1 /* 2 * Copyright 2013 Google Inc. 3 * Author: Willem de Bruijn (willemb@google.com) 4 * 5 * A basic test of packet socket fanout behavior. 6 * 7 * Control: 8 * - create fanout fails as expected with illegal flag combinations 9 * - join fanout fails as expected with diverging types or flags 10 * 11 * Datapath: 12 * Open a pair of packet sockets and a pair of INET sockets, send a known 13 * number of packets across the two INET sockets and count the number of 14 * packets enqueued onto the two packet sockets. 15 * 16 * The test currently runs for 17 * - PACKET_FANOUT_HASH 18 * - PACKET_FANOUT_HASH with PACKET_FANOUT_FLAG_ROLLOVER 19 * - PACKET_FANOUT_LB 20 * - PACKET_FANOUT_CPU 21 * - PACKET_FANOUT_ROLLOVER 22 * 23 * Todo: 24 * - functionality: PACKET_FANOUT_FLAG_DEFRAG 25 * 26 * License (GPLv2): 27 * 28 * This program is free software; you can redistribute it and/or modify it 29 * under the terms and conditions of the GNU General Public License, 30 * version 2, as published by the Free Software Foundation. 31 * 32 * This program is distributed in the hope it will be useful, but WITHOUT 33 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 34 * FITNESS FOR A PARTICULAR PURPOSE. * See the GNU General Public License for 35 * more details. 36 * 37 * You should have received a copy of the GNU General Public License along with 38 * this program; if not, write to the Free Software Foundation, Inc., 39 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. 40 */ 41 42 #define _GNU_SOURCE /* for sched_setaffinity */ 43 44 #include <arpa/inet.h> 45 #include <errno.h> 46 #include <fcntl.h> 47 #include <linux/filter.h> 48 #include <linux/if_packet.h> 49 #include <net/ethernet.h> 50 #include <netinet/ip.h> 51 #include <netinet/udp.h> 52 #include <poll.h> 53 #include <sched.h> 54 #include <stdint.h> 55 #include <stdio.h> 56 #include <stdlib.h> 57 #include <string.h> 58 #include <sys/mman.h> 59 #include <sys/socket.h> 60 #include <sys/stat.h> 61 #include <sys/types.h> 62 #include <unistd.h> 63 64 #define DATA_LEN 100 65 #define DATA_CHAR 'a' 66 #define RING_NUM_FRAMES 20 67 #define PORT_BASE 8000 68 69 static void pair_udp_open(int fds[], uint16_t port) 70 { 71 struct sockaddr_in saddr, daddr; 72 73 fds[0] = socket(PF_INET, SOCK_DGRAM, 0); 74 fds[1] = socket(PF_INET, SOCK_DGRAM, 0); 75 if (fds[0] == -1 || fds[1] == -1) { 76 fprintf(stderr, "ERROR: socket dgram\n"); 77 exit(1); 78 } 79 80 memset(&saddr, 0, sizeof(saddr)); 81 saddr.sin_family = AF_INET; 82 saddr.sin_port = htons(port); 83 saddr.sin_addr.s_addr = htonl(INADDR_LOOPBACK); 84 85 memset(&daddr, 0, sizeof(daddr)); 86 daddr.sin_family = AF_INET; 87 daddr.sin_port = htons(port + 1); 88 daddr.sin_addr.s_addr = htonl(INADDR_LOOPBACK); 89 90 /* must bind both to get consistent hash result */ 91 if (bind(fds[1], (void *) &daddr, sizeof(daddr))) { 92 perror("bind"); 93 exit(1); 94 } 95 if (bind(fds[0], (void *) &saddr, sizeof(saddr))) { 96 perror("bind"); 97 exit(1); 98 } 99 if (connect(fds[0], (void *) &daddr, sizeof(daddr))) { 100 perror("connect"); 101 exit(1); 102 } 103 } 104 105 static void pair_udp_send(int fds[], int num) 106 { 107 char buf[DATA_LEN], rbuf[DATA_LEN]; 108 109 memset(buf, DATA_CHAR, sizeof(buf)); 110 while (num--) { 111 /* Should really handle EINTR and EAGAIN */ 112 if (write(fds[0], buf, sizeof(buf)) != sizeof(buf)) { 113 fprintf(stderr, "ERROR: send failed left=%d\n", num); 114 exit(1); 115 } 116 if (read(fds[1], rbuf, sizeof(rbuf)) != sizeof(rbuf)) { 117 fprintf(stderr, "ERROR: recv failed left=%d\n", num); 118 exit(1); 119 } 120 if (memcmp(buf, rbuf, sizeof(buf))) { 121 fprintf(stderr, "ERROR: data failed left=%d\n", num); 122 exit(1); 123 } 124 } 125 } 126 127 static void sock_fanout_setfilter(int fd) 128 { 129 struct sock_filter bpf_filter[] = { 130 { 0x80, 0, 0, 0x00000000 }, /* LD pktlen */ 131 { 0x35, 0, 5, DATA_LEN }, /* JGE DATA_LEN [f goto nomatch]*/ 132 { 0x30, 0, 0, 0x00000050 }, /* LD ip[80] */ 133 { 0x15, 0, 3, DATA_CHAR }, /* JEQ DATA_CHAR [f goto nomatch]*/ 134 { 0x30, 0, 0, 0x00000051 }, /* LD ip[81] */ 135 { 0x15, 0, 1, DATA_CHAR }, /* JEQ DATA_CHAR [f goto nomatch]*/ 136 { 0x6, 0, 0, 0x00000060 }, /* RET match */ 137 /* nomatch */ { 0x6, 0, 0, 0x00000000 }, /* RET no match */ 138 }; 139 struct sock_fprog bpf_prog; 140 141 bpf_prog.filter = bpf_filter; 142 bpf_prog.len = sizeof(bpf_filter) / sizeof(struct sock_filter); 143 if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &bpf_prog, 144 sizeof(bpf_prog))) { 145 perror("setsockopt SO_ATTACH_FILTER"); 146 exit(1); 147 } 148 } 149 150 /* Open a socket in a given fanout mode. 151 * @return -1 if mode is bad, a valid socket otherwise */ 152 static int sock_fanout_open(uint16_t typeflags, int num_packets) 153 { 154 int fd, val; 155 156 fd = socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_IP)); 157 if (fd < 0) { 158 perror("socket packet"); 159 exit(1); 160 } 161 162 /* fanout group ID is always 0: tests whether old groups are deleted */ 163 val = ((int) typeflags) << 16; 164 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT, &val, sizeof(val))) { 165 if (close(fd)) { 166 perror("close packet"); 167 exit(1); 168 } 169 return -1; 170 } 171 172 sock_fanout_setfilter(fd); 173 return fd; 174 } 175 176 static char *sock_fanout_open_ring(int fd) 177 { 178 struct tpacket_req req = { 179 .tp_block_size = getpagesize(), 180 .tp_frame_size = getpagesize(), 181 .tp_block_nr = RING_NUM_FRAMES, 182 .tp_frame_nr = RING_NUM_FRAMES, 183 }; 184 char *ring; 185 int val = TPACKET_V2; 186 187 if (setsockopt(fd, SOL_PACKET, PACKET_VERSION, (void *) &val, 188 sizeof(val))) { 189 perror("packetsock ring setsockopt version"); 190 exit(1); 191 } 192 if (setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *) &req, 193 sizeof(req))) { 194 perror("packetsock ring setsockopt"); 195 exit(1); 196 } 197 198 ring = mmap(0, req.tp_block_size * req.tp_block_nr, 199 PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); 200 if (!ring) { 201 fprintf(stderr, "packetsock ring mmap\n"); 202 exit(1); 203 } 204 205 return ring; 206 } 207 208 static int sock_fanout_read_ring(int fd, void *ring) 209 { 210 struct tpacket2_hdr *header = ring; 211 int count = 0; 212 213 while (header->tp_status & TP_STATUS_USER && count < RING_NUM_FRAMES) { 214 count++; 215 header = ring + (count * getpagesize()); 216 } 217 218 return count; 219 } 220 221 static int sock_fanout_read(int fds[], char *rings[], const int expect[]) 222 { 223 int ret[2]; 224 225 ret[0] = sock_fanout_read_ring(fds[0], rings[0]); 226 ret[1] = sock_fanout_read_ring(fds[1], rings[1]); 227 228 fprintf(stderr, "info: count=%d,%d, expect=%d,%d\n", 229 ret[0], ret[1], expect[0], expect[1]); 230 231 if ((!(ret[0] == expect[0] && ret[1] == expect[1])) && 232 (!(ret[0] == expect[1] && ret[1] == expect[0]))) { 233 fprintf(stderr, "ERROR: incorrect queue lengths\n"); 234 return 1; 235 } 236 237 return 0; 238 } 239 240 /* Test illegal mode + flag combination */ 241 static void test_control_single(void) 242 { 243 fprintf(stderr, "test: control single socket\n"); 244 245 if (sock_fanout_open(PACKET_FANOUT_ROLLOVER | 246 PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) { 247 fprintf(stderr, "ERROR: opened socket with dual rollover\n"); 248 exit(1); 249 } 250 } 251 252 /* Test illegal group with different modes or flags */ 253 static void test_control_group(void) 254 { 255 int fds[2]; 256 257 fprintf(stderr, "test: control multiple sockets\n"); 258 259 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH, 20); 260 if (fds[0] == -1) { 261 fprintf(stderr, "ERROR: failed to open HASH socket\n"); 262 exit(1); 263 } 264 if (sock_fanout_open(PACKET_FANOUT_HASH | 265 PACKET_FANOUT_FLAG_DEFRAG, 10) != -1) { 266 fprintf(stderr, "ERROR: joined group with wrong flag defrag\n"); 267 exit(1); 268 } 269 if (sock_fanout_open(PACKET_FANOUT_HASH | 270 PACKET_FANOUT_FLAG_ROLLOVER, 10) != -1) { 271 fprintf(stderr, "ERROR: joined group with wrong flag ro\n"); 272 exit(1); 273 } 274 if (sock_fanout_open(PACKET_FANOUT_CPU, 10) != -1) { 275 fprintf(stderr, "ERROR: joined group with wrong mode\n"); 276 exit(1); 277 } 278 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, 20); 279 if (fds[1] == -1) { 280 fprintf(stderr, "ERROR: failed to join group\n"); 281 exit(1); 282 } 283 if (close(fds[1]) || close(fds[0])) { 284 fprintf(stderr, "ERROR: closing sockets\n"); 285 exit(1); 286 } 287 } 288 289 static int test_datapath(uint16_t typeflags, int port_off, 290 const int expect1[], const int expect2[]) 291 { 292 const int expect0[] = { 0, 0 }; 293 char *rings[2]; 294 int fds[2], fds_udp[2][2], ret; 295 296 fprintf(stderr, "test: datapath 0x%hx\n", typeflags); 297 298 fds[0] = sock_fanout_open(typeflags, 20); 299 fds[1] = sock_fanout_open(typeflags, 20); 300 if (fds[0] == -1 || fds[1] == -1) { 301 fprintf(stderr, "ERROR: failed open\n"); 302 exit(1); 303 } 304 rings[0] = sock_fanout_open_ring(fds[0]); 305 rings[1] = sock_fanout_open_ring(fds[1]); 306 pair_udp_open(fds_udp[0], PORT_BASE); 307 pair_udp_open(fds_udp[1], PORT_BASE + port_off); 308 sock_fanout_read(fds, rings, expect0); 309 310 /* Send data, but not enough to overflow a queue */ 311 pair_udp_send(fds_udp[0], 15); 312 pair_udp_send(fds_udp[1], 5); 313 ret = sock_fanout_read(fds, rings, expect1); 314 315 /* Send more data, overflow the queue */ 316 pair_udp_send(fds_udp[0], 15); 317 /* TODO: ensure consistent order between expect1 and expect2 */ 318 ret |= sock_fanout_read(fds, rings, expect2); 319 320 if (munmap(rings[1], RING_NUM_FRAMES * getpagesize()) || 321 munmap(rings[0], RING_NUM_FRAMES * getpagesize())) { 322 fprintf(stderr, "close rings\n"); 323 exit(1); 324 } 325 if (close(fds_udp[1][1]) || close(fds_udp[1][0]) || 326 close(fds_udp[0][1]) || close(fds_udp[0][0]) || 327 close(fds[1]) || close(fds[0])) { 328 fprintf(stderr, "close datapath\n"); 329 exit(1); 330 } 331 332 return ret; 333 } 334 335 static int set_cpuaffinity(int cpuid) 336 { 337 cpu_set_t mask; 338 339 CPU_ZERO(&mask); 340 CPU_SET(cpuid, &mask); 341 if (sched_setaffinity(0, sizeof(mask), &mask)) { 342 if (errno != EINVAL) { 343 fprintf(stderr, "setaffinity %d\n", cpuid); 344 exit(1); 345 } 346 return 1; 347 } 348 349 return 0; 350 } 351 352 int main(int argc, char **argv) 353 { 354 const int expect_hash[2][2] = { { 15, 5 }, { 20, 5 } }; 355 const int expect_hash_rb[2][2] = { { 15, 5 }, { 20, 15 } }; 356 const int expect_lb[2][2] = { { 10, 10 }, { 18, 17 } }; 357 const int expect_rb[2][2] = { { 20, 0 }, { 20, 15 } }; 358 const int expect_cpu0[2][2] = { { 20, 0 }, { 20, 0 } }; 359 const int expect_cpu1[2][2] = { { 0, 20 }, { 0, 20 } }; 360 int port_off = 2, tries = 5, ret; 361 362 test_control_single(); 363 test_control_group(); 364 365 /* find a set of ports that do not collide onto the same socket */ 366 ret = test_datapath(PACKET_FANOUT_HASH, port_off, 367 expect_hash[0], expect_hash[1]); 368 while (ret && tries--) { 369 fprintf(stderr, "info: trying alternate ports (%d)\n", tries); 370 ret = test_datapath(PACKET_FANOUT_HASH, ++port_off, 371 expect_hash[0], expect_hash[1]); 372 } 373 374 ret |= test_datapath(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_ROLLOVER, 375 port_off, expect_hash_rb[0], expect_hash_rb[1]); 376 ret |= test_datapath(PACKET_FANOUT_LB, 377 port_off, expect_lb[0], expect_lb[1]); 378 ret |= test_datapath(PACKET_FANOUT_ROLLOVER, 379 port_off, expect_rb[0], expect_rb[1]); 380 381 set_cpuaffinity(0); 382 ret |= test_datapath(PACKET_FANOUT_CPU, port_off, 383 expect_cpu0[0], expect_cpu0[1]); 384 if (!set_cpuaffinity(1)) 385 /* TODO: test that choice alternates with previous */ 386 ret |= test_datapath(PACKET_FANOUT_CPU, port_off, 387 expect_cpu1[0], expect_cpu1[1]); 388 389 if (ret) 390 return 1; 391 392 printf("OK. All tests passed\n"); 393 return 0; 394 } 395