dual_if < m_threads_per_dual_if ) { printf("ERROR number of threads asked per dual if is %d lower than max %d \n", (int)m_threads_per_dual_if, (int)m_max_threads_per_dual_if); exit(1); } return (true); } void CPlatformSocketInfoConfig::dump(FILE *fd){ fprintf(fd," core_mask %llx \n",(unsigned long long)get_cores_mask()); fprintf(fd," sockets :"); int i; for (i=0; i> 1; i++) { m_socket_per_dual_if[i]=0; } m_num_dual_if=0; m_threads_per_dual_if=0; m_rx_is_enabled=false; m_max_threads_per_dual_if=0; } void CPlatformSocketInfoConfig::Delete(){ } bool CPlatformSocketInfoConfig::is_sockets_enable(socket_id_t socket){ assert(socket>1)]); } void CPlatformSocketInfoConfig::set_rx_thread_is_enabled(bool enable){ m_rx_is_enabled =enable; } void CPlatformSocketInfoConfig::set_number_of_dual_ports(uint8_t num_dual_ports){ m_num_dual_if = num_dual_ports; } void CPlatformSocketInfoConfig::set_number_of_threads_per_ports(uint8_t num_threads){ m_threads_per_dual_if =num_threads; } bool CPlatformSocketInfoConfig::sanity_check(){ return (init()); } /* return the core mask */ uint64_t CPlatformSocketInfoConfig::get_cores_mask(){ int i; uint64_t mask=0; for (i=0; i=64) { printf(" ERROR phy threads can't be higher than 64 \n"); exit(1); } mask |=(1<m_master_thread); assert(m_platform->m_master_thread<64); if (m_rx_is_enabled) { mask |=(1<m_rx_thread); assert(m_platform->m_rx_thread<64); } return (mask); } virtual_thread_id_t CPlatformSocketInfoConfig::thread_phy_to_virt(physical_thread_id_t phy_id){ return (m_thread_phy_to_virtual[phy_id]); } physical_thread_id_t CPlatformSocketInfoConfig::thread_virt_to_phy(virtual_thread_id_t virt_id){ return ( m_thread_virt_to_phy[virt_id]); } bool CPlatformSocketInfoConfig::thread_phy_is_master(physical_thread_id_t phy_id){ return (m_platform->m_master_thread==phy_id?true:false); } bool CPlatformSocketInfoConfig::thread_phy_is_rx(physical_thread_id_t phy_id){ return (m_platform->m_rx_thread == phy_id?true:false); } //////////////////////////////////////// bool CPlatformSocketInfo::Create(CPlatformCoresYamlInfo * platform){ if ( (platform) && (platform->m_is_exists) ) { CPlatformSocketInfoConfig * lp=new CPlatformSocketInfoConfig(); assert(lp); lp->Create(platform); m_obj= lp; }else{ m_obj= new CPlatformSocketInfoNoConfig(); } return(true); } void CPlatformSocketInfo::Delete(){ if ( m_obj ){ delete m_obj; m_obj=NULL; } } bool CPlatformSocketInfo::is_sockets_enable(socket_id_t socket){ return ( m_obj->is_sockets_enable(socket) ); } socket_id_t CPlatformSocketInfo::max_num_active_sockets(){ return ( m_obj->max_num_active_sockets() ); } socket_id_t CPlatformSocketInfo::port_to_socket(port_id_t port){ return ( m_obj->port_to_socket(port) ); } void CPlatformSocketInfo::set_rx_thread_is_enabled(bool enable){ m_obj->set_rx_thread_is_enabled(enable); } void CPlatformSocketInfo::set_number_of_dual_ports(uint8_t num_dual_ports){ m_obj->set_number_of_dual_ports(num_dual_ports); } void CPlatformSocketInfo::set_number_of_threads_per_ports(uint8_t num_threads){ m_obj->set_number_of_threads_per_ports(num_threads); } bool CPlatformSocketInfo::sanity_check(){ return ( m_obj->sanity_check()); } /* return the core mask */ uint64_t CPlatformSocketInfo::get_cores_mask(){ return ( m_obj->get_cores_mask()); } virtual_thread_id_t CPlatformSocketInfo::thread_phy_to_virt(physical_thread_id_t phy_id){ return ( m_obj->thread_phy_to_virt(phy_id)); } physical_thread_id_t CPlatformSocketInfo::thread_virt_to_phy(virtual_thread_id_t virt_id){ return ( m_obj->thread_virt_to_phy(virt_id)); } bool CPlatformSocketInfo::thread_phy_is_master(physical_thread_id_t phy_id){ return ( m_obj->thread_phy_is_master(phy_id)); } bool CPlatformSocketInfo::thread_phy_is_rx(physical_thread_id_t phy_id) { return ( m_obj->thread_phy_is_rx(phy_id)); } void CPlatformSocketInfo::dump(FILE *fd){ m_obj->dump(fd); } //////////////////////////////////////// void CRteMemPool::dump_in_case_of_error(FILE *fd){ fprintf(fd," ERROR,there is no enough memory in socket %d \n",m_pool_id); fprintf(fd," Try to enlarge the memory values in the configuration file -/etc/trex_cfg.yaml ,see manual for more detail \n"); dump(fd); } std::string CRteMemPool::add_to_json( std::string name, rte_mempool_t * pool, bool last){ uint32_t p_free = rte_mempool_count(pool); uint32_t p_size = pool->size; char buff[200]; sprintf(buff,"\"%s\":[%llu,%llu]",name.c_str(),(unsigned long long)p_free,(unsigned long long)p_size); std::string json = std::string(buff) + (last?std::string(""):std::string(",")); return (json); } std::string CRteMemPool::dump_as_json(uint8_t id,bool last){ char buff[200]; sprintf(buff,"\"socket-%d\":{", (int)id); std::string json=std::string(buff); json+=add_to_json("64b",m_small_mbuf_pool); json+=add_to_json("128b",m_mbuf_pool_128); json+=add_to_json("256b",m_mbuf_pool_256); json+=add_to_json("512b",m_mbuf_pool_512); json+=add_to_json("1024b",m_mbuf_pool_1024); json+=add_to_json("2048b",m_mbuf_pool_2048); json+=add_to_json("4096b",m_mbuf_pool_4096); json+=add_to_json("9kb",m_mbuf_pool_9k,true); json+="}" ; if (last==false) { json+="," ; } return (json); } void CRteMemPool::dump(FILE *fd){ #define DUMP_MBUF(a,b) { float p=(100.0*(float)rte_mempool_count(b)/(float)b->size); fprintf(fd," %-30s : %.2f %% %s \n",a,p,(p<5.0?"<-":"OK") ); } DUMP_MBUF("mbuf_64",m_small_mbuf_pool); DUMP_MBUF("mbuf_128",m_mbuf_pool_128); DUMP_MBUF("mbuf_256",m_mbuf_pool_256); DUMP_MBUF("mbuf_512",m_mbuf_pool_512); DUMP_MBUF("mbuf_1024",m_mbuf_pool_1024); DUMP_MBUF("mbuf_2048",m_mbuf_pool_2048); DUMP_MBUF("mbuf_4096",m_mbuf_pool_4096); DUMP_MBUF("mbuf_9k",m_mbuf_pool_9k); } //////////////////////////////////////// std::string CGlobalInfo::dump_pool_as_json(void){ std::string json="\"mbuf_stats\":{"; CPlatformSocketInfo * lpSocket =&m_socket; int last_socket=-1; /* calc the last socket */ int i; for (i=0; i<(int)MAX_SOCKETS_SUPPORTED; i++) { if (lpSocket->is_sockets_enable((socket_id_t)i)) { last_socket=i; } } for (i=0; i<(int)MAX_SOCKETS_SUPPORTED; i++) { if (lpSocket->is_sockets_enable((socket_id_t)i)) { json+=m_mem_pool[i].dump_as_json(i,last_socket==i?true:false); } } json+="},"; return json; } void CGlobalInfo::free_pools(){ CPlatformSocketInfo * lpSocket =&m_socket; CRteMemPool * lpmem; int i; for (i=0; i<(int)MAX_SOCKETS_SUPPORTED; i++) { if (lpSocket->is_sockets_enable((socket_id_t)i)) { lpmem= &m_mem_pool[i]; utl_rte_mempool_delete(lpmem->m_small_mbuf_pool); utl_rte_mempool_delete(lpmem->m_mbuf_pool_128); utl_rte_mempool_delete(lpmem->m_mbuf_pool_256); utl_rte_mempool_delete(lpmem->m_mbuf_pool_512); utl_rte_mempool_delete(lpmem->m_mbuf_pool_1024); utl_rte_mempool_delete(lpmem->m_mbuf_pool_2048); utl_rte_mempool_delete(lpmem->m_mbuf_pool_4096); utl_rte_mempool_delete(lpmem->m_mbuf_pool_9k); } utl_rte_mempool_delete(m_mem_pool[0].m_mbuf_global_nodes); } } void CGlobalInfo::init_pools(uint32_t rx_buffers){ /* this include the pkt from 64- */ CGlobalMemory * lp=&CGlobalInfo::m_memory_cfg; CPlatformSocketInfo * lpSocket =&m_socket; CRteMemPool * lpmem; int i; for (i=0; i<(int)MAX_SOCKETS_SUPPORTED; i++) { if (lpSocket->is_sockets_enable((socket_id_t)i)) { lpmem= &m_mem_pool[i]; lpmem->m_pool_id=i; /* this include the packet from 0-64 this is for small packets */ lpmem->m_small_mbuf_pool =utl_rte_mempool_create("small-pkt-const", lp->m_mbuf[MBUF_64], CONST_SMALL_MBUF_SIZE, 32,(i<<5)+ 2,i); assert(lpmem->m_small_mbuf_pool); lpmem->m_mbuf_pool_128=utl_rte_mempool_create("_128-pkt-const", lp->m_mbuf[MBUF_128], CONST_128_MBUF_SIZE, 32,(i<<5)+ 6,i); assert(lpmem->m_mbuf_pool_128); lpmem->m_mbuf_pool_256=utl_rte_mempool_create("_256-pkt-const", lp->m_mbuf[MBUF_256], CONST_256_MBUF_SIZE, 32,(i<<5)+ 3,i); assert(lpmem->m_mbuf_pool_256); lpmem->m_mbuf_pool_512=utl_rte_mempool_create("_512_-pkt-const", lp->m_mbuf[MBUF_512], CONST_512_MBUF_SIZE, 32,(i<<5)+ 4,i); assert(lpmem->m_mbuf_pool_512); lpmem->m_mbuf_pool_1024=utl_rte_mempool_create("_1024-pkt-const", lp->m_mbuf[MBUF_1024], CONST_1024_MBUF_SIZE, 32,(i<<5)+ 5,i); assert(lpmem->m_mbuf_pool_1024); lpmem->m_mbuf_pool_2048=utl_rte_mempool_create("_2048-pkt-const", lp->m_mbuf[MBUF_2048], CONST_2048_MBUF_SIZE, 32,(i<<5)+ 5,i); assert(lpmem->m_mbuf_pool_2048); lpmem->m_mbuf_pool_4096=utl_rte_mempool_create("_4096-pkt-const", lp->m_mbuf[MBUF_4096], CONST_4096_MBUF_SIZE, 32,(i<<5)+ 5,i); assert(lpmem->m_mbuf_pool_4096); lpmem->m_mbuf_pool_9k=utl_rte_mempool_create("_9k-pkt-const", lp->m_mbuf[MBUF_9k]+rx_buffers, CONST_9k_MBUF_SIZE, 32,(i<<5)+ 5,i); assert(lpmem->m_mbuf_pool_9k); } } /* global always from socket 0 */ m_mem_pool[0].m_mbuf_global_nodes = utl_rte_mempool_create_non_pkt("global-nodes", lp->m_mbuf[MBUF_GLOBAL_FLOWS], sizeof(CGenNode), 128, 0 , SOCKET_ID_ANY); assert(m_mem_pool[0].m_mbuf_global_nodes); } void CFlowYamlInfo::Dump(FILE *fd){ fprintf(fd,"name : %s \n",m_name.c_str()); fprintf(fd,"cps : %f \n",m_k_cps); fprintf(fd,"ipg : %f \n",m_ipg_sec); fprintf(fd,"rtt : %f \n",m_rtt_sec); fprintf(fd,"w : %d \n",m_w); fprintf(fd,"wlength : %d \n",m_wlength); fprintf(fd,"limit : %d \n",m_limit); fprintf(fd,"limit_was_set : %d \n",m_limit_was_set?1:0); fprintf(fd,"cap_mode : %d \n",m_cap_mode?1:0); fprintf(fd,"plugin_id : %d \n",m_plugin_id); fprintf(fd,"one_server : %d \n",m_one_app_server?1:0); fprintf(fd,"one_server_was_set : %d \n",m_one_app_server_was_set?1:0); if (m_dpPkt) { m_dpPkt->Dump(fd); } } void dump_mac_addr(FILE* fd,uint8_t *p){ int i; for (i=0; i<6; i++) { uint8_t a=p[i]; if (i==5) { fprintf(fd,"%02x",a); }else{ fprintf(fd,"%02x:",a); } } } static uint8_t human_tbl[]={ ' ', 'K', 'M', 'G', 'T' }; std::string double_to_human_str(double num, std::string units, human_kbyte_t etype){ double abs_num=num; if (num<0.0) { abs_num=-num; } int i=0; int max_cnt=sizeof(human_tbl)/sizeof(human_tbl[0]); double div =1.0; double f=1000.0; if (etype ==KBYE_1024){ f=1024.0; } while ((abs_num > f ) && (i < max_cnt - 1)){ abs_num/=f; div*=f; i++; } char buf [100]; sprintf(buf,"%10.2f %c%s",num/div,human_tbl[i],units.c_str()); std::string res(buf); return (res); } void CPreviewMode::Dump(FILE *fd){ fprintf(fd," flags : %x\n", m_flags); fprintf(fd," write_file : %d\n", getFileWrite()?1:0); fprintf(fd," verbose : %d\n", (int)getVMode() ); fprintf(fd," realtime : %d\n", (int)getRealTime() ); fprintf(fd," flip : %d\n", (int)getClientServerFlip() ); fprintf(fd," cores : %d\n", (int)getCores() ); fprintf(fd," single core : %d\n", (int)getSingleCore() ); fprintf(fd," flow-flip : %d\n", (int)getClientServerFlowFlip() ); fprintf(fd," no clean close : %d\n", (int)getNoCleanFlowClose() ); fprintf(fd," 1g mode : %d\n", (int)get_1g_mode() ); fprintf(fd," zmq_publish : %d\n", (int)get_zmq_publish_enable() ); fprintf(fd," vlan_enable : %d\n", (int)get_vlan_mode_enable() ); fprintf(fd," mbuf_cache_disable : %d\n", (int)isMbufCacheDisabled() ); fprintf(fd," mac_ip_features : %d\n", (int)get_mac_ip_features_enable()?1:0 ); fprintf(fd," mac_ip_map : %d\n", (int)get_mac_ip_mapping_enable()?1:0 ); fprintf(fd," vm mode : %d\n", (int)get_vm_one_queue_enable()?1:0 ); } void CFlowGenStats::clear(){ m_nat_lookup_no_flow_id=0; m_total_bytes=0; m_total_pkt=0; m_total_open_flows =0; m_total_close_flows =0; m_nat_lookup_no_flow_id=0; m_nat_lookup_remove_flow_id=0; m_nat_lookup_add_flow_id=0; m_nat_flow_timeout=0; m_nat_flow_learn_error=0; } void CFlowGenStats::dump(FILE *fd){ std::string s_bytes=double_to_human_str((double )(m_total_bytes), "bytes", KBYE_1024); std::string s_pkt=double_to_human_str((double )(m_total_pkt), "pkt", KBYE_1000); std::string s_flows=double_to_human_str((double )(m_total_open_flows), "flows", KBYE_1000); DP_S(m_total_bytes,s_bytes); DP_S(m_total_pkt,s_pkt); DP_S(m_total_open_flows,s_flows); DP(m_total_pkt); DP(m_total_open_flows); DP(m_total_close_flows); DP_name("active",(m_total_open_flows-m_total_close_flows)); DP(m_total_bytes); DP(m_nat_lookup_no_flow_id); DP(m_nat_lookup_no_flow_id); DP(m_nat_lookup_remove_flow_id); DP(m_nat_lookup_add_flow_id); DP(m_nat_flow_timeout); DP_name("active_nat",(m_nat_lookup_add_flow_id-m_nat_lookup_remove_flow_id)); DP(m_nat_flow_learn_error); } int CErfIF::open_file(std::string file_name){ BP_ASSERT(m_writer==0); if ( m_preview_mode->getFileWrite() ){ capture_type_e file_type=ERF; if ( m_preview_mode->get_pcap_mode_enable() ){ file_type=LIBPCAP; } m_writer = CCapWriterFactory::CreateWriter(file_type,(char *)file_name.c_str()); if (m_writer == NULL) { fprintf(stderr,"ERROR can't create cap file %s ",(char *)file_name.c_str()); return (-1); } } m_raw = new CCapPktRaw(); return (0); } int CErfIF::write_pkt(CCapPktRaw *pkt_raw){ BP_ASSERT(m_writer); if ( m_preview_mode->getFileWrite() ){ BP_ASSERT(m_writer); bool res=m_writer->write_packet(pkt_raw); if (res != true) { fprintf(stderr,"ERROR can't write to cap file"); return (-1); } } return (0); } int CErfIF::close_file(void){ if (m_raw) { delete m_raw; m_raw = NULL; } if ( m_preview_mode->getFileWrite() ){ if (m_writer) { delete m_writer; m_writer = NULL; } } return (0); } void CFlowKey::Clean(){ m_ipaddr1=0; m_ipaddr2=0; m_port1=0; m_port2=0; m_ip_proto=0; m_l2_proto=0; m_vrfid=0; } void CFlowKey::Dump(FILE *fd){ fprintf(fd," %x:%x:%x:%x:%x:%x:%x\n",m_ipaddr1,m_ipaddr2,m_port1,m_port2,m_ip_proto,m_l2_proto,m_vrfid); } void CPacketDescriptor::Dump(FILE *fd){ fprintf(fd," IsSwapTuple : %d \n",IsSwapTuple()?1:0); fprintf(fd," IsSInitDir : %d \n",IsInitSide()?1:0); fprintf(fd," Isvalid : %d ",IsValidPkt()?1:0); fprintf(fd," IsRtt : %d ",IsRtt()?1:0); fprintf(fd," IsLearn : %d ",IsLearn()?1:0); if (IsTcp() ) { fprintf(fd," TCP "); }else{ fprintf(fd," UDP "); } fprintf(fd," IsLast Pkt : %d ", IsLastPkt() ?1:0); fprintf(fd," id : %d \n",getId() ); fprintf(fd," flow_ID : %d , max_pkts : %u, max_aging: %d sec , pkt_id : %u, init: %d ( dir:%d dir_max :%d ) bid:%d \n",getFlowId(), GetMaxPktsPerFlow(), GetMaxFlowTimeout() , getFlowPktNum(), IsInitSide(), GetDirInfo()->GetPktNum(), GetDirInfo()->GetMaxPkts(), IsBiDirectionalFlow()?1:0 ); fprintf(fd,"\n"); } void CPacketIndication::UpdateOffsets(){ m_ether_offset = getEtherOffset(); m_ip_offset = getIpOffset(); m_udp_tcp_offset = getTcpOffset(); m_payload_offset = getPayloadOffset(); } void CPacketIndication::UpdatePacketPadding(){ m_packet_padding = m_packet->getTotalLen() - (l3.m_ipv4->getTotalLength()+ getIpOffset()); } void CPacketIndication::RefreshPointers(){ char *pobase=getBasePtr(); m_ether = (EthernetHeader *) (pobase + m_ether_offset); l3.m_ipv4 = (IPHeader *) (pobase + m_ip_offset); l4.m_tcp= (TCPHeader *)(pobase + m_udp_tcp_offset); if ( m_payload_offset ){ m_payload =(uint8_t *)(pobase + m_payload_offset); }else{ m_payload =(uint8_t *)(0); } } // copy ref assume pkt point to a fresh void CPacketIndication::Clone(CPacketIndication * obj,CCapPktRaw * pkt){ Clean(); m_cap_ipg = obj->m_cap_ipg; m_packet = pkt; char *pobase=getBasePtr(); m_flow = obj->m_flow; m_ether = (EthernetHeader *) (pobase + obj->getEtherOffset()); l3.m_ipv4 = (IPHeader *) (pobase + obj->getIpOffset()); m_is_ipv6 = obj->m_is_ipv6; l4.m_tcp= (TCPHeader *)(pobase + obj->getTcpOffset()); if ( obj->getPayloadOffset() ){ m_payload =(uint8_t *)(pobase + obj->getPayloadOffset()); }else{ m_payload =(uint8_t *)(0); } m_payload_len = obj->m_payload_len; m_flow_key = obj->m_flow_key; m_desc = obj->m_desc; m_packet_padding = obj->m_packet_padding; /* copy offsets*/ m_ether_offset = obj->m_ether_offset; m_ip_offset = obj->m_ip_offset; m_udp_tcp_offset = obj->m_udp_tcp_offset;; m_payload_offset = obj->m_payload_offset; } void CPacketIndication::Dump(FILE *fd,int verbose){ fprintf(fd," ipg : %f \n",m_cap_ipg); fprintf(fd," key \n"); fprintf(fd," ------\n"); m_flow_key.Dump(fd); fprintf(fd," L2 info \n"); fprintf(fd," ------\n"); m_packet->Dump(fd,verbose); fprintf(fd," Descriptor \n"); fprintf(fd," ------\n"); m_desc.Dump(fd); if ( m_desc.IsValidPkt() ) { fprintf(fd," ipv4 \n"); l3.m_ipv4->dump(fd); if ( m_desc.IsUdp() ) { l4.m_udp->dump(fd); }else{ l4.m_tcp->dump(fd); } fprintf(fd," payload len : %d \n",m_payload_len); }else{ fprintf(fd," not valid packet \n"); } } void CPacketIndication::Clean(){ m_desc.Clear(); m_ether=0; l3.m_ipv4=0; l4.m_tcp=0; m_payload=0; m_payload_len=0; } uint64_t CCPacketParserCounters::getTotalErrors(){ uint64_t res= m_non_ip+ m_arp+ m_mpls+ m_non_valid_ipv4_ver+ m_ip_checksum_error+ m_ip_length_error+ m_ip_not_first_fragment_error+ m_ip_ttl_is_zero_error+ m_ip_multicast_error+ m_non_tcp_udp_ah+ m_non_tcp_udp_esp+ m_non_tcp_udp_icmp+ m_non_tcp_udp_gre+ m_non_tcp_udp_ip+ m_tcp_udp_pkt_length_error; return (res); } void CCPacketParserCounters::Clear(){ m_pkt=0; m_non_ip=0; m_vlan=0; m_arp=0; m_mpls=0; m_non_valid_ipv4_ver=0; m_ip_checksum_error=0; m_ip_length_error=0; m_ip_not_first_fragment_error=0; m_ip_ttl_is_zero_error=0; m_ip_multicast_error=0; m_ip_header_options=0; m_non_tcp_udp=0; m_non_tcp_udp_ah=0; m_non_tcp_udp_esp=0; m_non_tcp_udp_icmp=0; m_non_tcp_udp_gre=0; m_non_tcp_udp_ip=0; m_tcp_header_options=0; m_tcp_udp_pkt_length_error=0; m_tcp=0; m_udp=0; m_valid_udp_tcp=0; } void CCPacketParserCounters::Dump(FILE *fd){ DP (m_pkt); DP (m_non_ip); DP (m_vlan); DP (m_arp); DP (m_mpls); DP (m_non_valid_ipv4_ver); DP (m_ip_checksum_error); DP (m_ip_length_error); DP (m_ip_not_first_fragment_error); DP (m_ip_ttl_is_zero_error); DP (m_ip_multicast_error); DP (m_ip_header_options); DP (m_non_tcp_udp); DP (m_non_tcp_udp_ah); DP (m_non_tcp_udp_esp); DP (m_non_tcp_udp_icmp); DP (m_non_tcp_udp_gre); DP (m_non_tcp_udp_ip); DP (m_tcp_header_options); DP (m_tcp_udp_pkt_length_error); DP (m_tcp); DP (m_udp); DP (m_valid_udp_tcp); } bool CPacketParser::Create(){ m_counter.Clear(); return (true); } void CPacketParser::Delete(){ } bool CPacketParser::ProcessPacket(CPacketIndication * pkt_indication, CCapPktRaw * raw_packet){ BP_ASSERT(pkt_indication); pkt_indication->ProcessPacket(this,raw_packet); if (pkt_indication->m_desc.IsValidPkt()) { return (true); } return (false); } void CPacketParser::Dump(FILE *fd){ fprintf(fd," parser statistic \n"); fprintf(fd," ===================== \n"); m_counter.Dump(fd); } void CPacketIndication::SetKey(void){ uint32_t ip_src, ip_dst; m_desc.SetIsValidPkt(true); if (is_ipv6()){ uint16_t ipv6_src[8]; uint16_t ipv6_dst[8]; l3.m_ipv6->getSourceIpv6(&ipv6_src[0]); l3.m_ipv6->getDestIpv6(&ipv6_dst[0]); ip_src=(ipv6_src[6] << 16) | ipv6_src[7]; ip_dst=(ipv6_dst[6] << 16) | ipv6_dst[7]; m_flow_key.m_ip_proto = l3.m_ipv6->getNextHdr(); }else{ ip_src=l3.m_ipv4->getSourceIp(); ip_dst=l3.m_ipv4->getDestIp(); m_flow_key.m_ip_proto = l3.m_ipv4->getProtocol(); } /* UDP/TCP has same place */ uint16_t src_port = l4.m_udp->getSourcePort(); uint16_t dst_port = l4.m_udp->getDestPort(); if (ip_src > ip_dst ) { m_flow_key.m_ipaddr1 =ip_dst; m_flow_key.m_ipaddr2 =ip_src; m_flow_key.m_port1 = dst_port; m_flow_key.m_port2 = src_port; }else{ m_desc.SetSwapTuple(true); m_flow_key.m_ipaddr1 = ip_src; m_flow_key.m_ipaddr2 = ip_dst; m_flow_key.m_port1 = src_port; m_flow_key.m_port2 = dst_port; } m_flow_key.m_l2_proto = 0; m_flow_key.m_vrfid = 0; } uint8_t CPacketIndication::ProcessIpPacketProtocol(CCPacketParserCounters *m_cnt, uint8_t protocol, int *offset){ char * packetBase = m_packet->raw; TCPHeader * tcp=0; UDPHeader * udp=0; uint16_t tcp_header_len=0; switch (protocol) { case IPHeader::Protocol::TCP : m_desc.SetIsTcp(true); tcp =(TCPHeader *)(packetBase +*offset); l4.m_tcp = tcp; tcp_header_len = tcp->getHeaderLength(); if ( tcp_header_len > (5*4) ){ // we have ip option m_cnt->m_tcp_header_options++; } *offset += tcp_header_len; m_cnt->m_tcp++; break; case IPHeader::Protocol::UDP : m_desc.SetIsUdp(true); udp =(UDPHeader *)(packetBase +*offset); l4.m_udp = udp; *offset += 8; m_cnt->m_udp++; break; case IPHeader::Protocol::AH: m_cnt->m_non_tcp_udp_ah++; return (1); break; case IPHeader::Protocol::ESP: m_cnt->m_non_tcp_udp_esp++; return (1); break; case IPHeader::Protocol::ICMP: case IPHeader::Protocol::IPV6_ICMP: m_cnt->m_non_tcp_udp_icmp++; return (1); break; case IPHeader::Protocol::GRE: m_cnt->m_non_tcp_udp_gre++; return (1); break; case IPHeader::Protocol::IP: m_cnt->m_non_ip++; return (1); break; default: m_cnt->m_non_tcp_udp++; return (1); break; } /* out of packet */ if ( *offset > m_packet->getTotalLen() ) { m_cnt->m_tcp_udp_pkt_length_error++; return (1); } return (0); } void CPacketIndication::ProcessIpPacket(CPacketParser *parser, int offset){ char * packetBase; CCPacketParserCounters * m_cnt=&parser->m_counter; packetBase = m_packet->raw; uint8_t protocol; BP_ASSERT(l3.m_ipv4); parser->m_counter.m_pkt++; if ( l3.m_ipv4->getVersion() == 4 ){ m_cnt->m_ipv4++; }else{ m_cnt->m_non_valid_ipv4_ver++; return; } // check the IP Length if( (uint32_t)(l3.m_ipv4->getTotalLength()+offset) > (uint32_t)( m_packet->getTotalLen()) ){ m_cnt->m_ip_length_error++; return; } uint16_t ip_offset=offset; uint16_t ip_header_length = l3.m_ipv4->getHeaderLength(); if ( ip_header_length >(5*4) ){ m_cnt->m_ip_header_options++; } if ( (uint32_t)(ip_header_length + offset) > (uint32_t)m_packet->getTotalLen() ) { m_cnt->m_ip_length_error++; return; } offset += ip_header_length; if( l3.m_ipv4->getTimeToLive() ==0 ){ m_cnt->m_ip_ttl_is_zero_error++; return; } if( l3.m_ipv4->isNotFirstFragment() ) { m_cnt->m_ip_not_first_fragment_error++; return; } protocol = l3.m_ipv4->getProtocol(); if (ProcessIpPacketProtocol(m_cnt,protocol,&offset) != 0) { return; }; uint16_t payload_offset_from_ip = offset-ip_offset; if ( payload_offset_from_ip > l3.m_ipv4->getTotalLength() ) { m_cnt->m_tcp_udp_pkt_length_error++; return; } if ( m_packet->pkt_len > MAX_PKT_SIZE ){ m_cnt->m_tcp_udp_pkt_length_error++; printf("ERROR packet is too big, not supported jumbo packets that larger than %d \n",MAX_PKT_SIZE); return; } // Set packet length and include padding if needed m_packet->pkt_len = l3.m_ipv4->getTotalLength() + getIpOffset(); if (m_packet->pkt_len < 60) { m_packet->pkt_len = 60; } m_cnt->m_valid_udp_tcp++; m_payload_len = l3.m_ipv4->getTotalLength() - (payload_offset_from_ip); m_payload = (uint8_t *)(packetBase +offset); UpdatePacketPadding(); SetKey(); } void CPacketIndication::ProcessIpv6Packet(CPacketParser *parser, int offset){ char * packetBase = m_packet->raw; CCPacketParserCounters * m_cnt=&parser->m_counter; uint16_t src_ipv6[6]; uint16_t dst_ipv6[6]; uint16_t idx; uint8_t protocol; BP_ASSERT(l3.m_ipv6); parser->m_counter.m_pkt++; if ( l3.m_ipv6->getVersion() == 6 ){ m_cnt->m_ipv6++; }else{ m_cnt->m_non_valid_ipv6_ver++; return; } // Check length if ((uint32_t)(l3.m_ipv6->getPayloadLen()+offset+l3.m_ipv6->getHeaderLength()) > (uint32_t)( m_packet->getTotalLen()) ){ m_cnt->m_ipv6_length_error++; return; } for (idx=0; idx<6; idx++){ src_ipv6[idx] = CGlobalInfo::m_options.m_src_ipv6[idx]; dst_ipv6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; } l3.m_ipv6->updateMSBIpv6Src(&src_ipv6[0]); l3.m_ipv6->updateMSBIpv6Dst(&dst_ipv6[0]); offset += l3.m_ipv6->getHeaderLength(); protocol = l3.m_ipv6->getNextHdr(); if (ProcessIpPacketProtocol(m_cnt,protocol,&offset) != 0) { return; }; // Set packet length and include padding if needed uint16_t real_pkt_size = l3.m_ipv6->getPayloadLen()+ getIpOffset() + l3.m_ipv6->getHeaderLength(); m_packet->pkt_len = real_pkt_size; if (m_packet->pkt_len < 60) { m_packet->pkt_len = 60; } m_cnt->m_valid_udp_tcp++; m_payload_len = l3.m_ipv6->getPayloadLen(); m_payload = (uint8_t *)(packetBase +offset); m_packet_padding = m_packet->getTotalLen() - real_pkt_size; assert( m_packet->getTotalLen()>= real_pkt_size ); SetKey(); } static uint8_t cbuff[MAX_PKT_SIZE]; bool CPacketIndication::ConvertPacketToIpv6InPlace(CCapPktRaw * pkt, int offset){ // Copy l2 data and set l2 type to ipv6 memcpy(cbuff, pkt->raw, offset); *(uint16_t*)(cbuff+12) = PKT_HTONS(EthernetHeader::Protocol::IPv6); // Create the ipv6 header IPHeader *ipv4 = (IPHeader *) (pkt->raw+offset); IPv6Header *ipv6 = (IPv6Header *) (cbuff+offset); uint8_t ipv6_hdrlen = ipv6->getHeaderLength(); memset(ipv6,0,ipv6_hdrlen); ipv6->setVersion(6); if (ipv4->getTotalLength() < ipv4->getHeaderLength()) { return(false); } // Calculate the payload length uint16_t p_len = ipv4->getTotalLength() - ipv4->getHeaderLength(); ipv6->setPayloadLen(p_len); uint8_t l4_proto = ipv4->getProtocol(); ipv6->setNextHdr(l4_proto); ipv6->setHopLimit(64); // Update the least signficant 32-bits of ipv6 address // using the ipv4 address ipv6->updateLSBIpv6Src(ipv4->getSourceIp()); ipv6->updateLSBIpv6Dst(ipv4->getDestIp()); // Copy rest of packet uint16_t ipv4_offset = offset + ipv4->getHeaderLength(); uint16_t ipv6_offset = offset + ipv6_hdrlen; memcpy(cbuff+ipv6_offset,pkt->raw+ipv4_offset,p_len); ipv6_offset+=p_len; memcpy(pkt->raw,cbuff,ipv6_offset); // Set packet length pkt->pkt_len = ipv6_offset; m_is_ipv6 = true; return (true); } void CPacketIndication::ProcessPacket(CPacketParser *parser, CCapPktRaw * pkt){ _ProcessPacket(parser,pkt); if ( m_desc.IsValidPkt() ){ UpdateOffsets(); /* update fast offsets */ } } /* process packet */ void CPacketIndication::_ProcessPacket(CPacketParser *parser, CCapPktRaw * pkt){ BP_ASSERT(pkt); m_packet =pkt; Clean(); CCPacketParserCounters * m_cnt=&parser->m_counter; int offset = 0; char * packetBase; packetBase = m_packet->raw; BP_ASSERT(packetBase); m_ether = (EthernetHeader *)packetBase; m_is_ipv6 = false; // IP switch( m_ether->getNextProtocol() ) { case EthernetHeader::Protocol::IP : offset = 14; l3.m_ipv4 =(IPHeader *)(packetBase+offset); break; case EthernetHeader::Protocol::IPv6 : offset = 14; l3.m_ipv6 =(IPv6Header *)(packetBase+offset); m_is_ipv6 = true; break; case EthernetHeader::Protocol::VLAN : m_cnt->m_vlan++; switch ( m_ether->getVlanProtocol() ){ case EthernetHeader::Protocol::IP: offset = 18; l3.m_ipv4 =(IPHeader *)(packetBase+offset); break; case EthernetHeader::Protocol::IPv6 : offset = 18; l3.m_ipv6 =(IPv6Header *)(packetBase+offset); m_is_ipv6 = true; break; case EthernetHeader::Protocol::MPLS_Multicast : case EthernetHeader::Protocol::MPLS_Unicast : m_cnt->m_mpls++; return; case EthernetHeader::Protocol::ARP : m_cnt->m_arp++; return; default: m_cnt->m_non_ip++; return ; /* Non IP */ } break; case EthernetHeader::Protocol::ARP : m_cnt->m_arp++; return; /* Non IP */ break; case EthernetHeader::Protocol::MPLS_Multicast : case EthernetHeader::Protocol::MPLS_Unicast : m_cnt->m_mpls++; return; /* Non IP */ break; default: m_cnt->m_non_ip++; return; /* Non IP */ } if (is_ipv6() == false) { if( (14+20) > (uint32_t)( m_packet->getTotalLen()) ){ m_cnt->m_ip_length_error++; return; } } // For now, we can not mix ipv4 and ipv4 packets // so we require --ipv6 option be set for ipv6 packets if ((m_is_ipv6) && (CGlobalInfo::is_ipv6_enable() == false)){ fprintf(stderr,"ERROR --ipv6 must be set to process ipv6 packets\n"); exit(-1); } // Convert to Ipv6 if requested and not already Ipv6 if ((CGlobalInfo::is_ipv6_enable()) && (is_ipv6() == false )) { if (ConvertPacketToIpv6InPlace(pkt, offset) == false){ /* Move to next packet as this was erroneous */ printf(" unable to convert packet to IPv6, skipping...\n"); return; } } if (is_ipv6()){ ProcessIpv6Packet(parser,offset); }else{ ProcessIpPacket(parser,offset); } } void CFlowTableStats::Clear(){ m_lookup=0; m_found=0; m_fif=0; m_add=0; m_remove=0; m_fif_err=0; m_active=0; } void CFlowTableStats::Dump(FILE *fd){ DP (m_lookup); DP (m_found); DP (m_fif); DP (m_add); DP (m_remove); DP (m_fif_err); DP (m_active); } void CFlow::Dump(FILE *fd){ fprintf(fd," fif is swap : %d \n",is_fif_swap); } void CFlowTableManagerBase::Dump(FILE *fd){ m_stats.Dump(fd); } // Return flow that has given key. If flow does not exist, create one, and add to CFlow data structure. // key - key to lookup by. // is_fif - return: true if flow did not exist (This is the first packet we see in this flow). // false if flow already existed CFlow * CFlowTableManagerBase::process(const CFlowKey & key, bool & is_fif) { m_stats.m_lookup++; is_fif=false; CFlow * lp=lookup(key); if ( lp ) { m_stats.m_found++; return (lp); }else{ m_stats.m_fif++; m_stats.m_active++; m_stats.m_add++; is_fif=true; lp= add(key ); if (lp) { }else{ m_stats.m_fif_err++; } } return (lp); } bool CFlowTableMap::Create(int max_size){ m_stats.Clear(); return (true); } void CFlowTableMap::Delete(){ remove_all(); } void CFlowTableMap::remove(const CFlowKey & key ) { CFlow *lp=lookup(key); if ( lp ) { delete lp; m_stats.m_remove++; m_stats.m_active--; m_map.erase(key); }else{ BP_ASSERT(0); } } CFlow * CFlowTableMap::lookup(const CFlowKey & key ) { flow_map_t::iterator iter; iter = m_map.find(key); if (iter != m_map.end() ) { return ( (*iter).second ); }else{ return (( CFlow*)0); } } CFlow * CFlowTableMap::add(const CFlowKey & key ) { CFlow * flow = new CFlow(); m_map.insert(flow_map_t::value_type(key,flow)); return (flow); } void CFlowTableMap::remove_all(){ if ( m_map.empty() ) return; flow_map_iter_t it; for (it= m_map.begin(); it != m_map.end(); ++it) { CFlow *lp = it->second; delete lp; } m_map.clear(); } uint64_t CFlowTableMap::count(){ return ( m_map.size()); } /* * This function will insert an IP option header containing metadata for the * rx-check feature. * * An mbuf is created to hold the new option header plus the portion of the * packet after the base IP header (includes any IP options header that might * exist). This mbuf is then linked into the existing mbufs (becoming the * second mbuf). * * Note that the rxcheck option header is inserted as the first option header, * and any existing IP option headers are placed after it. */ void CFlowPktInfo::do_generate_new_mbuf_rxcheck(rte_mbuf_t * m, CGenNode * node, bool single_port){ /* retrieve size of rx-check header, must be multiple of 8 */ uint16_t opt_len = RX_CHECK_LEN; uint16_t current_opt_len = 0; assert( (opt_len % 8) == 0 ); /* determine starting move location */ char *mp1 = rte_pktmbuf_mtod(m, char*); uint16_t mp1_offset = m_pkt_indication.getFastIpOffsetFast(); if (unlikely (m_pkt_indication.is_ipv6()) ) { mp1_offset += IPv6Header::DefaultSize; }else{ mp1_offset += IPHeader::DefaultSize; } char *move_from = mp1 + mp1_offset; /* determine size of new mbuf required */ uint16_t move_len = m->data_len - mp1_offset; uint16_t new_mbuf_size = move_len + opt_len; uint16_t mp2_offset = opt_len; /* obtain a new mbuf */ rte_mbuf_t * new_mbuf = CGlobalInfo::pktmbuf_alloc(node->get_socket_id(), new_mbuf_size); assert(new_mbuf); char * mp2 = rte_pktmbuf_append(new_mbuf, new_mbuf_size); char * move_to = mp2 + mp2_offset; /* move part of packet from first mbuf to new mbuf */ memmove(move_to, move_from, move_len); /* trim first mbuf and set pointer to option header*/ CRx_check_header *rxhdr; uint16_t buf_adjust = move_len; rxhdr = (CRx_check_header *)mp2; m->data_len -= buf_adjust; /* insert rx-check data as an IPv4 option header or IPv6 extension header*/ CFlowPktInfo * lp=node->m_pkt_info; CPacketDescriptor * desc=&lp->m_pkt_indication.m_desc; /* set option type and update ip header length */ IPHeader * ipv4=(IPHeader *)(mp1 + 14); if (unlikely (m_pkt_indication.is_ipv6()) ) { IPv6Header * ipv6=(IPv6Header *)(mp1 + 14); uint8_t save_header= ipv6->getNextHdr(); ipv6->setNextHdr(RX_CHECK_V6_OPT_TYPE); ipv6->setHopLimit(TTL_RESERVE_DUPLICATE); ipv6->setPayloadLen( ipv6->getPayloadLen() + sizeof(CRx_check_header)); rxhdr->m_option_type = save_header; rxhdr->m_option_len = RX_CHECK_V6_OPT_LEN; }else{ current_opt_len = ipv4->getHeaderLength(); ipv4->setHeaderLength(current_opt_len+opt_len); ipv4->setTotalLength(ipv4->getTotalLength()+opt_len); ipv4->setTimeToLive(TTL_RESERVE_DUPLICATE); rxhdr->m_option_type = RX_CHECK_V4_OPT_TYPE; rxhdr->m_option_len = RX_CHECK_V4_OPT_LEN; } /* fill in the rx-check metadata in the options header */ if ( CGlobalInfo::m_options.is_rxcheck_const_ts() ){ /* Runtime flag to use a constant value for the timestamp field. */ /* This is used by simulation to provide consistency across runs. */ rxhdr->m_time_stamp = 0xB3B2B1B0; }else{ rxhdr->m_time_stamp = os_get_hr_tick_32(); } rxhdr->m_magic = RX_CHECK_MAGIC; rxhdr->m_flow_id = node->m_flow_id | ( ( (uint64_t)(desc->getFlowId() & 0xf))<<52 ) ; // include thread_id, node->flow_id, sub_flow in case of multi-flow template rxhdr->m_flags = 0; rxhdr->m_aging_sec = desc->GetMaxFlowTimeout(); rxhdr->m_template_id = (uint8_t)desc->getId(); /* add the flow packets goes to the same port */ if (single_port) { rxhdr->m_pkt_id = desc->getFlowPktNum(); rxhdr->m_flow_size = desc->GetMaxPktsPerFlow(); }else{ rxhdr->m_pkt_id = desc->GetDirInfo()->GetPktNum(); rxhdr->m_flow_size = desc->GetDirInfo()->GetMaxPkts(); /* set dir */ rxhdr->set_dir(desc->IsInitSide()?1:0); rxhdr->set_both_dir(desc->IsBiDirectionalFlow()?1:0); } /* update checksum for IPv4, split across 2 mbufs */ if (likely ( ! m_pkt_indication.is_ipv6()) ) { ipv4->updateCheckSum2((uint8_t *)ipv4, current_opt_len, (uint8_t *)rxhdr, opt_len); } /* link new mbuf */ new_mbuf->next = m->next; new_mbuf->nb_segs++; m->next = new_mbuf; m->nb_segs++; m->pkt_len += opt_len; } char * CFlowPktInfo::push_ipv4_option_offline(uint8_t bytes){ /* must be align by 4*/ assert( (bytes % 4)== 0 ); assert(m_pkt_indication.is_ipv6()==false); if ( m_pkt_indication.l3.m_ipv4->getHeaderLength()+bytes>60 ){ printf(" ERROR ipv4 options size is too big, should be able to add %d bytes for internal info \n",bytes); return((char *)0); } /* now we can do that !*/ /* add more bytes to the packet */ m_packet->append(bytes); uint8_t ip_offset_to_move= m_pkt_indication.getFastIpOffsetFast()+IPHeader::DefaultSize; char *p=m_packet->raw+ip_offset_to_move; uint16_t bytes_to_move= m_packet->pkt_len - ip_offset_to_move -bytes; /* move the start of ipv4 options */ memmove(p+bytes ,p, bytes_to_move); /* fix all other stuff */ if ( m_pkt_indication.m_udp_tcp_offset ){ m_pkt_indication.m_udp_tcp_offset+=bytes; } if ( m_pkt_indication.m_payload_offset ) { m_pkt_indication.m_payload_offset+=bytes; } m_pkt_indication.RefreshPointers(); /* now pointer are updated we can manipulate ipv4 header */ IPHeader * ipv4=m_pkt_indication.l3.m_ipv4; ipv4->setTotalLength(ipv4->getTotalLength()+bytes); ipv4->setHeaderLength(ipv4->getHeaderLength()+(bytes)); m_pkt_indication.UpdatePacketPadding(); /* refresh the global mbuf */ free_const_mbuf(); alloc_const_mbuf(); return (p); } void CFlowPktInfo::mask_as_learn(){ CNatOption *lpNat; if ( m_pkt_indication.is_ipv6() ) { lpNat=(CNatOption *)push_ipv6_option_offline(CNatOption::noOPTION_LEN); lpNat->set_init_ipv6_header(); lpNat->set_fid(0); lpNat->set_thread_id(0); } else { if (CGlobalInfo::is_learn_mode(CParserOption::LEARN_MODE_IP_OPTION)) { // Make space in IPv4 header for NAT option lpNat=(CNatOption *)push_ipv4_option_offline(CNatOption::noOPTION_LEN); lpNat->set_init_ipv4_header(); lpNat->set_fid(0); lpNat->set_thread_id(0); m_pkt_indication.l3.m_ipv4->updateCheckSum(); } /* learn is true */ m_pkt_indication.m_desc.SetLearn(true); } } char * CFlowPktInfo::push_ipv6_option_offline(uint8_t bytes){ /* must be align by 8*/ assert( (bytes % 8)== 0 ); assert(m_pkt_indication.is_ipv6()==true); /* add more bytes to the packet */ m_packet->append(bytes); uint8_t ip_offset_to_move= m_pkt_indication.getFastIpOffsetFast()+IPv6Header::DefaultSize; char *p=m_packet->raw+ip_offset_to_move; uint16_t bytes_to_move= m_packet->pkt_len - ip_offset_to_move -bytes; /* move the start of ipv4 options */ memmove(p+bytes ,p, bytes_to_move); /* fix all other stuff */ if ( m_pkt_indication.m_udp_tcp_offset ){ m_pkt_indication.m_udp_tcp_offset+=bytes; } if ( m_pkt_indication.m_payload_offset ) { m_pkt_indication.m_payload_offset+=bytes; } m_pkt_indication.RefreshPointers(); /* now pointer are updated we can manipulate ipv6 header */ IPv6Header * ipv6=m_pkt_indication.l3.m_ipv6; ipv6->setPayloadLen(ipv6->getPayloadLen()+bytes); uint8_t save_header= ipv6->getNextHdr(); *p=save_header; /* copy next header */ ipv6->setNextHdr(CNatOption::noIPV6_OPTION); m_pkt_indication.UpdatePacketPadding(); /* refresh the global mbuf */ free_const_mbuf(); alloc_const_mbuf(); return (p); } void CFlowPktInfo::alloc_const_mbuf(){ if ( m_packet->pkt_len > FIRST_PKT_SIZE ) { /* pkt size is bigger than FIRST_PKT_SIZE let's create an offline buffer */ int i; for (i=0; ipkt_len - FIRST_PKT_SIZE); m = CGlobalInfo::pktmbuf_alloc(i,pkt_s); BP_ASSERT(m); char *p=rte_pktmbuf_append(m, pkt_s); rte_memcpy(p,(m_packet->raw+FIRST_PKT_SIZE),pkt_s); assert(m_big_mbuf[i]==NULL); m_big_mbuf[i]=m; } } } } void CFlowPktInfo::free_const_mbuf(){ int i; for (i=0; im_packet); /* clone of the offsets */ m_pkt_indication.Clone(pkt_ind,m_packet); int i; for (i=0; iwrite_packet(lp->m_packet); BP_ASSERT(res); } delete lpWriter; } struct CTmpFlowPerDirInfo { CTmpFlowPerDirInfo(){ m_pkt_id=0; } uint16_t m_pkt_id; }; class CTmpFlowInfo { public: CTmpFlowInfo(){ m_max_pkts=0; m_max_aging_sec=0.0; m_last_pkt=0.0; } ~CTmpFlowInfo(){ } public: uint32_t m_max_pkts; dsec_t m_max_aging_sec; dsec_t m_last_pkt; CTmpFlowPerDirInfo m_per_dir[CS_NUM]; }; typedef CTmpFlowInfo * flow_tmp_t; typedef std::map flow_tmp_map_t; typedef flow_tmp_map_t::iterator flow_tmp_map_iter_t; bool CCapFileFlowInfo::is_valid_template_load_time(std::string & err){ err=""; int i; for (i=0; im_pkt_indication; if ( lpd->getEtherOffset() !=0 ){ err=" supported template Ether offset start is 0 \n"; return (false); } if ( lpd->getIpOffset() !=14 ){ err=" supported template ip offset is 14 \n"; return (false); } if ( lpd->is_ipv6() ){ if ( lpd->getTcpOffset() != (14+40) ){ err=" supported template tcp/udp offset is 54, no ipv6 option header is supported \n"; return (false); } }else{ if ( lpd->getTcpOffset() != (14+20) ){ err=" supported template tcp/udp offset is 34, no ipv4 option is allowed in this version \n"; return (false); } } } if ( CGlobalInfo::is_learn_mode() ) { if ( GetPacket(0)->m_pkt_indication.m_desc.IsPluginEnable() ) { err="plugins are not supported with --learn mode \n"; return(false); } } return(true); } /** * update global info * 1. maximum aging * 2. per sub-flow pkt_num/max-pkt per dir and per global */ void CCapFileFlowInfo::update_info(){ flow_tmp_map_iter_t iter; flow_tmp_map_t ft; CTmpFlowInfo * lpFlow; int i; dsec_t ctime=0.0; // first iteration, lern all the info into a temp flow table for (i=0; im_pkt_indication.m_desc; uint16_t flow_id = desc->getFlowId(); CPacketDescriptorPerDir * lpCurPacket = desc->GetDirInfo(); pkt_dir_t dir=desc->IsInitSide()?CLIENT_SIDE:SERVER_SIDE; // with respect to the first sub-flow in the template //update lpFlow iter = ft.find(flow_id); if (iter != ft.end() ) { lpFlow=(*iter).second; }else{ lpFlow = new CTmpFlowInfo(); assert(lpFlow); ft.insert(flow_tmp_map_t::value_type(flow_id,lpFlow)); //add it } // main info lpCurPacket->SetPktNum(lpFlow->m_per_dir[dir].m_pkt_id); lpFlow->m_max_pkts++; lpFlow->m_per_dir[dir].m_pkt_id++; dsec_t delta = ctime - lpFlow->m_last_pkt ; lpFlow->m_last_pkt = ctime; if (delta > lpFlow->m_max_aging_sec) { lpFlow->m_max_aging_sec = delta; } // per direction info if (im_pkt_indication.m_cap_ipg; } } for (i=0; im_pkt_indication.m_desc; uint16_t flow_id = desc->getFlowId(); CPacketDescriptorPerDir * lpCurPacket = desc->GetDirInfo(); pkt_dir_t dir=desc->IsInitSide()?CLIENT_SIDE:SERVER_SIDE; // with respect to the first sub-flow in the template iter = ft.find(flow_id); assert( iter != ft.end() ); lpFlow=(*iter).second; if ( (lpFlow->m_per_dir[0].m_pkt_id >0) && (lpFlow->m_per_dir[1].m_pkt_id >0) ) { /* we have both dir */ lp->m_pkt_indication.m_desc.SetBiPluginEnable(true); } lpCurPacket->SetMaxPkts(lpFlow->m_per_dir[dir].m_pkt_id); lp->m_pkt_indication.m_desc.SetMaxPktsPerFlow(lpFlow->m_max_pkts); lp->m_pkt_indication.m_desc.SetMaxFlowTimeout(lpFlow->m_max_aging_sec); } /* in case of learn mode , we need to mark the first packet */ if ( CGlobalInfo::is_learn_mode() ) { CFlowPktInfo * lp= GetPacket(0); assert(lp); /* only for bi directionl traffic mask the learn flag , only for the first packet */ if ( lp->m_pkt_indication.m_desc.IsBiDirectionalFlow() ){ lp->mask_as_learn(); } } if ( ft.empty() ) return; flow_tmp_map_iter_t it; for (it= ft.begin(); it != ft.end(); ++it) { CTmpFlowInfo *lp = it->second; assert(lp); delete lp; } ft.clear(); } enum CCapFileFlowInfo::load_cap_file_err CCapFileFlowInfo::load_cap_file(std::string cap_file, uint16_t _id, uint8_t plugin_id) { RemoveAll(); fprintf(stdout," -- loading cap file %s \n",cap_file.c_str()); CPacketParser parser; CPacketIndication pkt_indication; CCapReaderBase * lp=CCapReaderFactory::CreateReader((char *)cap_file.c_str(),0); if (lp == 0) { printf(" ERROR file %s does not exist or not supported \n",(char *)cap_file.c_str()); return kFileNotExist; } bool multi_flow_enable =( (plugin_id!=0)?true:false); CFlowTableMap flow; parser.Create(); flow.Create(0); m_total_bytes=0; m_total_flows=0; m_total_errors=0; CFlow * first_flow=0; bool first_flow_fif_is_swap=false; bool time_was_set=false; double last_time=0.0; CCapPktRaw raw_packet; int cnt=0; while ( true ) { /* read packet */ if ( lp->ReadPacket(&raw_packet) ==false ){ break; } cnt++; if ( !time_was_set ){ last_time=raw_packet.get_time(); time_was_set=true; }else{ if (raw_packet.get_time()pkt_len; pkt_indication.m_cap_ipg = raw_packet.get_time(); pkt_indication.m_flow =lpflow; pkt_indication.m_desc.SetFlowPktNum(lpflow->pkt_id); /* inc pkt_id inside the flow */ lpflow->pkt_id++; /* check that we don't have reserve TTL for duplication */ uint8_t ttl = pkt_indication.getTTL(); if ( (ttl == TTL_RESERVE_DUPLICATE) || (ttl == (TTL_RESERVE_DUPLICATE-1)) ) { pkt_indication.setTTL(TTL_RESERVE_DUPLICATE-4); } // Validation for first packet in flow if (is_fif) { lpflow->flow_id = m_total_flows; pkt_indication.m_desc.SetFlowId(lpflow->flow_id); if (m_total_flows == 0) { /* first flow */ first_flow =lpflow;/* save it for single flow support , to signal error */ lpflow->is_fif_swap =pkt_indication.m_desc.IsSwapTuple(); first_flow_fif_is_swap = pkt_indication.m_desc.IsSwapTuple(); pkt_indication.m_desc.SetInitSide(true); Append(&pkt_indication); m_total_flows++; } else { if ( multi_flow_enable ) { lpflow->is_fif_swap = pkt_indication.m_desc.IsSwapTuple(); /* in respect to the first flow */ bool init_side_in_repect_to_first_flow = ((first_flow_fif_is_swap?true:false) == lpflow->is_fif_swap)?true:false; pkt_indication.m_desc.SetInitSide(init_side_in_repect_to_first_flow); Append(&pkt_indication); m_total_flows++; } else { printf("More than one flow in this cap. Ignoring it !! \n"); pkt_indication.m_flow_key.Dump(stderr); m_total_errors++; } } if (CGlobalInfo::is_learn_mode(CParserOption::LEARN_MODE_TCP_ACK)) { // in this mode, first TCP packet must be SYN from client. if (pkt_indication.getIpProto() == IPPROTO_TCP) { TCPHeader *tcp = (TCPHeader *)(pkt_indication.getBasePtr() + pkt_indication.getTcpOffset()); if ( (! pkt_indication.m_desc.IsInitSide()) || (! tcp->getSynFlag()) ) { fprintf(stderr, "Error: In the chosen learn mode, first TCP packet should be SYN from client side.\n"); fprintf(stderr, " In cap file, first packet side direction is %s. TCP header is:\n", pkt_indication.m_desc.IsInitSide() ? "outside":"inside"); tcp->dump(stderr); fprintf(stderr, " Please give different CAP file, or try different --learn-mode\n"); return kNoSyn; } // We want at least the TCP flags to be inside first mbuf if (pkt_indication.getTcpOffset() + 14 > FIRST_PKT_SIZE) { fprintf(stderr, "Error: In the chosen learn mode, first TCP packet TCP flags offset should be less than %d, but it is %d.\n" , FIRST_PKT_SIZE, pkt_indication.getTcpOffset() + 14); fprintf(stderr, " Please give different CAP file, or try different --learn-mode\n"); return kTCPOffsetTooBig; } } } }else{ /* no FIF */ pkt_indication.m_desc.SetFlowId(lpflow->flow_id); if ( multi_flow_enable ==false ){ if (lpflow == first_flow) { // add to bool init_side= ((lpflow->is_fif_swap?true:false) == pkt_indication.m_desc.IsSwapTuple())?true:false; pkt_indication.m_desc.SetInitSide( init_side ); Append(&pkt_indication); }else{ //printf(" more than one flow in this cap ignot it !! \n"); m_total_errors++; } }else{ /* support multi-flow, */ /* work in respect to first flow */ bool init_side= ((first_flow_fif_is_swap?true:false) == pkt_indication.m_desc.IsSwapTuple())?true:false; pkt_indication.m_desc.SetInitSide( init_side ); Append(&pkt_indication); } } if (CGlobalInfo::is_learn_mode(CParserOption::LEARN_MODE_TCP_ACK)) { // This test must be down here, after initializing init side indication if (pkt_indication.getIpProto() != IPPROTO_TCP && !pkt_indication.m_desc.IsInitSide()) { fprintf(stderr, "Error: In the chosen learn mode, all packets from server to client in CAP file should be TCP.\n"); fprintf(stderr, " Please give different CAP file, or try different --learn-mode\n"); return kNoTCPFromServer; } } }else{ fprintf(stderr, "ERROR packet %d is not supported, should be Ethernet/IP(0x0800)/(TCP|UDP) format try to convert it using Wireshark !\n",cnt); return kPktNotSupp; } }else{ fprintf(stderr, "ERROR packet %d is not supported, should be Ethernet/IP(0x0800)/(TCP|UDP) format try to convert it using Wireshark !\n",cnt); return kPktProcessFail; } } /* set the last */ CFlowPktInfo * last_pkt =GetPacket((uint32_t)(Size()-1)); last_pkt->m_pkt_indication.m_desc.SetIsLastPkt(true); int i; for (i=1; im_pkt_indication.m_cap_ipg = lp->m_pkt_indication.m_cap_ipg- lp_prev->m_pkt_indication.m_cap_ipg; if ( lp->m_pkt_indication.m_desc.IsInitSide() != lp_prev->m_pkt_indication.m_desc.IsInitSide()) { lp_prev->m_pkt_indication.m_desc.SetRtt(true); } } GetPacket((uint32_t)Size()-1)->m_pkt_indication.m_cap_ipg=0.0; m_total_errors += parser.m_counter.getTotalErrors(); /* dump the flow */ //Dump(stdout); //flow.Dump(stdout); flow.Delete(); //parser.Dump(stdout); parser.Delete(); //fprintf(stdout," -- finish loading cap file \n"); //fprintf(stdout,"\n"); delete lp; if ( m_total_errors > 0 ) { parser.m_counter.Dump(stdout); fprintf(stderr, " ERORR in one of the cap file, you should have one flow per cap file or valid plugin \n"); return kCapFileErr; } return kOK; } void CCapFileFlowInfo::update_pcap_mode(){ int i; for (i=0; i<(int)Size(); i++) { CFlowPktInfo * lp=GetPacket((uint32_t)i); lp->m_pkt_indication.m_desc.SetPcapTiming(true); } } void CCapFileFlowInfo::get_total_memory(CCCapFileMemoryUsage & memory){ memory.clear(); int i; for (i=0; i<(int)Size(); i++) { CFlowPktInfo * lp=GetPacket((uint32_t)i); if ( lp->m_packet->pkt_len > FIRST_PKT_SIZE ) { memory.add_size(lp->m_packet->pkt_len - FIRST_PKT_SIZE); } } } double CCapFileFlowInfo::get_cap_file_length_sec(){ dsec_t sum=0.0; int i; for (i=0; i<(int)Size(); i++) { CFlowPktInfo * lp=GetPacket((uint32_t)i); sum+=lp->m_pkt_indication.m_cap_ipg; } return (sum); } void CCapFileFlowInfo::update_min_ipg(dsec_t min_ipg, dsec_t override_ipg){ int i; for (i=0; i<(int)Size(); i++) { CFlowPktInfo * lp=GetPacket((uint32_t)i); if ( lp->m_pkt_indication.m_cap_ipg < min_ipg ){ lp->m_pkt_indication.m_cap_ipg=override_ipg; } if ( lp->m_pkt_indication.m_cap_ipg < override_ipg ){ lp->m_pkt_indication.m_cap_ipg=override_ipg; } } } void CCapFileFlowInfo::Dump(FILE *fd){ int i; //CCapPacket::DumpHeader(fd); for (i=0; i<(int)Size(); i++) { fprintf(fd,"pkt_id : %d \n",i+1); fprintf(fd,"-----------\n"); CFlowPktInfo * lp=GetPacket((uint32_t)i); lp->Dump(fd); } } // add pkt indication void CCapFileFlowInfo::Append(CPacketIndication * pkt_indication){ CFlowPktInfo * lp; lp = new CFlowPktInfo(); lp->Create( pkt_indication ); m_flow_pkts.push_back(lp); } void CCCapFileMemoryUsage::Add(const CCCapFileMemoryUsage & obj){ int i; for (i=0; igenerate_new_mbuf(&node); //rte_pktmbuf_dump(buf, buf->pkt_len); rte_pktmbuf_free(buf); } } void CCapFileFlowInfo::RemoveAll(){ int i; m_total_bytes=0; m_total_errors = 0; m_total_flows = 0; for (i=0; i<(int)Size(); i++) { flow_pkt_info_t lp=GetPacket((uint32_t)i); lp->Delete(); delete lp; } // free all the pointers m_flow_pkts.clear(); } void CCapFileFlowInfo::Delete(){ RemoveAll(); } void operator >> (const YAML::Node& node, mac_mapping_t &fi) { utl_yaml_read_ip_addr(node,"ip", fi.ip); const YAML::Node& mac_info = node["mac"]; for(unsigned i=0;i> value; fi.mac.mac[i] = value; } } void operator >> (const YAML::Node& node, std::map &mac_info) { const YAML::Node& mac_node = node["items"]; mac_mapping_t mac_mapping; for (unsigned i=0;i> mac_mapping; mac_info[mac_mapping.ip] = mac_mapping.mac; } } void operator >> (const YAML::Node& node, CFlowYamlDpPkt & fi) { uint32_t val; node["pkt_id"] >> val; fi.m_pkt_id =(uint8_t)val; node["pyld_offset"] >> val; fi.m_pyld_offset =(uint8_t)val; node["type"] >> val; fi.m_type =(uint8_t)val; node["len"] >> val; fi.m_len =(uint8_t)val; node["mask"] >> val; fi.m_pkt_mask =val; } void operator >> (const YAML::Node& node, CVlanYamlInfo & fi) { uint32_t tmp; if ( node.FindValue("enable") ){ node["enable"] >> tmp ; fi.m_enable=tmp; node["vlan0"] >> tmp; fi.m_vlan_per_port[0] = tmp; node["vlan1"] >> tmp; fi.m_vlan_per_port[1] = tmp; } } void operator >> (const YAML::Node& node, CFlowYamlInfo & fi) { node["name"] >> fi.m_name; if ( node.FindValue("client_pool") ){ node["client_pool"] >> fi.m_client_pool_name; }else{ fi.m_client_pool_name = "default"; } if ( node.FindValue("server_pool") ){ node["server_pool"] >> fi.m_server_pool_name; }else{ fi.m_server_pool_name = "default"; } node["cps"] >> fi.m_k_cps; fi.m_k_cps = fi.m_k_cps/1000.0; double t; node["ipg"] >> t; fi.m_ipg_sec =t/1000000.0; node["rtt"] >> t; fi.m_rtt_sec = t/1000000.0; node["w"] >> fi.m_w; if ( node.FindValue("cap_ipg") ){ node["cap_ipg"] >> fi.m_cap_mode; fi.m_cap_mode_was_set =true; }else{ fi.m_cap_mode_was_set =false; } if ( node.FindValue("wlength") ){ node["wlength"] >> fi.m_wlength; fi.m_wlength_set=true; }else{ fi.m_wlength_set=false; fi.m_wlength =500; } if ( node.FindValue("limit") ){ node["limit"] >> fi.m_limit; fi.m_limit_was_set = true; }else{ fi.m_limit_was_set = false; fi.m_limit = 0; } if ( node.FindValue("plugin_id") ){ uint32_t plugin_val; node["plugin_id"] >> plugin_val; fi.m_plugin_id=plugin_val; }else{ fi.m_plugin_id=0; } fi.m_one_app_server_was_set = false; fi.m_one_app_server = false; if ( utl_yaml_read_ip_addr(node, "server_addr", fi.m_server_addr) ){ try { node["one_app_server"] >> fi.m_one_app_server; fi.m_one_app_server_was_set=true; } catch ( const std::exception& e ) { fi.m_one_app_server_was_set = false; fi.m_one_app_server = false; } } if ( ( fi.m_limit_was_set ) && (fi.m_plugin_id !=0) ){ fprintf(stderr," limit can't be non zero when plugin is set, you must have only one of the options set"); exit(-1); } if ( node.FindValue("dyn_pyload") ){ const YAML::Node& dyn_pyload = node["dyn_pyload"]; for(unsigned i=0;i> fd; if ( fi.m_dpPkt == 0 ){ fi.m_dpPkt = new CFlowYamlDynamicPyloadPlugin(); if (fi.m_plugin_id == 0) { fi.m_plugin_id = mpDYN_PYLOAD; }else{ fprintf(stderr," plugin should be zero with dynamic pyload program"); exit(-1); } } fi.m_dpPkt->Add(fd); } }else{ fi.m_dpPkt=0; } } void operator >> (const YAML::Node& node, CFlowsYamlInfo & flows_info) { node["duration"] >> flows_info.m_duration_sec; if ( node.FindValue("generator") ) { node["generator"] >> flows_info.m_tuple_gen; flows_info.m_tuple_gen_was_set =true; }else{ flows_info.m_tuple_gen_was_set =false; } // m_ipv6_set will be true if and only if both src_ipv6 // and dst_ipv6 are provided. These are used to set // the most significant 96-bits of the IPv6 address; the // least significant 32-bits come from the ipv4 address // (what is set above). // // If the IPv6 src/dst is not provided in the yaml file, // then the most significant 96-bits will be set to 0 // which represents an IPv4-compatible IPv6 address. // // If desired, an IPv4-mapped IPv6 address can be // formed by providing src_ipv6,dst_ipv6 and specifying // {0,0,0,0,0,0xffff} flows_info.m_ipv6_set=true; if ( node.FindValue("src_ipv6") ) { const YAML::Node& src_ipv6_info = node["src_ipv6"]; if (src_ipv6_info.size() == 6 ){ for(unsigned i=0;i> fi; flows_info.m_src_ipv6.push_back(fi); } } }else{ flows_info.m_ipv6_set=false; } if ( node.FindValue("dst_ipv6") ) { const YAML::Node& dst_ipv6_info = node["dst_ipv6"]; if (dst_ipv6_info.size() == 6 ){ for(unsigned i=0;i> fi; flows_info.m_dst_ipv6.push_back(fi); } } }else{ flows_info.m_ipv6_set=false; } if ( node.FindValue("cap_ipg") ) { node["cap_ipg"] >> flows_info.m_cap_mode; flows_info.m_cap_mode_set=true; }else{ flows_info.m_cap_mode=false; flows_info.m_cap_mode_set=false; } double t=0.0; if ( node.FindValue("cap_ipg_min") ) { node["cap_ipg_min"] >> t ; flows_info.m_cap_ipg_min = t/1000000.0; flows_info.m_cap_ipg_min_set=true; }else{ flows_info.m_cap_ipg_min_set=false; flows_info.m_cap_ipg_min = 20; } if ( node.FindValue("cap_override_ipg") ) { node["cap_override_ipg"] >> t; flows_info.m_cap_overide_ipg = t/1000000.0; flows_info.m_cap_overide_ipg_set = true; }else{ flows_info.m_cap_overide_ipg_set = false; flows_info.m_cap_overide_ipg = 0; } if (node.FindValue("wlength")) { node["wlength"] >> flows_info.m_wlength; flows_info.m_wlength_set=true; }else{ flows_info.m_wlength_set=false; flows_info.m_wlength =100; } if (node.FindValue("one_app_server")) { printf("one_app_server should be configured per template. \n" "Will ignore this configuration\n"); } flows_info.m_one_app_server =false; flows_info.m_one_app_server_was_set=false; if (node.FindValue("vlan")) { node["vlan"] >> flows_info.m_vlan_info; } if (node.FindValue("mac_override_by_ip")) { node["mac_override_by_ip"] >> flows_info.m_mac_replace_by_ip; }else{ flows_info.m_mac_replace_by_ip =false; } const YAML::Node& mac_info = node["mac"]; for(unsigned i=0;i> fi; flows_info.m_mac_base.push_back(fi); } const YAML::Node& cap_info = node["cap_info"]; for(unsigned i=0;i> fi; fi.m_client_pool_idx = flows_info.m_tuple_gen.get_client_pool_id(fi.m_client_pool_name); fi.m_server_pool_idx = flows_info.m_tuple_gen.get_server_pool_id(fi.m_server_pool_name); flows_info.m_vec.push_back(fi); } } void CVlanYamlInfo::Dump(FILE *fd){ fprintf(fd," vlan enable : %d \n",m_enable); fprintf(fd," vlan val : %d ,%d \n",m_vlan_per_port[0],m_vlan_per_port[1]); } void CFlowsYamlInfo::Dump(FILE *fd){ fprintf(fd," duration : %f sec \n",m_duration_sec); fprintf(fd,"\n"); if (CGlobalInfo::is_ipv6_enable()) { int idx; fprintf(fd," src_ipv6 : "); for (idx=0; idx<5; idx++){ fprintf(fd,"%04x:", CGlobalInfo::m_options.m_src_ipv6[idx]); } fprintf(fd,"%04x\n", CGlobalInfo::m_options.m_src_ipv6[5]); fprintf(fd," dst_ipv6 : "); for (idx=0; idx<5; idx++){ fprintf(fd,"%04x:", CGlobalInfo::m_options.m_dst_ipv6[idx]); } fprintf(fd,"%04x\n", CGlobalInfo::m_options.m_dst_ipv6[5]); } if ( !m_cap_mode_set ) { fprintf(fd," cap_ipg : wasn't set \n"); }else{ fprintf(fd," cap_ipg : %d \n",m_cap_mode?1:0); } if ( !m_cap_ipg_min_set ){ fprintf(fd," cap_ipg_min : wasn't set \n"); }else{ fprintf(fd," cap_ipg_min : %f \n",m_cap_ipg_min); } if ( !m_cap_overide_ipg_set ){ fprintf(fd," cap_override_ipg : wasn't set \n"); }else{ fprintf(fd," cap_override_ipg : %f \n",m_cap_overide_ipg); } if ( !m_wlength_set ){ fprintf(fd," wlength : wasn't set \n"); }else{ fprintf(fd," m_wlength : %d \n",m_wlength); } fprintf(fd," one_server_for_application : %d \n",m_one_app_server?1:0); fprintf(fd," one_server_for_application_was_set : %d \n",m_one_app_server_was_set?1:0); m_vlan_info.Dump(fd); fprintf(fd," mac base : "); int i; for (i=0; i<(int)m_mac_base.size(); i++) { if (i< (int)(m_mac_base.size()-1) ) { fprintf(fd,"0x%02x,",m_mac_base[i]); }else{ fprintf(fd,"0x%02x",m_mac_base[i]); } } fprintf(fd,"\n"); fprintf(fd," cap file info \n"); fprintf(fd," ------------- \n"); for (i=0; i<(int)m_vec.size(); i++) { m_vec[i].Dump(fd); } } /* example for YAML file - duration : 10.0 cap_info : - name: hey1.pcap cps : 12.0 ipg : 0.0001 - name: hey2.pcap cps : 11.0 ipg : 0.0001 */ bool CFlowsYamlInfo::verify_correctness(uint32_t num_threads) { if ( m_tuple_gen_was_set ==false ){ printf(" ERROR there must be a generator field in YAML , the old format is deprecated \n"); printf(" This is not supported : \n"); printf(" min_src_ip : 0x10000001 \n"); printf(" max_src_ip : 0x50000001 \n"); printf(" min_dst_ip : 0x60000001 \n"); printf(" max_dst_ip : 0x60000010 \n"); printf(" This is supported : \n"); printf("generator : \n"); printf(" distribution : \"seq\" \n"); printf(" clients_start : \"16.0.0.1\" \n"); printf(" clients_end : \"16.0.1.255\" \n"); printf(" servers_start : \"48.0.0.1\" \n"); printf(" servers_end : \"48.0.0.255\" \n"); printf(" clients_per_gb : 201 \n"); printf(" min_clients : 101 \n"); printf(" dual_port_mask : \"1.0.0.0\" \n"); printf(" tcp_aging : 1 \n"); printf(" udp_aging : 1 \n"); return(false); } if ( !m_tuple_gen.is_valid(num_threads,is_any_plugin_configured()) ){ return (false); } /* patch defect trex-54 */ if ( is_any_plugin_configured() ){ /*Plugin is configured. in that case due to a limitation ( defect trex-54 ) the number of servers should be bigger than number of clients */ int i; for (i=0; i<(int)m_vec.size(); i++) { CFlowYamlInfo * lp=&m_vec[i]; if ( lp->m_plugin_id ){ uint8_t c_idx = lp->m_client_pool_idx; uint8_t s_idx = lp->m_server_pool_idx; uint32_t total_clients = m_tuple_gen.m_client_pool[c_idx].getTotalIps(); uint32_t total_servers = m_tuple_gen.m_server_pool[s_idx].getTotalIps(); if ( total_servers < total_clients ){ printf(" Plugin is configured. in that case due to a limitation ( defect trex-54 ) \n"); printf(" the number of servers should be bigger than number of clients \n"); printf(" client_pool_name : %s \n", lp->m_client_pool_name.c_str()); printf(" server_pool_name : %s \n", lp->m_server_pool_name.c_str()); return (false); } uint32_t mul = total_servers / total_clients; uint32_t new_server_num = mul * total_clients; if ( new_server_num != total_servers ) { printf(" Plugin is configured. in that case due to a limitation ( defect trex-54 ) \n"); printf(" the number of servers should be exact multiplication of the number of clients \n"); printf(" client_pool_name : %s clients %d \n", lp->m_client_pool_name.c_str(),total_clients); printf(" server_pool_name : %s servers %d should be %d \n", lp->m_server_pool_name.c_str(),total_servers,new_server_num); return (false); } } } } return(true); } int CFlowsYamlInfo::load_from_yaml_file(std::string file_name){ m_vec.clear(); if ( !utl_is_file_exists (file_name) ){ printf(" ERROR file %s does not exist \n",file_name.c_str()); exit(-1); } try { std::ifstream fin((char *)file_name.c_str()); YAML::Parser parser(fin); YAML::Node doc; parser.GetNextDocument(doc); for(unsigned i=0;i> *this; break; } } catch ( const std::exception& e ) { std::cout << e.what() << "\n"; exit(-1); } /* update from user input */ if (CGlobalInfo::m_options.m_duration > 0.1) { m_duration_sec = CGlobalInfo::m_options.m_duration; } int i; m_is_plugin_configured=false; for (i=0; i<(int)m_vec.size(); i++) { m_vec[i].m_k_cps =m_vec[i].m_k_cps*CGlobalInfo::m_options.m_factor; if (( ! m_vec[i].m_cap_mode_was_set ) && (m_cap_mode_set ) ){ m_vec[i].m_cap_mode = m_cap_mode; } if (( ! m_vec[i].m_wlength_set ) && (m_wlength_set ) ){ m_vec[i].m_wlength = m_wlength; } if (( ! m_vec[i].m_one_app_server_was_set ) && (m_one_app_server_was_set ) ){ m_vec[i].m_one_app_server = m_one_app_server; } if ( m_cap_overide_ipg_set ){ m_vec[i].m_ipg_sec = m_cap_overide_ipg; m_vec[i].m_rtt_sec = m_cap_overide_ipg; } if ( m_vec[i].m_plugin_id ){ m_is_plugin_configured=true; } } return 0; } void CFlowStats::Clear(){ m_id=0; m_name=""; m_pkt=0.0; m_bytes=0.0; m_cps=0.0; m_mb_sec=0.0; m_mB_sec=0.0; m_c_flows=0.0; m_pps =0.0; m_total_Mbytes=00 ; m_errors =0; m_flows =0 ; m_memory.clear(); } void CFlowStats::Add(const CFlowStats & obj){ m_pkt += obj.m_pkt ; m_bytes += obj.m_bytes ; m_cps += obj.m_cps ; m_mb_sec += obj.m_mb_sec ; m_mB_sec += obj.m_mB_sec ; m_c_flows += obj.m_c_flows ; m_pps += obj.m_pps ; m_total_Mbytes +=obj.m_total_Mbytes ; m_errors +=obj.m_errors; m_flows +=obj.m_flows ; m_memory.Add(obj.m_memory); } void CFlowStats::DumpHeader(FILE *fd){ fprintf(fd," %2s,%-40s,%4s,%4s,%5s,%7s,%9s,%9s,%9s,%10s,%5s,%7s,%4s,%4s \n", "id","name","tps","cps","f-pkts","f-bytes","duration","Mb/sec","MB/sec","c-flows","PPS","total-Mbytes-duration","errors","flows"); } void CFlowStats::Dump(FILE *fd){ //"name","cps","f-pkts","f-bytes","Mb/sec","MB/sec","c-flows","PPS","total-Mbytes-duration","errors","flows" fprintf(fd," %02d, %-40s ,%4.2f,%4.2f, %5.0f , %7.0f ,%7.2f ,%7.2f , %7.2f , %10.0f , %5.0f , %7.0f , %llu , %llu \n", m_id, m_name.c_str(), m_cps, get_normal_cps(), m_pkt, m_bytes, duration_sec, m_mb_sec, m_mB_sec, m_c_flows, m_pps, m_total_Mbytes, (unsigned long long)m_errors, (unsigned long long)m_flows); } bool CFlowGeneratorRecPerThread::Create(CTupleGeneratorSmart * global_gen, CFlowYamlInfo * info, CFlowsYamlInfo * yaml_flow_info, CCapFileFlowInfo * flow_info, uint16_t _id, uint32_t thread_id){ BP_ASSERT(info); m_thread_id =thread_id ; tuple_gen.Create(global_gen, info->m_client_pool_idx, info->m_server_pool_idx); CTupleGenYamlInfo * lpt; lpt = &yaml_flow_info->m_tuple_gen; tuple_gen.SetSingleServer(info->m_one_app_server, info->m_server_addr, getDualPortId(thread_id), lpt->m_client_pool[info->m_client_pool_idx].getDualMask() ); tuple_gen.SetW(info->m_w); m_id =_id; m_info =info; m_flows_info = yaml_flow_info; // set policer give bucket size for bursts m_policer.set_cir(info->m_k_cps*1000.0); m_policer.set_level(0.0); m_policer.set_bucket_size(100.0); /* pointer to global */ m_flow_info = flow_info; return (true); } void CFlowGeneratorRecPerThread::Delete(){ tuple_gen.Delete(); } void CFlowGeneratorRecPerThread::Dump(FILE *fd){ fprintf(fd," configuration info "); fprintf(fd," -----------------"); m_info->Dump(fd); fprintf(fd," -----------------"); m_flow_info->Dump(fd); } void CFlowGeneratorRecPerThread::getFlowStats(CFlowStats * stats){ double t_pkt=(double)m_flow_info->Size(); double t_bytes=(double)m_flow_info->get_total_bytes(); double cps=m_info->m_k_cps *1000.0; double mb_sec = (cps*t_bytes*8.0)/(_1Mb_DOUBLE); double mB_sec = (cps*t_bytes)/(_1Mb_DOUBLE); double c_flow_windows_sec=0.0; if (m_info->m_cap_mode) { c_flow_windows_sec = m_flow_info->get_cap_file_length_sec(); }else{ c_flow_windows_sec = t_pkt * m_info->m_ipg_sec; } double c_flows = cps*c_flow_windows_sec*m_flow_info->get_total_flows(); double pps =cps*t_pkt; double total_Mbytes = mB_sec * m_flows_info->m_duration_sec; uint64_t errors = m_flow_info->get_total_errors(); uint64_t flows = m_flow_info->get_total_flows(); stats->m_id = m_id; stats->m_pkt = t_pkt; stats->m_bytes = t_bytes; stats->duration_sec = c_flow_windows_sec; stats->m_name = m_info->m_name.c_str(); stats->m_cps = cps; stats->m_mb_sec = mb_sec; stats->m_mB_sec = mB_sec; stats->m_c_flows = c_flows; stats->m_pps = pps; stats->m_total_Mbytes = total_Mbytes; stats->m_errors = errors; stats->m_flows = flows; } void CFlowGeneratorRec::Dump(FILE *fd){ fprintf(fd," configuration info "); fprintf(fd," -----------------"); m_info->Dump(fd); fprintf(fd," -----------------"); m_flow_info.Dump(fd); } void CFlowGeneratorRec::getFlowStats(CFlowStats * stats){ double t_pkt=(double)m_flow_info.Size(); double t_bytes=(double)m_flow_info.get_total_bytes(); double cps=m_info->m_k_cps *1000.0; double mb_sec = (cps*t_bytes*8.0)/(_1Mb_DOUBLE); double mB_sec = (cps*t_bytes)/(_1Mb_DOUBLE); double c_flow_windows_sec=0.0; if (m_info->m_cap_mode) { c_flow_windows_sec = m_flow_info.get_cap_file_length_sec(); }else{ c_flow_windows_sec = t_pkt * m_info->m_ipg_sec; } m_flow_info.get_total_memory(stats->m_memory); double c_flows = cps*c_flow_windows_sec; double pps =cps*t_pkt; double total_Mbytes = mB_sec * m_flows_info->m_duration_sec; uint64_t errors = m_flow_info.get_total_errors(); uint64_t flows = m_flow_info.get_total_flows(); stats->m_id = m_id; stats->m_pkt = t_pkt; stats->m_bytes = t_bytes; stats->duration_sec = c_flow_windows_sec; stats->m_name = m_info->m_name.c_str(); stats->m_cps = cps; stats->m_mb_sec = mb_sec; stats->m_mB_sec = mB_sec; stats->m_c_flows = c_flows; stats->m_pps = pps; stats->m_total_Mbytes = total_Mbytes; stats->m_errors = errors; stats->m_flows = flows; } void CFlowGeneratorRec::fixup_ipg_if_needed(void){ if ( m_flows_info->m_cap_mode ) { m_flow_info.update_pcap_mode(); } if ( (m_flows_info->m_cap_mode) && (m_flows_info->m_cap_ipg_min_set) && (m_flows_info->m_cap_overide_ipg_set) ){ m_flow_info.update_min_ipg(m_flows_info->m_cap_ipg_min, m_flows_info->m_cap_overide_ipg); } } bool CFlowGeneratorRec::Create(CFlowYamlInfo * info, CFlowsYamlInfo * flows_info, uint16_t _id){ BP_ASSERT(info); m_id=_id; m_info=info; m_flows_info=flows_info; m_flow_info.Create(); // set policer give bucket size for bursts m_policer.set_cir(info->m_k_cps*1000.0); m_policer.set_level(0.0); m_policer.set_bucket_size(100.0); int res=m_flow_info.load_cap_file(info->m_name.c_str(),_id,m_info->m_plugin_id); if ( res==0 ) { fixup_ipg_if_needed(); std::string err; /* verify that template are valid */ bool is_valid=m_flow_info.is_valid_template_load_time(err); if (!is_valid) { printf("\n ERROR template file is not valid '%s' \n",err.c_str()); return (false); } m_flow_info.update_info(); return (true); }else{ return (false); } } void CFlowGeneratorRec::Delete(){ m_flow_info.Delete(); } void CGenNode::DumpHeader(FILE *fd){ fprintf(fd," pkt_id,time,fid,pkt_info,pkt,len,type,is_init,is_last,type,thread_id,src_ip,dest_ip,src_port \n"); } void CGenNode::free_gen_node(){ rte_mbuf_t * m=get_cache_mbuf(); if ( unlikely(m != NULL) ) { rte_pktmbuf_free(m); m_plugin_info=0; } } void CGenNode::Dump(FILE *fd){ fprintf(fd,"%.6f,%llx,%p,%llu,%d,%d,%d,%d,%d,%d,%x,%x,%d\n", m_time, (unsigned long long)m_flow_id, m_pkt_info, (unsigned long long)m_pkt_info->m_pkt_indication.m_packet->pkt_cnt, m_pkt_info->m_pkt_indication.m_packet->pkt_len, m_pkt_info->m_pkt_indication.m_desc.getId(), (m_pkt_info->m_pkt_indication.m_desc.IsInitSide()?1:0), m_pkt_info->m_pkt_indication.m_desc.IsLastPkt(), m_type, m_thread_id, m_src_ip, m_dest_ip, m_src_port); } void CNodeGenerator::set_vif(CVirtualIF * v_if){ m_v_if = v_if; } bool CNodeGenerator::Create(CFlowGenListPerThread * parent){ m_v_if =0; m_parent=parent; m_socket_id =0; m_is_realtime =CGlobalInfo::is_realtime(); m_realtime_his.Create(); return(true); } void CNodeGenerator::Delete(){ m_realtime_his.Delete(); } void CNodeGenerator::add_node(CGenNode * mynode){ m_p_queue.push(mynode); } void CNodeGenerator::remove_all(CFlowGenListPerThread * thread){ CGenNode *node; while (!m_p_queue.empty()) { node = m_p_queue.top(); m_p_queue.pop(); /* sanity check */ if (node->m_type == CGenNode::STATELESS_PKT) { CGenNodeStateless * p=(CGenNodeStateless *)node; /* need to be changed in Pause support */ assert(p->is_mask_for_free()); } thread->free_node( node); } } int CNodeGenerator::open_file(std::string file_name, CPreviewMode * preview_mode){ BP_ASSERT(m_v_if); m_preview_mode =*preview_mode; /* ser preview mode */ m_v_if->set_review_mode(preview_mode); m_v_if->open_file(file_name); m_cnt = 0; m_non_active = 0; m_limit = 0; return (0); } int CNodeGenerator::close_file(CFlowGenListPerThread * thread){ remove_all(thread); BP_ASSERT(m_v_if); m_v_if->close_file(); return (0); } int CNodeGenerator::update_stl_stats(CGenNodeStateless *node_sl){ m_cnt++; if (!node_sl->is_node_active()) { m_non_active++; } #ifdef _DEBUG if ( m_preview_mode.getVMode() >2 ){ fprintf(stdout," %4lu ,", (ulong)m_cnt); fprintf(stdout," %4lu ,", (ulong)m_non_active); node_sl->Dump(stdout); } #endif return (0); } int CNodeGenerator::update_stats(CGenNode * node){ if ( m_preview_mode.getVMode() >2 ){ fprintf(stdout," %llu ,", (unsigned long long)m_cnt); node->Dump(stdout); m_cnt++; } return (0); } bool CNodeGenerator::has_limit_reached() { /* do we have a limit and has it passed ? */ return ( (m_limit > 0) && (m_cnt >= m_limit) ); } bool CFlowGenListPerThread::Create(uint32_t thread_id, uint32_t core_id, CFlowGenList * flow_list, uint32_t max_threads){ m_non_active_nodes = 0; m_terminated_by_master=false; m_flow_list =flow_list; m_core_id= core_id; m_tcp_dpc= 0; m_udp_dpc=0; m_max_threads=max_threads; m_thread_id=thread_id; m_cpu_cp_u.Create(&m_cpu_dp_u); uint32_t socket_id=rte_lcore_to_socket_id(m_core_id); char name[100]; sprintf(name,"nodes-%d",m_core_id); //printf(" create thread %d %s socket: %d \n",m_core_id,name,socket_id); m_node_pool = utl_rte_mempool_create_non_pkt(name, CGlobalInfo::m_memory_cfg.get_each_core_dp_flows(), sizeof(CGenNode), 128, 0 , socket_id); //printf(" pool %p \n",m_node_pool); m_node_gen.Create(this); m_flow_id_to_node_lookup.Create(); /* split the clients to threads */ CTupleGenYamlInfo * tuple_gen = &m_flow_list->m_yaml_info.m_tuple_gen; m_smart_gen.Create(0,m_thread_id,m_flow_list->get_is_mac_conf()); /* split the clients to threads using the mask */ CIpPortion portion; for (int i=0;im_client_pool.size();i++) { split_ips(m_thread_id, m_max_threads, getDualPortId(), tuple_gen->m_client_pool[i], portion); m_smart_gen.add_client_pool(tuple_gen->m_client_pool[i].m_dist, portion.m_ip_start, portion.m_ip_end, get_longest_flow(i,true), get_total_kcps(i,true)*1000, &m_flow_list->m_mac_info, tuple_gen->m_client_pool[i].m_tcp_aging_sec, tuple_gen->m_client_pool[i].m_udp_aging_sec ); } for (int i=0;im_server_pool.size();i++) { split_ips(m_thread_id, m_max_threads, getDualPortId(), tuple_gen->m_server_pool[i], portion); m_smart_gen.add_server_pool(tuple_gen->m_server_pool[i].m_dist, portion.m_ip_start, portion.m_ip_end, get_longest_flow(i,false), get_total_kcps(i,false)*1000, tuple_gen->m_server_pool[i].m_is_bundling); } init_from_global(portion); CMessagingManager * rx_dp=CMsgIns::Ins()->getRxDp(); m_ring_from_rx = rx_dp->getRingCpToDp(thread_id); m_ring_to_rx =rx_dp->getRingDpToCp(thread_id); assert(m_ring_from_rx); assert(m_ring_to_rx); /* create the info required for stateless DP core */ m_stateless_dp_info.create(thread_id, this); return (true); } /* return the client ip , port */ FORCE_NO_INLINE void CFlowGenListPerThread::handler_defer_job(CGenNode *p){ CGenNodeDeferPort * defer=(CGenNodeDeferPort *)p; int i; for (i=0; im_cnt; i++) { m_smart_gen.FreePort(defer->m_pool_idx[i], defer->m_clients[i],defer->m_ports[i]); } } FORCE_NO_INLINE void CFlowGenListPerThread::handler_defer_job_flush(void){ /* flush the pending job of free ports */ if (m_tcp_dpc) { handler_defer_job((CGenNode *)m_tcp_dpc); free_node((CGenNode *)m_tcp_dpc); m_tcp_dpc=0; } if (m_udp_dpc) { handler_defer_job((CGenNode *)m_udp_dpc); free_node((CGenNode *)m_udp_dpc); m_udp_dpc=0; } } void CFlowGenListPerThread::defer_client_port_free(bool is_tcp, uint32_t c_idx, uint16_t port, uint8_t c_pool_idx, CTupleGeneratorSmart * gen){ /* free is not required in this case */ if (!gen->IsFreePortRequired(c_pool_idx) ){ return; } CGenNodeDeferPort * defer; if (is_tcp) { if (gen->get_tcp_aging(c_pool_idx)==0) { gen->FreePort(c_pool_idx,c_idx,port); return; } defer=get_tcp_defer(); }else{ if (gen->get_udp_aging(c_pool_idx)==0) { gen->FreePort(c_pool_idx, c_idx,port); return; } defer=get_udp_defer(); } if ( defer->add_client(c_pool_idx, c_idx,port) ){ if (is_tcp) { m_node_gen.schedule_node((CGenNode *)defer,gen->get_tcp_aging(c_pool_idx)); m_tcp_dpc=0; }else{ m_node_gen.schedule_node((CGenNode *)defer,gen->get_udp_aging(c_pool_idx)); m_udp_dpc=0; } } } void CFlowGenListPerThread::defer_client_port_free(CGenNode *p){ defer_client_port_free(p->m_pkt_info->m_pkt_indication.m_desc.IsTcp(), p->m_src_idx,p->m_src_port,p->m_template_info->m_client_pool_idx, p->m_tuple_gen); } /* copy all info from global and div by num of threads */ void CFlowGenListPerThread::init_from_global(CIpPortion& portion){ /* copy generator , it is the same */ m_yaml_info =m_flow_list->m_yaml_info; /* copy first the flow info */ int i; for (i=0; i<(int)m_flow_list->m_cap_gen.size(); i++) { CFlowGeneratorRec * lp=m_flow_list->m_cap_gen[i]; CFlowGeneratorRecPerThread * lp_thread=new CFlowGeneratorRecPerThread(); /* TBD leak of memory */ CFlowYamlInfo * yaml_info =new CFlowYamlInfo(); yaml_info->m_name = lp->m_info->m_name; yaml_info->m_k_cps = lp->m_info->m_k_cps/(double)m_max_threads; yaml_info->m_ipg_sec = lp->m_info->m_ipg_sec; yaml_info->m_rtt_sec = lp->m_info->m_rtt_sec; yaml_info->m_w = lp->m_info->m_w; yaml_info->m_cap_mode =lp->m_info->m_cap_mode; yaml_info->m_wlength =lp->m_info->m_wlength; yaml_info->m_plugin_id = lp->m_info->m_plugin_id; yaml_info->m_one_app_server = lp->m_info->m_one_app_server; yaml_info->m_server_addr = lp->m_info->m_server_addr; yaml_info->m_dpPkt =lp->m_info->m_dpPkt; yaml_info->m_server_pool_idx=lp->m_info->m_server_pool_idx; yaml_info->m_client_pool_idx=lp->m_info->m_client_pool_idx; yaml_info->m_server_pool_name=lp->m_info->m_server_pool_name; yaml_info->m_client_pool_name=lp->m_info->m_client_pool_name; /* fix this */ assert(m_max_threads>0); if ( m_max_threads == 1 ) { /* we have one thread the limit */ yaml_info->m_limit = lp->m_info->m_limit; }else{ yaml_info->m_limit = lp->m_info->m_limit/m_max_threads; /* thread is zero base */ if ( m_thread_id == 0){ yaml_info->m_limit += lp->m_info->m_limit % m_max_threads; } if (yaml_info->m_limit==0) { yaml_info->m_limit=1; } } yaml_info->m_limit_was_set = lp->m_info->m_limit_was_set; yaml_info->m_flowcnt = 0; yaml_info->m_restart_time = ( yaml_info->m_limit_was_set ) ? (yaml_info->m_limit / (yaml_info->m_k_cps * 1000.0)) : 0; lp_thread->Create(&m_smart_gen, yaml_info, lp->m_flows_info, &lp->m_flow_info, lp->m_id, m_thread_id); m_cap_gen.push_back(lp_thread); } } static void free_map_flow_id_to_node(CGenNode *p){ CGlobalInfo::free_node(p); } void CFlowGenListPerThread::Delete(){ // free all current maps m_flow_id_to_node_lookup.remove_all(free_map_flow_id_to_node); // free object m_flow_id_to_node_lookup.Delete(); m_smart_gen.Delete(); m_node_gen.Delete(); Clean(); m_cpu_cp_u.Delete(); utl_rte_mempool_delete(m_node_pool); } void CFlowGenListPerThread::Clean(){ int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRecPerThread * lp=m_cap_gen[i]; if (lp->m_tuple_gen_was_set) { CTupleGeneratorSmart *gen; gen = lp->tuple_gen.get_gen(); gen->Delete(); delete gen; } lp->Delete(); delete lp; } m_cap_gen.clear(); } //uint64_t _start_time; void CNodeGenerator::dump_json(std::string & json){ json="{\"name\":\"tx-gen\",\"type\":0,\"data\":{"; m_realtime_his.dump_json("realtime-hist",json); json+="\"unknown\":0}}" ; } void CNodeGenerator::add_exit_node(CFlowGenListPerThread * thread, dsec_t max_time){ if ( max_time > 0 ) { CGenNode *exit_node = thread->create_node(); exit_node->m_type = CGenNode::EXIT_SCHED; exit_node->m_time = max_time; add_node(exit_node); } } inline bool CNodeGenerator::handle_stl_node(CGenNode * node, CFlowGenListPerThread * thread){ uint8_t type=node->m_type; if ( likely( type == CGenNode::STATELESS_PKT ) ) { m_p_queue.pop(); CGenNodeStateless *node_sl = (CGenNodeStateless *)node; /* if the stream has been deactivated - end */ if ( unlikely( node_sl->is_mask_for_free() ) ) { thread->free_node(node); } else { /* count before handle - node might be destroyed */ #ifdef TREX_SIM update_stl_stats(node_sl); #endif node_sl->handle(thread); #ifdef TREX_SIM if (has_limit_reached()) { thread->m_stateless_dp_info.stop_traffic(node_sl->get_port_id(), false, 0); } #endif } return (true); } return(false); } inline bool CNodeGenerator::do_work_stl(CGenNode * node, CFlowGenListPerThread * thread, bool always){ if ( handle_stl_node(node,thread)){ return (false); }else{ return (handle_slow_messages(node->m_type,node,thread,always)); } } inline bool CNodeGenerator::do_work_both(CGenNode * node, CFlowGenListPerThread * thread, dsec_t d_time, bool always ){ bool exit_scheduler=false; uint8_t type=node->m_type; bool done; if ( handle_stl_node (node,thread) ){ }else{ if ( likely( type == CGenNode::FLOW_PKT ) ) { /* PKT */ if ( !(node->is_repeat_flow()) || (always==false)) { flush_one_node_to_file(node); #ifdef _DEBUG update_stats(node); #endif } m_p_queue.pop(); if ( node->is_last_in_flow() ) { if ((node->is_repeat_flow()) && (always==false)) { /* Flow is repeated, reschedule it */ thread->reschedule_flow( node); }else{ /* Flow will not be repeated, so free node */ thread->free_last_flow_node( node); } }else{ node->update_next_pkt_in_flow(); m_p_queue.push(node); } }else{ if ((type == CGenNode::FLOW_FIF)) { /* callback to our method */ m_p_queue.pop(); if ( always == false) { thread->m_cur_time_sec = node->m_time ; thread->generate_flows_roundrobin(&done); if (!done) { node->m_time +=d_time; m_p_queue.push(node); }else{ thread->free_node(node); } }else{ thread->free_node(node); } }else{ exit_scheduler = handle_slow_messages(type,node,thread,always); } } } return (exit_scheduler); } template inline bool CNodeGenerator::do_work(CGenNode * node, CFlowGenListPerThread * thread, dsec_t d_time, bool always ){ /* template filter in compile time */ if ( SCH_MODE == smSTATELESS ) { return ( do_work_stl(node,thread,always) ); }else{ /* smSTATEFUL */ return ( do_work_both(node,thread,d_time,always) ); } } inline void CNodeGenerator::do_sleep(dsec_t & cur_time, CFlowGenListPerThread * thread, dsec_t n_time){ thread->m_cpu_dp_u.commit1(); dsec_t dt; /* TBD make this better using calculation, minimum now_sec() */ while ( true ) { cur_time = now_sec(); dt = cur_time - n_time ; if (dt> WAIT_WINDOW_SIZE ) { break; } rte_pause(); } thread->m_cpu_dp_u.start_work1(); } inline int CNodeGenerator::teardown(CFlowGenListPerThread * thread, bool always, double &old_offset, double offset){ thread->m_cpu_dp_u.commit1(); /* to do */ if ( thread->is_terminated_by_master() ) { return (0); } if (!always) { old_offset =offset; }else{ // free the left other thread->handler_defer_job_flush(); } return (0); } template inline int CNodeGenerator::flush_file_realtime(dsec_t max_time, dsec_t d_time, bool always, CFlowGenListPerThread * thread, double &old_offset){ CGenNode * node; dsec_t offset=0.0; dsec_t cur_time; dsec_t n_time; if (always) { offset=old_offset; }else{ add_exit_node(thread,max_time); } thread->m_cpu_dp_u.start_work1(); sch_state_t state = scINIT; node = m_p_queue.top(); n_time = node->m_time + offset; cur_time = now_sec(); while (state!=scTERMINATE) { switch (state) { case scINIT: cur_time = now_sec(); { dsec_t dt = cur_time - n_time ; if (dt>0) { state=scWORK; if (dt > BURST_OFFSET_DTIME) { offset += dt; } }else{ state=scWAIT; } } ; break; case scWORK: do { bool s=do_work(node,thread,d_time,always); if (s) { // can we remove this IF ? state=scTERMINATE; break; } node = m_p_queue.top(); n_time = node->m_time + offset; if ((n_time-cur_time)>EAT_WINDOW_DTIME) { state=scINIT; break; } } while ( true ); break; case scWAIT: do_sleep(cur_time,thread,n_time); // estimate loop state=scWORK; break; default: assert(0); } /* switch */ }/* while*/ return (teardown(thread,always,old_offset,offset)); } FORCE_NO_INLINE int CNodeGenerator::flush_file_sim(dsec_t max_time, dsec_t d_time, bool always, CFlowGenListPerThread * thread, double &old_offset){ CGenNode * node; if (!always) { add_exit_node(thread,max_time); } while (true) { node = m_p_queue.top(); bool do_exit; if ( get_is_stateless() ) { do_exit=do_work(node,thread,d_time,always); }else{ do_exit=do_work(node,thread,d_time,always); } if ( do_exit ){ break; } } return (teardown(thread,always,old_offset,0)); } int CNodeGenerator::flush_file(dsec_t max_time, dsec_t d_time, bool always, CFlowGenListPerThread * thread, double &old_offset){ #ifdef TREX_SIM return ( flush_file_sim(max_time, d_time,always,thread,old_offset) ); #else if ( get_is_stateless() ) { return ( flush_file_realtime(max_time, d_time,always,thread,old_offset) ); }else{ return ( flush_file_realtime(max_time, d_time,always,thread,old_offset) ); } #endif } void CNodeGenerator::handle_flow_pkt(CGenNode *node, CFlowGenListPerThread *thread) { /*repeat and NAT is not supported */ if ( node->is_nat_first_state() ) { node->set_nat_wait_state(); flush_one_node_to_file(node); #ifdef _DEBUG update_stats(node); #endif } else { if ( node->is_nat_wait_state() ) { if (node->is_responder_pkt()) { m_p_queue.pop(); /* time out, need to free the flow and remove the association , we didn't get convertion yet*/ thread->terminate_nat_flows(node); return; } else { flush_one_node_to_file(node); #ifdef _DEBUG update_stats(node); #endif } } else { assert(0); } } m_p_queue.pop(); if ( node->is_last_in_flow() ) { thread->free_last_flow_node( node); } else { node->update_next_pkt_in_flow(); m_p_queue.push(node); } } void CNodeGenerator::handle_flow_sync(CGenNode *node, CFlowGenListPerThread *thread, bool &exit_scheduler) { /* tickle the watchdog */ thread->tickle(); /* flow sync message is a sync point for time */ thread->m_cur_time_sec = node->m_time; /* first pop the node */ m_p_queue.pop(); thread->check_msgs(); /* check messages */ m_v_if->flush_tx_queue(); /* flush pkt each timeout */ /* exit in case this is the last node*/ if ( m_p_queue.size() == m_parent->m_non_active_nodes ) { thread->free_node(node); exit_scheduler = true; } else { /* schedule for next maintenace */ node->m_time += SYNC_TIME_OUT; m_p_queue.push(node); } } void CNodeGenerator::handle_command(CGenNode *node, CFlowGenListPerThread *thread, bool &exit_scheduler) { m_p_queue.pop(); CGenNodeCommand *node_cmd = (CGenNodeCommand *)node; TrexStatelessCpToDpMsgBase * cmd=node_cmd->m_cmd; cmd->handle(&thread->m_stateless_dp_info); exit_scheduler = cmd->is_quit(); thread->free_node((CGenNode *)node_cmd);/* free the node */ } void CNodeGenerator::handle_pcap_pkt(CGenNode *node, CFlowGenListPerThread *thread) { m_p_queue.pop(); CGenNodePCAP *node_pcap = (CGenNodePCAP *)node; /* might have been marked for free */ if ( unlikely( node_pcap->is_marked_for_free() ) ) { thread->free_node(node); } else { node_pcap->handle(thread); } } bool CNodeGenerator::handle_slow_messages(uint8_t type, CGenNode * node, CFlowGenListPerThread * thread, bool always){ /* should we continue after */ bool exit_scheduler = false; switch (type) { case CGenNode::PCAP_PKT: handle_pcap_pkt(node, thread); break; case CGenNode::FLOW_DEFER_PORT_RELEASE: m_p_queue.pop(); thread->handler_defer_job(node); thread->free_node(node); break; case CGenNode::FLOW_PKT_NAT: handle_flow_pkt(node, thread); break; case CGenNode::FLOW_SYNC: handle_flow_sync(node, thread, exit_scheduler); break; case CGenNode::EXIT_SCHED: m_p_queue.pop(); thread->free_node(node); exit_scheduler = true; break; case CGenNode::COMMAND: handle_command(node, thread, exit_scheduler); break; default: assert(0); } return (exit_scheduler); } void CFlowGenListPerThread::Dump(FILE *fd){ fprintf(fd,"yaml info "); m_yaml_info.Dump(fd); fprintf(fd,"\n"); fprintf(fd,"cap file info"); int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRecPerThread * lp=m_cap_gen[i]; lp->Dump(stdout); } } void CFlowGenListPerThread::DumpStats(FILE *fd){ m_stats.dump(fd); } void CFlowGenListPerThread::DumpCsv(FILE *fd){ CFlowStats::DumpHeader(fd); CFlowStats stats; CFlowStats sum; int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRecPerThread * lp=m_cap_gen[i]; lp->getFlowStats(&stats); stats.Dump(fd); sum.Add(stats); } fprintf(fd,"\n"); sum.m_name= "sum"; sum.Dump(fd); } uint32_t CFlowGenListPerThread::getDualPortId(){ return ( ::getDualPortId(m_thread_id) ); } double CFlowGenListPerThread::get_longest_flow(uint8_t pool_idx, bool is_client){ int i; double longest_flow = 0.0; for (i=0;i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRecPerThread * lp=m_cap_gen[i]; if (is_client && lp->m_info->m_client_pool_idx != pool_idx) continue; if (!is_client && lp->m_info->m_server_pool_idx != pool_idx) continue; double tmp_len; tmp_len = lp->m_flow_info->get_cap_file_length_sec(); if (longest_flow < tmp_len ) { longest_flow = tmp_len; } } return longest_flow; } double CFlowGenListPerThread::get_longest_flow(){ int i; double longest_flow = 0.0; for (i=0;i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRecPerThread * lp=m_cap_gen[i]; double tmp_len; tmp_len = lp->m_flow_info->get_cap_file_length_sec(); if (longest_flow < tmp_len ) { longest_flow = tmp_len; } } return longest_flow; } double CFlowGenListPerThread::get_total_kcps(uint8_t pool_idx, bool is_client){ int i; double total=0.0; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRecPerThread * lp=m_cap_gen[i]; if (is_client && lp->m_info->m_client_pool_idx != pool_idx) continue; if (!is_client && lp->m_info->m_server_pool_idx != pool_idx) continue; total +=lp->m_info->m_k_cps; } return (total); } double CFlowGenListPerThread::get_total_kcps(){ int i; double total=0.0; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRecPerThread * lp=m_cap_gen[i]; total +=lp->m_info->m_k_cps; } return (total); } double CFlowGenListPerThread::get_delta_flow_is_sec(){ return (1.0/(1000.0*get_total_kcps())); } void CFlowGenListPerThread::inc_current_template(void){ m_cur_template++; if (m_cur_template == m_cap_gen.size()) { m_cur_template=0; } } int CFlowGenListPerThread::generate_flows_roundrobin(bool *done){ // round robin CFlowGeneratorRecPerThread * cur; bool found=false; // try current int i; *done = true; for (i=0;i<(int)m_cap_gen.size();i++ ) { cur=m_cap_gen[m_cur_template]; if (!(cur->m_info->m_limit_was_set) || (cur->m_info->m_flowcnt < cur->m_info->m_limit)) { *done = false; if ( cur->m_policer.update(1.0,m_cur_time_sec) ){ cur->m_info->m_flowcnt++; found=true; break; } } inc_current_template(); } if (found) { /* generate the flow into the generator*/ CGenNode * node= create_node() ; cur->generate_flow(&m_node_gen,m_cur_time_sec,m_cur_flow_id,node); m_cur_flow_id++; /* this is estimation */ m_stats.m_total_open_flows += cur->m_flow_info->get_total_flows(); m_stats.m_total_bytes += cur->m_flow_info->get_total_bytes(); m_stats.m_total_pkt += cur->m_flow_info->Size(); inc_current_template(); } return (0); } int CFlowGenListPerThread::reschedule_flow(CGenNode *node){ // Re-schedule the node node->reset_pkt_in_flow(); node->m_time += node->m_template_info->m_restart_time; m_node_gen.add_node(node); m_stats.m_total_bytes += node->m_flow_info->get_total_bytes(); m_stats.m_total_pkt += node->m_flow_info->Size(); return (0); } void CFlowGenListPerThread::terminate_nat_flows(CGenNode *p){ m_stats.m_nat_flow_timeout++; m_stats.m_nat_lookup_remove_flow_id++; m_flow_id_to_node_lookup.remove_no_lookup(p->get_short_fid()); free_last_flow_node( p); } void CFlowGenListPerThread::handle_latency_pkt_msg(CGenNodeLatencyPktInfo * msg){ /* send the packet */ #ifdef RX_QUEUE_TRACE_ printf(" latency msg dir %d\n",msg->m_dir); struct rte_mbuf * m; m=msg->m_pkt; rte_pktmbuf_dump(stdout,m, rte_pktmbuf_pkt_len(m)); #endif /* update timestamp */ struct rte_mbuf * m; m=msg->m_pkt; uint8_t *p=rte_pktmbuf_mtod(m, uint8_t*); latency_header * h=(latency_header *)(p+msg->m_latency_offset); h->time_stamp = os_get_hr_tick_64(); m_node_gen.m_v_if->send_one_pkt((pkt_dir_t)msg->m_dir,msg->m_pkt); } void CFlowGenListPerThread::handle_nat_msg(CGenNodeNatInfo * msg){ int i; for (i=0; im_cnt; i++) { CNatFlowInfo * nat_msg=&msg->m_data[i]; CGenNode * node=m_flow_id_to_node_lookup.lookup(nat_msg->m_fid); if (!node) { /* this should be move to a notification module */ #ifdef NAT_TRACE_ printf(" ERORR not valid flow_id %d probably flow was aged \n",nat_msg->m_fid); #endif m_stats.m_nat_lookup_no_flow_id++; continue; } #ifdef NAT_TRACE_ printf(" %.03f RX :set node %p:%x %x:%x:%x \n",now_sec() ,node,nat_msg->m_fid,nat_msg->m_external_ip,nat_msg->m_external_ip_server,nat_msg->m_external_port); #endif node->set_nat_ipv4_addr(nat_msg->m_external_ip); node->set_nat_ipv4_port(nat_msg->m_external_port); node->set_nat_ipv4_addr_server(nat_msg->m_external_ip_server); assert(node->is_nat_wait_state()); if ( CGlobalInfo::is_learn_verify_mode() ){ if (!node->is_external_is_eq_to_internal_ip() ){ m_stats.m_nat_flow_learn_error++; } } node->set_nat_learn_state(); /* remove from the hash */ m_flow_id_to_node_lookup.remove_no_lookup(nat_msg->m_fid); m_stats.m_nat_lookup_remove_flow_id++; } } void CFlowGenListPerThread::check_msgs(void) { /* inlined for performance */ m_stateless_dp_info.periodic_check_for_cp_messages(); if ( likely ( m_ring_from_rx->isEmpty() ) ) { return; } #ifdef NAT_TRACE_ printf(" %.03f got message from RX \n",now_sec()); #endif while ( true ) { CGenNode * node; if ( m_ring_from_rx->Dequeue(node)!=0 ){ break; } assert(node); //printf ( " message: thread %d, node->m_flow_id : %d \n", m_thread_id,node->m_flow_id); /* only one message is supported right now */ CGenNodeMsgBase * msg=(CGenNodeMsgBase *)node; uint8_t msg_type = msg->m_msg_type; switch (msg_type ) { case CGenNodeMsgBase::NAT_FIRST: handle_nat_msg((CGenNodeNatInfo * )msg); break; case CGenNodeMsgBase::LATENCY_PKT: handle_latency_pkt_msg((CGenNodeLatencyPktInfo *) msg); break; default: printf("ERROR pkt-thread message type is not valid %d \n",msg_type); assert(0); } CGlobalInfo::free_node(node); } } void CFlowGenListPerThread::start_stateless_simulation_file(std::string erf_file_name, CPreviewMode &preview, uint64_t limit){ m_preview_mode = preview; m_node_gen.open_file(erf_file_name,&m_preview_mode); m_node_gen.set_packet_limit(limit); } void CFlowGenListPerThread::stop_stateless_simulation_file(){ m_node_gen.m_v_if->close_file(); } void CFlowGenListPerThread::start_stateless_daemon_simulation(){ m_cur_time_sec = 0; /* if no pending CP messages - the core will simply be stuck forever */ if (m_stateless_dp_info.are_any_pending_cp_messages()) { m_stateless_dp_info.run_once(); } } /* return true if we need to shedule next_stream, */ bool CFlowGenListPerThread::set_stateless_next_node( CGenNodeStateless * cur_node, CGenNodeStateless * next_node){ return ( m_stateless_dp_info.set_stateless_next_node(cur_node,next_node) ); } void CFlowGenListPerThread::start_stateless_daemon(CPreviewMode &preview){ m_cur_time_sec = 0; /* set per thread global info, for performance */ m_preview_mode = preview; m_node_gen.open_file("",&m_preview_mode); m_stateless_dp_info.start(); } void CFlowGenListPerThread::start_generate_stateful(std::string erf_file_name, CPreviewMode & preview){ /* now we are ready to generate*/ if ( m_cap_gen.size()==0 ){ fprintf(stderr," nothing to generate no template loaded \n"); return; } m_preview_mode = preview; m_node_gen.open_file(erf_file_name,&m_preview_mode); dsec_t d_time_flow=get_delta_flow_is_sec(); m_cur_time_sec = 0.01+m_thread_id*m_flow_list->get_delta_flow_is_sec(); if ( CGlobalInfo::is_realtime() ){ m_cur_time_sec += now_sec() + 0.5 ; } dsec_t c_stop_sec = m_cur_time_sec + m_yaml_info.m_duration_sec; m_stop_time_sec =c_stop_sec; m_cur_flow_id =1; m_cur_template =(m_thread_id % m_cap_gen.size()); m_stats.clear(); fprintf(stdout," Generating erf file ... \n"); CGenNode * node= create_node() ; /* add periodic */ node->m_type = CGenNode::FLOW_FIF; node->m_time = m_cur_time_sec; m_node_gen.add_node(node); double old_offset=0.0; node= create_node() ; node->m_type = CGenNode::FLOW_SYNC; node->m_time = m_cur_time_sec + SYNC_TIME_OUT ; m_node_gen.add_node(node); #ifdef _DEBUG if ( m_preview_mode.getVMode() >2 ){ CGenNode::DumpHeader(stdout); } #endif m_node_gen.flush_file(c_stop_sec,d_time_flow, false,this,old_offset); #ifdef VALG CALLGRIND_STOP_INSTRUMENTATION; printf (" %llu \n",os_get_hr_tick_64()-_start_time); #endif if ( !CGlobalInfo::m_options.preview.getNoCleanFlowClose() && (is_terminated_by_master()==false) ){ /* clean close */ m_node_gen.flush_file(m_cur_time_sec, d_time_flow, true,this,old_offset); } if (m_preview_mode.getVMode() > 1 ) { fprintf(stdout,"\n\n"); fprintf(stdout,"\n\n"); fprintf(stdout,"file stats \n"); fprintf(stdout,"=================\n"); m_stats.dump(stdout); } m_node_gen.close_file(this); } void CFlowGenList::Delete(){ clean_p_thread_info(); Clean(); if (CPluginCallback::callback) { delete CPluginCallback::callback; CPluginCallback::callback = NULL; } } bool CFlowGenList::Create(){ check_objects_sizes(); CPluginCallback::callback= new CPluginCallbackSimple(); return (true); } void CFlowGenList::generate_p_thread_info(uint32_t num_threads){ clean_p_thread_info(); BP_ASSERT(num_threads < 64); int i; for (i=0; i<(int)num_threads; i++) { CFlowGenListPerThread * lp= new CFlowGenListPerThread(); lp->Create(i,i,this,num_threads); m_threads_info.push_back(lp); } } void CFlowGenList::clean_p_thread_info(void){ int i; for (i=0; i<(int)m_threads_info.size(); i++) { CFlowGenListPerThread * lp=m_threads_info[i]; lp->Delete(); delete lp; } m_threads_info.clear(); } int CFlowGenList::load_from_mac_file(std::string file_name) { if ( !utl_is_file_exists (file_name) ){ printf(" ERROR no mac_file is set, file %s does not exist \n",file_name.c_str()); exit(-1); } m_mac_info.set_configured(true); try { std::ifstream fin((char *)file_name.c_str()); YAML::Parser parser(fin); YAML::Node doc; parser.GetNextDocument(doc); doc[0] >> m_mac_info.get_mac_info(); } catch ( const std::exception& e ) { std::cout << e.what() << "\n"; m_mac_info.clear(); exit(-1); } return (0); } int CFlowGenList::load_from_yaml(std::string file_name, uint32_t num_threads){ uint8_t idx; m_yaml_info.load_from_yaml_file(file_name); if (m_yaml_info.verify_correctness(num_threads) ==false){ exit(0); } /* move it to global info, better CPU D-cache usage */ CGlobalInfo::m_options.preview.set_vlan_mode_enable(m_yaml_info.m_vlan_info.m_enable); CGlobalInfo::m_options.m_vlan_port[0] = m_yaml_info.m_vlan_info.m_vlan_per_port[0]; CGlobalInfo::m_options.m_vlan_port[1] = m_yaml_info.m_vlan_info.m_vlan_per_port[1]; CGlobalInfo::m_options.preview.set_mac_ip_overide_enable(m_yaml_info.m_mac_replace_by_ip); if ( m_yaml_info.m_mac_base.size() != 6 ){ printf(" mac addr is not valid \n"); exit(0); } if (m_yaml_info.m_ipv6_set == true) { // Copy the most significant 96-bits from yaml data for (idx=0; idx<6; idx++){ CGlobalInfo::m_options.m_src_ipv6[idx] = m_yaml_info.m_src_ipv6[idx]; CGlobalInfo::m_options.m_dst_ipv6[idx] = m_yaml_info.m_dst_ipv6[idx]; } }else{ // Set the most signifcant 96-bits to zero which represents an // IPv4-compatible IPv6 address for (idx=0; idx<6; idx++){ CGlobalInfo::m_options.m_src_ipv6[idx] = 0; CGlobalInfo::m_options.m_dst_ipv6[idx] = 0; } } int i=0; Clean(); bool all_template_has_one_direction=true; for (i=0; i<(int)m_yaml_info.m_vec.size(); i++) { CFlowGeneratorRec * lp=new CFlowGeneratorRec(); if ( lp->Create(&m_yaml_info.m_vec[i],&m_yaml_info,i) == false){ fprintf(stdout,"\n ERROR reading YAML template files, please verify that they are valid \n\n"); exit(-1); return (-1); } m_cap_gen.push_back(lp); if (lp->m_flow_info.GetPacket(0)->m_pkt_indication.m_desc.IsBiDirectionalFlow() ) { all_template_has_one_direction=false; } } if ( CGlobalInfo::is_learn_mode() && all_template_has_one_direction ) { fprintf(stdout,"\n Warning --learn mode has nothing to do when all templates are one directional, please remove it \n"); } return (0); } void CFlowGenList::Clean(){ int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRec * lp=m_cap_gen[i]; lp->Delete(); delete lp; } m_cap_gen.clear(); } double CFlowGenList::GetCpuUtil(){ int i; double c=0.0; for (i=0; i<(int)m_threads_info.size(); i++) { CFlowGenListPerThread * lp=m_threads_info[i]; c+=lp->m_cpu_cp_u.GetVal(); } return (c/m_threads_info.size()); } void CFlowGenList::UpdateFast(){ int i; for (i=0; i<(int)m_threads_info.size(); i++) { CFlowGenListPerThread * lp=m_threads_info[i]; lp->Update(); } } void CFlowGenList::Dump(FILE *fd){ fprintf(fd,"yaml info \n"); fprintf(fd,"--------------\n"); m_yaml_info.Dump(fd); fprintf(fd,"\n"); fprintf(fd,"cap file info \n"); fprintf(fd,"----------------------\n"); int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRec * lp=m_cap_gen[i]; lp->Dump(stdout); } } void CFlowGenList::DumpPktSize(){ int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRec * lp=m_cap_gen[i]; lp->m_flow_info.dump_pkt_sizes(); } } void CFlowGenList::DumpCsv(FILE *fd){ CFlowStats::DumpHeader(fd); CFlowStats stats; CFlowStats sum; int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRec * lp=m_cap_gen[i]; lp->getFlowStats(&stats); stats.Dump(fd); sum.Add(stats); } fprintf(fd,"\n"); sum.m_name= "sum"; sum.Dump(fd); sum.m_memory.dump(fd); } uint32_t CFlowGenList::get_total_repeat_flows(){ uint32_t flows=0; int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRec * lp=m_cap_gen[i]; flows+=lp->m_info->m_limit ; } return (flows); } double CFlowGenList::get_total_tx_bps(){ CFlowStats stats; double total=0.0; int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRec * lp=m_cap_gen[i]; lp->getFlowStats(&stats); total+=(stats.m_mb_sec); } return (_1Mb_DOUBLE*total); } double CFlowGenList::get_total_pps(){ CFlowStats stats; double total=0.0; int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRec * lp=m_cap_gen[i]; lp->getFlowStats(&stats); total+=stats.m_pps; } return (total); } double CFlowGenList::get_total_kcps(){ CFlowStats stats; double total=0.0; int i; for (i=0; i<(int)m_cap_gen.size(); i++) { CFlowGeneratorRec * lp=m_cap_gen[i]; lp->getFlowStats(&stats); total+= stats.get_normal_cps(); } return ((total/1000.0)); } double CFlowGenList::get_delta_flow_is_sec(){ return (1.0/(1000.0*get_total_kcps())); } bool CPolicer::update(double dsize,double now_sec){ if ( m_last_time ==0.0 ) { /* first time */ m_last_time = now_sec; return (true); } if (m_cir == 0.0) { return (false); } // check if there is a need to add tokens if(now_sec > m_last_time) { dsec_t dtime=(now_sec - m_last_time); dsec_t dsize =dtime*m_cir; m_level +=dsize; if (m_level > m_bucket_size) { m_level = m_bucket_size; } m_last_time = now_sec; } if (m_level > dsize) { m_level -= dsize; return (true); }else{ return (false); } } float CPPSMeasure::add(uint64_t pkts){ if ( false == m_start ){ m_start=true; m_last_time_msec = os_get_time_msec() ; m_last_pkts=pkts; return (0.0); } uint32_t ctime=os_get_time_msec(); if ((ctime - m_last_time_msec) max) { std::cerr << "Value " << val << " for option " << opt_name << " is out of range. Should be (" << min << "-" << max << ")." << std::endl; return false; } return true; } void CParserOption::dump(FILE *fd){ preview.Dump(fd); fprintf(fd," cfg file : %s \n",cfg_file.c_str()); fprintf(fd," mac file : %s \n",mac_file.c_str()); fprintf(fd," out file : %s \n",out_file.c_str()); fprintf(fd," duration : %.0f \n",m_duration); fprintf(fd," factor : %.0f \n",m_factor); fprintf(fd," mbuf_factor : %.0f \n",m_mbuf_factor); fprintf(fd," latency : %d pkt/sec \n",m_latency_rate); fprintf(fd," zmq_port : %d \n",m_zmq_port); fprintf(fd," telnet_port : %d \n",m_telnet_port); fprintf(fd," expected_ports : %d \n",m_expected_portd); if (preview.get_vlan_mode_enable() ) { fprintf(fd," vlans : [%d,%d] \n",m_vlan_port[0],m_vlan_port[1]); } fprintf(fd," mac spreading: %d \n",(int)m_mac_splitter); int i; for (i = 0; i < TREX_MAX_PORTS; i++) { fprintf(fd," port : %d dst:",i); CMacAddrCfg * lp=&m_mac_addr[i]; dump_mac_addr(fd,lp->u.m_mac.dest); fprintf(fd," src:"); dump_mac_addr(fd,lp->u.m_mac.src); fprintf(fd,"\n"); } } #if 0 void CTupleGlobalGenerator::Dump(FILE *fd){ fprintf(fd," src:%x dest: %x \n",m_result_src_ip,m_result_dest_ip); } bool CTupleGlobalGenerator::Create(){ was_generated=false; return (true); } void CTupleGlobalGenerator::Copy(CTupleGlobalGenerator * gen){ was_generated=false; m_min_src_ip = gen->m_min_src_ip; m_max_src_ip = gen->m_max_src_ip; m_min_dest_ip = gen->m_min_dest_ip; m_max_dest_ip = gen->m_max_dest_ip; } void CTupleGlobalGenerator::Delete(){ was_generated=false; } #endif static uint32_t get_rand_32(uint32_t MinimumRange , uint32_t MaximumRange ); #if 0 void CTupleGlobalGenerator::Generate(uint32_t thread_id, uint32_t num_addr ){ if ( was_generated == false) { /* first time */ was_generated = true; cur_src_ip = m_min_src_ip; cur_dst_ip = m_min_dest_ip; } if ( ( cur_src_ip + num_addr ) > m_max_src_ip ) { cur_src_ip = m_min_src_ip; } /* copy the results */ m_result_src_ip = cur_src_ip; m_result_dest_ip = cur_dst_ip; cur_src_ip += num_addr; cur_dst_ip += 1; if (cur_dst_ip > m_max_dest_ip ) { cur_dst_ip = m_min_dest_ip; } } void CTupleTemplateGenerator::Dump(FILE *fd){ fprintf(fd," id: %x, %x:%x - %x \n",m_id,m_result_src_ip,m_result_dest_ip,m_result_src_port); } bool CTupleTemplateGenerator::Create(CTupleGlobalGenerator * global_gen, uint16_t w, uint16_t wlength, uint32_t _id, uint32_t thread_id){ m_was_generated = false; m_thread_id = thread_id; m_lp_global_gen = global_gen; BP_ASSERT(m_lp_global_gen); m_cur_src_port = 1; m_cur_src_port_cnt=0; m_w = w; m_wlength = wlength; m_id = _id; m_was_init=true; return(true); } void CTupleTemplateGenerator::Delete(){ m_was_generated = false; m_was_init=false; } void CTupleTemplateGenerator::Generate_src_dest(){ /* TBD need to fix the 100*/ m_lp_global_gen->Generate(m_thread_id,m_wlength); m_result_src_ip = m_lp_global_gen->m_result_src_ip; m_dest_ip = m_lp_global_gen->m_result_dest_ip; m_result_dest_ip = update_dest_ip(m_dest_ip ); m_cnt=0; } uint16_t CTupleTemplateGenerator::GenerateOneSourcePort(){ /* handle port */ m_cur_src_port++; /* do not use port zero */ if (m_cur_src_port == 0) { m_cur_src_port=1; } m_result_src_port=m_cur_src_port; return (m_cur_src_port); } void CTupleTemplateGenerator::Generate(){ BP_ASSERT(m_was_init); if ( m_was_generated == false ) { /* first time */ Generate_src_dest(); m_was_generated = true; }else{ /* ip+cnt,dest+cnt*/ m_cnt++; if ( m_cnt >= m_wlength ) { m_cnt =0; m_result_src_ip -=m_wlength; m_result_dest_ip = m_dest_ip; m_cur_src_port_cnt++; if (m_cur_src_port_cnt >= m_w ) { Generate_src_dest(); m_cur_src_port_cnt=0; } } m_result_src_ip += 1; m_result_dest_ip = update_dest_ip(m_dest_ip +m_cnt ); } /* handle port */ m_cur_src_port++; /* do not use port zero */ if (m_cur_src_port == 0) { m_cur_src_port=1; } m_result_src_ip =update_src_ip( m_result_src_ip ); m_result_src_port=m_cur_src_port; } #endif static uint32_t get_rand_32(uint32_t MinimumRange, uint32_t MaximumRange) __attribute__ ((unused)); static uint32_t get_rand_32(uint32_t MinimumRange, uint32_t MaximumRange) { enum {RANDS_NUM = 2 , RAND_MAX_BITS = 0xf , UNSIGNED_INT_BITS = 0x20 , TWO_BITS_MASK = 0x3}; const double TWO_POWER_32_BITS = 0x10000000 * (double)0x10; uint32_t RandomNumber = 0; for (int i = 0 ; i < RANDS_NUM;i++) { RandomNumber = (RandomNumber<m_pkt_info; rte_mbuf_t * buf=lp->generate_new_mbuf(node); //rte_pktmbuf_dump(buf, buf->pkt_len); //sending it ?? // free it here as if driver does rte_pktmbuf_free(buf); #endif return (0); } void CErfIF::fill_raw_packet(rte_mbuf_t * m,CGenNode * node,pkt_dir_t dir){ fill_pkt(m_raw,m); CPktNsecTimeStamp t_c(node->m_time); m_raw->time_nsec = t_c.m_time_nsec; m_raw->time_sec = t_c.m_time_sec; uint8_t p_id = (uint8_t)dir; m_raw->setInterface(p_id); } pkt_dir_t CErfIFStl::port_id_to_dir(uint8_t port_id) { return ((pkt_dir_t)(port_id&1)); } int CErfIFStl::update_mac_addr_from_global_cfg(pkt_dir_t dir, uint8_t * p){ memcpy(p,CGlobalInfo::m_options.get_dst_src_mac_addr(dir),12); return (0); } int CErfIFStl::send_sl_node(CGenNodeStateless *node_sl) { pkt_dir_t dir=(pkt_dir_t)node_sl->get_mbuf_cache_dir(); rte_mbuf_t * m; if ( likely(node_sl->is_cache_mbuf_array()) ) { m=node_sl->cache_mbuf_array_get_cur(); fill_raw_packet(m,(CGenNode *)node_sl,dir); }else{ m=node_sl->get_cache_mbuf(); bool is_const = false; if (m) { is_const = true; rte_pktmbuf_refcnt_update(m,1); }else{ m=node_sl->alloc_node_with_vm(); assert(m); } if (node_sl->is_stat_needed() && (node_sl->get_stat_hw_id() >= MAX_FLOW_STATS) ) { /* latency packet. flow stat without latency handled like normal packet in simulation */ uint16_t hw_id = node_sl->get_stat_hw_id(); rte_mbuf_t *mi; struct flow_stat_payload_header *fsp_head; mi = node_sl->alloc_flow_stat_mbuf(m, fsp_head, is_const); fsp_head->seq = 0x12345678; fsp_head->hw_id = hw_id - MAX_FLOW_STATS; fsp_head->magic = FLOW_STAT_PAYLOAD_MAGIC; fsp_head->time_stamp = 0x8899aabbccddeeff; fill_raw_packet(mi, (CGenNode *)node_sl, dir); rte_pktmbuf_free(mi); } else { fill_raw_packet(m,(CGenNode *)node_sl,dir); rte_pktmbuf_free(m); } } /* check that we have mbuf */ int rc = write_pkt(m_raw); BP_ASSERT(rc == 0); return (rc); } int CErfIFStl::send_pcap_node(CGenNodePCAP *pcap_node) { rte_mbuf_t *m = pcap_node->get_pkt(); if (!m) { return (-1); } pkt_dir_t dir = (pkt_dir_t)pcap_node->get_mbuf_dir(); fill_raw_packet(m, (CGenNode*)pcap_node, dir); rte_pktmbuf_free(m); int rc = write_pkt(m_raw); BP_ASSERT(rc == 0); return (rc); } /* * This is the simulation stateless send_node. * in simulation (bp-sim-64) it is called instead of CCoreEthIFStateless::send_node * Purpose is to test the mbuf manipulation functions which are the same in simulation and "real" code */ int CErfIFStl::send_node(CGenNode * _no_to_use){ if ( m_preview_mode->getFileWrite() ) { switch (_no_to_use->m_type) { case CGenNode::STATELESS_PKT: return send_sl_node((CGenNodeStateless *) _no_to_use); case CGenNode::PCAP_PKT: return send_pcap_node((CGenNodePCAP *) _no_to_use); default: assert(0); } } return (0); } int CErfIF::send_node(CGenNode * node){ if ( m_preview_mode->getFileWrite() ){ CFlowPktInfo * lp=node->m_pkt_info; rte_mbuf_t * m=lp->generate_new_mbuf(node); pkt_dir_t dir=node->cur_interface_dir(); fill_raw_packet(m,node,dir); /* update mac addr dest/src 12 bytes */ uint8_t *p=(uint8_t *)m_raw->raw; int p_id=(int)dir; memcpy(p,CGlobalInfo::m_options.get_dst_src_mac_addr(p_id),12); /* If vlan is enabled, add vlan header */ if ( unlikely( CGlobalInfo::m_options.preview.get_vlan_mode_enable() ) ){ /* retrieve vlan ID and form vlan tag */ uint8_t vlan_port = (node->m_src_ip &1); uint16_t vlan_protocol = EthernetHeader::Protocol::VLAN; uint16_t vlan_id = CGlobalInfo::m_options.m_vlan_port[vlan_port]; uint32_t vlan_tag = (vlan_protocol << 16) | vlan_id; vlan_tag = PKT_HTONL(vlan_tag); /* insert vlan tag and adjust packet size */ memcpy(cbuff+4, p+12, m_raw->pkt_len-12); memcpy(cbuff, &vlan_tag, 4); memcpy(p+12, cbuff, m_raw->pkt_len-8); m_raw->pkt_len += 4; } //utl_DumpBuffer(stdout,p, 12,0); int rc = write_pkt(m_raw); BP_ASSERT(rc == 0); rte_pktmbuf_free(m); } return (0); } int CErfIF::flush_tx_queue(void){ return (0); } void CTcpSeq::update(uint8_t *p, CFlowPktInfo *pkt_info, int16_t s_size){ TCPHeader *tcp= (TCPHeader *)(p+pkt_info->m_pkt_indication.getFastTcpOffset()); uint32_t seqnum, acknum; // This routine will adjust the TCP segment size for packets // based on the modifications made by the plugins. // Basically it will keep track of the size changes // and adjust the TCP sequence numbers accordingly. bool is_init=pkt_info->m_pkt_indication.m_desc.IsInitSide(); // Update TCP seq number seqnum = tcp->getSeqNumber(); acknum = tcp->getAckNumber(); if (is_init) { // Packet is from client seqnum += client_seq_delta; acknum += server_seq_delta; } else { // Packet is from server seqnum += server_seq_delta; acknum += client_seq_delta; } tcp->setSeqNumber(seqnum); tcp->setAckNumber(acknum); // Adjust delta being tracked if (is_init) { client_seq_delta += s_size; } else { server_seq_delta += s_size; } } void on_node_first(uint8_t plugin_id,CGenNode * node, CFlowYamlInfo * template_info, CTupleTemplateGeneratorSmart * tuple_gen, CFlowGenListPerThread * flow_gen){ if (CPluginCallback::callback) { CPluginCallback::callback->on_node_first(plugin_id,node,template_info, tuple_gen,flow_gen); } } void on_node_last(uint8_t plugin_id,CGenNode * node){ if (CPluginCallback::callback) { CPluginCallback::callback->on_node_last(plugin_id,node); } } rte_mbuf_t * on_node_generate_mbuf(uint8_t plugin_id,CGenNode * node,CFlowPktInfo * pkt_info){ rte_mbuf_t * m; assert(CPluginCallback::callback); m=CPluginCallback::callback->on_node_generate_mbuf(plugin_id,node,pkt_info); assert(m); return(m); } class CPlugin_rtsp : public CTcpSeq { public: void * m_gen; uint16_t rtp_client_0; uint16_t rtp_client_1; }; void CPluginCallbackSimple::on_node_first(uint8_t plugin_id, CGenNode * node, CFlowYamlInfo * template_info, CTupleTemplateGeneratorSmart * tuple_gen, CFlowGenListPerThread * flow_gen ){ //printf(" on on_node_first callback %d node %x! \n",(int)plugin_id,node); /* generate 2 ports from client side */ if ( (plugin_id == mpRTSP) || (plugin_id == mpSIP_VOICE) ) { CPlugin_rtsp * lpP=new CPlugin_rtsp(); assert(lpP); /* TBD need to be fixed using new API */ lpP->rtp_client_0 = tuple_gen->GenerateOneSourcePort(); lpP->rtp_client_1 = tuple_gen->GenerateOneSourcePort(); lpP->m_gen=flow_gen; node->m_plugin_info = (void *)lpP; }else{ if (plugin_id ==mpDYN_PYLOAD) { /* nothing to do */ }else{ if (plugin_id ==mpAVL_HTTP_BROWSIN) { CTcpSeq * lpP=new CTcpSeq(); assert(lpP); node->m_plugin_info = (void *)lpP; }else{ /* do not support this */ assert(0); } } } } void CPluginCallbackSimple::on_node_last(uint8_t plugin_id,CGenNode * node){ //printf(" on on_node_last callback %d %x! \n",(int)plugin_id,node); if ( (plugin_id == mpRTSP) || (plugin_id == mpSIP_VOICE) ) { CPlugin_rtsp * lpP=(CPlugin_rtsp * )node->m_plugin_info; /* free the ports */ CFlowGenListPerThread * flow_gen=(CFlowGenListPerThread *) lpP->m_gen; bool is_tcp=node->m_pkt_info->m_pkt_indication.m_desc.IsTcp(); flow_gen->defer_client_port_free(is_tcp,node->m_src_idx,lpP->rtp_client_0, node->m_template_info->m_client_pool_idx,node->m_tuple_gen); flow_gen->defer_client_port_free(is_tcp,node->m_src_idx,lpP->rtp_client_1, node->m_template_info->m_client_pool_idx, node->m_tuple_gen); assert(lpP); delete lpP; node->m_plugin_info=0; }else{ if (plugin_id ==mpDYN_PYLOAD) { /* nothing to do */ }else{ if (plugin_id ==mpAVL_HTTP_BROWSIN) { /* nothing to do */ CTcpSeq * lpP=(CTcpSeq * )node->m_plugin_info; delete lpP; node->m_plugin_info=0; }else{ /* do not support this */ assert(0); } } } } rte_mbuf_t * CPluginCallbackSimple::http_plugin(uint8_t plugin_id, CGenNode * node, CFlowPktInfo * pkt_info){ CPacketDescriptor * lpd=&pkt_info->m_pkt_indication.m_desc; assert(lpd->getFlowId()==0); /* only one flow */ CMiniVMCmdBase * program[2]; CMiniVMReplaceIP replace_cmd; CMiniVMCmdBase eop_cmd; CTcpSeq * lpP=(CTcpSeq * )node->m_plugin_info; assert(lpP); rte_mbuf_t *mbuf; int16_t s_size=0; if ( likely (lpd->getFlowPktNum() != 3) ){ if (unlikely (CGlobalInfo::is_ipv6_enable()) ) { // Request a larger initial segment for IPv6 mbuf = pkt_info->do_generate_new_mbuf_big(node); }else{ mbuf = pkt_info->do_generate_new_mbuf(node); } }else{ CFlowInfo flow_info; flow_info.vm_program=0; flow_info.client_ip = node->m_src_ip; flow_info.server_ip = node->m_dest_ip; flow_info.client_port = node->m_src_port; flow_info.server_port = 0; flow_info.replace_server_port =false; flow_info.is_init_ip_dir = (node->cur_pkt_ip_addr_dir() == CLIENT_SIDE?true:false); flow_info.is_init_port_dir = (node->cur_pkt_port_addr_dir() ==CLIENT_SIDE?true:false); replace_cmd.m_cmd = VM_REPLACE_IP_OFFSET; replace_cmd.m_flags = 0; // Determine how much larger the packet needs to be to // handle the largest IP address. There is a single address // string of 8 bytes that needs to be replaced. if (CGlobalInfo::is_ipv6_enable() ) { // For IPv6, accomodate use of brackets (+2 bytes) replace_cmd.m_add_pkt_len = (INET6_ADDRSTRLEN + 2) - 8; // Mark as IPv6 and set the upper 96-bits replace_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ replace_cmd.m_server_ip.v6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; } } else { replace_cmd.m_add_pkt_len = INET_ADDRSTRLEN - 8; } // Set m_start_0/m_stop_1 at start/end of IP address to be replaced. // For this packet we know the IP addr string length is 8 bytes. replace_cmd.m_start_0 = 10+16; replace_cmd.m_stop_1 = replace_cmd.m_start_0 + 8; replace_cmd.m_server_ip.v4 = flow_info.server_ip; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; mbuf = pkt_info->do_generate_new_mbuf_ex_vm(node,&flow_info, &s_size); } // Fixup the TCP sequence numbers uint8_t *p=rte_pktmbuf_mtod(mbuf, uint8_t*); // Update TCP sequence numbers lpP->update(p, pkt_info, s_size); return(mbuf); } rte_mbuf_t * CPluginCallbackSimple::dyn_pyload_plugin(uint8_t plugin_id, CGenNode * node, CFlowPktInfo * pkt_info){ CMiniVMCmdBase * program[2]; CMiniVMDynPyload dyn_cmd; CMiniVMCmdBase eop_cmd; CPacketDescriptor * lpd=&pkt_info->m_pkt_indication.m_desc; CFlowYamlDynamicPyloadPlugin * lpt = node->m_template_info->m_dpPkt; assert(lpt); CFlowInfo flow_info; flow_info.vm_program=0; int16_t s_size=0; // IPv6 packets are not supported if (CGlobalInfo::is_ipv6_enable() ) { fprintf (stderr," IPv6 is not supported for dynamic pyload change\n"); exit(-1); } if ( lpd->getFlowId() == 0 ) { flow_info.client_ip = node->m_src_ip; flow_info.server_ip = node->m_dest_ip; flow_info.client_port = node->m_src_port; flow_info.server_port = 0; flow_info.replace_server_port =false; flow_info.is_init_ip_dir = (node->cur_pkt_ip_addr_dir() == CLIENT_SIDE?true:false); flow_info.is_init_port_dir = (node->cur_pkt_port_addr_dir() ==CLIENT_SIDE?true:false); uint32_t pkt_num = lpd->getFlowPktNum(); if (pkt_num < 253) { int i; /* fast filter */ for (i=0; im_num; i++) { if (lpt->m_pkt_ids[i] == pkt_num ) { //add a program here dyn_cmd.m_cmd = VM_DYN_PYLOAD; dyn_cmd.m_ptr= &lpt->m_program[i]; dyn_cmd.m_flags = 0; dyn_cmd.m_add_pkt_len = INET_ADDRSTRLEN - 8; dyn_cmd.m_ip.v4=node->m_src_ip; eop_cmd.m_cmd = VM_EOP; program[0] = &dyn_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } } } // only for the first flow }else{ fprintf (stderr," only one flow is allowed for dynamic pyload change \n"); exit(-1); }/* only for the first flow */ if ( unlikely( flow_info.vm_program != 0 ) ) { return ( pkt_info->do_generate_new_mbuf_ex_vm(node,&flow_info, &s_size) ); }else{ return ( pkt_info->do_generate_new_mbuf_ex(node,&flow_info) ); } } rte_mbuf_t * CPluginCallbackSimple::sip_voice_plugin(uint8_t plugin_id,CGenNode * node,CFlowPktInfo * pkt_info){ CMiniVMCmdBase * program[2]; CMiniVMReplaceIP_PORT_IP_IP_Port via_replace_cmd; CMiniVMCmdBase eop_cmd; CPacketDescriptor * lpd=&pkt_info->m_pkt_indication.m_desc; CPlugin_rtsp * lpP=(CPlugin_rtsp * )node->m_plugin_info; assert(lpP); // printf(" %d %d \n",lpd->getFlowId(),lpd->getFlowPktNum()); CFlowInfo flow_info; flow_info.vm_program=0; int16_t s_size=0; switch ( lpd->getFlowId() ) { /* flow - SIP , packet #0,#1 control */ case 0: flow_info.client_ip = node->m_src_ip; flow_info.server_ip = node->m_dest_ip; flow_info.client_port = node->m_src_port; flow_info.server_port = 0; flow_info.replace_server_port =false; flow_info.is_init_ip_dir = (node->cur_pkt_ip_addr_dir() == CLIENT_SIDE?true:false); flow_info.is_init_port_dir = (node->cur_pkt_port_addr_dir() ==CLIENT_SIDE?true:false); /* program to replace ip server */ switch ( lpd->getFlowPktNum() ) { case 0: { via_replace_cmd.m_cmd = VM_REPLACE_IPVIA_IP_IP_PORT; via_replace_cmd.m_flags = 0; via_replace_cmd.m_start_0 = 0; via_replace_cmd.m_stop_1 = 0; // Determine how much larger the packet needs to be to // handle the largest IP address. There are 3 address // strings (each 9 bytes) that needs to be replaced. // We also need to accomodate IPv6 use of brackets // (+2 bytes) in a URI. // There are also 2 port strings that needs to be // replaced (1 is 4 bytes the other is 5 bytes). if (CGlobalInfo::is_ipv6_enable() ) { via_replace_cmd.m_add_pkt_len = (((INET6_ADDRSTRLEN + 2) - 9) * 3) + ((INET_PORTSTRLEN * 2) - 9); // Mark as IPv6 and set the upper 96-bits via_replace_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ via_replace_cmd.m_ip.v6[idx] = CGlobalInfo::m_options.m_src_ipv6[idx]; via_replace_cmd.m_ip_via.v6[idx] = CGlobalInfo::m_options.m_src_ipv6[idx]; } } else { via_replace_cmd.m_add_pkt_len = ((INET_ADDRSTRLEN - 9) * 3) + ((INET_PORTSTRLEN * 2) - 9); } via_replace_cmd.m_ip.v4 =node->m_src_ip; via_replace_cmd.m_ip0_start = 377; via_replace_cmd.m_ip0_stop = 377+9; via_replace_cmd.m_ip1_start = 409; via_replace_cmd.m_ip1_stop = 409+9; via_replace_cmd.m_port =lpP->rtp_client_0; via_replace_cmd.m_port_start = 435; via_replace_cmd.m_port_stop = 435+5; via_replace_cmd.m_ip_via.v4 = node->m_src_ip; via_replace_cmd.m_port_via = node->m_src_port; via_replace_cmd.m_ip_via_start = 208; via_replace_cmd.m_ip_via_stop = 208+9+5; eop_cmd.m_cmd = VM_EOP; program[0] = &via_replace_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; case 1: { via_replace_cmd.m_cmd = VM_REPLACE_IPVIA_IP_IP_PORT; via_replace_cmd.m_flags = 0; via_replace_cmd.m_start_0 = 0; via_replace_cmd.m_stop_1 = 0; // Determine how much larger the packet needs to be to // handle the largest IP address. There are 3 address // strings (each 9 bytes) that needs to be replaced. // We also need to accomodate IPv6 use of brackets // (+2 bytes) in a URI. // There are also 2 port strings that needs to be // replaced (1 is 4 bytes the other is 5 bytes). if (CGlobalInfo::is_ipv6_enable() ) { via_replace_cmd.m_add_pkt_len = (((INET6_ADDRSTRLEN + 2) - 9) * 3) + ((INET_PORTSTRLEN * 2) - 9); // Mark as IPv6 and set the upper 96-bits via_replace_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ via_replace_cmd.m_ip.v6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; via_replace_cmd.m_ip_via.v6[idx] = CGlobalInfo::m_options.m_src_ipv6[idx]; } } else { via_replace_cmd.m_add_pkt_len = ((INET_ADDRSTRLEN - 9) * 3) + ((INET_PORTSTRLEN * 2) - 9); } via_replace_cmd.m_ip.v4 =node->m_dest_ip; via_replace_cmd.m_ip0_start = 370; via_replace_cmd.m_ip0_stop = 370+8; via_replace_cmd.m_ip1_start = 401; via_replace_cmd.m_ip1_stop = 401+8; via_replace_cmd.m_port =lpP->rtp_client_0; via_replace_cmd.m_port_start = 426; via_replace_cmd.m_port_stop = 426+5; via_replace_cmd.m_ip_via.v4 = node->m_src_ip; via_replace_cmd.m_port_via = node->m_src_port; via_replace_cmd.m_ip_via_start = 207; via_replace_cmd.m_ip_via_stop = 207+9+5; eop_cmd.m_cmd = VM_EOP; program[0] = &via_replace_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; }/* end of big switch on packet */ break; case 1: flow_info.client_ip = node->m_src_ip ; flow_info.server_ip = node->m_dest_ip; flow_info.client_port = lpP->rtp_client_0; /* this is tricky ..*/ flow_info.server_port = lpP->rtp_client_0; flow_info.replace_server_port = true; flow_info.is_init_ip_dir = (node->cur_pkt_ip_addr_dir() == CLIENT_SIDE?true:false); flow_info.is_init_port_dir = (node->cur_pkt_port_addr_dir() ==CLIENT_SIDE?true:false); break; default: assert(0); break; }; //printf(" c_ip:%x s_ip:%x c_po:%x s_po:%x init:%x replace:%x \n",flow_info.client_ip,flow_info.server_ip,flow_info.client_port,flow_info.server_port,flow_info.is_init_dir,flow_info.replace_server_port); //printf(" program %p \n",flow_info.vm_program); if ( unlikely( flow_info.vm_program != 0 ) ) { return ( pkt_info->do_generate_new_mbuf_ex_vm(node,&flow_info, &s_size) ); }else{ return ( pkt_info->do_generate_new_mbuf_ex(node,&flow_info) ); } } rte_mbuf_t * CPluginCallbackSimple::rtsp_plugin(uint8_t plugin_id,CGenNode * node,CFlowPktInfo * pkt_info){ CMiniVMCmdBase * program[2]; CMiniVMReplaceIP replace_cmd; CMiniVMCmdBase eop_cmd; CMiniVMReplaceIPWithPort replace_port_cmd; rte_mbuf_t *mbuf; CPacketDescriptor * lpd=&pkt_info->m_pkt_indication.m_desc; CPlugin_rtsp * lpP=(CPlugin_rtsp * )node->m_plugin_info; assert(lpP); // printf(" %d %d \n",lpd->getFlowId(),lpd->getFlowPktNum()); CFlowInfo flow_info; flow_info.vm_program=0; int16_t s_size=0; switch ( lpd->getFlowId() ) { /* flow - control */ case 0: flow_info.client_ip = node->m_src_ip; flow_info.server_ip = node->m_dest_ip; flow_info.client_port = node->m_src_port; flow_info.server_port = 0; flow_info.replace_server_port =false; flow_info.is_init_ip_dir = (node->cur_pkt_ip_addr_dir() == CLIENT_SIDE?true:false); flow_info.is_init_port_dir = (node->cur_pkt_port_addr_dir() ==CLIENT_SIDE?true:false); /* program to replace ip server */ switch ( lpd->getFlowPktNum() ) { case 3: { replace_cmd.m_cmd = VM_REPLACE_IP_OFFSET; replace_cmd.m_flags = 0; replace_cmd.m_start_0 = 16; replace_cmd.m_stop_1 = 16+9; // Determine how much larger the packet needs to be to // handle the largest IP address. There is a single address // string of 9 bytes that needs to be replaced. if (CGlobalInfo::is_ipv6_enable() ) { // For IPv6, accomodate use of brackets (+2 bytes) replace_cmd.m_add_pkt_len = (INET6_ADDRSTRLEN + 2) - 9; // Mark as IPv6 and set the upper 96-bits replace_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ replace_cmd.m_server_ip.v6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; } } else { replace_cmd.m_add_pkt_len = INET_ADDRSTRLEN - 9; } replace_cmd.m_server_ip.v4 = flow_info.server_ip; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; case 4: { replace_cmd.m_cmd = VM_REPLACE_IP_OFFSET; replace_cmd.m_flags = 0; replace_cmd.m_start_0 = 46; replace_cmd.m_stop_1 = 46+9; // Determine how much larger the packet needs to be to // handle the largest IP address. There is a single address // string of 9 bytes that needs to be replaced. if (CGlobalInfo::is_ipv6_enable() ) { // For IPv6, accomodate use of brackets (+2 bytes) replace_cmd.m_add_pkt_len = (INET6_ADDRSTRLEN + 2) - 9; // Mark as IPv6 and set the upper 96-bits replace_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ replace_cmd.m_server_ip.v6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; } } else { replace_cmd.m_add_pkt_len = INET_ADDRSTRLEN - 9; } replace_cmd.m_server_ip.v4 = flow_info.server_ip; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; case 5: { replace_port_cmd.m_cmd = VM_REPLACE_IP_PORT_OFFSET; replace_port_cmd.m_flags = 0; replace_port_cmd.m_start_0 = 13; replace_port_cmd.m_stop_1 = 13+9; // Determine how much larger the packet needs to be to // handle the largest IP address. There is a single address // string of 9 bytes that needs to be replaced. // There are also 2 port strings (8 bytes) that needs to be // replaced. if (CGlobalInfo::is_ipv6_enable() ) { // For IPv6, accomodate use of brackets (+2 bytes) replace_port_cmd.m_add_pkt_len = ((INET6_ADDRSTRLEN + 2) - 9) + ((INET_PORTSTRLEN * 2) - 8); // Mark as IPv6 and set the upper 96-bits replace_port_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ replace_port_cmd.m_server_ip.v6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; } } else { replace_port_cmd.m_add_pkt_len = (INET_ADDRSTRLEN - 9) + ((INET_PORTSTRLEN * 2) - 8); } replace_port_cmd.m_server_ip.v4 = flow_info.server_ip; replace_port_cmd.m_start_port = 164; replace_port_cmd.m_stop_port = 164+(4*2)+1; replace_port_cmd.m_client_port = lpP->rtp_client_0; replace_port_cmd.m_server_port =0; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_port_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; case 6: { replace_port_cmd.m_cmd = VM_REPLACE_IP_PORT_RESPONSE_OFFSET; replace_port_cmd.m_flags = 0; replace_port_cmd.m_start_0 = 0; replace_port_cmd.m_stop_1 = 0; // Determine how much larger the packet needs to be to // handle the largest port addresses. There are 4 port address // strings (16 bytes) that needs to be replaced. replace_port_cmd.m_add_pkt_len = ((INET_PORTSTRLEN * 4) - 16); replace_port_cmd.m_server_ip.v4 = flow_info.server_ip; replace_port_cmd.m_start_port = 247; replace_port_cmd.m_stop_port = 247+(4*4)+2+13; replace_port_cmd.m_client_port = lpP->rtp_client_0; replace_port_cmd.m_server_port = lpP->rtp_client_0; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_port_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; case 7: { replace_port_cmd.m_cmd = VM_REPLACE_IP_PORT_OFFSET; replace_port_cmd.m_flags = 0; replace_port_cmd.m_start_0 = 13; replace_port_cmd.m_stop_1 = 13+9; // Determine how much larger the packet needs to be to // handle the largest IP address. There is a single address // string of 9 bytes that needs to be replaced. // There are also 2 port strings (8 bytes) that needs to be // replaced. if (CGlobalInfo::is_ipv6_enable() ) { // For IPv6, accomodate use of brackets (+2 bytes) replace_port_cmd.m_add_pkt_len = ((INET6_ADDRSTRLEN + 2) - 9) + ((INET_PORTSTRLEN * 2) - 8); // Mark as IPv6 and set the upper 96-bits replace_port_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ replace_port_cmd.m_server_ip.v6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; } } else { replace_port_cmd.m_add_pkt_len = (INET_ADDRSTRLEN - 9) + ((INET_PORTSTRLEN * 2) - 8); } replace_port_cmd.m_server_ip.v4 = flow_info.server_ip; replace_port_cmd.m_start_port = 164; replace_port_cmd.m_stop_port = 164+(4*2)+1; replace_port_cmd.m_client_port = lpP->rtp_client_1; replace_port_cmd.m_server_port =0; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_port_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; case 8: { replace_port_cmd.m_cmd = VM_REPLACE_IP_PORT_RESPONSE_OFFSET; replace_port_cmd.m_flags = 0; replace_port_cmd.m_start_0 = 0; replace_port_cmd.m_stop_1 = 0; // Determine how much larger the packet needs to be to // handle the largest port addresses. There are 4 port address // strings (16 bytes) that needs to be replaced. replace_port_cmd.m_add_pkt_len = ((INET_PORTSTRLEN * 4) - 16); replace_port_cmd.m_server_ip.v4 = flow_info.server_ip; replace_port_cmd.m_start_port = 247; replace_port_cmd.m_stop_port = 247+(4*4)+2+13; replace_port_cmd.m_client_port = lpP->rtp_client_1; replace_port_cmd.m_server_port = lpP->rtp_client_1; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_port_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; /* PLAY */ case 9: { replace_cmd.m_cmd = VM_REPLACE_IP_OFFSET; replace_cmd.m_flags = 0; replace_cmd.m_start_0 = 12; replace_cmd.m_stop_1 = 12+9; // Determine how much larger the packet needs to be to // handle the largest IP address. There is a single address // string of 9 bytes that needs to be replaced. if (CGlobalInfo::is_ipv6_enable() ) { // For IPv6, accomodate use of brackets (+2 bytes) replace_cmd.m_add_pkt_len = (INET6_ADDRSTRLEN + 2) - 9; // Mark as IPv6 and set the upper 96-bits replace_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ replace_cmd.m_server_ip.v6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; } } else { replace_cmd.m_add_pkt_len = INET_ADDRSTRLEN - 9; } replace_cmd.m_server_ip.v4 = flow_info.server_ip; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; /*OPTION 0*/ case 12: { replace_cmd.m_cmd = VM_REPLACE_IP_OFFSET; replace_cmd.m_flags = 0; replace_cmd.m_start_0 = 15; replace_cmd.m_stop_1 = 15+9; // Determine how much larger the packet needs to be to // handle the largest IP address. There is a single address // string of 9 bytes that needs to be replaced. if (CGlobalInfo::is_ipv6_enable() ) { // For IPv6, accomodate use of brackets (+2 bytes) replace_cmd.m_add_pkt_len = (INET6_ADDRSTRLEN + 2) - 9; // Mark as IPv6 and set the upper 96-bits replace_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ replace_cmd.m_server_ip.v6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; } } else { replace_cmd.m_add_pkt_len = INET_ADDRSTRLEN - 9; } replace_cmd.m_server_ip.v4 = flow_info.server_ip; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; /* option #2*/ case 15: { replace_cmd.m_cmd = VM_REPLACE_IP_OFFSET; replace_cmd.m_flags = 0; replace_cmd.m_start_0 = 15; replace_cmd.m_stop_1 = 15+9; // Determine how much larger the packet needs to be to // handle the largest IP address. There is a single address // string of 9 bytes that needs to be replaced. if (CGlobalInfo::is_ipv6_enable() ) { // For IPv6, accomodate use of brackets (+2 bytes) replace_cmd.m_add_pkt_len = (INET6_ADDRSTRLEN + 2) - 9; // Mark as IPv6 and set the upper 96-bits replace_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ replace_cmd.m_server_ip.v6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; } } else { replace_cmd.m_add_pkt_len = INET_ADDRSTRLEN - 9; } replace_cmd.m_server_ip.v4 = flow_info.server_ip; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; case 18: { replace_cmd.m_cmd = VM_REPLACE_IP_OFFSET; replace_cmd.m_flags = 0; replace_cmd.m_start_0 = 16; replace_cmd.m_stop_1 = 16+9; // Determine how much larger the packet needs to be to // handle the largest IP address. There is a single address // string of 9 bytes that needs to be replaced. if (CGlobalInfo::is_ipv6_enable() ) { // For IPv6, accomodate use of brackets (+2 bytes) replace_cmd.m_add_pkt_len = (INET6_ADDRSTRLEN + 2) - 9; // Mark as IPv6 and set the upper 96-bits replace_cmd.m_flags |= CMiniVMCmdBase::MIN_VM_V6; for (uint8_t idx=0; idx<6; idx++){ replace_cmd.m_server_ip.v6[idx] = CGlobalInfo::m_options.m_dst_ipv6[idx]; } } else { replace_cmd.m_add_pkt_len = INET_ADDRSTRLEN - 9; } replace_cmd.m_server_ip.v4 = flow_info.server_ip; eop_cmd.m_cmd = VM_EOP; program[0] = &replace_cmd; program[1] = &eop_cmd; flow_info.vm_program = program; } break; }/* end of big switch on packet */ break; case 1: flow_info.client_ip = node->m_src_ip ; flow_info.server_ip = node->m_dest_ip; flow_info.client_port = lpP->rtp_client_0; /* this is tricky ..*/ flow_info.server_port = lpP->rtp_client_0; flow_info.replace_server_port = true; flow_info.is_init_ip_dir = (node->cur_pkt_ip_addr_dir() == CLIENT_SIDE?true:false); flow_info.is_init_port_dir = (node->cur_pkt_port_addr_dir() ==CLIENT_SIDE?true:false); break; case 2: flow_info.client_ip = node->m_src_ip ; flow_info.server_ip = node->m_dest_ip; flow_info.client_port = lpP->rtp_client_1; /* this is tricky ..*/ flow_info.server_port = lpP->rtp_client_1; flow_info.replace_server_port =true; flow_info.is_init_ip_dir = (node->cur_pkt_ip_addr_dir() == CLIENT_SIDE?true:false); flow_info.is_init_port_dir = (node->cur_pkt_port_addr_dir() ==CLIENT_SIDE?true:false); break; default: assert(0); break; }; //printf(" c_ip:%x s_ip:%x c_po:%x s_po:%x init:%x replace:%x \n",flow_info.client_ip,flow_info.server_ip,flow_info.client_port,flow_info.server_port,flow_info.is_init_dir,flow_info.replace_server_port); //printf(" program %p \n",flow_info.vm_program); if ( unlikely( flow_info.vm_program != 0 ) ) { mbuf = pkt_info->do_generate_new_mbuf_ex_vm(node,&flow_info, &s_size); }else{ if (unlikely (CGlobalInfo::is_ipv6_enable()) ) { // Request a larger initial segment for IPv6 mbuf = pkt_info->do_generate_new_mbuf_ex_big(node,&flow_info); }else{ mbuf = pkt_info->do_generate_new_mbuf_ex(node,&flow_info); } } // Fixup the TCP sequence numbers for the TCP flow if ( lpd->getFlowId() == 0 ) { uint8_t *p=rte_pktmbuf_mtod(mbuf, uint8_t*); // Update TCP sequence numbers lpP->update(p, pkt_info, s_size); } return(mbuf); } /* replace the tuples */ rte_mbuf_t * CPluginCallbackSimple::on_node_generate_mbuf(uint8_t plugin_id,CGenNode * node,CFlowPktInfo * pkt_info){ rte_mbuf_t * m=NULL; switch (plugin_id) { case mpRTSP: m=rtsp_plugin(plugin_id,node,pkt_info); break; case mpSIP_VOICE: m=sip_voice_plugin(plugin_id,node,pkt_info); break; case mpDYN_PYLOAD: m=dyn_pyload_plugin(plugin_id,node,pkt_info); break; case mpAVL_HTTP_BROWSIN: m=http_plugin(plugin_id,node,pkt_info); break; default: assert(0); } return (m); } int CMiniVM::mini_vm_run(CMiniVMCmdBase * cmds[]){ m_new_pkt_size=0; bool need_to_stop=false; int cnt=0; CMiniVMCmdBase * cmd=cmds[cnt]; while (! need_to_stop) { switch (cmd->m_cmd) { case VM_REPLACE_IP_OFFSET: mini_vm_replace_ip((CMiniVMReplaceIP *)cmd); break; case VM_REPLACE_IP_PORT_OFFSET: mini_vm_replace_port_ip((CMiniVMReplaceIPWithPort *)cmd); break; case VM_REPLACE_IP_PORT_RESPONSE_OFFSET: mini_vm_replace_ports((CMiniVMReplaceIPWithPort *)cmd); break; case VM_REPLACE_IP_IP_PORT: mini_vm_replace_ip_ip_ports((CMiniVMReplaceIP_IP_Port * )cmd); break; case VM_REPLACE_IPVIA_IP_IP_PORT: mini_vm_replace_ip_via_ip_ip_ports((CMiniVMReplaceIP_PORT_IP_IP_Port *)cmd); break; case VM_DYN_PYLOAD: mini_vm_dyn_payload((CMiniVMDynPyload *)cmd); break; case VM_EOP: need_to_stop=true; break; default: printf(" vm cmd %d does not exist \n",cmd->m_cmd); assert(0); } cnt++; cmd=cmds[cnt]; } return (0); } inline int cp_pkt_len(char *to,char *from,uint16_t from_offset,uint16_t len){ memcpy(to, from+from_offset , len); return (len); } /* not including the to_offset 0 1 x */ inline int cp_pkt_to_from(char *to,char *from,uint16_t from_offset,uint16_t to_offset){ memcpy(to, from+from_offset , to_offset-from_offset) ; return (to_offset-from_offset); } int CMiniVM::mini_vm_dyn_payload( CMiniVMDynPyload * cmd){ /* copy all the packet */ CFlowYamlDpPkt * dyn=(CFlowYamlDpPkt *)cmd->m_ptr; uint16_t l7_offset = m_pkt_info->m_pkt_indication.getFastPayloadOffset(); uint16_t len = m_pkt_info->m_packet->pkt_len - l7_offset ; char * original_l7_ptr=m_pkt_info->m_packet->raw+l7_offset; char * p=m_pyload_mbuf_ptr; /* copy payload */ memcpy(p,original_l7_ptr,len); if ( ( dyn->m_pyld_offset+ (dyn->m_len*4)) < ( len-4) ){ // we can change the packet int i; uint32_t *l=(uint32_t *)(p+dyn->m_pyld_offset); for (i=0; im_len; i++) { if ( dyn->m_type==0 ) { *l=(rand() & dyn->m_pkt_mask); }else if (dyn->m_type==1){ *l=(PKT_NTOHL(cmd->m_ip.v4) & dyn->m_pkt_mask); } l++; } } // Return packet size which hasn't changed m_new_pkt_size = m_pkt_info->m_packet->pkt_len; return (0); } int CMiniVM::mini_vm_replace_ip_via_ip_ip_ports(CMiniVMReplaceIP_PORT_IP_IP_Port * cmd){ uint16_t l7_offset = m_pkt_info->m_pkt_indication.getFastPayloadOffset(); uint16_t len = m_pkt_info->m_packet->pkt_len - l7_offset; char * original_l7_ptr=m_pkt_info->m_packet->raw+l7_offset; char * p=m_pyload_mbuf_ptr; p+=cp_pkt_to_from(p,original_l7_ptr, 0, cmd->m_ip_via_start); if (cmd->m_flags & CMiniVMCmdBase::MIN_VM_V6) { p+=ipv6_to_str(&cmd->m_ip_via,p); } else { p+=ip_to_str(cmd->m_ip_via.v4,p); } p+=sprintf(p,":%u",cmd->m_port_via); /* up to the IP */ p+=cp_pkt_to_from(p,original_l7_ptr, cmd->m_ip_via_stop, cmd->m_ip0_start); /*IP */ if (cmd->m_flags & CMiniVMCmdBase::MIN_VM_V6) { p[-2] = '6'; p+=ipv6_to_str(&cmd->m_ip,p); } else { p+=ip_to_str(cmd->m_ip.v4,p); } /* up to IP 2 */ p+=cp_pkt_to_from(p, original_l7_ptr , cmd->m_ip0_stop, cmd->m_ip1_start); /* IP2 */ if (cmd->m_flags & CMiniVMCmdBase::MIN_VM_V6) { p[-2] = '6'; p+=ipv6_to_str(&cmd->m_ip,p); } else { p+=ip_to_str(cmd->m_ip.v4,p); } /* up to port */ p+=cp_pkt_to_from(p, original_l7_ptr , cmd->m_ip1_stop, cmd->m_port_start); /* port */ p+=sprintf(p,"%u",cmd->m_port); /* up to end */ p+=cp_pkt_to_from(p, original_l7_ptr , cmd->m_port_stop, len); // Determine new packet size m_new_pkt_size= ((p+l7_offset) - m_pyload_mbuf_ptr); return (0); } int CMiniVM::mini_vm_replace_ip_ip_ports(CMiniVMReplaceIP_IP_Port * cmd){ uint16_t l7_offset = m_pkt_info->m_pkt_indication.getFastPayloadOffset(); uint16_t len = m_pkt_info->m_packet->pkt_len - l7_offset; char * original_l7_ptr=m_pkt_info->m_packet->raw+l7_offset; char * p=m_pyload_mbuf_ptr; /* up to the IP */ p+=cp_pkt_to_from(p,original_l7_ptr, 0, cmd->m_ip0_start); /*IP */ if (cmd->m_flags & CMiniVMCmdBase::MIN_VM_V6) { p+=ipv6_to_str(&cmd->m_ip,p); } else { p+=ip_to_str(cmd->m_ip.v4,p); } /* up to IP 2 */ p+=cp_pkt_to_from(p, original_l7_ptr , cmd->m_ip0_stop, cmd->m_ip1_start); /* IP2 */ if (cmd->m_flags & CMiniVMCmdBase::MIN_VM_V6) { p+=ipv6_to_str(&cmd->m_ip,p); } else { p+=ip_to_str(cmd->m_ip.v4,p); } /* up to port */ p+=cp_pkt_to_from(p, original_l7_ptr , cmd->m_ip1_stop, cmd->m_port_start); /* port */ p+=sprintf(p,"%u",cmd->m_port); /* up to end */ p+=cp_pkt_to_from(p, original_l7_ptr , cmd->m_port_stop, len); // Determine new packet size m_new_pkt_size= ((p+l7_offset) - m_pyload_mbuf_ptr); return (0); } int CMiniVM::mini_vm_replace_ports(CMiniVMReplaceIPWithPort * cmd){ uint16_t l7_offset = m_pkt_info->m_pkt_indication.getFastPayloadOffset(); uint16_t len = m_pkt_info->m_packet->pkt_len - l7_offset; char * original_l7_ptr=m_pkt_info->m_packet->raw+l7_offset; memcpy(m_pyload_mbuf_ptr, original_l7_ptr,cmd->m_start_port); char * p=m_pyload_mbuf_ptr+cmd->m_start_port; p+=sprintf(p,"%u-%u;server_port=%u-%u",cmd->m_client_port,cmd->m_client_port+1,cmd->m_server_port,cmd->m_server_port+1); memcpy(p, original_l7_ptr+cmd->m_stop_port,len-cmd->m_stop_port); p+=(len-cmd->m_stop_port); // Determine new packet size m_new_pkt_size= ((p+l7_offset) - m_pyload_mbuf_ptr); return (0); } int CMiniVM::mini_vm_replace_port_ip(CMiniVMReplaceIPWithPort * cmd){ uint16_t l7_offset=m_pkt_info->m_pkt_indication.getFastPayloadOffset(); uint16_t len = m_pkt_info->m_packet->pkt_len - l7_offset - 0; char * original_l7_ptr=m_pkt_info->m_packet->raw+l7_offset; memcpy(m_pyload_mbuf_ptr, original_l7_ptr,cmd->m_start_0); char *p=m_pyload_mbuf_ptr+cmd->m_start_0; if (cmd->m_flags & CMiniVMCmdBase::MIN_VM_V6) { p+=ipv6_to_str(&cmd->m_server_ip,p); } else { p+=ip_to_str(cmd->m_server_ip.v4,p); } /* copy until the port start offset */ int len1=cmd->m_start_port-cmd->m_stop_1 ; memcpy(p, original_l7_ptr+cmd->m_stop_1,len1); p+=len1; p+=sprintf(p,"%u-%u",cmd->m_client_port,cmd->m_client_port+1); memcpy(p, original_l7_ptr+cmd->m_stop_port,len-cmd->m_stop_port); p+=len-cmd->m_stop_port; // Determine new packet size m_new_pkt_size= ((p+l7_offset) - m_pyload_mbuf_ptr); return (0); } int CMiniVM::mini_vm_replace_ip(CMiniVMReplaceIP * cmd){ uint16_t l7_offset = m_pkt_info->m_pkt_indication.getFastPayloadOffset(); uint16_t len = m_pkt_info->m_packet->pkt_len - l7_offset; char * original_l7_ptr = m_pkt_info->m_packet->raw+l7_offset; memcpy(m_pyload_mbuf_ptr, original_l7_ptr,cmd->m_start_0); char *p=m_pyload_mbuf_ptr+cmd->m_start_0; int n_size=0; if (cmd->m_flags & CMiniVMCmdBase::MIN_VM_V6) { n_size=ipv6_to_str(&cmd->m_server_ip,p); } else { n_size=ip_to_str(cmd->m_server_ip.v4,p); } p+=n_size; memcpy(p, original_l7_ptr+cmd->m_stop_1,len-cmd->m_stop_1); // Determine new packet size m_new_pkt_size= ((p+l7_offset+(len-cmd->m_stop_1)) - m_pyload_mbuf_ptr); return (0); } void CFlowYamlDpPkt::Dump(FILE *fd){ fprintf(fd," pkt_id : %d \n",(int)m_pkt_id); fprintf(fd," offset : %d \n",(int)m_pyld_offset); fprintf(fd," offset : %d \n",(int)m_type); fprintf(fd," len : %d \n",(int)m_len); fprintf(fd," mask : 0x%x \n",(int)m_pkt_mask); } void CFlowYamlDynamicPyloadPlugin::Add(CFlowYamlDpPkt & fd){ if (m_num ==MAX_PYLOAD_PKT_CHANGE) { fprintf (stderr,"ERROR can set only %d rules \n",MAX_PYLOAD_PKT_CHANGE); exit(-1); } m_pkt_ids[m_num]=fd.m_pkt_id; m_program[m_num]=fd; m_num+=1; } void CFlowYamlDynamicPyloadPlugin::Dump(FILE *fd){ int i; fprintf(fd," pkts :"); for (i=0; igetTotalLength(); } if (m_ipv6) { ip_len=m_ipv6->getSize()+m_ipv6->getPayloadLen(); } return ( ip_len +m_vlan_offset+14); } uint16_t CSimplePacketParser::getIpId() { if (m_ipv4) { return ( m_ipv4->getId() ); } return (0); } uint8_t CSimplePacketParser::getTTl(){ if (m_ipv4) { return ( m_ipv4->getTimeToLive() ); } if (m_ipv6) { return ( m_ipv6->getHopLimit() ); } return (0); } bool CSimplePacketParser::Parse(){ rte_mbuf_t * m=m_m; uint8_t *p=rte_pktmbuf_mtod(m, uint8_t*); EthernetHeader *m_ether = (EthernetHeader *)p; IPHeader * ipv4=0; IPv6Header * ipv6=0; m_vlan_offset=0; m_option_offset=0; uint8_t protocol = 0; // Retrieve the protocol type from the packet switch( m_ether->getNextProtocol() ) { case EthernetHeader::Protocol::IP : // IPv4 packet ipv4=(IPHeader *)(p+14); m_l4 = (uint8_t *)ipv4 + ipv4->getHeaderLength(); protocol = ipv4->getProtocol(); m_option_offset = 14 + IPV4_HDR_LEN; break; case EthernetHeader::Protocol::IPv6 : // IPv6 packet ipv6=(IPv6Header *)(p+14); m_l4 = (uint8_t *)ipv6 + ipv6->getHeaderLength(); protocol = ipv6->getNextHdr(); m_option_offset = 14 +IPV6_HDR_LEN; break; case EthernetHeader::Protocol::VLAN : m_vlan_offset = 4; switch ( m_ether->getVlanProtocol() ){ case EthernetHeader::Protocol::IP: // IPv4 packet ipv4=(IPHeader *)(p+18); m_l4 = (uint8_t *)ipv4 + ipv4->getHeaderLength(); protocol = ipv4->getProtocol(); m_option_offset = 18+ IPV4_HDR_LEN; break; case EthernetHeader::Protocol::IPv6 : // IPv6 packet ipv6=(IPv6Header *)(p+18); m_l4 = (uint8_t *)ipv6 + ipv6->getHeaderLength(); protocol = ipv6->getNextHdr(); m_option_offset = 18 + IPV6_HDR_LEN; break; default: break; } default: break; } m_protocol =protocol; m_ipv4=ipv4; m_ipv6=ipv6; if ( protocol == 0 ){ return (false); } return (true); } /* free the right object. it is classic to use virtual function but we can't do it here and we don't even want to use callback function as we want to save space and in most cases there is nothing to free. this might be changed in the future */ void CGenNodeBase::free_base(){ if ( m_type == FLOW_PKT ) { CGenNode* p=(CGenNode*)this; p->free_gen_node(); return; } if (m_type==STATELESS_PKT) { CGenNodeStateless* p=(CGenNodeStateless*)this; p->free_stl_node(); return; } if (m_type == PCAP_PKT) { CGenNodePCAP *p = (CGenNodePCAP *)this; p->destroy(); return; } if ( m_type == COMMAND ) { CGenNodeCommand* p=(CGenNodeCommand*)this; p->free_command(); } }