VNFs/DPPD-PROX/handle_swap.c

   1 /*
   2 // Copyright (c) 2010-2020 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_mbuf.h>
  18 #include <rte_udp.h>
  19
  20 #include "task_init.h"
  21 #include "task_base.h"
  22 #include "lconf.h"
  23 #include "log.h"
  24 #include "prox_port_cfg.h"
  25 #include "mpls.h"
  26 #include "qinq.h"
  27 #include "gre.h"
  28 #include "prefetch.h"
  29 #include "defines.h"
  30 #include "igmp.h"
  31 #include "prox_cksum.h"
  32 #include "prox_compat.h"
  33
  34 struct task_swap {
  35         struct task_base base;
  36         struct rte_mempool *igmp_pool;
  37         uint32_t runtime_flags;
  38         uint32_t igmp_address;
  39         uint8_t src_dst_mac[12];
  40         uint32_t local_ipv4;
  41         int offload_crc;
  42         uint64_t last_echo_req_rcvd_tsc;
  43         uint64_t last_echo_rep_rcvd_tsc;
  44         uint32_t n_echo_req;
  45         uint32_t n_echo_rep;
  46 };
  47
  48 #define NB_IGMP_MBUF            1024
  49 #define IGMP_MBUF_SIZE          2048
  50 #define NB_CACHE_IGMP_MBUF      256
  51
  52 static void write_src_and_dst_mac(struct task_swap *task, struct rte_mbuf *mbuf)
  53 {
  54         prox_rte_ether_hdr *hdr;
  55         prox_rte_ether_addr mac;
  56
  57         if (unlikely((task->runtime_flags & (TASK_ARG_DST_MAC_SET|TASK_ARG_SRC_MAC_SET)) == (TASK_ARG_DST_MAC_SET|TASK_ARG_SRC_MAC_SET))) {
  58                 /* Source and Destination mac hardcoded */
  59                 hdr = rte_pktmbuf_mtod(mbuf, prox_rte_ether_hdr *);
  60                 rte_memcpy(hdr, task->src_dst_mac, sizeof(task->src_dst_mac));
  61         } else {
  62                 hdr = rte_pktmbuf_mtod(mbuf, prox_rte_ether_hdr *);
  63                 if (likely((task->runtime_flags & TASK_ARG_SRC_MAC_SET) == 0)) {
  64                         /* dst mac will be used as src mac */
  65                         prox_rte_ether_addr_copy(&hdr->d_addr, &mac);
  66                 }
  67
  68                 if (unlikely(task->runtime_flags & TASK_ARG_DST_MAC_SET))
  69                         prox_rte_ether_addr_copy((prox_rte_ether_addr *)&task->src_dst_mac[0], &hdr->d_addr);
  70                 else
  71                         prox_rte_ether_addr_copy(&hdr->s_addr, &hdr->d_addr);
  72
  73                 if (unlikely(task->runtime_flags & TASK_ARG_SRC_MAC_SET)) {
  74                         prox_rte_ether_addr_copy((prox_rte_ether_addr *)&task->src_dst_mac[6], &hdr->s_addr);
  75                 } else {
  76                         prox_rte_ether_addr_copy(&mac, &hdr->s_addr);
  77                 }
  78         }
  79 }
  80 static inline void build_mcast_mac(uint32_t ip, prox_rte_ether_addr *dst_mac)
  81 {
  82         // MAC address is 01:00:5e followed by 23 LSB of IP address
  83         uint64_t mac = 0x0000005e0001L | ((ip & 0xFFFF7F00L) << 16);
  84         memcpy(dst_mac, &mac, sizeof(prox_rte_ether_addr));
  85 }
  86
  87 static inline void build_icmp_reply_message(struct task_base *tbase, struct rte_mbuf *mbuf)
  88 {
  89         struct task_swap *task = (struct task_swap *)tbase;
  90         prox_rte_ether_hdr *hdr = rte_pktmbuf_mtod(mbuf, prox_rte_ether_hdr *);
  91         prox_rte_ether_addr dst_mac;
  92         prox_rte_ether_addr_copy(&hdr->s_addr, &dst_mac);
  93         prox_rte_ether_addr_copy(&hdr->d_addr, &hdr->s_addr);
  94         prox_rte_ether_addr_copy(&dst_mac, &hdr->d_addr);
  95         prox_rte_ipv4_hdr *ip_hdr = (prox_rte_ipv4_hdr *)(hdr + 1);
  96         ip_hdr->dst_addr = ip_hdr->src_addr;
  97         ip_hdr->src_addr = task->local_ipv4;
  98         prox_rte_icmp_hdr *picmp = (prox_rte_icmp_hdr *)(ip_hdr + 1);
  99         picmp->icmp_type = PROX_RTE_IP_ICMP_ECHO_REPLY;
 100 }
 101
 102 static inline void build_igmp_message(struct task_base *tbase, struct rte_mbuf *mbuf, uint32_t ip, uint8_t igmp_message)
 103 {
 104         struct task_swap *task = (struct task_swap *)tbase;
 105         prox_rte_ether_hdr *hdr = rte_pktmbuf_mtod(mbuf, prox_rte_ether_hdr *);
 106         prox_rte_ether_addr dst_mac;
 107         build_mcast_mac(ip, &dst_mac);
 108
 109         rte_pktmbuf_pkt_len(mbuf) = 46;
 110         rte_pktmbuf_data_len(mbuf) = 46;
 111         init_mbuf_seg(mbuf);
 112
 113         prox_rte_ether_addr_copy(&dst_mac, &hdr->d_addr);
 114         prox_rte_ether_addr_copy((prox_rte_ether_addr *)&task->src_dst_mac[6], &hdr->s_addr);
 115         hdr->ether_type = ETYPE_IPv4;
 116
 117         prox_rte_ipv4_hdr *ip_hdr = (prox_rte_ipv4_hdr *)(hdr + 1);
 118         ip_hdr->version_ihl = 0x45;             /**< version and header length */
 119         ip_hdr->type_of_service = 0;    /**< type of service */
 120         ip_hdr->total_length = rte_cpu_to_be_16(32);            /**< length of packet */
 121         ip_hdr->packet_id = 0;          /**< packet ID */
 122         ip_hdr->fragment_offset = 0;    /**< fragmentation offset */
 123         ip_hdr->time_to_live = 1;               /**< time to live */
 124         ip_hdr->next_proto_id = IPPROTO_IGMP;           /**< protocol ID */
 125         ip_hdr->hdr_checksum = 0;               /**< header checksum */
 126         ip_hdr->src_addr = task->local_ipv4;            /**< source address */
 127         ip_hdr->dst_addr = ip;  /**< destination address */
 128         struct igmpv2_hdr *pigmp = (struct igmpv2_hdr *)(ip_hdr + 1);
 129         pigmp->type = igmp_message;
 130         pigmp->max_resp_time = 0;
 131         pigmp->checksum = 0;
 132         pigmp->group_address = ip;
 133         prox_ip_udp_cksum(mbuf, ip_hdr, sizeof(prox_rte_ether_hdr), sizeof(prox_rte_ipv4_hdr), task->offload_crc);
 134 }
 135
 136 static int handle_swap_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
 137 {
 138         struct task_swap *task = (struct task_swap *)tbase;
 139         prox_rte_ether_hdr *hdr;
 140         prox_rte_ether_addr mac;
 141         prox_rte_ipv4_hdr *ip_hdr;
 142         prox_rte_udp_hdr *udp_hdr;
 143         struct gre_hdr *pgre;
 144         prox_rte_ipv4_hdr *inner_ip_hdr;
 145         uint32_t ip;
 146         uint16_t port;
 147         uint8_t out[64] = {0};
 148         struct mpls_hdr *mpls;
 149         uint32_t mpls_len = 0;
 150         struct qinq_hdr *qinq;
 151         prox_rte_vlan_hdr *vlan;
 152         uint16_t j;
 153         struct igmpv2_hdr *pigmp;
 154         prox_rte_icmp_hdr *picmp;
 155         uint8_t type;
 156
 157         for (j = 0; j < n_pkts; ++j) {
 158                 PREFETCH0(mbufs[j]);
 159         }
 160         for (j = 0; j < n_pkts; ++j) {
 161                 PREFETCH0(rte_pktmbuf_mtod(mbufs[j], void *));
 162         }
 163
 164         // TODO 1: check packet is long enough for Ethernet + IP + UDP = 42 bytes
 165         for (uint16_t j = 0; j < n_pkts; ++j) {
 166                 hdr = rte_pktmbuf_mtod(mbufs[j], prox_rte_ether_hdr *);
 167                 switch (hdr->ether_type) {
 168                 case ETYPE_MPLSU:
 169                         mpls = (struct mpls_hdr *)(hdr + 1);
 170                         while (!(mpls->bytes & 0x00010000)) {
 171                                 // TODO: verify pcket length
 172                                 mpls++;
 173                                 mpls_len += sizeof(struct mpls_hdr);
 174                         }
 175                         mpls_len += sizeof(struct mpls_hdr);
 176                         ip_hdr = (prox_rte_ipv4_hdr *)(mpls + 1);
 177                         break;
 178                 case ETYPE_8021ad:
 179                         qinq = (struct qinq_hdr *)hdr;
 180                         if (qinq->cvlan.eth_proto != ETYPE_VLAN) {
 181                                 plog_warn("Unexpected proto in QinQ = %#04x\n", qinq->cvlan.eth_proto);
 182                                 out[j] = OUT_DISCARD;
 183                                 continue;
 184                         }
 185                         ip_hdr = (prox_rte_ipv4_hdr *)(qinq + 1);
 186                         break;
 187                 case ETYPE_VLAN:
 188                         vlan = (prox_rte_vlan_hdr *)(hdr + 1);
 189                         if (vlan->eth_proto == ETYPE_IPv4) {
 190                                 ip_hdr = (prox_rte_ipv4_hdr *)(vlan + 1);
 191                         } else if (vlan->eth_proto == ETYPE_VLAN) {
 192                                 vlan = (prox_rte_vlan_hdr *)(vlan + 1);
 193                                 if (vlan->eth_proto == ETYPE_IPv4) {
 194                                         ip_hdr = (prox_rte_ipv4_hdr *)(vlan + 1);
 195                                 }
 196                                 else if (vlan->eth_proto == ETYPE_IPv6) {
 197                                         plog_warn("Unsupported IPv6\n");
 198                                         out[j] = OUT_DISCARD;
 199                                         continue;
 200                                 }
 201                                 else {
 202                                         plog_warn("Unsupported packet type\n");
 203                                         out[j] = OUT_DISCARD;
 204                                         continue;
 205                                 }
 206                         } else {
 207                                 plog_warn("Unsupported packet type\n");
 208                                 out[j] = OUT_DISCARD;
 209                                 continue;
 210                         }
 211                         break;
 212                 case ETYPE_IPv4:
 213                         ip_hdr = (prox_rte_ipv4_hdr *)(hdr + 1);
 214                         break;
 215                 case ETYPE_IPv6:
 216                         plog_warn("Unsupported IPv6\n");
 217                         out[j] = OUT_DISCARD;
 218                         continue;
 219                 case ETYPE_LLDP:
 220                         out[j] = OUT_DISCARD;
 221                         continue;
 222                 default:
 223                         plog_warn("Unsupported ether_type 0x%x\n", hdr->ether_type);
 224                         out[j] = OUT_DISCARD;
 225                         continue;
 226                 }
 227                 // TODO 2 : check packet is long enough for Ethernet + IP + UDP + extra header (VLAN, MPLS, ...)
 228                 ip = ip_hdr->dst_addr;
 229
 230                 switch (ip_hdr->next_proto_id) {
 231                 case IPPROTO_GRE:
 232                         ip_hdr->dst_addr = ip_hdr->src_addr;
 233                         ip_hdr->src_addr = ip;
 234
 235                         pgre = (struct gre_hdr *)(ip_hdr + 1);
 236                         inner_ip_hdr = ((prox_rte_ipv4_hdr *)(pgre + 1));
 237                         ip = inner_ip_hdr->dst_addr;
 238                         inner_ip_hdr->dst_addr = inner_ip_hdr->src_addr;
 239                         inner_ip_hdr->src_addr = ip;
 240
 241                         udp_hdr = (prox_rte_udp_hdr *)(inner_ip_hdr + 1);
 242                         // TODO 3.1 : verify proto is UPD or TCP
 243                         port = udp_hdr->dst_port;
 244                         udp_hdr->dst_port = udp_hdr->src_port;
 245                         udp_hdr->src_port = port;
 246                         write_src_and_dst_mac(task, mbufs[j]);
 247                         break;
 248                 case IPPROTO_UDP:
 249                 case IPPROTO_TCP:
 250                         if (task->igmp_address && PROX_RTE_IS_IPV4_MCAST(rte_be_to_cpu_32(ip))) {
 251                                 out[j] = OUT_DISCARD;
 252                                 continue;
 253                         }
 254                         udp_hdr = (prox_rte_udp_hdr *)(ip_hdr + 1);
 255                         ip_hdr->dst_addr = ip_hdr->src_addr;
 256                         ip_hdr->src_addr = ip;
 257
 258                         port = udp_hdr->dst_port;
 259                         udp_hdr->dst_port = udp_hdr->src_port;
 260                         udp_hdr->src_port = port;
 261                         write_src_and_dst_mac(task, mbufs[j]);
 262                         break;
 263                 case IPPROTO_ICMP:
 264                         picmp = (prox_rte_icmp_hdr *)(ip_hdr + 1);
 265                         type = picmp->icmp_type;
 266                         if (type == PROX_RTE_IP_ICMP_ECHO_REQUEST) {
 267                                 if (ip_hdr->dst_addr == task->local_ipv4) {
 268                                         task->n_echo_req++;
 269                                         if (rte_rdtsc() - task->last_echo_req_rcvd_tsc > rte_get_tsc_hz()) {
 270                                                 plog_info("Received %u Echo Request on IP "IPv4_BYTES_FMT" (last received from IP "IPv4_BYTES_FMT")\n", task->n_echo_req, IPv4_BYTES(((uint8_t*)&ip_hdr->dst_addr)), IPv4_BYTES(((uint8_t*)&ip_hdr->src_addr)));
 271                                                 task->n_echo_req = 0;
 272                                                 task->last_echo_req_rcvd_tsc = rte_rdtsc();
 273                                         }
 274                                         build_icmp_reply_message(tbase, mbufs[j]);
 275                                 } else {
 276                                         out[j] = OUT_DISCARD;
 277                                         continue;
 278                                 }
 279                         } else if (type == PROX_RTE_IP_ICMP_ECHO_REPLY) {
 280                                 if (ip_hdr->dst_addr == task->local_ipv4) {
 281                                         task->n_echo_rep++;
 282                                         if (rte_rdtsc() - task->last_echo_rep_rcvd_tsc > rte_get_tsc_hz()) {
 283                                                 plog_info("Received %u Echo Reply on IP "IPv4_BYTES_FMT" (last received from IP "IPv4_BYTES_FMT")\n", task->n_echo_rep, IPv4_BYTES(((uint8_t*)&ip_hdr->dst_addr)), IPv4_BYTES(((uint8_t*)&ip_hdr->src_addr)));
 284                                                 task->n_echo_rep = 0;
 285                                                 task->last_echo_rep_rcvd_tsc = rte_rdtsc();
 286                                         }
 287                                 } else {
 288                                         out[j] = OUT_DISCARD;
 289                                         continue;
 290                                 }
 291                         } else {
 292                                 out[j] = OUT_DISCARD;
 293                                 continue;
 294                         }
 295                         break;
 296                 case IPPROTO_IGMP:
 297                         pigmp = (struct igmpv2_hdr *)(ip_hdr + 1);
 298                         // TODO: check packet len
 299                         type = pigmp->type;
 300                         if (type == IGMP_MEMBERSHIP_QUERY) {
 301                                 if (task->igmp_address) {
 302                                         // We have an address registered
 303                                         if ((task->igmp_address == pigmp->group_address) || (pigmp->group_address == 0)) {
 304                                                 // We get a request for the registered address, or to 0.0.0.0
 305                                                 build_igmp_message(tbase, mbufs[j], task->igmp_address, IGMP_MEMBERSHIP_REPORT);        // replace Membership query packet with a response
 306                                         } else {
 307                                                 // Discard as either we are not registered or this is a query for a different group
 308                                                 out[j] = OUT_DISCARD;
 309                                                 continue;
 310                                         }
 311                                 } else {
 312                                         // Discard as either we are not registered
 313                                         out[j] = OUT_DISCARD;
 314                                         continue;
 315                                 }
 316                         } else {
 317                                 // Do not forward other IGMP packets back
 318                                 out[j] = OUT_DISCARD;
 319                                 continue;
 320                         }
 321                         break;
 322                 default:
 323                         plog_warn("Unsupported IP protocol 0x%x\n", ip_hdr->next_proto_id);
 324                         out[j] = OUT_DISCARD;
 325                         continue;
 326                 }
 327         }
 328         return task->base.tx_pkt(&task->base, mbufs, n_pkts, out);
 329 }
 330
 331 void igmp_join_group(struct task_base *tbase, uint32_t igmp_address)
 332 {
 333         struct task_swap *task = (struct task_swap *)tbase;
 334         struct rte_mbuf *igmp_mbuf;
 335         uint8_t out[64] = {0};
 336         int ret;
 337
 338         task->igmp_address = igmp_address;
 339         ret = rte_mempool_get(task->igmp_pool, (void **)&igmp_mbuf);
 340         if (ret != 0) {
 341                 plog_err("Unable to allocate igmp mbuf\n");
 342                 return;
 343         }
 344         build_igmp_message(tbase, igmp_mbuf, task->igmp_address, IGMP_MEMBERSHIP_REPORT);
 345         task->base.tx_pkt(&task->base, &igmp_mbuf, 1, out);
 346 }
 347
 348 void igmp_leave_group(struct task_base *tbase)
 349 {
 350         struct task_swap *task = (struct task_swap *)tbase;
 351         struct rte_mbuf *igmp_mbuf;
 352         uint8_t out[64] = {0};
 353         int ret;
 354
 355         task->igmp_address = 0;
 356         ret = rte_mempool_get(task->igmp_pool, (void **)&igmp_mbuf);
 357         if (ret != 0) {
 358                 plog_err("Unable to allocate igmp mbuf\n");
 359                 return;
 360         }
 361         build_igmp_message(tbase, igmp_mbuf, task->igmp_address, IGMP_LEAVE_GROUP);
 362         task->base.tx_pkt(&task->base, &igmp_mbuf, 1, out);
 363 }
 364
 365 static void init_task_swap(struct task_base *tbase, struct task_args *targ)
 366 {
 367         struct task_swap *task = (struct task_swap *)tbase;
 368         prox_rte_ether_addr *src_addr, *dst_addr;
 369
 370         /*
 371          * The destination MAC of the outgoing packet is based on the config file:
 372          *    - 'dst mac=xx:xx:xx:xx:xx:xx' => the pre-configured mac will be used as dst mac
 373          *    - 'dst mac=packet'            => the src mac of the incoming packet is used as dst mac
 374          *    - (default - no 'dst mac')    => the src mac from the incoming packet is used as dst mac
 375          *
 376          * The source MAC of the outgoing packet is based on the config file:
 377          *    - 'src mac=xx:xx:xx:xx:xx:xx' => the pre-configured mac will be used as src mac
 378          *    - 'src mac=packet'            => the dst mac of the incoming packet is used as src mac
 379          *    - 'src mac=hw'                => the mac address of the tx port is used as src mac
 380          *                                     An error is returned if there are no physical tx ports
 381          *    - (default - no 'src mac')    => if there is physical tx port, the mac of that port is used as src mac
 382          *    - (default - no 'src mac')       if there are no physical tx ports the dst mac of the incoming packet
 383          */
 384
 385         if (targ->flags & TASK_ARG_DST_MAC_SET) {
 386                 dst_addr = &targ->edaddr;
 387                 memcpy(&task->src_dst_mac[0], dst_addr, sizeof(*src_addr));
 388         }
 389
 390         PROX_PANIC(targ->flags & TASK_ARG_DO_NOT_SET_SRC_MAC, "src mac must be set in swap mode, by definition => src mac=no is not supported\n");
 391         PROX_PANIC(targ->flags & TASK_ARG_DO_NOT_SET_DST_MAC, "dst mac must be set in swap mode, by definition => dst mac=no is not supported\n");
 392
 393         if (targ->flags & TASK_ARG_SRC_MAC_SET) {
 394                 src_addr =  &targ->esaddr;
 395                 memcpy(&task->src_dst_mac[6], src_addr, sizeof(*dst_addr));
 396                 plog_info("\t\tCore %d: src mac set from config file\n", targ->lconf->id);
 397         } else {
 398                 if (targ->flags & TASK_ARG_HW_SRC_MAC)
 399                         PROX_PANIC(targ->nb_txports == 0, "src mac set to hw but no tx port\n");
 400                 if (targ->nb_txports) {
 401                         src_addr = &prox_port_cfg[task->base.tx_params_hw.tx_port_queue[0].port].eth_addr;
 402                         memcpy(&task->src_dst_mac[6], src_addr, sizeof(*dst_addr));
 403                         targ->flags |= TASK_ARG_SRC_MAC_SET;
 404                         plog_info("\t\tCore %d: src mac set from port\n", targ->lconf->id);
 405                 }
 406         }
 407         task->runtime_flags = targ->flags;
 408         task->igmp_address =  rte_cpu_to_be_32(targ->igmp_address);
 409         if (task->igmp_pool == NULL) {
 410                 static char name[] = "igmp0_pool";
 411                 name[4]++;
 412                 struct rte_mempool *ret = rte_mempool_create(name, NB_IGMP_MBUF, IGMP_MBUF_SIZE, NB_CACHE_IGMP_MBUF,
 413                         sizeof(struct rte_pktmbuf_pool_private), rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, 0,
 414                         rte_socket_id(), 0);
 415                 PROX_PANIC(ret == NULL, "Failed to allocate IGMP memory pool on socket %u with %u elements\n",
 416                         rte_socket_id(), NB_IGMP_MBUF);
 417                 plog_info("\t\tMempool %p (%s) size = %u * %u cache %u, socket %d\n", ret, name, NB_IGMP_MBUF,
 418                         IGMP_MBUF_SIZE, NB_CACHE_IGMP_MBUF, rte_socket_id());
 419                 task->igmp_pool = ret;
 420         }
 421         task->local_ipv4 = rte_cpu_to_be_32(targ->local_ipv4);
 422
 423         struct prox_port_cfg *port = find_reachable_port(targ);
 424         if (port) {
 425                 task->offload_crc = port->requested_tx_offload & (DEV_TX_OFFLOAD_IPV4_CKSUM | DEV_TX_OFFLOAD_UDP_CKSUM);
 426         }
 427 }
 428
 429 static struct task_init task_init_swap = {
 430         .mode_str = "swap",
 431         .init = init_task_swap,
 432         .handle = handle_swap_bulk,
 433         .flag_features = 0,
 434         .size = sizeof(struct task_swap),
 435 };
 436
 437 __attribute__((constructor)) static void reg_task_swap(void)
 438 {
 439         reg_task(&task_init_swap);
 440 }