xref: /linux/tools/testing/selftests/net/ip_defrag.c (revision 05909cd9a0c8811731b38697af13075e8954314f)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #define _GNU_SOURCE
4 
5 #include <arpa/inet.h>
6 #include <errno.h>
7 #include <error.h>
8 #include <linux/in.h>
9 #include <netinet/ip.h>
10 #include <netinet/ip6.h>
11 #include <netinet/udp.h>
12 #include <stdbool.h>
13 #include <stdio.h>
14 #include <stdlib.h>
15 #include <string.h>
16 #include <time.h>
17 #include <unistd.h>
18 
19 static bool		cfg_do_ipv4;
20 static bool		cfg_do_ipv6;
21 static bool		cfg_verbose;
22 static bool		cfg_overlap;
23 static bool		cfg_permissive;
24 static unsigned short	cfg_port = 9000;
25 
26 const struct in_addr addr4 = { .s_addr = __constant_htonl(INADDR_LOOPBACK + 2) };
27 const struct in6_addr addr6 = IN6ADDR_LOOPBACK_INIT;
28 
29 #define IP4_HLEN	(sizeof(struct iphdr))
30 #define IP6_HLEN	(sizeof(struct ip6_hdr))
31 #define UDP_HLEN	(sizeof(struct udphdr))
32 
33 /* IPv6 fragment header lenth. */
34 #define FRAG_HLEN	8
35 
36 static int payload_len;
37 static int max_frag_len;
38 
39 #define MSG_LEN_MAX	10000	/* Max UDP payload length. */
40 
41 #define IP4_MF		(1u << 13)  /* IPv4 MF flag. */
42 #define IP6_MF		(1)  /* IPv6 MF flag. */
43 
44 #define CSUM_MANGLED_0 (0xffff)
45 
46 static uint8_t udp_payload[MSG_LEN_MAX];
47 static uint8_t ip_frame[IP_MAXPACKET];
48 static uint32_t ip_id = 0xabcd;
49 static int msg_counter;
50 static int frag_counter;
51 static unsigned int seed;
52 
53 /* Receive a UDP packet. Validate it matches udp_payload. */
54 static void recv_validate_udp(int fd_udp)
55 {
56 	ssize_t ret;
57 	static uint8_t recv_buff[MSG_LEN_MAX];
58 
59 	ret = recv(fd_udp, recv_buff, payload_len, 0);
60 	msg_counter++;
61 
62 	if (cfg_overlap) {
63 		if (ret == -1 && (errno == ETIMEDOUT || errno == EAGAIN))
64 			return;  /* OK */
65 		if (!cfg_permissive) {
66 			if (ret != -1)
67 				error(1, 0, "recv: expected timeout; got %d",
68 					(int)ret);
69 			error(1, errno, "recv: expected timeout: %d", errno);
70 		}
71 	}
72 
73 	if (ret == -1)
74 		error(1, errno, "recv: payload_len = %d max_frag_len = %d",
75 			payload_len, max_frag_len);
76 	if (ret != payload_len)
77 		error(1, 0, "recv: wrong size: %d vs %d", (int)ret, payload_len);
78 	if (memcmp(udp_payload, recv_buff, payload_len))
79 		error(1, 0, "recv: wrong data");
80 }
81 
82 static uint32_t raw_checksum(uint8_t *buf, int len, uint32_t sum)
83 {
84 	int i;
85 
86 	for (i = 0; i < (len & ~1U); i += 2) {
87 		sum += (u_int16_t)ntohs(*((u_int16_t *)(buf + i)));
88 		if (sum > 0xffff)
89 			sum -= 0xffff;
90 	}
91 
92 	if (i < len) {
93 		sum += buf[i] << 8;
94 		if (sum > 0xffff)
95 			sum -= 0xffff;
96 	}
97 
98 	return sum;
99 }
100 
101 static uint16_t udp_checksum(struct ip *iphdr, struct udphdr *udphdr)
102 {
103 	uint32_t sum = 0;
104 	uint16_t res;
105 
106 	sum = raw_checksum((uint8_t *)&iphdr->ip_src, 2 * sizeof(iphdr->ip_src),
107 				IPPROTO_UDP + (uint32_t)(UDP_HLEN + payload_len));
108 	sum = raw_checksum((uint8_t *)udphdr, UDP_HLEN, sum);
109 	sum = raw_checksum((uint8_t *)udp_payload, payload_len, sum);
110 	res = 0xffff & ~sum;
111 	if (res)
112 		return htons(res);
113 	else
114 		return CSUM_MANGLED_0;
115 }
116 
117 static uint16_t udp6_checksum(struct ip6_hdr *iphdr, struct udphdr *udphdr)
118 {
119 	uint32_t sum = 0;
120 	uint16_t res;
121 
122 	sum = raw_checksum((uint8_t *)&iphdr->ip6_src, 2 * sizeof(iphdr->ip6_src),
123 				IPPROTO_UDP);
124 	sum = raw_checksum((uint8_t *)&udphdr->len, sizeof(udphdr->len), sum);
125 	sum = raw_checksum((uint8_t *)udphdr, UDP_HLEN, sum);
126 	sum = raw_checksum((uint8_t *)udp_payload, payload_len, sum);
127 	res = 0xffff & ~sum;
128 	if (res)
129 		return htons(res);
130 	else
131 		return CSUM_MANGLED_0;
132 }
133 
134 static void send_fragment(int fd_raw, struct sockaddr *addr, socklen_t alen,
135 				int offset, bool ipv6)
136 {
137 	int frag_len;
138 	int res;
139 	int payload_offset = offset > 0 ? offset - UDP_HLEN : 0;
140 	uint8_t *frag_start = ipv6 ? ip_frame + IP6_HLEN + FRAG_HLEN :
141 					ip_frame + IP4_HLEN;
142 
143 	if (offset == 0) {
144 		struct udphdr udphdr;
145 		udphdr.source = htons(cfg_port + 1);
146 		udphdr.dest = htons(cfg_port);
147 		udphdr.len = htons(UDP_HLEN + payload_len);
148 		udphdr.check = 0;
149 		if (ipv6)
150 			udphdr.check = udp6_checksum((struct ip6_hdr *)ip_frame, &udphdr);
151 		else
152 			udphdr.check = udp_checksum((struct ip *)ip_frame, &udphdr);
153 		memcpy(frag_start, &udphdr, UDP_HLEN);
154 	}
155 
156 	if (ipv6) {
157 		struct ip6_hdr *ip6hdr = (struct ip6_hdr *)ip_frame;
158 		struct ip6_frag *fraghdr = (struct ip6_frag *)(ip_frame + IP6_HLEN);
159 		if (payload_len - payload_offset <= max_frag_len && offset > 0) {
160 			/* This is the last fragment. */
161 			frag_len = FRAG_HLEN + payload_len - payload_offset;
162 			fraghdr->ip6f_offlg = htons(offset);
163 		} else {
164 			frag_len = FRAG_HLEN + max_frag_len;
165 			fraghdr->ip6f_offlg = htons(offset | IP6_MF);
166 		}
167 		ip6hdr->ip6_plen = htons(frag_len);
168 		if (offset == 0)
169 			memcpy(frag_start + UDP_HLEN, udp_payload,
170 				frag_len - FRAG_HLEN - UDP_HLEN);
171 		else
172 			memcpy(frag_start, udp_payload + payload_offset,
173 				frag_len - FRAG_HLEN);
174 		frag_len += IP6_HLEN;
175 	} else {
176 		struct ip *iphdr = (struct ip *)ip_frame;
177 		if (payload_len - payload_offset <= max_frag_len && offset > 0) {
178 			/* This is the last fragment. */
179 			frag_len = IP4_HLEN + payload_len - payload_offset;
180 			iphdr->ip_off = htons(offset / 8);
181 		} else {
182 			frag_len = IP4_HLEN + max_frag_len;
183 			iphdr->ip_off = htons(offset / 8 | IP4_MF);
184 		}
185 		iphdr->ip_len = htons(frag_len);
186 		if (offset == 0)
187 			memcpy(frag_start + UDP_HLEN, udp_payload,
188 				frag_len - IP4_HLEN - UDP_HLEN);
189 		else
190 			memcpy(frag_start, udp_payload + payload_offset,
191 				frag_len - IP4_HLEN);
192 	}
193 
194 	res = sendto(fd_raw, ip_frame, frag_len, 0, addr, alen);
195 	if (res < 0 && errno != EPERM)
196 		error(1, errno, "send_fragment");
197 	if (res >= 0 && res != frag_len)
198 		error(1, 0, "send_fragment: %d vs %d", res, frag_len);
199 
200 	frag_counter++;
201 }
202 
203 static void send_udp_frags(int fd_raw, struct sockaddr *addr,
204 				socklen_t alen, bool ipv6)
205 {
206 	struct ip *iphdr = (struct ip *)ip_frame;
207 	struct ip6_hdr *ip6hdr = (struct ip6_hdr *)ip_frame;
208 	int res;
209 	int offset;
210 	int frag_len;
211 
212 	/* Send the UDP datagram using raw IP fragments: the 0th fragment
213 	 * has the UDP header; other fragments are pieces of udp_payload
214 	 * split in chunks of frag_len size.
215 	 *
216 	 * Odd fragments (1st, 3rd, 5th, etc.) are sent out first, then
217 	 * even fragments (0th, 2nd, etc.) are sent out.
218 	 */
219 	if (ipv6) {
220 		struct ip6_frag *fraghdr = (struct ip6_frag *)(ip_frame + IP6_HLEN);
221 		((struct sockaddr_in6 *)addr)->sin6_port = 0;
222 		memset(ip6hdr, 0, sizeof(*ip6hdr));
223 		ip6hdr->ip6_flow = htonl(6<<28);  /* Version. */
224 		ip6hdr->ip6_nxt = IPPROTO_FRAGMENT;
225 		ip6hdr->ip6_hops = 255;
226 		ip6hdr->ip6_src = addr6;
227 		ip6hdr->ip6_dst = addr6;
228 		fraghdr->ip6f_nxt = IPPROTO_UDP;
229 		fraghdr->ip6f_reserved = 0;
230 		fraghdr->ip6f_ident = htonl(ip_id++);
231 	} else {
232 		memset(iphdr, 0, sizeof(*iphdr));
233 		iphdr->ip_hl = 5;
234 		iphdr->ip_v = 4;
235 		iphdr->ip_tos = 0;
236 		iphdr->ip_id = htons(ip_id++);
237 		iphdr->ip_ttl = 0x40;
238 		iphdr->ip_p = IPPROTO_UDP;
239 		iphdr->ip_src.s_addr = htonl(INADDR_LOOPBACK);
240 		iphdr->ip_dst = addr4;
241 		iphdr->ip_sum = 0;
242 	}
243 
244 	/* Occasionally test in-order fragments. */
245 	if (!cfg_overlap && (rand() % 100 < 15)) {
246 		offset = 0;
247 		while (offset < (UDP_HLEN + payload_len)) {
248 			send_fragment(fd_raw, addr, alen, offset, ipv6);
249 			offset += max_frag_len;
250 		}
251 		return;
252 	}
253 
254 	/* Occasionally test IPv4 "runs" (see net/ipv4/ip_fragment.c) */
255 	if (!cfg_overlap && (rand() % 100 < 20) &&
256 			(payload_len > 9 * max_frag_len)) {
257 		offset = 6 * max_frag_len;
258 		while (offset < (UDP_HLEN + payload_len)) {
259 			send_fragment(fd_raw, addr, alen, offset, ipv6);
260 			offset += max_frag_len;
261 		}
262 		offset = 3 * max_frag_len;
263 		while (offset < 6 * max_frag_len) {
264 			send_fragment(fd_raw, addr, alen, offset, ipv6);
265 			offset += max_frag_len;
266 		}
267 		offset = 0;
268 		while (offset < 3 * max_frag_len) {
269 			send_fragment(fd_raw, addr, alen, offset, ipv6);
270 			offset += max_frag_len;
271 		}
272 		return;
273 	}
274 
275 	/* Odd fragments. */
276 	offset = max_frag_len;
277 	while (offset < (UDP_HLEN + payload_len)) {
278 		send_fragment(fd_raw, addr, alen, offset, ipv6);
279 		/* IPv4 ignores duplicates, so randomly send a duplicate. */
280 		if (rand() % 100 == 1)
281 			send_fragment(fd_raw, addr, alen, offset, ipv6);
282 		offset += 2 * max_frag_len;
283 	}
284 
285 	if (cfg_overlap) {
286 		/* Send an extra random fragment.
287 		 *
288 		 * Duplicates and some fragments completely inside
289 		 * previously sent fragments are dropped/ignored. So
290 		 * random offset and frag_len can result in a dropped
291 		 * fragment instead of a dropped queue/packet. Thus we
292 		 * hard-code offset and frag_len.
293 		 */
294 		if (max_frag_len * 4 < payload_len || max_frag_len < 16) {
295 			/* not enough payload for random offset and frag_len. */
296 			offset = 8;
297 			frag_len = UDP_HLEN + max_frag_len;
298 		} else {
299 			offset = rand() % (payload_len / 2);
300 			frag_len = 2 * max_frag_len + 1 + rand() % 256;
301 		}
302 		if (ipv6) {
303 			struct ip6_frag *fraghdr = (struct ip6_frag *)(ip_frame + IP6_HLEN);
304 			/* sendto() returns EINVAL if offset + frag_len is too small. */
305 			/* In IPv6 if !!(frag_len % 8), the fragment is dropped. */
306 			frag_len &= ~0x7;
307 			fraghdr->ip6f_offlg = htons(offset / 8 | IP6_MF);
308 			ip6hdr->ip6_plen = htons(frag_len);
309 			frag_len += IP6_HLEN;
310 		} else {
311 			frag_len += IP4_HLEN;
312 			iphdr->ip_off = htons(offset / 8 | IP4_MF);
313 			iphdr->ip_len = htons(frag_len);
314 		}
315 		res = sendto(fd_raw, ip_frame, frag_len, 0, addr, alen);
316 		if (res < 0 && errno != EPERM)
317 			error(1, errno, "sendto overlap: %d", frag_len);
318 		if (res >= 0 && res != frag_len)
319 			error(1, 0, "sendto overlap: %d vs %d", (int)res, frag_len);
320 		frag_counter++;
321 	}
322 
323 	/* Event fragments. */
324 	offset = 0;
325 	while (offset < (UDP_HLEN + payload_len)) {
326 		send_fragment(fd_raw, addr, alen, offset, ipv6);
327 		/* IPv4 ignores duplicates, so randomly send a duplicate. */
328 		if (rand() % 100 == 1)
329 			send_fragment(fd_raw, addr, alen, offset, ipv6);
330 		offset += 2 * max_frag_len;
331 	}
332 }
333 
334 static void run_test(struct sockaddr *addr, socklen_t alen, bool ipv6)
335 {
336 	int fd_tx_raw, fd_rx_udp;
337 	/* Frag queue timeout is set to one second in the calling script;
338 	 * socket timeout should be just a bit longer to avoid tests interfering
339 	 * with each other.
340 	 */
341 	struct timeval tv = { .tv_sec = 1, .tv_usec = 10 };
342 	int idx;
343 	int min_frag_len = 8;
344 
345 	/* Initialize the payload. */
346 	for (idx = 0; idx < MSG_LEN_MAX; ++idx)
347 		udp_payload[idx] = idx % 256;
348 
349 	/* Open sockets. */
350 	fd_tx_raw = socket(addr->sa_family, SOCK_RAW, IPPROTO_RAW);
351 	if (fd_tx_raw == -1)
352 		error(1, errno, "socket tx_raw");
353 
354 	fd_rx_udp = socket(addr->sa_family, SOCK_DGRAM, 0);
355 	if (fd_rx_udp == -1)
356 		error(1, errno, "socket rx_udp");
357 	if (bind(fd_rx_udp, addr, alen))
358 		error(1, errno, "bind");
359 	/* Fail fast. */
360 	if (setsockopt(fd_rx_udp, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)))
361 		error(1, errno, "setsockopt rcv timeout");
362 
363 	for (payload_len = min_frag_len; payload_len < MSG_LEN_MAX;
364 			payload_len += (rand() % 4096)) {
365 		if (cfg_verbose)
366 			printf("payload_len: %d\n", payload_len);
367 
368 		if (cfg_overlap) {
369 			/* With overlaps, one send/receive pair below takes
370 			 * at least one second (== timeout) to run, so there
371 			 * is not enough test time to run a nested loop:
372 			 * the full overlap test takes 20-30 seconds.
373 			 */
374 			max_frag_len = min_frag_len +
375 				rand() % (1500 - FRAG_HLEN - min_frag_len);
376 			send_udp_frags(fd_tx_raw, addr, alen, ipv6);
377 			recv_validate_udp(fd_rx_udp);
378 		} else {
379 			/* Without overlaps, each packet reassembly (== one
380 			 * send/receive pair below) takes very little time to
381 			 * run, so we can easily afford more thourough testing
382 			 * with a nested loop: the full non-overlap test takes
383 			 * less than one second).
384 			 */
385 			max_frag_len = min_frag_len;
386 			do {
387 				send_udp_frags(fd_tx_raw, addr, alen, ipv6);
388 				recv_validate_udp(fd_rx_udp);
389 				max_frag_len += 8 * (rand() % 8);
390 			} while (max_frag_len < (1500 - FRAG_HLEN) &&
391 				 max_frag_len <= payload_len);
392 		}
393 	}
394 
395 	/* Cleanup. */
396 	if (close(fd_tx_raw))
397 		error(1, errno, "close tx_raw");
398 	if (close(fd_rx_udp))
399 		error(1, errno, "close rx_udp");
400 
401 	if (cfg_verbose)
402 		printf("processed %d messages, %d fragments\n",
403 			msg_counter, frag_counter);
404 
405 	fprintf(stderr, "PASS\n");
406 }
407 
408 
409 static void run_test_v4(void)
410 {
411 	struct sockaddr_in addr = {0};
412 
413 	addr.sin_family = AF_INET;
414 	addr.sin_port = htons(cfg_port);
415 	addr.sin_addr = addr4;
416 
417 	run_test((void *)&addr, sizeof(addr), false /* !ipv6 */);
418 }
419 
420 static void run_test_v6(void)
421 {
422 	struct sockaddr_in6 addr = {0};
423 
424 	addr.sin6_family = AF_INET6;
425 	addr.sin6_port = htons(cfg_port);
426 	addr.sin6_addr = addr6;
427 
428 	run_test((void *)&addr, sizeof(addr), true /* ipv6 */);
429 }
430 
431 static void parse_opts(int argc, char **argv)
432 {
433 	int c;
434 
435 	while ((c = getopt(argc, argv, "46opv")) != -1) {
436 		switch (c) {
437 		case '4':
438 			cfg_do_ipv4 = true;
439 			break;
440 		case '6':
441 			cfg_do_ipv6 = true;
442 			break;
443 		case 'o':
444 			cfg_overlap = true;
445 			break;
446 		case 'p':
447 			cfg_permissive = true;
448 			break;
449 		case 'v':
450 			cfg_verbose = true;
451 			break;
452 		default:
453 			error(1, 0, "%s: parse error", argv[0]);
454 		}
455 	}
456 }
457 
458 int main(int argc, char **argv)
459 {
460 	parse_opts(argc, argv);
461 	seed = time(NULL);
462 	srand(seed);
463 	/* Print the seed to track/reproduce potential failures. */
464 	printf("seed = %d\n", seed);
465 
466 	if (cfg_do_ipv4)
467 		run_test_v4();
468 	if (cfg_do_ipv6)
469 		run_test_v6();
470 
471 	return 0;
472 }
473