700 lines · c
1// SPDX-License-Identifier: GPL-2.0-only2/* Copyright (C) 2017 Cavium, Inc.3 */4#include <linux/bpf.h>5#include <linux/netlink.h>6#include <linux/rtnetlink.h>7#include <assert.h>8#include <errno.h>9#include <signal.h>10#include <stdio.h>11#include <stdlib.h>12#include <string.h>13#include <sys/socket.h>14#include <unistd.h>15#include <bpf/bpf.h>16#include <arpa/inet.h>17#include <fcntl.h>18#include <poll.h>19#include <net/if.h>20#include <netdb.h>21#include <sys/ioctl.h>22#include <sys/syscall.h>23#include "bpf_util.h"24#include <bpf/libbpf.h>25#include <libgen.h>26#include <getopt.h>27#include <pthread.h>28#include "xdp_sample_user.h"29#include "xdp_router_ipv4.skel.h"30 31static const char *__doc__ =32"XDP IPv4 router implementation\n"33"Usage: xdp_router_ipv4 <IFNAME-0> ... <IFNAME-N>\n";34 35static char buf[8192];36static int lpm_map_fd;37static int arp_table_map_fd;38static int exact_match_map_fd;39static int tx_port_map_fd;40 41static bool routes_thread_exit;42static int interval = 5;43 44static int mask = SAMPLE_RX_CNT | SAMPLE_REDIRECT_ERR_MAP_CNT |45 SAMPLE_DEVMAP_XMIT_CNT_MULTI | SAMPLE_EXCEPTION_CNT;46 47DEFINE_SAMPLE_INIT(xdp_router_ipv4);48 49static const struct option long_options[] = {50 { "help", no_argument, NULL, 'h' },51 { "skb-mode", no_argument, NULL, 'S' },52 { "force", no_argument, NULL, 'F' },53 { "interval", required_argument, NULL, 'i' },54 { "verbose", no_argument, NULL, 'v' },55 { "stats", no_argument, NULL, 's' },56 {}57};58 59static int get_route_table(int rtm_family);60 61static int recv_msg(struct sockaddr_nl sock_addr, int sock)62{63 struct nlmsghdr *nh;64 int len, nll = 0;65 char *buf_ptr;66 67 buf_ptr = buf;68 while (1) {69 len = recv(sock, buf_ptr, sizeof(buf) - nll, 0);70 if (len < 0)71 return len;72 73 nh = (struct nlmsghdr *)buf_ptr;74 75 if (nh->nlmsg_type == NLMSG_DONE)76 break;77 buf_ptr += len;78 nll += len;79 if ((sock_addr.nl_groups & RTMGRP_NEIGH) == RTMGRP_NEIGH)80 break;81 82 if ((sock_addr.nl_groups & RTMGRP_IPV4_ROUTE) == RTMGRP_IPV4_ROUTE)83 break;84 }85 return nll;86}87 88/* Function to parse the route entry returned by netlink89 * Updates the route entry related map entries90 */91static void read_route(struct nlmsghdr *nh, int nll)92{93 char dsts[24], gws[24], ifs[16], dsts_len[24], metrics[24];94 struct bpf_lpm_trie_key_u8 *prefix_key;95 struct rtattr *rt_attr;96 struct rtmsg *rt_msg;97 int rtm_family;98 int rtl;99 int i;100 struct route_table {101 int dst_len, iface, metric;102 __be32 dst, gw;103 __be64 mac;104 } route;105 struct arp_table {106 __be64 mac;107 __be32 dst;108 };109 110 struct direct_map {111 struct arp_table arp;112 int ifindex;113 __be64 mac;114 } direct_entry;115 116 memset(&route, 0, sizeof(route));117 for (; NLMSG_OK(nh, nll); nh = NLMSG_NEXT(nh, nll)) {118 rt_msg = (struct rtmsg *)NLMSG_DATA(nh);119 rtm_family = rt_msg->rtm_family;120 if (rtm_family == AF_INET)121 if (rt_msg->rtm_table != RT_TABLE_MAIN)122 continue;123 rt_attr = (struct rtattr *)RTM_RTA(rt_msg);124 rtl = RTM_PAYLOAD(nh);125 126 for (; RTA_OK(rt_attr, rtl); rt_attr = RTA_NEXT(rt_attr, rtl)) {127 switch (rt_attr->rta_type) {128 case NDA_DST:129 sprintf(dsts, "%u",130 (*((__be32 *)RTA_DATA(rt_attr))));131 break;132 case RTA_GATEWAY:133 sprintf(gws, "%u",134 *((__be32 *)RTA_DATA(rt_attr)));135 break;136 case RTA_OIF:137 sprintf(ifs, "%u",138 *((int *)RTA_DATA(rt_attr)));139 break;140 case RTA_METRICS:141 sprintf(metrics, "%u",142 *((int *)RTA_DATA(rt_attr)));143 default:144 break;145 }146 }147 sprintf(dsts_len, "%d", rt_msg->rtm_dst_len);148 route.dst = atoi(dsts);149 route.dst_len = atoi(dsts_len);150 route.gw = atoi(gws);151 route.iface = atoi(ifs);152 route.metric = atoi(metrics);153 assert(get_mac_addr(route.iface, &route.mac) == 0);154 assert(bpf_map_update_elem(tx_port_map_fd,155 &route.iface, &route.iface, 0) == 0);156 if (rtm_family == AF_INET) {157 struct trie_value {158 __u8 prefix[4];159 __be64 value;160 int ifindex;161 int metric;162 __be32 gw;163 } *prefix_value;164 165 prefix_key = alloca(sizeof(*prefix_key) + 4);166 prefix_value = alloca(sizeof(*prefix_value));167 168 prefix_key->prefixlen = 32;169 prefix_key->prefixlen = route.dst_len;170 direct_entry.mac = route.mac & 0xffffffffffff;171 direct_entry.ifindex = route.iface;172 direct_entry.arp.mac = 0;173 direct_entry.arp.dst = 0;174 if (route.dst_len == 32) {175 if (nh->nlmsg_type == RTM_DELROUTE) {176 assert(bpf_map_delete_elem(exact_match_map_fd,177 &route.dst) == 0);178 } else {179 if (bpf_map_lookup_elem(arp_table_map_fd,180 &route.dst,181 &direct_entry.arp.mac) == 0)182 direct_entry.arp.dst = route.dst;183 assert(bpf_map_update_elem(exact_match_map_fd,184 &route.dst,185 &direct_entry, 0) == 0);186 }187 }188 for (i = 0; i < 4; i++)189 prefix_key->data[i] = (route.dst >> i * 8) & 0xff;190 191 if (bpf_map_lookup_elem(lpm_map_fd, prefix_key,192 prefix_value) < 0) {193 for (i = 0; i < 4; i++)194 prefix_value->prefix[i] = prefix_key->data[i];195 prefix_value->value = route.mac & 0xffffffffffff;196 prefix_value->ifindex = route.iface;197 prefix_value->gw = route.gw;198 prefix_value->metric = route.metric;199 200 assert(bpf_map_update_elem(lpm_map_fd,201 prefix_key,202 prefix_value, 0203 ) == 0);204 } else {205 if (nh->nlmsg_type == RTM_DELROUTE) {206 assert(bpf_map_delete_elem(lpm_map_fd,207 prefix_key208 ) == 0);209 /* Rereading the route table to check if210 * there is an entry with the same211 * prefix but a different metric as the212 * deleted entry.213 */214 get_route_table(AF_INET);215 } else if (prefix_key->data[0] ==216 prefix_value->prefix[0] &&217 prefix_key->data[1] ==218 prefix_value->prefix[1] &&219 prefix_key->data[2] ==220 prefix_value->prefix[2] &&221 prefix_key->data[3] ==222 prefix_value->prefix[3] &&223 route.metric >= prefix_value->metric) {224 continue;225 } else {226 for (i = 0; i < 4; i++)227 prefix_value->prefix[i] =228 prefix_key->data[i];229 prefix_value->value =230 route.mac & 0xffffffffffff;231 prefix_value->ifindex = route.iface;232 prefix_value->gw = route.gw;233 prefix_value->metric = route.metric;234 assert(bpf_map_update_elem(lpm_map_fd,235 prefix_key,236 prefix_value,237 0) == 0);238 }239 }240 }241 memset(&route, 0, sizeof(route));242 memset(dsts, 0, sizeof(dsts));243 memset(dsts_len, 0, sizeof(dsts_len));244 memset(gws, 0, sizeof(gws));245 memset(ifs, 0, sizeof(ifs));246 memset(&route, 0, sizeof(route));247 }248}249 250/* Function to read the existing route table when the process is launched*/251static int get_route_table(int rtm_family)252{253 struct sockaddr_nl sa;254 struct nlmsghdr *nh;255 int sock, seq = 0;256 struct msghdr msg;257 struct iovec iov;258 int ret = 0;259 int nll;260 261 struct {262 struct nlmsghdr nl;263 struct rtmsg rt;264 char buf[8192];265 } req;266 267 sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);268 if (sock < 0) {269 fprintf(stderr, "open netlink socket: %s\n", strerror(errno));270 return -errno;271 }272 memset(&sa, 0, sizeof(sa));273 sa.nl_family = AF_NETLINK;274 if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {275 fprintf(stderr, "bind netlink socket: %s\n", strerror(errno));276 ret = -errno;277 goto cleanup;278 }279 memset(&req, 0, sizeof(req));280 req.nl.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg));281 req.nl.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;282 req.nl.nlmsg_type = RTM_GETROUTE;283 284 req.rt.rtm_family = rtm_family;285 req.rt.rtm_table = RT_TABLE_MAIN;286 req.nl.nlmsg_pid = 0;287 req.nl.nlmsg_seq = ++seq;288 memset(&msg, 0, sizeof(msg));289 iov.iov_base = (void *)&req.nl;290 iov.iov_len = req.nl.nlmsg_len;291 msg.msg_iov = &iov;292 msg.msg_iovlen = 1;293 ret = sendmsg(sock, &msg, 0);294 if (ret < 0) {295 fprintf(stderr, "send to netlink: %s\n", strerror(errno));296 ret = -errno;297 goto cleanup;298 }299 memset(buf, 0, sizeof(buf));300 nll = recv_msg(sa, sock);301 if (nll < 0) {302 fprintf(stderr, "recv from netlink: %s\n", strerror(nll));303 ret = nll;304 goto cleanup;305 }306 nh = (struct nlmsghdr *)buf;307 read_route(nh, nll);308cleanup:309 close(sock);310 return ret;311}312 313/* Function to parse the arp entry returned by netlink314 * Updates the arp entry related map entries315 */316static void read_arp(struct nlmsghdr *nh, int nll)317{318 struct rtattr *rt_attr;319 char dsts[24], mac[24];320 struct ndmsg *rt_msg;321 int rtl, ndm_family;322 323 struct arp_table {324 __be64 mac;325 __be32 dst;326 } arp_entry;327 struct direct_map {328 struct arp_table arp;329 int ifindex;330 __be64 mac;331 } direct_entry;332 333 for (; NLMSG_OK(nh, nll); nh = NLMSG_NEXT(nh, nll)) {334 rt_msg = (struct ndmsg *)NLMSG_DATA(nh);335 rt_attr = (struct rtattr *)RTM_RTA(rt_msg);336 ndm_family = rt_msg->ndm_family;337 rtl = RTM_PAYLOAD(nh);338 for (; RTA_OK(rt_attr, rtl); rt_attr = RTA_NEXT(rt_attr, rtl)) {339 switch (rt_attr->rta_type) {340 case NDA_DST:341 sprintf(dsts, "%u",342 *((__be32 *)RTA_DATA(rt_attr)));343 break;344 case NDA_LLADDR:345 sprintf(mac, "%lld",346 *((__be64 *)RTA_DATA(rt_attr)));347 break;348 default:349 break;350 }351 }352 arp_entry.dst = atoi(dsts);353 arp_entry.mac = atol(mac);354 355 if (ndm_family == AF_INET) {356 if (bpf_map_lookup_elem(exact_match_map_fd,357 &arp_entry.dst,358 &direct_entry) == 0) {359 if (nh->nlmsg_type == RTM_DELNEIGH) {360 direct_entry.arp.dst = 0;361 direct_entry.arp.mac = 0;362 } else if (nh->nlmsg_type == RTM_NEWNEIGH) {363 direct_entry.arp.dst = arp_entry.dst;364 direct_entry.arp.mac = arp_entry.mac;365 }366 assert(bpf_map_update_elem(exact_match_map_fd,367 &arp_entry.dst,368 &direct_entry, 0369 ) == 0);370 memset(&direct_entry, 0, sizeof(direct_entry));371 }372 if (nh->nlmsg_type == RTM_DELNEIGH) {373 assert(bpf_map_delete_elem(arp_table_map_fd,374 &arp_entry.dst) == 0);375 } else if (nh->nlmsg_type == RTM_NEWNEIGH) {376 assert(bpf_map_update_elem(arp_table_map_fd,377 &arp_entry.dst,378 &arp_entry.mac, 0379 ) == 0);380 }381 }382 memset(&arp_entry, 0, sizeof(arp_entry));383 memset(dsts, 0, sizeof(dsts));384 }385}386 387/* Function to read the existing arp table when the process is launched*/388static int get_arp_table(int rtm_family)389{390 struct sockaddr_nl sa;391 struct nlmsghdr *nh;392 int sock, seq = 0;393 struct msghdr msg;394 struct iovec iov;395 int ret = 0;396 int nll;397 struct {398 struct nlmsghdr nl;399 struct ndmsg rt;400 char buf[8192];401 } req;402 403 sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);404 if (sock < 0) {405 fprintf(stderr, "open netlink socket: %s\n", strerror(errno));406 return -errno;407 }408 memset(&sa, 0, sizeof(sa));409 sa.nl_family = AF_NETLINK;410 if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {411 fprintf(stderr, "bind netlink socket: %s\n", strerror(errno));412 ret = -errno;413 goto cleanup;414 }415 memset(&req, 0, sizeof(req));416 req.nl.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg));417 req.nl.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;418 req.nl.nlmsg_type = RTM_GETNEIGH;419 req.rt.ndm_state = NUD_REACHABLE;420 req.rt.ndm_family = rtm_family;421 req.nl.nlmsg_pid = 0;422 req.nl.nlmsg_seq = ++seq;423 memset(&msg, 0, sizeof(msg));424 iov.iov_base = (void *)&req.nl;425 iov.iov_len = req.nl.nlmsg_len;426 msg.msg_iov = &iov;427 msg.msg_iovlen = 1;428 ret = sendmsg(sock, &msg, 0);429 if (ret < 0) {430 fprintf(stderr, "send to netlink: %s\n", strerror(errno));431 ret = -errno;432 goto cleanup;433 }434 memset(buf, 0, sizeof(buf));435 nll = recv_msg(sa, sock);436 if (nll < 0) {437 fprintf(stderr, "recv from netlink: %s\n", strerror(nll));438 ret = nll;439 goto cleanup;440 }441 nh = (struct nlmsghdr *)buf;442 read_arp(nh, nll);443cleanup:444 close(sock);445 return ret;446}447 448/* Function to keep track and update changes in route and arp table449 * Give regular statistics of packets forwarded450 */451static void *monitor_routes_thread(void *arg)452{453 struct pollfd fds_route, fds_arp;454 struct sockaddr_nl la, lr;455 int sock, sock_arp, nll;456 struct nlmsghdr *nh;457 458 sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);459 if (sock < 0) {460 fprintf(stderr, "open netlink socket: %s\n", strerror(errno));461 return NULL;462 }463 464 fcntl(sock, F_SETFL, O_NONBLOCK);465 memset(&lr, 0, sizeof(lr));466 lr.nl_family = AF_NETLINK;467 lr.nl_groups = RTMGRP_IPV6_ROUTE | RTMGRP_IPV4_ROUTE | RTMGRP_NOTIFY;468 if (bind(sock, (struct sockaddr *)&lr, sizeof(lr)) < 0) {469 fprintf(stderr, "bind netlink socket: %s\n", strerror(errno));470 close(sock);471 return NULL;472 }473 474 fds_route.fd = sock;475 fds_route.events = POLL_IN;476 477 sock_arp = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);478 if (sock_arp < 0) {479 fprintf(stderr, "open netlink socket: %s\n", strerror(errno));480 close(sock);481 return NULL;482 }483 484 fcntl(sock_arp, F_SETFL, O_NONBLOCK);485 memset(&la, 0, sizeof(la));486 la.nl_family = AF_NETLINK;487 la.nl_groups = RTMGRP_NEIGH | RTMGRP_NOTIFY;488 if (bind(sock_arp, (struct sockaddr *)&la, sizeof(la)) < 0) {489 fprintf(stderr, "bind netlink socket: %s\n", strerror(errno));490 goto cleanup;491 }492 493 fds_arp.fd = sock_arp;494 fds_arp.events = POLL_IN;495 496 /* dump route and arp tables */497 if (get_arp_table(AF_INET) < 0) {498 fprintf(stderr, "Failed reading arp table\n");499 goto cleanup;500 }501 502 if (get_route_table(AF_INET) < 0) {503 fprintf(stderr, "Failed reading route table\n");504 goto cleanup;505 }506 507 while (!routes_thread_exit) {508 memset(buf, 0, sizeof(buf));509 if (poll(&fds_route, 1, 3) == POLL_IN) {510 nll = recv_msg(lr, sock);511 if (nll < 0) {512 fprintf(stderr, "recv from netlink: %s\n",513 strerror(nll));514 goto cleanup;515 }516 517 nh = (struct nlmsghdr *)buf;518 read_route(nh, nll);519 }520 521 memset(buf, 0, sizeof(buf));522 if (poll(&fds_arp, 1, 3) == POLL_IN) {523 nll = recv_msg(la, sock_arp);524 if (nll < 0) {525 fprintf(stderr, "recv from netlink: %s\n",526 strerror(nll));527 goto cleanup;528 }529 530 nh = (struct nlmsghdr *)buf;531 read_arp(nh, nll);532 }533 534 sleep(interval);535 }536 537cleanup:538 close(sock_arp);539 close(sock);540 return NULL;541}542 543static void usage(char *argv[], const struct option *long_options,544 const char *doc, int mask, bool error,545 struct bpf_object *obj)546{547 sample_usage(argv, long_options, doc, mask, error);548}549 550int main(int argc, char **argv)551{552 bool error = true, generic = false, force = false;553 int opt, ret = EXIT_FAIL_BPF;554 struct xdp_router_ipv4 *skel;555 int i, total_ifindex = argc - 1;556 char **ifname_list = argv + 1;557 pthread_t routes_thread;558 int longindex = 0;559 560 if (libbpf_set_strict_mode(LIBBPF_STRICT_ALL) < 0) {561 fprintf(stderr, "Failed to set libbpf strict mode: %s\n",562 strerror(errno));563 goto end;564 }565 566 skel = xdp_router_ipv4__open();567 if (!skel) {568 fprintf(stderr, "Failed to xdp_router_ipv4__open: %s\n",569 strerror(errno));570 goto end;571 }572 573 ret = sample_init_pre_load(skel);574 if (ret < 0) {575 fprintf(stderr, "Failed to sample_init_pre_load: %s\n",576 strerror(-ret));577 ret = EXIT_FAIL_BPF;578 goto end_destroy;579 }580 581 ret = xdp_router_ipv4__load(skel);582 if (ret < 0) {583 fprintf(stderr, "Failed to xdp_router_ipv4__load: %s\n",584 strerror(errno));585 goto end_destroy;586 }587 588 ret = sample_init(skel, mask);589 if (ret < 0) {590 fprintf(stderr, "Failed to initialize sample: %s\n", strerror(-ret));591 ret = EXIT_FAIL;592 goto end_destroy;593 }594 595 while ((opt = getopt_long(argc, argv, "si:SFvh",596 long_options, &longindex)) != -1) {597 switch (opt) {598 case 's':599 mask |= SAMPLE_REDIRECT_MAP_CNT;600 total_ifindex--;601 ifname_list++;602 break;603 case 'i':604 interval = strtoul(optarg, NULL, 0);605 total_ifindex -= 2;606 ifname_list += 2;607 break;608 case 'S':609 generic = true;610 total_ifindex--;611 ifname_list++;612 break;613 case 'F':614 force = true;615 total_ifindex--;616 ifname_list++;617 break;618 case 'v':619 sample_switch_mode();620 total_ifindex--;621 ifname_list++;622 break;623 case 'h':624 error = false;625 default:626 usage(argv, long_options, __doc__, mask, error, skel->obj);627 goto end_destroy;628 }629 }630 631 ret = EXIT_FAIL_OPTION;632 if (optind == argc) {633 usage(argv, long_options, __doc__, mask, true, skel->obj);634 goto end_destroy;635 }636 637 lpm_map_fd = bpf_map__fd(skel->maps.lpm_map);638 if (lpm_map_fd < 0) {639 fprintf(stderr, "Failed loading lpm_map %s\n",640 strerror(-lpm_map_fd));641 goto end_destroy;642 }643 arp_table_map_fd = bpf_map__fd(skel->maps.arp_table);644 if (arp_table_map_fd < 0) {645 fprintf(stderr, "Failed loading arp_table_map_fd %s\n",646 strerror(-arp_table_map_fd));647 goto end_destroy;648 }649 exact_match_map_fd = bpf_map__fd(skel->maps.exact_match);650 if (exact_match_map_fd < 0) {651 fprintf(stderr, "Failed loading exact_match_map_fd %s\n",652 strerror(-exact_match_map_fd));653 goto end_destroy;654 }655 tx_port_map_fd = bpf_map__fd(skel->maps.tx_port);656 if (tx_port_map_fd < 0) {657 fprintf(stderr, "Failed loading tx_port_map_fd %s\n",658 strerror(-tx_port_map_fd));659 goto end_destroy;660 }661 662 ret = EXIT_FAIL_XDP;663 for (i = 0; i < total_ifindex; i++) {664 int index = if_nametoindex(ifname_list[i]);665 666 if (!index) {667 fprintf(stderr, "Interface %s not found %s\n",668 ifname_list[i], strerror(-tx_port_map_fd));669 goto end_destroy;670 }671 if (sample_install_xdp(skel->progs.xdp_router_ipv4_prog,672 index, generic, force) < 0)673 goto end_destroy;674 }675 676 ret = pthread_create(&routes_thread, NULL, monitor_routes_thread, NULL);677 if (ret) {678 fprintf(stderr, "Failed creating routes_thread: %s\n", strerror(-ret));679 ret = EXIT_FAIL;680 goto end_destroy;681 }682 683 ret = sample_run(interval, NULL, NULL);684 routes_thread_exit = true;685 686 if (ret < 0) {687 fprintf(stderr, "Failed during sample run: %s\n", strerror(-ret));688 ret = EXIT_FAIL;689 goto end_thread_wait;690 }691 ret = EXIT_OK;692 693end_thread_wait:694 pthread_join(routes_thread, NULL);695end_destroy:696 xdp_router_ipv4__destroy(skel);697end:698 sample_exit(ret);699}700