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"
27 #include "packet_utils.h"
29 static inline int find_ip(struct ether_hdr_arp *pkt, uint16_t len, uint32_t *ip_dst)
31 prox_rte_vlan_hdr *vlan_hdr;
32 prox_rte_ether_hdr *eth_hdr = (prox_rte_ether_hdr*)pkt;
33 prox_rte_ipv4_hdr *ip;
34 uint16_t ether_type = eth_hdr->ether_type;
35 uint16_t l2_len = sizeof(prox_rte_ether_hdr);
38 while (((ether_type == ETYPE_8021ad) || (ether_type == ETYPE_VLAN)) && (l2_len + sizeof(prox_rte_vlan_hdr) < len)) {
39 vlan_hdr = (prox_rte_vlan_hdr *)((uint8_t *)pkt + l2_len);
41 ether_type = vlan_hdr->eth_proto;
47 // In case of MPLS, next hop MAC is based on MPLS, not destination IP
59 plog_warn("Unsupported packet type %x - CRC might be wrong\n", ether_type);
63 if (l2_len && (l2_len + sizeof(prox_rte_ipv4_hdr) <= len)) {
64 prox_rte_ipv4_hdr *ip = (prox_rte_ipv4_hdr *)((uint8_t *)pkt + l2_len);
65 // TODO: implement LPM => replace ip_dst by next hop IP DST
66 *ip_dst = ip->dst_addr;
72 /* This implementation could be improved: instead of checking each time we send a packet whether we need also
73 to send an ARP, we should only check whether the MAC is valid.
74 We should check arp_update_time in the master process. This would also require the generating task to clear its arp ring
75 to avoid sending many ARP while starting after a long stop.
76 We could also check for arp_timeout in the master so that dataplane has only to check whether MAC is available
77 but this would require either thread safety, or the the exchange of information between master and generating core.
80 int write_dst_mac(struct task_base *tbase, struct rte_mbuf *mbuf, uint32_t *ip_dst, uint64_t **time)
82 const uint64_t hz = rte_get_tsc_hz();
83 struct ether_hdr_arp *packet = rte_pktmbuf_mtod(mbuf, struct ether_hdr_arp *);
84 prox_rte_ether_addr *mac = &packet->ether_hdr.d_addr;
86 uint64_t tsc = rte_rdtsc();
87 struct l3_base *l3 = &(tbase->l3);
89 if (likely((l3->flags & FLAG_DST_MAC_KNOWN) && (tsc < l3->gw.arp_update_time) && (tsc < l3->gw.arp_timeout))) {
90 memcpy(mac, &l3->gw.mac, sizeof(prox_rte_ether_addr));
92 } else if (tsc > l3->gw.arp_update_time) {
93 // long time since we have successfully sent an arp, send arp
94 // If sending ARP failed (ring full) then arp_update_time is not updated to avoid having to wait 1 sec to send ARP REQ again
95 *time = &l3->gw.arp_update_time;
97 if ((l3->flags & FLAG_DST_MAC_KNOWN) && (tsc < l3->gw.arp_timeout)){
98 // MAC is valid in the table => send also the mbuf
99 memcpy(mac, &l3->gw.mac, sizeof(prox_rte_ether_addr));
100 return SEND_MBUF_AND_ARP;
102 // MAC still unknown, or timed out => only send ARP
106 // MAC is unknown and we already sent an ARP recently, drop mbuf and wait for ARP reply
111 uint16_t len = rte_pktmbuf_pkt_len(mbuf);
112 if (find_ip(packet, len, ip_dst) != 0) {
113 // Unable to find IP address => non IP packet => send it as it
116 if (likely(l3->n_pkts < 4)) {
117 for (unsigned int idx = 0; idx < l3->n_pkts; idx++) {
118 if (*ip_dst == l3->optimized_arp_table[idx].ip) {
119 // IP address already in table
120 if ((tsc < l3->optimized_arp_table[idx].arp_update_time) && (tsc < l3->optimized_arp_table[idx].arp_timeout)) {
121 // MAC address was recently updated in table, use it
122 memcpy(mac, &l3->optimized_arp_table[idx].mac, sizeof(prox_rte_ether_addr));
124 } else if (tsc > l3->optimized_arp_table[idx].arp_update_time) {
125 // ARP not sent since a long time, send ARP
126 *time = &l3->optimized_arp_table[idx].arp_update_time;
127 if (tsc < l3->optimized_arp_table[idx].arp_timeout) {
128 // MAC still valid => also send mbuf
129 memcpy(mac, &l3->optimized_arp_table[idx].mac, sizeof(prox_rte_ether_addr));
130 return SEND_MBUF_AND_ARP;
132 // MAC unvalid => only send ARP
136 // ARP timeout elapsed, MAC not valid anymore but waiting for ARP reply
141 // IP address not found in table
142 l3->optimized_arp_table[l3->n_pkts].ip = *ip_dst;
143 *time = &l3->optimized_arp_table[l3->n_pkts].arp_update_time;
146 if (l3->n_pkts < 4) {
150 // We have too many IP addresses to search linearly; lets use hash table instead => copy all entries in hash table
151 for (uint32_t idx = 0; idx < l3->n_pkts; idx++) {
152 uint32_t ip = l3->optimized_arp_table[idx].ip;
153 int ret = rte_hash_add_key(l3->ip_hash, (const void *)&ip);
155 // This should not happen as few entries so far.
156 // If it happens, we still send the ARP as easier:
157 // If the ARP corresponds to this error, the ARP reply will be ignored
158 // If ARP does not correspond to this error/ip, then ARP reply will be handled.
159 plogx_err("Unable add ip %d.%d.%d.%d in mac_hash (already %d entries)\n", IP4(ip), idx);
161 memcpy(&l3->arp_table[ret], &l3->optimized_arp_table[idx], sizeof(struct arp_table));
166 // Find IP in lookup table. Send ARP if not found
167 int ret = rte_hash_lookup(l3->ip_hash, (const void *)ip_dst);
168 if (unlikely(ret < 0)) {
169 // IP not found, try to send an ARP
170 int ret = rte_hash_add_key(l3->ip_hash, (const void *)ip_dst);
172 // No reason to send ARP, as reply would be anyhow ignored
173 plogx_err("Unable to add ip %d.%d.%d.%d in mac_hash\n", IP4(*ip_dst));
176 l3->arp_table[ret].ip = *ip_dst;
177 *time = &l3->arp_table[ret].arp_update_time;
182 if (likely((tsc < l3->arp_table[ret].arp_update_time) && (tsc < l3->arp_table[ret].arp_timeout))) {
183 // MAC still valid and ARP sent recently
184 memcpy(mac, &l3->arp_table[ret].mac, sizeof(prox_rte_ether_addr));
186 } else if (tsc > l3->arp_table[ret].arp_update_time) {
187 // ARP not sent since a long time, send ARP
188 *time = &l3->arp_table[ret].arp_update_time;
189 if (tsc < l3->arp_table[ret].arp_timeout) {
190 // MAC still valid => send also MBUF
191 memcpy(mac, &l3->arp_table[ret].mac, sizeof(prox_rte_ether_addr));
192 return SEND_MBUF_AND_ARP;
205 void task_init_l3(struct task_base *tbase, struct task_args *targ)
207 static char hash_name[30];
208 uint32_t n_entries = MAX_ARP_ENTRIES * 4;
209 const int socket_id = rte_lcore_to_socket_id(targ->lconf->id);
210 sprintf(hash_name, "A%03d_%03d_mac_table", targ->lconf->id, targ->id);
214 struct rte_hash_parameters hash_params = {
216 .entries = n_entries,
217 .key_len = sizeof(uint32_t),
218 .hash_func = rte_hash_crc,
219 .hash_func_init_val = 0,
221 tbase->l3.ip_hash = rte_hash_create(&hash_params);
222 PROX_PANIC(tbase->l3.ip_hash == NULL, "Failed to set up ip hash table\n");
224 tbase->l3.arp_table = (struct arp_table *)prox_zmalloc(n_entries * sizeof(struct arp_table), socket_id);
225 PROX_PANIC(tbase->l3.arp_table == NULL, "Failed to allocate memory for %u entries in arp table\n", n_entries);
226 plog_info("\tarp table, with %d entries of size %ld\n", n_entries, sizeof(struct l3_base));
228 targ->lconf->ctrl_func_p[targ->task] = handle_ctrl_plane_pkts;
229 targ->lconf->ctrl_timeout = freq_to_tsc(targ->ctrl_freq);
230 tbase->l3.gw.ip = rte_cpu_to_be_32(targ->gateway_ipv4);
231 tbase->flags |= TASK_L3;
232 tbase->l3.core_id = targ->lconf->id;
233 tbase->l3.task_id = targ->id;
234 tbase->l3.tmaster = targ->tmaster;
235 tbase->l3.seed = (uint)rte_rdtsc();
236 if (targ->arp_timeout != 0)
237 tbase->l3.arp_timeout = targ->arp_timeout;
239 tbase->l3.arp_timeout = DEFAULT_ARP_TIMEOUT;
240 if (targ->arp_update_time != 0)
241 tbase->l3.arp_update_time = targ->arp_update_time;
243 tbase->l3.arp_update_time = DEFAULT_ARP_UPDATE_TIME;
246 void task_start_l3(struct task_base *tbase, struct task_args *targ)
248 const int NB_ARP_MBUF = 1024;
249 const int ARP_MBUF_SIZE = 2048;
250 const int NB_CACHE_ARP_MBUF = 256;
252 struct prox_port_cfg *port = find_reachable_port(targ);
253 if (port && (tbase->l3.arp_pool == NULL)) {
254 static char name[] = "arp0_pool";
255 tbase->l3.reachable_port_id = port - prox_port_cfg;
256 if (targ->local_ipv4) {
257 tbase->local_ipv4 = rte_be_to_cpu_32(targ->local_ipv4);
258 register_ip_to_ctrl_plane(tbase->l3.tmaster, tbase->local_ipv4, tbase->l3.reachable_port_id, targ->lconf->id, targ->id);
260 master_init_vdev(tbase->l3.tmaster, tbase->l3.reachable_port_id, targ->lconf->id, targ->id);
262 struct rte_mempool *ret = rte_mempool_create(name, NB_ARP_MBUF, ARP_MBUF_SIZE, NB_CACHE_ARP_MBUF,
263 sizeof(struct rte_pktmbuf_pool_private), rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, 0,
265 PROX_PANIC(ret == NULL, "Failed to allocate ARP memory pool on socket %u with %u elements\n",
266 rte_socket_id(), NB_ARP_MBUF);
267 plog_info("\t\tMempool %p (%s) size = %u * %u cache %u, socket %d\n", ret, name, NB_ARP_MBUF,
268 ARP_MBUF_SIZE, NB_CACHE_ARP_MBUF, rte_socket_id());
269 tbase->l3.arp_pool = ret;
273 void task_set_gateway_ip(struct task_base *tbase, uint32_t ip)
275 tbase->l3.gw.ip = ip;
276 tbase->flags &= ~FLAG_DST_MAC_KNOWN;
279 void task_set_local_ip(struct task_base *tbase, uint32_t ip)
281 tbase->local_ipv4 = ip;
284 static void reset_arp_update_time(struct l3_base *l3, uint32_t ip)
287 plogx_info("\tMAC entry for IP "IPv4_BYTES_FMT" timeout in kernel\n", IP4(ip));
288 if (ip == l3->gw.ip) {
289 l3->gw.arp_update_time = 0;
290 } else if (l3->n_pkts < 4) {
291 for (idx = 0; idx < l3->n_pkts; idx++) {
292 uint32_t ip_dst = l3->optimized_arp_table[idx].ip;
296 if (idx < l3->n_pkts) {
297 l3->optimized_arp_table[idx].arp_update_time = 0;
300 int ret = rte_hash_lookup(l3->ip_hash, (const void *)&ip);
302 l3->arp_table[ret].arp_update_time = 0;
307 void handle_ctrl_plane_pkts(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
310 const uint64_t hz = rte_get_tsc_hz();
311 uint32_t ip, ip_dst, idx;
314 struct ether_hdr_arp *hdr;
315 struct l3_base *l3 = &tbase->l3;
316 uint64_t tsc= rte_rdtsc();
317 uint64_t update_time = l3->arp_timeout * hz / 1000;
319 for (j = 0; j < n_pkts; ++j) {
322 for (j = 0; j < n_pkts; ++j) {
323 PREFETCH0(rte_pktmbuf_mtod(mbufs[j], void *));
326 for (j = 0; j < n_pkts; ++j) {
327 out[0] = OUT_HANDLED;
328 command = mbufs[j]->udata64 & 0xFFFF;
329 plogx_dbg("\tReceived %s mbuf %p\n", actions_string[command], mbufs[j]);
331 case UPDATE_FROM_CTRL:
332 hdr = rte_pktmbuf_mtod(mbufs[j], struct ether_hdr_arp *);
333 ip = (mbufs[j]->udata64 >> 32) & 0xFFFFFFFF;
335 if (prox_rte_is_zero_ether_addr(&hdr->arp.data.sha)) {
336 // MAC timeout or deleted from kernel table => reset update_time
337 // This will cause us to send new ARP request
338 // However, as arp_timeout not touched, we should continue sending our regular IP packets
339 reset_arp_update_time(l3, ip);
340 plogx_info("\tTimeout for MAC entry for IP "IPv4_BYTES_FMT"\n", IP4(ip));
343 plogx_dbg("\tUpdating MAC entry for IP "IPv4_BYTES_FMT" with MAC "MAC_BYTES_FMT"\n",
344 IP4(ip), MAC_BYTES(hdr->arp.data.sha.addr_bytes));
345 if (ip == l3->gw.ip) {
346 // MAC address of the gateway
347 memcpy(&l3->gw.mac, &hdr->arp.data.sha, 6);
348 l3->flags |= FLAG_DST_MAC_KNOWN;
349 l3->gw.arp_timeout = tsc + update_time;
350 update_arp_update_time(l3, &l3->gw.arp_update_time, l3->arp_update_time);
351 } else if (l3->n_pkts < 4) {
352 // Few packets tracked - should be faster to loop through them thean using a hash table
353 for (idx = 0; idx < l3->n_pkts; idx++) {
354 ip_dst = l3->optimized_arp_table[idx].ip;
358 if (idx < l3->n_pkts) {
359 memcpy(&l3->optimized_arp_table[idx].mac, &(hdr->arp.data.sha), sizeof(prox_rte_ether_addr));
360 l3->optimized_arp_table[idx].arp_timeout = tsc + update_time;
361 update_arp_update_time(l3, &l3->optimized_arp_table[idx].arp_update_time, l3->arp_update_time);
364 int ret = rte_hash_add_key(l3->ip_hash, (const void *)&ip);
366 plogx_info("Unable add ip %d.%d.%d.%d in mac_hash\n", IP4(ip));
368 memcpy(&l3->arp_table[ret].mac, &(hdr->arp.data.sha), sizeof(prox_rte_ether_addr));
369 l3->arp_table[ret].arp_timeout = tsc + update_time;
370 update_arp_update_time(l3, &l3->arp_table[ret].arp_update_time, l3->arp_update_time);
375 case ARP_REPLY_FROM_CTRL:
377 case ARP_REQ_FROM_CTRL:
380 // tx_ctrlplane_pkt does not drop packets
381 plogx_dbg("\tForwarding (ARP/PING) packet from master\n");
382 tbase->aux->tx_ctrlplane_pkt(tbase, &mbufs[j], 1, out);
383 TASK_STATS_ADD_TX_NON_DP(&tbase->aux->stats, 1);