/* // Copyright (c) 2010-2017 Intel Corporation // // 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 #include #include #include #include #include #include "prox_malloc.h" #include "task_base.h" #include "tx_pkt.h" #include "rx_pkt.h" #include "etypes.h" #include "log.h" #include "quit.h" #include "qinq.h" #include "lconf.h" #include "prefetch.h" #include "defines.h" #include "prox_cfg.h" #include "hash_utils.h" #include "handle_lb_net.h" #include "toeplitz.h" #if RTE_VERSION < RTE_VERSION_NUM(1,8,0,0) #define RTE_CACHE_LINE_SIZE CACHE_LINE_SIZE #endif /* Load balancing based on one byte, figures out what type of packet is passed and depending on the type, pass the packet to the correct worker thread. If an unsupported packet type is used, the packet is simply dropped. This Load balancer can only handling QinQ packets (i.e. packets comming from the vCPE). */ int handle_lb_qinq_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts); int handle_lb_qinq_bulk_set_port(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts); struct task_lb_qinq { struct task_base base; uint8_t *worker_table; uint8_t bit_mask; uint8_t protocols_mask; uint8_t nb_worker_threads; uint16_t qinq_tag; }; static void init_task_lb_qinq(struct task_base *tbase, struct task_args *targ) { struct task_lb_qinq *task = (struct task_lb_qinq *)tbase; const int socket_id = rte_lcore_to_socket_id(targ->lconf->id); task->qinq_tag = targ->qinq_tag; task->nb_worker_threads = targ->nb_worker_threads; task->bit_mask = rte_is_power_of_2(targ->nb_worker_threads) ? targ->nb_worker_threads - 1 : 0xff; /* The load distributor is sending to a set of cores. These cores are responsible for handling a set of flows identified by a qinq tag. The load distributor identifies the flows and forwards them to the appropriate worker. The mapping from flow to worker is stored within the work_table. Build the worker_table by asking each worker which flows are handled. */ task->worker_table = prox_zmalloc(0x1000000, socket_id); for (int i = 0; i < targ->nb_worker_threads; ++i) { struct core_task ct = targ->core_task_set[0].core_task[i]; struct task_args *t = core_targ_get(ct.core, ct.task); PROX_PANIC(t->task_init->flow_iter.beg == NULL, "Load distributor can't find flows owned by destination worker %d\n", i); struct flow_iter *it = &t->task_init->flow_iter; int cnt = 0; for (it->beg(it, t); !it->is_end(it, t); it->next(it, t)) { uint16_t svlan = it->get_svlan(it, t); uint16_t cvlan = it->get_cvlan(it, t); task->worker_table[PKT_TO_LUTQINQ(svlan, cvlan)] = i; } } /* Check which protocols we are allowed to send to worker tasks */ for (int i = 0; i < MAX_PROTOCOLS; ++i) { int is_active = !!targ->core_task_set[i].n_elems; task->protocols_mask |= is_active << i; } plog_info("\t\ttask_lb_qinq protocols_mask = 0x%x\n", task->protocols_mask); if (targ->task_init->flag_features & TASK_FEATURE_LUT_QINQ_RSS) tbase->flags |= BASE_FLAG_LUT_QINQ_RSS; if (targ->task_init->flag_features & TASK_FEATURE_LUT_QINQ_HASH) tbase->flags |= BASE_FLAG_LUT_QINQ_HASH; plog_info("\t\ttask_lb_qinq flags = 0x%x\n", tbase->flags); } static struct task_init task_init_lb_qinq = { .mode_str = "lbqinq", .init = init_task_lb_qinq, .handle = handle_lb_qinq_bulk, .size = sizeof(struct task_lb_qinq) }; /* Add correct port id to mbufs coming from a DPDK ring port in the loadbalancer. For the split-bng using DPDK rings between the vSwitch and the VMs we need to know the port from which a packet was received. The ring PMD in dpdk does not update the port field in the mbuf and thus we have no control over the port numbers that are being used. This submode allows the loadbalancer to set the port number on which it received the mbuf. */ static struct task_init task_init_lb_qinq_set_port = { .mode_str = "lbqinq", .sub_mode_str = "lut_qinq_set_port", .init = init_task_lb_qinq, .handle = handle_lb_qinq_bulk_set_port, .size = sizeof(struct task_lb_qinq) }; /* Load Balance on Hash of combination of cvlan and svlan */ static struct task_init task_init_lb_qinq_hash_friend = { .mode_str = "lbqinq", .sub_mode_str ="lut_qinq_hash_friend", .init = init_task_lb_qinq, .handle = handle_lb_qinq_bulk, .flag_features = TASK_FEATURE_LUT_QINQ_HASH, .size = sizeof(struct task_lb_qinq) }; /* Load Balance on rss of combination of cvlan and svlan. This could be used to compare with HW implementations. */ static struct task_init task_init_lb_qinq_rss_friend = { .mode_str = "lbqinq", .sub_mode_str ="lut_qinq_rss_friend", .init = init_task_lb_qinq, .handle = handle_lb_qinq_bulk, .flag_features = TASK_FEATURE_LUT_QINQ_RSS, .size = sizeof(struct task_lb_qinq) }; __attribute__((constructor)) static void reg_task_lb_qinq(void) { reg_task(&task_init_lb_qinq); reg_task(&task_init_lb_qinq_hash_friend); reg_task(&task_init_lb_qinq_rss_friend); reg_task(&task_init_lb_qinq_set_port); } static inline uint8_t handle_lb_qinq(struct task_lb_qinq *task, struct rte_mbuf *mbuf); int handle_lb_qinq_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts) { struct task_lb_qinq *task = (struct task_lb_qinq *)tbase; uint8_t out[MAX_PKT_BURST]; uint16_t j; prefetch_first(mbufs, n_pkts); for (j = 0; j + PREFETCH_OFFSET < n_pkts; ++j) { #ifdef PROX_PREFETCH_OFFSET PREFETCH0(mbufs[j + PREFETCH_OFFSET]); PREFETCH0(rte_pktmbuf_mtod(mbufs[j + PREFETCH_OFFSET - 1], void *)); #endif out[j] = handle_lb_qinq(task, mbufs[j]); } #ifdef PROX_PREFETCH_OFFSET PREFETCH0(rte_pktmbuf_mtod(mbufs[n_pkts - 1], void *)); for (; j < n_pkts; ++j) { out[j] = handle_lb_qinq(task, mbufs[j]); } #endif return task->base.tx_pkt(&task->base, mbufs, n_pkts, out); } int handle_lb_qinq_bulk_set_port(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts) { struct task_lb_qinq *task = (struct task_lb_qinq *)tbase; uint8_t out[MAX_PKT_BURST]; uint16_t j; #if RTE_VERSION < RTE_VERSION_NUM(1,8,0,0) uint32_t port_id = mbufs[0]->pkt.in_port; #else uint32_t port_id = mbufs[0]->port; #endif if (tbase->rx_pkt == rx_pkt_hw) { port_id = tbase->rx_params_hw.last_read_portid + tbase->rx_params_hw.nb_rxports; port_id = ( port_id - 1 ) % tbase->rx_params_hw.nb_rxports; port_id = tbase->rx_params_hw.rx_pq[port_id].port; } else if (tbase->rx_pkt == rx_pkt_hw1) { port_id = tbase->rx_params_hw1.rx_pq.port; } prefetch_first(mbufs, n_pkts); for (j = 0; j + PREFETCH_OFFSET < n_pkts; ++j) { #ifdef PROX_PREFETCH_OFFSET PREFETCH0(mbufs[j + PREFETCH_OFFSET]); PREFETCH0(rte_pktmbuf_mtod(mbufs[j + PREFETCH_OFFSET - 1], void *)); #endif #if RTE_VERSION < RTE_VERSION_NUM(1,8,0,0) mbufs[j]->pkt.in_port = port_id; #else mbufs[j]->port = port_id; #endif out[j] = handle_lb_qinq(task, mbufs[j]); } #ifdef PROX_PREFETCH_OFFSET PREFETCH0(rte_pktmbuf_mtod(mbufs[n_pkts - 1], void *)); for (; j < n_pkts; ++j) { #if RTE_VERSION < RTE_VERSION_NUM(1,8,0,0) mbufs[j]->pkt.in_port = port_id; #else mbufs[j]->port = port_id; #endif out[j] = handle_lb_qinq(task, mbufs[j]); } #endif return task->base.tx_pkt(&task->base, mbufs, n_pkts, out); } struct qinq_packet { struct qinq_hdr qinq_hdr; union { prox_rte_ipv4_hdr ipv4_hdr; prox_rte_ipv6_hdr ipv6_hdr; }; } __attribute__((packed)) __attribute__((__aligned__(2))); struct qinq_packet_data { prox_rte_ether_addr d_addr; prox_rte_ether_addr s_addr; uint64_t qinq; } __attribute__((packed)) __attribute__((__aligned__(2))); struct ether_packet { prox_rte_ether_hdr ether_hdr; union { prox_rte_ipv4_hdr ipv4_hdr; prox_rte_ipv6_hdr ipv6_hdr; }; } __attribute__((packed)) __attribute__((__aligned__(2))); struct cpe_packet { union { struct qinq_packet qp; struct ether_packet ep; struct qinq_packet_data qd; }; }; static inline uint8_t get_worker(struct task_lb_qinq *task, struct cpe_packet *packet) { uint8_t worker = 0; if (((struct task_base *)task)->flags & BASE_FLAG_LUT_QINQ_HASH) { // Load Balance on Hash of combination of cvlan and svlan uint64_t qinq_net = packet->qd.qinq; qinq_net = qinq_net & 0xFF0F0000FF0F0000; // Mask Proto and QoS bits if (task->bit_mask != 0xff) { worker = rte_hash_crc(&qinq_net,8,0) & task->bit_mask; } else { worker = rte_hash_crc(&qinq_net,8,0) % task->nb_worker_threads; } plogx_dbg("Sending packet svlan=%x, cvlan=%x, pseudo_qinq=%lx to worker %d\n", rte_bswap16(0xFF0F & packet->qp.qinq_hdr.svlan.vlan_tci), rte_bswap16(0xFF0F & packet->qp.qinq_hdr.cvlan.vlan_tci), qinq_net, worker); } else if (((struct task_base *)task)->flags & BASE_FLAG_LUT_QINQ_RSS){ // Load Balance on rss of combination of cvlan and svlan uint32_t qinq = (packet->qp.qinq_hdr.cvlan.vlan_tci & 0xFF0F) << 16; uint32_t rss = toeplitz_hash((uint8_t *)&qinq, 4); if (task->bit_mask != 0xff) { worker = rss & task->bit_mask; } else { worker = (0x1ff & rss) % task->nb_worker_threads; } plogx_dbg("Sending packet svlan=%x, cvlan=%x, rss_input=%x, rss=%x to worker %d\n", rte_bswap16(0xFF0F & packet->qp.qinq_hdr.svlan.vlan_tci), rte_bswap16(0xFF0F & packet->qp.qinq_hdr.cvlan.vlan_tci), qinq, rss, worker); } else { uint16_t svlan = packet->qp.qinq_hdr.svlan.vlan_tci; uint16_t cvlan = packet->qp.qinq_hdr.cvlan.vlan_tci; prefetch_nta(&task->worker_table[PKT_TO_LUTQINQ(svlan, cvlan)]); worker = task->worker_table[PKT_TO_LUTQINQ(svlan, cvlan)]; const size_t pos = offsetof(struct cpe_packet, qp.qinq_hdr.cvlan.vlan_tci); plogx_dbg("qinq = %u, worker = %u, pos = %lu\n", rte_be_to_cpu_16(cvlan), worker, pos); } return worker; } static inline uint8_t handle_lb_qinq(struct task_lb_qinq *task, struct rte_mbuf *mbuf) { struct cpe_packet *packet = rte_pktmbuf_mtod(mbuf, struct cpe_packet*); if (packet->ep.ether_hdr.ether_type == ETYPE_IPv4) { if (unlikely((packet->ep.ipv4_hdr.version_ihl >> 4) != 4)) { plogx_err("Invalid Version %u for ETYPE_IPv4\n", packet->ep.ipv4_hdr.version_ihl); return OUT_DISCARD; } /* use 24 bits from the IP, clients are from the 10.0.0.0/8 network */ const uint32_t tmp = rte_bswap32(packet->ep.ipv4_hdr.src_addr) & 0x00FFFFFF; const uint32_t svlan = rte_bswap16(tmp >> 12); const uint32_t cvlan = rte_bswap16(tmp & 0x0FFF); prefetch_nta(&task->worker_table[PKT_TO_LUTQINQ(svlan, cvlan)]); uint8_t worker = task->worker_table[PKT_TO_LUTQINQ(svlan, cvlan)]; return worker + IPV4 * task->nb_worker_threads; } else if (unlikely(packet->qp.qinq_hdr.svlan.eth_proto != task->qinq_tag)) { /* might receive LLDP from the L2 switch... */ if (packet->qp.qinq_hdr.svlan.eth_proto != ETYPE_LLDP) { plogdx_err(mbuf, "Invalid packet for LB in QinQ mode\n"); } return OUT_DISCARD; } uint8_t worker = 0; uint8_t proto = 0xFF; switch (packet->qp.qinq_hdr.ether_type) { case ETYPE_IPv4: { if (unlikely((packet->qp.ipv4_hdr.version_ihl >> 4) != 4)) { plogx_err("Invalid Version %u for ETYPE_IPv4\n", packet->qp.ipv4_hdr.version_ihl); return OUT_DISCARD; } worker = get_worker(task, packet); proto = IPV4; break; } case ETYPE_IPv6: { if (unlikely((packet->qp.ipv4_hdr.version_ihl >> 4) != 6)) { plogx_err("Invalid Version %u for ETYPE_IPv6\n", packet->qp.ipv4_hdr.version_ihl); return OUT_DISCARD; } /* Use IP Destination when IPV6 QinQ */ if (task->bit_mask != 0xff) { worker = ((uint8_t *)packet)[61] & task->bit_mask; } else { worker = ((uint8_t *)packet)[61] % task->nb_worker_threads; } proto = IPV6; break; } case ETYPE_ARP: { // We can only send to ARP ring if it exists if (0 != (task->protocols_mask & (1 << ARP))) { proto = ARP; } else { proto = IPV4; } worker = get_worker(task, packet); break; } default: plogx_warn("Error in ETYPE_8021ad: ether_type = %#06x\n", packet->qp.qinq_hdr.ether_type); return OUT_DISCARD; } return worker + proto * task->nb_worker_threads; }