kvm: x86: optimize dr6 restore
[linux/fpc-iii.git] / samples / bpf / xdp_router_ipv4_user.c
blobb2b4dfa776c8c1a40fa04bac053dc36a49c9d550
1 /* Copyright (C) 2017 Cavium, Inc.
3 * This program is free software; you can redistribute it and/or modify it
4 * under the terms of version 2 of the GNU General Public License
5 * as published by the Free Software Foundation.
6 */
7 #include <linux/bpf.h>
8 #include <linux/netlink.h>
9 #include <linux/rtnetlink.h>
10 #include <assert.h>
11 #include <errno.h>
12 #include <signal.h>
13 #include <stdio.h>
14 #include <stdlib.h>
15 #include <string.h>
16 #include <sys/socket.h>
17 #include <unistd.h>
18 #include "bpf_load.h"
19 #include <bpf/bpf.h>
20 #include <arpa/inet.h>
21 #include <fcntl.h>
22 #include <poll.h>
23 #include <net/if.h>
24 #include <netdb.h>
25 #include <sys/ioctl.h>
26 #include <sys/syscall.h>
27 #include "bpf_util.h"
29 int sock, sock_arp, flags = 0;
30 static int total_ifindex;
31 int *ifindex_list;
32 char buf[8192];
34 static int get_route_table(int rtm_family);
35 static void int_exit(int sig)
37 int i = 0;
39 for (i = 0; i < total_ifindex; i++)
40 bpf_set_link_xdp_fd(ifindex_list[i], -1, flags);
41 exit(0);
44 static void close_and_exit(int sig)
46 int i = 0;
48 close(sock);
49 close(sock_arp);
51 for (i = 0; i < total_ifindex; i++)
52 bpf_set_link_xdp_fd(ifindex_list[i], -1, flags);
53 exit(0);
56 /* Get the mac address of the interface given interface name */
57 static __be64 getmac(char *iface)
59 struct ifreq ifr;
60 __be64 mac = 0;
61 int fd, i;
63 fd = socket(AF_INET, SOCK_DGRAM, 0);
64 ifr.ifr_addr.sa_family = AF_INET;
65 strncpy(ifr.ifr_name, iface, IFNAMSIZ - 1);
66 if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0) {
67 printf("ioctl failed leaving....\n");
68 return -1;
70 for (i = 0; i < 6 ; i++)
71 *((__u8 *)&mac + i) = (__u8)ifr.ifr_hwaddr.sa_data[i];
72 close(fd);
73 return mac;
76 static int recv_msg(struct sockaddr_nl sock_addr, int sock)
78 struct nlmsghdr *nh;
79 int len, nll = 0;
80 char *buf_ptr;
82 buf_ptr = buf;
83 while (1) {
84 len = recv(sock, buf_ptr, sizeof(buf) - nll, 0);
85 if (len < 0)
86 return len;
88 nh = (struct nlmsghdr *)buf_ptr;
90 if (nh->nlmsg_type == NLMSG_DONE)
91 break;
92 buf_ptr += len;
93 nll += len;
94 if ((sock_addr.nl_groups & RTMGRP_NEIGH) == RTMGRP_NEIGH)
95 break;
97 if ((sock_addr.nl_groups & RTMGRP_IPV4_ROUTE) == RTMGRP_IPV4_ROUTE)
98 break;
100 return nll;
103 /* Function to parse the route entry returned by netlink
104 * Updates the route entry related map entries
106 static void read_route(struct nlmsghdr *nh, int nll)
108 char dsts[24], gws[24], ifs[16], dsts_len[24], metrics[24];
109 struct bpf_lpm_trie_key *prefix_key;
110 struct rtattr *rt_attr;
111 struct rtmsg *rt_msg;
112 int rtm_family;
113 int rtl;
114 int i;
115 struct route_table {
116 int dst_len, iface, metric;
117 char *iface_name;
118 __be32 dst, gw;
119 __be64 mac;
120 } route;
121 struct arp_table {
122 __be64 mac;
123 __be32 dst;
126 struct direct_map {
127 struct arp_table arp;
128 int ifindex;
129 __be64 mac;
130 } direct_entry;
132 if (nh->nlmsg_type == RTM_DELROUTE)
133 printf("DELETING Route entry\n");
134 else if (nh->nlmsg_type == RTM_GETROUTE)
135 printf("READING Route entry\n");
136 else if (nh->nlmsg_type == RTM_NEWROUTE)
137 printf("NEW Route entry\n");
138 else
139 printf("%d\n", nh->nlmsg_type);
141 memset(&route, 0, sizeof(route));
142 printf("Destination\t\tGateway\t\tGenmask\t\tMetric\t\tIface\n");
143 for (; NLMSG_OK(nh, nll); nh = NLMSG_NEXT(nh, nll)) {
144 rt_msg = (struct rtmsg *)NLMSG_DATA(nh);
145 rtm_family = rt_msg->rtm_family;
146 if (rtm_family == AF_INET)
147 if (rt_msg->rtm_table != RT_TABLE_MAIN)
148 continue;
149 rt_attr = (struct rtattr *)RTM_RTA(rt_msg);
150 rtl = RTM_PAYLOAD(nh);
152 for (; RTA_OK(rt_attr, rtl); rt_attr = RTA_NEXT(rt_attr, rtl)) {
153 switch (rt_attr->rta_type) {
154 case NDA_DST:
155 sprintf(dsts, "%u",
156 (*((__be32 *)RTA_DATA(rt_attr))));
157 break;
158 case RTA_GATEWAY:
159 sprintf(gws, "%u",
160 *((__be32 *)RTA_DATA(rt_attr)));
161 break;
162 case RTA_OIF:
163 sprintf(ifs, "%u",
164 *((int *)RTA_DATA(rt_attr)));
165 break;
166 case RTA_METRICS:
167 sprintf(metrics, "%u",
168 *((int *)RTA_DATA(rt_attr)));
169 default:
170 break;
173 sprintf(dsts_len, "%d", rt_msg->rtm_dst_len);
174 route.dst = atoi(dsts);
175 route.dst_len = atoi(dsts_len);
176 route.gw = atoi(gws);
177 route.iface = atoi(ifs);
178 route.metric = atoi(metrics);
179 route.iface_name = alloca(sizeof(char *) * IFNAMSIZ);
180 route.iface_name = if_indextoname(route.iface, route.iface_name);
181 route.mac = getmac(route.iface_name);
182 if (route.mac == -1) {
183 int i = 0;
185 for (i = 0; i < total_ifindex; i++)
186 bpf_set_link_xdp_fd(ifindex_list[i], -1, flags);
187 exit(0);
189 assert(bpf_map_update_elem(map_fd[4], &route.iface, &route.iface, 0) == 0);
190 if (rtm_family == AF_INET) {
191 struct trie_value {
192 __u8 prefix[4];
193 __be64 value;
194 int ifindex;
195 int metric;
196 __be32 gw;
197 } *prefix_value;
199 prefix_key = alloca(sizeof(*prefix_key) + 3);
200 prefix_value = alloca(sizeof(*prefix_value));
202 prefix_key->prefixlen = 32;
203 prefix_key->prefixlen = route.dst_len;
204 direct_entry.mac = route.mac & 0xffffffffffff;
205 direct_entry.ifindex = route.iface;
206 direct_entry.arp.mac = 0;
207 direct_entry.arp.dst = 0;
208 if (route.dst_len == 32) {
209 if (nh->nlmsg_type == RTM_DELROUTE) {
210 assert(bpf_map_delete_elem(map_fd[3], &route.dst) == 0);
211 } else {
212 if (bpf_map_lookup_elem(map_fd[2], &route.dst, &direct_entry.arp.mac) == 0)
213 direct_entry.arp.dst = route.dst;
214 assert(bpf_map_update_elem(map_fd[3], &route.dst, &direct_entry, 0) == 0);
217 for (i = 0; i < 4; i++)
218 prefix_key->data[i] = (route.dst >> i * 8) & 0xff;
220 printf("%3d.%d.%d.%d\t\t%3x\t\t%d\t\t%d\t\t%s\n",
221 (int)prefix_key->data[0],
222 (int)prefix_key->data[1],
223 (int)prefix_key->data[2],
224 (int)prefix_key->data[3],
225 route.gw, route.dst_len,
226 route.metric,
227 route.iface_name);
228 if (bpf_map_lookup_elem(map_fd[0], prefix_key,
229 prefix_value) < 0) {
230 for (i = 0; i < 4; i++)
231 prefix_value->prefix[i] = prefix_key->data[i];
232 prefix_value->value = route.mac & 0xffffffffffff;
233 prefix_value->ifindex = route.iface;
234 prefix_value->gw = route.gw;
235 prefix_value->metric = route.metric;
237 assert(bpf_map_update_elem(map_fd[0],
238 prefix_key,
239 prefix_value, 0
240 ) == 0);
241 } else {
242 if (nh->nlmsg_type == RTM_DELROUTE) {
243 printf("deleting entry\n");
244 printf("prefix key=%d.%d.%d.%d/%d",
245 prefix_key->data[0],
246 prefix_key->data[1],
247 prefix_key->data[2],
248 prefix_key->data[3],
249 prefix_key->prefixlen);
250 assert(bpf_map_delete_elem(map_fd[0],
251 prefix_key
252 ) == 0);
253 /* Rereading the route table to check if
254 * there is an entry with the same
255 * prefix but a different metric as the
256 * deleted enty.
258 get_route_table(AF_INET);
259 } else if (prefix_key->data[0] ==
260 prefix_value->prefix[0] &&
261 prefix_key->data[1] ==
262 prefix_value->prefix[1] &&
263 prefix_key->data[2] ==
264 prefix_value->prefix[2] &&
265 prefix_key->data[3] ==
266 prefix_value->prefix[3] &&
267 route.metric >= prefix_value->metric) {
268 continue;
269 } else {
270 for (i = 0; i < 4; i++)
271 prefix_value->prefix[i] =
272 prefix_key->data[i];
273 prefix_value->value =
274 route.mac & 0xffffffffffff;
275 prefix_value->ifindex = route.iface;
276 prefix_value->gw = route.gw;
277 prefix_value->metric = route.metric;
278 assert(bpf_map_update_elem(
279 map_fd[0],
280 prefix_key,
281 prefix_value,
282 0) == 0);
286 memset(&route, 0, sizeof(route));
287 memset(dsts, 0, sizeof(dsts));
288 memset(dsts_len, 0, sizeof(dsts_len));
289 memset(gws, 0, sizeof(gws));
290 memset(ifs, 0, sizeof(ifs));
291 memset(&route, 0, sizeof(route));
295 /* Function to read the existing route table when the process is launched*/
296 static int get_route_table(int rtm_family)
298 struct sockaddr_nl sa;
299 struct nlmsghdr *nh;
300 int sock, seq = 0;
301 struct msghdr msg;
302 struct iovec iov;
303 int ret = 0;
304 int nll;
306 struct {
307 struct nlmsghdr nl;
308 struct rtmsg rt;
309 char buf[8192];
310 } req;
312 sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
313 if (sock < 0) {
314 printf("open netlink socket: %s\n", strerror(errno));
315 return -1;
317 memset(&sa, 0, sizeof(sa));
318 sa.nl_family = AF_NETLINK;
319 if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
320 printf("bind to netlink: %s\n", strerror(errno));
321 ret = -1;
322 goto cleanup;
324 memset(&req, 0, sizeof(req));
325 req.nl.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg));
326 req.nl.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
327 req.nl.nlmsg_type = RTM_GETROUTE;
329 req.rt.rtm_family = rtm_family;
330 req.rt.rtm_table = RT_TABLE_MAIN;
331 req.nl.nlmsg_pid = 0;
332 req.nl.nlmsg_seq = ++seq;
333 memset(&msg, 0, sizeof(msg));
334 iov.iov_base = (void *)&req.nl;
335 iov.iov_len = req.nl.nlmsg_len;
336 msg.msg_iov = &iov;
337 msg.msg_iovlen = 1;
338 ret = sendmsg(sock, &msg, 0);
339 if (ret < 0) {
340 printf("send to netlink: %s\n", strerror(errno));
341 ret = -1;
342 goto cleanup;
344 memset(buf, 0, sizeof(buf));
345 nll = recv_msg(sa, sock);
346 if (nll < 0) {
347 printf("recv from netlink: %s\n", strerror(nll));
348 ret = -1;
349 goto cleanup;
351 nh = (struct nlmsghdr *)buf;
352 read_route(nh, nll);
353 cleanup:
354 close(sock);
355 return ret;
358 /* Function to parse the arp entry returned by netlink
359 * Updates the arp entry related map entries
361 static void read_arp(struct nlmsghdr *nh, int nll)
363 struct rtattr *rt_attr;
364 char dsts[24], mac[24];
365 struct ndmsg *rt_msg;
366 int rtl, ndm_family;
368 struct arp_table {
369 __be64 mac;
370 __be32 dst;
371 } arp_entry;
372 struct direct_map {
373 struct arp_table arp;
374 int ifindex;
375 __be64 mac;
376 } direct_entry;
378 if (nh->nlmsg_type == RTM_GETNEIGH)
379 printf("READING arp entry\n");
380 printf("Address\tHwAddress\n");
381 for (; NLMSG_OK(nh, nll); nh = NLMSG_NEXT(nh, nll)) {
382 rt_msg = (struct ndmsg *)NLMSG_DATA(nh);
383 rt_attr = (struct rtattr *)RTM_RTA(rt_msg);
384 ndm_family = rt_msg->ndm_family;
385 rtl = RTM_PAYLOAD(nh);
386 for (; RTA_OK(rt_attr, rtl); rt_attr = RTA_NEXT(rt_attr, rtl)) {
387 switch (rt_attr->rta_type) {
388 case NDA_DST:
389 sprintf(dsts, "%u",
390 *((__be32 *)RTA_DATA(rt_attr)));
391 break;
392 case NDA_LLADDR:
393 sprintf(mac, "%lld",
394 *((__be64 *)RTA_DATA(rt_attr)));
395 break;
396 default:
397 break;
400 arp_entry.dst = atoi(dsts);
401 arp_entry.mac = atol(mac);
402 printf("%x\t\t%llx\n", arp_entry.dst, arp_entry.mac);
403 if (ndm_family == AF_INET) {
404 if (bpf_map_lookup_elem(map_fd[3], &arp_entry.dst,
405 &direct_entry) == 0) {
406 if (nh->nlmsg_type == RTM_DELNEIGH) {
407 direct_entry.arp.dst = 0;
408 direct_entry.arp.mac = 0;
409 } else if (nh->nlmsg_type == RTM_NEWNEIGH) {
410 direct_entry.arp.dst = arp_entry.dst;
411 direct_entry.arp.mac = arp_entry.mac;
413 assert(bpf_map_update_elem(map_fd[3],
414 &arp_entry.dst,
415 &direct_entry, 0
416 ) == 0);
417 memset(&direct_entry, 0, sizeof(direct_entry));
419 if (nh->nlmsg_type == RTM_DELNEIGH) {
420 assert(bpf_map_delete_elem(map_fd[2], &arp_entry.dst) == 0);
421 } else if (nh->nlmsg_type == RTM_NEWNEIGH) {
422 assert(bpf_map_update_elem(map_fd[2],
423 &arp_entry.dst,
424 &arp_entry.mac, 0
425 ) == 0);
428 memset(&arp_entry, 0, sizeof(arp_entry));
429 memset(dsts, 0, sizeof(dsts));
433 /* Function to read the existing arp table when the process is launched*/
434 static int get_arp_table(int rtm_family)
436 struct sockaddr_nl sa;
437 struct nlmsghdr *nh;
438 int sock, seq = 0;
439 struct msghdr msg;
440 struct iovec iov;
441 int ret = 0;
442 int nll;
443 struct {
444 struct nlmsghdr nl;
445 struct ndmsg rt;
446 char buf[8192];
447 } req;
449 sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
450 if (sock < 0) {
451 printf("open netlink socket: %s\n", strerror(errno));
452 return -1;
454 memset(&sa, 0, sizeof(sa));
455 sa.nl_family = AF_NETLINK;
456 if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
457 printf("bind to netlink: %s\n", strerror(errno));
458 ret = -1;
459 goto cleanup;
461 memset(&req, 0, sizeof(req));
462 req.nl.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg));
463 req.nl.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
464 req.nl.nlmsg_type = RTM_GETNEIGH;
465 req.rt.ndm_state = NUD_REACHABLE;
466 req.rt.ndm_family = rtm_family;
467 req.nl.nlmsg_pid = 0;
468 req.nl.nlmsg_seq = ++seq;
469 memset(&msg, 0, sizeof(msg));
470 iov.iov_base = (void *)&req.nl;
471 iov.iov_len = req.nl.nlmsg_len;
472 msg.msg_iov = &iov;
473 msg.msg_iovlen = 1;
474 ret = sendmsg(sock, &msg, 0);
475 if (ret < 0) {
476 printf("send to netlink: %s\n", strerror(errno));
477 ret = -1;
478 goto cleanup;
480 memset(buf, 0, sizeof(buf));
481 nll = recv_msg(sa, sock);
482 if (nll < 0) {
483 printf("recv from netlink: %s\n", strerror(nll));
484 ret = -1;
485 goto cleanup;
487 nh = (struct nlmsghdr *)buf;
488 read_arp(nh, nll);
489 cleanup:
490 close(sock);
491 return ret;
494 /* Function to keep track and update changes in route and arp table
495 * Give regular statistics of packets forwarded
497 static int monitor_route(void)
499 unsigned int nr_cpus = bpf_num_possible_cpus();
500 const unsigned int nr_keys = 256;
501 struct pollfd fds_route, fds_arp;
502 __u64 prev[nr_keys][nr_cpus];
503 struct sockaddr_nl la, lr;
504 __u64 values[nr_cpus];
505 struct nlmsghdr *nh;
506 int nll, ret = 0;
507 int interval = 5;
508 __u32 key;
509 int i;
511 sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
512 if (sock < 0) {
513 printf("open netlink socket: %s\n", strerror(errno));
514 return -1;
517 fcntl(sock, F_SETFL, O_NONBLOCK);
518 memset(&lr, 0, sizeof(lr));
519 lr.nl_family = AF_NETLINK;
520 lr.nl_groups = RTMGRP_IPV6_ROUTE | RTMGRP_IPV4_ROUTE | RTMGRP_NOTIFY;
521 if (bind(sock, (struct sockaddr *)&lr, sizeof(lr)) < 0) {
522 printf("bind to netlink: %s\n", strerror(errno));
523 ret = -1;
524 goto cleanup;
526 fds_route.fd = sock;
527 fds_route.events = POLL_IN;
529 sock_arp = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
530 if (sock_arp < 0) {
531 printf("open netlink socket: %s\n", strerror(errno));
532 return -1;
535 fcntl(sock_arp, F_SETFL, O_NONBLOCK);
536 memset(&la, 0, sizeof(la));
537 la.nl_family = AF_NETLINK;
538 la.nl_groups = RTMGRP_NEIGH | RTMGRP_NOTIFY;
539 if (bind(sock_arp, (struct sockaddr *)&la, sizeof(la)) < 0) {
540 printf("bind to netlink: %s\n", strerror(errno));
541 ret = -1;
542 goto cleanup;
544 fds_arp.fd = sock_arp;
545 fds_arp.events = POLL_IN;
547 memset(prev, 0, sizeof(prev));
548 do {
549 signal(SIGINT, close_and_exit);
550 signal(SIGTERM, close_and_exit);
552 sleep(interval);
553 for (key = 0; key < nr_keys; key++) {
554 __u64 sum = 0;
556 assert(bpf_map_lookup_elem(map_fd[1], &key, values) == 0);
557 for (i = 0; i < nr_cpus; i++)
558 sum += (values[i] - prev[key][i]);
559 if (sum)
560 printf("proto %u: %10llu pkt/s\n",
561 key, sum / interval);
562 memcpy(prev[key], values, sizeof(values));
565 memset(buf, 0, sizeof(buf));
566 if (poll(&fds_route, 1, 3) == POLL_IN) {
567 nll = recv_msg(lr, sock);
568 if (nll < 0) {
569 printf("recv from netlink: %s\n", strerror(nll));
570 ret = -1;
571 goto cleanup;
574 nh = (struct nlmsghdr *)buf;
575 printf("Routing table updated.\n");
576 read_route(nh, nll);
578 memset(buf, 0, sizeof(buf));
579 if (poll(&fds_arp, 1, 3) == POLL_IN) {
580 nll = recv_msg(la, sock_arp);
581 if (nll < 0) {
582 printf("recv from netlink: %s\n", strerror(nll));
583 ret = -1;
584 goto cleanup;
587 nh = (struct nlmsghdr *)buf;
588 read_arp(nh, nll);
591 } while (1);
592 cleanup:
593 close(sock);
594 return ret;
597 int main(int ac, char **argv)
599 char filename[256];
600 char **ifname_list;
601 int i = 1;
603 snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
604 if (ac < 2) {
605 printf("usage: %s [-S] Interface name list\n", argv[0]);
606 return 1;
608 if (!strcmp(argv[1], "-S")) {
609 flags = XDP_FLAGS_SKB_MODE;
610 total_ifindex = ac - 2;
611 ifname_list = (argv + 2);
612 } else {
613 flags = 0;
614 total_ifindex = ac - 1;
615 ifname_list = (argv + 1);
617 if (load_bpf_file(filename)) {
618 printf("%s", bpf_log_buf);
619 return 1;
621 printf("\n**************loading bpf file*********************\n\n\n");
622 if (!prog_fd[0]) {
623 printf("load_bpf_file: %s\n", strerror(errno));
624 return 1;
626 ifindex_list = (int *)malloc(total_ifindex * sizeof(int *));
627 for (i = 0; i < total_ifindex; i++) {
628 ifindex_list[i] = if_nametoindex(ifname_list[i]);
629 if (!ifindex_list[i]) {
630 printf("Couldn't translate interface name: %s",
631 strerror(errno));
632 return 1;
635 for (i = 0; i < total_ifindex; i++) {
636 if (bpf_set_link_xdp_fd(ifindex_list[i], prog_fd[0], flags) < 0) {
637 printf("link set xdp fd failed\n");
638 int recovery_index = i;
640 for (i = 0; i < recovery_index; i++)
641 bpf_set_link_xdp_fd(ifindex_list[i], -1, flags);
643 return 1;
645 printf("Attached to %d\n", ifindex_list[i]);
647 signal(SIGINT, int_exit);
648 signal(SIGTERM, int_exit);
650 printf("*******************ROUTE TABLE*************************\n\n\n");
651 get_route_table(AF_INET);
652 printf("*******************ARP TABLE***************************\n\n\n");
653 get_arp_table(AF_INET);
654 if (monitor_route() < 0) {
655 printf("Error in receiving route update");
656 return 1;
659 return 0;