xref: /dpdk/lib/node/ip4_lookup.c (revision be4c0cb4901fc0703786e0d3da4e0123306e4539)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(C) 2020 Marvell International Ltd.
3  */
4 
5 #include <arpa/inet.h>
6 #include <sys/socket.h>
7 
8 #include <rte_ethdev.h>
9 #include <rte_ether.h>
10 #include <rte_graph.h>
11 #include <rte_graph_worker.h>
12 #include <rte_ip.h>
13 #include <rte_lpm.h>
14 
15 #include "rte_node_ip4_api.h"
16 
17 #include "node_private.h"
18 
19 #define IPV4_L3FWD_LPM_MAX_RULES 1024
20 #define IPV4_L3FWD_LPM_NUMBER_TBL8S (1 << 8)
21 
22 /* IP4 Lookup global data struct */
23 struct ip4_lookup_node_main {
24 	struct rte_lpm *lpm_tbl[RTE_MAX_NUMA_NODES];
25 };
26 
27 struct ip4_lookup_node_ctx {
28 	/* Socket's LPM table */
29 	struct rte_lpm *lpm;
30 	/* Dynamic offset to mbuf priv1 */
31 	int mbuf_priv1_off;
32 };
33 
34 int node_mbuf_priv1_dynfield_offset = -1;
35 
36 static struct ip4_lookup_node_main ip4_lookup_nm;
37 
38 #define IP4_LOOKUP_NODE_LPM(ctx) \
39 	(((struct ip4_lookup_node_ctx *)ctx)->lpm)
40 
41 #define IP4_LOOKUP_NODE_PRIV1_OFF(ctx) \
42 	(((struct ip4_lookup_node_ctx *)ctx)->mbuf_priv1_off)
43 
44 #if defined(__ARM_NEON)
45 #include "ip4_lookup_neon.h"
46 #elif defined(RTE_ARCH_X86)
47 #include "ip4_lookup_sse.h"
48 #endif
49 
50 static uint16_t
51 ip4_lookup_node_process_scalar(struct rte_graph *graph, struct rte_node *node,
52 			void **objs, uint16_t nb_objs)
53 {
54 	struct rte_lpm *lpm = IP4_LOOKUP_NODE_LPM(node->ctx);
55 	const int dyn = IP4_LOOKUP_NODE_PRIV1_OFF(node->ctx);
56 	struct rte_ipv4_hdr *ipv4_hdr;
57 	void **to_next, **from;
58 	uint16_t last_spec = 0;
59 	struct rte_mbuf *mbuf;
60 	rte_edge_t next_index;
61 	uint16_t held = 0;
62 	uint32_t drop_nh;
63 	int i, rc;
64 
65 	/* Speculative next */
66 	next_index = RTE_NODE_IP4_LOOKUP_NEXT_REWRITE;
67 	/* Drop node */
68 	drop_nh = ((uint32_t)RTE_NODE_IP4_LOOKUP_NEXT_PKT_DROP) << 16;
69 	from = objs;
70 
71 	/* Get stream for the speculated next node */
72 	to_next = rte_node_next_stream_get(graph, node, next_index, nb_objs);
73 	for (i = 0; i < nb_objs; i++) {
74 		uint32_t next_hop;
75 		uint16_t next;
76 
77 		mbuf = (struct rte_mbuf *)objs[i];
78 
79 		/* Extract DIP of mbuf0 */
80 		ipv4_hdr = rte_pktmbuf_mtod_offset(mbuf, struct rte_ipv4_hdr *,
81 				sizeof(struct rte_ether_hdr));
82 		/* Extract cksum, ttl as ipv4 hdr is in cache */
83 		node_mbuf_priv1(mbuf, dyn)->cksum = ipv4_hdr->hdr_checksum;
84 		node_mbuf_priv1(mbuf, dyn)->ttl = ipv4_hdr->time_to_live;
85 
86 		rc = rte_lpm_lookup(lpm, rte_be_to_cpu_32(ipv4_hdr->dst_addr),
87 				    &next_hop);
88 		next_hop = (rc == 0) ? next_hop : drop_nh;
89 		NODE_INCREMENT_XSTAT_ID(node, 0, rc != 0, 1);
90 
91 		node_mbuf_priv1(mbuf, dyn)->nh = (uint16_t)next_hop;
92 		next_hop = next_hop >> 16;
93 		next = (uint16_t)next_hop;
94 
95 		if (unlikely(next_index != next)) {
96 			/* Copy things successfully speculated till now */
97 			rte_memcpy(to_next, from, last_spec * sizeof(from[0]));
98 			from += last_spec;
99 			to_next += last_spec;
100 			held += last_spec;
101 			last_spec = 0;
102 
103 			rte_node_enqueue_x1(graph, node, next, from[0]);
104 			from += 1;
105 		} else {
106 			last_spec += 1;
107 		}
108 	}
109 
110 	/* !!! Home run !!! */
111 	if (likely(last_spec == nb_objs)) {
112 		rte_node_next_stream_move(graph, node, next_index);
113 		return nb_objs;
114 	}
115 	held += last_spec;
116 	rte_memcpy(to_next, from, last_spec * sizeof(from[0]));
117 	rte_node_next_stream_put(graph, node, next_index, held);
118 
119 	return nb_objs;
120 }
121 
122 int
123 rte_node_ip4_route_add(uint32_t ip, uint8_t depth, uint16_t next_hop,
124 		       enum rte_node_ip4_lookup_next next_node)
125 {
126 	char abuf[INET6_ADDRSTRLEN];
127 	struct in_addr in;
128 	uint8_t socket;
129 	uint32_t val;
130 	int ret;
131 
132 	in.s_addr = htonl(ip);
133 	inet_ntop(AF_INET, &in, abuf, sizeof(abuf));
134 	/* Embedded next node id into 24 bit next hop */
135 	val = ((next_node << 16) | next_hop) & ((1ull << 24) - 1);
136 	node_dbg("ip4_lookup", "LPM: Adding route %s / %d nh (0x%x)", abuf,
137 		 depth, val);
138 
139 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
140 		if (!ip4_lookup_nm.lpm_tbl[socket])
141 			continue;
142 
143 		ret = rte_lpm_add(ip4_lookup_nm.lpm_tbl[socket],
144 				  ip, depth, val);
145 		if (ret < 0) {
146 			node_err("ip4_lookup",
147 				 "Unable to add entry %s / %d nh (%x) to LPM table on sock %d, rc=%d",
148 				 abuf, depth, val, socket, ret);
149 			return ret;
150 		}
151 	}
152 
153 	return 0;
154 }
155 
156 static int
157 setup_lpm(struct ip4_lookup_node_main *nm, int socket)
158 {
159 	struct rte_lpm_config config_ipv4;
160 	char s[RTE_LPM_NAMESIZE];
161 
162 	/* One LPM table per socket */
163 	if (nm->lpm_tbl[socket])
164 		return 0;
165 
166 	/* create the LPM table */
167 	config_ipv4.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
168 	config_ipv4.number_tbl8s = IPV4_L3FWD_LPM_NUMBER_TBL8S;
169 	config_ipv4.flags = 0;
170 	snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socket);
171 	nm->lpm_tbl[socket] = rte_lpm_create(s, socket, &config_ipv4);
172 	if (nm->lpm_tbl[socket] == NULL)
173 		return -rte_errno;
174 
175 	return 0;
176 }
177 
178 static int
179 ip4_lookup_node_init(const struct rte_graph *graph, struct rte_node *node)
180 {
181 	uint16_t socket, lcore_id;
182 	static uint8_t init_once;
183 	int rc;
184 
185 	RTE_SET_USED(graph);
186 	RTE_BUILD_BUG_ON(sizeof(struct ip4_lookup_node_ctx) > RTE_NODE_CTX_SZ);
187 
188 	if (!init_once) {
189 		node_mbuf_priv1_dynfield_offset = rte_mbuf_dynfield_register(
190 				&node_mbuf_priv1_dynfield_desc);
191 		if (node_mbuf_priv1_dynfield_offset < 0)
192 			return -rte_errno;
193 
194 		/* Setup LPM tables for all sockets */
195 		RTE_LCORE_FOREACH(lcore_id)
196 		{
197 			socket = rte_lcore_to_socket_id(lcore_id);
198 			rc = setup_lpm(&ip4_lookup_nm, socket);
199 			if (rc) {
200 				node_err("ip4_lookup",
201 					 "Failed to setup lpm tbl for sock %u, rc=%d",
202 					 socket, rc);
203 				return rc;
204 			}
205 		}
206 		init_once = 1;
207 	}
208 
209 	/* Update socket's LPM and mbuf dyn priv1 offset in node ctx */
210 	IP4_LOOKUP_NODE_LPM(node->ctx) = ip4_lookup_nm.lpm_tbl[graph->socket];
211 	IP4_LOOKUP_NODE_PRIV1_OFF(node->ctx) = node_mbuf_priv1_dynfield_offset;
212 
213 #if defined(__ARM_NEON) || defined(RTE_ARCH_X86)
214 	if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128)
215 		node->process = ip4_lookup_node_process_vec;
216 #endif
217 
218 	node_dbg("ip4_lookup", "Initialized ip4_lookup node");
219 
220 	return 0;
221 }
222 
223 static struct rte_node_xstats ip4_lookup_xstats = {
224 	.nb_xstats = 1,
225 	.xstat_desc = {
226 		[0] = "ip4_lookup_error",
227 	},
228 };
229 
230 static struct rte_node_register ip4_lookup_node = {
231 	.process = ip4_lookup_node_process_scalar,
232 	.name = "ip4_lookup",
233 
234 	.init = ip4_lookup_node_init,
235 	.xstats = &ip4_lookup_xstats,
236 
237 	.nb_edges = RTE_NODE_IP4_LOOKUP_NEXT_PKT_DROP + 1,
238 	.next_nodes = {
239 		[RTE_NODE_IP4_LOOKUP_NEXT_IP4_LOCAL] = "ip4_local",
240 		[RTE_NODE_IP4_LOOKUP_NEXT_REWRITE] = "ip4_rewrite",
241 		[RTE_NODE_IP4_LOOKUP_NEXT_PKT_DROP] = "pkt_drop",
242 	},
243 };
244 
245 RTE_NODE_REGISTER(ip4_lookup_node);
246