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 	SYS(fail, "ip netns add ns_dst");
145 	SYS(fail, "ip link add veth1_1 type veth peer name veth2_1 netns ns_dst");
146 	SYS(fail, "ip link add veth1_2 type veth peer name veth2_2 netns ns_dst");
147 
148 	SYS(fail, "ip link add bond1 type bond mode %s xmit_hash_policy %s",
149 	    mode_names[mode], xmit_policy_names[xmit_policy]);
150 	SYS(fail, "ip link set bond1 up address " BOND1_MAC_STR " addrgenmode none");
151 	SYS(fail, "ip -netns ns_dst link add bond2 type bond mode %s xmit_hash_policy %s",
152 	    mode_names[mode], xmit_policy_names[xmit_policy]);
153 	SYS(fail, "ip -netns ns_dst link set bond2 up address " BOND2_MAC_STR " addrgenmode none");
154 
155 	SYS(fail, "ip link set veth1_1 master bond1");
156 	if (bond_both_attach == BOND_BOTH_AND_ATTACH) {
157 		SYS(fail, "ip link set veth1_2 master bond1");
158 	} else {
159 		SYS(fail, "ip link set veth1_2 up addrgenmode none");
160 
161 		if (xdp_attach(skeletons, skeletons->xdp_dummy->progs.xdp_dummy_prog, "veth1_2"))
162 			return -1;
163 	}
164 
165 	SYS(fail, "ip -netns ns_dst link set veth2_1 master bond2");
166 
167 	if (bond_both_attach == BOND_BOTH_AND_ATTACH)
168 		SYS(fail, "ip -netns ns_dst link set veth2_2 master bond2");
169 	else
170 		SYS(fail, "ip -netns ns_dst link set veth2_2 up addrgenmode none");
171 
172 	/* Load a dummy program on sending side as with veth peer needs to have a
173 	 * XDP program loaded as well.
174 	 */
175 	if (xdp_attach(skeletons, skeletons->xdp_dummy->progs.xdp_dummy_prog, "bond1"))
176 		return -1;
177 
178 	if (bond_both_attach == BOND_BOTH_AND_ATTACH) {
179 		if (!ASSERT_OK(setns_by_name("ns_dst"), "set netns to ns_dst"))
180 			return -1;
181 
182 		if (xdp_attach(skeletons, skeletons->xdp_tx->progs.xdp_tx, "bond2"))
183 			return -1;
184 
185 		restore_root_netns();
186 	}
187 
188 	return 0;
189 fail:
190 	return -1;
191 }
192 
193 static void bonding_cleanup(struct skeletons *skeletons)
194 {
195 	restore_root_netns();
196 	while (skeletons->nlinks) {
197 		skeletons->nlinks--;
198 		bpf_link__destroy(skeletons->links[skeletons->nlinks]);
199 	}
200 	ASSERT_OK(system("ip link delete bond1"), "delete bond1");
201 	ASSERT_OK(system("ip link delete veth1_1"), "delete veth1_1");
202 	ASSERT_OK(system("ip link delete veth1_2"), "delete veth1_2");
203 	ASSERT_OK(system("ip netns delete ns_dst"), "delete ns_dst");
204 }
205 
206 static int send_udp_packets(int vary_dst_ip)
207 {
208 	struct ethhdr eh = {
209 		.h_source = BOND1_MAC,
210 		.h_dest = BOND2_MAC,
211 		.h_proto = htons(ETH_P_IP),
212 	};
213 	struct iphdr iph = {};
214 	struct udphdr uh = {};
215 	uint8_t buf[128];
216 	int i, s = -1;
217 	int ifindex;
218 
219 	s = socket(AF_PACKET, SOCK_RAW, IPPROTO_RAW);
220 	if (!ASSERT_GE(s, 0, "socket"))
221 		goto err;
222 
223 	ifindex = if_nametoindex("bond1");
224 	if (!ASSERT_GT(ifindex, 0, "get bond1 ifindex"))
225 		goto err;
226 
227 	iph.ihl = 5;
228 	iph.version = 4;
229 	iph.tos = 16;
230 	iph.id = 1;
231 	iph.ttl = 64;
232 	iph.protocol = IPPROTO_UDP;
233 	iph.saddr = 1;
234 	iph.daddr = 2;
235 	iph.tot_len = htons(sizeof(buf) - ETH_HLEN);
236 	iph.check = 0;
237 
238 	for (i = 1; i <= NPACKETS; i++) {
239 		int n;
240 		struct sockaddr_ll saddr_ll = {
241 			.sll_ifindex = ifindex,
242 			.sll_halen = ETH_ALEN,
243 			.sll_addr = BOND2_MAC,
244 		};
245 
246 		/* vary the UDP destination port for even distribution with roundrobin/xor modes */
247 		uh.dest++;
248 
249 		if (vary_dst_ip)
250 			iph.daddr++;
251 
252 		/* construct a packet */
253 		memcpy(buf, &eh, sizeof(eh));
254 		memcpy(buf + sizeof(eh), &iph, sizeof(iph));
255 		memcpy(buf + sizeof(eh) + sizeof(iph), &uh, sizeof(uh));
256 
257 		n = sendto(s, buf, sizeof(buf), 0, (struct sockaddr *)&saddr_ll, sizeof(saddr_ll));
258 		if (!ASSERT_EQ(n, sizeof(buf), "sendto"))
259 			goto err;
260 	}
261 
262 	return 0;
263 
264 err:
265 	if (s >= 0)
266 		close(s);
267 	return -1;
268 }
269 
270 static void test_xdp_bonding_with_mode(struct skeletons *skeletons, int mode, int xmit_policy)
271 {
272 	int bond1_rx;
273 
274 	if (bonding_setup(skeletons, mode, xmit_policy, BOND_BOTH_AND_ATTACH))
275 		goto out;
276 
277 	if (send_udp_packets(xmit_policy != BOND_XMIT_POLICY_LAYER34))
278 		goto out;
279 
280 	bond1_rx = get_rx_packets("bond1");
281 	ASSERT_EQ(bond1_rx, NPACKETS, "expected more received packets");
282 
283 	switch (mode) {
284 	case BOND_MODE_ROUNDROBIN:
285 	case BOND_MODE_XOR: {
286 		int veth1_rx = get_rx_packets("veth1_1");
287 		int veth2_rx = get_rx_packets("veth1_2");
288 		int diff = abs(veth1_rx - veth2_rx);
289 
290 		ASSERT_GE(veth1_rx + veth2_rx, NPACKETS, "expected more packets");
291 
292 		switch (xmit_policy) {
293 		case BOND_XMIT_POLICY_LAYER2:
294 			ASSERT_GE(diff, NPACKETS,
295 				  "expected packets on only one of the interfaces");
296 			break;
297 		case BOND_XMIT_POLICY_LAYER23:
298 		case BOND_XMIT_POLICY_LAYER34:
299 			ASSERT_LT(diff, NPACKETS/2,
300 				  "expected even distribution of packets");
301 			break;
302 		default:
303 			PRINT_FAIL("Unimplemented xmit_policy=%d\n", xmit_policy);
304 			break;
305 		}
306 		break;
307 	}
308 	case BOND_MODE_ACTIVEBACKUP: {
309 		int veth1_rx = get_rx_packets("veth1_1");
310 		int veth2_rx = get_rx_packets("veth1_2");
311 		int diff = abs(veth1_rx - veth2_rx);
312 
313 		ASSERT_GE(diff, NPACKETS,
314 			  "expected packets on only one of the interfaces");
315 		break;
316 	}
317 	default:
318 		PRINT_FAIL("Unimplemented xmit_policy=%d\n", xmit_policy);
319 		break;
320 	}
321 
322 out:
323 	bonding_cleanup(skeletons);
324 }
325 
326 /* Test the broadcast redirection using xdp_redirect_map_multi_prog and adding
327  * all the interfaces to it and checking that broadcasting won't send the packet
328  * to neither the ingress bond device (bond2) or its slave (veth2_1).
329  */
330 static void test_xdp_bonding_redirect_multi(struct skeletons *skeletons)
331 {
332 	static const char * const ifaces[] = {"bond2", "veth2_1", "veth2_2"};
333 	int veth1_1_rx, veth1_2_rx;
334 	int err;
335 
336 	if (bonding_setup(skeletons, BOND_MODE_ROUNDROBIN, BOND_XMIT_POLICY_LAYER23,
337 			  BOND_ONE_NO_ATTACH))
338 		goto out;
339 
340 
341 	if (!ASSERT_OK(setns_by_name("ns_dst"), "could not set netns to ns_dst"))
342 		goto out;
343 
344 	/* populate the devmap with the relevant interfaces */
345 	for (int i = 0; i < ARRAY_SIZE(ifaces); i++) {
346 		int ifindex = if_nametoindex(ifaces[i]);
347 		int map_fd = bpf_map__fd(skeletons->xdp_redirect_multi_kern->maps.map_all);
348 
349 		if (!ASSERT_GT(ifindex, 0, "could not get interface index"))
350 			goto out;
351 
352 		err = bpf_map_update_elem(map_fd, &ifindex, &ifindex, 0);
353 		if (!ASSERT_OK(err, "add interface to map_all"))
354 			goto out;
355 	}
356 
357 	if (xdp_attach(skeletons,
358 		       skeletons->xdp_redirect_multi_kern->progs.xdp_redirect_map_multi_prog,
359 		       "bond2"))
360 		goto out;
361 
362 	restore_root_netns();
363 
364 	if (send_udp_packets(BOND_MODE_ROUNDROBIN))
365 		goto out;
366 
367 	veth1_1_rx = get_rx_packets("veth1_1");
368 	veth1_2_rx = get_rx_packets("veth1_2");
369 
370 	ASSERT_EQ(veth1_1_rx, 0, "expected no packets on veth1_1");
371 	ASSERT_GE(veth1_2_rx, NPACKETS, "expected packets on veth1_2");
372 
373 out:
374 	restore_root_netns();
375 	bonding_cleanup(skeletons);
376 }
377 
378 /* Test that XDP programs cannot be attached to both the bond master and slaves simultaneously */
379 static void test_xdp_bonding_attach(struct skeletons *skeletons)
380 {
381 	struct bpf_link *link = NULL;
382 	struct bpf_link *link2 = NULL;
383 	int veth, bond, err;
384 
385 	if (!ASSERT_OK(system("ip link add veth type veth"), "add veth"))
386 		goto out;
387 	if (!ASSERT_OK(system("ip link add bond type bond"), "add bond"))
388 		goto out;
389 
390 	veth = if_nametoindex("veth");
391 	if (!ASSERT_GE(veth, 0, "if_nametoindex veth"))
392 		goto out;
393 	bond = if_nametoindex("bond");
394 	if (!ASSERT_GE(bond, 0, "if_nametoindex bond"))
395 		goto out;
396 
397 	/* enslaving with a XDP program loaded is allowed */
398 	link = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, veth);
399 	if (!ASSERT_OK_PTR(link, "attach program to veth"))
400 		goto out;
401 
402 	err = system("ip link set veth master bond");
403 	if (!ASSERT_OK(err, "set veth master"))
404 		goto out;
405 
406 	bpf_link__destroy(link);
407 	link = NULL;
408 
409 	/* attaching to slave when master has no program is allowed */
410 	link = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, veth);
411 	if (!ASSERT_OK_PTR(link, "attach program to slave when enslaved"))
412 		goto out;
413 
414 	/* attaching to master not allowed when slave has program loaded */
415 	link2 = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, bond);
416 	if (!ASSERT_ERR_PTR(link2, "attach program to master when slave has program"))
417 		goto out;
418 
419 	bpf_link__destroy(link);
420 	link = NULL;
421 
422 	/* attaching XDP program to master allowed when slave has no program */
423 	link = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, bond);
424 	if (!ASSERT_OK_PTR(link, "attach program to master"))
425 		goto out;
426 
427 	/* attaching to slave not allowed when master has program loaded */
428 	link2 = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, veth);
429 	if (!ASSERT_ERR_PTR(link2, "attach program to slave when master has program"))
430 		goto out;
431 
432 	bpf_link__destroy(link);
433 	link = NULL;
434 
435 	/* test program unwinding with a non-XDP slave */
436 	if (!ASSERT_OK(system("ip link add vxlan type vxlan id 1 remote 1.2.3.4 dstport 0 dev lo"),
437 		       "add vxlan"))
438 		goto out;
439 
440 	err = system("ip link set vxlan master bond");
441 	if (!ASSERT_OK(err, "set vxlan master"))
442 		goto out;
443 
444 	/* attaching not allowed when one slave does not support XDP */
445 	link = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, bond);
446 	if (!ASSERT_ERR_PTR(link, "attach program to master when slave does not support XDP"))
447 		goto out;
448 
449 out:
450 	bpf_link__destroy(link);
451 	bpf_link__destroy(link2);
452 
453 	system("ip link del veth");
454 	system("ip link del bond");
455 	system("ip link del vxlan");
456 }
457 
458 /* Test with nested bonding devices to catch issue with negative jump label count */
459 static void test_xdp_bonding_nested(struct skeletons *skeletons)
460 {
461 	struct bpf_link *link = NULL;
462 	int bond, err;
463 
464 	if (!ASSERT_OK(system("ip link add bond type bond"), "add bond"))
465 		goto out;
466 
467 	bond = if_nametoindex("bond");
468 	if (!ASSERT_GE(bond, 0, "if_nametoindex bond"))
469 		goto out;
470 
471 	if (!ASSERT_OK(system("ip link add bond_nest1 type bond"), "add bond_nest1"))
472 		goto out;
473 
474 	err = system("ip link set bond_nest1 master bond");
475 	if (!ASSERT_OK(err, "set bond_nest1 master"))
476 		goto out;
477 
478 	if (!ASSERT_OK(system("ip link add bond_nest2 type bond"), "add bond_nest1"))
479 		goto out;
480 
481 	err = system("ip link set bond_nest2 master bond_nest1");
482 	if (!ASSERT_OK(err, "set bond_nest2 master"))
483 		goto out;
484 
485 	link = bpf_program__attach_xdp(skeletons->xdp_dummy->progs.xdp_dummy_prog, bond);
486 	ASSERT_OK_PTR(link, "attach program to master");
487 
488 out:
489 	bpf_link__destroy(link);
490 	system("ip link del bond");
491 	system("ip link del bond_nest1");
492 	system("ip link del bond_nest2");
493 }
494 
495 static int libbpf_debug_print(enum libbpf_print_level level,
496 			      const char *format, va_list args)
497 {
498 	if (level != LIBBPF_WARN)
499 		vprintf(format, args);
500 	return 0;
501 }
502 
503 struct bond_test_case {
504 	char *name;
505 	int mode;
506 	int xmit_policy;
507 };
508 
509 static struct bond_test_case bond_test_cases[] = {
510 	{ "xdp_bonding_roundrobin", BOND_MODE_ROUNDROBIN, BOND_XMIT_POLICY_LAYER23, },
511 	{ "xdp_bonding_activebackup", BOND_MODE_ACTIVEBACKUP, BOND_XMIT_POLICY_LAYER23 },
512 
513 	{ "xdp_bonding_xor_layer2", BOND_MODE_XOR, BOND_XMIT_POLICY_LAYER2, },
514 	{ "xdp_bonding_xor_layer23", BOND_MODE_XOR, BOND_XMIT_POLICY_LAYER23, },
515 	{ "xdp_bonding_xor_layer34", BOND_MODE_XOR, BOND_XMIT_POLICY_LAYER34, },
516 };
517 
518 void serial_test_xdp_bonding(void)
519 {
520 	libbpf_print_fn_t old_print_fn;
521 	struct skeletons skeletons = {};
522 	int i;
523 
524 	old_print_fn = libbpf_set_print(libbpf_debug_print);
525 
526 	root_netns_fd = open("/proc/self/ns/net", O_RDONLY);
527 	if (!ASSERT_GE(root_netns_fd, 0, "open /proc/self/ns/net"))
528 		goto out;
529 
530 	skeletons.xdp_dummy = xdp_dummy__open_and_load();
531 	if (!ASSERT_OK_PTR(skeletons.xdp_dummy, "xdp_dummy__open_and_load"))
532 		goto out;
533 
534 	skeletons.xdp_tx = xdp_tx__open_and_load();
535 	if (!ASSERT_OK_PTR(skeletons.xdp_tx, "xdp_tx__open_and_load"))
536 		goto out;
537 
538 	skeletons.xdp_redirect_multi_kern = xdp_redirect_multi_kern__open_and_load();
539 	if (!ASSERT_OK_PTR(skeletons.xdp_redirect_multi_kern,
540 			   "xdp_redirect_multi_kern__open_and_load"))
541 		goto out;
542 
543 	if (test__start_subtest("xdp_bonding_attach"))
544 		test_xdp_bonding_attach(&skeletons);
545 
546 	if (test__start_subtest("xdp_bonding_nested"))
547 		test_xdp_bonding_nested(&skeletons);
548 
549 	for (i = 0; i < ARRAY_SIZE(bond_test_cases); i++) {
550 		struct bond_test_case *test_case = &bond_test_cases[i];
551 
552 		if (test__start_subtest(test_case->name))
553 			test_xdp_bonding_with_mode(
554 				&skeletons,
555 				test_case->mode,
556 				test_case->xmit_policy);
557 	}
558 
559 	if (test__start_subtest("xdp_bonding_redirect_multi"))
560 		test_xdp_bonding_redirect_multi(&skeletons);
561 
562 out:
563 	xdp_dummy__destroy(skeletons.xdp_dummy);
564 	xdp_tx__destroy(skeletons.xdp_tx);
565 	xdp_redirect_multi_kern__destroy(skeletons.xdp_redirect_multi_kern);
566 
567 	libbpf_set_print(old_print_fn);
568 	if (root_netns_fd >= 0)
569 		close(root_netns_fd);
570 }
571