Merge pull request #709 from weiyanhua100/icmp-bug-0127

conf/dpvs.bond.conf.sample

@@ -250,6 +250,7 @@ worker_defs {
     <init> worker cpu8 {
         type    slave
         cpu_id  8
+        icmp_redirect_core
         port    bond0 {
             rx_queue_ids     7
             tx_queue_ids     7

conf/dpvs.conf.items

@@ -137,6 +137,7 @@ worker_defs {
     <init> worker cpu5 {
         type    kni
         cpu_id  5
+        icmp_redirect_core
         port dpdk0 {
             tx_queue_ids 6
         }

conf/dpvs.conf.sample

@@ -201,6 +201,7 @@ worker_defs {
     <init> worker   cpu8 {
         type        slave
         cpu_id      8
+        icmp_redirect_core
         port        dpdk0 {
             rx_queue_ids     7
             tx_queue_ids     7

conf/dpvs.conf.single-bond.sample

@@ -158,6 +158,7 @@ worker_defs {
     <init> worker cpu8 {
         type    slave
         cpu_id  8
+        icmp_redirect_core
         port    bond0 {
             rx_queue_ids     7
             tx_queue_ids     7

conf/dpvs.conf.single-nic.sample

@@ -130,6 +130,7 @@ worker_defs {
     <init> worker   cpu8 {
         type        slave
         cpu_id      8
+        icmp_redirect_core
         port        dpdk0 {
             rx_queue_ids     7
             tx_queue_ids     7

include/icmp.h

@@ -26,4 +26,9 @@ void icmp_send(struct rte_mbuf *imbuf, int type, int code, uint32_t info);
 
 #define icmp4_id(icmph)      (((icmph)->un).echo.id)
 
+#ifdef CONFIG_ICMP_REDIRECT_CORE
+int icmp_recv_proc(struct rte_mbuf *mbuf);
+void icmp_redirect_proc(void *args);
+extern lcoreid_t g_icmp_redirect_lcore_id;
+#endif
 #endif /* __DPVS_ICMP_H__ */

include/ipv4.h

@@ -119,6 +119,7 @@ int ip4_defrag(struct rte_mbuf *mbuf, int user);
 
 uint32_t ip4_select_id(struct ipv4_hdr *iph);
 int ipv4_local_out(struct rte_mbuf *mbuf);
+int ipv4_rcv_fin(struct rte_mbuf *mbuf);
 
 /* helper functions */
 static inline struct ipv4_hdr *ip4_hdr(const struct rte_mbuf *mbuf)

include/netif.h

@@ -316,7 +316,7 @@ int netif_ctrl_term(void); /* netif ctrl plane cleanup */
 void netif_cfgfile_init(void);
 void netif_keyword_value_init(void);
 void install_netif_keywords(void);
-
+void kni_ingress(struct rte_mbuf *mbuf, struct netif_port *dev);
 
 static inline void *netif_priv(struct netif_port *dev)
 {

src/config.mk

@@ -44,6 +44,7 @@ CFLAGS += -D DPVS_MAX_LCORE=64
 #CFLAGS += -D CONFIG_NDISC_DEBUG
 #CFLAGS += -D CONFIG_MSG_DEBUG
 #CFLAGS += -D CONFIG_DPVS_MP_DEBUG
+#CFLAGS += -D CONFIG_ICMP_REDIRECT_CORE
 
 ifeq ($(CONFIG_PDUMP), y)
 CFLAGS += -D CONFIG_DPVS_PDUMP

src/icmp.c

@@ -20,6 +20,11 @@
 #include "icmp.h"
 #include "netinet/in.h"
 #include "netinet/ip_icmp.h"
+#ifdef CONFIG_ICMP_REDIRECT_CORE
+#include "netif.h"
+#include "scheduler.h"
+#include "global_data.h"
+#endif
 
 #define ICMP
 #define RTE_LOGTYPE_ICMP    RTE_LOGTYPE_USER1
@@ -318,10 +323,113 @@ static struct inet_protocol icmp_protocol = {
     .handler    = icmp_rcv,
 };
 
+#ifdef CONFIG_ICMP_REDIRECT_CORE
+static struct rte_ring *icmp_redirect_ring;
+#define ICMP_RING_SIZE 2048
+lcoreid_t g_icmp_redirect_lcore_id = 0;
+
+static struct dpvs_lcore_job icmp_redirect = {
+    .name = "icmp_redirect_proc",
+    .type = LCORE_JOB_LOOP,
+    .func = icmp_redirect_proc,
+    .data = NULL,
+};
+
+static int icmp_redirect_init(void)
+{
+    int ret = 0;
+    int socket_id;
+
+    socket_id = rte_socket_id();
+    icmp_redirect_ring = rte_ring_create("icmp_redirect_ring", ICMP_RING_SIZE, socket_id, RING_F_SC_DEQ);
+    if (icmp_redirect_ring == NULL) {
+        rte_panic("create ring:icmp_redirect_ring  failed!\n");
+        return EDPVS_NOMEM;
+    }
+
+    ret = dpvs_lcore_job_register(&icmp_redirect, LCORE_ROLE_FWD_WORKER);
+    if (ret < 0) {
+        rte_ring_free(icmp_redirect_ring);
+        return ret;
+    }
+
+    return EDPVS_OK;
+}
+
+int icmp_recv_proc(struct rte_mbuf *mbuf)
+{
+    int ret = 0;
+    ret = rte_ring_enqueue(icmp_redirect_ring, mbuf);
+    if (unlikely(-EDQUOT == ret)) {
+        RTE_LOG(WARNING, ICMP, "%s: icmp ring quota exceeded\n", __func__);
+    }
+    else if (ret < 0) {
+        RTE_LOG(WARNING, ICMP, "%s: icmp ring enqueue failed\n", __func__);
+        rte_pktmbuf_free(mbuf);
+    }
+
+    return 0;
+}
+
+void icmp_redirect_proc(void *args)
+{
+    int ret = 0;
+    int i = 0;
+    lcoreid_t cid;
+    struct rte_mbuf *mbufs[NETIF_MAX_PKT_BURST];
+    uint16_t nb_rb = 0;
+    uint16_t data_off;
+
+    cid = rte_lcore_id();
+    if (cid != g_icmp_redirect_lcore_id)
+        return;
+
+    nb_rb = rte_ring_dequeue_burst(icmp_redirect_ring, (void**)mbufs, NETIF_MAX_PKT_BURST, NULL);
+    if (nb_rb <= 0) {
+        return;
+    }
+
+    for (i = 0; i < nb_rb; i++) {
+        struct rte_mbuf *mbuf = mbufs[i];
+        struct netif_port *dev = netif_port_get(mbuf->port);
+
+        /* Remove ether_hdr at the beginning of an mbuf */
+        data_off = mbuf->data_off;
+        if (unlikely(NULL == rte_pktmbuf_adj(mbuf, sizeof(struct ether_hdr)))) {
+            rte_pktmbuf_free(mbuf);
+            return;
+        }
+
+        ret = INET_HOOK(AF_INET, INET_HOOK_PRE_ROUTING,
+                     mbuf, dev, NULL, ipv4_rcv_fin);
+        if (ret == EDPVS_KNICONTINUE) {
+            if (dev->flag & NETIF_PORT_FLAG_FORWARD2KNI) {
+                rte_pktmbuf_free(mbuf);
+                return;
+            }
+            if (likely(NULL != rte_pktmbuf_prepend(mbuf,
+                (mbuf->data_off - data_off)))) {
+                    kni_ingress(mbuf, dev);
+            } else {
+                rte_pktmbuf_free(mbuf);
+            }
+        }
+    }
+
+    return;
+}
+#endif
+
 int icmp_init(void)
 {
     int err;
 
+#ifdef CONFIG_ICMP_REDIRECT_CORE
+    err = icmp_redirect_init();
+    if (err)
+        return err;
+#endif
+
     err = ipv4_register_protocol(&icmp_protocol, IPPROTO_ICMP);
 
     return err;

src/ipv4.c

@@ -326,7 +326,7 @@ static int ip4_rcv_options(struct rte_mbuf *mbuf)
     return EDPVS_OK;
 }
 
-static int ipv4_rcv_fin(struct rte_mbuf *mbuf)
+int ipv4_rcv_fin(struct rte_mbuf *mbuf)
 {
     int err;
     struct route_entry *rt = NULL;
@@ -378,6 +378,9 @@ static int ipv4_rcv_fin(struct rte_mbuf *mbuf)
 
 static int ipv4_rcv(struct rte_mbuf *mbuf, struct netif_port *port)
 {
+#ifdef CONFIG_ICMP_REDIRECT_CORE
+        struct icmphdr *ich, _icmph;
+#endif
     struct ipv4_hdr *iph;
     uint16_t hlen, len;
     eth_type_t etype = mbuf->packet_type; /* FIXME: use other field ? */
@@ -430,6 +433,18 @@ static int ipv4_rcv(struct rte_mbuf *mbuf, struct netif_port *port)
 
     if (unlikely(iph->next_proto_id == IPPROTO_OSPF))
         return EDPVS_KNICONTINUE;
+#ifdef CONFIG_ICMP_REDIRECT_CORE
+    else if (unlikely(iph->next_proto_id == IPPROTO_ICMP)) {
+        ich = mbuf_header_pointer(mbuf, hlen, sizeof(_icmph), &_icmph);
+        if (unlikely(!ich))
+            goto drop;
+        if (ich->type == ICMP_ECHOREPLY || ich->type == ICMP_ECHO) {
+            rte_pktmbuf_prepend(mbuf, (uint16_t)sizeof(struct ether_hdr));
+            icmp_recv_proc(mbuf);
+            return EDPVS_OK;
+        }
+    }
+#endif
 
     return INET_HOOK(AF_INET, INET_HOOK_PRE_ROUTING,
                      mbuf, port, NULL, ipv4_rcv_fin);

src/netif.c

@@ -43,6 +43,9 @@
 #include <netinet/in.h>
 #include <arpa/inet.h>
 #include <ipvs/redirect.h>
+#ifdef CONFIG_ICMP_REDIRECT_CORE
+#include "icmp.h"
+#endif
 
 #define NETIF_PKTPOOL_NB_MBUF_DEF   65535
 #define NETIF_PKTPOOL_NB_MBUF_MIN   1023
@@ -145,7 +148,6 @@ static struct list_head port_ntab[NETIF_PORT_TABLE_BUCKETS]; /* hashed by name *
 #define NETIF_CTRL_BUFFER_LEN     4096
 
 /* function declarations */
-static void kni_ingress(struct rte_mbuf *mbuf, struct netif_port *dev);
 static void kni_lcore_loop(void *dummy);
 
 
@@ -734,6 +736,18 @@ static void cpu_id_handler(vector_t tokens)
     FREE_PTR(str);
 }
 
+#ifdef CONFIG_ICMP_REDIRECT_CORE
+static void cpu_icmp_redirect_handler(vector_t tokens)
+{
+    struct worker_conf_stream *current_worker = list_entry(worker_list.next,
+            struct worker_conf_stream, worker_list_node);
+
+    RTE_LOG(INFO, NETIF, "%s(%d) used to redirect icmp packets\n",
+        current_worker->name, current_worker->cpu_id);
+    g_icmp_redirect_lcore_id = current_worker->cpu_id;
+}
+#endif
+
 static void worker_port_handler(vector_t tokens)
 {
     assert(VECTOR_SIZE(tokens) >= 1);
@@ -918,6 +932,9 @@ void install_netif_keywords(void)
     install_sublevel();
     install_keyword("type", worker_type_handler, KW_TYPE_INIT);
     install_keyword("cpu_id", cpu_id_handler, KW_TYPE_INIT);
+#ifdef CONFIG_ICMP_REDIRECT_CORE
+    install_keyword("icmp_redirect_core", cpu_icmp_redirect_handler, KW_TYPE_INIT);
+#endif
     install_keyword("port", worker_port_handler, KW_TYPE_INIT);
     install_sublevel();
     install_keyword("rx_queue_ids", rx_queue_ids_handler, KW_TYPE_INIT);
@@ -2691,7 +2708,7 @@ static inline void free_mbufs(struct rte_mbuf **pkts, unsigned num)
     }
 }
 
-static void kni_ingress(struct rte_mbuf *mbuf, struct netif_port *dev)
+void kni_ingress(struct rte_mbuf *mbuf, struct netif_port *dev)
 {
     if (!kni_dev_exist(dev))
         goto freepkt;