ip_defrag.c 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393
  1. // SPDX-License-Identifier: GPL-2.0
  2. #define _GNU_SOURCE
  3. #include <arpa/inet.h>
  4. #include <errno.h>
  5. #include <error.h>
  6. #include <linux/in.h>
  7. #include <netinet/ip.h>
  8. #include <netinet/ip6.h>
  9. #include <netinet/udp.h>
  10. #include <stdbool.h>
  11. #include <stdio.h>
  12. #include <stdlib.h>
  13. #include <string.h>
  14. #include <time.h>
  15. #include <unistd.h>
  16. static bool cfg_do_ipv4;
  17. static bool cfg_do_ipv6;
  18. static bool cfg_verbose;
  19. static bool cfg_overlap;
  20. static unsigned short cfg_port = 9000;
  21. const struct in_addr addr4 = { .s_addr = __constant_htonl(INADDR_LOOPBACK + 2) };
  22. const struct in6_addr addr6 = IN6ADDR_LOOPBACK_INIT;
  23. #define IP4_HLEN (sizeof(struct iphdr))
  24. #define IP6_HLEN (sizeof(struct ip6_hdr))
  25. #define UDP_HLEN (sizeof(struct udphdr))
  26. /* IPv6 fragment header lenth. */
  27. #define FRAG_HLEN 8
  28. static int payload_len;
  29. static int max_frag_len;
  30. #define MSG_LEN_MAX 60000 /* Max UDP payload length. */
  31. #define IP4_MF (1u << 13) /* IPv4 MF flag. */
  32. #define IP6_MF (1) /* IPv6 MF flag. */
  33. #define CSUM_MANGLED_0 (0xffff)
  34. static uint8_t udp_payload[MSG_LEN_MAX];
  35. static uint8_t ip_frame[IP_MAXPACKET];
  36. static uint32_t ip_id = 0xabcd;
  37. static int msg_counter;
  38. static int frag_counter;
  39. static unsigned int seed;
  40. /* Receive a UDP packet. Validate it matches udp_payload. */
  41. static void recv_validate_udp(int fd_udp)
  42. {
  43. ssize_t ret;
  44. static uint8_t recv_buff[MSG_LEN_MAX];
  45. ret = recv(fd_udp, recv_buff, payload_len, 0);
  46. msg_counter++;
  47. if (cfg_overlap) {
  48. if (ret != -1)
  49. error(1, 0, "recv: expected timeout; got %d",
  50. (int)ret);
  51. if (errno != ETIMEDOUT && errno != EAGAIN)
  52. error(1, errno, "recv: expected timeout: %d",
  53. errno);
  54. return; /* OK */
  55. }
  56. if (ret == -1)
  57. error(1, errno, "recv: payload_len = %d max_frag_len = %d",
  58. payload_len, max_frag_len);
  59. if (ret != payload_len)
  60. error(1, 0, "recv: wrong size: %d vs %d", (int)ret, payload_len);
  61. if (memcmp(udp_payload, recv_buff, payload_len))
  62. error(1, 0, "recv: wrong data");
  63. }
  64. static uint32_t raw_checksum(uint8_t *buf, int len, uint32_t sum)
  65. {
  66. int i;
  67. for (i = 0; i < (len & ~1U); i += 2) {
  68. sum += (u_int16_t)ntohs(*((u_int16_t *)(buf + i)));
  69. if (sum > 0xffff)
  70. sum -= 0xffff;
  71. }
  72. if (i < len) {
  73. sum += buf[i] << 8;
  74. if (sum > 0xffff)
  75. sum -= 0xffff;
  76. }
  77. return sum;
  78. }
  79. static uint16_t udp_checksum(struct ip *iphdr, struct udphdr *udphdr)
  80. {
  81. uint32_t sum = 0;
  82. uint16_t res;
  83. sum = raw_checksum((uint8_t *)&iphdr->ip_src, 2 * sizeof(iphdr->ip_src),
  84. IPPROTO_UDP + (uint32_t)(UDP_HLEN + payload_len));
  85. sum = raw_checksum((uint8_t *)udphdr, UDP_HLEN, sum);
  86. sum = raw_checksum((uint8_t *)udp_payload, payload_len, sum);
  87. res = 0xffff & ~sum;
  88. if (res)
  89. return htons(res);
  90. else
  91. return CSUM_MANGLED_0;
  92. }
  93. static uint16_t udp6_checksum(struct ip6_hdr *iphdr, struct udphdr *udphdr)
  94. {
  95. uint32_t sum = 0;
  96. uint16_t res;
  97. sum = raw_checksum((uint8_t *)&iphdr->ip6_src, 2 * sizeof(iphdr->ip6_src),
  98. IPPROTO_UDP);
  99. sum = raw_checksum((uint8_t *)&udphdr->len, sizeof(udphdr->len), sum);
  100. sum = raw_checksum((uint8_t *)udphdr, UDP_HLEN, sum);
  101. sum = raw_checksum((uint8_t *)udp_payload, payload_len, sum);
  102. res = 0xffff & ~sum;
  103. if (res)
  104. return htons(res);
  105. else
  106. return CSUM_MANGLED_0;
  107. }
  108. static void send_fragment(int fd_raw, struct sockaddr *addr, socklen_t alen,
  109. int offset, bool ipv6)
  110. {
  111. int frag_len;
  112. int res;
  113. int payload_offset = offset > 0 ? offset - UDP_HLEN : 0;
  114. uint8_t *frag_start = ipv6 ? ip_frame + IP6_HLEN + FRAG_HLEN :
  115. ip_frame + IP4_HLEN;
  116. if (offset == 0) {
  117. struct udphdr udphdr;
  118. udphdr.source = htons(cfg_port + 1);
  119. udphdr.dest = htons(cfg_port);
  120. udphdr.len = htons(UDP_HLEN + payload_len);
  121. udphdr.check = 0;
  122. if (ipv6)
  123. udphdr.check = udp6_checksum((struct ip6_hdr *)ip_frame, &udphdr);
  124. else
  125. udphdr.check = udp_checksum((struct ip *)ip_frame, &udphdr);
  126. memcpy(frag_start, &udphdr, UDP_HLEN);
  127. }
  128. if (ipv6) {
  129. struct ip6_hdr *ip6hdr = (struct ip6_hdr *)ip_frame;
  130. struct ip6_frag *fraghdr = (struct ip6_frag *)(ip_frame + IP6_HLEN);
  131. if (payload_len - payload_offset <= max_frag_len && offset > 0) {
  132. /* This is the last fragment. */
  133. frag_len = FRAG_HLEN + payload_len - payload_offset;
  134. fraghdr->ip6f_offlg = htons(offset);
  135. } else {
  136. frag_len = FRAG_HLEN + max_frag_len;
  137. fraghdr->ip6f_offlg = htons(offset | IP6_MF);
  138. }
  139. ip6hdr->ip6_plen = htons(frag_len);
  140. if (offset == 0)
  141. memcpy(frag_start + UDP_HLEN, udp_payload,
  142. frag_len - FRAG_HLEN - UDP_HLEN);
  143. else
  144. memcpy(frag_start, udp_payload + payload_offset,
  145. frag_len - FRAG_HLEN);
  146. frag_len += IP6_HLEN;
  147. } else {
  148. struct ip *iphdr = (struct ip *)ip_frame;
  149. if (payload_len - payload_offset <= max_frag_len && offset > 0) {
  150. /* This is the last fragment. */
  151. frag_len = IP4_HLEN + payload_len - payload_offset;
  152. iphdr->ip_off = htons(offset / 8);
  153. } else {
  154. frag_len = IP4_HLEN + max_frag_len;
  155. iphdr->ip_off = htons(offset / 8 | IP4_MF);
  156. }
  157. iphdr->ip_len = htons(frag_len);
  158. if (offset == 0)
  159. memcpy(frag_start + UDP_HLEN, udp_payload,
  160. frag_len - IP4_HLEN - UDP_HLEN);
  161. else
  162. memcpy(frag_start, udp_payload + payload_offset,
  163. frag_len - IP4_HLEN);
  164. }
  165. res = sendto(fd_raw, ip_frame, frag_len, 0, addr, alen);
  166. if (res < 0)
  167. error(1, errno, "send_fragment");
  168. if (res != frag_len)
  169. error(1, 0, "send_fragment: %d vs %d", res, frag_len);
  170. frag_counter++;
  171. }
  172. static void send_udp_frags(int fd_raw, struct sockaddr *addr,
  173. socklen_t alen, bool ipv6)
  174. {
  175. struct ip *iphdr = (struct ip *)ip_frame;
  176. struct ip6_hdr *ip6hdr = (struct ip6_hdr *)ip_frame;
  177. int res;
  178. int offset;
  179. int frag_len;
  180. /* Send the UDP datagram using raw IP fragments: the 0th fragment
  181. * has the UDP header; other fragments are pieces of udp_payload
  182. * split in chunks of frag_len size.
  183. *
  184. * Odd fragments (1st, 3rd, 5th, etc.) are sent out first, then
  185. * even fragments (0th, 2nd, etc.) are sent out.
  186. */
  187. if (ipv6) {
  188. struct ip6_frag *fraghdr = (struct ip6_frag *)(ip_frame + IP6_HLEN);
  189. ((struct sockaddr_in6 *)addr)->sin6_port = 0;
  190. memset(ip6hdr, 0, sizeof(*ip6hdr));
  191. ip6hdr->ip6_flow = htonl(6<<28); /* Version. */
  192. ip6hdr->ip6_nxt = IPPROTO_FRAGMENT;
  193. ip6hdr->ip6_hops = 255;
  194. ip6hdr->ip6_src = addr6;
  195. ip6hdr->ip6_dst = addr6;
  196. fraghdr->ip6f_nxt = IPPROTO_UDP;
  197. fraghdr->ip6f_reserved = 0;
  198. fraghdr->ip6f_ident = htonl(ip_id++);
  199. } else {
  200. memset(iphdr, 0, sizeof(*iphdr));
  201. iphdr->ip_hl = 5;
  202. iphdr->ip_v = 4;
  203. iphdr->ip_tos = 0;
  204. iphdr->ip_id = htons(ip_id++);
  205. iphdr->ip_ttl = 0x40;
  206. iphdr->ip_p = IPPROTO_UDP;
  207. iphdr->ip_src.s_addr = htonl(INADDR_LOOPBACK);
  208. iphdr->ip_dst = addr4;
  209. iphdr->ip_sum = 0;
  210. }
  211. /* Odd fragments. */
  212. offset = max_frag_len;
  213. while (offset < (UDP_HLEN + payload_len)) {
  214. send_fragment(fd_raw, addr, alen, offset, ipv6);
  215. offset += 2 * max_frag_len;
  216. }
  217. if (cfg_overlap) {
  218. /* Send an extra random fragment. */
  219. offset = rand() % (UDP_HLEN + payload_len - 1);
  220. /* sendto() returns EINVAL if offset + frag_len is too small. */
  221. if (ipv6) {
  222. struct ip6_frag *fraghdr = (struct ip6_frag *)(ip_frame + IP6_HLEN);
  223. frag_len = max_frag_len + rand() % 256;
  224. /* In IPv6 if !!(frag_len % 8), the fragment is dropped. */
  225. frag_len &= ~0x7;
  226. fraghdr->ip6f_offlg = htons(offset / 8 | IP6_MF);
  227. ip6hdr->ip6_plen = htons(frag_len);
  228. frag_len += IP6_HLEN;
  229. } else {
  230. frag_len = IP4_HLEN + UDP_HLEN + rand() % 256;
  231. iphdr->ip_off = htons(offset / 8 | IP4_MF);
  232. iphdr->ip_len = htons(frag_len);
  233. }
  234. res = sendto(fd_raw, ip_frame, frag_len, 0, addr, alen);
  235. if (res < 0)
  236. error(1, errno, "sendto overlap");
  237. if (res != frag_len)
  238. error(1, 0, "sendto overlap: %d vs %d", (int)res, frag_len);
  239. frag_counter++;
  240. }
  241. /* Event fragments. */
  242. offset = 0;
  243. while (offset < (UDP_HLEN + payload_len)) {
  244. send_fragment(fd_raw, addr, alen, offset, ipv6);
  245. offset += 2 * max_frag_len;
  246. }
  247. }
  248. static void run_test(struct sockaddr *addr, socklen_t alen, bool ipv6)
  249. {
  250. int fd_tx_raw, fd_rx_udp;
  251. struct timeval tv = { .tv_sec = 0, .tv_usec = 10 * 1000 };
  252. int idx;
  253. int min_frag_len = ipv6 ? 1280 : 8;
  254. /* Initialize the payload. */
  255. for (idx = 0; idx < MSG_LEN_MAX; ++idx)
  256. udp_payload[idx] = idx % 256;
  257. /* Open sockets. */
  258. fd_tx_raw = socket(addr->sa_family, SOCK_RAW, IPPROTO_RAW);
  259. if (fd_tx_raw == -1)
  260. error(1, errno, "socket tx_raw");
  261. fd_rx_udp = socket(addr->sa_family, SOCK_DGRAM, 0);
  262. if (fd_rx_udp == -1)
  263. error(1, errno, "socket rx_udp");
  264. if (bind(fd_rx_udp, addr, alen))
  265. error(1, errno, "bind");
  266. /* Fail fast. */
  267. if (setsockopt(fd_rx_udp, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)))
  268. error(1, errno, "setsockopt rcv timeout");
  269. for (payload_len = min_frag_len; payload_len < MSG_LEN_MAX;
  270. payload_len += (rand() % 4096)) {
  271. if (cfg_verbose)
  272. printf("payload_len: %d\n", payload_len);
  273. max_frag_len = min_frag_len;
  274. do {
  275. send_udp_frags(fd_tx_raw, addr, alen, ipv6);
  276. recv_validate_udp(fd_rx_udp);
  277. max_frag_len += 8 * (rand() % 8);
  278. } while (max_frag_len < (1500 - FRAG_HLEN) && max_frag_len <= payload_len);
  279. }
  280. /* Cleanup. */
  281. if (close(fd_tx_raw))
  282. error(1, errno, "close tx_raw");
  283. if (close(fd_rx_udp))
  284. error(1, errno, "close rx_udp");
  285. if (cfg_verbose)
  286. printf("processed %d messages, %d fragments\n",
  287. msg_counter, frag_counter);
  288. fprintf(stderr, "PASS\n");
  289. }
  290. static void run_test_v4(void)
  291. {
  292. struct sockaddr_in addr = {0};
  293. addr.sin_family = AF_INET;
  294. addr.sin_port = htons(cfg_port);
  295. addr.sin_addr = addr4;
  296. run_test((void *)&addr, sizeof(addr), false /* !ipv6 */);
  297. }
  298. static void run_test_v6(void)
  299. {
  300. struct sockaddr_in6 addr = {0};
  301. addr.sin6_family = AF_INET6;
  302. addr.sin6_port = htons(cfg_port);
  303. addr.sin6_addr = addr6;
  304. run_test((void *)&addr, sizeof(addr), true /* ipv6 */);
  305. }
  306. static void parse_opts(int argc, char **argv)
  307. {
  308. int c;
  309. while ((c = getopt(argc, argv, "46ov")) != -1) {
  310. switch (c) {
  311. case '4':
  312. cfg_do_ipv4 = true;
  313. break;
  314. case '6':
  315. cfg_do_ipv6 = true;
  316. break;
  317. case 'o':
  318. cfg_overlap = true;
  319. break;
  320. case 'v':
  321. cfg_verbose = true;
  322. break;
  323. default:
  324. error(1, 0, "%s: parse error", argv[0]);
  325. }
  326. }
  327. }
  328. int main(int argc, char **argv)
  329. {
  330. parse_opts(argc, argv);
  331. seed = time(NULL);
  332. srand(seed);
  333. /* Print the seed to track/reproduce potential failures. */
  334. printf("seed = %d\n", seed);
  335. if (cfg_do_ipv4)
  336. run_test_v4();
  337. if (cfg_do_ipv6)
  338. run_test_v6();
  339. return 0;
  340. }