VNFs/DPPD-PROX/handle_swap.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_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 "igmp.h"
  30 #include "prox_cksum.h"
  31
  32 struct task_swap {
  33         struct task_base base;
  34         struct rte_mempool *igmp_pool;
  35         uint32_t runtime_flags;
  36         uint32_t igmp_address;
  37         uint8_t src_dst_mac[12];
  38         uint32_t local_ipv4;
  39         int offload_crc;
  40 };
  41
  42 #define NB_IGMP_MBUF            1024
  43 #define IGMP_MBUF_SIZE          2048
  44 #define NB_CACHE_IGMP_MBUF      256
  45
  46 static void write_src_and_dst_mac(struct task_swap *task, struct rte_mbuf *mbuf)
  47 {
  48         struct ether_hdr *hdr;
  49         struct ether_addr mac;
  50
  51         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))) {
  52                 /* Source and Destination mac hardcoded */
  53                 hdr = rte_pktmbuf_mtod(mbuf, struct ether_hdr *);
  54                 rte_memcpy(hdr, task->src_dst_mac, sizeof(task->src_dst_mac));
  55         } else {
  56                 hdr = rte_pktmbuf_mtod(mbuf, struct ether_hdr *);
  57                 if (likely((task->runtime_flags & TASK_ARG_SRC_MAC_SET) == 0)) {
  58                         /* dst mac will be used as src mac */
  59                         ether_addr_copy(&hdr->d_addr, &mac);
  60                 }
  61
  62                 if (unlikely(task->runtime_flags & TASK_ARG_DST_MAC_SET))
  63                         ether_addr_copy((struct ether_addr *)&task->src_dst_mac[0], &hdr->d_addr);
  64                 else
  65                         ether_addr_copy(&hdr->s_addr, &hdr->d_addr);
  66
  67                 if (unlikely(task->runtime_flags & TASK_ARG_SRC_MAC_SET)) {
  68                         ether_addr_copy((struct ether_addr *)&task->src_dst_mac[6], &hdr->s_addr);
  69                 } else {
  70                         ether_addr_copy(&mac, &hdr->s_addr);
  71                 }
  72         }
  73 }
  74
  75 static inline void build_mcast_mac(uint32_t ip, struct ether_addr *dst_mac)
  76 {
  77         // MAC address is 01:00:5e followed by 23 LSB of IP address
  78         uint64_t mac = 0x0000005e0001L | ((ip & 0xFFFF7F00L) << 16);
  79         memcpy(dst_mac, &mac, sizeof(struct ether_addr));
  80 }
  81
  82 static inline void build_igmp_message(struct task_base *tbase, struct rte_mbuf *mbuf, uint32_t ip, uint8_t igmp_message)
  83 {
  84         struct task_swap *task = (struct task_swap *)tbase;
  85         struct ether_hdr *hdr = rte_pktmbuf_mtod(mbuf, struct ether_hdr *);
  86         struct ether_addr dst_mac;
  87         build_mcast_mac(ip, &dst_mac);
  88
  89         rte_pktmbuf_pkt_len(mbuf) = 46;
  90         rte_pktmbuf_data_len(mbuf) = 46;
  91         init_mbuf_seg(mbuf);
  92
  93         ether_addr_copy(&dst_mac, &hdr->d_addr);
  94         ether_addr_copy((struct ether_addr *)&task->src_dst_mac[6], &hdr->s_addr);
  95         hdr->ether_type = ETYPE_IPv4;
  96
  97         struct ipv4_hdr *ip_hdr = (struct ipv4_hdr *)(hdr + 1);
  98         ip_hdr->version_ihl = 0x45;             /**< version and header length */
  99         ip_hdr->type_of_service = 0;    /**< type of service */
 100         ip_hdr->total_length = rte_cpu_to_be_16(32);            /**< length of packet */
 101         ip_hdr->packet_id = 0;          /**< packet ID */
 102         ip_hdr->fragment_offset = 0;    /**< fragmentation offset */
 103         ip_hdr->time_to_live = 1;               /**< time to live */
 104         ip_hdr->next_proto_id = IPPROTO_IGMP;           /**< protocol ID */
 105         ip_hdr->hdr_checksum = 0;               /**< header checksum */
 106         ip_hdr->src_addr = task->local_ipv4;            /**< source address */
 107         ip_hdr->dst_addr = ip;  /**< destination address */
 108         struct igmpv2_hdr *pigmp = (struct igmpv2_hdr *)(ip_hdr + 1);
 109         pigmp->type = igmp_message;
 110         pigmp->max_resp_time = 0;
 111         pigmp->checksum = 0;
 112         pigmp->group_address = ip;
 113         prox_ip_udp_cksum(mbuf, ip_hdr, sizeof(struct ether_hdr), sizeof(struct ipv4_hdr), task->offload_crc);
 114 }
 115
 116 static int handle_swap_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
 117 {
 118         struct task_swap *task = (struct task_swap *)tbase;
 119         struct ether_hdr *hdr;
 120         struct ether_addr mac;
 121         struct ipv4_hdr *ip_hdr;
 122         struct udp_hdr *udp_hdr;
 123         struct gre_hdr *pgre;
 124         struct ipv4_hdr *inner_ip_hdr;
 125         uint32_t ip;
 126         uint16_t port;
 127         uint8_t out[64] = {0};
 128         struct mpls_hdr *mpls;
 129         uint32_t mpls_len = 0;
 130         struct qinq_hdr *qinq;
 131         struct vlan_hdr *vlan;
 132         uint16_t j;
 133         struct igmpv2_hdr *pigmp;
 134         uint8_t type;
 135
 136         for (j = 0; j < n_pkts; ++j) {
 137                 PREFETCH0(mbufs[j]);
 138         }
 139         for (j = 0; j < n_pkts; ++j) {
 140                 PREFETCH0(rte_pktmbuf_mtod(mbufs[j], void *));
 141         }
 142
 143         // TODO 1: check packet is long enough for Ethernet + IP + UDP = 42 bytes
 144         for (uint16_t j = 0; j < n_pkts; ++j) {
 145                 hdr = rte_pktmbuf_mtod(mbufs[j], struct ether_hdr *);
 146                 switch (hdr->ether_type) {
 147                 case ETYPE_MPLSU:
 148                         mpls = (struct mpls_hdr *)(hdr + 1);
 149                         while (!(mpls->bytes & 0x00010000)) {
 150                                 // TODO: verify pcket length
 151                                 mpls++;
 152                                 mpls_len += sizeof(struct mpls_hdr);
 153                         }
 154                         mpls_len += sizeof(struct mpls_hdr);
 155                         ip_hdr = (struct ipv4_hdr *)(mpls + 1);
 156                         break;
 157                 case ETYPE_8021ad:
 158                         qinq = (struct qinq_hdr *)hdr;
 159                         if (qinq->cvlan.eth_proto != ETYPE_VLAN) {
 160                                 plog_warn("Unexpected proto in QinQ = %#04x\n", qinq->cvlan.eth_proto);
 161                                 out[j] = OUT_DISCARD;
 162                                 continue;
 163                         }
 164                         ip_hdr = (struct ipv4_hdr *)(qinq + 1);
 165                         break;
 166                 case ETYPE_VLAN:
 167                         vlan = (struct vlan_hdr *)(hdr + 1);
 168                         if (vlan->eth_proto == ETYPE_IPv4) {
 169                                 ip_hdr = (struct ipv4_hdr *)(vlan + 1);
 170                         } else if (vlan->eth_proto == ETYPE_VLAN) {
 171                                 vlan = (struct vlan_hdr *)(vlan + 1);
 172                                 if (vlan->eth_proto == ETYPE_IPv4) {
 173                                         ip_hdr = (struct ipv4_hdr *)(vlan + 1);
 174                                 }
 175                                 else if (vlan->eth_proto == ETYPE_IPv6) {
 176                                         plog_warn("Unsupported IPv6\n");
 177                                         out[j] = OUT_DISCARD;
 178                                         continue;
 179                                 }
 180                                 else {
 181                                         plog_warn("Unsupported packet type\n");
 182                                         out[j] = OUT_DISCARD;
 183                                         continue;
 184                                 }
 185                         } else {
 186                                 plog_warn("Unsupported packet type\n");
 187                                 out[j] = OUT_DISCARD;
 188                                 continue;
 189                         }
 190                         break;
 191                 case ETYPE_IPv4:
 192                         ip_hdr = (struct ipv4_hdr *)(hdr + 1);
 193                         break;
 194                 case ETYPE_IPv6:
 195                         plog_warn("Unsupported IPv6\n");
 196                         out[j] = OUT_DISCARD;
 197                         continue;
 198                 case ETYPE_LLDP:
 199                         out[j] = OUT_DISCARD;
 200                         continue;
 201                 default:
 202                         plog_warn("Unsupported ether_type 0x%x\n", hdr->ether_type);
 203                         out[j] = OUT_DISCARD;
 204                         continue;
 205                 }
 206                 // TODO 2 : check packet is long enough for Ethernet + IP + UDP + extra header (VLAN, MPLS, ...)
 207                 ip = ip_hdr->dst_addr;
 208
 209                 switch (ip_hdr->next_proto_id) {
 210                 case IPPROTO_GRE:
 211                         ip_hdr->dst_addr = ip_hdr->src_addr;
 212                         ip_hdr->src_addr = ip;
 213
 214                         pgre = (struct gre_hdr *)(ip_hdr + 1);
 215                         inner_ip_hdr = ((struct ipv4_hdr *)(pgre + 1));
 216                         ip = inner_ip_hdr->dst_addr;
 217                         inner_ip_hdr->dst_addr = inner_ip_hdr->src_addr;
 218                         inner_ip_hdr->src_addr = ip;
 219
 220                         udp_hdr = (struct udp_hdr *)(inner_ip_hdr + 1);
 221                         // TODO 3.1 : verify proto is UPD or TCP
 222                         port = udp_hdr->dst_port;
 223                         udp_hdr->dst_port = udp_hdr->src_port;
 224                         udp_hdr->src_port = port;
 225                         write_src_and_dst_mac(task, mbufs[j]);
 226                         break;
 227                 case IPPROTO_UDP:
 228                 case IPPROTO_TCP:
 229                         if (task->igmp_address && IS_IPV4_MCAST(rte_be_to_cpu_32(ip))) {
 230                                 out[j] = OUT_DISCARD;
 231                                 continue;
 232                         }
 233                         udp_hdr = (struct udp_hdr *)(ip_hdr + 1);
 234                         ip_hdr->dst_addr = ip_hdr->src_addr;
 235                         ip_hdr->src_addr = ip;
 236
 237                         port = udp_hdr->dst_port;
 238                         udp_hdr->dst_port = udp_hdr->src_port;
 239                         udp_hdr->src_port = port;
 240                         write_src_and_dst_mac(task, mbufs[j]);
 241                         break;
 242                 case IPPROTO_IGMP:
 243                         pigmp = (struct igmpv2_hdr *)(ip_hdr + 1);
 244                         // TODO: check packet len
 245                         type = pigmp->type;
 246                         if (type == IGMP_MEMBERSHIP_QUERY) {
 247                                 if (task->igmp_address) {
 248                                         // We have an address registered
 249                                         if ((task->igmp_address == pigmp->group_address) || (pigmp->group_address == 0)) {
 250                                                 // We get a request for the registered address, or to 0.0.0.0
 251                                                 build_igmp_message(tbase, mbufs[j], task->igmp_address, IGMP_MEMBERSHIP_REPORT);        // replace Membership query packet with a response
 252                                         } else {
 253                                                 // Discard as either we are not registered or this is a query for a different group
 254                                                 out[j] = OUT_DISCARD;
 255                                                 continue;
 256                                         }
 257                                 } else {
 258                                         // Discard as either we are not registered
 259                                         out[j] = OUT_DISCARD;
 260                                         continue;
 261                                 }
 262                         } else {
 263                                 // Do not forward other IGMP packets back
 264                                 out[j] = OUT_DISCARD;
 265                                 continue;
 266                         }
 267                         break;
 268                 default:
 269                         plog_warn("Unsupported IP protocol 0x%x\n", ip_hdr->next_proto_id);
 270                         out[j] = OUT_DISCARD;
 271                         continue;
 272                 }
 273         }
 274         return task->base.tx_pkt(&task->base, mbufs, n_pkts, out);
 275 }
 276
 277 void igmp_join_group(struct task_base *tbase, uint32_t igmp_address)
 278 {
 279         struct task_swap *task = (struct task_swap *)tbase;
 280         struct rte_mbuf *igmp_mbuf;
 281         uint8_t out[64] = {0};
 282         int ret;
 283
 284         task->igmp_address = igmp_address;
 285         ret = rte_mempool_get(task->igmp_pool, (void **)&igmp_mbuf);
 286         if (ret != 0) {
 287                 plog_err("Unable to allocate igmp mbuf\n");
 288                 return;
 289         }
 290         build_igmp_message(tbase, igmp_mbuf, task->igmp_address, IGMP_MEMBERSHIP_REPORT);
 291         task->base.tx_pkt(&task->base, &igmp_mbuf, 1, out);
 292 }
 293
 294 void igmp_leave_group(struct task_base *tbase)
 295 {
 296         struct task_swap *task = (struct task_swap *)tbase;
 297         struct rte_mbuf *igmp_mbuf;
 298         uint8_t out[64] = {0};
 299         int ret;
 300
 301         task->igmp_address = 0;
 302         ret = rte_mempool_get(task->igmp_pool, (void **)&igmp_mbuf);
 303         if (ret != 0) {
 304                 plog_err("Unable to allocate igmp mbuf\n");
 305                 return;
 306         }
 307         build_igmp_message(tbase, igmp_mbuf, task->igmp_address, IGMP_LEAVE_GROUP);
 308         task->base.tx_pkt(&task->base, &igmp_mbuf, 1, out);
 309 }
 310
 311 static void init_task_swap(struct task_base *tbase, struct task_args *targ)
 312 {
 313         struct task_swap *task = (struct task_swap *)tbase;
 314         struct ether_addr *src_addr, *dst_addr;
 315
 316         /*
 317          * The destination MAC of the outgoing packet is based on the config file:
 318          *    - 'dst mac=xx:xx:xx:xx:xx:xx' => the pre-configured mac will be used as dst mac
 319          *    - 'dst mac=packet'            => the src mac of the incoming packet is used as dst mac
 320          *    - (default - no 'dst mac')    => the src mac from the incoming packet is used as dst mac
 321          *
 322          * The source MAC of the outgoing packet is based on the config file:
 323          *    - 'src mac=xx:xx:xx:xx:xx:xx' => the pre-configured mac will be used as src mac
 324          *    - 'src mac=packet'            => the dst mac of the incoming packet is used as src mac
 325          *    - 'src mac=hw'                => the mac address of the tx port is used as src mac
 326          *                                     An error is returned if there are no physical tx ports
 327          *    - (default - no 'src mac')    => if there is physical tx port, the mac of that port is used as src mac
 328          *    - (default - no 'src mac')       if there are no physical tx ports the dst mac of the incoming packet
 329          */
 330
 331         if (targ->flags & TASK_ARG_DST_MAC_SET) {
 332                 dst_addr = &targ->edaddr;
 333                 memcpy(&task->src_dst_mac[0], dst_addr, sizeof(*src_addr));
 334         }
 335
 336         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");
 337         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");
 338
 339         if (targ->flags & TASK_ARG_SRC_MAC_SET) {
 340                 src_addr =  &targ->esaddr;
 341                 memcpy(&task->src_dst_mac[6], src_addr, sizeof(*dst_addr));
 342                 plog_info("\t\tCore %d: src mac set from config file\n", targ->lconf->id);
 343         } else {
 344                 if (targ->flags & TASK_ARG_HW_SRC_MAC)
 345                         PROX_PANIC(targ->nb_txports == 0, "src mac set to hw but no tx port\n");
 346                 if (targ->nb_txports) {
 347                         src_addr = &prox_port_cfg[task->base.tx_params_hw.tx_port_queue[0].port].eth_addr;
 348                         memcpy(&task->src_dst_mac[6], src_addr, sizeof(*dst_addr));
 349                         targ->flags |= TASK_ARG_SRC_MAC_SET;
 350                         plog_info("\t\tCore %d: src mac set from port\n", targ->lconf->id);
 351                 }
 352         }
 353         task->runtime_flags = targ->flags;
 354         task->igmp_address =  rte_cpu_to_be_32(targ->igmp_address);
 355         if (task->igmp_pool == NULL) {
 356                 static char name[] = "igmp0_pool";
 357                 name[4]++;
 358                 struct rte_mempool *ret = rte_mempool_create(name, NB_IGMP_MBUF, IGMP_MBUF_SIZE, NB_CACHE_IGMP_MBUF,
 359                         sizeof(struct rte_pktmbuf_pool_private), rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, 0,
 360                         rte_socket_id(), 0);
 361                 PROX_PANIC(ret == NULL, "Failed to allocate IGMP memory pool on socket %u with %u elements\n",
 362                         rte_socket_id(), NB_IGMP_MBUF);
 363                 plog_info("\t\tMempool %p (%s) size = %u * %u cache %u, socket %d\n", ret, name, NB_IGMP_MBUF,
 364                         IGMP_MBUF_SIZE, NB_CACHE_IGMP_MBUF, rte_socket_id());
 365                 task->igmp_pool = ret;
 366         }
 367         task->local_ipv4 = rte_cpu_to_be_32(targ->local_ipv4);
 368
 369         struct prox_port_cfg *port = find_reachable_port(targ);
 370         if (port) {
 371                 task->offload_crc = port->requested_tx_offload & (DEV_TX_OFFLOAD_IPV4_CKSUM | DEV_TX_OFFLOAD_UDP_CKSUM);
 372         }
 373 }
 374
 375 static struct task_init task_init_swap = {
 376         .mode_str = "swap",
 377         .init = init_task_swap,
 378         .handle = handle_swap_bulk,
 379         .flag_features = 0,
 380         .size = sizeof(struct task_swap),
 381 };
 382
 383 __attribute__((constructor)) static void reg_task_swap(void)
 384 {
 385         reg_task(&task_init_swap);
 386 }