2 // Copyright (c) 2010-2020 Intel Corporation
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
8 // http://www.apache.org/licenses/LICENSE-2.0
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.
17 #include <rte_lcore.h>
19 #include <rte_hash_crc.h>
20 #include "task_base.h"
25 #include "handle_master.h"
26 #include "prox_port_cfg.h"
28 static inline int find_ip(struct ether_hdr_arp *pkt, uint16_t len, uint32_t *ip_dst)
30 prox_rte_vlan_hdr *vlan_hdr;
31 prox_rte_ether_hdr *eth_hdr = (prox_rte_ether_hdr*)pkt;
32 prox_rte_ipv4_hdr *ip;
33 uint16_t ether_type = eth_hdr->ether_type;
34 uint16_t l2_len = sizeof(prox_rte_ether_hdr);
37 while (((ether_type == ETYPE_8021ad) || (ether_type == ETYPE_VLAN)) && (l2_len + sizeof(prox_rte_vlan_hdr) < len)) {
38 vlan_hdr = (prox_rte_vlan_hdr *)((uint8_t *)pkt + l2_len);
40 ether_type = vlan_hdr->eth_proto;
46 // In case of MPLS, next hop MAC is based on MPLS, not destination IP
58 plog_warn("Unsupported packet type %x - CRC might be wrong\n", ether_type);
62 if (l2_len && (l2_len + sizeof(prox_rte_ipv4_hdr) <= len)) {
63 prox_rte_ipv4_hdr *ip = (prox_rte_ipv4_hdr *)((uint8_t *)pkt + l2_len);
64 // TODO: implement LPM => replace ip_dst by next hop IP DST
65 *ip_dst = ip->dst_addr;
71 /* This implementation could be improved: instead of checking each time we send a packet whether we need also
72 to send an ARP, we should only check whether the MAC is valid.
73 We should check arp_update_time in the master process. This would also require the generating task to clear its arp ring
74 to avoid sending many ARP while starting after a long stop.
75 We could also check for arp_timeout in the master so that dataplane has only to check whether MAC is available
76 but this would require either thread safety, or the the exchange of information between master and generating core.
79 int write_dst_mac(struct task_base *tbase, struct rte_mbuf *mbuf, uint32_t *ip_dst)
81 const uint64_t hz = rte_get_tsc_hz();
82 struct ether_hdr_arp *packet = rte_pktmbuf_mtod(mbuf, struct ether_hdr_arp *);
83 prox_rte_ether_addr *mac = &packet->ether_hdr.d_addr;
85 uint64_t tsc = rte_rdtsc();
86 struct l3_base *l3 = &(tbase->l3);
88 if (likely((l3->flags & FLAG_DST_MAC_KNOWN) && (tsc < l3->gw.arp_update_time) && (tsc < l3->gw.arp_timeout))) {
89 memcpy(mac, &l3->gw.mac, sizeof(prox_rte_ether_addr));
91 } else if (tsc > l3->gw.arp_update_time) {
92 // long time since we have sent an arp, send arp
93 l3->gw.arp_update_time = tsc + l3->arp_update_time * hz / 1000;
95 if ((l3->flags & FLAG_DST_MAC_KNOWN) && (tsc < l3->gw.arp_timeout)){
96 // MAC is valid in the table => send also the mbuf
97 memcpy(mac, &l3->gw.mac, sizeof(prox_rte_ether_addr));
98 return SEND_MBUF_AND_ARP;
100 // MAC still unknown, or timed out => only send ARP
104 // MAC is unknown and we already sent an ARP recently, drop mbuf and wait for ARP reply
109 uint16_t len = rte_pktmbuf_pkt_len(mbuf);
110 if (find_ip(packet, len, ip_dst) != 0) {
111 // Unable to find IP address => non IP packet => send it as it
114 if (likely(l3->n_pkts < 4)) {
115 for (unsigned int idx = 0; idx < l3->n_pkts; idx++) {
116 if (*ip_dst == l3->optimized_arp_table[idx].ip) {
117 // IP address already in table
118 if ((tsc < l3->optimized_arp_table[idx].arp_update_time) && (tsc < l3->optimized_arp_table[idx].arp_timeout)) {
119 // MAC address was recently updated in table, use it
120 memcpy(mac, &l3->optimized_arp_table[idx].mac, sizeof(prox_rte_ether_addr));
122 } else if (tsc > l3->optimized_arp_table[idx].arp_update_time) {
123 // ARP not sent since a long time, send ARP
124 l3->optimized_arp_table[idx].arp_update_time = tsc + l3->arp_update_time * hz / 1000;
125 if (tsc < l3->optimized_arp_table[idx].arp_timeout) {
126 // MAC still valid => also send mbuf
127 memcpy(mac, &l3->optimized_arp_table[idx].mac, sizeof(prox_rte_ether_addr));
128 return SEND_MBUF_AND_ARP;
130 // MAC unvalid => only send ARP
134 // ARP timeout elapsed, MAC not valid anymore but waiting for ARP reply
139 // IP address not found in table
140 l3->optimized_arp_table[l3->n_pkts].ip = *ip_dst;
141 l3->optimized_arp_table[l3->n_pkts].arp_update_time = tsc + l3->arp_update_time * hz / 1000;
144 if (l3->n_pkts < 4) {
148 // We have too many IP addresses to search linearly; lets use hash table instead => copy all entries in hash table
149 for (uint32_t idx = 0; idx < l3->n_pkts; idx++) {
150 uint32_t ip = l3->optimized_arp_table[idx].ip;
151 int ret = rte_hash_add_key(l3->ip_hash, (const void *)&ip);
153 // This should not happen as few entries so far.
154 // If it happens, we still send the ARP as easier:
155 // If the ARP corresponds to this error, the ARP reply will be ignored
156 // If ARP does not correspond to this error/ip, then ARP reply will be handled.
157 plogx_err("Unable add ip %d.%d.%d.%d in mac_hash (already %d entries)\n", IP4(ip), idx);
159 memcpy(&l3->arp_table[ret], &l3->optimized_arp_table[idx], sizeof(struct arp_table));
164 // Find IP in lookup table. Send ARP if not found
165 int ret = rte_hash_lookup(l3->ip_hash, (const void *)ip_dst);
166 if (unlikely(ret < 0)) {
167 // IP not found, try to send an ARP
168 int ret = rte_hash_add_key(l3->ip_hash, (const void *)ip_dst);
170 // No reason to send ARP, as reply would be anyhow ignored
171 plogx_err("Unable to add ip %d.%d.%d.%d in mac_hash\n", IP4(*ip_dst));
174 l3->arp_table[ret].ip = *ip_dst;
175 l3->arp_table[ret].arp_update_time = tsc + l3->arp_update_time * hz / 1000;
180 if (likely((tsc < l3->arp_table[ret].arp_update_time) && (tsc < l3->arp_table[ret].arp_timeout))) {
181 // MAC still valid and ARP sent recently
182 memcpy(mac, &l3->arp_table[ret].mac, sizeof(prox_rte_ether_addr));
184 } else if (tsc > l3->arp_table[ret].arp_update_time) {
185 // ARP not sent since a long time, send ARP
186 l3->arp_table[ret].arp_update_time = tsc + l3->arp_update_time * hz / 1000;
187 if (tsc < l3->arp_table[ret].arp_timeout) {
188 // MAC still valid => send also MBUF
189 memcpy(mac, &l3->arp_table[ret].mac, sizeof(prox_rte_ether_addr));
190 return SEND_MBUF_AND_ARP;
203 void task_init_l3(struct task_base *tbase, struct task_args *targ)
205 static char hash_name[30];
206 uint32_t n_entries = MAX_ARP_ENTRIES * 4;
207 const int socket_id = rte_lcore_to_socket_id(targ->lconf->id);
208 sprintf(hash_name, "A%03d_%03d_mac_table", targ->lconf->id, targ->id);
212 struct rte_hash_parameters hash_params = {
214 .entries = n_entries,
215 .key_len = sizeof(uint32_t),
216 .hash_func = rte_hash_crc,
217 .hash_func_init_val = 0,
219 tbase->l3.ip_hash = rte_hash_create(&hash_params);
220 PROX_PANIC(tbase->l3.ip_hash == NULL, "Failed to set up ip hash table\n");
222 tbase->l3.arp_table = (struct arp_table *)prox_zmalloc(n_entries * sizeof(struct arp_table), socket_id);
223 PROX_PANIC(tbase->l3.arp_table == NULL, "Failed to allocate memory for %u entries in arp table\n", n_entries);
224 plog_info("\tarp table, with %d entries of size %ld\n", n_entries, sizeof(struct l3_base));
226 targ->lconf->ctrl_func_p[targ->task] = handle_ctrl_plane_pkts;
227 targ->lconf->ctrl_timeout = freq_to_tsc(targ->ctrl_freq);
228 tbase->l3.gw.ip = rte_cpu_to_be_32(targ->gateway_ipv4);
229 tbase->flags |= TASK_L3;
230 tbase->l3.core_id = targ->lconf->id;
231 tbase->l3.task_id = targ->id;
232 tbase->l3.tmaster = targ->tmaster;
233 if (targ->arp_timeout != 0)
234 tbase->l3.arp_timeout = targ->arp_timeout;
236 tbase->l3.arp_timeout = DEFAULT_ARP_TIMEOUT;
237 if (targ->arp_update_time != 0)
238 tbase->l3.arp_update_time = targ->arp_update_time;
240 tbase->l3.arp_update_time = DEFAULT_ARP_UPDATE_TIME;
243 void task_start_l3(struct task_base *tbase, struct task_args *targ)
245 const int NB_ARP_MBUF = 1024;
246 const int ARP_MBUF_SIZE = 2048;
247 const int NB_CACHE_ARP_MBUF = 256;
249 struct prox_port_cfg *port = find_reachable_port(targ);
250 if (port && (tbase->l3.arp_pool == NULL)) {
251 static char name[] = "arp0_pool";
252 tbase->l3.reachable_port_id = port - prox_port_cfg;
253 if (targ->local_ipv4) {
254 tbase->local_ipv4 = rte_be_to_cpu_32(targ->local_ipv4);
255 register_ip_to_ctrl_plane(tbase->l3.tmaster, tbase->local_ipv4, tbase->l3.reachable_port_id, targ->lconf->id, targ->id);
257 master_init_vdev(tbase->l3.tmaster, tbase->l3.reachable_port_id, targ->lconf->id, targ->id);
259 struct rte_mempool *ret = rte_mempool_create(name, NB_ARP_MBUF, ARP_MBUF_SIZE, NB_CACHE_ARP_MBUF,
260 sizeof(struct rte_pktmbuf_pool_private), rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, 0,
262 PROX_PANIC(ret == NULL, "Failed to allocate ARP memory pool on socket %u with %u elements\n",
263 rte_socket_id(), NB_ARP_MBUF);
264 plog_info("\t\tMempool %p (%s) size = %u * %u cache %u, socket %d\n", ret, name, NB_ARP_MBUF,
265 ARP_MBUF_SIZE, NB_CACHE_ARP_MBUF, rte_socket_id());
266 tbase->l3.arp_pool = ret;
270 void task_set_gateway_ip(struct task_base *tbase, uint32_t ip)
272 tbase->l3.gw.ip = ip;
273 tbase->flags &= ~FLAG_DST_MAC_KNOWN;
276 void task_set_local_ip(struct task_base *tbase, uint32_t ip)
278 tbase->local_ipv4 = ip;
281 static void reset_arp_update_time(struct l3_base *l3, uint32_t ip)
284 plogx_info("\tMAC entry for IP "IPv4_BYTES_FMT" timeout in kernel\n", IP4(ip));
285 if (ip == l3->gw.ip) {
286 l3->gw.arp_update_time = 0;
287 } else if (l3->n_pkts < 4) {
288 for (idx = 0; idx < l3->n_pkts; idx++) {
289 uint32_t ip_dst = l3->optimized_arp_table[idx].ip;
293 if (idx < l3->n_pkts) {
294 l3->optimized_arp_table[idx].arp_update_time = 0;
297 int ret = rte_hash_lookup(l3->ip_hash, (const void *)&ip);
299 l3->arp_table[ret].arp_update_time = 0;
304 void handle_ctrl_plane_pkts(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
307 const uint64_t hz = rte_get_tsc_hz();
308 uint32_t ip, ip_dst, idx;
311 struct ether_hdr_arp *hdr;
312 struct l3_base *l3 = &tbase->l3;
313 uint64_t tsc= rte_rdtsc();
314 uint64_t update_time = l3->arp_timeout * hz / 1000;
316 for (j = 0; j < n_pkts; ++j) {
319 for (j = 0; j < n_pkts; ++j) {
320 PREFETCH0(rte_pktmbuf_mtod(mbufs[j], void *));
323 for (j = 0; j < n_pkts; ++j) {
324 out[0] = OUT_HANDLED;
325 command = mbufs[j]->udata64 & 0xFFFF;
326 plogx_dbg("\tReceived %s mbuf %p\n", actions_string[command], mbufs[j]);
328 case UPDATE_FROM_CTRL:
329 hdr = rte_pktmbuf_mtod(mbufs[j], struct ether_hdr_arp *);
330 ip = (mbufs[j]->udata64 >> 32) & 0xFFFFFFFF;
332 if (prox_rte_is_zero_ether_addr(&hdr->arp.data.sha)) {
333 // MAC timeout or deleted from kernel table => reset update_time
334 // This will cause us to send new ARP request
335 // However, as arp_timeout not touched, we should continue sending our regular IP packets
336 reset_arp_update_time(l3, ip);
337 plogx_info("\tTimeout for MAC entry for IP "IPv4_BYTES_FMT"\n", IP4(ip));
340 plogx_dbg("\tUpdating MAC entry for IP "IPv4_BYTES_FMT" with MAC "MAC_BYTES_FMT"\n",
341 IP4(ip), MAC_BYTES(hdr->arp.data.sha.addr_bytes));
342 if (ip == l3->gw.ip) {
343 // MAC address of the gateway
344 memcpy(&l3->gw.mac, &hdr->arp.data.sha, 6);
345 l3->flags |= FLAG_DST_MAC_KNOWN;
346 l3->gw.arp_timeout = tsc + update_time;
347 } else if (l3->n_pkts < 4) {
348 // Few packets tracked - should be faster to loop through them thean using a hash table
349 for (idx = 0; idx < l3->n_pkts; idx++) {
350 ip_dst = l3->optimized_arp_table[idx].ip;
354 if (idx < l3->n_pkts) {
355 memcpy(&l3->optimized_arp_table[idx].mac, &(hdr->arp.data.sha), sizeof(prox_rte_ether_addr));
356 l3->optimized_arp_table[idx].arp_timeout = tsc + update_time;
359 int ret = rte_hash_add_key(l3->ip_hash, (const void *)&ip);
361 plogx_info("Unable add ip %d.%d.%d.%d in mac_hash\n", IP4(ip));
363 memcpy(&l3->arp_table[ret].mac, &(hdr->arp.data.sha), sizeof(prox_rte_ether_addr));
364 l3->arp_table[ret].arp_timeout = tsc + update_time;
369 case ARP_REPLY_FROM_CTRL:
371 case ARP_REQ_FROM_CTRL:
374 // tx_ctrlplane_pkt does not drop packets
375 plogx_dbg("\tForwarding (ARP/PING) packet from master\n");
376 tbase->aux->tx_ctrlplane_pkt(tbase, &mbufs[j], 1, out);
377 TASK_STATS_ADD_TX_NON_DP(&tbase->aux->stats, 1);