Bug fix: Flow Directory. (test_flow_dir)

docs/NF_Dev.md

@@ -24,6 +24,8 @@ Each NF needs to have a packet handler function.  It must match this specificati
   - `ONVM_NF_ACTION_NEXT`: Forward the packet using the rule from the SDN controller stored in the flow table
   - `ONVM_NF_ACTION_TONF`: Forward the packet to the specified NF
   - `ONVM_NF_ACTION_OUT`: Forward the packet to the specified NIC port
+
+NFs that set destination ID to 255 will be forwarded with the `ONVM_NF_ACTION_NEXT` action.
 
 NF Library
 --

examples/test_flow_dir/README.md

@@ -3,6 +3,8 @@ Flow Director Test
 
 This NF demonstrates how to use ONVM's Flow Director. When a packet arrives the NF checks whether it is from a flow that already has a service chain rule. If not, it creates a new rule so the packet will be sent to the destination NF indicated on the command line. Packets that match a rule are processed with the ONVM_NF_ACTION_NEXT action.
 
+This NF demonstrates how to populate the Manager's flow table as well as how to create service chain rules for specific flows. To enable this functionality use the -s flag.
+
 Compilation and Execution
 --
 ```
@@ -24,6 +26,7 @@ App Specific Arguments
 --
   - `-d <dst>`: destination service ID to foward to
   - `-p <print_delay>`: number of packets between each print, e.g. `-p 1` prints every packets.
+  - `s`: Prepopulate sample flow table rules.
 
 Config File Support
 --

examples/test_flow_dir/test_flow_dir.c

@@ -66,11 +66,18 @@
 
 #define NF_TAG "test_flow_dir"
 
-/* number of package between each print */
+/* Number of packets between each print. */
 static uint32_t print_delay = 1000000;
 
 static uint32_t destination;
 
+/* Populate flow disabled by default */
+static int populate_flow = 0;
+
+static struct onvm_flow_entry *flow_entry = NULL;
+
+/* Pointer to the manager flow table */
+extern struct onvm_ft* sdn_ft;
 /*
  * Print a usage message
  */
@@ -82,6 +89,8 @@ usage(const char *progname) {
         printf("Flags:\n");
         printf(" - `-d <dst>`: destination service ID to foward to\n");
         printf(" - `-p <print_delay>`: number of packets between each print, e.g. `-p 1` prints every packets.\n");
+        printf(" -  -s : Prepopulate sample flow table rules. \n");
+
 }
 
 /*
@@ -91,14 +100,17 @@ static int
 parse_app_args(int argc, char *argv[], const char *progname) {
         int c;
 
-        while ((c = getopt(argc, argv, "d:p:")) != -1) {
+        while ((c = getopt(argc, argv, "d:p:s")) != -1) {
                 switch (c) {
                         case 'd':
                                 destination = strtoul(optarg, NULL, 10);
                                 break;
                         case 'p':
                                 print_delay = strtoul(optarg, NULL, 10);
                                 break;
+                        case 's':
+                                populate_flow = 1;
+                                break;
                         case '?':
                                 usage(progname);
                                 if (optopt == 'd')
@@ -151,32 +163,103 @@ do_stats_display(struct rte_mbuf *pkt) {
         }
 }
 
+/*
+ * This function populates a set of flows to the sdn flow table.
+ * For each new flow a service chain is created and appended.
+ * Each service chain is of max 4 length. This function is not enabled by default.
+ * Users may update the predefined struct with their own rules here.
+ */
+
+static void
+populate_sample_ipv4(void) {
+        struct onvm_flow_entry *flow_entry = (struct onvm_flow_entry *)
+                rte_calloc(NULL, 1, sizeof(struct onvm_flow_entry), 0);
+        if (flow_entry == NULL) {
+                rte_exit(EXIT_FAILURE, "Unable to populate flow.\n");
+        }
+        struct test_add_key {
+                struct onvm_ft_ipv4_5tuple key;
+                uint8_t destinations[ONVM_MAX_CHAIN_LENGTH];
+        };
+
+        struct test_add_key keys[] = {
+                {{IPv4(100, 10, 0, 0), IPv4(100, 10, 0, 0),  1, 1, IPPROTO_TCP}, {2,3}},
+                {{IPv4(102, 10, 0, 0), IPv4(101, 10, 0, 1),  0, 1, IPPROTO_TCP}, {4,3,2,1}},
+                {{IPv4(103, 10, 0, 0), IPv4(102, 10, 0, 1),  0, 1, IPPROTO_TCP}, {2,1}},
+                {{IPv4(10, 11, 1, 17), IPv4(10, 11, 1, 17),  1234, 1234, IPPROTO_UDP}, {4,3,2}},
+        };
+
+        uint32_t num_keys = RTE_DIM(keys);
+        for (uint32_t i = 0; i < num_keys; i++) {
+                struct onvm_ft_ipv4_5tuple key;
+                key = keys[i].key;
+                int ret = onvm_ft_lookup_key(sdn_ft, &key, (char **)&flow_entry);
+                if (ret >= 0) {
+                        continue;
+                } else {
+                        printf("\nAdding Key: ");
+                        _onvm_ft_print_key(&key);
+                        ret = onvm_ft_add_key(sdn_ft, &key, (char **)&flow_entry);
+                        if (ret < 0) {
+                                printf("Unable to add key.");
+                                continue;
+                        }
+                        printf("Creating new service chain.\n");
+                        flow_entry->sc = onvm_sc_create();
+                        uint32_t num_dest = RTE_DIM(keys[i].destinations);
+                        for (uint32_t j = 0; j < num_dest; j++) {
+                                if (keys[i].destinations[j] == 0) {
+                                        continue;
+                                }
+                                printf("Appending Destination: %d \n", keys[i].destinations[j]);
+                                onvm_sc_append_entry(flow_entry->sc, ONVM_NF_ACTION_TONF, keys[i].destinations[j]);
+                        }
+                }
+        }
+        flow_entry = NULL;
+        rte_free(flow_entry);
+}
+
+static void
+nf_setup(__attribute__((unused)) struct onvm_nf_local_ctx *nf_local_ctx) {
+        flow_entry = (struct onvm_flow_entry *) rte_calloc(NULL, 1, sizeof(struct onvm_flow_entry), 0);
+        if (flow_entry == NULL) {
+                rte_exit(EXIT_FAILURE, "Unable to allocate flow entry\n");
+        }
+}
+
 static int
 packet_handler(struct rte_mbuf *pkt, struct onvm_pkt_meta *meta,
                __attribute__((unused)) struct onvm_nf_local_ctx *nf_local_ctx) {
         static uint32_t counter = 0;
-        struct onvm_flow_entry *flow_entry = NULL;
         int ret;
 
+        if (!onvm_pkt_is_ipv4(pkt)) {
+                meta->action = ONVM_NF_ACTION_DROP;
+                return 0;
+        }
+
         if (++counter == print_delay) {
                 do_stats_display(pkt);
                 counter = 0;
         }
 
-        ret = onvm_flow_dir_get_pkt(pkt, &flow_entry);
+        struct onvm_ft_ipv4_5tuple key;
+        onvm_ft_fill_key(&key, pkt);
+        ret = onvm_ft_lookup_key(sdn_ft, &key, (char **)&flow_entry);
         if (ret >= 0) {
                 meta->action = ONVM_NF_ACTION_NEXT;
         } else {
-                ret = onvm_flow_dir_add_pkt(pkt, &flow_entry);
+                ret = onvm_ft_add_key(sdn_ft, &key, (char **)&flow_entry);
                 if (ret < 0) {
                         meta->action = ONVM_NF_ACTION_DROP;
                         meta->destination = 0;
                         return 0;
                 }
-                memset(flow_entry, 0, sizeof(struct onvm_flow_entry));
                 flow_entry->sc = onvm_sc_create();
                 onvm_sc_append_entry(flow_entry->sc, ONVM_NF_ACTION_TONF, destination);
-                // onvm_sc_print(flow_entry->sc);
+                meta->action = ONVM_NF_ACTION_TONF;
+                meta->destination = destination;
         }
         return 0;
 }
@@ -193,6 +276,7 @@ main(int argc, char *argv[]) {
 
         nf_function_table = onvm_nflib_init_nf_function_table();
         nf_function_table->pkt_handler = &packet_handler;
+        nf_function_table->setup = &nf_setup;
 
         if ((arg_offset = onvm_nflib_init(argc, argv, NF_TAG, nf_local_ctx, nf_function_table)) < 0) {
                 onvm_nflib_stop(nf_local_ctx);
@@ -214,9 +298,16 @@ main(int argc, char *argv[]) {
         /* Map the sdn_ft table */
         onvm_flow_dir_nf_init();
 
+        if (populate_flow) {
+                printf("Populating flow table. \n");
+                populate_sample_ipv4();
+        }
+
         onvm_nflib_run(nf_local_ctx);
 
         onvm_nflib_stop(nf_local_ctx);
+        flow_entry = NULL;
+        rte_free(flow_entry);
         printf("If we reach here, program is ending\n");
         return 0;
 }

onvm/onvm_mgr/onvm_init.c

@@ -251,7 +251,7 @@ init(int argc, char *argv[]) {
                 printf("Chain length can not be larger than the maximum chain length\n");
                 exit(1);
         }
-        printf("Default service chain: send to sdn NF\n");
+        printf("Default service chain: send to NF with service ID 1\n");
 
         /* set up service chain pointer shared to NFs*/
         mz_scp = rte_memzone_reserve(MZ_SCP_INFO, sizeof(struct onvm_service_chain *), rte_socket_id(), NO_FLAGS);

onvm/onvm_nflib/onvm_common.h

@@ -79,6 +79,8 @@
 #define ONVM_NF_ACTION_TONF 2  // send to the NF specified in the argument field (assume it is on the same host)
 #define ONVM_NF_ACTION_OUT  3  // send the packet out the NIC port set in the argument field
 
+#define ACTION_NEXT_DEST_ID 255 // Destination service ID that will set the manager to perform flow table lookup.
+
 #define PKT_WAKEUP_THRESHOLD 1 // for shared core mode, how many packets are required to wake up the NF
 #define MSG_WAKEUP_THRESHOLD 1 // for shared core mode, how many messages on an NF's ring are required to wake up the NF
 

onvm/onvm_nflib/onvm_flow_table.h

@@ -126,10 +126,10 @@ onvm_ft_free(struct onvm_ft *table);
 
 static inline void
 _onvm_ft_print_key(struct onvm_ft_ipv4_5tuple *key) {
-        printf("IP: %" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8, key->src_addr & 0xFF, (key->src_addr >> 8) & 0xFF,
-               (key->src_addr >> 16) & 0xFF, (key->src_addr >> 24) & 0xFF);
-        printf("-%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8 " ", key->dst_addr & 0xFF, (key->dst_addr >> 8) & 0xFF,
-               (key->dst_addr >> 16) & 0xFF, (key->dst_addr >> 24) & 0xFF);
+        printf("IP: %" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8, (key->src_addr >> 24) & 0xFF,
+                (key->src_addr >> 16) & 0xFF, (key->src_addr >> 8) & 0xFF, key->src_addr & 0xFF);
+        printf("-%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8 " ", (key->dst_addr >> 24) & 0xFF,
+                (key->dst_addr >> 16) & 0xFF, (key->dst_addr >> 8) & 0xFF, key->dst_addr & 0xFF);
         printf("Port: %d %d Proto: %d\n", key->src_port, key->dst_port, key->proto);
 }
 
@@ -150,16 +150,16 @@ onvm_ft_fill_key(struct onvm_ft_ipv4_5tuple *key, struct rte_mbuf *pkt) {
         ipv4_hdr = onvm_pkt_ipv4_hdr(pkt);
         memset(key, 0, sizeof(struct onvm_ft_ipv4_5tuple));
         key->proto = ipv4_hdr->next_proto_id;
-        key->src_addr = ipv4_hdr->src_addr;
-        key->dst_addr = ipv4_hdr->dst_addr;
+        key->src_addr = rte_be_to_cpu_32(ipv4_hdr->src_addr);
+        key->dst_addr = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
         if (key->proto == IP_PROTOCOL_TCP) {
                 tcp_hdr = onvm_pkt_tcp_hdr(pkt);
-                key->src_port = tcp_hdr->src_port;
-                key->dst_port = tcp_hdr->dst_port;
+                key->src_port = rte_be_to_cpu_16(tcp_hdr->src_port);
+                key->dst_port = rte_be_to_cpu_16(tcp_hdr->dst_port);
         } else if (key->proto == IP_PROTOCOL_UDP) {
                 udp_hdr = onvm_pkt_udp_hdr(pkt);
-                key->src_port = udp_hdr->src_port;
-                key->dst_port = udp_hdr->dst_port;
+                key->src_port = rte_be_to_cpu_16(udp_hdr->src_port);
+                key->dst_port = rte_be_to_cpu_16(udp_hdr->dst_port);
         } else {
                 key->src_port = 0;
                 key->dst_port = 0;
@@ -224,10 +224,10 @@ onvm_softrss(struct onvm_ft_ipv4_5tuple *key) {
 
         rte_convert_rss_key((uint32_t *)rss_symmetric_key, (uint32_t *)rss_key_be, RTE_DIM(rss_symmetric_key));
 
-        tuple.v4.src_addr = rte_be_to_cpu_32(key->src_addr);
-        tuple.v4.dst_addr = rte_be_to_cpu_32(key->dst_addr);
-        tuple.v4.sport = rte_be_to_cpu_16(key->src_port);
-        tuple.v4.dport = rte_be_to_cpu_16(key->dst_port);
+        tuple.v4.src_addr = key->src_addr;
+        tuple.v4.dst_addr = key->dst_addr;
+        tuple.v4.sport = key->src_port;
+        tuple.v4.dport = key->dst_port;
 
         rss_l3l4 = rte_softrss_be((uint32_t *)&tuple, RTE_THASH_V4_L4_LEN, rss_key_be);
 

onvm/onvm_nflib/onvm_nflib.c

@@ -555,6 +555,9 @@ onvm_nflib_start_nf(struct onvm_nf_local_ctx *nf_local_ctx, struct onvm_nf_init_
 
 int
 onvm_nflib_run(struct onvm_nf_local_ctx *nf_local_ctx) {
+        /* Map the sdn_ft table */
+        onvm_flow_dir_nf_init();
+
         int ret;
 
         pthread_t main_loop_thread;

onvm/onvm_nflib/onvm_pkt_common.c

@@ -102,9 +102,8 @@ onvm_pkt_process_tx_batch(struct queue_mgr *tx_mgr, struct rte_mbuf *pkts[], uin
                         // and !<return value> is 1.
                         nf->stats.act_drop++;
                         nf->stats.tx += !onvm_pkt_drop(pkts[i]);
-                } else if (meta->action == ONVM_NF_ACTION_NEXT) {
-                        /* TODO: Here we drop the packet : there will be a flow table
-                        in the future to know what to do with the packet next */
+                } else if (meta->action == ONVM_NF_ACTION_NEXT || meta->destination == ACTION_NEXT_DEST_ID) {
+                        // Perform next action is configured by the manager's flow table
                         nf->stats.act_next++;
                         onvm_pkt_process_next_action(tx_mgr, pkts[i], nf);
                 } else if (meta->action == ONVM_NF_ACTION_TONF) {