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 * - PACKET_FANOUT_CBPF 23 * - PACKET_FANOUT_EBPF 24 * 25 * Todo: 26 * - functionality: PACKET_FANOUT_FLAG_DEFRAG 27 * 28 * License (GPLv2): 29 * 30 * This program is free software; you can redistribute it and/or modify it 31 * under the terms and conditions of the GNU General Public License, 32 * version 2, as published by the Free Software Foundation. 33 * 34 * This program is distributed in the hope it will be useful, but WITHOUT 35 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 36 * FITNESS FOR A PARTICULAR PURPOSE. * See the GNU General Public License for 37 * more details. 38 * 39 * You should have received a copy of the GNU General Public License along with 40 * this program; if not, write to the Free Software Foundation, Inc., 41 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. 42 */ 43 44#define _GNU_SOURCE /* for sched_setaffinity */ 45 46#include <arpa/inet.h> 47#include <errno.h> 48#include <fcntl.h> 49#include <linux/unistd.h> /* for __NR_bpf */ 50#include <linux/filter.h> 51#include <linux/bpf.h> 52#include <linux/if_packet.h> 53#include <net/ethernet.h> 54#include <netinet/ip.h> 55#include <netinet/udp.h> 56#include <poll.h> 57#include <sched.h> 58#include <stdint.h> 59#include <stdio.h> 60#include <stdlib.h> 61#include <string.h> 62#include <sys/mman.h> 63#include <sys/socket.h> 64#include <sys/stat.h> 65#include <sys/types.h> 66#include <unistd.h> 67 68#include "psock_lib.h" 69 70#define RING_NUM_FRAMES 20 71 72/* Open a socket in a given fanout mode. 73 * @return -1 if mode is bad, a valid socket otherwise */ 74static int sock_fanout_open(uint16_t typeflags, uint16_t group_id) 75{ 76 int fd, val; 77 78 fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_IP)); 79 if (fd < 0) { 80 perror("socket packet"); 81 exit(1); 82 } 83 84 val = (((int) typeflags) << 16) | group_id; 85 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT, &val, sizeof(val))) { 86 if (close(fd)) { 87 perror("close packet"); 88 exit(1); 89 } 90 return -1; 91 } 92 93 pair_udp_setfilter(fd); 94 return fd; 95} 96 97static void sock_fanout_set_cbpf(int fd) 98{ 99 struct sock_filter bpf_filter[] = { 100 BPF_STMT(BPF_LD+BPF_B+BPF_ABS, 80), /* ldb [80] */ 101 BPF_STMT(BPF_RET+BPF_A, 0), /* ret A */ 102 }; 103 struct sock_fprog bpf_prog; 104 105 bpf_prog.filter = bpf_filter; 106 bpf_prog.len = sizeof(bpf_filter) / sizeof(struct sock_filter); 107 108 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &bpf_prog, 109 sizeof(bpf_prog))) { 110 perror("fanout data cbpf"); 111 exit(1); 112 } 113} 114 115static void sock_fanout_getopts(int fd, uint16_t *typeflags, uint16_t *group_id) 116{ 117 int sockopt; 118 socklen_t sockopt_len = sizeof(sockopt); 119 120 if (getsockopt(fd, SOL_PACKET, PACKET_FANOUT, 121 &sockopt, &sockopt_len)) { 122 perror("failed to getsockopt"); 123 exit(1); 124 } 125 *typeflags = sockopt >> 16; 126 *group_id = sockopt & 0xfffff; 127} 128 129static void sock_fanout_set_ebpf(int fd) 130{ 131 const int len_off = __builtin_offsetof(struct __sk_buff, len); 132 struct bpf_insn prog[] = { 133 { BPF_ALU64 | BPF_MOV | BPF_X, 6, 1, 0, 0 }, 134 { BPF_LDX | BPF_W | BPF_MEM, 0, 6, len_off, 0 }, 135 { BPF_JMP | BPF_JGE | BPF_K, 0, 0, 1, DATA_LEN }, 136 { BPF_JMP | BPF_JA | BPF_K, 0, 0, 4, 0 }, 137 { BPF_LD | BPF_B | BPF_ABS, 0, 0, 0, 0x50 }, 138 { BPF_JMP | BPF_JEQ | BPF_K, 0, 0, 2, DATA_CHAR }, 139 { BPF_JMP | BPF_JEQ | BPF_K, 0, 0, 1, DATA_CHAR_1 }, 140 { BPF_ALU | BPF_MOV | BPF_K, 0, 0, 0, 0 }, 141 { BPF_JMP | BPF_EXIT, 0, 0, 0, 0 } 142 }; 143 char log_buf[512]; 144 union bpf_attr attr; 145 int pfd; 146 147 memset(&attr, 0, sizeof(attr)); 148 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; 149 attr.insns = (unsigned long) prog; 150 attr.insn_cnt = sizeof(prog) / sizeof(prog[0]); 151 attr.license = (unsigned long) "GPL"; 152 attr.log_buf = (unsigned long) log_buf, 153 attr.log_size = sizeof(log_buf), 154 attr.log_level = 1, 155 156 pfd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr)); 157 if (pfd < 0) { 158 perror("bpf"); 159 fprintf(stderr, "bpf verifier:\n%s\n", log_buf); 160 exit(1); 161 } 162 163 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &pfd, sizeof(pfd))) { 164 perror("fanout data ebpf"); 165 exit(1); 166 } 167 168 if (close(pfd)) { 169 perror("close ebpf"); 170 exit(1); 171 } 172} 173 174static char *sock_fanout_open_ring(int fd) 175{ 176 struct tpacket_req req = { 177 .tp_block_size = getpagesize(), 178 .tp_frame_size = getpagesize(), 179 .tp_block_nr = RING_NUM_FRAMES, 180 .tp_frame_nr = RING_NUM_FRAMES, 181 }; 182 char *ring; 183 int val = TPACKET_V2; 184 185 if (setsockopt(fd, SOL_PACKET, PACKET_VERSION, (void *) &val, 186 sizeof(val))) { 187 perror("packetsock ring setsockopt version"); 188 exit(1); 189 } 190 if (setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *) &req, 191 sizeof(req))) { 192 perror("packetsock ring setsockopt"); 193 exit(1); 194 } 195 196 ring = mmap(0, req.tp_block_size * req.tp_block_nr, 197 PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); 198 if (ring == MAP_FAILED) { 199 perror("packetsock ring mmap"); 200 exit(1); 201 } 202 203 return ring; 204} 205 206static int sock_fanout_read_ring(int fd, void *ring) 207{ 208 struct tpacket2_hdr *header = ring; 209 int count = 0; 210 211 while (count < RING_NUM_FRAMES && header->tp_status & TP_STATUS_USER) { 212 count++; 213 header = ring + (count * getpagesize()); 214 } 215 216 return count; 217} 218 219static int sock_fanout_read(int fds[], char *rings[], const int expect[]) 220{ 221 int ret[2]; 222 223 ret[0] = sock_fanout_read_ring(fds[0], rings[0]); 224 ret[1] = sock_fanout_read_ring(fds[1], rings[1]); 225 226 fprintf(stderr, "info: count=%d,%d, expect=%d,%d\n", 227 ret[0], ret[1], expect[0], expect[1]); 228 229 if ((!(ret[0] == expect[0] && ret[1] == expect[1])) && 230 (!(ret[0] == expect[1] && ret[1] == expect[0]))) { 231 fprintf(stderr, "ERROR: incorrect queue lengths\n"); 232 return 1; 233 } 234 235 return 0; 236} 237 238/* Test illegal mode + flag combination */ 239static void test_control_single(void) 240{ 241 fprintf(stderr, "test: control single socket\n"); 242 243 if (sock_fanout_open(PACKET_FANOUT_ROLLOVER | 244 PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) { 245 fprintf(stderr, "ERROR: opened socket with dual rollover\n"); 246 exit(1); 247 } 248} 249 250/* Test illegal group with different modes or flags */ 251static void test_control_group(void) 252{ 253 int fds[2]; 254 255 fprintf(stderr, "test: control multiple sockets\n"); 256 257 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH, 0); 258 if (fds[0] == -1) { 259 fprintf(stderr, "ERROR: failed to open HASH socket\n"); 260 exit(1); 261 } 262 if (sock_fanout_open(PACKET_FANOUT_HASH | 263 PACKET_FANOUT_FLAG_DEFRAG, 0) != -1) { 264 fprintf(stderr, "ERROR: joined group with wrong flag defrag\n"); 265 exit(1); 266 } 267 if (sock_fanout_open(PACKET_FANOUT_HASH | 268 PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) { 269 fprintf(stderr, "ERROR: joined group with wrong flag ro\n"); 270 exit(1); 271 } 272 if (sock_fanout_open(PACKET_FANOUT_CPU, 0) != -1) { 273 fprintf(stderr, "ERROR: joined group with wrong mode\n"); 274 exit(1); 275 } 276 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, 0); 277 if (fds[1] == -1) { 278 fprintf(stderr, "ERROR: failed to join group\n"); 279 exit(1); 280 } 281 if (close(fds[1]) || close(fds[0])) { 282 fprintf(stderr, "ERROR: closing sockets\n"); 283 exit(1); 284 } 285} 286 287/* Test creating a unique fanout group ids */ 288static void test_unique_fanout_group_ids(void) 289{ 290 int fds[3]; 291 uint16_t typeflags, first_group_id, second_group_id; 292 293 fprintf(stderr, "test: unique ids\n"); 294 295 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH | 296 PACKET_FANOUT_FLAG_UNIQUEID, 0); 297 if (fds[0] == -1) { 298 fprintf(stderr, "ERROR: failed to create a unique id group.\n"); 299 exit(1); 300 } 301 302 sock_fanout_getopts(fds[0], &typeflags, &first_group_id); 303 if (typeflags != PACKET_FANOUT_HASH) { 304 fprintf(stderr, "ERROR: unexpected typeflags %x\n", typeflags); 305 exit(1); 306 } 307 308 if (sock_fanout_open(PACKET_FANOUT_CPU, first_group_id) != -1) { 309 fprintf(stderr, "ERROR: joined group with wrong type.\n"); 310 exit(1); 311 } 312 313 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, first_group_id); 314 if (fds[1] == -1) { 315 fprintf(stderr, 316 "ERROR: failed to join previously created group.\n"); 317 exit(1); 318 } 319 320 fds[2] = sock_fanout_open(PACKET_FANOUT_HASH | 321 PACKET_FANOUT_FLAG_UNIQUEID, 0); 322 if (fds[2] == -1) { 323 fprintf(stderr, 324 "ERROR: failed to create a second unique id group.\n"); 325 exit(1); 326 } 327 328 sock_fanout_getopts(fds[2], &typeflags, &second_group_id); 329 if (sock_fanout_open(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_UNIQUEID, 330 second_group_id) != -1) { 331 fprintf(stderr, 332 "ERROR: specified a group id when requesting unique id\n"); 333 exit(1); 334 } 335 336 if (close(fds[0]) || close(fds[1]) || close(fds[2])) { 337 fprintf(stderr, "ERROR: closing sockets\n"); 338 exit(1); 339 } 340} 341 342static int test_datapath(uint16_t typeflags, int port_off, 343 const int expect1[], const int expect2[]) 344{ 345 const int expect0[] = { 0, 0 }; 346 char *rings[2]; 347 uint8_t type = typeflags & 0xFF; 348 int fds[2], fds_udp[2][2], ret; 349 350 fprintf(stderr, "test: datapath 0x%hx\n", typeflags); 351 352 fds[0] = sock_fanout_open(typeflags, 0); 353 fds[1] = sock_fanout_open(typeflags, 0); 354 if (fds[0] == -1 || fds[1] == -1) { 355 fprintf(stderr, "ERROR: failed open\n"); 356 exit(1); 357 } 358 if (type == PACKET_FANOUT_CBPF) 359 sock_fanout_set_cbpf(fds[0]); 360 else if (type == PACKET_FANOUT_EBPF) 361 sock_fanout_set_ebpf(fds[0]); 362 363 rings[0] = sock_fanout_open_ring(fds[0]); 364 rings[1] = sock_fanout_open_ring(fds[1]); 365 pair_udp_open(fds_udp[0], PORT_BASE); 366 pair_udp_open(fds_udp[1], PORT_BASE + port_off); 367 sock_fanout_read(fds, rings, expect0); 368 369 /* Send data, but not enough to overflow a queue */ 370 pair_udp_send(fds_udp[0], 15); 371 pair_udp_send_char(fds_udp[1], 5, DATA_CHAR_1); 372 ret = sock_fanout_read(fds, rings, expect1); 373 374 /* Send more data, overflow the queue */ 375 pair_udp_send_char(fds_udp[0], 15, DATA_CHAR_1); 376 /* TODO: ensure consistent order between expect1 and expect2 */ 377 ret |= sock_fanout_read(fds, rings, expect2); 378 379 if (munmap(rings[1], RING_NUM_FRAMES * getpagesize()) || 380 munmap(rings[0], RING_NUM_FRAMES * getpagesize())) { 381 fprintf(stderr, "close rings\n"); 382 exit(1); 383 } 384 if (close(fds_udp[1][1]) || close(fds_udp[1][0]) || 385 close(fds_udp[0][1]) || close(fds_udp[0][0]) || 386 close(fds[1]) || close(fds[0])) { 387 fprintf(stderr, "close datapath\n"); 388 exit(1); 389 } 390 391 return ret; 392} 393 394static int set_cpuaffinity(int cpuid) 395{ 396 cpu_set_t mask; 397 398 CPU_ZERO(&mask); 399 CPU_SET(cpuid, &mask); 400 if (sched_setaffinity(0, sizeof(mask), &mask)) { 401 if (errno != EINVAL) { 402 fprintf(stderr, "setaffinity %d\n", cpuid); 403 exit(1); 404 } 405 return 1; 406 } 407 408 return 0; 409} 410 411int main(int argc, char **argv) 412{ 413 const int expect_hash[2][2] = { { 15, 5 }, { 20, 5 } }; 414 const int expect_hash_rb[2][2] = { { 15, 5 }, { 20, 15 } }; 415 const int expect_lb[2][2] = { { 10, 10 }, { 18, 17 } }; 416 const int expect_rb[2][2] = { { 15, 5 }, { 20, 15 } }; 417 const int expect_cpu0[2][2] = { { 20, 0 }, { 20, 0 } }; 418 const int expect_cpu1[2][2] = { { 0, 20 }, { 0, 20 } }; 419 const int expect_bpf[2][2] = { { 15, 5 }, { 15, 20 } }; 420 const int expect_uniqueid[2][2] = { { 20, 20}, { 20, 20 } }; 421 int port_off = 2, tries = 5, ret; 422 423 test_control_single(); 424 test_control_group(); 425 test_unique_fanout_group_ids(); 426 427 /* find a set of ports that do not collide onto the same socket */ 428 ret = test_datapath(PACKET_FANOUT_HASH, port_off, 429 expect_hash[0], expect_hash[1]); 430 while (ret && tries--) { 431 fprintf(stderr, "info: trying alternate ports (%d)\n", tries); 432 ret = test_datapath(PACKET_FANOUT_HASH, ++port_off, 433 expect_hash[0], expect_hash[1]); 434 } 435 436 ret |= test_datapath(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_ROLLOVER, 437 port_off, expect_hash_rb[0], expect_hash_rb[1]); 438 ret |= test_datapath(PACKET_FANOUT_LB, 439 port_off, expect_lb[0], expect_lb[1]); 440 ret |= test_datapath(PACKET_FANOUT_ROLLOVER, 441 port_off, expect_rb[0], expect_rb[1]); 442 443 ret |= test_datapath(PACKET_FANOUT_CBPF, 444 port_off, expect_bpf[0], expect_bpf[1]); 445 ret |= test_datapath(PACKET_FANOUT_EBPF, 446 port_off, expect_bpf[0], expect_bpf[1]); 447 448 set_cpuaffinity(0); 449 ret |= test_datapath(PACKET_FANOUT_CPU, port_off, 450 expect_cpu0[0], expect_cpu0[1]); 451 if (!set_cpuaffinity(1)) 452 /* TODO: test that choice alternates with previous */ 453 ret |= test_datapath(PACKET_FANOUT_CPU, port_off, 454 expect_cpu1[0], expect_cpu1[1]); 455 456 ret |= test_datapath(PACKET_FANOUT_FLAG_UNIQUEID, port_off, 457 expect_uniqueid[0], expect_uniqueid[1]); 458 459 if (ret) 460 return 1; 461 462 printf("OK. All tests passed\n"); 463 return 0; 464} 465