1
2
3
4#include <linux/bpf.h>
5#include <linux/if_link.h>
6#include <assert.h>
7#include <errno.h>
8#include <signal.h>
9#include <stdio.h>
10#include <stdlib.h>
11#include <string.h>
12#include <net/if.h>
13#include <sys/resource.h>
14#include <arpa/inet.h>
15#include <netinet/ether.h>
16#include <unistd.h>
17#include <time.h>
18#include "libbpf.h"
19#include <bpf/bpf.h>
20#include "bpf_util.h"
21#include "xdp_tx_iptunnel_common.h"
22
23#define STATS_INTERVAL_S 2U
24
25static int ifindex = -1;
26static __u32 xdp_flags = XDP_FLAGS_UPDATE_IF_NOEXIST;
27static int rxcnt_map_fd;
28static __u32 prog_id;
29
30static void int_exit(int sig)
31{
32 __u32 curr_prog_id = 0;
33
34 if (ifindex > -1) {
35 if (bpf_get_link_xdp_id(ifindex, &curr_prog_id, xdp_flags)) {
36 printf("bpf_get_link_xdp_id failed\n");
37 exit(1);
38 }
39 if (prog_id == curr_prog_id)
40 bpf_set_link_xdp_fd(ifindex, -1, xdp_flags);
41 else if (!curr_prog_id)
42 printf("couldn't find a prog id on a given iface\n");
43 else
44 printf("program on interface changed, not removing\n");
45 }
46 exit(0);
47}
48
49
50
51static void poll_stats(unsigned int kill_after_s)
52{
53 const unsigned int nr_protos = 256;
54 unsigned int nr_cpus = bpf_num_possible_cpus();
55 time_t started_at = time(NULL);
56 __u64 values[nr_cpus], prev[nr_protos][nr_cpus];
57 __u32 proto;
58 int i;
59
60 memset(prev, 0, sizeof(prev));
61
62 while (!kill_after_s || time(NULL) - started_at <= kill_after_s) {
63 sleep(STATS_INTERVAL_S);
64
65 for (proto = 0; proto < nr_protos; proto++) {
66 __u64 sum = 0;
67
68 assert(bpf_map_lookup_elem(rxcnt_map_fd, &proto,
69 values) == 0);
70 for (i = 0; i < nr_cpus; i++)
71 sum += (values[i] - prev[proto][i]);
72
73 if (sum)
74 printf("proto %u: sum:%10llu pkts, rate:%10llu pkts/s\n",
75 proto, sum, sum / STATS_INTERVAL_S);
76 memcpy(prev[proto], values, sizeof(values));
77 }
78 }
79}
80
81static void usage(const char *cmd)
82{
83 printf("Start a XDP prog which encapsulates incoming packets\n"
84 "in an IPv4/v6 header and XDP_TX it out. The dst <VIP:PORT>\n"
85 "is used to select packets to encapsulate\n\n");
86 printf("Usage: %s [...]\n", cmd);
87 printf(" -i <ifname|ifindex> Interface\n");
88 printf(" -a <vip-service-address> IPv4 or IPv6\n");
89 printf(" -p <vip-service-port> A port range (e.g. 433-444) is also allowed\n");
90 printf(" -s <source-ip> Used in the IPTunnel header\n");
91 printf(" -d <dest-ip> Used in the IPTunnel header\n");
92 printf(" -m <dest-MAC> Used in sending the IP Tunneled pkt\n");
93 printf(" -T <stop-after-X-seconds> Default: 0 (forever)\n");
94 printf(" -P <IP-Protocol> Default is TCP\n");
95 printf(" -S use skb-mode\n");
96 printf(" -N enforce native mode\n");
97 printf(" -F Force loading the XDP prog\n");
98 printf(" -h Display this help\n");
99}
100
101static int parse_ipstr(const char *ipstr, unsigned int *addr)
102{
103 if (inet_pton(AF_INET6, ipstr, addr) == 1) {
104 return AF_INET6;
105 } else if (inet_pton(AF_INET, ipstr, addr) == 1) {
106 addr[1] = addr[2] = addr[3] = 0;
107 return AF_INET;
108 }
109
110 fprintf(stderr, "%s is an invalid IP\n", ipstr);
111 return AF_UNSPEC;
112}
113
114static int parse_ports(const char *port_str, int *min_port, int *max_port)
115{
116 char *end;
117 long tmp_min_port;
118 long tmp_max_port;
119
120 tmp_min_port = strtol(optarg, &end, 10);
121 if (tmp_min_port < 1 || tmp_min_port > 65535) {
122 fprintf(stderr, "Invalid port(s):%s\n", optarg);
123 return 1;
124 }
125
126 if (*end == '-') {
127 end++;
128 tmp_max_port = strtol(end, NULL, 10);
129 if (tmp_max_port < 1 || tmp_max_port > 65535) {
130 fprintf(stderr, "Invalid port(s):%s\n", optarg);
131 return 1;
132 }
133 } else {
134 tmp_max_port = tmp_min_port;
135 }
136
137 if (tmp_min_port > tmp_max_port) {
138 fprintf(stderr, "Invalid port(s):%s\n", optarg);
139 return 1;
140 }
141
142 if (tmp_max_port - tmp_min_port + 1 > MAX_IPTNL_ENTRIES) {
143 fprintf(stderr, "Port range (%s) is larger than %u\n",
144 port_str, MAX_IPTNL_ENTRIES);
145 return 1;
146 }
147 *min_port = tmp_min_port;
148 *max_port = tmp_max_port;
149
150 return 0;
151}
152
153int main(int argc, char **argv)
154{
155 struct bpf_prog_load_attr prog_load_attr = {
156 .prog_type = BPF_PROG_TYPE_XDP,
157 };
158 struct rlimit r = {RLIM_INFINITY, RLIM_INFINITY};
159 int min_port = 0, max_port = 0, vip2tnl_map_fd;
160 const char *optstr = "i:a:p:s:d:m:T:P:FSNh";
161 unsigned char opt_flags[256] = {};
162 struct bpf_prog_info info = {};
163 __u32 info_len = sizeof(info);
164 unsigned int kill_after_s = 0;
165 struct iptnl_info tnl = {};
166 struct bpf_object *obj;
167 struct vip vip = {};
168 char filename[256];
169 int opt, prog_fd;
170 int i, err;
171
172 tnl.family = AF_UNSPEC;
173 vip.protocol = IPPROTO_TCP;
174
175 for (i = 0; i < strlen(optstr); i++)
176 if (optstr[i] != 'h' && 'a' <= optstr[i] && optstr[i] <= 'z')
177 opt_flags[(unsigned char)optstr[i]] = 1;
178
179 while ((opt = getopt(argc, argv, optstr)) != -1) {
180 unsigned short family;
181 unsigned int *v6;
182
183 switch (opt) {
184 case 'i':
185 ifindex = if_nametoindex(optarg);
186 if (!ifindex)
187 ifindex = atoi(optarg);
188 break;
189 case 'a':
190 vip.family = parse_ipstr(optarg, vip.daddr.v6);
191 if (vip.family == AF_UNSPEC)
192 return 1;
193 break;
194 case 'p':
195 if (parse_ports(optarg, &min_port, &max_port))
196 return 1;
197 break;
198 case 'P':
199 vip.protocol = atoi(optarg);
200 break;
201 case 's':
202 case 'd':
203 if (opt == 's')
204 v6 = tnl.saddr.v6;
205 else
206 v6 = tnl.daddr.v6;
207
208 family = parse_ipstr(optarg, v6);
209 if (family == AF_UNSPEC)
210 return 1;
211 if (tnl.family == AF_UNSPEC) {
212 tnl.family = family;
213 } else if (tnl.family != family) {
214 fprintf(stderr,
215 "The IP version of the src and dst addresses used in the IP encapsulation does not match\n");
216 return 1;
217 }
218 break;
219 case 'm':
220 if (!ether_aton_r(optarg,
221 (struct ether_addr *)tnl.dmac)) {
222 fprintf(stderr, "Invalid mac address:%s\n",
223 optarg);
224 return 1;
225 }
226 break;
227 case 'T':
228 kill_after_s = atoi(optarg);
229 break;
230 case 'S':
231 xdp_flags |= XDP_FLAGS_SKB_MODE;
232 break;
233 case 'N':
234 xdp_flags |= XDP_FLAGS_DRV_MODE;
235 break;
236 case 'F':
237 xdp_flags &= ~XDP_FLAGS_UPDATE_IF_NOEXIST;
238 break;
239 default:
240 usage(argv[0]);
241 return 1;
242 }
243 opt_flags[opt] = 0;
244 }
245
246 for (i = 0; i < strlen(optstr); i++) {
247 if (opt_flags[(unsigned int)optstr[i]]) {
248 fprintf(stderr, "Missing argument -%c\n", optstr[i]);
249 usage(argv[0]);
250 return 1;
251 }
252 }
253
254 if (setrlimit(RLIMIT_MEMLOCK, &r)) {
255 perror("setrlimit(RLIMIT_MEMLOCK, RLIM_INFINITY)");
256 return 1;
257 }
258
259 if (!ifindex) {
260 fprintf(stderr, "Invalid ifname\n");
261 return 1;
262 }
263
264 snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
265 prog_load_attr.file = filename;
266
267 if (bpf_prog_load_xattr(&prog_load_attr, &obj, &prog_fd))
268 return 1;
269
270 if (!prog_fd) {
271 printf("load_bpf_file: %s\n", strerror(errno));
272 return 1;
273 }
274
275 rxcnt_map_fd = bpf_object__find_map_fd_by_name(obj, "rxcnt");
276 vip2tnl_map_fd = bpf_object__find_map_fd_by_name(obj, "vip2tnl");
277 if (vip2tnl_map_fd < 0 || rxcnt_map_fd < 0) {
278 printf("bpf_object__find_map_fd_by_name failed\n");
279 return 1;
280 }
281
282 signal(SIGINT, int_exit);
283 signal(SIGTERM, int_exit);
284
285 while (min_port <= max_port) {
286 vip.dport = htons(min_port++);
287 if (bpf_map_update_elem(vip2tnl_map_fd, &vip, &tnl,
288 BPF_NOEXIST)) {
289 perror("bpf_map_update_elem(&vip2tnl)");
290 return 1;
291 }
292 }
293
294 if (bpf_set_link_xdp_fd(ifindex, prog_fd, xdp_flags) < 0) {
295 printf("link set xdp fd failed\n");
296 return 1;
297 }
298
299 err = bpf_obj_get_info_by_fd(prog_fd, &info, &info_len);
300 if (err) {
301 printf("can't get prog info - %s\n", strerror(errno));
302 return err;
303 }
304 prog_id = info.id;
305
306 poll_stats(kill_after_s);
307
308 bpf_set_link_xdp_fd(ifindex, -1, xdp_flags);
309
310 return 0;
311}
312
