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_cycles.h>
  19
  20 #include "mbuf_utils.h"
  21 #include "prox_malloc.h"
  22 #include "prox_lua.h"
  23 #include "prox_lua_types.h"
  24 #include "handle_qinq_encap4.h"
  25 #include "handle_qinq_decap4.h"
  26 #include "prox_args.h"
  27 #include "defines.h"
  28 #include "tx_pkt.h"
  29 #include "prefetch.h"
  30 #include "pkt_prototypes.h"
  31 #include "hash_entry_types.h"
  32 #include "task_init.h"
  33 #include "bng_pkts.h"
  34 #include "prox_cksum.h"
  35 #include "hash_utils.h"
  36 #include "quit.h"
  37 #include "prox_port_cfg.h"
  38 #include "handle_lb_net.h"
  39 #include "prox_cfg.h"
  40 #include "cfgfile.h"
  41 #include "toeplitz.h"
  42 #include "prox_shared.h"
  43
  44 static struct cpe_table_data *read_cpe_table_config(const char *name, uint8_t socket)
  45 {
  46         struct lua_State *L = prox_lua();
  47         struct cpe_table_data *ret = NULL;
  48
  49         lua_getglobal(L, name);
  50         PROX_PANIC(lua_isnil(L, -1), "Coudn't find cpe_table data\n");
  51
  52         return ret;
  53 }
  54
  55 struct qinq_gre_map *get_qinq_gre_map(struct task_args *targ)
  56 {
  57         const int socket_id = rte_lcore_to_socket_id(targ->lconf->id);
  58         struct qinq_gre_map *ret = prox_sh_find_socket(socket_id, "qinq_gre_map");
  59
  60         if (!ret) {
  61                 PROX_PANIC(!strcmp(targ->user_table, ""), "No user table defined\n");
  62                 int rv = lua_to_qinq_gre_map(prox_lua(), GLOBAL, targ->user_table, socket_id, &ret);
  63                 PROX_PANIC(rv, "Error reading mapping between qinq and gre from qinq_gre_map: \n%s\n",
  64                            get_lua_to_errors());
  65                 prox_sh_add_socket(socket_id, "qinq_gre_map", ret);
  66         }
  67         return ret;
  68 }
  69
  70 /* Encapsulate IPv4 packets in QinQ. QinQ tags are derived from gre_id. */
  71 int handle_qinq_encap4_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts);
  72 static void arp_msg(struct task_base *tbase, void **data, uint16_t n_msgs);
  73
  74 static void fill_table(struct task_args *targ, struct rte_table_hash *table)
  75 {
  76         struct cpe_table_data *cpe_table_data;
  77         const int socket_id = rte_lcore_to_socket_id(targ->lconf->id);
  78         int ret = lua_to_cpe_table_data(prox_lua(), GLOBAL, targ->cpe_table_name, socket_id, &cpe_table_data);
  79         const uint8_t n_slaves = targ->nb_slave_threads;
  80         const uint8_t worker_id = targ->worker_thread_id;
  81
  82         for (uint32_t i = 0; i < cpe_table_data->n_entries; ++i) {
  83                 if (rte_bswap32(cpe_table_data->entries[i].ip) % n_slaves != worker_id) {
  84                         continue;
  85                 }
  86                 struct cpe_table_entry *entry = &cpe_table_data->entries[i];
  87
  88                 uint32_t port_idx = prox_cfg.cpe_table_ports[entry->port_idx];
  89                 PROX_PANIC(targ->mapping[port_idx] == 255, "Error reading cpe table: Mapping for port %d is missing", port_idx);
  90
  91                 struct cpe_key key = {
  92                         .ip = entry->ip,
  93                         .gre_id = entry->gre_id,
  94                 };
  95
  96                 struct cpe_data data = {
  97                         .qinq_svlan = entry->svlan,
  98                         .qinq_cvlan = entry->cvlan,
  99                         .user = entry->user,
 100                         .mac_port = {
 101                                 .mac = entry->eth_addr,
 102                                 .out_idx = targ->mapping[port_idx],
 103                         },
 104                         .tsc = UINT64_MAX,
 105                 };
 106
 107                 int key_found;
 108                 void* entry_in_hash;
 109                 rte_table_hash_key8_ext_dosig_ops.f_add(table, &key, &data, &key_found, &entry_in_hash);
 110         }
 111 }
 112
 113 static void init_task_qinq_encap4(struct task_base *tbase, struct task_args *targ)
 114 {
 115         struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)(tbase);
 116         int socket_id = rte_lcore_to_socket_id(targ->lconf->id);
 117
 118         task->qinq_tag = targ->qinq_tag;
 119         task->cpe_table = targ->cpe_table;
 120         task->cpe_timeout = msec_to_tsc(targ->cpe_table_timeout_ms);
 121
 122         if (!strcmp(targ->task_init->sub_mode_str, "pe")) {
 123                 PROX_PANIC(!strcmp(targ->cpe_table_name, ""), "CPE table not configured\n");
 124                 fill_table(targ, task->cpe_table);
 125         }
 126
 127 #ifdef ENABLE_EXTRA_USER_STATISTICS
 128         task->n_users = targ->n_users;
 129         task->stats_per_user = prox_zmalloc(targ->n_users * sizeof(uint32_t), socket_id);
 130 #endif
 131         if (targ->runtime_flags & TASK_CLASSIFY) {
 132                 PROX_PANIC(!strcmp(targ->dscp, ""), "DSCP table not specified\n");
 133                 task->dscp = prox_sh_find_socket(socket_id, targ->dscp);
 134                 if (!task->dscp) {
 135                         int ret = lua_to_dscp(prox_lua(), GLOBAL, targ->dscp, socket_id, &task->dscp);
 136                         PROX_PANIC(ret, "Failed to create dscp table from config:\n%s\n",
 137                                    get_lua_to_errors());
 138                         prox_sh_add_socket(socket_id, targ->dscp, task->dscp);
 139                 }
 140         }
 141
 142         task->runtime_flags = targ->runtime_flags;
 143
 144         for (uint32_t i = 0; i < 64; ++i) {
 145                 task->fake_packets[i] = (struct rte_mbuf*)((uint8_t*)&task->keys[i] - sizeof (struct rte_mbuf));
 146         }
 147
 148         targ->lconf->ctrl_timeout = freq_to_tsc(targ->ctrl_freq);
 149         targ->lconf->ctrl_func_m[targ->task] = arp_msg;
 150
 151         struct prox_port_cfg *port = find_reachable_port(targ);
 152         if (port) {
 153                 task->offload_crc = port->capabilities.tx_offload_cksum;
 154         }
 155
 156         /* TODO: check if it is not necessary to limit reverse mapping
 157            for the elements that have been changing in mapping? */
 158
 159         for (uint32_t i =0 ; i < sizeof(targ->mapping)/sizeof(targ->mapping[0]); ++i) {
 160                 task->src_mac[targ->mapping[i]] = *(uint64_t*)&prox_port_cfg[i].eth_addr;
 161         }
 162
 163         /* task->src_mac[entry->port_idx] = *(uint64_t*)&prox_port_cfg[entry->port_idx].eth_addr; */
 164 }
 165
 166 static void arp_msg(struct task_base *tbase, void **data, uint16_t n_msgs)
 167 {
 168         struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)tbase;
 169         struct arp_msg **msgs = (struct arp_msg **)data;
 170
 171         arp_update_from_msg(task->cpe_table, msgs, n_msgs, task->cpe_timeout);
 172 }
 173
 174 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)
 175 {
 176         struct qinq_gre_data entry = {
 177                 .gre_id = qinq_gre_map->entries[i].gre_id,
 178                 .user = qinq_gre_map->entries[i].user,
 179         };
 180
 181 #ifdef USE_QINQ
 182         struct vlans qinq2 = {
 183                 .svlan = {.eth_proto = targ->qinq_tag, .vlan_tci = qinq_gre_map->entries[i].svlan},
 184                 .cvlan = {.eth_proto = ETYPE_VLAN,     .vlan_tci = qinq_gre_map->entries[i].cvlan}
 185         };
 186
 187         int key_found = 0;
 188         void* entry_in_hash = NULL;
 189         rte_table_hash_key8_ext_dosig_ops.f_add(qinq_gre_table, &qinq2, &entry, &key_found, &entry_in_hash);
 190
 191         plog_dbg("Core %u adding user %u (tag %x svlan %x cvlan %x), rss=%x\n",
 192                  targ->lconf->id, qinq_gre_map->entries[i].user, qinq2.svlan.eth_proto,
 193                  rte_bswap16(qinq_gre_map->entries[i].svlan),
 194                  rte_bswap16(qinq_gre_map->entries[i].cvlan),
 195                  qinq_gre_map->entries[i].rss);
 196 #else
 197         /* lower 3 bytes of IPv4 address contain svlan/cvlan. */
 198         uint64_t ip = ((uint32_t)rte_bswap16(qinq_gre_map->entries[i].svlan) << 12) |
 199                 rte_bswap16(qinq_gre_map->entries[i].cvlan);
 200         int key_found = 0;
 201         void* entry_in_hash = NULL;
 202         rte_table_hash_key8_ext_dosig_ops.f_add(qinq_gre_table, &ip, &entry, &key_found, &entry_in_hash);
 203
 204         plog_dbg("Core %u hash table add: key = %016"PRIx64"\n",
 205                  targ->lconf->id, ip);
 206 #endif
 207         (*count)++;
 208 }
 209
 210 void init_qinq_gre_table(struct task_args *targ, struct qinq_gre_map *qinq_gre_map)
 211 {
 212         struct rte_table_hash* qinq_gre_table;
 213         uint8_t table_part = targ->nb_slave_threads;
 214         if (!rte_is_power_of_2(table_part)) {
 215                 table_part = rte_align32pow2(table_part) >> 1;
 216         }
 217
 218         if (table_part == 0)
 219                 table_part = 1;
 220
 221         uint32_t n_entries = MAX_GRE / table_part;
 222
 223         struct rte_table_hash_key8_ext_params table_hash_params = {
 224                 .n_entries = n_entries,
 225                 .n_entries_ext = n_entries >> 1,
 226                 .f_hash = hash_crc32,
 227                 .seed = 0,
 228                 .signature_offset = HASH_METADATA_OFFSET(8),
 229                 .key_offset = HASH_METADATA_OFFSET(0),
 230         };
 231
 232         qinq_gre_table = rte_table_hash_key8_ext_dosig_ops.
 233                 f_create(&table_hash_params, rte_lcore_to_socket_id(targ->lconf->id), sizeof(struct qinq_gre_data));
 234
 235         // LB configuration known from Network Load Balancer
 236         // Find LB network Load balancer, i.e. ENCAP friend.
 237         for (uint8_t task_id = 0; task_id < targ->lconf->n_tasks_all; ++task_id) {
 238                 enum task_mode smode = targ->lconf->targs[task_id].mode;
 239                 if (QINQ_ENCAP4 == smode) {
 240                         targ->lb_friend_core =  targ->lconf->targs[task_id].lb_friend_core;
 241                         targ->lb_friend_task =  targ->lconf->targs[task_id].lb_friend_task;
 242                 }
 243         }
 244         // Packet coming from Load balancer. LB could balance on gre_id LSB, qinq hash or qinq RSS
 245         uint32_t flag_features = 0;
 246         if (targ->lb_friend_core != 0xFF) {
 247                 struct task_args *lb_targ = &lcore_cfg[targ->lb_friend_core].targs[targ->lb_friend_task];
 248                 flag_features = lb_targ->task_init->flag_features;
 249                 plog_info("\t\tWT %d Updated features to %x from friend %d\n", targ->lconf->id, flag_features, targ->lb_friend_core);
 250         } else {
 251                 plog_info("\t\tWT %d has no friend\n", targ->lconf->id);
 252         }
 253         if (targ->nb_slave_threads == 0)  {
 254                 // No slave threads, i.e. using RSS
 255                 plog_info("feature was %x is now %x\n", flag_features, TASK_FEATURE_LUT_QINQ_RSS);
 256                 flag_features = TASK_FEATURE_LUT_QINQ_RSS;
 257         }
 258         if ((flag_features & (TASK_FEATURE_GRE_ID|TASK_FEATURE_LUT_QINQ_RSS|TASK_FEATURE_LUT_QINQ_HASH)) == 0) {
 259                 plog_info("\t\tCould not find flag feature from Load balancer => supposing TASK_FEATURE_GRE_ID\n");
 260                 flag_features = TASK_FEATURE_GRE_ID;
 261         }
 262
 263         /* Only store QinQ <-> GRE mapping for packets that are handled by this worker thread */
 264         uint32_t count = 0;
 265         if (flag_features & TASK_FEATURE_LUT_QINQ_RSS) {
 266                 // If there is a load balancer, number of worker thread is indicated by targ->nb_slave_threads and n_rxq = 0
 267                 // If there is no load balancers, number of worker thread is indicated by n_rxq and nb_slave_threads = 0
 268                 uint8_t nb_worker_threads, worker_thread_id;
 269                 if (targ->nb_slave_threads) {
 270                         nb_worker_threads = targ->nb_slave_threads;
 271                         worker_thread_id = targ->worker_thread_id;
 272                 } else if (prox_port_cfg[targ->rx_port_queue[0].port].n_rxq) {
 273                         nb_worker_threads = prox_port_cfg[targ->rx_port_queue[0].port].n_rxq;
 274                         worker_thread_id = targ->rx_port_queue[0].queue;
 275                 } else {
 276                         PROX_PANIC(1, "Unexpected: unknown number of worker thread\n");
 277                 }
 278                 plog_info("\t\tUsing %d worker_threads id %d\n", nb_worker_threads, worker_thread_id);
 279                 for (uint32_t i = 0; i < qinq_gre_map->count; ++i) {
 280                         if (targ->nb_slave_threads == 0 || rss_to_queue(qinq_gre_map->entries[i].rss, nb_worker_threads) == worker_thread_id) {
 281                                 add_key(targ, qinq_gre_map, qinq_gre_table, i, &count);
 282                                 //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);
 283                         }
 284                 }
 285                 plog_info("\t\tAdded %d entries to worker thread %d\n", count,  worker_thread_id);
 286         } else if (flag_features & TASK_FEATURE_LUT_QINQ_HASH) {
 287                 for (uint32_t i = 0; i < qinq_gre_map->count; ++i) {
 288                         uint64_t cvlan = rte_bswap16(qinq_gre_map->entries[i].cvlan & 0xFF0F);
 289                         uint64_t svlan = rte_bswap16((qinq_gre_map->entries[i].svlan & 0xFF0F));
 290                         uint64_t qinq = rte_bswap64((svlan << 32) | cvlan);
 291                         uint8_t queue = hash_crc32(&qinq, 8, 0) % targ->nb_slave_threads;
 292                         if (queue == targ->worker_thread_id) {
 293                                 add_key(targ, qinq_gre_map, qinq_gre_table, i, &count);
 294                         }
 295                 }
 296                 plog_info("\t\tAdded %d entries to WT %d\n", count,  targ->worker_thread_id);
 297         } else if (flag_features & TASK_FEATURE_GRE_ID) {
 298                 for (uint32_t i = 0; i < qinq_gre_map->count; ++i) {
 299                         if (qinq_gre_map->entries[i].gre_id % targ->nb_slave_threads == targ->worker_thread_id) {
 300                                 add_key(targ, qinq_gre_map, qinq_gre_table, i, &count);
 301                         }
 302                 }
 303         }
 304
 305         for (uint8_t task_id = 0; task_id < targ->lconf->n_tasks_all; ++task_id) {
 306                 enum task_mode smode = targ->lconf->targs[task_id].mode;
 307                 if (QINQ_DECAP4 == smode) {
 308                         targ->lconf->targs[task_id].qinq_gre_table = qinq_gre_table;
 309                 }
 310
 311         }
 312 }
 313
 314 void init_cpe4_table(struct task_args *targ)
 315 {
 316         char name[64];
 317         sprintf(name, "core_%u_CPEv4Table", targ->lconf->id);
 318
 319         uint8_t table_part = targ->nb_slave_threads;
 320         if (!rte_is_power_of_2(table_part)) {
 321                 table_part = rte_align32pow2(table_part) >> 1;
 322         }
 323
 324         if (table_part == 0)
 325                 table_part = 1;
 326
 327         uint32_t n_entries = MAX_GRE / table_part;
 328         struct rte_table_hash_key8_ext_params table_hash_params = {
 329                 .n_entries = n_entries,
 330                 .n_entries_ext = n_entries >> 1,
 331                 .f_hash = hash_crc32,
 332                 .seed = 0,
 333                 .signature_offset = HASH_METADATA_OFFSET(8),
 334                 .key_offset = HASH_METADATA_OFFSET(0),
 335         };
 336         size_t entry_size = sizeof(struct cpe_data);
 337         if (!rte_is_power_of_2(entry_size)) {
 338                 entry_size = rte_align32pow2(entry_size);
 339         }
 340
 341         struct rte_table_hash* phash = rte_table_hash_key8_ext_dosig_ops.
 342                 f_create(&table_hash_params, rte_lcore_to_socket_id(targ->lconf->id), entry_size);
 343         PROX_PANIC(NULL == phash, "Unable to allocate memory for IPv4 hash table on core %u\n", targ->lconf->id);
 344
 345         /* for locality, copy the pointer to the port structure where it is needed at packet handling time */
 346         for (uint8_t task_id = 0; task_id < targ->lconf->n_tasks_all; ++task_id) {
 347                 enum task_mode smode = targ->lconf->targs[task_id].mode;
 348                 if (QINQ_ENCAP4 == smode || QINQ_DECAP4 == smode) {
 349                         targ->lconf->targs[task_id].cpe_table = phash;
 350                 }
 351         }
 352 }
 353
 354 static void early_init_table(struct task_args* targ)
 355 {
 356         if (!targ->cpe_table) {
 357                 init_cpe4_table(targ);
 358         }
 359 }
 360
 361 static inline void restore_cpe(struct cpe_pkt *packet, struct cpe_data *table, __attribute__((unused)) uint16_t qinq_tag, uint64_t *src_mac)
 362 {
 363 #ifdef USE_QINQ
 364         struct qinq_hdr *pqinq = &packet->qinq_hdr;
 365         rte_memcpy(pqinq, &qinq_proto, sizeof(struct qinq_hdr));
 366         (*(uint64_t *)(&pqinq->d_addr)) = table->mac_port_8bytes;
 367         /* set source as well now */
 368         *((uint64_t *)(&pqinq->s_addr)) = *((uint64_t *)&src_mac[table->mac_port.out_idx]);
 369         pqinq->svlan.vlan_tci = table->qinq_svlan;
 370         pqinq->cvlan.vlan_tci = table->qinq_cvlan;
 371         pqinq->svlan.eth_proto = qinq_tag;
 372         pqinq->cvlan.eth_proto = ETYPE_VLAN;
 373         pqinq->ether_type = ETYPE_IPv4;
 374 #else
 375         (*(uint64_t *)(&packet->ether_hdr.d_addr)) = table->mac_port_8bytes;
 376         /* set source as well now */
 377         *((uint64_t *)(&packet->ether_hdr.s_addr)) = *((uint64_t *)&src_mac[table->mac_port.out_idx]);
 378         packet->ether_hdr.ether_type = ETYPE_IPv4;
 379
 380         packet->ipv4_hdr.dst_addr = rte_bswap32(10 << 24 | rte_bswap16(table->qinq_svlan) << 12 | rte_bswap16(table->qinq_cvlan));
 381 #endif
 382 }
 383
 384 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);
 385
 386 /* Same functionality as handle_qinq_encap_v4_bulk but untag MPLS as well. */
 387 static int handle_qinq_encap4_untag_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
 388 {
 389         struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)tbase;
 390         uint8_t out[MAX_PKT_BURST];
 391         prefetch_pkts(mbufs, n_pkts);
 392
 393         for (uint16_t j = 0; j < n_pkts; ++j) {
 394                 if (likely(mpls_untag(mbufs[j]))) {
 395                         struct cpe_pkt* cpe_pkt = (struct cpe_pkt*) rte_pktmbuf_adj(mbufs[j], UPSTREAM_DELTA);
 396                         out[j] = handle_qinq_encap4(task, cpe_pkt, mbufs[j], NULL);
 397                 }
 398                 else {
 399                         out[j] = OUT_DISCARD;
 400                 }
 401         }
 402
 403         return task->base.tx_pkt(&task->base, mbufs, n_pkts, out);
 404 }
 405
 406 static inline void extract_key_bulk(struct task_qinq_encap4 *task, struct rte_mbuf **mbufs, uint16_t n_pkts)
 407 {
 408         for (uint16_t j = 0; j < n_pkts; ++j) {
 409                 extract_key_core(mbufs[j], &task->keys[j]);
 410         }
 411 }
 412
 413 __attribute__((cold)) static void handle_error(struct rte_mbuf *mbuf)
 414 {
 415         struct core_net_pkt* core_pkt = rte_pktmbuf_mtod(mbuf, struct core_net_pkt *);
 416         uint32_t dst_ip = core_pkt->ip_hdr.dst_addr;
 417         uint32_t le_gre_id = rte_be_to_cpu_32(core_pkt->gre_hdr.gre_id);
 418
 419         plogx_dbg("Unknown IP %x/gre_id %x\n", dst_ip, le_gre_id);
 420 }
 421
 422 static int handle_qinq_encap4_bulk_pe(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
 423 {
 424         struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)tbase;
 425         uint64_t pkts_mask = RTE_LEN2MASK(n_pkts, uint64_t);
 426         struct cpe_data* entries[64];
 427         uint8_t out[MAX_PKT_BURST];
 428         uint64_t lookup_hit_mask;
 429
 430         prefetch_pkts(mbufs, n_pkts);
 431
 432         for (uint16_t j = 0; j < n_pkts; ++j) {
 433                 struct ipv4_hdr* ip = (struct ipv4_hdr *)(rte_pktmbuf_mtod(mbufs[j], struct ether_hdr *) + 1);
 434                 task->keys[j] = (uint64_t)ip->dst_addr;
 435         }
 436         rte_table_hash_key8_ext_dosig_ops.f_lookup(task->cpe_table, task->fake_packets, pkts_mask, &lookup_hit_mask, (void**)entries);
 437
 438         if (likely(lookup_hit_mask == pkts_mask)) {
 439                 for (uint16_t j = 0; j < n_pkts; ++j) {
 440                         struct cpe_pkt* cpe_pkt = (struct cpe_pkt*) rte_pktmbuf_prepend(mbufs[j], sizeof(struct qinq_hdr) - sizeof(struct ether_hdr));
 441                         uint16_t padlen = mbuf_calc_padlen(mbufs[j], cpe_pkt, &cpe_pkt->ipv4_hdr);
 442
 443                         if (padlen) {
 444                                 rte_pktmbuf_trim(mbufs[j], padlen);
 445                         }
 446                         out[j] = handle_qinq_encap4(task, cpe_pkt, mbufs[j], entries[j]);
 447                 }
 448         }
 449         else {
 450                 for (uint16_t j = 0; j < n_pkts; ++j) {
 451                         if (unlikely(!((lookup_hit_mask >> j) & 0x1))) {
 452                                 handle_error(mbufs[j]);
 453                                 out[j] = OUT_DISCARD;
 454                                 continue;
 455                         }
 456                         struct cpe_pkt* cpe_pkt = (struct cpe_pkt*) rte_pktmbuf_prepend(mbufs[j], sizeof(struct qinq_hdr) - sizeof(struct ether_hdr));
 457                         uint16_t padlen = mbuf_calc_padlen(mbufs[j], cpe_pkt, &cpe_pkt->ipv4_hdr);
 458
 459                         if (padlen) {
 460                                 rte_pktmbuf_trim(mbufs[j], padlen);
 461                         }
 462                         out[j] = handle_qinq_encap4(task, cpe_pkt, mbufs[j], entries[j]);
 463                 }
 464         }
 465
 466         return task->base.tx_pkt(&task->base, mbufs, n_pkts, out);
 467 }
 468 int handle_qinq_encap4_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
 469 {
 470         struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)tbase;
 471         uint64_t pkts_mask = RTE_LEN2MASK(n_pkts, uint64_t);
 472         struct cpe_data* entries[64];
 473         uint8_t out[MAX_PKT_BURST];
 474         uint64_t lookup_hit_mask;
 475
 476         prefetch_pkts(mbufs, n_pkts);
 477
 478         // From GRE ID and IP address, retrieve QinQ and MAC addresses
 479         extract_key_bulk(task, mbufs, n_pkts);
 480         rte_table_hash_key8_ext_dosig_ops.f_lookup(task->cpe_table, task->fake_packets, pkts_mask, &lookup_hit_mask, (void**)entries);
 481
 482         if (likely(lookup_hit_mask == pkts_mask)) {
 483                 for (uint16_t j = 0; j < n_pkts; ++j) {
 484                         struct cpe_pkt* cpe_pkt = (struct cpe_pkt*) rte_pktmbuf_adj(mbufs[j], UPSTREAM_DELTA);
 485                         // We are receiving GRE tunnelled packets (and removing UPSTRAM_DELTA bytes), whose length is > 64 bytes
 486                         // So there should be no padding, but in case the is one, remove it
 487                         uint16_t padlen = mbuf_calc_padlen(mbufs[j], cpe_pkt, &cpe_pkt->ipv4_hdr);
 488
 489                         if (padlen) {
 490                                 rte_pktmbuf_trim(mbufs[j], padlen);
 491                         }
 492                         out[j] = handle_qinq_encap4(task, cpe_pkt, mbufs[j], entries[j]);
 493                 }
 494         }
 495         else {
 496                 for (uint16_t j = 0; j < n_pkts; ++j) {
 497                         if (unlikely(!((lookup_hit_mask >> j) & 0x1))) {
 498                                 handle_error(mbufs[j]);
 499                                 out[j] = OUT_DISCARD;
 500                                 continue;
 501                         }
 502                         struct cpe_pkt* cpe_pkt = (struct cpe_pkt*) rte_pktmbuf_adj(mbufs[j], UPSTREAM_DELTA);
 503                         uint16_t padlen = mbuf_calc_padlen(mbufs[j], cpe_pkt, &cpe_pkt->ipv4_hdr);
 504
 505                         if (padlen) {
 506                                 rte_pktmbuf_trim(mbufs[j], padlen);
 507                         }
 508                         out[j] = handle_qinq_encap4(task, cpe_pkt, mbufs[j], entries[j]);
 509                 }
 510         }
 511
 512         return task->base.tx_pkt(&task->base, mbufs, n_pkts, out);
 513 }
 514
 515 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)
 516 {
 517         PROX_ASSERT(cpe_pkt);
 518
 519         if (cpe_pkt->ipv4_hdr.time_to_live) {
 520                 cpe_pkt->ipv4_hdr.time_to_live--;
 521         }
 522         else {
 523                 plog_info("TTL = 0 => Dropping\n");
 524                 return OUT_DISCARD;
 525         }
 526         cpe_pkt->ipv4_hdr.hdr_checksum = 0;
 527
 528         restore_cpe(cpe_pkt, entry, task->qinq_tag, task->src_mac);
 529
 530         if (task->runtime_flags & TASK_CLASSIFY) {
 531                 uint8_t queue = task->dscp[cpe_pkt->ipv4_hdr.type_of_service >> 2] & 0x3;
 532                 uint8_t tc = task->dscp[cpe_pkt->ipv4_hdr.type_of_service >> 2] >> 2;
 533
 534                 rte_sched_port_pkt_write(mbuf, 0, entry->user, tc, queue, 0);
 535         }
 536 #ifdef ENABLE_EXTRA_USER_STATISTICS
 537         task->stats_per_user[entry->user]++;
 538 #endif
 539         if (task->runtime_flags & TASK_TX_CRC) {
 540                 prox_ip_cksum(mbuf, &cpe_pkt->ipv4_hdr, sizeof(struct qinq_hdr), sizeof(struct ipv4_hdr), task->offload_crc);
 541         }
 542         return entry->mac_port.out_idx;
 543 }
 544
 545 static void flow_iter_next(struct flow_iter *iter, struct task_args *targ)
 546 {
 547         do {
 548                 iter->idx++;
 549                 uint8_t flag_features = iter->data;
 550
 551                 if (flag_features & TASK_FEATURE_LUT_QINQ_RSS) {
 552                         // If there is a load balancer, number of worker thread is indicated by targ->nb_slave_threads and n_rxq = 0
 553                         // If there is no load balancers, number of worker thread is indicated by n_rxq and nb_slave_threads = 0
 554                         uint8_t nb_worker_threads, worker_thread_id;
 555                         nb_worker_threads = 1;
 556                         worker_thread_id = 1;
 557                         if (targ->nb_slave_threads) {
 558                                 nb_worker_threads = targ->nb_slave_threads;
 559                                 worker_thread_id = targ->worker_thread_id;
 560                         } else if (prox_port_cfg[targ->rx_port_queue[0].port].n_rxq) {
 561                                 nb_worker_threads = prox_port_cfg[targ->rx_port_queue[0].port].n_rxq;
 562                                 worker_thread_id = targ->rx_port_queue[0].queue;
 563                         } else {
 564                                 plog_err("Unexpected: unknown number of worker thread\n");
 565                         }
 566
 567                         if (targ->nb_slave_threads == 0 || rss_to_queue(get_qinq_gre_map(targ)->entries[iter->idx].rss, nb_worker_threads) == worker_thread_id)
 568                                 break;
 569                 } else if (flag_features & TASK_FEATURE_LUT_QINQ_HASH) {
 570                         uint64_t cvlan = rte_bswap16(get_qinq_gre_map(targ)->entries[iter->idx].cvlan & 0xFF0F);
 571                         uint64_t svlan = rte_bswap16(get_qinq_gre_map(targ)->entries[iter->idx].svlan & 0xFF0F);
 572                         uint64_t qinq = rte_bswap64((svlan << 32) | cvlan);
 573                         uint8_t queue = hash_crc32(&qinq, 8, 0) % targ->nb_slave_threads;
 574                         if (queue == targ->worker_thread_id)
 575                                 break;
 576                 } else if (flag_features & TASK_FEATURE_GRE_ID) {
 577                         if (get_qinq_gre_map(targ)->entries[iter->idx].gre_id % targ->nb_slave_threads == targ->worker_thread_id)
 578                                 break;
 579                 }
 580         } while (iter->idx != (int)get_qinq_gre_map(targ)->count);
 581 }
 582
 583 static void flow_iter_beg(struct flow_iter *iter, struct task_args *targ)
 584 {
 585         uint32_t flag_features = 0;
 586         if (targ->lb_friend_core != 0xFF) {
 587                 struct task_args *lb_targ = &lcore_cfg[targ->lb_friend_core].targs[targ->lb_friend_task];
 588                 flag_features = lb_targ->task_init->flag_features;
 589                 plog_info("\t\tWT %d Updated features to %x from friend %d\n", targ->lconf->id, flag_features, targ->lb_friend_core);
 590         } else {
 591                 plog_info("\t\tWT %d has no friend\n", targ->lconf->id);
 592         }
 593         if (targ->nb_slave_threads == 0)  {
 594                 // No slave threads, i.e. using RSS
 595                 plog_info("feature was %x is now %x\n", flag_features, TASK_FEATURE_LUT_QINQ_RSS);
 596                 flag_features = TASK_FEATURE_LUT_QINQ_RSS;
 597         }
 598         if ((flag_features & (TASK_FEATURE_GRE_ID|TASK_FEATURE_LUT_QINQ_RSS|TASK_FEATURE_LUT_QINQ_HASH)) == 0) {
 599                 plog_info("\t\tCould not find flag feature from Load balancer => supposing TASK_FEATURE_GRE_ID\n");
 600                 flag_features = TASK_FEATURE_GRE_ID;
 601         }
 602
 603         iter->idx = -1;
 604         flow_iter_next(iter, targ);
 605 }
 606
 607 static int flow_iter_is_end(struct flow_iter *iter, struct task_args *targ)
 608 {
 609         return iter->idx == (int)get_qinq_gre_map(targ)->count;
 610 }
 611
 612 static uint32_t flow_iter_get_gre_id(struct flow_iter *iter, struct task_args *targ)
 613 {
 614         return get_qinq_gre_map(targ)->entries[iter->idx].gre_id;
 615 }
 616
 617 static struct task_init task_init_qinq_encap4_table = {
 618         .mode = QINQ_ENCAP4,
 619         .mode_str = "qinqencapv4",
 620         .early_init = early_init_table,
 621         .init = init_task_qinq_encap4,
 622         .handle = handle_qinq_encap4_bulk,
 623         /* In this case user in qinq_lookup table is the QoS user
 624            (from user_table), i.e. usually from 0 to 32K Otherwise it
 625            would have been a user from (0 to n_interface x 32K) */
 626         .flow_iter = {
 627                 .beg        = flow_iter_beg,
 628                 .is_end     = flow_iter_is_end,
 629                 .next       = flow_iter_next,
 630                 .get_gre_id = flow_iter_get_gre_id,
 631         },
 632         .flag_features = TASK_FEATURE_CLASSIFY,
 633         .size = sizeof(struct task_qinq_encap4)
 634 };
 635
 636 static struct task_init task_init_qinq_encap4_table_pe = {
 637         .mode = QINQ_ENCAP4,
 638         .mode_str = "qinqencapv4",
 639         .sub_mode_str = "pe",
 640         .early_init = early_init_table,
 641         .init = init_task_qinq_encap4,
 642         .handle = handle_qinq_encap4_bulk_pe,
 643         .flag_features = TASK_FEATURE_CLASSIFY,
 644         .size = sizeof(struct task_qinq_encap4)
 645 };
 646
 647 static struct task_init task_init_qinq_encap4_untag = {
 648         .mode = QINQ_ENCAP4,
 649         .sub_mode_str = "unmpls",
 650         .mode_str = "qinqencapv4",
 651         .init = init_task_qinq_encap4,
 652         .handle = handle_qinq_encap4_untag_bulk,
 653         .flag_features = TASK_FEATURE_CLASSIFY,
 654         .size = sizeof(struct task_qinq_encap4)
 655 };
 656
 657 __attribute__((constructor)) static void reg_task_qinq_encap4(void)
 658 {
 659         reg_task(&task_init_qinq_encap4_table);
 660         reg_task(&task_init_qinq_encap4_table_pe);
 661         reg_task(&task_init_qinq_encap4_untag);
 662 }