xref: /linux/tools/testing/selftests/bpf/prog_tests/xdp_metadata.c (revision 6e7fd890f1d6ac83805409e9c346240de2705584)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <test_progs.h>
3 #include <network_helpers.h>
4 #include "xdp_metadata.skel.h"
5 #include "xdp_metadata2.skel.h"
6 #include "xdp_metadata.h"
7 #include "xsk.h"
8 
9 #include <bpf/btf.h>
10 #include <linux/errqueue.h>
11 #include <linux/if_link.h>
12 #include <linux/net_tstamp.h>
13 #include <linux/udp.h>
14 #include <sys/mman.h>
15 #include <net/if.h>
16 #include <poll.h>
17 
18 #define TX_NAME "veTX"
19 #define RX_NAME "veRX"
20 
21 #define UDP_PAYLOAD_BYTES 4
22 
23 #define UDP_SOURCE_PORT 1234
24 #define AF_XDP_CONSUMER_PORT 8080
25 
26 #define UMEM_NUM 16
27 #define UMEM_FRAME_SIZE XSK_UMEM__DEFAULT_FRAME_SIZE
28 #define UMEM_SIZE (UMEM_FRAME_SIZE * UMEM_NUM)
29 #define XDP_FLAGS XDP_FLAGS_DRV_MODE
30 #define QUEUE_ID 0
31 
32 #define TX_ADDR "10.0.0.1"
33 #define RX_ADDR "10.0.0.2"
34 #define PREFIX_LEN "8"
35 #define FAMILY AF_INET
36 #define TX_NETNS_NAME "xdp_metadata_tx"
37 #define RX_NETNS_NAME "xdp_metadata_rx"
38 #define TX_MAC "00:00:00:00:00:01"
39 #define RX_MAC "00:00:00:00:00:02"
40 
41 #define VLAN_ID 59
42 #define VLAN_PROTO "802.1Q"
43 #define VLAN_PID htons(ETH_P_8021Q)
44 #define TX_NAME_VLAN TX_NAME "." TO_STR(VLAN_ID)
45 
46 #define XDP_RSS_TYPE_L4 BIT(3)
47 #define VLAN_VID_MASK 0xfff
48 
49 struct xsk {
50 	void *umem_area;
51 	struct xsk_umem *umem;
52 	struct xsk_ring_prod fill;
53 	struct xsk_ring_cons comp;
54 	struct xsk_ring_prod tx;
55 	struct xsk_ring_cons rx;
56 	struct xsk_socket *socket;
57 };
58 
59 static int open_xsk(int ifindex, struct xsk *xsk)
60 {
61 	int mmap_flags = MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE;
62 	const struct xsk_socket_config socket_config = {
63 		.rx_size = XSK_RING_PROD__DEFAULT_NUM_DESCS,
64 		.tx_size = XSK_RING_PROD__DEFAULT_NUM_DESCS,
65 		.bind_flags = XDP_COPY,
66 	};
67 	const struct xsk_umem_config umem_config = {
68 		.fill_size = XSK_RING_PROD__DEFAULT_NUM_DESCS,
69 		.comp_size = XSK_RING_CONS__DEFAULT_NUM_DESCS,
70 		.frame_size = XSK_UMEM__DEFAULT_FRAME_SIZE,
71 		.flags = XDP_UMEM_UNALIGNED_CHUNK_FLAG | XDP_UMEM_TX_SW_CSUM |
72 			 XDP_UMEM_TX_METADATA_LEN,
73 		.tx_metadata_len = sizeof(struct xsk_tx_metadata),
74 	};
75 	__u32 idx;
76 	u64 addr;
77 	int ret;
78 	int i;
79 
80 	xsk->umem_area = mmap(NULL, UMEM_SIZE, PROT_READ | PROT_WRITE, mmap_flags, -1, 0);
81 	if (!ASSERT_NEQ(xsk->umem_area, MAP_FAILED, "mmap"))
82 		return -1;
83 
84 	ret = xsk_umem__create(&xsk->umem,
85 			       xsk->umem_area, UMEM_SIZE,
86 			       &xsk->fill,
87 			       &xsk->comp,
88 			       &umem_config);
89 	if (!ASSERT_OK(ret, "xsk_umem__create"))
90 		return ret;
91 
92 	ret = xsk_socket__create(&xsk->socket, ifindex, QUEUE_ID,
93 				 xsk->umem,
94 				 &xsk->rx,
95 				 &xsk->tx,
96 				 &socket_config);
97 	if (!ASSERT_OK(ret, "xsk_socket__create"))
98 		return ret;
99 
100 	/* First half of umem is for TX. This way address matches 1-to-1
101 	 * to the completion queue index.
102 	 */
103 
104 	for (i = 0; i < UMEM_NUM / 2; i++) {
105 		addr = i * UMEM_FRAME_SIZE;
106 		printf("%p: tx_desc[%d] -> %lx\n", xsk, i, addr);
107 	}
108 
109 	/* Second half of umem is for RX. */
110 
111 	ret = xsk_ring_prod__reserve(&xsk->fill, UMEM_NUM / 2, &idx);
112 	if (!ASSERT_EQ(UMEM_NUM / 2, ret, "xsk_ring_prod__reserve"))
113 		return ret;
114 	if (!ASSERT_EQ(idx, 0, "fill idx != 0"))
115 		return -1;
116 
117 	for (i = 0; i < UMEM_NUM / 2; i++) {
118 		addr = (UMEM_NUM / 2 + i) * UMEM_FRAME_SIZE;
119 		printf("%p: rx_desc[%d] -> %lx\n", xsk, i, addr);
120 		*xsk_ring_prod__fill_addr(&xsk->fill, i) = addr;
121 	}
122 	xsk_ring_prod__submit(&xsk->fill, ret);
123 
124 	return 0;
125 }
126 
127 static void close_xsk(struct xsk *xsk)
128 {
129 	if (xsk->umem)
130 		xsk_umem__delete(xsk->umem);
131 	if (xsk->socket)
132 		xsk_socket__delete(xsk->socket);
133 	munmap(xsk->umem_area, UMEM_SIZE);
134 }
135 
136 static void ip_csum(struct iphdr *iph)
137 {
138 	__u32 sum = 0;
139 	__u16 *p;
140 	int i;
141 
142 	iph->check = 0;
143 	p = (void *)iph;
144 	for (i = 0; i < sizeof(*iph) / sizeof(*p); i++)
145 		sum += p[i];
146 
147 	while (sum >> 16)
148 		sum = (sum & 0xffff) + (sum >> 16);
149 
150 	iph->check = ~sum;
151 }
152 
153 static int generate_packet(struct xsk *xsk, __u16 dst_port)
154 {
155 	struct xsk_tx_metadata *meta;
156 	struct xdp_desc *tx_desc;
157 	struct udphdr *udph;
158 	struct ethhdr *eth;
159 	struct iphdr *iph;
160 	void *data;
161 	__u32 idx;
162 	int ret;
163 
164 	ret = xsk_ring_prod__reserve(&xsk->tx, 1, &idx);
165 	if (!ASSERT_EQ(ret, 1, "xsk_ring_prod__reserve"))
166 		return -1;
167 
168 	tx_desc = xsk_ring_prod__tx_desc(&xsk->tx, idx);
169 	tx_desc->addr = idx % (UMEM_NUM / 2) * UMEM_FRAME_SIZE + sizeof(struct xsk_tx_metadata);
170 	printf("%p: tx_desc[%u]->addr=%llx\n", xsk, idx, tx_desc->addr);
171 	data = xsk_umem__get_data(xsk->umem_area, tx_desc->addr);
172 
173 	meta = data - sizeof(struct xsk_tx_metadata);
174 	memset(meta, 0, sizeof(*meta));
175 	meta->flags = XDP_TXMD_FLAGS_TIMESTAMP;
176 
177 	eth = data;
178 	iph = (void *)(eth + 1);
179 	udph = (void *)(iph + 1);
180 
181 	memcpy(eth->h_dest, "\x00\x00\x00\x00\x00\x02", ETH_ALEN);
182 	memcpy(eth->h_source, "\x00\x00\x00\x00\x00\x01", ETH_ALEN);
183 	eth->h_proto = htons(ETH_P_IP);
184 
185 	iph->version = 0x4;
186 	iph->ihl = 0x5;
187 	iph->tos = 0x9;
188 	iph->tot_len = htons(sizeof(*iph) + sizeof(*udph) + UDP_PAYLOAD_BYTES);
189 	iph->id = 0;
190 	iph->frag_off = 0;
191 	iph->ttl = 0;
192 	iph->protocol = IPPROTO_UDP;
193 	ASSERT_EQ(inet_pton(FAMILY, TX_ADDR, &iph->saddr), 1, "inet_pton(TX_ADDR)");
194 	ASSERT_EQ(inet_pton(FAMILY, RX_ADDR, &iph->daddr), 1, "inet_pton(RX_ADDR)");
195 	ip_csum(iph);
196 
197 	udph->source = htons(UDP_SOURCE_PORT);
198 	udph->dest = htons(dst_port);
199 	udph->len = htons(sizeof(*udph) + UDP_PAYLOAD_BYTES);
200 	udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
201 					 ntohs(udph->len), IPPROTO_UDP, 0);
202 
203 	memset(udph + 1, 0xAA, UDP_PAYLOAD_BYTES);
204 
205 	meta->flags |= XDP_TXMD_FLAGS_CHECKSUM;
206 	meta->request.csum_start = sizeof(*eth) + sizeof(*iph);
207 	meta->request.csum_offset = offsetof(struct udphdr, check);
208 
209 	tx_desc->len = sizeof(*eth) + sizeof(*iph) + sizeof(*udph) + UDP_PAYLOAD_BYTES;
210 	tx_desc->options |= XDP_TX_METADATA;
211 	xsk_ring_prod__submit(&xsk->tx, 1);
212 
213 	ret = sendto(xsk_socket__fd(xsk->socket), NULL, 0, MSG_DONTWAIT, NULL, 0);
214 	if (!ASSERT_GE(ret, 0, "sendto"))
215 		return ret;
216 
217 	return 0;
218 }
219 
220 static int generate_packet_inet(void)
221 {
222 	char udp_payload[UDP_PAYLOAD_BYTES];
223 	struct sockaddr_in rx_addr;
224 	int sock_fd, err = 0;
225 
226 	/* Build a packet */
227 	memset(udp_payload, 0xAA, UDP_PAYLOAD_BYTES);
228 	rx_addr.sin_addr.s_addr = inet_addr(RX_ADDR);
229 	rx_addr.sin_family = AF_INET;
230 	rx_addr.sin_port = htons(AF_XDP_CONSUMER_PORT);
231 
232 	sock_fd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
233 	if (!ASSERT_GE(sock_fd, 0, "socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP)"))
234 		return sock_fd;
235 
236 	err = sendto(sock_fd, udp_payload, UDP_PAYLOAD_BYTES, MSG_DONTWAIT,
237 		     (void *)&rx_addr, sizeof(rx_addr));
238 	ASSERT_GE(err, 0, "sendto");
239 
240 	close(sock_fd);
241 	return err;
242 }
243 
244 static void complete_tx(struct xsk *xsk)
245 {
246 	struct xsk_tx_metadata *meta;
247 	__u64 addr;
248 	void *data;
249 	__u32 idx;
250 
251 	if (ASSERT_EQ(xsk_ring_cons__peek(&xsk->comp, 1, &idx), 1, "xsk_ring_cons__peek")) {
252 		addr = *xsk_ring_cons__comp_addr(&xsk->comp, idx);
253 
254 		printf("%p: complete tx idx=%u addr=%llx\n", xsk, idx, addr);
255 
256 		data = xsk_umem__get_data(xsk->umem_area, addr);
257 		meta = data - sizeof(struct xsk_tx_metadata);
258 
259 		ASSERT_NEQ(meta->completion.tx_timestamp, 0, "tx_timestamp");
260 
261 		xsk_ring_cons__release(&xsk->comp, 1);
262 	}
263 }
264 
265 static void refill_rx(struct xsk *xsk, __u64 addr)
266 {
267 	__u32 idx;
268 
269 	if (ASSERT_EQ(xsk_ring_prod__reserve(&xsk->fill, 1, &idx), 1, "xsk_ring_prod__reserve")) {
270 		printf("%p: complete idx=%u addr=%llx\n", xsk, idx, addr);
271 		*xsk_ring_prod__fill_addr(&xsk->fill, idx) = addr;
272 		xsk_ring_prod__submit(&xsk->fill, 1);
273 	}
274 }
275 
276 static int verify_xsk_metadata(struct xsk *xsk, bool sent_from_af_xdp)
277 {
278 	const struct xdp_desc *rx_desc;
279 	struct pollfd fds = {};
280 	struct xdp_meta *meta;
281 	struct udphdr *udph;
282 	struct ethhdr *eth;
283 	struct iphdr *iph;
284 	__u64 comp_addr;
285 	void *data;
286 	__u64 addr;
287 	__u32 idx = 0;
288 	int ret;
289 
290 	ret = recvfrom(xsk_socket__fd(xsk->socket), NULL, 0, MSG_DONTWAIT, NULL, NULL);
291 	if (!ASSERT_EQ(ret, 0, "recvfrom"))
292 		return -1;
293 
294 	fds.fd = xsk_socket__fd(xsk->socket);
295 	fds.events = POLLIN;
296 
297 	ret = poll(&fds, 1, 1000);
298 	if (!ASSERT_GT(ret, 0, "poll"))
299 		return -1;
300 
301 	ret = xsk_ring_cons__peek(&xsk->rx, 1, &idx);
302 	if (!ASSERT_EQ(ret, 1, "xsk_ring_cons__peek"))
303 		return -2;
304 
305 	rx_desc = xsk_ring_cons__rx_desc(&xsk->rx, idx);
306 	comp_addr = xsk_umem__extract_addr(rx_desc->addr);
307 	addr = xsk_umem__add_offset_to_addr(rx_desc->addr);
308 	printf("%p: rx_desc[%u]->addr=%llx addr=%llx comp_addr=%llx\n",
309 	       xsk, idx, rx_desc->addr, addr, comp_addr);
310 	data = xsk_umem__get_data(xsk->umem_area, addr);
311 
312 	/* Make sure we got the packet offset correctly. */
313 
314 	eth = data;
315 	ASSERT_EQ(eth->h_proto, htons(ETH_P_IP), "eth->h_proto");
316 	iph = (void *)(eth + 1);
317 	ASSERT_EQ((int)iph->version, 4, "iph->version");
318 	udph = (void *)(iph + 1);
319 
320 	/* custom metadata */
321 
322 	meta = data - sizeof(struct xdp_meta);
323 
324 	if (!ASSERT_NEQ(meta->rx_timestamp, 0, "rx_timestamp"))
325 		return -1;
326 
327 	if (!ASSERT_NEQ(meta->rx_hash, 0, "rx_hash"))
328 		return -1;
329 
330 	if (!sent_from_af_xdp) {
331 		if (!ASSERT_NEQ(meta->rx_hash_type & XDP_RSS_TYPE_L4, 0, "rx_hash_type"))
332 			return -1;
333 
334 		if (!ASSERT_EQ(meta->rx_vlan_tci & VLAN_VID_MASK, VLAN_ID, "rx_vlan_tci"))
335 			return -1;
336 
337 		if (!ASSERT_EQ(meta->rx_vlan_proto, VLAN_PID, "rx_vlan_proto"))
338 			return -1;
339 		goto done;
340 	}
341 
342 	ASSERT_EQ(meta->rx_hash_type, 0, "rx_hash_type");
343 
344 	/* checksum offload */
345 	ASSERT_EQ(udph->check, htons(0x721c), "csum");
346 
347 done:
348 	xsk_ring_cons__release(&xsk->rx, 1);
349 	refill_rx(xsk, comp_addr);
350 
351 	return 0;
352 }
353 
354 static void switch_ns_to_rx(struct nstoken **tok)
355 {
356 	close_netns(*tok);
357 	*tok = open_netns(RX_NETNS_NAME);
358 }
359 
360 static void switch_ns_to_tx(struct nstoken **tok)
361 {
362 	close_netns(*tok);
363 	*tok = open_netns(TX_NETNS_NAME);
364 }
365 
366 void test_xdp_metadata(void)
367 {
368 	struct xdp_metadata2 *bpf_obj2 = NULL;
369 	struct xdp_metadata *bpf_obj = NULL;
370 	struct bpf_program *new_prog, *prog;
371 	struct nstoken *tok = NULL;
372 	__u32 queue_id = QUEUE_ID;
373 	struct bpf_map *prog_arr;
374 	struct xsk tx_xsk = {};
375 	struct xsk rx_xsk = {};
376 	__u32 val, key = 0;
377 	int retries = 10;
378 	int rx_ifindex;
379 	int tx_ifindex;
380 	int sock_fd;
381 	int ret;
382 
383 	/* Setup new networking namespaces, with a veth pair. */
384 	SYS(out, "ip netns add " TX_NETNS_NAME);
385 	SYS(out, "ip netns add " RX_NETNS_NAME);
386 
387 	tok = open_netns(TX_NETNS_NAME);
388 	if (!ASSERT_OK_PTR(tok, "setns"))
389 		goto out;
390 	SYS(out, "ip link add numtxqueues 1 numrxqueues 1 " TX_NAME
391 	    " type veth peer " RX_NAME " numtxqueues 1 numrxqueues 1");
392 	SYS(out, "ip link set " RX_NAME " netns " RX_NETNS_NAME);
393 
394 	SYS(out, "ip link set dev " TX_NAME " address " TX_MAC);
395 	SYS(out, "ip link set dev " TX_NAME " up");
396 
397 	SYS(out, "ip link add link " TX_NAME " " TX_NAME_VLAN
398 		 " type vlan proto " VLAN_PROTO " id " TO_STR(VLAN_ID));
399 	SYS(out, "ip link set dev " TX_NAME_VLAN " up");
400 	SYS(out, "ip addr add " TX_ADDR "/" PREFIX_LEN " dev " TX_NAME_VLAN);
401 
402 	/* Avoid ARP calls */
403 	SYS(out, "ip -4 neigh add " RX_ADDR " lladdr " RX_MAC " dev " TX_NAME_VLAN);
404 
405 	switch_ns_to_rx(&tok);
406 	if (!ASSERT_OK_PTR(tok, "setns rx"))
407 		goto out;
408 
409 	SYS(out, "ip link set dev " RX_NAME " address " RX_MAC);
410 	SYS(out, "ip link set dev " RX_NAME " up");
411 	SYS(out, "ip addr add " RX_ADDR "/" PREFIX_LEN " dev " RX_NAME);
412 
413 	rx_ifindex = if_nametoindex(RX_NAME);
414 
415 	/* Setup separate AF_XDP for RX interface. */
416 
417 	ret = open_xsk(rx_ifindex, &rx_xsk);
418 	if (!ASSERT_OK(ret, "open_xsk(RX_NAME)"))
419 		goto out;
420 
421 	bpf_obj = xdp_metadata__open();
422 	if (!ASSERT_OK_PTR(bpf_obj, "open skeleton"))
423 		goto out;
424 
425 	prog = bpf_object__find_program_by_name(bpf_obj->obj, "rx");
426 	bpf_program__set_ifindex(prog, rx_ifindex);
427 	bpf_program__set_flags(prog, BPF_F_XDP_DEV_BOUND_ONLY);
428 
429 	if (!ASSERT_OK(xdp_metadata__load(bpf_obj), "load skeleton"))
430 		goto out;
431 
432 	/* Make sure we can't add dev-bound programs to prog maps. */
433 	prog_arr = bpf_object__find_map_by_name(bpf_obj->obj, "prog_arr");
434 	if (!ASSERT_OK_PTR(prog_arr, "no prog_arr map"))
435 		goto out;
436 
437 	val = bpf_program__fd(prog);
438 	if (!ASSERT_ERR(bpf_map__update_elem(prog_arr, &key, sizeof(key),
439 					     &val, sizeof(val), BPF_ANY),
440 			"update prog_arr"))
441 		goto out;
442 
443 	/* Attach BPF program to RX interface. */
444 
445 	ret = bpf_xdp_attach(rx_ifindex,
446 			     bpf_program__fd(bpf_obj->progs.rx),
447 			     XDP_FLAGS, NULL);
448 	if (!ASSERT_GE(ret, 0, "bpf_xdp_attach"))
449 		goto out;
450 
451 	sock_fd = xsk_socket__fd(rx_xsk.socket);
452 	ret = bpf_map_update_elem(bpf_map__fd(bpf_obj->maps.xsk), &queue_id, &sock_fd, 0);
453 	if (!ASSERT_GE(ret, 0, "bpf_map_update_elem"))
454 		goto out;
455 
456 	switch_ns_to_tx(&tok);
457 	if (!ASSERT_OK_PTR(tok, "setns tx"))
458 		goto out;
459 
460 	/* Setup separate AF_XDP for TX interface nad send packet to the RX socket. */
461 	tx_ifindex = if_nametoindex(TX_NAME);
462 	ret = open_xsk(tx_ifindex, &tx_xsk);
463 	if (!ASSERT_OK(ret, "open_xsk(TX_NAME)"))
464 		goto out;
465 
466 	if (!ASSERT_GE(generate_packet(&tx_xsk, AF_XDP_CONSUMER_PORT), 0,
467 		       "generate AF_XDP_CONSUMER_PORT"))
468 		goto out;
469 
470 	switch_ns_to_rx(&tok);
471 	if (!ASSERT_OK_PTR(tok, "setns rx"))
472 		goto out;
473 
474 	/* Verify packet sent from AF_XDP has proper metadata. */
475 	if (!ASSERT_GE(verify_xsk_metadata(&rx_xsk, true), 0,
476 		       "verify_xsk_metadata"))
477 		goto out;
478 
479 	switch_ns_to_tx(&tok);
480 	if (!ASSERT_OK_PTR(tok, "setns tx"))
481 		goto out;
482 	complete_tx(&tx_xsk);
483 
484 	/* Now check metadata of packet, generated with network stack */
485 	if (!ASSERT_GE(generate_packet_inet(), 0, "generate UDP packet"))
486 		goto out;
487 
488 	switch_ns_to_rx(&tok);
489 	if (!ASSERT_OK_PTR(tok, "setns rx"))
490 		goto out;
491 
492 	if (!ASSERT_GE(verify_xsk_metadata(&rx_xsk, false), 0,
493 		       "verify_xsk_metadata"))
494 		goto out;
495 
496 	/* Make sure freplace correctly picks up original bound device
497 	 * and doesn't crash.
498 	 */
499 
500 	bpf_obj2 = xdp_metadata2__open();
501 	if (!ASSERT_OK_PTR(bpf_obj2, "open skeleton"))
502 		goto out;
503 
504 	new_prog = bpf_object__find_program_by_name(bpf_obj2->obj, "freplace_rx");
505 	bpf_program__set_attach_target(new_prog, bpf_program__fd(prog), "rx");
506 
507 	if (!ASSERT_OK(xdp_metadata2__load(bpf_obj2), "load freplace skeleton"))
508 		goto out;
509 
510 	if (!ASSERT_OK(xdp_metadata2__attach(bpf_obj2), "attach freplace"))
511 		goto out;
512 
513 	switch_ns_to_tx(&tok);
514 	if (!ASSERT_OK_PTR(tok, "setns tx"))
515 		goto out;
516 
517 	/* Send packet to trigger . */
518 	if (!ASSERT_GE(generate_packet(&tx_xsk, AF_XDP_CONSUMER_PORT), 0,
519 		       "generate freplace packet"))
520 		goto out;
521 
522 	switch_ns_to_rx(&tok);
523 	if (!ASSERT_OK_PTR(tok, "setns rx"))
524 		goto out;
525 
526 	while (!retries--) {
527 		if (bpf_obj2->bss->called)
528 			break;
529 		usleep(10);
530 	}
531 	ASSERT_GT(bpf_obj2->bss->called, 0, "not called");
532 
533 out:
534 	close_xsk(&rx_xsk);
535 	close_xsk(&tx_xsk);
536 	xdp_metadata2__destroy(bpf_obj2);
537 	xdp_metadata__destroy(bpf_obj);
538 	if (tok)
539 		close_netns(tok);
540 	SYS_NOFAIL("ip netns del " RX_NETNS_NAME);
541 	SYS_NOFAIL("ip netns del " TX_NETNS_NAME);
542 }
543