1 // SPDX-License-Identifier: GPL-2.0
2 
3 /**
4  * Test XDP bonding support
5  *
6  * Sets up two bonded veth pairs between two fresh namespaces
7  * and verifies that XDP_TX program loaded on a bond device
8  * are correctly loaded onto the slave devices and XDP_TX'd
9  * packets are balanced using bonding.
10  */
11 
12 #define _GNU_SOURCE
13 #include <sched.h>
14 #include <net/if.h>
15 #include <linux/if_link.h>
16 #include "test_progs.h"
17 #include "network_helpers.h"
18 #include <linux/if_bonding.h>
19 #include <linux/limits.h>
20 #include <linux/udp.h>
21 
22 #include "xdp_dummy.skel.h"
23 #include "xdp_redirect_multi_kern.skel.h"
24 #include "xdp_tx.skel.h"
25 
26 #define BOND1_MAC {0x00, 0x11, 0x22, 0x33, 0x44, 0x55}
27 #define BOND1_MAC_STR "00:11:22:33:44:55"
28 #define BOND2_MAC {0x00, 0x22, 0x33, 0x44, 0x55, 0x66}
29 #define BOND2_MAC_STR "00:22:33:44:55:66"
30 #define NPACKETS 100
31 
32 static int root_netns_fd = -1;
33 
34 static void restore_root_netns(void)
35 {
36 	ASSERT_OK(setns(root_netns_fd, CLONE_NEWNET), "restore_root_netns");
37 }
38 
39 static int setns_by_name(char *name)
40 {
41 	int nsfd, err;
42 	char nspath[PATH_MAX];
43 
44 	snprintf(nspath, sizeof(nspath), "%s/%s", "/var/run/netns", name);
45 	nsfd = open(nspath, O_RDONLY | O_CLOEXEC);
46 	if (nsfd < 0)
47 		return -1;
48 
49 	err = setns(nsfd, CLONE_NEWNET);
50 	close(nsfd);
51 	return err;
52 }
53 
54 static int get_rx_packets(const char *iface)
55 {
56 	FILE *f;
57 	char line[512];
58 	int iface_len = strlen(iface);
59 
60 	f = fopen("/proc/net/dev", "r");
61 	if (!f)
62 		return -1;
63 
64 	while (fgets(line, sizeof(line), f)) {
65 		char *p = line;
66 
67 		while (*p == ' ')
68 			p++; /* skip whitespace */
69 		if (!strncmp(p, iface, iface_len)) {
70 			p += iface_len;
71 			if (*p++ != ':')
72 				continue;
73 			while (*p == ' ')
74 				p++; /* skip whitespace */
75 			while (*p && *p != ' ')
76 				p++; /* skip rx bytes */
77 			while (*p == ' ')
78 				p++; /* skip whitespace */
79 			fclose(f);
80 			return atoi(p);
81 		}
82 	}
83 	fclose(f);
84 	return -1;
85 }
86 
87 #define MAX_BPF_LINKS 8
88 
89 struct skeletons {
90 	struct xdp_dummy *xdp_dummy;
91 	struct xdp_tx *xdp_tx;
92 	struct xdp_redirect_multi_kern *xdp_redirect_multi_kern;
93 
94 	int nlinks;
95 	struct bpf_link *links[MAX_BPF_LINKS];
96 };
97 
98 static int xdp_attach(struct skeletons *skeletons, struct bpf_program *prog, char *iface)
99 {
100 	struct bpf_link *link;
101 	int ifindex;
102 
103 	ifindex = if_nametoindex(iface);
104 	if (!ASSERT_GT(ifindex, 0, "get ifindex"))
105 		return -1;
106 
107 	if (!ASSERT_LE(skeletons->nlinks+1, MAX_BPF_LINKS, "too many XDP programs attached"))
108 		return -1;
109 
110 	link = bpf_program__attach_xdp(prog, ifindex);
111 	if (!ASSERT_OK_PTR(link, "attach xdp program"))
112 		return -1;
113 
114 	skeletons->links[skeletons->nlinks++] = link;
115 	return 0;
116 }
117 
118 enum {
119 	BOND_ONE_NO_ATTACH = 0,
120 	BOND_BOTH_AND_ATTACH,
121 };
122 
123 static const char * const mode_names[] = {
124 	[BOND_MODE_ROUNDROBIN]   = "balance-rr",
125 	[BOND_MODE_ACTIVEBACKUP] = "active-backup",
126 	[BOND_MODE_XOR]          = "balance-xor",
127 	[BOND_MODE_BROADCAST]    = "broadcast",
128 	[BOND_MODE_8023AD]       = "802.3ad",
129 	[BOND_MODE_TLB]          = "balance-tlb",
130 	[BOND_MODE_ALB]          = "balance-alb",
131 };
132 
133 static const char * const xmit_policy_names[] = {
134 	[BOND_XMIT_POLICY_LAYER2]       = "layer2",
135 	[BOND_XMIT_POLICY_LAYER34]      = "layer3+4",
136 	[BOND_XMIT_POLICY_LAYER23]      = "layer2+3",
137 	[BOND_XMIT_POLICY_ENCAP23]      = "encap2+3",
138 	[BOND_XMIT_POLICY_ENCAP34]      = "encap3+4",
139 };
140 
141 static int bonding_setup(struct skeletons *skeletons, int mode, int xmit_policy,
142 			 int bond_both_attach)
143 {
144 #define SYS(fmt, ...)						\
145 	({							\
146 		char cmd[1024];					\
147 		snprintf(cmd, sizeof(cmd), fmt, ##__VA_ARGS__);	\
148 		if (!ASSERT_OK(system(cmd), cmd))		\
149 			return -1;				\
150 	})
151 
152 	SYS("ip netns add ns_dst");
153 	SYS("ip link add veth1_1 type veth peer name veth2_1 netns ns_dst");
154 	SYS("ip link add veth1_2 type veth peer name veth2_2 netns ns_dst");
155 
156 	SYS("ip link add bond1 type bond mode %s xmit_hash_policy %s",
157 	    mode_names[mode], xmit_policy_names[xmit_policy]);
158 	SYS("ip link set bond1 up address " BOND1_MAC_STR " addrgenmode none");
159 	SYS("ip -netns ns_dst link add bond2 type bond mode %s xmit_hash_policy %s",
160 	    mode_names[mode], xmit_policy_names[xmit_policy]);
161 	SYS("ip -netns ns_dst link set bond2 up address " BOND2_MAC_STR " addrgenmode none");
162 
163 	SYS("ip link set veth1_1 master bond1");
164 	if (bond_both_attach == BOND_BOTH_AND_ATTACH) {
165 		SYS("ip link set veth1_2 master bond1");
166 	} else {
167 		SYS("ip link set veth1_2 up addrgenmode none");
168 
169 		if (xdp_attach(skeletons, skeletons->xdp_dummy->progs.xdp_dummy_prog, "veth1_2"))
170 			return -1;
171 	}
172 
173 	SYS("ip -netns ns_dst link set veth2_1 master bond2");
174 
175 	if (bond_both_attach == BOND_BOTH_AND_ATTACH)
176 		SYS("ip -netns ns_dst link set veth2_2 master bond2");
177 	else
178 		SYS("ip -netns ns_dst link set veth2_2 up addrgenmode none");
179 
180 	/* Load a dummy program on sending side as with veth peer needs to have a
181 	 * XDP program loaded as well.
182 	 */
183 	if (xdp_attach(skeletons, skeletons->xdp_dummy->progs.xdp_dummy_prog, "bond1"))
184 		return -1;
185 
186 	if (bond_both_attach == BOND_BOTH_AND_ATTACH) {
187 		if (!ASSERT_OK(setns_by_name("ns_dst"), "set netns to ns_dst"))
188 			return -1;
189 
190 		if (xdp_attach(skeletons, skeletons->xdp_tx->progs.xdp_tx, "bond2"))
191 			return -1;
192 
193 		restore_root_netns();
194 	}
195 
196 	return 0;
197 
198 #undef SYS
199 }
200 
201 static void bonding_cleanup(struct skeletons *skeletons)
202 {
203 	restore_root_netns();
204 	while (skeletons->nlinks) {
205 		skeletons->nlinks--;
206 		bpf_link__destroy(skeletons->links[skeletons->nlinks]);
207 	}
208 	ASSERT_OK(system("ip link delete bond1"), "delete bond1");
209 	ASSERT_OK(system("ip link delete veth1_1"), "delete veth1_1");
210 	ASSERT_OK(system("ip link delete veth1_2"), "delete veth1_2");
211 	ASSERT_OK(system("ip netns delete ns_dst"), "delete ns_dst");
212 }
213 
214 static int send_udp_packets(int vary_dst_ip)
215 {
216 	struct ethhdr eh = {
217 		.h_source = BOND1_MAC,
218 		.h_dest = BOND2_MAC,
219 		.h_proto = htons(ETH_P_IP),
220 	};
221 	struct iphdr iph = {};
222 	struct udphdr uh = {};
223 	uint8_t buf[128];
224 	int i, s = -1;
225 	int ifindex;
226 
227 	s = socket(AF_PACKET, SOCK_RAW, IPPROTO_RAW);
228 	if (!ASSERT_GE(s, 0, "socket"))
229 		goto err;
230 
231 	ifindex = if_nametoindex("bond1");
232 	if (!ASSERT_GT(ifindex, 0, "get bond1 ifindex"))
233 		goto err;
234 
235 	iph.ihl = 5;
236 	iph.version = 4;
237 	iph.tos = 16;
238 	iph.id = 1;
239 	iph.ttl = 64;
240 	iph.protocol = IPPROTO_UDP;
241 	iph.saddr = 1;
242 	iph.daddr = 2;
243 	iph.tot_len = htons(sizeof(buf) - ETH_HLEN);
244 	iph.check = 0;
245 
246 	for (i = 1; i <= NPACKETS; i++) {
247 		int n;
248 		struct sockaddr_ll saddr_ll = {
249 			.sll_ifindex = ifindex,
250 			.sll_halen = ETH_ALEN,
251 			.sll_addr = BOND2_MAC,
252 		};
253 
254 		/* vary the UDP destination port for even distribution with roundrobin/xor modes */
255 		uh.dest++;
256 
257 		if (vary_dst_ip)
258 			iph.daddr++;
259 
260 		/* construct a packet */
261 		memcpy(buf, &eh, sizeof(eh));
262 		memcpy(buf + sizeof(eh), &iph, sizeof(iph));
263 		memcpy(buf + sizeof(eh) + sizeof(iph), &uh, sizeof(uh));
264 
265 		n = sendto(s, buf, sizeof(buf), 0, (struct sockaddr *)&saddr_ll, sizeof(saddr_ll));
266 		if (!ASSERT_EQ(n, sizeof(buf), "sendto"))
267 			goto err;
268 	}
269 
270 	return 0;
271 
272 err:
273 	if (s >= 0)
274 		close(s);
275 	return -1;
276 }
277 
278 static void test_xdp_bonding_with_mode(struct skeletons *skeletons, int mode, int xmit_policy)
279 {
280 	int bond1_rx;
281 
282 	if (bonding_setup(skeletons, mode, xmit_policy, BOND_BOTH_AND_ATTACH))
283 		goto out;
284 
285 	if (send_udp_packets(xmit_policy != BOND_XMIT_POLICY_LAYER34))
286 		goto out;
287 
288 	bond1_rx = get_rx_packets("bond1");
289 	ASSERT_EQ(bond1_rx, NPACKETS, "expected more received packets");
290 
291 	switch (mode) {
292 	case BOND_MODE_ROUNDROBIN:
293 	case BOND_MODE_XOR: {
294 		int veth1_rx = get_rx_packets("veth1_1");
295 		int veth2_rx = get_rx_packets("veth1_2");
296 		int diff = abs(veth1_rx - veth2_rx);
297 
298 		ASSERT_GE(veth1_rx + veth2_rx, NPACKETS, "expected more packets");
299 
300 		switch (xmit_policy) {
301 		case BOND_XMIT_POLICY_LAYER2:
302 			ASSERT_GE(diff, NPACKETS,
303 				  "expected packets on only one of the interfaces");
304 			break;
305 		case BOND_XMIT_POLICY_LAYER23:
306 		case BOND_XMIT_POLICY_LAYER34:
307 			ASSERT_LT(diff, NPACKETS/2,
308 				  "expected even distribution of packets");
309 			break;
310 		default:
311 			PRINT_FAIL("Unimplemented xmit_policy=%d\n", xmit_policy);
312 			break;
313 		}
314 		break;
315 	}
316 	case BOND_MODE_ACTIVEBACKUP: {
317 		int veth1_rx = get_rx_packets("veth1_1");
318 		int veth2_rx = get_rx_packets("veth1_2");
319 		int diff = abs(veth1_rx - veth2_rx);
320 
321 		ASSERT_GE(diff, NPACKETS,
322 			  "expected packets on only one of the interfaces");
323 		break;
324 	}
325 	default:
326 		PRINT_FAIL("Unimplemented xmit_policy=%d\n", xmit_policy);
327 		break;
328 	}
329 
330 out:
331 	bonding_cleanup(skeletons);
332 }
333 
334 /* Test the broadcast redirection using xdp_redirect_map_multi_prog and adding
335  * all the interfaces to it and checking that broadcasting won't send the packet
336  * to neither the ingress bond device (bond2) or its slave (veth2_1).
337  */
338 static void test_xdp_bonding_redirect_multi(struct skeletons *skeletons)
339 {
340 	static const char * const ifaces[] = {"bond2", "veth2_1", "veth2_2"};
341 	int veth1_1_rx, veth1_2_rx;
342 	int err;
343 
344 	if (bonding_setup(skeletons, BOND_MODE_ROUNDROBIN, BOND_XMIT_POLICY_LAYER23,
345 			  BOND_ONE_NO_ATTACH))
346 		goto out;
347 
348 
349 	if (!ASSERT_OK(setns_by_name("ns_dst"), "could not set netns to ns_dst"))
350 		goto out;
351 
352 	/* populate the devmap with the relevant interfaces */
353 	for (int i = 0; i < ARRAY_SIZE(ifaces); i++) {
354 		int ifindex = if_nametoindex(ifaces[i]);
355 		int map_fd = bpf_map__fd(skeletons->xdp_redirect_multi_kern->maps.map_all);
356 
357 		if (!ASSERT_GT(ifindex, 0, "could not get interface index"))
358 			goto out;
359 
360 		err = bpf_map_update_elem(map_fd, &ifindex, &ifindex, 0);
361 		if (!ASSERT_OK(err, "add interface to map_all"))
362 			goto out;
363 	}
364 
365 	if (xdp_attach(skeletons,
366 		       skeletons->xdp_redirect_multi_kern->progs.xdp_redirect_map_multi_prog,
367 		       "bond2"))
368 		goto out;
369 
370 	restore_root_netns();
371 
372 	if (send_udp_packets(BOND_MODE_ROUNDROBIN))
373 		goto out;
374 
375 	veth1_1_rx = get_rx_packets("veth1_1");
376 	veth1_2_rx = get_rx_packets("veth1_2");
377 
378 	ASSERT_EQ(veth1_1_rx, 0, "expected no packets on veth1_1");
379 	ASSERT_GE(veth1_2_rx, NPACKETS, "expected packets on veth1_2");
380 
381 out:
382 	restore_root_netns();
383 	bonding_cleanup(skeletons);
384 }
385 
386 /* Test that XDP programs cannot be attached to both the bond master and slaves simultaneously */
387 static void test_xdp_bonding_attach(struct skeletons *skeletons)
388 {
389 	struct bpf_link *link = NULL;
390 	struct bpf_link *link2 = NULL;
391 	int veth, bond, err;
392 
393 	if (!ASSERT_OK(system("ip link add veth type veth"), "add veth"))
394 		goto out;
395 	if (!ASSERT_OK(system("ip link add bond type bond"), "add bond"))
396 		goto out;
397 
398 	veth = if_nametoindex("veth");
399 	if (!ASSERT_GE(veth, 0, "if_nametoindex veth"))
400 		goto out;
401 	bond = if_nametoindex("bond");
402 	if (!ASSERT_GE(bond, 0, "if_nametoindex bond"))
403 		goto out;
404 
405 	/* enslaving with a XDP program loaded is allowed */
406 	link = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, veth);
407 	if (!ASSERT_OK_PTR(link, "attach program to veth"))
408 		goto out;
409 
410 	err = system("ip link set veth master bond");
411 	if (!ASSERT_OK(err, "set veth master"))
412 		goto out;
413 
414 	bpf_link__destroy(link);
415 	link = NULL;
416 
417 	/* attaching to slave when master has no program is allowed */
418 	link = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, veth);
419 	if (!ASSERT_OK_PTR(link, "attach program to slave when enslaved"))
420 		goto out;
421 
422 	/* attaching to master not allowed when slave has program loaded */
423 	link2 = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, bond);
424 	if (!ASSERT_ERR_PTR(link2, "attach program to master when slave has program"))
425 		goto out;
426 
427 	bpf_link__destroy(link);
428 	link = NULL;
429 
430 	/* attaching XDP program to master allowed when slave has no program */
431 	link = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, bond);
432 	if (!ASSERT_OK_PTR(link, "attach program to master"))
433 		goto out;
434 
435 	/* attaching to slave not allowed when master has program loaded */
436 	link2 = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, veth);
437 	if (!ASSERT_ERR_PTR(link2, "attach program to slave when master has program"))
438 		goto out;
439 
440 	bpf_link__destroy(link);
441 	link = NULL;
442 
443 	/* test program unwinding with a non-XDP slave */
444 	if (!ASSERT_OK(system("ip link add vxlan type vxlan id 1 remote 1.2.3.4 dstport 0 dev lo"),
445 		       "add vxlan"))
446 		goto out;
447 
448 	err = system("ip link set vxlan master bond");
449 	if (!ASSERT_OK(err, "set vxlan master"))
450 		goto out;
451 
452 	/* attaching not allowed when one slave does not support XDP */
453 	link = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, bond);
454 	if (!ASSERT_ERR_PTR(link, "attach program to master when slave does not support XDP"))
455 		goto out;
456 
457 out:
458 	bpf_link__destroy(link);
459 	bpf_link__destroy(link2);
460 
461 	system("ip link del veth");
462 	system("ip link del bond");
463 	system("ip link del vxlan");
464 }
465 
466 /* Test with nested bonding devices to catch issue with negative jump label count */
467 static void test_xdp_bonding_nested(struct skeletons *skeletons)
468 {
469 	struct bpf_link *link = NULL;
470 	int bond, err;
471 
472 	if (!ASSERT_OK(system("ip link add bond type bond"), "add bond"))
473 		goto out;
474 
475 	bond = if_nametoindex("bond");
476 	if (!ASSERT_GE(bond, 0, "if_nametoindex bond"))
477 		goto out;
478 
479 	if (!ASSERT_OK(system("ip link add bond_nest1 type bond"), "add bond_nest1"))
480 		goto out;
481 
482 	err = system("ip link set bond_nest1 master bond");
483 	if (!ASSERT_OK(err, "set bond_nest1 master"))
484 		goto out;
485 
486 	if (!ASSERT_OK(system("ip link add bond_nest2 type bond"), "add bond_nest1"))
487 		goto out;
488 
489 	err = system("ip link set bond_nest2 master bond_nest1");
490 	if (!ASSERT_OK(err, "set bond_nest2 master"))
491 		goto out;
492 
493 	link = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, bond);
494 	ASSERT_OK_PTR(link, "attach program to master");
495 
496 out:
497 	bpf_link__destroy(link);
498 	system("ip link del bond");
499 	system("ip link del bond_nest1");
500 	system("ip link del bond_nest2");
501 }
502 
503 static int libbpf_debug_print(enum libbpf_print_level level,
504 			      const char *format, va_list args)
505 {
506 	if (level != LIBBPF_WARN)
507 		vprintf(format, args);
508 	return 0;
509 }
510 
511 struct bond_test_case {
512 	char *name;
513 	int mode;
514 	int xmit_policy;
515 };
516 
517 static struct bond_test_case bond_test_cases[] = {
518 	{ "xdp_bonding_roundrobin", BOND_MODE_ROUNDROBIN, BOND_XMIT_POLICY_LAYER23, },
519 	{ "xdp_bonding_activebackup", BOND_MODE_ACTIVEBACKUP, BOND_XMIT_POLICY_LAYER23 },
520 
521 	{ "xdp_bonding_xor_layer2", BOND_MODE_XOR, BOND_XMIT_POLICY_LAYER2, },
522 	{ "xdp_bonding_xor_layer23", BOND_MODE_XOR, BOND_XMIT_POLICY_LAYER23, },
523 	{ "xdp_bonding_xor_layer34", BOND_MODE_XOR, BOND_XMIT_POLICY_LAYER34, },
524 };
525 
526 void serial_test_xdp_bonding(void)
527 {
528 	libbpf_print_fn_t old_print_fn;
529 	struct skeletons skeletons = {};
530 	int i;
531 
532 	old_print_fn = libbpf_set_print(libbpf_debug_print);
533 
534 	root_netns_fd = open("/proc/self/ns/net", O_RDONLY);
535 	if (!ASSERT_GE(root_netns_fd, 0, "open /proc/self/ns/net"))
536 		goto out;
537 
538 	skeletons.xdp_dummy = xdp_dummy__open_and_load();
539 	if (!ASSERT_OK_PTR(skeletons.xdp_dummy, "xdp_dummy__open_and_load"))
540 		goto out;
541 
542 	skeletons.xdp_tx = xdp_tx__open_and_load();
543 	if (!ASSERT_OK_PTR(skeletons.xdp_tx, "xdp_tx__open_and_load"))
544 		goto out;
545 
546 	skeletons.xdp_redirect_multi_kern = xdp_redirect_multi_kern__open_and_load();
547 	if (!ASSERT_OK_PTR(skeletons.xdp_redirect_multi_kern,
548 			   "xdp_redirect_multi_kern__open_and_load"))
549 		goto out;
550 
551 	if (test__start_subtest("xdp_bonding_attach"))
552 		test_xdp_bonding_attach(&skeletons);
553 
554 	if (test__start_subtest("xdp_bonding_nested"))
555 		test_xdp_bonding_nested(&skeletons);
556 
557 	for (i = 0; i < ARRAY_SIZE(bond_test_cases); i++) {
558 		struct bond_test_case *test_case = &bond_test_cases[i];
559 
560 		if (test__start_subtest(test_case->name))
561 			test_xdp_bonding_with_mode(
562 				&skeletons,
563 				test_case->mode,
564 				test_case->xmit_policy);
565 	}
566 
567 	if (test__start_subtest("xdp_bonding_redirect_multi"))
568 		test_xdp_bonding_redirect_multi(&skeletons);
569 
570 out:
571 	xdp_dummy__destroy(skeletons.xdp_dummy);
572 	xdp_tx__destroy(skeletons.xdp_tx);
573 	xdp_redirect_multi_kern__destroy(skeletons.xdp_redirect_multi_kern);
574 
575 	libbpf_set_print(old_print_fn);
576 	if (root_netns_fd >= 0)
577 		close(root_netns_fd);
578 }
579