tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
xdp: Sample xdp program implementing ip forward
Implements port to port forwarding with route table and arp table
lookup for ipv4 packets using bpf_redirect helper function and
lpm_trie map.
Signed-off-by: Christina Jacob <Christina.Jacob@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
authored by
Christina Jacob
and committed by
David S. Miller
3e29cd0e
4ad1ceec
+849
+4
samples/bpf/Makefile
+4
samples/bpf/Makefile
···
29
29
hostprogs-y += test_cgrp2_sock2
30
30
hostprogs-y += xdp1
31
31
hostprogs-y += xdp2
32
+
hostprogs-y += xdp_router_ipv4
32
33
hostprogs-y += test_current_task_under_cgroup
33
34
hostprogs-y += trace_event
34
35
hostprogs-y += sampleip
···
77
76
xdp1-objs := bpf_load.o $(LIBBPF) xdp1_user.o
78
77
# reuse xdp1 source intentionally
79
78
xdp2-objs := bpf_load.o $(LIBBPF) xdp1_user.o
79
+
xdp_router_ipv4-objs := bpf_load.o $(LIBBPF) xdp_router_ipv4_user.o
80
80
test_current_task_under_cgroup-objs := bpf_load.o $(LIBBPF) $(CGROUP_HELPERS) \
81
81
test_current_task_under_cgroup_user.o
82
82
trace_event-objs := bpf_load.o $(LIBBPF) trace_event_user.o
···
120
118
always += test_cgrp2_tc_kern.o
121
119
always += xdp1_kern.o
122
120
always += xdp2_kern.o
121
+
always += xdp_router_ipv4_kern.o
123
122
always += test_current_task_under_cgroup_kern.o
124
123
always += trace_event_kern.o
125
124
always += sampleip_kern.o
···
169
166
HOSTLOADLIBES_test_overhead += -lelf -lrt
170
167
HOSTLOADLIBES_xdp1 += -lelf
171
168
HOSTLOADLIBES_xdp2 += -lelf
169
+
HOSTLOADLIBES_xdp_router_ipv4 += -lelf
172
170
HOSTLOADLIBES_test_current_task_under_cgroup += -lelf
173
171
HOSTLOADLIBES_trace_event += -lelf
174
172
HOSTLOADLIBES_sampleip += -lelf
+186
samples/bpf/xdp_router_ipv4_kern.c
+186
samples/bpf/xdp_router_ipv4_kern.c
···
1
+
/* Copyright (C) 2017 Cavium, Inc.
2
+
*
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
+
#define KBUILD_MODNAME "foo"
8
+
#include <uapi/linux/bpf.h>
9
+
#include <linux/in.h>
10
+
#include <linux/if_ether.h>
11
+
#include <linux/if_packet.h>
12
+
#include <linux/if_vlan.h>
13
+
#include <linux/ip.h>
14
+
#include <linux/ipv6.h>
15
+
#include "bpf_helpers.h"
16
+
#include <linux/slab.h>
17
+
#include <net/ip_fib.h>
18
+
19
+
struct trie_value {
20
+
__u8 prefix[4];
21
+
__be64 value;
22
+
int ifindex;
23
+
int metric;
24
+
__be32 gw;
25
+
};
26
+
27
+
/* Key for lpm_trie*/
28
+
union key_4 {
29
+
u32 b32[2];
30
+
u8 b8[8];
31
+
};
32
+
33
+
struct arp_entry {
34
+
__be64 mac;
35
+
__be32 dst;
36
+
};
37
+
38
+
struct direct_map {
39
+
struct arp_entry arp;
40
+
int ifindex;
41
+
__be64 mac;
42
+
};
43
+
44
+
/* Map for trie implementation*/
45
+
struct bpf_map_def SEC("maps") lpm_map = {
46
+
.type = BPF_MAP_TYPE_LPM_TRIE,
47
+
.key_size = 8,
48
+
.value_size = sizeof(struct trie_value),
49
+
.max_entries = 50,
50
+
.map_flags = BPF_F_NO_PREALLOC,
51
+
};
52
+
53
+
/* Map for counter*/
54
+
struct bpf_map_def SEC("maps") rxcnt = {
55
+
.type = BPF_MAP_TYPE_PERCPU_ARRAY,
56
+
.key_size = sizeof(u32),
57
+
.value_size = sizeof(u64),
58
+
.max_entries = 256,
59
+
};
60
+
61
+
/* Map for ARP table*/
62
+
struct bpf_map_def SEC("maps") arp_table = {
63
+
.type = BPF_MAP_TYPE_HASH,
64
+
.key_size = sizeof(__be32),
65
+
.value_size = sizeof(__be64),
66
+
.max_entries = 50,
67
+
};
68
+
69
+
/* Map to keep the exact match entries in the route table*/
70
+
struct bpf_map_def SEC("maps") exact_match = {
71
+
.type = BPF_MAP_TYPE_HASH,
72
+
.key_size = sizeof(__be32),
73
+
.value_size = sizeof(struct direct_map),
74
+
.max_entries = 50,
75
+
};
76
+
77
+
struct bpf_map_def SEC("maps") tx_port = {
78
+
.type = BPF_MAP_TYPE_DEVMAP,
79
+
.key_size = sizeof(int),
80
+
.value_size = sizeof(int),
81
+
.max_entries = 100,
82
+
};
83
+
84
+
/* Function to set source and destination mac of the packet */
85
+
static inline void set_src_dst_mac(void *data, void *src, void *dst)
86
+
{
87
+
unsigned short *source = src;
88
+
unsigned short *dest = dst;
89
+
unsigned short *p = data;
90
+
91
+
__builtin_memcpy(p, dest, 6);
92
+
__builtin_memcpy(p + 3, source, 6);
93
+
}
94
+
95
+
/* Parse IPV4 packet to get SRC, DST IP and protocol */
96
+
static inline int parse_ipv4(void *data, u64 nh_off, void *data_end,
97
+
__be32 *src, __be32 *dest)
98
+
{
99
+
struct iphdr *iph = data + nh_off;
100
+
101
+
if (iph + 1 > data_end)
102
+
return 0;
103
+
*src = iph->saddr;
104
+
*dest = iph->daddr;
105
+
return iph->protocol;
106
+
}
107
+
108
+
SEC("xdp_router_ipv4")
109
+
int xdp_router_ipv4_prog(struct xdp_md *ctx)
110
+
{
111
+
void *data_end = (void *)(long)ctx->data_end;
112
+
__be64 *dest_mac = NULL, *src_mac = NULL;
113
+
void *data = (void *)(long)ctx->data;
114
+
struct trie_value *prefix_value;
115
+
int rc = XDP_DROP, forward_to;
116
+
struct ethhdr *eth = data;
117
+
union key_4 key4;
118
+
long *value;
119
+
u16 h_proto;
120
+
u32 ipproto;
121
+
u64 nh_off;
122
+
123
+
nh_off = sizeof(*eth);
124
+
if (data + nh_off > data_end)
125
+
return rc;
126
+
127
+
h_proto = eth->h_proto;
128
+
129
+
if (h_proto == htons(ETH_P_8021Q) || h_proto == htons(ETH_P_8021AD)) {
130
+
struct vlan_hdr *vhdr;
131
+
132
+
vhdr = data + nh_off;
133
+
nh_off += sizeof(struct vlan_hdr);
134
+
if (data + nh_off > data_end)
135
+
return rc;
136
+
h_proto = vhdr->h_vlan_encapsulated_proto;
137
+
}
138
+
if (h_proto == htons(ETH_P_ARP)) {
139
+
return XDP_PASS;
140
+
} else if (h_proto == htons(ETH_P_IP)) {
141
+
struct direct_map *direct_entry;
142
+
__be32 src_ip = 0, dest_ip = 0;
143
+
144
+
ipproto = parse_ipv4(data, nh_off, data_end, &src_ip, &dest_ip);
145
+
direct_entry = bpf_map_lookup_elem(&exact_match, &dest_ip);
146
+
/* Check for exact match, this would give a faster lookup*/
147
+
if (direct_entry && direct_entry->mac && direct_entry->arp.mac) {
148
+
src_mac = &direct_entry->mac;
149
+
dest_mac = &direct_entry->arp.mac;
150
+
forward_to = direct_entry->ifindex;
151
+
} else {
152
+
/* Look up in the trie for lpm*/
153
+
key4.b32[0] = 32;
154
+
key4.b8[4] = dest_ip & 0xff;
155
+
key4.b8[5] = (dest_ip >> 8) & 0xff;
156
+
key4.b8[6] = (dest_ip >> 16) & 0xff;
157
+
key4.b8[7] = (dest_ip >> 24) & 0xff;
158
+
prefix_value = bpf_map_lookup_elem(&lpm_map, &key4);
159
+
if (!prefix_value)
160
+
return XDP_DROP;
161
+
src_mac = &prefix_value->value;
162
+
if (!src_mac)
163
+
return XDP_DROP;
164
+
dest_mac = bpf_map_lookup_elem(&arp_table, &dest_ip);
165
+
if (!dest_mac) {
166
+
if (!prefix_value->gw)
167
+
return XDP_DROP;
168
+
dest_ip = prefix_value->gw;
169
+
dest_mac = bpf_map_lookup_elem(&arp_table, &dest_ip);
170
+
}
171
+
forward_to = prefix_value->ifindex;
172
+
}
173
+
} else {
174
+
ipproto = 0;
175
+
}
176
+
if (src_mac && dest_mac) {
177
+
set_src_dst_mac(data, src_mac, dest_mac);
178
+
value = bpf_map_lookup_elem(&rxcnt, &ipproto);
179
+
if (value)
180
+
*value += 1;
181
+
return bpf_redirect_map(&tx_port, forward_to, 0);
182
+
}
183
+
return rc;
184
+
}
185
+
186
+
char _license[] SEC("license") = "GPL";
+659
samples/bpf/xdp_router_ipv4_user.c
+659
samples/bpf/xdp_router_ipv4_user.c
···
1
+
/* Copyright (C) 2017 Cavium, Inc.
2
+
*
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 "libbpf.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"
28
+
29
+
int sock, sock_arp, flags = 0;
30
+
static int total_ifindex;
31
+
int *ifindex_list;
32
+
char buf[8192];
33
+
34
+
static int get_route_table(int rtm_family);
35
+
static void int_exit(int sig)
36
+
{
37
+
int i = 0;
38
+
39
+
for (i = 0; i < total_ifindex; i++)
40
+
set_link_xdp_fd(ifindex_list[i], -1, flags);
41
+
exit(0);
42
+
}
43
+
44
+
static void close_and_exit(int sig)
45
+
{
46
+
int i = 0;
47
+
48
+
close(sock);
49
+
close(sock_arp);
50
+
51
+
for (i = 0; i < total_ifindex; i++)
52
+
set_link_xdp_fd(ifindex_list[i], -1, flags);
53
+
exit(0);
54
+
}
55
+
56
+
/* Get the mac address of the interface given interface name */
57
+
static __be64 getmac(char *iface)
58
+
{
59
+
struct ifreq ifr;
60
+
__be64 mac = 0;
61
+
int fd, i;
62
+
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;
69
+
}
70
+
for (i = 0; i < 6 ; i++)
71
+
*((__u8 *)&mac + i) = (__u8)ifr.ifr_hwaddr.sa_data[i];
72
+
close(fd);
73
+
return mac;
74
+
}
75
+
76
+
static int recv_msg(struct sockaddr_nl sock_addr, int sock)
77
+
{
78
+
struct nlmsghdr *nh;
79
+
int len, nll = 0;
80
+
char *buf_ptr;
81
+
82
+
buf_ptr = buf;
83
+
while (1) {
84
+
len = recv(sock, buf_ptr, sizeof(buf) - nll, 0);
85
+
if (len < 0)
86
+
return len;
87
+
88
+
nh = (struct nlmsghdr *)buf_ptr;
89
+
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;
96
+
97
+
if ((sock_addr.nl_groups & RTMGRP_IPV4_ROUTE) == RTMGRP_IPV4_ROUTE)
98
+
break;
99
+
}
100
+
return nll;
101
+
}
102
+
103
+
/* Function to parse the route entry returned by netlink
104
+
* Updates the route entry related map entries
105
+
*/
106
+
static void read_route(struct nlmsghdr *nh, int nll)
107
+
{
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;
124
+
};
125
+
126
+
struct direct_map {
127
+
struct arp_table arp;
128
+
int ifindex;
129
+
__be64 mac;
130
+
} direct_entry;
131
+
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);
140
+
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);
151
+
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;
171
+
}
172
+
}
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;
184
+
185
+
for (i = 0; i < total_ifindex; i++)
186
+
set_link_xdp_fd(ifindex_list[i], -1, flags);
187
+
exit(0);
188
+
}
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;
198
+
199
+
prefix_key = alloca(sizeof(*prefix_key) + 3);
200
+
prefix_value = alloca(sizeof(*prefix_value));
201
+
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);
215
+
}
216
+
for (i = 0; i < 4; i++)
217
+
prefix_key->data[i] = (route.dst >> i * 8) & 0xff;
218
+
219
+
printf("%3d.%d.%d.%d\t\t%3x\t\t%d\t\t%d\t\t%s\n",
220
+
(int)prefix_key->data[0],
221
+
(int)prefix_key->data[1],
222
+
(int)prefix_key->data[2],
223
+
(int)prefix_key->data[3],
224
+
route.gw, route.dst_len,
225
+
route.metric,
226
+
route.iface_name);
227
+
if (bpf_map_lookup_elem(map_fd[0], prefix_key,
228
+
prefix_value) < 0) {
229
+
for (i = 0; i < 4; i++)
230
+
prefix_value->prefix[i] = prefix_key->data[i];
231
+
prefix_value->value = route.mac & 0xffffffffffff;
232
+
prefix_value->ifindex = route.iface;
233
+
prefix_value->gw = route.gw;
234
+
prefix_value->metric = route.metric;
235
+
236
+
assert(bpf_map_update_elem(map_fd[0],
237
+
prefix_key,
238
+
prefix_value, 0
239
+
) == 0);
240
+
} else {
241
+
if (nh->nlmsg_type == RTM_DELROUTE) {
242
+
printf("deleting entry\n");
243
+
printf("prefix key=%d.%d.%d.%d/%d",
244
+
prefix_key->data[0],
245
+
prefix_key->data[1],
246
+
prefix_key->data[2],
247
+
prefix_key->data[3],
248
+
prefix_key->prefixlen);
249
+
assert(bpf_map_delete_elem(map_fd[0],
250
+
prefix_key
251
+
) == 0);
252
+
/* Rereading the route table to check if
253
+
* there is an entry with the same
254
+
* prefix but a different metric as the
255
+
* deleted enty.
256
+
*/
257
+
get_route_table(AF_INET);
258
+
} else if (prefix_key->data[0] ==
259
+
prefix_value->prefix[0] &&
260
+
prefix_key->data[1] ==
261
+
prefix_value->prefix[1] &&
262
+
prefix_key->data[2] ==
263
+
prefix_value->prefix[2] &&
264
+
prefix_key->data[3] ==
265
+
prefix_value->prefix[3] &&
266
+
route.metric >= prefix_value->metric) {
267
+
continue;
268
+
} else {
269
+
for (i = 0; i < 4; i++)
270
+
prefix_value->prefix[i] =
271
+
prefix_key->data[i];
272
+
prefix_value->value =
273
+
route.mac & 0xffffffffffff;
274
+
prefix_value->ifindex = route.iface;
275
+
prefix_value->gw = route.gw;
276
+
prefix_value->metric = route.metric;
277
+
assert(bpf_map_update_elem(
278
+
map_fd[0],
279
+
prefix_key,
280
+
prefix_value,
281
+
0) == 0);
282
+
}
283
+
}
284
+
}
285
+
memset(&route, 0, sizeof(route));
286
+
memset(dsts, 0, sizeof(dsts));
287
+
memset(dsts_len, 0, sizeof(dsts_len));
288
+
memset(gws, 0, sizeof(gws));
289
+
memset(ifs, 0, sizeof(ifs));
290
+
memset(&route, 0, sizeof(route));
291
+
}
292
+
}
293
+
294
+
/* Function to read the existing route table when the process is launched*/
295
+
static int get_route_table(int rtm_family)
296
+
{
297
+
struct sockaddr_nl sa;
298
+
struct nlmsghdr *nh;
299
+
int sock, seq = 0;
300
+
struct msghdr msg;
301
+
struct iovec iov;
302
+
int ret = 0;
303
+
int nll;
304
+
305
+
struct {
306
+
struct nlmsghdr nl;
307
+
struct rtmsg rt;
308
+
char buf[8192];
309
+
} req;
310
+
311
+
sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
312
+
if (sock < 0) {
313
+
printf("open netlink socket: %s\n", strerror(errno));
314
+
return -1;
315
+
}
316
+
memset(&sa, 0, sizeof(sa));
317
+
sa.nl_family = AF_NETLINK;
318
+
if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
319
+
printf("bind to netlink: %s\n", strerror(errno));
320
+
ret = -1;
321
+
goto cleanup;
322
+
}
323
+
memset(&req, 0, sizeof(req));
324
+
req.nl.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg));
325
+
req.nl.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
326
+
req.nl.nlmsg_type = RTM_GETROUTE;
327
+
328
+
req.rt.rtm_family = rtm_family;
329
+
req.rt.rtm_table = RT_TABLE_MAIN;
330
+
req.nl.nlmsg_pid = 0;
331
+
req.nl.nlmsg_seq = ++seq;
332
+
memset(&msg, 0, sizeof(msg));
333
+
iov.iov_base = (void *)&req.nl;
334
+
iov.iov_len = req.nl.nlmsg_len;
335
+
msg.msg_iov = &iov;
336
+
msg.msg_iovlen = 1;
337
+
ret = sendmsg(sock, &msg, 0);
338
+
if (ret < 0) {
339
+
printf("send to netlink: %s\n", strerror(errno));
340
+
ret = -1;
341
+
goto cleanup;
342
+
}
343
+
memset(buf, 0, sizeof(buf));
344
+
nll = recv_msg(sa, sock);
345
+
if (nll < 0) {
346
+
printf("recv from netlink: %s\n", strerror(nll));
347
+
ret = -1;
348
+
goto cleanup;
349
+
}
350
+
nh = (struct nlmsghdr *)buf;
351
+
read_route(nh, nll);
352
+
cleanup:
353
+
close(sock);
354
+
return ret;
355
+
}
356
+
357
+
/* Function to parse the arp entry returned by netlink
358
+
* Updates the arp entry related map entries
359
+
*/
360
+
static void read_arp(struct nlmsghdr *nh, int nll)
361
+
{
362
+
struct rtattr *rt_attr;
363
+
char dsts[24], mac[24];
364
+
struct ndmsg *rt_msg;
365
+
int rtl, ndm_family;
366
+
367
+
struct arp_table {
368
+
__be64 mac;
369
+
__be32 dst;
370
+
} arp_entry;
371
+
struct direct_map {
372
+
struct arp_table arp;
373
+
int ifindex;
374
+
__be64 mac;
375
+
} direct_entry;
376
+
377
+
if (nh->nlmsg_type == RTM_GETNEIGH)
378
+
printf("READING arp entry\n");
379
+
printf("Address\tHwAddress\n");
380
+
for (; NLMSG_OK(nh, nll); nh = NLMSG_NEXT(nh, nll)) {
381
+
rt_msg = (struct ndmsg *)NLMSG_DATA(nh);
382
+
rt_attr = (struct rtattr *)RTM_RTA(rt_msg);
383
+
ndm_family = rt_msg->ndm_family;
384
+
rtl = RTM_PAYLOAD(nh);
385
+
for (; RTA_OK(rt_attr, rtl); rt_attr = RTA_NEXT(rt_attr, rtl)) {
386
+
switch (rt_attr->rta_type) {
387
+
case NDA_DST:
388
+
sprintf(dsts, "%u",
389
+
*((__be32 *)RTA_DATA(rt_attr)));
390
+
break;
391
+
case NDA_LLADDR:
392
+
sprintf(mac, "%lld",
393
+
*((__be64 *)RTA_DATA(rt_attr)));
394
+
break;
395
+
default:
396
+
break;
397
+
}
398
+
}
399
+
arp_entry.dst = atoi(dsts);
400
+
arp_entry.mac = atol(mac);
401
+
printf("%x\t\t%llx\n", arp_entry.dst, arp_entry.mac);
402
+
if (ndm_family == AF_INET) {
403
+
if (bpf_map_lookup_elem(map_fd[3], &arp_entry.dst,
404
+
&direct_entry) == 0) {
405
+
if (nh->nlmsg_type == RTM_DELNEIGH) {
406
+
direct_entry.arp.dst = 0;
407
+
direct_entry.arp.mac = 0;
408
+
} else if (nh->nlmsg_type == RTM_NEWNEIGH) {
409
+
direct_entry.arp.dst = arp_entry.dst;
410
+
direct_entry.arp.mac = arp_entry.mac;
411
+
}
412
+
assert(bpf_map_update_elem(map_fd[3],
413
+
&arp_entry.dst,
414
+
&direct_entry, 0
415
+
) == 0);
416
+
memset(&direct_entry, 0, sizeof(direct_entry));
417
+
}
418
+
if (nh->nlmsg_type == RTM_DELNEIGH) {
419
+
assert(bpf_map_delete_elem(map_fd[2], &arp_entry.dst) == 0);
420
+
} else if (nh->nlmsg_type == RTM_NEWNEIGH) {
421
+
assert(bpf_map_update_elem(map_fd[2],
422
+
&arp_entry.dst,
423
+
&arp_entry.mac, 0
424
+
) == 0);
425
+
}
426
+
}
427
+
memset(&arp_entry, 0, sizeof(arp_entry));
428
+
memset(dsts, 0, sizeof(dsts));
429
+
}
430
+
}
431
+
432
+
/* Function to read the existing arp table when the process is launched*/
433
+
static int get_arp_table(int rtm_family)
434
+
{
435
+
struct sockaddr_nl sa;
436
+
struct nlmsghdr *nh;
437
+
int sock, seq = 0;
438
+
struct msghdr msg;
439
+
struct iovec iov;
440
+
int ret = 0;
441
+
int nll;
442
+
struct {
443
+
struct nlmsghdr nl;
444
+
struct ndmsg rt;
445
+
char buf[8192];
446
+
} req;
447
+
448
+
sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
449
+
if (sock < 0) {
450
+
printf("open netlink socket: %s\n", strerror(errno));
451
+
return -1;
452
+
}
453
+
memset(&sa, 0, sizeof(sa));
454
+
sa.nl_family = AF_NETLINK;
455
+
if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
456
+
printf("bind to netlink: %s\n", strerror(errno));
457
+
ret = -1;
458
+
goto cleanup;
459
+
}
460
+
memset(&req, 0, sizeof(req));
461
+
req.nl.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg));
462
+
req.nl.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
463
+
req.nl.nlmsg_type = RTM_GETNEIGH;
464
+
req.rt.ndm_state = NUD_REACHABLE;
465
+
req.rt.ndm_family = rtm_family;
466
+
req.nl.nlmsg_pid = 0;
467
+
req.nl.nlmsg_seq = ++seq;
468
+
memset(&msg, 0, sizeof(msg));
469
+
iov.iov_base = (void *)&req.nl;
470
+
iov.iov_len = req.nl.nlmsg_len;
471
+
msg.msg_iov = &iov;
472
+
msg.msg_iovlen = 1;
473
+
ret = sendmsg(sock, &msg, 0);
474
+
if (ret < 0) {
475
+
printf("send to netlink: %s\n", strerror(errno));
476
+
ret = -1;
477
+
goto cleanup;
478
+
}
479
+
memset(buf, 0, sizeof(buf));
480
+
nll = recv_msg(sa, sock);
481
+
if (nll < 0) {
482
+
printf("recv from netlink: %s\n", strerror(nll));
483
+
ret = -1;
484
+
goto cleanup;
485
+
}
486
+
nh = (struct nlmsghdr *)buf;
487
+
read_arp(nh, nll);
488
+
cleanup:
489
+
close(sock);
490
+
return ret;
491
+
}
492
+
493
+
/* Function to keep track and update changes in route and arp table
494
+
* Give regular statistics of packets forwarded
495
+
*/
496
+
static int monitor_route(void)
497
+
{
498
+
unsigned int nr_cpus = bpf_num_possible_cpus();
499
+
const unsigned int nr_keys = 256;
500
+
struct pollfd fds_route, fds_arp;
501
+
__u64 prev[nr_keys][nr_cpus];
502
+
struct sockaddr_nl la, lr;
503
+
__u64 values[nr_cpus];
504
+
struct nlmsghdr *nh;
505
+
int nll, ret = 0;
506
+
int interval = 5;
507
+
__u32 key;
508
+
int i;
509
+
510
+
sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
511
+
if (sock < 0) {
512
+
printf("open netlink socket: %s\n", strerror(errno));
513
+
return -1;
514
+
}
515
+
516
+
fcntl(sock, F_SETFL, O_NONBLOCK);
517
+
memset(&lr, 0, sizeof(lr));
518
+
lr.nl_family = AF_NETLINK;
519
+
lr.nl_groups = RTMGRP_IPV6_ROUTE | RTMGRP_IPV4_ROUTE | RTMGRP_NOTIFY;
520
+
if (bind(sock, (struct sockaddr *)&lr, sizeof(lr)) < 0) {
521
+
printf("bind to netlink: %s\n", strerror(errno));
522
+
ret = -1;
523
+
goto cleanup;
524
+
}
525
+
fds_route.fd = sock;
526
+
fds_route.events = POLL_IN;
527
+
528
+
sock_arp = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
529
+
if (sock_arp < 0) {
530
+
printf("open netlink socket: %s\n", strerror(errno));
531
+
return -1;
532
+
}
533
+
534
+
fcntl(sock_arp, F_SETFL, O_NONBLOCK);
535
+
memset(&la, 0, sizeof(la));
536
+
la.nl_family = AF_NETLINK;
537
+
la.nl_groups = RTMGRP_NEIGH | RTMGRP_NOTIFY;
538
+
if (bind(sock_arp, (struct sockaddr *)&la, sizeof(la)) < 0) {
539
+
printf("bind to netlink: %s\n", strerror(errno));
540
+
ret = -1;
541
+
goto cleanup;
542
+
}
543
+
fds_arp.fd = sock_arp;
544
+
fds_arp.events = POLL_IN;
545
+
546
+
memset(prev, 0, sizeof(prev));
547
+
do {
548
+
signal(SIGINT, close_and_exit);
549
+
signal(SIGTERM, close_and_exit);
550
+
551
+
sleep(interval);
552
+
for (key = 0; key < nr_keys; key++) {
553
+
__u64 sum = 0;
554
+
555
+
assert(bpf_map_lookup_elem(map_fd[1], &key, values) == 0);
556
+
for (i = 0; i < nr_cpus; i++)
557
+
sum += (values[i] - prev[key][i]);
558
+
if (sum)
559
+
printf("proto %u: %10llu pkt/s\n",
560
+
key, sum / interval);
561
+
memcpy(prev[key], values, sizeof(values));
562
+
}
563
+
564
+
memset(buf, 0, sizeof(buf));
565
+
if (poll(&fds_route, 1, 3) == POLL_IN) {
566
+
nll = recv_msg(lr, sock);
567
+
if (nll < 0) {
568
+
printf("recv from netlink: %s\n", strerror(nll));
569
+
ret = -1;
570
+
goto cleanup;
571
+
}
572
+
573
+
nh = (struct nlmsghdr *)buf;
574
+
printf("Routing table updated.\n");
575
+
read_route(nh, nll);
576
+
}
577
+
memset(buf, 0, sizeof(buf));
578
+
if (poll(&fds_arp, 1, 3) == POLL_IN) {
579
+
nll = recv_msg(la, sock_arp);
580
+
if (nll < 0) {
581
+
printf("recv from netlink: %s\n", strerror(nll));
582
+
ret = -1;
583
+
goto cleanup;
584
+
}
585
+
586
+
nh = (struct nlmsghdr *)buf;
587
+
read_arp(nh, nll);
588
+
}
589
+
590
+
} while (1);
591
+
cleanup:
592
+
close(sock);
593
+
return ret;
594
+
}
595
+
596
+
int main(int ac, char **argv)
597
+
{
598
+
char filename[256];
599
+
char **ifname_list;
600
+
int i = 1;
601
+
602
+
snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
603
+
if (ac < 2) {
604
+
printf("usage: %s [-S] Interface name list\n", argv[0]);
605
+
return 1;
606
+
}
607
+
if (!strcmp(argv[1], "-S")) {
608
+
flags = XDP_FLAGS_SKB_MODE;
609
+
total_ifindex = ac - 2;
610
+
ifname_list = (argv + 2);
611
+
} else {
612
+
flags = 0;
613
+
total_ifindex = ac - 1;
614
+
ifname_list = (argv + 1);
615
+
}
616
+
if (load_bpf_file(filename)) {
617
+
printf("%s", bpf_log_buf);
618
+
return 1;
619
+
}
620
+
printf("\n**************loading bpf file*********************\n\n\n");
621
+
if (!prog_fd[0]) {
622
+
printf("load_bpf_file: %s\n", strerror(errno));
623
+
return 1;
624
+
}
625
+
ifindex_list = (int *)malloc(total_ifindex * sizeof(int *));
626
+
for (i = 0; i < total_ifindex; i++) {
627
+
ifindex_list[i] = if_nametoindex(ifname_list[i]);
628
+
if (!ifindex_list[i]) {
629
+
printf("Couldn't translate interface name: %s",
630
+
strerror(errno));
631
+
return 1;
632
+
}
633
+
}
634
+
for (i = 0; i < total_ifindex; i++) {
635
+
if (set_link_xdp_fd(ifindex_list[i], prog_fd[0], flags) < 0) {
636
+
printf("link set xdp fd failed\n");
637
+
int recovery_index = i;
638
+
639
+
for (i = 0; i < recovery_index; i++)
640
+
set_link_xdp_fd(ifindex_list[i], -1, flags);
641
+
642
+
return 1;
643
+
}
644
+
printf("Attached to %d\n", ifindex_list[i]);
645
+
}
646
+
signal(SIGINT, int_exit);
647
+
signal(SIGTERM, int_exit);
648
+
649
+
printf("*******************ROUTE TABLE*************************\n\n\n");
650
+
get_route_table(AF_INET);
651
+
printf("*******************ARP TABLE***************************\n\n\n");
652
+
get_arp_table(AF_INET);
653
+
if (monitor_route() < 0) {
654
+
printf("Error in receiving route update");
655
+
return 1;
656
+
}
657
+
658
+
return 0;
659
+
}