1 // SPDX-License-Identifier: GPL-2.0
2 #include <test_progs.h>
3 #include <network_helpers.h>
4 #include <net/if.h>
5 #include <linux/if_ether.h>
6 #include <linux/if_packet.h>
7 #include <linux/if_link.h>
8 #include <linux/ipv6.h>
9 #include <linux/in6.h>
10 #include <netinet/udp.h>
11 #include <bpf/bpf_endian.h>
12 #include <uapi/linux/netdev.h>
13 #include "test_xdp_do_redirect.skel.h"
14 #include "xdp_dummy.skel.h"
15
16 struct udp_packet {
17 struct ethhdr eth;
18 struct ipv6hdr iph;
19 struct udphdr udp;
20 __u8 payload[64 - sizeof(struct udphdr)
21 - sizeof(struct ethhdr) - sizeof(struct ipv6hdr)];
22 } __packed;
23
24 static struct udp_packet pkt_udp = {
25 .eth.h_proto = __bpf_constant_htons(ETH_P_IPV6),
26 .eth.h_dest = {0x00, 0x11, 0x22, 0x33, 0x44, 0x55},
27 .eth.h_source = {0x66, 0x77, 0x88, 0x99, 0xaa, 0xbb},
28 .iph.version = 6,
29 .iph.nexthdr = IPPROTO_UDP,
30 .iph.payload_len = bpf_htons(sizeof(struct udp_packet)
31 - offsetof(struct udp_packet, udp)),
32 .iph.hop_limit = 2,
33 .iph.saddr.s6_addr16 = {bpf_htons(0xfc00), 0, 0, 0, 0, 0, 0, bpf_htons(1)},
34 .iph.daddr.s6_addr16 = {bpf_htons(0xfc00), 0, 0, 0, 0, 0, 0, bpf_htons(2)},
35 .udp.source = bpf_htons(1),
36 .udp.dest = bpf_htons(1),
37 .udp.len = bpf_htons(sizeof(struct udp_packet)
38 - offsetof(struct udp_packet, udp)),
39 .payload = {0x42}, /* receiver XDP program matches on this */
40 };
41
attach_tc_prog(struct bpf_tc_hook * hook,int fd)42 static int attach_tc_prog(struct bpf_tc_hook *hook, int fd)
43 {
44 DECLARE_LIBBPF_OPTS(bpf_tc_opts, opts, .handle = 1, .priority = 1, .prog_fd = fd);
45 int ret;
46
47 ret = bpf_tc_hook_create(hook);
48 if (!ASSERT_OK(ret, "create tc hook"))
49 return ret;
50
51 ret = bpf_tc_attach(hook, &opts);
52 if (!ASSERT_OK(ret, "bpf_tc_attach")) {
53 bpf_tc_hook_destroy(hook);
54 return ret;
55 }
56
57 return 0;
58 }
59
60 /* The maximum permissible size is: PAGE_SIZE - sizeof(struct xdp_page_head) -
61 * SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) - XDP_PACKET_HEADROOM =
62 * 3408 bytes for 64-byte cacheline and 3216 for 256-byte one.
63 */
64 #if defined(__s390x__)
65 #define MAX_PKT_SIZE 3216
66 #else
67 #define MAX_PKT_SIZE 3408
68 #endif
test_max_pkt_size(int fd)69 static void test_max_pkt_size(int fd)
70 {
71 char data[MAX_PKT_SIZE + 1] = {};
72 int err;
73 DECLARE_LIBBPF_OPTS(bpf_test_run_opts, opts,
74 .data_in = &data,
75 .data_size_in = MAX_PKT_SIZE,
76 .flags = BPF_F_TEST_XDP_LIVE_FRAMES,
77 .repeat = 1,
78 );
79 err = bpf_prog_test_run_opts(fd, &opts);
80 ASSERT_OK(err, "prog_run_max_size");
81
82 opts.data_size_in += 1;
83 err = bpf_prog_test_run_opts(fd, &opts);
84 ASSERT_EQ(err, -EINVAL, "prog_run_too_big");
85 }
86
87 #define NUM_PKTS 10000
test_xdp_do_redirect(void)88 void test_xdp_do_redirect(void)
89 {
90 int err, xdp_prog_fd, tc_prog_fd, ifindex_src, ifindex_dst;
91 char data[sizeof(pkt_udp) + sizeof(__u64)];
92 struct test_xdp_do_redirect *skel = NULL;
93 struct nstoken *nstoken = NULL;
94 struct bpf_link *link;
95 LIBBPF_OPTS(bpf_xdp_query_opts, query_opts);
96 struct xdp_md ctx_in = { .data = sizeof(__u64),
97 .data_end = sizeof(data) };
98 DECLARE_LIBBPF_OPTS(bpf_test_run_opts, opts,
99 .data_in = &data,
100 .data_size_in = sizeof(data),
101 .ctx_in = &ctx_in,
102 .ctx_size_in = sizeof(ctx_in),
103 .flags = BPF_F_TEST_XDP_LIVE_FRAMES,
104 .repeat = NUM_PKTS,
105 .batch_size = 64,
106 );
107 DECLARE_LIBBPF_OPTS(bpf_tc_hook, tc_hook,
108 .attach_point = BPF_TC_INGRESS);
109
110 memcpy(&data[sizeof(__u64)], &pkt_udp, sizeof(pkt_udp));
111 ((__u32 *)data)[0] = 0x42; /* metadata test value */
112 ((__u32 *)data)[1] = 0;
113
114 skel = test_xdp_do_redirect__open();
115 if (!ASSERT_OK_PTR(skel, "skel"))
116 return;
117
118 /* The XDP program we run with bpf_prog_run() will cycle through all
119 * three xmit (PASS/TX/REDIRECT) return codes starting from above, and
120 * ending up with PASS, so we should end up with two packets on the dst
121 * iface and NUM_PKTS-2 in the TC hook. We match the packets on the UDP
122 * payload.
123 */
124 SYS(out, "ip netns add testns");
125 nstoken = open_netns("testns");
126 if (!ASSERT_OK_PTR(nstoken, "setns"))
127 goto out;
128
129 SYS(out, "ip link add veth_src type veth peer name veth_dst");
130 SYS(out, "ip link set dev veth_src address 00:11:22:33:44:55");
131 SYS(out, "ip link set dev veth_dst address 66:77:88:99:aa:bb");
132 SYS(out, "ip link set dev veth_src up");
133 SYS(out, "ip link set dev veth_dst up");
134 SYS(out, "ip addr add dev veth_src fc00::1/64");
135 SYS(out, "ip addr add dev veth_dst fc00::2/64");
136 SYS(out, "ip neigh add fc00::2 dev veth_src lladdr 66:77:88:99:aa:bb");
137
138 /* We enable forwarding in the test namespace because that will cause
139 * the packets that go through the kernel stack (with XDP_PASS) to be
140 * forwarded back out the same interface (because of the packet dst
141 * combined with the interface addresses). When this happens, the
142 * regular forwarding path will end up going through the same
143 * veth_xdp_xmit() call as the XDP_REDIRECT code, which can cause a
144 * deadlock if it happens on the same CPU. There's a local_bh_disable()
145 * in the test_run code to prevent this, but an earlier version of the
146 * code didn't have this, so we keep the test behaviour to make sure the
147 * bug doesn't resurface.
148 */
149 SYS(out, "sysctl -qw net.ipv6.conf.all.forwarding=1");
150
151 ifindex_src = if_nametoindex("veth_src");
152 ifindex_dst = if_nametoindex("veth_dst");
153 if (!ASSERT_NEQ(ifindex_src, 0, "ifindex_src") ||
154 !ASSERT_NEQ(ifindex_dst, 0, "ifindex_dst"))
155 goto out;
156
157 /* Check xdp features supported by veth driver */
158 err = bpf_xdp_query(ifindex_src, XDP_FLAGS_DRV_MODE, &query_opts);
159 if (!ASSERT_OK(err, "veth_src bpf_xdp_query"))
160 goto out;
161
162 if (!ASSERT_EQ(query_opts.feature_flags,
163 NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
164 NETDEV_XDP_ACT_RX_SG,
165 "veth_src query_opts.feature_flags"))
166 goto out;
167
168 err = bpf_xdp_query(ifindex_dst, XDP_FLAGS_DRV_MODE, &query_opts);
169 if (!ASSERT_OK(err, "veth_dst bpf_xdp_query"))
170 goto out;
171
172 if (!ASSERT_EQ(query_opts.feature_flags,
173 NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
174 NETDEV_XDP_ACT_RX_SG,
175 "veth_dst query_opts.feature_flags"))
176 goto out;
177
178 /* Enable GRO */
179 SYS(out, "ethtool -K veth_src gro on");
180 SYS(out, "ethtool -K veth_dst gro on");
181
182 err = bpf_xdp_query(ifindex_src, XDP_FLAGS_DRV_MODE, &query_opts);
183 if (!ASSERT_OK(err, "veth_src bpf_xdp_query gro on"))
184 goto out;
185
186 if (!ASSERT_EQ(query_opts.feature_flags,
187 NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
188 NETDEV_XDP_ACT_NDO_XMIT | NETDEV_XDP_ACT_RX_SG |
189 NETDEV_XDP_ACT_NDO_XMIT_SG,
190 "veth_src query_opts.feature_flags gro on"))
191 goto out;
192
193 err = bpf_xdp_query(ifindex_dst, XDP_FLAGS_DRV_MODE, &query_opts);
194 if (!ASSERT_OK(err, "veth_dst bpf_xdp_query gro on"))
195 goto out;
196
197 if (!ASSERT_EQ(query_opts.feature_flags,
198 NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
199 NETDEV_XDP_ACT_NDO_XMIT | NETDEV_XDP_ACT_RX_SG |
200 NETDEV_XDP_ACT_NDO_XMIT_SG,
201 "veth_dst query_opts.feature_flags gro on"))
202 goto out;
203
204 memcpy(skel->rodata->expect_dst, &pkt_udp.eth.h_dest, ETH_ALEN);
205 skel->rodata->ifindex_out = ifindex_src; /* redirect back to the same iface */
206 skel->rodata->ifindex_in = ifindex_src;
207 ctx_in.ingress_ifindex = ifindex_src;
208 tc_hook.ifindex = ifindex_src;
209
210 if (!ASSERT_OK(test_xdp_do_redirect__load(skel), "load"))
211 goto out;
212
213 link = bpf_program__attach_xdp(skel->progs.xdp_count_pkts, ifindex_dst);
214 if (!ASSERT_OK_PTR(link, "prog_attach"))
215 goto out;
216 skel->links.xdp_count_pkts = link;
217
218 tc_prog_fd = bpf_program__fd(skel->progs.tc_count_pkts);
219 if (attach_tc_prog(&tc_hook, tc_prog_fd))
220 goto out;
221
222 xdp_prog_fd = bpf_program__fd(skel->progs.xdp_redirect);
223 err = bpf_prog_test_run_opts(xdp_prog_fd, &opts);
224 if (!ASSERT_OK(err, "prog_run"))
225 goto out_tc;
226
227 /* wait for the packets to be flushed */
228 kern_sync_rcu();
229
230 /* There will be one packet sent through XDP_REDIRECT and one through
231 * XDP_TX; these will show up on the XDP counting program, while the
232 * rest will be counted at the TC ingress hook (and the counting program
233 * resets the packet payload so they don't get counted twice even though
234 * they are re-xmited out the veth device
235 */
236 ASSERT_EQ(skel->bss->pkts_seen_xdp, 2, "pkt_count_xdp");
237 ASSERT_EQ(skel->bss->pkts_seen_zero, 2, "pkt_count_zero");
238 ASSERT_EQ(skel->bss->pkts_seen_tc, NUM_PKTS - 2, "pkt_count_tc");
239
240 test_max_pkt_size(bpf_program__fd(skel->progs.xdp_count_pkts));
241
242 out_tc:
243 bpf_tc_hook_destroy(&tc_hook);
244 out:
245 if (nstoken)
246 close_netns(nstoken);
247 SYS_NOFAIL("ip netns del testns");
248 test_xdp_do_redirect__destroy(skel);
249 }
250
251 #define NS_NB 3
252 #define NS0 "NS0"
253 #define NS1 "NS1"
254 #define NS2 "NS2"
255 #define IPV4_NETWORK "10.1.1"
256 #define VETH1_INDEX 111
257 #define VETH2_INDEX 222
258
259 struct test_data {
260 struct netns_obj *ns[NS_NB];
261 u32 xdp_flags;
262 };
263
cleanup(struct test_data * data)264 static void cleanup(struct test_data *data)
265 {
266 int i;
267
268 for (i = 0; i < NS_NB; i++)
269 netns_free(data->ns[i]);
270 }
271
272 /**
273 * ping_setup -
274 * Create two veth peers and forward packets in-between using XDP
275 *
276 * ------------ ------------
277 * | NS1 | | NS2 |
278 * | veth0 | | veth0 |
279 * | 10.1.1.1 | | 10.1.1.2 |
280 * -----|------ ------|-----
281 * | |
282 * | |
283 * -----|-----------------------|-------
284 * | veth1 veth2 |
285 * | (id:111) (id:222) |
286 * | | | |
287 * | ----- xdp forwarding ----- |
288 * | |
289 * | NS0 |
290 * -------------------------------------
291 */
ping_setup(struct test_data * data)292 static int ping_setup(struct test_data *data)
293 {
294 int i;
295
296 data->ns[0] = netns_new(NS0, false);
297 if (!ASSERT_OK_PTR(data->ns[0], "create ns"))
298 return -1;
299
300 for (i = 1; i < NS_NB; i++) {
301 char ns_name[4] = {};
302
303 snprintf(ns_name, 4, "NS%d", i);
304 data->ns[i] = netns_new(ns_name, false);
305 if (!ASSERT_OK_PTR(data->ns[i], "create ns"))
306 goto fail;
307
308 SYS(fail,
309 "ip -n %s link add veth%d index %d%d%d type veth peer name veth0 netns %s",
310 NS0, i, i, i, i, ns_name);
311 SYS(fail, "ip -n %s link set veth%d up", NS0, i);
312
313 SYS(fail, "ip -n %s addr add %s.%d/24 dev veth0", ns_name, IPV4_NETWORK, i);
314 SYS(fail, "ip -n %s link set veth0 up", ns_name);
315 }
316
317 return 0;
318
319 fail:
320 cleanup(data);
321 return -1;
322 }
323
ping_test(struct test_data * data)324 static void ping_test(struct test_data *data)
325 {
326 struct test_xdp_do_redirect *skel = NULL;
327 struct xdp_dummy *skel_dummy = NULL;
328 struct nstoken *nstoken = NULL;
329 int i, ret;
330
331 skel_dummy = xdp_dummy__open_and_load();
332 if (!ASSERT_OK_PTR(skel_dummy, "open and load xdp_dummy skeleton"))
333 goto close;
334
335 for (i = 1; i < NS_NB; i++) {
336 char ns_name[4] = {};
337
338 snprintf(ns_name, 4, "NS%d", i);
339 nstoken = open_netns(ns_name);
340 if (!ASSERT_OK_PTR(nstoken, "open ns"))
341 goto close;
342
343 ret = bpf_xdp_attach(if_nametoindex("veth0"),
344 bpf_program__fd(skel_dummy->progs.xdp_dummy_prog),
345 data->xdp_flags, NULL);
346 if (!ASSERT_GE(ret, 0, "bpf_xdp_attach dummy_prog"))
347 goto close;
348
349 close_netns(nstoken);
350 nstoken = NULL;
351 }
352
353 skel = test_xdp_do_redirect__open_and_load();
354 if (!ASSERT_OK_PTR(skel, "open and load skeleton"))
355 goto close;
356
357 nstoken = open_netns(NS0);
358 if (!ASSERT_OK_PTR(nstoken, "open NS0"))
359 goto close;
360
361 ret = bpf_xdp_attach(VETH2_INDEX,
362 bpf_program__fd(skel->progs.xdp_redirect_to_111),
363 data->xdp_flags, NULL);
364 if (!ASSERT_GE(ret, 0, "bpf_xdp_attach"))
365 goto close;
366
367 ret = bpf_xdp_attach(VETH1_INDEX,
368 bpf_program__fd(skel->progs.xdp_redirect_to_222),
369 data->xdp_flags, NULL);
370 if (!ASSERT_GE(ret, 0, "bpf_xdp_attach"))
371 goto close;
372
373 close_netns(nstoken);
374 nstoken = NULL;
375
376 nstoken = open_netns(NS1);
377 if (!ASSERT_OK_PTR(nstoken, "open NS1"))
378 goto close;
379
380 SYS(close, "ping -c 1 %s.2 > /dev/null", IPV4_NETWORK);
381
382 close:
383 close_netns(nstoken);
384 xdp_dummy__destroy(skel_dummy);
385 test_xdp_do_redirect__destroy(skel);
386 }
387
388
xdp_redirect_ping(u32 xdp_flags)389 static void xdp_redirect_ping(u32 xdp_flags)
390 {
391 struct test_data data = {};
392
393 if (ping_setup(&data) < 0)
394 return;
395
396 data.xdp_flags = xdp_flags;
397 ping_test(&data);
398 cleanup(&data);
399 }
400
test_xdp_index_redirect(void)401 void test_xdp_index_redirect(void)
402 {
403 if (test__start_subtest("noflag"))
404 xdp_redirect_ping(0);
405
406 if (test__start_subtest("drvflag"))
407 xdp_redirect_ping(XDP_FLAGS_DRV_MODE);
408
409 if (test__start_subtest("skbflag"))
410 xdp_redirect_ping(XDP_FLAGS_SKB_MODE);
411 }
412
413