VNFs/DPPD-PROX/handle_qinq_encap4.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_table_hash.h>
  18 #include <rte_hash_crc.h>
  19 #include <rte_cycles.h>
  20
  21 #include "mbuf_utils.h"
  22 #include "prox_malloc.h"
  23 #include "prox_lua.h"
  24 #include "prox_lua_types.h"
  25 #include "handle_qinq_encap4.h"
  26 #include "handle_qinq_decap4.h"
  27 #include "prox_args.h"
  28 #include "defines.h"
  29 #include "tx_pkt.h"
  30 #include "prefetch.h"
  31 #include "pkt_prototypes.h"
  32 #include "hash_entry_types.h"
  33 #include "task_init.h"
  34 #include "bng_pkts.h"
  35 #include "prox_cksum.h"
  36 #include "hash_utils.h"
  37 #include "quit.h"
  38 #include "prox_port_cfg.h"
  39 #include "handle_lb_net.h"
  40 #include "prox_cfg.h"
  41 #include "cfgfile.h"
  42 #include "toeplitz.h"
  43 #include "prox_shared.h"
  44 #include "prox_compat.h"
  45
  46 static struct cpe_table_data *read_cpe_table_config(const char *name, uint8_t socket)
  47 {
  48         struct lua_State *L = prox_lua();
  49         struct cpe_table_data *ret = NULL;
  50
  51         lua_getglobal(L, name);
  52         PROX_PANIC(lua_isnil(L, -1), "Coudn't find cpe_table data\n");
  53
  54         return ret;
  55 }
  56
  57 struct qinq_gre_map *get_qinq_gre_map(struct task_args *targ)
  58 {
  59         const int socket_id = rte_lcore_to_socket_id(targ->lconf->id);
  60         struct qinq_gre_map *ret = prox_sh_find_socket(socket_id, "qinq_gre_map");
  61
  62         if (!ret) {
  63                 PROX_PANIC(!strcmp(targ->user_table, ""), "No user table defined\n");
  64                 int rv = lua_to_qinq_gre_map(prox_lua(), GLOBAL, targ->user_table, socket_id, &ret);
  65                 PROX_PANIC(rv, "Error reading mapping between qinq and gre from qinq_gre_map: \n%s\n",
  66                            get_lua_to_errors());
  67                 prox_sh_add_socket(socket_id, "qinq_gre_map", ret);
  68         }
  69         return ret;
  70 }
  71
  72 /* Encapsulate IPv4 packets in QinQ. QinQ tags are derived from gre_id. */
  73 int handle_qinq_encap4_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts);
  74 static void arp_msg(struct task_base *tbase, void **data, uint16_t n_msgs);
  75
  76 static void fill_table(struct task_args *targ, struct rte_table_hash *table)
  77 {
  78         struct cpe_table_data *cpe_table_data;
  79         const int socket_id = rte_lcore_to_socket_id(targ->lconf->id);
  80         int ret = lua_to_cpe_table_data(prox_lua(), GLOBAL, targ->cpe_table_name, socket_id, &cpe_table_data);
  81         const uint8_t n_slaves = targ->nb_slave_threads;
  82         const uint8_t worker_id = targ->worker_thread_id;
  83
  84         for (uint32_t i = 0; i < cpe_table_data->n_entries; ++i) {
  85                 if (rte_bswap32(cpe_table_data->entries[i].ip) % n_slaves != worker_id) {
  86                         continue;
  87                 }
  88                 struct cpe_table_entry *entry = &cpe_table_data->entries[i];
  89
  90                 uint32_t port_idx = prox_cfg.cpe_table_ports[entry->port_idx];
  91                 PROX_PANIC(targ->mapping[port_idx] == 255, "Error reading cpe table: Mapping for port %d is missing", port_idx);
  92
  93                 struct cpe_key key = {
  94                         .ip = entry->ip,
  95                         .gre_id = entry->gre_id,
  96                 };
  97
  98                 struct cpe_data data = {
  99                         .qinq_svlan = entry->svlan,
 100                         .qinq_cvlan = entry->cvlan,
 101                         .user = entry->user,
 102                         .mac_port = {
 103                                 .mac = entry->eth_addr,
 104                                 .out_idx = targ->mapping[port_idx],
 105                         },
 106                         .tsc = UINT64_MAX,
 107                 };
 108
 109                 int key_found;
 110                 void* entry_in_hash;
 111                 prox_rte_table_key8_add(table, &key, &data, &key_found, &entry_in_hash);
 112         }
 113 }
 114
 115 static void init_task_qinq_encap4(struct task_base *tbase, struct task_args *targ)
 116 {
 117         struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)(tbase);
 118         int socket_id = rte_lcore_to_socket_id(targ->lconf->id);
 119
 120         task->qinq_tag = targ->qinq_tag;
 121         task->cpe_table = targ->cpe_table;
 122         task->cpe_timeout = msec_to_tsc(targ->cpe_table_timeout_ms);
 123
 124         if (!strcmp(targ->task_init->sub_mode_str, "pe")) {
 125                 PROX_PANIC(!strcmp(targ->cpe_table_name, ""), "CPE table not configured\n");
 126                 fill_table(targ, task->cpe_table);
 127         }
 128
 129 #ifdef ENABLE_EXTRA_USER_STATISTICS
 130         task->n_users = targ->n_users;
 131         task->stats_per_user = prox_zmalloc(targ->n_users * sizeof(uint32_t), socket_id);
 132 #endif
 133         if (targ->runtime_flags & TASK_CLASSIFY) {
 134                 PROX_PANIC(!strcmp(targ->dscp, ""), "DSCP table not specified\n");
 135                 task->dscp = prox_sh_find_socket(socket_id, targ->dscp);
 136                 if (!task->dscp) {
 137                         int ret = lua_to_dscp(prox_lua(), GLOBAL, targ->dscp, socket_id, &task->dscp);
 138                         PROX_PANIC(ret, "Failed to create dscp table from config:\n%s\n",
 139                                    get_lua_to_errors());
 140                         prox_sh_add_socket(socket_id, targ->dscp, task->dscp);
 141                 }
 142         }
 143
 144         task->runtime_flags = targ->runtime_flags;
 145
 146         for (uint32_t i = 0; i < 64; ++i) {
 147                 task->fake_packets[i] = (struct rte_mbuf*)((uint8_t*)&task->keys[i] - sizeof (struct rte_mbuf));
 148         }
 149
 150         targ->lconf->ctrl_timeout = freq_to_tsc(targ->ctrl_freq);
 151         targ->lconf->ctrl_func_m[targ->task] = arp_msg;
 152
 153         struct prox_port_cfg *port = find_reachable_port(targ);
 154         if (port) {
 155                 task->offload_crc = port->requested_tx_offload & (DEV_TX_OFFLOAD_IPV4_CKSUM | DEV_TX_OFFLOAD_UDP_CKSUM);
 156         }
 157
 158         /* TODO: check if it is not necessary to limit reverse mapping
 159            for the elements that have been changing in mapping? */
 160
 161         for (uint32_t i =0 ; i < sizeof(targ->mapping)/sizeof(targ->mapping[0]); ++i) {
 162                 task->src_mac[targ->mapping[i]] = *(uint64_t*)&prox_port_cfg[i].eth_addr;
 163         }
 164
 165         /* task->src_mac[entry->port_idx] = *(uint64_t*)&prox_port_cfg[entry->port_idx].eth_addr; */
 166 }
 167
 168 static void arp_msg(struct task_base *tbase, void **data, uint16_t n_msgs)
 169 {
 170         struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)tbase;
 171         struct arp_msg **msgs = (struct arp_msg **)data;
 172
 173         arp_update_from_msg(task->cpe_table, msgs, n_msgs, task->cpe_timeout);
 174 }
 175
 176 static inline void add_key(struct task_args *targ, struct qinq_gre_map *qinq_gre_map, struct rte_table_hash* qinq_gre_table, uint32_t i, uint32_t *count)
 177 {
 178         struct qinq_gre_data entry = {
 179                 .gre_id = qinq_gre_map->entries[i].gre_id,
 180                 .user = qinq_gre_map->entries[i].user,
 181         };
 182
 183 #ifdef USE_QINQ
 184         struct vlans qinq2 = {
 185                 .svlan = {.eth_proto = targ->qinq_tag, .vlan_tci = qinq_gre_map->entries[i].svlan},
 186                 .cvlan = {.eth_proto = ETYPE_VLAN,     .vlan_tci = qinq_gre_map->entries[i].cvlan}
 187         };
 188
 189         int key_found = 0;
 190         void* entry_in_hash = NULL;
 191         prox_rte_table_key8_add(qinq_gre_table, &qinq2, &entry, &key_found, &entry_in_hash);
 192
 193         plog_dbg("Core %u adding user %u (tag %x svlan %x cvlan %x), rss=%x\n",
 194                  targ->lconf->id, qinq_gre_map->entries[i].user, qinq2.svlan.eth_proto,
 195                  rte_bswap16(qinq_gre_map->entries[i].svlan),
 196                  rte_bswap16(qinq_gre_map->entries[i].cvlan),
 197                  qinq_gre_map->entries[i].rss);
 198 #else
 199         /* lower 3 bytes of IPv4 address contain svlan/cvlan. */
 200         uint64_t ip = ((uint32_t)rte_bswap16(qinq_gre_map->entries[i].svlan) << 12) |
 201                 rte_bswap16(qinq_gre_map->entries[i].cvlan);
 202         int key_found = 0;
 203         void* entry_in_hash = NULL;
 204         prox_rte_table_key8_add(qinq_gre_table, &ip, &entry, &key_found, &entry_in_hash);
 205
 206         plog_dbg("Core %u hash table add: key = %016"PRIx64"\n",
 207                  targ->lconf->id, ip);
 208 #endif
 209         (*count)++;
 210 }
 211
 212 void init_qinq_gre_table(struct task_args *targ, struct qinq_gre_map *qinq_gre_map)
 213 {
 214         struct rte_table_hash* qinq_gre_table;
 215         uint8_t table_part = targ->nb_slave_threads;
 216         if (!rte_is_power_of_2(table_part)) {
 217                 table_part = rte_align32pow2(table_part) >> 1;
 218         }
 219
 220         if (table_part == 0)
 221                 table_part = 1;
 222
 223         uint32_t n_entries = MAX_GRE / table_part;
 224         static char hash_name[30];
 225         sprintf(hash_name, "qinq_gre_hash_table_%03d", targ->lconf->id);
 226
 227         struct prox_rte_table_params table_hash_params = {
 228                 .name = hash_name,
 229                 .key_size = 8,
 230                 .n_keys = n_entries,
 231                 .n_buckets = n_entries,
 232                 .f_hash = (rte_table_hash_op_hash)hash_crc32,
 233                 .seed = 0,
 234                 .key_offset = HASH_METADATA_OFFSET(0),
 235                 .key_mask = NULL
 236         };
 237
 238         qinq_gre_table = prox_rte_table_create(&table_hash_params, rte_lcore_to_socket_id(targ->lconf->id), sizeof(struct qinq_gre_data));
 239
 240         // LB configuration known from Network Load Balancer
 241         // Find LB network Load balancer, i.e. ENCAP friend.
 242         for (uint8_t task_id = 0; task_id < targ->lconf->n_tasks_all; ++task_id) {
 243                 enum task_mode smode = targ->lconf->targs[task_id].mode;
 244                 if (QINQ_ENCAP4 == smode) {
 245                         targ->lb_friend_core =  targ->lconf->targs[task_id].lb_friend_core;
 246                         targ->lb_friend_task =  targ->lconf->targs[task_id].lb_friend_task;
 247                 }
 248         }
 249         // Packet coming from Load balancer. LB could balance on gre_id LSB, qinq hash or qinq RSS
 250         uint32_t flag_features = 0;
 251         if (targ->lb_friend_core != 0xFF) {
 252                 struct task_args *lb_targ = &lcore_cfg[targ->lb_friend_core].targs[targ->lb_friend_task];
 253                 flag_features = lb_targ->task_init->flag_features;
 254                 plog_info("\t\tWT %d Updated features to %x from friend %d\n", targ->lconf->id, flag_features, targ->lb_friend_core);
 255         } else {
 256                 plog_info("\t\tWT %d has no friend\n", targ->lconf->id);
 257         }
 258         if (targ->nb_slave_threads == 0)  {
 259                 // No slave threads, i.e. using RSS
 260                 plog_info("feature was %x is now %x\n", flag_features, TASK_FEATURE_LUT_QINQ_RSS);
 261                 flag_features = TASK_FEATURE_LUT_QINQ_RSS;
 262         }
 263         if ((flag_features & (TASK_FEATURE_GRE_ID|TASK_FEATURE_LUT_QINQ_RSS|TASK_FEATURE_LUT_QINQ_HASH)) == 0) {
 264                 plog_info("\t\tCould not find flag feature from Load balancer => supposing TASK_FEATURE_GRE_ID\n");
 265                 flag_features = TASK_FEATURE_GRE_ID;
 266         }
 267
 268         /* Only store QinQ <-> GRE mapping for packets that are handled by this worker thread */
 269         uint32_t count = 0;
 270         if (flag_features & TASK_FEATURE_LUT_QINQ_RSS) {
 271                 // If there is a load balancer, number of worker thread is indicated by targ->nb_slave_threads and n_rxq = 0
 272                 // If there is no load balancers, number of worker thread is indicated by n_rxq and nb_slave_threads = 0
 273                 uint8_t nb_worker_threads, worker_thread_id;
 274                 if (targ->nb_slave_threads) {
 275                         nb_worker_threads = targ->nb_slave_threads;
 276                         worker_thread_id = targ->worker_thread_id;
 277                 } else if (prox_port_cfg[targ->rx_port_queue[0].port].n_rxq) {
 278                         nb_worker_threads = prox_port_cfg[targ->rx_port_queue[0].port].n_rxq;
 279                         worker_thread_id = targ->rx_port_queue[0].queue;
 280                 } else {
 281                         PROX_PANIC(1, "Unexpected: unknown number of worker thread\n");
 282                 }
 283                 plog_info("\t\tUsing %d worker_threads id %d\n", nb_worker_threads, worker_thread_id);
 284                 for (uint32_t i = 0; i < qinq_gre_map->count; ++i) {
 285                         if (targ->nb_slave_threads == 0 || rss_to_queue(qinq_gre_map->entries[i].rss, nb_worker_threads) == worker_thread_id) {
 286                                 add_key(targ, qinq_gre_map, qinq_gre_table, i, &count);
 287                                 //plog_info("Queue %d adding key %16lx, svlan %x cvlan %x, rss=%x\n", targ->rx_queue, *(uint64_t *)q, qinq_to_gre_lookup[i].svlan,  qinq_to_gre_lookup[i].cvlan, qinq_to_gre_lookup[i].rss);
 288                         }
 289                 }
 290                 plog_info("\t\tAdded %d entries to worker thread %d\n", count,  worker_thread_id);
 291         } else if (flag_features & TASK_FEATURE_LUT_QINQ_HASH) {
 292                 for (uint32_t i = 0; i < qinq_gre_map->count; ++i) {
 293                         uint64_t cvlan = rte_bswap16(qinq_gre_map->entries[i].cvlan & 0xFF0F);
 294                         uint64_t svlan = rte_bswap16((qinq_gre_map->entries[i].svlan & 0xFF0F));
 295                         uint64_t qinq = rte_bswap64((svlan << 32) | cvlan);
 296                         uint8_t queue = rte_hash_crc(&qinq, 8, 0) % targ->nb_slave_threads;
 297                         if (queue == targ->worker_thread_id) {
 298                                 add_key(targ, qinq_gre_map, qinq_gre_table, i, &count);
 299                         }
 300                 }
 301                 plog_info("\t\tAdded %d entries to WT %d\n", count,  targ->worker_thread_id);
 302         } else if (flag_features & TASK_FEATURE_GRE_ID) {
 303                 for (uint32_t i = 0; i < qinq_gre_map->count; ++i) {
 304                         if (qinq_gre_map->entries[i].gre_id % targ->nb_slave_threads == targ->worker_thread_id) {
 305                                 add_key(targ, qinq_gre_map, qinq_gre_table, i, &count);
 306                         }
 307                 }
 308         }
 309
 310         for (uint8_t task_id = 0; task_id < targ->lconf->n_tasks_all; ++task_id) {
 311                 enum task_mode smode = targ->lconf->targs[task_id].mode;
 312                 if (QINQ_DECAP4 == smode) {
 313                         targ->lconf->targs[task_id].qinq_gre_table = qinq_gre_table;
 314                 }
 315
 316         }
 317 }
 318
 319 void init_cpe4_table(struct task_args *targ)
 320 {
 321         char name[64];
 322         sprintf(name, "core_%u_CPEv4Table", targ->lconf->id);
 323
 324         uint8_t table_part = targ->nb_slave_threads;
 325         if (!rte_is_power_of_2(table_part)) {
 326                 table_part = rte_align32pow2(table_part) >> 1;
 327         }
 328
 329         if (table_part == 0)
 330                 table_part = 1;
 331
 332         uint32_t n_entries = MAX_GRE / table_part;
 333
 334         static char hash_name[30];
 335         sprintf(hash_name, "cpe4_table_%03d", targ->lconf->id);
 336
 337         struct prox_rte_table_params table_hash_params = {
 338                 .name = hash_name,
 339                 .key_size = 8,
 340                 .n_keys = n_entries,
 341                 .n_buckets = n_entries >> 1,
 342                 .f_hash = (rte_table_hash_op_hash)hash_crc32,
 343                 .seed = 0,
 344                 .key_offset = HASH_METADATA_OFFSET(0),
 345                 .key_mask = NULL
 346         };
 347         size_t entry_size = sizeof(struct cpe_data);
 348         if (!rte_is_power_of_2(entry_size)) {
 349                 entry_size = rte_align32pow2(entry_size);
 350         }
 351
 352         struct rte_table_hash* phash = prox_rte_table_create(&table_hash_params, rte_lcore_to_socket_id(targ->lconf->id), entry_size);
 353         PROX_PANIC(NULL == phash, "Unable to allocate memory for IPv4 hash table on core %u\n", targ->lconf->id);
 354
 355         /* for locality, copy the pointer to the port structure where it is needed at packet handling time */
 356         for (uint8_t task_id = 0; task_id < targ->lconf->n_tasks_all; ++task_id) {
 357                 enum task_mode smode = targ->lconf->targs[task_id].mode;
 358                 if (QINQ_ENCAP4 == smode || QINQ_DECAP4 == smode) {
 359                         targ->lconf->targs[task_id].cpe_table = phash;
 360                 }
 361         }
 362 }
 363
 364 static void early_init_table(struct task_args* targ)
 365 {
 366         if (!targ->cpe_table) {
 367                 init_cpe4_table(targ);
 368         }
 369 }
 370
 371 static inline void restore_cpe(struct cpe_pkt *packet, struct cpe_data *table, __attribute__((unused)) uint16_t qinq_tag, uint64_t *src_mac)
 372 {
 373 #ifdef USE_QINQ
 374         struct qinq_hdr *pqinq = &packet->qinq_hdr;
 375         rte_memcpy(pqinq, &qinq_proto, sizeof(struct qinq_hdr));
 376         (*(uint64_t *)(&pqinq->d_addr)) = table->mac_port_8bytes;
 377         /* set source as well now */
 378         *((uint64_t *)(&pqinq->s_addr)) = *((uint64_t *)&src_mac[table->mac_port.out_idx]);
 379         pqinq->svlan.vlan_tci = table->qinq_svlan;
 380         pqinq->cvlan.vlan_tci = table->qinq_cvlan;
 381         pqinq->svlan.eth_proto = qinq_tag;
 382         pqinq->cvlan.eth_proto = ETYPE_VLAN;
 383         pqinq->ether_type = ETYPE_IPv4;
 384 #else
 385         (*(uint64_t *)(&packet->ether_hdr.d_addr)) = table->mac_port_8bytes;
 386         /* set source as well now */
 387         *((uint64_t *)(&packet->ether_hdr.s_addr)) = *((uint64_t *)&src_mac[table->mac_port.out_idx]);
 388         packet->ether_hdr.ether_type = ETYPE_IPv4;
 389
 390         packet->ipv4_hdr.dst_addr = rte_bswap32(10 << 24 | rte_bswap16(table->qinq_svlan) << 12 | rte_bswap16(table->qinq_cvlan));
 391 #endif
 392 }
 393
 394 static inline uint8_t handle_qinq_encap4(struct task_qinq_encap4 *task, struct cpe_pkt *cpe_pkt, struct rte_mbuf *mbuf, struct cpe_data *entry);
 395
 396 /* Same functionality as handle_qinq_encap_v4_bulk but untag MPLS as well. */
 397 static int handle_qinq_encap4_untag_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
 398 {
 399         struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)tbase;
 400         uint8_t out[MAX_PKT_BURST];
 401         prefetch_pkts(mbufs, n_pkts);
 402
 403         for (uint16_t j = 0; j < n_pkts; ++j) {
 404                 if (likely(mpls_untag(mbufs[j]))) {
 405                         struct cpe_pkt* cpe_pkt = (struct cpe_pkt*) rte_pktmbuf_adj(mbufs[j], UPSTREAM_DELTA);
 406                         out[j] = handle_qinq_encap4(task, cpe_pkt, mbufs[j], NULL);
 407                 }
 408                 else {
 409                         out[j] = OUT_DISCARD;
 410                 }
 411         }
 412
 413         return task->base.tx_pkt(&task->base, mbufs, n_pkts, out);
 414 }
 415
 416 static inline void extract_key_bulk(struct task_qinq_encap4 *task, struct rte_mbuf **mbufs, uint16_t n_pkts)
 417 {
 418         for (uint16_t j = 0; j < n_pkts; ++j) {
 419                 extract_key_core(mbufs[j], &task->keys[j]);
 420         }
 421 }
 422
 423 __attribute__((cold)) static void handle_error(struct rte_mbuf *mbuf)
 424 {
 425         struct core_net_pkt* core_pkt = rte_pktmbuf_mtod(mbuf, struct core_net_pkt *);
 426         uint32_t dst_ip = core_pkt->ip_hdr.dst_addr;
 427         uint32_t le_gre_id = rte_be_to_cpu_32(core_pkt->gre_hdr.gre_id);
 428
 429         plogx_dbg("Unknown IP %x/gre_id %x\n", dst_ip, le_gre_id);
 430 }
 431
 432 static int handle_qinq_encap4_bulk_pe(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
 433 {
 434         struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)tbase;
 435         uint64_t pkts_mask = RTE_LEN2MASK(n_pkts, uint64_t);
 436         struct cpe_data* entries[64];
 437         uint8_t out[MAX_PKT_BURST];
 438         uint64_t lookup_hit_mask;
 439
 440         prefetch_pkts(mbufs, n_pkts);
 441
 442         for (uint16_t j = 0; j < n_pkts; ++j) {
 443                 struct ipv4_hdr* ip = (struct ipv4_hdr *)(rte_pktmbuf_mtod(mbufs[j], struct ether_hdr *) + 1);
 444                 task->keys[j] = (uint64_t)ip->dst_addr;
 445         }
 446         prox_rte_table_key8_lookup(task->cpe_table, task->fake_packets, pkts_mask, &lookup_hit_mask, (void**)entries);
 447
 448         if (likely(lookup_hit_mask == pkts_mask)) {
 449                 for (uint16_t j = 0; j < n_pkts; ++j) {
 450                         struct cpe_pkt* cpe_pkt = (struct cpe_pkt*) rte_pktmbuf_prepend(mbufs[j], sizeof(struct qinq_hdr) - sizeof(struct ether_hdr));
 451                         uint16_t padlen = mbuf_calc_padlen(mbufs[j], cpe_pkt, &cpe_pkt->ipv4_hdr);
 452
 453                         if (padlen) {
 454                                 rte_pktmbuf_trim(mbufs[j], padlen);
 455                         }
 456                         out[j] = handle_qinq_encap4(task, cpe_pkt, mbufs[j], entries[j]);
 457                 }
 458         }
 459         else {
 460                 for (uint16_t j = 0; j < n_pkts; ++j) {
 461                         if (unlikely(!((lookup_hit_mask >> j) & 0x1))) {
 462                                 handle_error(mbufs[j]);
 463                                 out[j] = OUT_DISCARD;
 464                                 continue;
 465                         }
 466                         struct cpe_pkt* cpe_pkt = (struct cpe_pkt*) rte_pktmbuf_prepend(mbufs[j], sizeof(struct qinq_hdr) - sizeof(struct ether_hdr));
 467                         uint16_t padlen = mbuf_calc_padlen(mbufs[j], cpe_pkt, &cpe_pkt->ipv4_hdr);
 468
 469                         if (padlen) {
 470                                 rte_pktmbuf_trim(mbufs[j], padlen);
 471                         }
 472                         out[j] = handle_qinq_encap4(task, cpe_pkt, mbufs[j], entries[j]);
 473                 }
 474         }
 475
 476         return task->base.tx_pkt(&task->base, mbufs, n_pkts, out);
 477 }
 478 int handle_qinq_encap4_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
 479 {
 480         struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)tbase;
 481         uint64_t pkts_mask = RTE_LEN2MASK(n_pkts, uint64_t);
 482         struct cpe_data* entries[64];
 483         uint8_t out[MAX_PKT_BURST];
 484         uint64_t lookup_hit_mask;
 485
 486         prefetch_pkts(mbufs, n_pkts);
 487
 488         // From GRE ID and IP address, retrieve QinQ and MAC addresses
 489         extract_key_bulk(task, mbufs, n_pkts);
 490         prox_rte_table_key8_lookup(task->cpe_table, task->fake_packets, pkts_mask, &lookup_hit_mask, (void**)entries);
 491
 492         if (likely(lookup_hit_mask == pkts_mask)) {
 493                 for (uint16_t j = 0; j < n_pkts; ++j) {
 494                         struct cpe_pkt* cpe_pkt = (struct cpe_pkt*) rte_pktmbuf_adj(mbufs[j], UPSTREAM_DELTA);
 495                         // We are receiving GRE tunnelled packets (and removing UPSTRAM_DELTA bytes), whose length is > 64 bytes
 496                         // So there should be no padding, but in case the is one, remove it
 497                         uint16_t padlen = mbuf_calc_padlen(mbufs[j], cpe_pkt, &cpe_pkt->ipv4_hdr);
 498
 499                         if (padlen) {
 500                                 rte_pktmbuf_trim(mbufs[j], padlen);
 501                         }
 502                         out[j] = handle_qinq_encap4(task, cpe_pkt, mbufs[j], entries[j]);
 503                 }
 504         }
 505         else {
 506                 for (uint16_t j = 0; j < n_pkts; ++j) {
 507                         if (unlikely(!((lookup_hit_mask >> j) & 0x1))) {
 508                                 handle_error(mbufs[j]);
 509                                 out[j] = OUT_DISCARD;
 510                                 continue;
 511                         }
 512                         struct cpe_pkt* cpe_pkt = (struct cpe_pkt*) rte_pktmbuf_adj(mbufs[j], UPSTREAM_DELTA);
 513                         uint16_t padlen = mbuf_calc_padlen(mbufs[j], cpe_pkt, &cpe_pkt->ipv4_hdr);
 514
 515                         if (padlen) {
 516                                 rte_pktmbuf_trim(mbufs[j], padlen);
 517                         }
 518                         out[j] = handle_qinq_encap4(task, cpe_pkt, mbufs[j], entries[j]);
 519                 }
 520         }
 521
 522         return task->base.tx_pkt(&task->base, mbufs, n_pkts, out);
 523 }
 524
 525 static inline uint8_t handle_qinq_encap4(struct task_qinq_encap4 *task, struct cpe_pkt *cpe_pkt, struct rte_mbuf *mbuf, struct cpe_data *entry)
 526 {
 527         PROX_ASSERT(cpe_pkt);
 528
 529         if (cpe_pkt->ipv4_hdr.time_to_live) {
 530                 cpe_pkt->ipv4_hdr.time_to_live--;
 531         }
 532         else {
 533                 plog_info("TTL = 0 => Dropping\n");
 534                 return OUT_DISCARD;
 535         }
 536         cpe_pkt->ipv4_hdr.hdr_checksum = 0;
 537
 538         restore_cpe(cpe_pkt, entry, task->qinq_tag, task->src_mac);
 539
 540         if (task->runtime_flags & TASK_CLASSIFY) {
 541                 uint8_t queue = task->dscp[cpe_pkt->ipv4_hdr.type_of_service >> 2] & 0x3;
 542                 uint8_t tc = task->dscp[cpe_pkt->ipv4_hdr.type_of_service >> 2] >> 2;
 543
 544                 rte_sched_port_pkt_write(mbuf, 0, entry->user, tc, queue, 0);
 545         }
 546 #ifdef ENABLE_EXTRA_USER_STATISTICS
 547         task->stats_per_user[entry->user]++;
 548 #endif
 549         if (task->runtime_flags & TASK_TX_CRC) {
 550                 prox_ip_cksum(mbuf, &cpe_pkt->ipv4_hdr, sizeof(struct qinq_hdr), sizeof(struct ipv4_hdr), task->offload_crc);
 551         }
 552         return entry->mac_port.out_idx;
 553 }
 554
 555 static void flow_iter_next(struct flow_iter *iter, struct task_args *targ)
 556 {
 557         do {
 558                 iter->idx++;
 559                 uint8_t flag_features = iter->data;
 560
 561                 if (flag_features & TASK_FEATURE_LUT_QINQ_RSS) {
 562                         // If there is a load balancer, number of worker thread is indicated by targ->nb_slave_threads and n_rxq = 0
 563                         // If there is no load balancers, number of worker thread is indicated by n_rxq and nb_slave_threads = 0
 564                         uint8_t nb_worker_threads, worker_thread_id;
 565                         nb_worker_threads = 1;
 566                         worker_thread_id = 1;
 567                         if (targ->nb_slave_threads) {
 568                                 nb_worker_threads = targ->nb_slave_threads;
 569                                 worker_thread_id = targ->worker_thread_id;
 570                         } else if (prox_port_cfg[targ->rx_port_queue[0].port].n_rxq) {
 571                                 nb_worker_threads = prox_port_cfg[targ->rx_port_queue[0].port].n_rxq;
 572                                 worker_thread_id = targ->rx_port_queue[0].queue;
 573                         } else {
 574                                 plog_err("Unexpected: unknown number of worker thread\n");
 575                         }
 576
 577                         if (targ->nb_slave_threads == 0 || rss_to_queue(get_qinq_gre_map(targ)->entries[iter->idx].rss, nb_worker_threads) == worker_thread_id)
 578                                 break;
 579                 } else if (flag_features & TASK_FEATURE_LUT_QINQ_HASH) {
 580                         uint64_t cvlan = rte_bswap16(get_qinq_gre_map(targ)->entries[iter->idx].cvlan & 0xFF0F);
 581                         uint64_t svlan = rte_bswap16(get_qinq_gre_map(targ)->entries[iter->idx].svlan & 0xFF0F);
 582                         uint64_t qinq = rte_bswap64((svlan << 32) | cvlan);
 583                         uint8_t queue = rte_hash_crc(&qinq, 8, 0) % targ->nb_slave_threads;
 584                         if (queue == targ->worker_thread_id)
 585                                 break;
 586                 } else if (flag_features & TASK_FEATURE_GRE_ID) {
 587                         if (get_qinq_gre_map(targ)->entries[iter->idx].gre_id % targ->nb_slave_threads == targ->worker_thread_id)
 588                                 break;
 589                 }
 590         } while (iter->idx != (int)get_qinq_gre_map(targ)->count);
 591 }
 592
 593 static void flow_iter_beg(struct flow_iter *iter, struct task_args *targ)
 594 {
 595         uint32_t flag_features = 0;
 596         if (targ->lb_friend_core != 0xFF) {
 597                 struct task_args *lb_targ = &lcore_cfg[targ->lb_friend_core].targs[targ->lb_friend_task];
 598                 flag_features = lb_targ->task_init->flag_features;
 599                 plog_info("\t\tWT %d Updated features to %x from friend %d\n", targ->lconf->id, flag_features, targ->lb_friend_core);
 600         } else {
 601                 plog_info("\t\tWT %d has no friend\n", targ->lconf->id);
 602         }
 603         if (targ->nb_slave_threads == 0)  {
 604                 // No slave threads, i.e. using RSS
 605                 plog_info("feature was %x is now %x\n", flag_features, TASK_FEATURE_LUT_QINQ_RSS);
 606                 flag_features = TASK_FEATURE_LUT_QINQ_RSS;
 607         }
 608         if ((flag_features & (TASK_FEATURE_GRE_ID|TASK_FEATURE_LUT_QINQ_RSS|TASK_FEATURE_LUT_QINQ_HASH)) == 0) {
 609                 plog_info("\t\tCould not find flag feature from Load balancer => supposing TASK_FEATURE_GRE_ID\n");
 610                 flag_features = TASK_FEATURE_GRE_ID;
 611         }
 612
 613         iter->idx = -1;
 614         flow_iter_next(iter, targ);
 615 }
 616
 617 static int flow_iter_is_end(struct flow_iter *iter, struct task_args *targ)
 618 {
 619         return iter->idx == (int)get_qinq_gre_map(targ)->count;
 620 }
 621
 622 static uint32_t flow_iter_get_gre_id(struct flow_iter *iter, struct task_args *targ)
 623 {
 624         return get_qinq_gre_map(targ)->entries[iter->idx].gre_id;
 625 }
 626
 627 static struct task_init task_init_qinq_encap4_table = {
 628         .mode = QINQ_ENCAP4,
 629         .mode_str = "qinqencapv4",
 630         .early_init = early_init_table,
 631         .init = init_task_qinq_encap4,
 632         .handle = handle_qinq_encap4_bulk,
 633         /* In this case user in qinq_lookup table is the QoS user
 634            (from user_table), i.e. usually from 0 to 32K Otherwise it
 635            would have been a user from (0 to n_interface x 32K) */
 636         .flow_iter = {
 637                 .beg        = flow_iter_beg,
 638                 .is_end     = flow_iter_is_end,
 639                 .next       = flow_iter_next,
 640                 .get_gre_id = flow_iter_get_gre_id,
 641         },
 642         .flag_features = TASK_FEATURE_CLASSIFY,
 643         .size = sizeof(struct task_qinq_encap4)
 644 };
 645
 646 static struct task_init task_init_qinq_encap4_table_pe = {
 647         .mode = QINQ_ENCAP4,
 648         .mode_str = "qinqencapv4",
 649         .sub_mode_str = "pe",
 650         .early_init = early_init_table,
 651         .init = init_task_qinq_encap4,
 652         .handle = handle_qinq_encap4_bulk_pe,
 653         .flag_features = TASK_FEATURE_CLASSIFY,
 654         .size = sizeof(struct task_qinq_encap4)
 655 };
 656
 657 static struct task_init task_init_qinq_encap4_untag = {
 658         .mode = QINQ_ENCAP4,
 659         .sub_mode_str = "unmpls",
 660         .mode_str = "qinqencapv4",
 661         .init = init_task_qinq_encap4,
 662         .handle = handle_qinq_encap4_untag_bulk,
 663         .flag_features = TASK_FEATURE_CLASSIFY,
 664         .size = sizeof(struct task_qinq_encap4)
 665 };
 666
 667 __attribute__((constructor)) static void reg_task_qinq_encap4(void)
 668 {
 669         reg_task(&task_init_qinq_encap4_table);
 670         reg_task(&task_init_qinq_encap4_table_pe);
 671         reg_task(&task_init_qinq_encap4_untag);
 672 }