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 void stop_swap(struct task_base *tbase)
 137 {
 138         struct task_swap *task = (struct task_swap *)tbase;
 139         if (task->igmp_pool) {
 140                 rte_mempool_free(task->igmp_pool);
 141                 task->igmp_pool = NULL;
 142         }
 143 }
 144
 145 static int handle_swap_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
 146 {
 147         struct task_swap *task = (struct task_swap *)tbase;
 148         prox_rte_ether_hdr *hdr;
 149         prox_rte_ether_addr mac;
 150         prox_rte_ipv4_hdr *ip_hdr;
 151         prox_rte_udp_hdr *udp_hdr;
 152         struct gre_hdr *pgre;
 153         prox_rte_ipv4_hdr *inner_ip_hdr;
 154         uint32_t ip;
 155         uint16_t port;
 156         uint8_t out[64] = {0};
 157         struct mpls_hdr *mpls;
 158         uint32_t mpls_len = 0;
 159         struct qinq_hdr *qinq;
 160         prox_rte_vlan_hdr *vlan;
 161         uint16_t j;
 162         struct igmpv2_hdr *pigmp;
 163         prox_rte_icmp_hdr *picmp;
 164         uint8_t type;
 165
 166         for (j = 0; j < n_pkts; ++j) {
 167                 PREFETCH0(mbufs[j]);
 168         }
 169         for (j = 0; j < n_pkts; ++j) {
 170                 PREFETCH0(rte_pktmbuf_mtod(mbufs[j], void *));
 171         }
 172
 173         // TODO 1: check packet is long enough for Ethernet + IP + UDP = 42 bytes
 174         for (uint16_t j = 0; j < n_pkts; ++j) {
 175                 hdr = rte_pktmbuf_mtod(mbufs[j], prox_rte_ether_hdr *);
 176                 switch (hdr->ether_type) {
 177                 case ETYPE_MPLSU:
 178                         mpls = (struct mpls_hdr *)(hdr + 1);
 179                         while (!(mpls->bytes & 0x00010000)) {
 180                                 // TODO: verify pcket length
 181                                 mpls++;
 182                                 mpls_len += sizeof(struct mpls_hdr);
 183                         }
 184                         mpls_len += sizeof(struct mpls_hdr);
 185                         ip_hdr = (prox_rte_ipv4_hdr *)(mpls + 1);
 186                         break;
 187                 case ETYPE_8021ad:
 188                         qinq = (struct qinq_hdr *)hdr;
 189                         if (qinq->cvlan.eth_proto != ETYPE_VLAN) {
 190                                 plog_warn("Unexpected proto in QinQ = %#04x\n", qinq->cvlan.eth_proto);
 191                                 out[j] = OUT_DISCARD;
 192                                 continue;
 193                         }
 194                         ip_hdr = (prox_rte_ipv4_hdr *)(qinq + 1);
 195                         break;
 196                 case ETYPE_VLAN:
 197                         vlan = (prox_rte_vlan_hdr *)(hdr + 1);
 198                         if (vlan->eth_proto == ETYPE_IPv4) {
 199                                 ip_hdr = (prox_rte_ipv4_hdr *)(vlan + 1);
 200                         } else if (vlan->eth_proto == ETYPE_VLAN) {
 201                                 vlan = (prox_rte_vlan_hdr *)(vlan + 1);
 202                                 if (vlan->eth_proto == ETYPE_IPv4) {
 203                                         ip_hdr = (prox_rte_ipv4_hdr *)(vlan + 1);
 204                                 }
 205                                 else if (vlan->eth_proto == ETYPE_IPv6) {
 206                                         plog_warn("Unsupported IPv6\n");
 207                                         out[j] = OUT_DISCARD;
 208                                         continue;
 209                                 }
 210                                 else {
 211                                         plog_warn("Unsupported packet type\n");
 212                                         out[j] = OUT_DISCARD;
 213                                         continue;
 214                                 }
 215                         } else {
 216                                 plog_warn("Unsupported packet type\n");
 217                                 out[j] = OUT_DISCARD;
 218                                 continue;
 219                         }
 220                         break;
 221                 case ETYPE_IPv4:
 222                         ip_hdr = (prox_rte_ipv4_hdr *)(hdr + 1);
 223                         break;
 224                 case ETYPE_IPv6:
 225                         plog_warn("Unsupported IPv6\n");
 226                         out[j] = OUT_DISCARD;
 227                         continue;
 228                 case ETYPE_LLDP:
 229                         out[j] = OUT_DISCARD;
 230                         continue;
 231                 default:
 232                         plog_warn("Unsupported ether_type 0x%x\n", hdr->ether_type);
 233                         out[j] = OUT_DISCARD;
 234                         continue;
 235                 }
 236                 // TODO 2 : check packet is long enough for Ethernet + IP + UDP + extra header (VLAN, MPLS, ...)
 237                 ip = ip_hdr->dst_addr;
 238
 239                 switch (ip_hdr->next_proto_id) {
 240                 case IPPROTO_GRE:
 241                         ip_hdr->dst_addr = ip_hdr->src_addr;
 242                         ip_hdr->src_addr = ip;
 243
 244                         pgre = (struct gre_hdr *)(ip_hdr + 1);
 245                         inner_ip_hdr = ((prox_rte_ipv4_hdr *)(pgre + 1));
 246                         ip = inner_ip_hdr->dst_addr;
 247                         inner_ip_hdr->dst_addr = inner_ip_hdr->src_addr;
 248                         inner_ip_hdr->src_addr = ip;
 249
 250                         udp_hdr = (prox_rte_udp_hdr *)(inner_ip_hdr + 1);
 251                         // TODO 3.1 : verify proto is UPD or TCP
 252                         port = udp_hdr->dst_port;
 253                         udp_hdr->dst_port = udp_hdr->src_port;
 254                         udp_hdr->src_port = port;
 255                         write_src_and_dst_mac(task, mbufs[j]);
 256                         break;
 257                 case IPPROTO_UDP:
 258                 case IPPROTO_TCP:
 259                         if (task->igmp_address && PROX_RTE_IS_IPV4_MCAST(rte_be_to_cpu_32(ip))) {
 260                                 out[j] = OUT_DISCARD;
 261                                 continue;
 262                         }
 263                         udp_hdr = (prox_rte_udp_hdr *)(ip_hdr + 1);
 264                         ip_hdr->dst_addr = ip_hdr->src_addr;
 265                         ip_hdr->src_addr = ip;
 266
 267                         port = udp_hdr->dst_port;
 268                         udp_hdr->dst_port = udp_hdr->src_port;
 269                         udp_hdr->src_port = port;
 270                         write_src_and_dst_mac(task, mbufs[j]);
 271                         break;
 272                 case IPPROTO_ICMP:
 273                         picmp = (prox_rte_icmp_hdr *)(ip_hdr + 1);
 274                         type = picmp->icmp_type;
 275                         if (type == PROX_RTE_IP_ICMP_ECHO_REQUEST) {
 276                                 if (ip_hdr->dst_addr == task->local_ipv4) {
 277                                         task->n_echo_req++;
 278                                         if (rte_rdtsc() - task->last_echo_req_rcvd_tsc > rte_get_tsc_hz()) {
 279                                                 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)));
 280                                                 task->n_echo_req = 0;
 281                                                 task->last_echo_req_rcvd_tsc = rte_rdtsc();
 282                                         }
 283                                         build_icmp_reply_message(tbase, mbufs[j]);
 284                                 } else {
 285                                         out[j] = OUT_DISCARD;
 286                                         continue;
 287                                 }
 288                         } else if (type == PROX_RTE_IP_ICMP_ECHO_REPLY) {
 289                                 if (ip_hdr->dst_addr == task->local_ipv4) {
 290                                         task->n_echo_rep++;
 291                                         if (rte_rdtsc() - task->last_echo_rep_rcvd_tsc > rte_get_tsc_hz()) {
 292                                                 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)));
 293                                                 task->n_echo_rep = 0;
 294                                                 task->last_echo_rep_rcvd_tsc = rte_rdtsc();
 295                                         }
 296                                 } else {
 297                                         out[j] = OUT_DISCARD;
 298                                         continue;
 299                                 }
 300                         } else {
 301                                 out[j] = OUT_DISCARD;
 302                                 continue;
 303                         }
 304                         break;
 305                 case IPPROTO_IGMP:
 306                         pigmp = (struct igmpv2_hdr *)(ip_hdr + 1);
 307                         // TODO: check packet len
 308                         type = pigmp->type;
 309                         if (type == IGMP_MEMBERSHIP_QUERY) {
 310                                 if (task->igmp_address) {
 311                                         // We have an address registered
 312                                         if ((task->igmp_address == pigmp->group_address) || (pigmp->group_address == 0)) {
 313                                                 // We get a request for the registered address, or to 0.0.0.0
 314                                                 build_igmp_message(tbase, mbufs[j], task->igmp_address, IGMP_MEMBERSHIP_REPORT);        // replace Membership query packet with a response
 315                                         } else {
 316                                                 // Discard as either we are not registered or this is a query for a different group
 317                                                 out[j] = OUT_DISCARD;
 318                                                 continue;
 319                                         }
 320                                 } else {
 321                                         // Discard as either we are not registered
 322                                         out[j] = OUT_DISCARD;
 323                                         continue;
 324                                 }
 325                         } else {
 326                                 // Do not forward other IGMP packets back
 327                                 out[j] = OUT_DISCARD;
 328                                 continue;
 329                         }
 330                         break;
 331                 default:
 332                         plog_warn("Unsupported IP protocol 0x%x\n", ip_hdr->next_proto_id);
 333                         out[j] = OUT_DISCARD;
 334                         continue;
 335                 }
 336         }
 337         return task->base.tx_pkt(&task->base, mbufs, n_pkts, out);
 338 }
 339
 340 void igmp_join_group(struct task_base *tbase, uint32_t igmp_address)
 341 {
 342         struct task_swap *task = (struct task_swap *)tbase;
 343         struct rte_mbuf *igmp_mbuf;
 344         uint8_t out[64] = {0};
 345         int ret;
 346
 347         task->igmp_address = igmp_address;
 348         ret = rte_mempool_get(task->igmp_pool, (void **)&igmp_mbuf);
 349         if (ret != 0) {
 350                 plog_err("Unable to allocate igmp mbuf\n");
 351                 return;
 352         }
 353         build_igmp_message(tbase, igmp_mbuf, task->igmp_address, IGMP_MEMBERSHIP_REPORT);
 354         task->base.tx_pkt(&task->base, &igmp_mbuf, 1, out);
 355 }
 356
 357 void igmp_leave_group(struct task_base *tbase)
 358 {
 359         struct task_swap *task = (struct task_swap *)tbase;
 360         struct rte_mbuf *igmp_mbuf;
 361         uint8_t out[64] = {0};
 362         int ret;
 363
 364         task->igmp_address = 0;
 365         ret = rte_mempool_get(task->igmp_pool, (void **)&igmp_mbuf);
 366         if (ret != 0) {
 367                 plog_err("Unable to allocate igmp mbuf\n");
 368                 return;
 369         }
 370         build_igmp_message(tbase, igmp_mbuf, task->igmp_address, IGMP_LEAVE_GROUP);
 371         task->base.tx_pkt(&task->base, &igmp_mbuf, 1, out);
 372 }
 373
 374 static void init_task_swap(struct task_base *tbase, struct task_args *targ)
 375 {
 376         struct task_swap *task = (struct task_swap *)tbase;
 377         prox_rte_ether_addr *src_addr, *dst_addr;
 378
 379         /*
 380          * The destination MAC of the outgoing packet is based on the config file:
 381          *    - 'dst mac=xx:xx:xx:xx:xx:xx' => the pre-configured mac will be used as dst mac
 382          *    - 'dst mac=packet'            => the src mac of the incoming packet is used as dst mac
 383          *    - (default - no 'dst mac')    => the src mac from the incoming packet is used as dst mac
 384          *
 385          * The source MAC of the outgoing packet is based on the config file:
 386          *    - 'src mac=xx:xx:xx:xx:xx:xx' => the pre-configured mac will be used as src mac
 387          *    - 'src mac=packet'            => the dst mac of the incoming packet is used as src mac
 388          *    - 'src mac=hw'                => the mac address of the tx port is used as src mac
 389          *                                     An error is returned if there are no physical tx ports
 390          *    - (default - no 'src mac')    => if there is physical tx port, the mac of that port is used as src mac
 391          *    - (default - no 'src mac')       if there are no physical tx ports the dst mac of the incoming packet
 392          */
 393
 394         if (targ->flags & TASK_ARG_DST_MAC_SET) {
 395                 dst_addr = &targ->edaddr;
 396                 memcpy(&task->src_dst_mac[0], dst_addr, sizeof(*src_addr));
 397         }
 398
 399         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");
 400         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");
 401
 402         if (targ->flags & TASK_ARG_SRC_MAC_SET) {
 403                 src_addr =  &targ->esaddr;
 404                 memcpy(&task->src_dst_mac[6], src_addr, sizeof(*dst_addr));
 405                 plog_info("\t\tCore %d: src mac set from config file\n", targ->lconf->id);
 406         } else {
 407                 if (targ->flags & TASK_ARG_HW_SRC_MAC)
 408                         PROX_PANIC(targ->nb_txports == 0, "src mac set to hw but no tx port\n");
 409                 if (targ->nb_txports) {
 410                         src_addr = &prox_port_cfg[task->base.tx_params_hw.tx_port_queue[0].port].eth_addr;
 411                         memcpy(&task->src_dst_mac[6], src_addr, sizeof(*dst_addr));
 412                         targ->flags |= TASK_ARG_SRC_MAC_SET;
 413                         plog_info("\t\tCore %d: src mac set from port\n", targ->lconf->id);
 414                 }
 415         }
 416         task->runtime_flags = targ->flags;
 417         task->igmp_address =  rte_cpu_to_be_32(targ->igmp_address);
 418         if (task->igmp_pool == NULL) {
 419                 static char name[] = "igmp0_pool";
 420                 name[4]++;
 421                 struct rte_mempool *ret = rte_mempool_create(name, NB_IGMP_MBUF, IGMP_MBUF_SIZE, NB_CACHE_IGMP_MBUF,
 422                         sizeof(struct rte_pktmbuf_pool_private), rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, 0,
 423                         rte_socket_id(), 0);
 424                 PROX_PANIC(ret == NULL, "Failed to allocate IGMP memory pool on socket %u with %u elements\n",
 425                         rte_socket_id(), NB_IGMP_MBUF);
 426                 plog_info("\t\tMempool %p (%s) size = %u * %u cache %u, socket %d\n", ret, name, NB_IGMP_MBUF,
 427                         IGMP_MBUF_SIZE, NB_CACHE_IGMP_MBUF, rte_socket_id());
 428                 task->igmp_pool = ret;
 429         }
 430         task->local_ipv4 = rte_cpu_to_be_32(targ->local_ipv4);
 431
 432         struct prox_port_cfg *port = find_reachable_port(targ);
 433         if (port) {
 434                 task->offload_crc = port->requested_tx_offload & (DEV_TX_OFFLOAD_IPV4_CKSUM | DEV_TX_OFFLOAD_UDP_CKSUM);
 435         }
 436 }
 437
 438 static struct task_init task_init_swap = {
 439         .mode_str = "swap",
 440         .init = init_task_swap,
 441         .handle = handle_swap_bulk,
 442         .flag_features = 0,
 443         .size = sizeof(struct task_swap),
 444         .stop_last = stop_swap
 445 };
 446
 447 __attribute__((constructor)) static void reg_task_swap(void)
 448 {
 449         reg_task(&task_init_swap);
 450 }