VNFs/DPPD-PROX/handle_routing.c

   1 /*
   2 // Copyright (c) 2010-2017 Intel Corporation
   3 //
   4 // Licensed under the Apache License, Version 2.0 (the "License");
   5 // you may not use this file except in compliance with the License.
   6 // You may obtain a copy of the License at
   7 //
   8 //     http://www.apache.org/licenses/LICENSE-2.0
   9 //
  10 // Unless required by applicable law or agreed to in writing, software
  11 // distributed under the License is distributed on an "AS IS" BASIS,
  12 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13 // See the License for the specific language governing permissions and
  14 // limitations under the License.
  15 */
  16
  17 #include <rte_lpm.h>
  18 #include <rte_cycles.h>
  19 #include <string.h>
  20 #include <rte_version.h>
  21 #include <rte_ip.h>
  22 #include <rte_byteorder.h>
  23
  24 #include "prox_lua.h"
  25 #include "prox_lua_types.h"
  26
  27 #include "quit.h"
  28 #include "log.h"
  29 #include "handle_routing.h"
  30 #include "tx_pkt.h"
  31 #include "gre.h"
  32 #include "lconf.h"
  33 #include "prox_port_cfg.h"
  34 #include "etypes.h"
  35 #include "prefetch.h"
  36 #include "hash_entry_types.h"
  37 #include "mpls.h"
  38 #include "qinq.h"
  39 #include "prox_cfg.h"
  40 #include "ip6_addr.h"
  41 #include "prox_shared.h"
  42 #include "prox_cksum.h"
  43 #include "mbuf_utils.h"
  44 #include "prox_compat.h"
  45
  46 struct task_routing {
  47         struct task_base                base;
  48         uint8_t                         runtime_flags;
  49         struct lcore_cfg                *lconf;
  50         struct rte_lpm                  *ipv4_lpm;
  51         struct next_hop                 *next_hops;
  52         uint32_t                        number_free_rules;
  53         uint16_t                        qinq_tag;
  54         uint32_t                        marking[4];
  55         uint64_t                        src_mac[PROX_MAX_PORTS];
  56 };
  57
  58 static void routing_update(struct task_base *tbase, void **data, uint16_t n_msgs)
  59 {
  60         struct task_routing *task = (struct task_routing *)tbase;
  61         struct route_msg *msg;
  62
  63         for (uint16_t i = 0; i < n_msgs; ++i) {
  64                 msg = (struct route_msg *)data[i];
  65
  66                 if (task->number_free_rules == 0) {
  67                         plog_warn("Failed adding route: %u.%u.%u.%u/%u: lpm table full\n",
  68                                 msg->ip_bytes[0], msg->ip_bytes[1], msg->ip_bytes[2],
  69                                 msg->ip_bytes[3], msg->prefix);
  70                 } else {
  71                         if (rte_lpm_add(task->ipv4_lpm, rte_bswap32(msg->ip), msg->prefix, msg->nh)) {
  72                                 plog_warn("Failed adding route: %u.%u.%u.%u/%u\n",
  73                                         msg->ip_bytes[0], msg->ip_bytes[1], msg->ip_bytes[2],
  74                                         msg->ip_bytes[3], msg->prefix);
  75                         } else {
  76                                 task->number_free_rules--;
  77                         }
  78                 }
  79         }
  80 }
  81
  82 static void init_task_routing(struct task_base *tbase, struct task_args *targ)
  83 {
  84         struct task_routing *task = (struct task_routing *)tbase;
  85         const int socket_id = rte_lcore_to_socket_id(targ->lconf->id);
  86         struct lpm4 *lpm;
  87
  88         task->lconf = targ->lconf;
  89         task->qinq_tag = targ->qinq_tag;
  90         task->runtime_flags = targ->runtime_flags;
  91
  92         PROX_PANIC(!strcmp(targ->route_table, ""), "route table not specified\n");
  93         if (targ->flags & TASK_ARG_LOCAL_LPM) {
  94                 int ret = lua_to_lpm4(prox_lua(), GLOBAL, targ->route_table, socket_id, &lpm);
  95                 PROX_PANIC(ret, "Failed to load IPv4 LPM:\n%s\n", get_lua_to_errors());
  96                 prox_sh_add_socket(socket_id, targ->route_table, lpm);
  97
  98                 task->number_free_rules = lpm->n_free_rules;
  99         }
 100         else {
 101                 lpm = prox_sh_find_socket(socket_id, targ->route_table);
 102                 if (!lpm) {
 103                         int ret = lua_to_lpm4(prox_lua(), GLOBAL, targ->route_table, socket_id, &lpm);
 104                         PROX_PANIC(ret, "Failed to load IPv4 LPM:\n%s\n", get_lua_to_errors());
 105                         prox_sh_add_socket(socket_id, targ->route_table, lpm);
 106                 }
 107         }
 108         task->ipv4_lpm = lpm->rte_lpm;
 109         task->next_hops = lpm->next_hops;
 110         task->number_free_rules = lpm->n_free_rules;
 111
 112         for (uint32_t i = 0; i < MAX_HOP_INDEX; i++) {
 113                 int tx_port = task->next_hops[i].mac_port.out_idx;
 114                 if ((tx_port > targ->nb_txports - 1) && (tx_port > targ->nb_txrings - 1)) {
 115                         PROX_PANIC(1, "Routing Table contains port %d but only %d tx port/ %d ring:\n", tx_port, targ->nb_txports, targ->nb_txrings);
 116                 }
 117         }
 118
 119         if (targ->nb_txrings) {
 120                 struct task_args *dtarg;
 121                 struct core_task ct;
 122                 for (uint32_t i = 0; i < targ->nb_txrings; ++i) {
 123                         ct = targ->core_task_set[0].core_task[i];
 124                         dtarg = core_targ_get(ct.core, ct.task);
 125                         dtarg = find_reachable_task_sending_to_port(dtarg);
 126                         if (task->runtime_flags & TASK_MPLS_TAGGING) {
 127                                 task->src_mac[i] = (0x0000ffffffffffff & ((*(uint64_t*)&prox_port_cfg[dtarg->tx_port_queue[0].port].eth_addr))) | ((uint64_t)ETYPE_MPLSU << (64 - 16));
 128                         } else {
 129                                 task->src_mac[i] = (0x0000ffffffffffff & ((*(uint64_t*)&prox_port_cfg[dtarg->tx_port_queue[0].port].eth_addr))) | ((uint64_t)ETYPE_IPv4 << (64 - 16));
 130                         }
 131                 }
 132         } else {
 133                 for (uint32_t i = 0; i < targ->nb_txports; ++i) {
 134                         if (task->runtime_flags & TASK_MPLS_TAGGING) {
 135                                 task->src_mac[i] = (0x0000ffffffffffff & ((*(uint64_t*)&prox_port_cfg[targ->tx_port_queue[i].port].eth_addr))) | ((uint64_t)ETYPE_MPLSU << (64 - 16));
 136                         } else {
 137                                 task->src_mac[i] = (0x0000ffffffffffff & ((*(uint64_t*)&prox_port_cfg[targ->tx_port_queue[i].port].eth_addr))) | ((uint64_t)ETYPE_IPv4 << (64 - 16));
 138                         }
 139                 }
 140         }
 141
 142         for (uint32_t i = 0; i < 4; ++i) {
 143                 task->marking[i] = rte_bswap32(targ->marking[i] << 9);
 144         }
 145
 146         struct prox_port_cfg *port = find_reachable_port(targ);
 147
 148         targ->lconf->ctrl_func_m[targ->task] = routing_update;
 149         targ->lconf->ctrl_timeout = freq_to_tsc(20);
 150 }
 151
 152 static inline uint8_t handle_routing(struct task_routing *task, struct rte_mbuf *mbuf);
 153
 154 static int handle_routing_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
 155 {
 156         struct task_routing *task = (struct task_routing *)tbase;
 157         uint8_t out[MAX_PKT_BURST];
 158         uint16_t j;
 159
 160         prefetch_first(mbufs, n_pkts);
 161
 162         for (j = 0; j + PREFETCH_OFFSET < n_pkts; ++j) {
 163 #ifdef PROX_PREFETCH_OFFSET
 164                 PREFETCH0(mbufs[j + PREFETCH_OFFSET]);
 165                 PREFETCH0(rte_pktmbuf_mtod(mbufs[j + PREFETCH_OFFSET - 1], void *));
 166 #endif
 167                 out[j] = handle_routing(task, mbufs[j]);
 168         }
 169 #ifdef PROX_PREFETCH_OFFSET
 170         PREFETCH0(rte_pktmbuf_mtod(mbufs[n_pkts - 1], void *));
 171         for (; j < n_pkts; ++j) {
 172                 out[j] = handle_routing(task, mbufs[j]);
 173         }
 174 #endif
 175
 176         return task->base.tx_pkt(&task->base, mbufs, n_pkts, out);
 177 }
 178
 179 static void set_l2(struct task_routing *task, struct rte_mbuf *mbuf, uint8_t nh_idx)
 180 {
 181         prox_rte_ether_hdr *peth = rte_pktmbuf_mtod(mbuf, prox_rte_ether_hdr *);
 182         *((uint64_t *)(&peth->d_addr)) = task->next_hops[nh_idx].mac_port_8bytes;
 183         *((uint64_t *)(&peth->s_addr)) = task->src_mac[task->next_hops[nh_idx].mac_port.out_idx];
 184 }
 185
 186 static void set_l2_mpls(struct task_routing *task, struct rte_mbuf *mbuf, uint8_t nh_idx)
 187 {
 188         prox_rte_ether_hdr *peth = (prox_rte_ether_hdr *)rte_pktmbuf_prepend(mbuf, sizeof(struct mpls_hdr));
 189
 190         *((uint64_t *)(&peth->d_addr)) = task->next_hops[nh_idx].mac_port_8bytes;
 191         *((uint64_t *)(&peth->s_addr)) = task->src_mac[task->next_hops[nh_idx].mac_port.out_idx];
 192         /* MPLSU ether_type written as high word of 64bit src_mac prepared by init_task_routing */
 193         struct mpls_hdr *mpls = (struct mpls_hdr *)(peth + 1);
 194
 195         if (task->runtime_flags & TASK_MARK) {
 196                   enum prox_rte_color color = rte_sched_port_pkt_read_color(mbuf);
 197                 *(uint32_t *)mpls = task->next_hops[nh_idx].mpls | task->marking[color] | 0x00010000; // Set BoS to 1
 198         }
 199         else {
 200                 *(uint32_t *)mpls = task->next_hops[nh_idx].mpls | 0x00010000; // Set BoS to 1
 201         }
 202 }
 203
 204 static uint8_t route_ipv4(struct task_routing *task, uint8_t *beg, uint32_t ip_offset, struct rte_mbuf *mbuf)
 205 {
 206         prox_rte_ipv4_hdr *ip = (prox_rte_ipv4_hdr*)(beg + ip_offset);
 207         prox_rte_ether_hdr *peth_out;
 208         uint8_t tx_port;
 209         uint32_t dst_ip;
 210
 211         if (unlikely(ip->version_ihl >> 4 != 4)) {
 212                 plog_warn("Offset: %d\n", ip_offset);
 213                 plog_warn("Expected to receive IPv4 packet but IP version was %d\n",
 214                         ip->version_ihl >> 4);
 215                 return OUT_DISCARD;
 216         }
 217
 218         switch(ip->next_proto_id) {
 219         case IPPROTO_GRE: {
 220                 struct gre_hdr *pgre = (struct gre_hdr *)(ip + 1);
 221                 dst_ip = ((prox_rte_ipv4_hdr *)(pgre + 1))->dst_addr;
 222                 break;
 223         }
 224         case IPPROTO_TCP:
 225         case IPPROTO_UDP:
 226                 dst_ip = ip->dst_addr;
 227                 break;
 228         default:
 229                 /* Routing for other protocols is not implemented */
 230                 return OUT_DISCARD;
 231         }
 232
 233 #if RTE_VERSION >= RTE_VERSION_NUM(16,4,0,1)
 234         uint32_t next_hop_index;
 235 #else
 236         uint8_t next_hop_index;
 237 #endif
 238         if (unlikely(rte_lpm_lookup(task->ipv4_lpm, rte_bswap32(dst_ip), &next_hop_index) != 0)) {
 239                 uint8_t* dst_ipp = (uint8_t*)&dst_ip;
 240                 plog_warn("lpm_lookup failed for ip %d.%d.%d.%d: rc = %d\n",
 241                         dst_ipp[0], dst_ipp[1], dst_ipp[2], dst_ipp[3], -ENOENT);
 242                 return OUT_DISCARD;
 243         }
 244
 245         tx_port = task->next_hops[next_hop_index].mac_port.out_idx;
 246         if (task->runtime_flags & TASK_MPLS_TAGGING) {
 247                 uint16_t padlen = rte_pktmbuf_pkt_len(mbuf) - rte_be_to_cpu_16(ip->total_length) - ip_offset;
 248                 if (padlen) {
 249                         rte_pktmbuf_trim(mbuf, padlen);
 250                 }
 251
 252                 set_l2_mpls(task, mbuf, next_hop_index);
 253         }
 254         else {
 255                 set_l2(task, mbuf, next_hop_index);
 256         }
 257         return tx_port;
 258 }
 259
 260 static inline uint8_t handle_routing(struct task_routing *task, struct rte_mbuf *mbuf)
 261 {
 262         struct qinq_hdr *qinq;
 263         prox_rte_ether_hdr *peth = rte_pktmbuf_mtod(mbuf, prox_rte_ether_hdr *);
 264
 265         switch (peth->ether_type) {
 266         case ETYPE_8021ad: {
 267                 struct qinq_hdr *qinq = (struct qinq_hdr *)peth;
 268                 if ((qinq->cvlan.eth_proto != ETYPE_VLAN)) {
 269                         plog_warn("Unexpected proto in QinQ = %#04x\n", qinq->cvlan.eth_proto);
 270                         return OUT_DISCARD;
 271                 }
 272
 273                 return route_ipv4(task, (uint8_t*)qinq, sizeof(*qinq), mbuf);
 274         }
 275         case ETYPE_IPv4:
 276                 return route_ipv4(task, (uint8_t*)peth, sizeof(*peth), mbuf);
 277         case ETYPE_MPLSU: {
 278                 /* skip MPLS headers if any for routing */
 279                 struct mpls_hdr *mpls = (struct mpls_hdr *)(peth + 1);
 280                 uint32_t count = sizeof(prox_rte_ether_hdr);
 281                 while (!(mpls->bytes & 0x00010000)) {
 282                         mpls++;
 283                         count += sizeof(struct mpls_hdr);
 284                 }
 285                 count += sizeof(struct mpls_hdr);
 286
 287                 return route_ipv4(task, (uint8_t*)peth, count, mbuf);
 288         }
 289         default:
 290                 if (peth->ether_type == task->qinq_tag) {
 291                         struct qinq_hdr *qinq = (struct qinq_hdr *)peth;
 292                         if ((qinq->cvlan.eth_proto != ETYPE_VLAN)) {
 293                                 plog_warn("Unexpected proto in QinQ = %#04x\n", qinq->cvlan.eth_proto);
 294                                 return OUT_DISCARD;
 295                         }
 296
 297                         return route_ipv4(task, (uint8_t*)qinq, sizeof(*qinq), mbuf);
 298                 }
 299                 plog_warn("Failed routing packet: ether_type %#06x is unknown\n", peth->ether_type);
 300                 return OUT_DISCARD;
 301         }
 302 }
 303
 304 static struct task_init task_init_routing = {
 305         .mode_str = "routing",
 306         .init = init_task_routing,
 307         .handle = handle_routing_bulk,
 308         .flag_features = TASK_FEATURE_ROUTING|TASK_FEATURE_TXQ_FLAGS_NOOFFLOADS,
 309         .size = sizeof(struct task_routing)
 310 };
 311
 312 __attribute__((constructor)) static void reg_task_routing(void)
 313 {
 314         reg_task(&task_init_routing);
 315 }