1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright 2013 Google Inc. 4 * Author: Willem de Bruijn (willemb@google.com) 5 * 6 * A basic test of packet socket fanout behavior. 7 * 8 * Control: 9 * - create fanout fails as expected with illegal flag combinations 10 * - join fanout fails as expected with diverging types or flags 11 * 12 * Datapath: 13 * Open a pair of packet sockets and a pair of INET sockets, send a known 14 * number of packets across the two INET sockets and count the number of 15 * packets enqueued onto the two packet sockets. 16 * 17 * The test currently runs for 18 * - PACKET_FANOUT_HASH 19 * - PACKET_FANOUT_HASH with PACKET_FANOUT_FLAG_ROLLOVER 20 * - PACKET_FANOUT_LB 21 * - PACKET_FANOUT_CPU 22 * - PACKET_FANOUT_ROLLOVER 23 * - PACKET_FANOUT_CBPF 24 * - PACKET_FANOUT_EBPF 25 * 26 * Todo: 27 * - functionality: PACKET_FANOUT_FLAG_DEFRAG 28 */ 29 30 #define _GNU_SOURCE /* for sched_setaffinity */ 31 32 #include <arpa/inet.h> 33 #include <errno.h> 34 #include <fcntl.h> 35 #include <linux/unistd.h> /* for __NR_bpf */ 36 #include <linux/filter.h> 37 #include <linux/bpf.h> 38 #include <linux/if_packet.h> 39 #include <net/if.h> 40 #include <net/ethernet.h> 41 #include <netinet/ip.h> 42 #include <netinet/udp.h> 43 #include <poll.h> 44 #include <sched.h> 45 #include <stdint.h> 46 #include <stdio.h> 47 #include <stdlib.h> 48 #include <string.h> 49 #include <sys/mman.h> 50 #include <sys/socket.h> 51 #include <sys/stat.h> 52 #include <sys/types.h> 53 #include <unistd.h> 54 55 #include "psock_lib.h" 56 57 #define RING_NUM_FRAMES 20 58 59 static uint32_t cfg_max_num_members; 60 61 /* Open a socket in a given fanout mode. 62 * @return -1 if mode is bad, a valid socket otherwise */ 63 static int sock_fanout_open(uint16_t typeflags, uint16_t group_id) 64 { 65 struct sockaddr_ll addr = {0}; 66 struct fanout_args args; 67 int fd, val, err; 68 69 fd = socket(PF_PACKET, SOCK_RAW, 0); 70 if (fd < 0) { 71 perror("socket packet"); 72 exit(1); 73 } 74 75 pair_udp_setfilter(fd); 76 77 addr.sll_family = AF_PACKET; 78 addr.sll_protocol = htons(ETH_P_IP); 79 addr.sll_ifindex = if_nametoindex("lo"); 80 if (addr.sll_ifindex == 0) { 81 perror("if_nametoindex"); 82 exit(1); 83 } 84 if (bind(fd, (void *) &addr, sizeof(addr))) { 85 perror("bind packet"); 86 exit(1); 87 } 88 89 if (cfg_max_num_members) { 90 args.id = group_id; 91 args.type_flags = typeflags; 92 args.max_num_members = cfg_max_num_members; 93 err = setsockopt(fd, SOL_PACKET, PACKET_FANOUT, &args, 94 sizeof(args)); 95 } else { 96 val = (((int) typeflags) << 16) | group_id; 97 err = setsockopt(fd, SOL_PACKET, PACKET_FANOUT, &val, 98 sizeof(val)); 99 } 100 if (err) { 101 if (close(fd)) { 102 perror("close packet"); 103 exit(1); 104 } 105 return -1; 106 } 107 108 return fd; 109 } 110 111 static void sock_fanout_set_cbpf(int fd) 112 { 113 struct sock_filter bpf_filter[] = { 114 BPF_STMT(BPF_LD | BPF_B | BPF_ABS, 80), /* ldb [80] */ 115 BPF_STMT(BPF_RET | BPF_A, 0), /* ret A */ 116 }; 117 struct sock_fprog bpf_prog; 118 119 bpf_prog.filter = bpf_filter; 120 bpf_prog.len = sizeof(bpf_filter) / sizeof(struct sock_filter); 121 122 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &bpf_prog, 123 sizeof(bpf_prog))) { 124 perror("fanout data cbpf"); 125 exit(1); 126 } 127 } 128 129 static void sock_fanout_getopts(int fd, uint16_t *typeflags, uint16_t *group_id) 130 { 131 int sockopt; 132 socklen_t sockopt_len = sizeof(sockopt); 133 134 if (getsockopt(fd, SOL_PACKET, PACKET_FANOUT, 135 &sockopt, &sockopt_len)) { 136 perror("failed to getsockopt"); 137 exit(1); 138 } 139 *typeflags = sockopt >> 16; 140 *group_id = sockopt & 0xfffff; 141 } 142 143 static void sock_fanout_set_ebpf(int fd) 144 { 145 static char log_buf[65536]; 146 147 const int len_off = __builtin_offsetof(struct __sk_buff, len); 148 struct bpf_insn prog[] = { 149 { BPF_ALU64 | BPF_MOV | BPF_X, 6, 1, 0, 0 }, 150 { BPF_LDX | BPF_W | BPF_MEM, 0, 6, len_off, 0 }, 151 { BPF_JMP | BPF_JGE | BPF_K, 0, 0, 1, DATA_LEN }, 152 { BPF_JMP | BPF_JA | BPF_K, 0, 0, 4, 0 }, 153 { BPF_LD | BPF_B | BPF_ABS, 0, 0, 0, 0x50 }, 154 { BPF_JMP | BPF_JEQ | BPF_K, 0, 0, 2, DATA_CHAR }, 155 { BPF_JMP | BPF_JEQ | BPF_K, 0, 0, 1, DATA_CHAR_1 }, 156 { BPF_ALU | BPF_MOV | BPF_K, 0, 0, 0, 0 }, 157 { BPF_JMP | BPF_EXIT, 0, 0, 0, 0 } 158 }; 159 union bpf_attr attr; 160 int pfd; 161 162 memset(&attr, 0, sizeof(attr)); 163 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; 164 attr.insns = (unsigned long) prog; 165 attr.insn_cnt = sizeof(prog) / sizeof(prog[0]); 166 attr.license = (unsigned long) "GPL"; 167 attr.log_buf = (unsigned long) log_buf, 168 attr.log_size = sizeof(log_buf), 169 attr.log_level = 1, 170 171 pfd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr)); 172 if (pfd < 0) { 173 perror("bpf"); 174 fprintf(stderr, "bpf verifier:\n%s\n", log_buf); 175 exit(1); 176 } 177 178 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &pfd, sizeof(pfd))) { 179 perror("fanout data ebpf"); 180 exit(1); 181 } 182 183 if (close(pfd)) { 184 perror("close ebpf"); 185 exit(1); 186 } 187 } 188 189 static char *sock_fanout_open_ring(int fd) 190 { 191 struct tpacket_req req = { 192 .tp_block_size = getpagesize(), 193 .tp_frame_size = getpagesize(), 194 .tp_block_nr = RING_NUM_FRAMES, 195 .tp_frame_nr = RING_NUM_FRAMES, 196 }; 197 char *ring; 198 int val = TPACKET_V2; 199 200 if (setsockopt(fd, SOL_PACKET, PACKET_VERSION, (void *) &val, 201 sizeof(val))) { 202 perror("packetsock ring setsockopt version"); 203 exit(1); 204 } 205 if (setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *) &req, 206 sizeof(req))) { 207 perror("packetsock ring setsockopt"); 208 exit(1); 209 } 210 211 ring = mmap(0, req.tp_block_size * req.tp_block_nr, 212 PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); 213 if (ring == MAP_FAILED) { 214 perror("packetsock ring mmap"); 215 exit(1); 216 } 217 218 return ring; 219 } 220 221 static int sock_fanout_read_ring(int fd, void *ring) 222 { 223 struct tpacket2_hdr *header = ring; 224 int count = 0; 225 226 while (count < RING_NUM_FRAMES && header->tp_status & TP_STATUS_USER) { 227 count++; 228 header = ring + (count * getpagesize()); 229 } 230 231 return count; 232 } 233 234 static int sock_fanout_read(int fds[], char *rings[], const int expect[]) 235 { 236 int ret[2]; 237 238 ret[0] = sock_fanout_read_ring(fds[0], rings[0]); 239 ret[1] = sock_fanout_read_ring(fds[1], rings[1]); 240 241 fprintf(stderr, "info: count=%d,%d, expect=%d,%d\n", 242 ret[0], ret[1], expect[0], expect[1]); 243 244 if ((!(ret[0] == expect[0] && ret[1] == expect[1])) && 245 (!(ret[0] == expect[1] && ret[1] == expect[0]))) { 246 fprintf(stderr, "warning: incorrect queue lengths\n"); 247 return 1; 248 } 249 250 return 0; 251 } 252 253 /* Test illegal mode + flag combination */ 254 static void test_control_single(void) 255 { 256 fprintf(stderr, "test: control single socket\n"); 257 258 if (sock_fanout_open(PACKET_FANOUT_ROLLOVER | 259 PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) { 260 fprintf(stderr, "ERROR: opened socket with dual rollover\n"); 261 exit(1); 262 } 263 } 264 265 /* Test illegal group with different modes or flags */ 266 static void test_control_group(void) 267 { 268 int fds[2]; 269 270 fprintf(stderr, "test: control multiple sockets\n"); 271 272 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH, 0); 273 if (fds[0] == -1) { 274 fprintf(stderr, "ERROR: failed to open HASH socket\n"); 275 exit(1); 276 } 277 if (sock_fanout_open(PACKET_FANOUT_HASH | 278 PACKET_FANOUT_FLAG_DEFRAG, 0) != -1) { 279 fprintf(stderr, "ERROR: joined group with wrong flag defrag\n"); 280 exit(1); 281 } 282 if (sock_fanout_open(PACKET_FANOUT_HASH | 283 PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) { 284 fprintf(stderr, "ERROR: joined group with wrong flag ro\n"); 285 exit(1); 286 } 287 if (sock_fanout_open(PACKET_FANOUT_CPU, 0) != -1) { 288 fprintf(stderr, "ERROR: joined group with wrong mode\n"); 289 exit(1); 290 } 291 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, 0); 292 if (fds[1] == -1) { 293 fprintf(stderr, "ERROR: failed to join group\n"); 294 exit(1); 295 } 296 if (close(fds[1]) || close(fds[0])) { 297 fprintf(stderr, "ERROR: closing sockets\n"); 298 exit(1); 299 } 300 } 301 302 /* Test illegal max_num_members values */ 303 static void test_control_group_max_num_members(void) 304 { 305 int fds[3]; 306 307 fprintf(stderr, "test: control multiple sockets, max_num_members\n"); 308 309 /* expected failure on greater than PACKET_FANOUT_MAX */ 310 cfg_max_num_members = (1 << 16) + 1; 311 if (sock_fanout_open(PACKET_FANOUT_HASH, 0) != -1) { 312 fprintf(stderr, "ERROR: max_num_members > PACKET_FANOUT_MAX\n"); 313 exit(1); 314 } 315 316 cfg_max_num_members = 256; 317 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH, 0); 318 if (fds[0] == -1) { 319 fprintf(stderr, "ERROR: failed open\n"); 320 exit(1); 321 } 322 323 /* expected failure on joining group with different max_num_members */ 324 cfg_max_num_members = 257; 325 if (sock_fanout_open(PACKET_FANOUT_HASH, 0) != -1) { 326 fprintf(stderr, "ERROR: set different max_num_members\n"); 327 exit(1); 328 } 329 330 /* success on joining group with same max_num_members */ 331 cfg_max_num_members = 256; 332 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, 0); 333 if (fds[1] == -1) { 334 fprintf(stderr, "ERROR: failed to join group\n"); 335 exit(1); 336 } 337 338 /* success on joining group with max_num_members unspecified */ 339 cfg_max_num_members = 0; 340 fds[2] = sock_fanout_open(PACKET_FANOUT_HASH, 0); 341 if (fds[2] == -1) { 342 fprintf(stderr, "ERROR: failed to join group\n"); 343 exit(1); 344 } 345 346 if (close(fds[2]) || close(fds[1]) || close(fds[0])) { 347 fprintf(stderr, "ERROR: closing sockets\n"); 348 exit(1); 349 } 350 } 351 352 /* Test creating a unique fanout group ids */ 353 static void test_unique_fanout_group_ids(void) 354 { 355 int fds[3]; 356 uint16_t typeflags, first_group_id, second_group_id; 357 358 fprintf(stderr, "test: unique ids\n"); 359 360 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH | 361 PACKET_FANOUT_FLAG_UNIQUEID, 0); 362 if (fds[0] == -1) { 363 fprintf(stderr, "ERROR: failed to create a unique id group.\n"); 364 exit(1); 365 } 366 367 sock_fanout_getopts(fds[0], &typeflags, &first_group_id); 368 if (typeflags != PACKET_FANOUT_HASH) { 369 fprintf(stderr, "ERROR: unexpected typeflags %x\n", typeflags); 370 exit(1); 371 } 372 373 if (sock_fanout_open(PACKET_FANOUT_CPU, first_group_id) != -1) { 374 fprintf(stderr, "ERROR: joined group with wrong type.\n"); 375 exit(1); 376 } 377 378 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, first_group_id); 379 if (fds[1] == -1) { 380 fprintf(stderr, 381 "ERROR: failed to join previously created group.\n"); 382 exit(1); 383 } 384 385 fds[2] = sock_fanout_open(PACKET_FANOUT_HASH | 386 PACKET_FANOUT_FLAG_UNIQUEID, 0); 387 if (fds[2] == -1) { 388 fprintf(stderr, 389 "ERROR: failed to create a second unique id group.\n"); 390 exit(1); 391 } 392 393 sock_fanout_getopts(fds[2], &typeflags, &second_group_id); 394 if (sock_fanout_open(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_UNIQUEID, 395 second_group_id) != -1) { 396 fprintf(stderr, 397 "ERROR: specified a group id when requesting unique id\n"); 398 exit(1); 399 } 400 401 if (close(fds[0]) || close(fds[1]) || close(fds[2])) { 402 fprintf(stderr, "ERROR: closing sockets\n"); 403 exit(1); 404 } 405 } 406 407 static int test_datapath(uint16_t typeflags, int port_off, 408 const int expect1[], const int expect2[]) 409 { 410 const int expect0[] = { 0, 0 }; 411 char *rings[2]; 412 uint8_t type = typeflags & 0xFF; 413 int fds[2], fds_udp[2][2], ret; 414 415 fprintf(stderr, "\ntest: datapath 0x%hx ports %hu,%hu\n", 416 typeflags, (uint16_t)PORT_BASE, 417 (uint16_t)(PORT_BASE + port_off)); 418 419 fds[0] = sock_fanout_open(typeflags, 0); 420 fds[1] = sock_fanout_open(typeflags, 0); 421 if (fds[0] == -1 || fds[1] == -1) { 422 fprintf(stderr, "ERROR: failed open\n"); 423 exit(1); 424 } 425 if (type == PACKET_FANOUT_CBPF) 426 sock_fanout_set_cbpf(fds[0]); 427 else if (type == PACKET_FANOUT_EBPF) 428 sock_fanout_set_ebpf(fds[0]); 429 430 rings[0] = sock_fanout_open_ring(fds[0]); 431 rings[1] = sock_fanout_open_ring(fds[1]); 432 pair_udp_open(fds_udp[0], PORT_BASE); 433 pair_udp_open(fds_udp[1], PORT_BASE + port_off); 434 sock_fanout_read(fds, rings, expect0); 435 436 /* Send data, but not enough to overflow a queue */ 437 pair_udp_send(fds_udp[0], 15); 438 pair_udp_send_char(fds_udp[1], 5, DATA_CHAR_1); 439 ret = sock_fanout_read(fds, rings, expect1); 440 441 /* Send more data, overflow the queue */ 442 pair_udp_send_char(fds_udp[0], 15, DATA_CHAR_1); 443 /* TODO: ensure consistent order between expect1 and expect2 */ 444 ret |= sock_fanout_read(fds, rings, expect2); 445 446 if (munmap(rings[1], RING_NUM_FRAMES * getpagesize()) || 447 munmap(rings[0], RING_NUM_FRAMES * getpagesize())) { 448 fprintf(stderr, "close rings\n"); 449 exit(1); 450 } 451 if (close(fds_udp[1][1]) || close(fds_udp[1][0]) || 452 close(fds_udp[0][1]) || close(fds_udp[0][0]) || 453 close(fds[1]) || close(fds[0])) { 454 fprintf(stderr, "close datapath\n"); 455 exit(1); 456 } 457 458 return ret; 459 } 460 461 static int set_cpuaffinity(int cpuid) 462 { 463 cpu_set_t mask; 464 465 CPU_ZERO(&mask); 466 CPU_SET(cpuid, &mask); 467 if (sched_setaffinity(0, sizeof(mask), &mask)) { 468 if (errno != EINVAL) { 469 fprintf(stderr, "setaffinity %d\n", cpuid); 470 exit(1); 471 } 472 return 1; 473 } 474 475 return 0; 476 } 477 478 int main(int argc, char **argv) 479 { 480 const int expect_hash[2][2] = { { 15, 5 }, { 20, 5 } }; 481 const int expect_hash_rb[2][2] = { { 15, 5 }, { 20, 15 } }; 482 const int expect_lb[2][2] = { { 10, 10 }, { 18, 17 } }; 483 const int expect_rb[2][2] = { { 15, 5 }, { 20, 15 } }; 484 const int expect_cpu0[2][2] = { { 20, 0 }, { 20, 0 } }; 485 const int expect_cpu1[2][2] = { { 0, 20 }, { 0, 20 } }; 486 const int expect_bpf[2][2] = { { 15, 5 }, { 15, 20 } }; 487 const int expect_uniqueid[2][2] = { { 20, 20}, { 20, 20 } }; 488 int port_off = 2, tries = 20, ret; 489 490 test_control_single(); 491 test_control_group(); 492 test_control_group_max_num_members(); 493 test_unique_fanout_group_ids(); 494 495 /* PACKET_FANOUT_MAX */ 496 cfg_max_num_members = 1 << 16; 497 /* find a set of ports that do not collide onto the same socket */ 498 ret = test_datapath(PACKET_FANOUT_HASH, port_off, 499 expect_hash[0], expect_hash[1]); 500 while (ret) { 501 fprintf(stderr, "info: trying alternate ports (%d)\n", tries); 502 ret = test_datapath(PACKET_FANOUT_HASH, ++port_off, 503 expect_hash[0], expect_hash[1]); 504 if (!--tries) { 505 fprintf(stderr, "too many collisions\n"); 506 return 1; 507 } 508 } 509 510 ret |= test_datapath(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_ROLLOVER, 511 port_off, expect_hash_rb[0], expect_hash_rb[1]); 512 ret |= test_datapath(PACKET_FANOUT_LB, 513 port_off, expect_lb[0], expect_lb[1]); 514 ret |= test_datapath(PACKET_FANOUT_ROLLOVER, 515 port_off, expect_rb[0], expect_rb[1]); 516 517 ret |= test_datapath(PACKET_FANOUT_CBPF, 518 port_off, expect_bpf[0], expect_bpf[1]); 519 ret |= test_datapath(PACKET_FANOUT_EBPF, 520 port_off, expect_bpf[0], expect_bpf[1]); 521 522 set_cpuaffinity(0); 523 ret |= test_datapath(PACKET_FANOUT_CPU, port_off, 524 expect_cpu0[0], expect_cpu0[1]); 525 if (!set_cpuaffinity(1)) 526 /* TODO: test that choice alternates with previous */ 527 ret |= test_datapath(PACKET_FANOUT_CPU, port_off, 528 expect_cpu1[0], expect_cpu1[1]); 529 530 ret |= test_datapath(PACKET_FANOUT_FLAG_UNIQUEID, port_off, 531 expect_uniqueid[0], expect_uniqueid[1]); 532 533 if (ret) 534 return 1; 535 536 printf("OK. All tests passed\n"); 537 return 0; 538 } 539