/*
-// Copyright (c) 2010-2017 Intel Corporation
+// Copyright (c) 2010-2020 Intel Corporation
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
#include "lconf.h"
#include "prefetch.h"
#include "log.h"
+#include "defines.h"
#include "handle_master.h"
#include "prox_port_cfg.h"
+#include "packet_utils.h"
static inline int find_ip(struct ether_hdr_arp *pkt, uint16_t len, uint32_t *ip_dst)
{
- struct vlan_hdr *vlan_hdr;
- struct ether_hdr *eth_hdr = (struct ether_hdr*)pkt;
- struct ipv4_hdr *ip;
+ prox_rte_vlan_hdr *vlan_hdr;
+ prox_rte_ether_hdr *eth_hdr = (prox_rte_ether_hdr*)pkt;
+ prox_rte_ipv4_hdr *ip;
uint16_t ether_type = eth_hdr->ether_type;
- uint16_t l2_len = sizeof(struct ether_hdr);
+ uint16_t l2_len = sizeof(prox_rte_ether_hdr);
// Unstack VLAN tags
- while (((ether_type == ETYPE_8021ad) || (ether_type == ETYPE_VLAN)) && (l2_len + sizeof(struct vlan_hdr) < len)) {
- vlan_hdr = (struct vlan_hdr *)((uint8_t *)pkt + l2_len);
+ while (((ether_type == ETYPE_8021ad) || (ether_type == ETYPE_VLAN)) && (l2_len + sizeof(prox_rte_vlan_hdr) < len)) {
+ vlan_hdr = (prox_rte_vlan_hdr *)((uint8_t *)pkt + l2_len);
l2_len +=4;
ether_type = vlan_hdr->eth_proto;
}
break;
}
- if (l2_len && (l2_len + sizeof(struct ipv4_hdr) <= len)) {
- struct ipv4_hdr *ip = (struct ipv4_hdr *)((uint8_t *)pkt + l2_len);
+ if (l2_len && (l2_len + sizeof(prox_rte_ipv4_hdr) <= len)) {
+ prox_rte_ipv4_hdr *ip = (prox_rte_ipv4_hdr *)((uint8_t *)pkt + l2_len);
// TODO: implement LPM => replace ip_dst by next hop IP DST
*ip_dst = ip->dst_addr;
return 0;
return -1;
}
-int write_dst_mac(struct task_base *tbase, struct rte_mbuf *mbuf, uint32_t *ip_dst)
+/* This implementation could be improved: instead of checking each time we send a packet whether we need also
+ to send an ARP, we should only check whether the MAC is valid.
+ We should check arp_update_time in the master process. This would also require the generating task to clear its arp ring
+ to avoid sending many ARP while starting after a long stop.
+ We could also check for arp_timeout in the master so that dataplane has only to check whether MAC is available
+ but this would require either thread safety, or the the exchange of information between master and generating core.
+*/
+
+int write_dst_mac(struct task_base *tbase, struct rte_mbuf *mbuf, uint32_t *ip_dst, uint64_t **time)
{
const uint64_t hz = rte_get_tsc_hz();
struct ether_hdr_arp *packet = rte_pktmbuf_mtod(mbuf, struct ether_hdr_arp *);
- struct ether_addr *mac = &packet->ether_hdr.d_addr;
+ prox_rte_ether_addr *mac = &packet->ether_hdr.d_addr;
uint64_t tsc = rte_rdtsc();
struct l3_base *l3 = &(tbase->l3);
if (l3->gw.ip) {
if (likely((l3->flags & FLAG_DST_MAC_KNOWN) && (tsc < l3->gw.arp_update_time) && (tsc < l3->gw.arp_timeout))) {
- memcpy(mac, &l3->gw.mac, sizeof(struct ether_addr));
- return 0;
+ memcpy(mac, &l3->gw.mac, sizeof(prox_rte_ether_addr));
+ return SEND_MBUF;
} else if (tsc > l3->gw.arp_update_time) {
- // long time since we have sent an arp, send arp
- l3->gw.arp_update_time = tsc + hz;
+ // long time since we have successfully sent an arp, send arp
+ // 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
+ *time = &l3->gw.arp_update_time;
*ip_dst = l3->gw.ip;
- return -1;
+ if ((l3->flags & FLAG_DST_MAC_KNOWN) && (tsc < l3->gw.arp_timeout)){
+ // MAC is valid in the table => send also the mbuf
+ memcpy(mac, &l3->gw.mac, sizeof(prox_rte_ether_addr));
+ return SEND_MBUF_AND_ARP;
+ } else {
+ // MAC still unknown, or timed out => only send ARP
+ return SEND_ARP;
+ }
+ } else {
+ // MAC is unknown and we already sent an ARP recently, drop mbuf and wait for ARP reply
+ return DROP_MBUF;
}
- return -2;
}
uint16_t len = rte_pktmbuf_pkt_len(mbuf);
if (find_ip(packet, len, ip_dst) != 0) {
- return 0;
+ // Unable to find IP address => non IP packet => send it as it
+ return SEND_MBUF;
}
if (likely(l3->n_pkts < 4)) {
for (unsigned int idx = 0; idx < l3->n_pkts; idx++) {
if (*ip_dst == l3->optimized_arp_table[idx].ip) {
+ // IP address already in table
if ((tsc < l3->optimized_arp_table[idx].arp_update_time) && (tsc < l3->optimized_arp_table[idx].arp_timeout)) {
- memcpy(mac, &l3->optimized_arp_table[idx].mac, sizeof(struct ether_addr));
- return 0;
+ // MAC address was recently updated in table, use it
+ memcpy(mac, &l3->optimized_arp_table[idx].mac, sizeof(prox_rte_ether_addr));
+ return SEND_MBUF;
} else if (tsc > l3->optimized_arp_table[idx].arp_update_time) {
- l3->optimized_arp_table[idx].arp_update_time = tsc + hz;
- return -1;
+ // ARP not sent since a long time, send ARP
+ *time = &l3->optimized_arp_table[idx].arp_update_time;
+ if (tsc < l3->optimized_arp_table[idx].arp_timeout) {
+ // MAC still valid => also send mbuf
+ memcpy(mac, &l3->optimized_arp_table[idx].mac, sizeof(prox_rte_ether_addr));
+ return SEND_MBUF_AND_ARP;
+ } else {
+ // MAC unvalid => only send ARP
+ return SEND_ARP;
+ }
} else {
- return -2;
+ // ARP timeout elapsed, MAC not valid anymore but waiting for ARP reply
+ return DROP_MBUF;
}
}
}
+ // IP address not found in table
l3->optimized_arp_table[l3->n_pkts].ip = *ip_dst;
- l3->optimized_arp_table[l3->n_pkts].arp_update_time = tsc + hz;
+ *time = &l3->optimized_arp_table[l3->n_pkts].arp_update_time;
l3->n_pkts++;
- if (l3->n_pkts < 4)
- return -1;
+ if (l3->n_pkts < 4) {
+ return SEND_ARP;
+ }
- // We have ** many ** IP addresses; lets use hash table instead
+ // We have too many IP addresses to search linearly; lets use hash table instead => copy all entries in hash table
for (uint32_t idx = 0; idx < l3->n_pkts; idx++) {
uint32_t ip = l3->optimized_arp_table[idx].ip;
int ret = rte_hash_add_key(l3->ip_hash, (const void *)&ip);
if (ret < 0) {
- plogx_info("Unable add ip %d.%d.%d.%d in mac_hash\n", IP4(ip));
+ // This should not happen as few entries so far.
+ // If it happens, we still send the ARP as easier:
+ // If the ARP corresponds to this error, the ARP reply will be ignored
+ // If ARP does not correspond to this error/ip, then ARP reply will be handled.
+ plogx_err("Unable add ip %d.%d.%d.%d in mac_hash (already %d entries)\n", IP4(ip), idx);
} else {
memcpy(&l3->arp_table[ret], &l3->optimized_arp_table[idx], sizeof(struct arp_table));
}
}
- return -1;
+ return SEND_ARP;
} else {
- // Find mac in lookup table. Send ARP if not found
+ // Find IP in lookup table. Send ARP if not found
int ret = rte_hash_lookup(l3->ip_hash, (const void *)ip_dst);
if (unlikely(ret < 0)) {
+ // IP not found, try to send an ARP
int ret = rte_hash_add_key(l3->ip_hash, (const void *)ip_dst);
if (ret < 0) {
- plogx_info("Unable add ip %d.%d.%d.%d in mac_hash\n", IP4(*ip_dst));
- return -2;
+ // No reason to send ARP, as reply would be anyhow ignored
+ plogx_err("Unable to add ip %d.%d.%d.%d in mac_hash\n", IP4(*ip_dst));
+ return DROP_MBUF;
} else {
l3->arp_table[ret].ip = *ip_dst;
- l3->arp_table[ret].arp_update_time = tsc + hz;
+ *time = &l3->arp_table[ret].arp_update_time;
}
- return -1;
+ return SEND_ARP;
} else {
- if ((tsc < l3->arp_table[ret].arp_update_time) && (tsc < l3->arp_table[ret].arp_timeout)) {
- memcpy(mac, &l3->arp_table[ret].mac, sizeof(struct ether_addr));
- return 0;
+ // IP has been found
+ if (likely((tsc < l3->arp_table[ret].arp_update_time) && (tsc < l3->arp_table[ret].arp_timeout))) {
+ // MAC still valid and ARP sent recently
+ memcpy(mac, &l3->arp_table[ret].mac, sizeof(prox_rte_ether_addr));
+ return SEND_MBUF;
} else if (tsc > l3->arp_table[ret].arp_update_time) {
- l3->arp_table[ret].arp_update_time = tsc + hz;
- return -1;
+ // ARP not sent since a long time, send ARP
+ *time = &l3->arp_table[ret].arp_update_time;
+ if (tsc < l3->arp_table[ret].arp_timeout) {
+ // MAC still valid => send also MBUF
+ memcpy(mac, &l3->arp_table[ret].mac, sizeof(prox_rte_ether_addr));
+ return SEND_MBUF_AND_ARP;
+ } else {
+ return SEND_ARP;
+ }
} else {
- return -2;
+ return DROP_MBUF;
}
}
}
- return 0;
+ // Should not happen
+ return DROP_MBUF;
}
void task_init_l3(struct task_base *tbase, struct task_args *targ)
tbase->l3.core_id = targ->lconf->id;
tbase->l3.task_id = targ->id;
tbase->l3.tmaster = targ->tmaster;
+ tbase->l3.seed = (uint)rte_rdtsc();
+ if (targ->arp_timeout != 0)
+ tbase->l3.arp_timeout = targ->arp_timeout;
+ else
+ tbase->l3.arp_timeout = DEFAULT_ARP_TIMEOUT;
+ if (targ->arp_update_time != 0)
+ tbase->l3.arp_update_time = targ->arp_update_time;
+ else
+ tbase->l3.arp_update_time = DEFAULT_ARP_UPDATE_TIME;
}
void task_start_l3(struct task_base *tbase, struct task_args *targ)
{
+ const int NB_ARP_MBUF = 1024;
+ const int ARP_MBUF_SIZE = 2048;
+ const int NB_CACHE_ARP_MBUF = 256;
+
struct prox_port_cfg *port = find_reachable_port(targ);
- if (port) {
+ if (port && (tbase->l3.arp_pool == NULL)) {
+ static char name[] = "arp0_pool";
tbase->l3.reachable_port_id = port - prox_port_cfg;
if (targ->local_ipv4) {
tbase->local_ipv4 = rte_be_to_cpu_32(targ->local_ipv4);
register_ip_to_ctrl_plane(tbase->l3.tmaster, tbase->local_ipv4, tbase->l3.reachable_port_id, targ->lconf->id, targ->id);
}
+ master_init_vdev(tbase->l3.tmaster, tbase->l3.reachable_port_id, targ->lconf->id, targ->id);
+ name[3]++;
+ struct rte_mempool *ret = rte_mempool_create(name, NB_ARP_MBUF, ARP_MBUF_SIZE, NB_CACHE_ARP_MBUF,
+ sizeof(struct rte_pktmbuf_pool_private), rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, 0,
+ rte_socket_id(), 0);
+ PROX_PANIC(ret == NULL, "Failed to allocate ARP memory pool on socket %u with %u elements\n",
+ rte_socket_id(), NB_ARP_MBUF);
+ plog_info("\t\tMempool %p (%s) size = %u * %u cache %u, socket %d\n", ret, name, NB_ARP_MBUF,
+ ARP_MBUF_SIZE, NB_CACHE_ARP_MBUF, rte_socket_id());
+ tbase->l3.arp_pool = ret;
}
}
tbase->local_ipv4 = ip;
}
+static void reset_arp_update_time(struct l3_base *l3, uint32_t ip)
+{
+ uint32_t idx;
+ plogx_info("\tMAC entry for IP "IPv4_BYTES_FMT" timeout in kernel\n", IP4(ip));
+ if (ip == l3->gw.ip) {
+ l3->gw.arp_update_time = 0;
+ } else if (l3->n_pkts < 4) {
+ for (idx = 0; idx < l3->n_pkts; idx++) {
+ uint32_t ip_dst = l3->optimized_arp_table[idx].ip;
+ if (ip_dst == ip)
+ break;
+ }
+ if (idx < l3->n_pkts) {
+ l3->optimized_arp_table[idx].arp_update_time = 0;
+ }
+ } else {
+ int ret = rte_hash_lookup(l3->ip_hash, (const void *)&ip);
+ if (ret >= 0)
+ l3->arp_table[ret].arp_update_time = 0;
+ }
+ return;
+}
+
void handle_ctrl_plane_pkts(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
{
uint8_t out[1];
struct ether_hdr_arp *hdr;
struct l3_base *l3 = &tbase->l3;
uint64_t tsc= rte_rdtsc();
+ uint64_t update_time = l3->arp_timeout * hz / 1000;
for (j = 0; j < n_pkts; ++j) {
PREFETCH0(mbufs[j]);
hdr = rte_pktmbuf_mtod(mbufs[j], struct ether_hdr_arp *);
ip = (mbufs[j]->udata64 >> 32) & 0xFFFFFFFF;
+ if (prox_rte_is_zero_ether_addr(&hdr->arp.data.sha)) {
+ // MAC timeout or deleted from kernel table => reset update_time
+ // This will cause us to send new ARP request
+ // However, as arp_timeout not touched, we should continue sending our regular IP packets
+ reset_arp_update_time(l3, ip);
+ plogx_info("\tTimeout for MAC entry for IP "IPv4_BYTES_FMT"\n", IP4(ip));
+ return;
+ } else
+ plogx_dbg("\tUpdating MAC entry for IP "IPv4_BYTES_FMT" with MAC "MAC_BYTES_FMT"\n",
+ IP4(ip), MAC_BYTES(hdr->arp.data.sha.addr_bytes));
if (ip == l3->gw.ip) {
// MAC address of the gateway
memcpy(&l3->gw.mac, &hdr->arp.data.sha, 6);
l3->flags |= FLAG_DST_MAC_KNOWN;
- l3->gw.arp_timeout = tsc + 30 * hz;
+ l3->gw.arp_timeout = tsc + update_time;
+ update_arp_update_time(l3, &l3->gw.arp_update_time, l3->arp_update_time);
} else if (l3->n_pkts < 4) {
// Few packets tracked - should be faster to loop through them thean using a hash table
for (idx = 0; idx < l3->n_pkts; idx++) {
break;
}
if (idx < l3->n_pkts) {
- // IP not found; this is a reply while we never asked for the request!
- memcpy(&l3->optimized_arp_table[idx].mac, &(hdr->arp.data.sha), sizeof(struct ether_addr));
- l3->optimized_arp_table[idx].arp_timeout = tsc + 30 * hz;
+ memcpy(&l3->optimized_arp_table[idx].mac, &(hdr->arp.data.sha), sizeof(prox_rte_ether_addr));
+ l3->optimized_arp_table[idx].arp_timeout = tsc + update_time;
+ update_arp_update_time(l3, &l3->optimized_arp_table[idx].arp_update_time, l3->arp_update_time);
}
} else {
int ret = rte_hash_add_key(l3->ip_hash, (const void *)&ip);
if (ret < 0) {
plogx_info("Unable add ip %d.%d.%d.%d in mac_hash\n", IP4(ip));
} else {
- memcpy(&l3->arp_table[ret].mac, &(hdr->arp.data.sha), sizeof(struct ether_addr));
- l3->arp_table[ret].arp_timeout = tsc + 30 * hz;
+ memcpy(&l3->arp_table[ret].mac, &(hdr->arp.data.sha), sizeof(prox_rte_ether_addr));
+ l3->arp_table[ret].arp_timeout = tsc + update_time;
+ update_arp_update_time(l3, &l3->arp_table[ret].arp_update_time, l3->arp_update_time);
}
}
tx_drop(mbufs[j]);
break;
case ARP_REPLY_FROM_CTRL:
+ case ICMP_FROM_CTRL:
case ARP_REQ_FROM_CTRL:
- TASK_STATS_ADD_TX_NON_DP(&tbase->aux->stats, 1);
+ case PKT_FROM_TAP:
out[0] = 0;
- tbase->aux->tx_pkt_l2(tbase, &mbufs[j], 1, out);
+ // tx_ctrlplane_pkt does not drop packets
+ plogx_dbg("\tForwarding (ARP/PING) packet from master\n");
+ tbase->aux->tx_ctrlplane_pkt(tbase, &mbufs[j], 1, out);
+ TASK_STATS_ADD_TX_NON_DP(&tbase->aux->stats, 1);
break;
}
}