2 * Copyright 2013 Google Inc.
3 * Author: Willem de Bruijn (willemb@google.com)
5 * A basic test of packet socket fanout behavior.
8 * - create fanout fails as expected with illegal flag combinations
9 * - join fanout fails as expected with diverging types or flags
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.
16 * The test currently runs for
17 * - PACKET_FANOUT_HASH
18 * - PACKET_FANOUT_HASH with PACKET_FANOUT_FLAG_ROLLOVER
21 * - PACKET_FANOUT_ROLLOVER
22 * - PACKET_FANOUT_CBPF
23 * - PACKET_FANOUT_EBPF
26 * - functionality: PACKET_FANOUT_FLAG_DEFRAG
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.
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
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.
44 #define _GNU_SOURCE /* for sched_setaffinity */
46 #include <arpa/inet.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>
54 #include <net/ethernet.h>
55 #include <netinet/ip.h>
56 #include <netinet/udp.h>
64 #include <sys/socket.h>
66 #include <sys/types.h>
69 #include "psock_lib.h"
71 #define RING_NUM_FRAMES 20
73 /* Open a socket in a given fanout mode.
74 * @return -1 if mode is bad, a valid socket otherwise */
75 static int sock_fanout_open(uint16_t typeflags, uint16_t group_id)
77 struct sockaddr_ll addr = {0};
80 fd = socket(PF_PACKET, SOCK_RAW, 0);
82 perror("socket packet");
86 pair_udp_setfilter(fd);
88 addr.sll_family = AF_PACKET;
89 addr.sll_protocol = htons(ETH_P_IP);
90 addr.sll_ifindex = if_nametoindex("lo");
91 if (addr.sll_ifindex == 0) {
92 perror("if_nametoindex");
95 if (bind(fd, (void *) &addr, sizeof(addr))) {
96 perror("bind packet");
100 val = (((int) typeflags) << 16) | group_id;
101 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT, &val, sizeof(val))) {
103 perror("close packet");
112 static void sock_fanout_set_cbpf(int fd)
114 struct sock_filter bpf_filter[] = {
115 BPF_STMT(BPF_LD+BPF_B+BPF_ABS, 80), /* ldb [80] */
116 BPF_STMT(BPF_RET+BPF_A, 0), /* ret A */
118 struct sock_fprog bpf_prog;
120 bpf_prog.filter = bpf_filter;
121 bpf_prog.len = sizeof(bpf_filter) / sizeof(struct sock_filter);
123 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &bpf_prog,
125 perror("fanout data cbpf");
130 static void sock_fanout_getopts(int fd, uint16_t *typeflags, uint16_t *group_id)
133 socklen_t sockopt_len = sizeof(sockopt);
135 if (getsockopt(fd, SOL_PACKET, PACKET_FANOUT,
136 &sockopt, &sockopt_len)) {
137 perror("failed to getsockopt");
140 *typeflags = sockopt >> 16;
141 *group_id = sockopt & 0xfffff;
144 static void sock_fanout_set_ebpf(int fd)
146 static char log_buf[65536];
148 const int len_off = __builtin_offsetof(struct __sk_buff, len);
149 struct bpf_insn prog[] = {
150 { BPF_ALU64 | BPF_MOV | BPF_X, 6, 1, 0, 0 },
151 { BPF_LDX | BPF_W | BPF_MEM, 0, 6, len_off, 0 },
152 { BPF_JMP | BPF_JGE | BPF_K, 0, 0, 1, DATA_LEN },
153 { BPF_JMP | BPF_JA | BPF_K, 0, 0, 4, 0 },
154 { BPF_LD | BPF_B | BPF_ABS, 0, 0, 0, 0x50 },
155 { BPF_JMP | BPF_JEQ | BPF_K, 0, 0, 2, DATA_CHAR },
156 { BPF_JMP | BPF_JEQ | BPF_K, 0, 0, 1, DATA_CHAR_1 },
157 { BPF_ALU | BPF_MOV | BPF_K, 0, 0, 0, 0 },
158 { BPF_JMP | BPF_EXIT, 0, 0, 0, 0 }
163 memset(&attr, 0, sizeof(attr));
164 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
165 attr.insns = (unsigned long) prog;
166 attr.insn_cnt = sizeof(prog) / sizeof(prog[0]);
167 attr.license = (unsigned long) "GPL";
168 attr.log_buf = (unsigned long) log_buf,
169 attr.log_size = sizeof(log_buf),
172 pfd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr));
175 fprintf(stderr, "bpf verifier:\n%s\n", log_buf);
179 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &pfd, sizeof(pfd))) {
180 perror("fanout data ebpf");
185 perror("close ebpf");
190 static char *sock_fanout_open_ring(int fd)
192 struct tpacket_req req = {
193 .tp_block_size = getpagesize(),
194 .tp_frame_size = getpagesize(),
195 .tp_block_nr = RING_NUM_FRAMES,
196 .tp_frame_nr = RING_NUM_FRAMES,
199 int val = TPACKET_V2;
201 if (setsockopt(fd, SOL_PACKET, PACKET_VERSION, (void *) &val,
203 perror("packetsock ring setsockopt version");
206 if (setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *) &req,
208 perror("packetsock ring setsockopt");
212 ring = mmap(0, req.tp_block_size * req.tp_block_nr,
213 PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
214 if (ring == MAP_FAILED) {
215 perror("packetsock ring mmap");
222 static int sock_fanout_read_ring(int fd, void *ring)
224 struct tpacket2_hdr *header = ring;
227 while (count < RING_NUM_FRAMES && header->tp_status & TP_STATUS_USER) {
229 header = ring + (count * getpagesize());
235 static int sock_fanout_read(int fds[], char *rings[], const int expect[])
239 ret[0] = sock_fanout_read_ring(fds[0], rings[0]);
240 ret[1] = sock_fanout_read_ring(fds[1], rings[1]);
242 fprintf(stderr, "info: count=%d,%d, expect=%d,%d\n",
243 ret[0], ret[1], expect[0], expect[1]);
245 if ((!(ret[0] == expect[0] && ret[1] == expect[1])) &&
246 (!(ret[0] == expect[1] && ret[1] == expect[0]))) {
247 fprintf(stderr, "warning: incorrect queue lengths\n");
254 /* Test illegal mode + flag combination */
255 static void test_control_single(void)
257 fprintf(stderr, "test: control single socket\n");
259 if (sock_fanout_open(PACKET_FANOUT_ROLLOVER |
260 PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) {
261 fprintf(stderr, "ERROR: opened socket with dual rollover\n");
266 /* Test illegal group with different modes or flags */
267 static void test_control_group(void)
271 fprintf(stderr, "test: control multiple sockets\n");
273 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH, 0);
275 fprintf(stderr, "ERROR: failed to open HASH socket\n");
278 if (sock_fanout_open(PACKET_FANOUT_HASH |
279 PACKET_FANOUT_FLAG_DEFRAG, 0) != -1) {
280 fprintf(stderr, "ERROR: joined group with wrong flag defrag\n");
283 if (sock_fanout_open(PACKET_FANOUT_HASH |
284 PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) {
285 fprintf(stderr, "ERROR: joined group with wrong flag ro\n");
288 if (sock_fanout_open(PACKET_FANOUT_CPU, 0) != -1) {
289 fprintf(stderr, "ERROR: joined group with wrong mode\n");
292 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, 0);
294 fprintf(stderr, "ERROR: failed to join group\n");
297 if (close(fds[1]) || close(fds[0])) {
298 fprintf(stderr, "ERROR: closing sockets\n");
303 /* Test creating a unique fanout group ids */
304 static void test_unique_fanout_group_ids(void)
307 uint16_t typeflags, first_group_id, second_group_id;
309 fprintf(stderr, "test: unique ids\n");
311 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH |
312 PACKET_FANOUT_FLAG_UNIQUEID, 0);
314 fprintf(stderr, "ERROR: failed to create a unique id group.\n");
318 sock_fanout_getopts(fds[0], &typeflags, &first_group_id);
319 if (typeflags != PACKET_FANOUT_HASH) {
320 fprintf(stderr, "ERROR: unexpected typeflags %x\n", typeflags);
324 if (sock_fanout_open(PACKET_FANOUT_CPU, first_group_id) != -1) {
325 fprintf(stderr, "ERROR: joined group with wrong type.\n");
329 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, first_group_id);
332 "ERROR: failed to join previously created group.\n");
336 fds[2] = sock_fanout_open(PACKET_FANOUT_HASH |
337 PACKET_FANOUT_FLAG_UNIQUEID, 0);
340 "ERROR: failed to create a second unique id group.\n");
344 sock_fanout_getopts(fds[2], &typeflags, &second_group_id);
345 if (sock_fanout_open(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_UNIQUEID,
346 second_group_id) != -1) {
348 "ERROR: specified a group id when requesting unique id\n");
352 if (close(fds[0]) || close(fds[1]) || close(fds[2])) {
353 fprintf(stderr, "ERROR: closing sockets\n");
358 static int test_datapath(uint16_t typeflags, int port_off,
359 const int expect1[], const int expect2[])
361 const int expect0[] = { 0, 0 };
363 uint8_t type = typeflags & 0xFF;
364 int fds[2], fds_udp[2][2], ret;
366 fprintf(stderr, "\ntest: datapath 0x%hx ports %hu,%hu\n",
367 typeflags, (uint16_t)PORT_BASE,
368 (uint16_t)(PORT_BASE + port_off));
370 fds[0] = sock_fanout_open(typeflags, 0);
371 fds[1] = sock_fanout_open(typeflags, 0);
372 if (fds[0] == -1 || fds[1] == -1) {
373 fprintf(stderr, "ERROR: failed open\n");
376 if (type == PACKET_FANOUT_CBPF)
377 sock_fanout_set_cbpf(fds[0]);
378 else if (type == PACKET_FANOUT_EBPF)
379 sock_fanout_set_ebpf(fds[0]);
381 rings[0] = sock_fanout_open_ring(fds[0]);
382 rings[1] = sock_fanout_open_ring(fds[1]);
383 pair_udp_open(fds_udp[0], PORT_BASE);
384 pair_udp_open(fds_udp[1], PORT_BASE + port_off);
385 sock_fanout_read(fds, rings, expect0);
387 /* Send data, but not enough to overflow a queue */
388 pair_udp_send(fds_udp[0], 15);
389 pair_udp_send_char(fds_udp[1], 5, DATA_CHAR_1);
390 ret = sock_fanout_read(fds, rings, expect1);
392 /* Send more data, overflow the queue */
393 pair_udp_send_char(fds_udp[0], 15, DATA_CHAR_1);
394 /* TODO: ensure consistent order between expect1 and expect2 */
395 ret |= sock_fanout_read(fds, rings, expect2);
397 if (munmap(rings[1], RING_NUM_FRAMES * getpagesize()) ||
398 munmap(rings[0], RING_NUM_FRAMES * getpagesize())) {
399 fprintf(stderr, "close rings\n");
402 if (close(fds_udp[1][1]) || close(fds_udp[1][0]) ||
403 close(fds_udp[0][1]) || close(fds_udp[0][0]) ||
404 close(fds[1]) || close(fds[0])) {
405 fprintf(stderr, "close datapath\n");
412 static int set_cpuaffinity(int cpuid)
417 CPU_SET(cpuid, &mask);
418 if (sched_setaffinity(0, sizeof(mask), &mask)) {
419 if (errno != EINVAL) {
420 fprintf(stderr, "setaffinity %d\n", cpuid);
429 int main(int argc, char **argv)
431 const int expect_hash[2][2] = { { 15, 5 }, { 20, 5 } };
432 const int expect_hash_rb[2][2] = { { 15, 5 }, { 20, 15 } };
433 const int expect_lb[2][2] = { { 10, 10 }, { 18, 17 } };
434 const int expect_rb[2][2] = { { 15, 5 }, { 20, 15 } };
435 const int expect_cpu0[2][2] = { { 20, 0 }, { 20, 0 } };
436 const int expect_cpu1[2][2] = { { 0, 20 }, { 0, 20 } };
437 const int expect_bpf[2][2] = { { 15, 5 }, { 15, 20 } };
438 const int expect_uniqueid[2][2] = { { 20, 20}, { 20, 20 } };
439 int port_off = 2, tries = 20, ret;
441 test_control_single();
442 test_control_group();
443 test_unique_fanout_group_ids();
445 /* find a set of ports that do not collide onto the same socket */
446 ret = test_datapath(PACKET_FANOUT_HASH, port_off,
447 expect_hash[0], expect_hash[1]);
449 fprintf(stderr, "info: trying alternate ports (%d)\n", tries);
450 ret = test_datapath(PACKET_FANOUT_HASH, ++port_off,
451 expect_hash[0], expect_hash[1]);
453 fprintf(stderr, "too many collisions\n");
458 ret |= test_datapath(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_ROLLOVER,
459 port_off, expect_hash_rb[0], expect_hash_rb[1]);
460 ret |= test_datapath(PACKET_FANOUT_LB,
461 port_off, expect_lb[0], expect_lb[1]);
462 ret |= test_datapath(PACKET_FANOUT_ROLLOVER,
463 port_off, expect_rb[0], expect_rb[1]);
465 ret |= test_datapath(PACKET_FANOUT_CBPF,
466 port_off, expect_bpf[0], expect_bpf[1]);
467 ret |= test_datapath(PACKET_FANOUT_EBPF,
468 port_off, expect_bpf[0], expect_bpf[1]);
471 ret |= test_datapath(PACKET_FANOUT_CPU, port_off,
472 expect_cpu0[0], expect_cpu0[1]);
473 if (!set_cpuaffinity(1))
474 /* TODO: test that choice alternates with previous */
475 ret |= test_datapath(PACKET_FANOUT_CPU, port_off,
476 expect_cpu1[0], expect_cpu1[1]);
478 ret |= test_datapath(PACKET_FANOUT_FLAG_UNIQUEID, port_off,
479 expect_uniqueid[0], expect_uniqueid[1]);
484 printf("OK. All tests passed\n");