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>
53 #include <net/ethernet.h>
54 #include <netinet/ip.h>
55 #include <netinet/udp.h>
63 #include <sys/socket.h>
65 #include <sys/types.h>
68 #include "psock_lib.h"
70 #define RING_NUM_FRAMES 20
72 /* Open a socket in a given fanout mode.
73 * @return -1 if mode is bad, a valid socket otherwise */
74 static int sock_fanout_open(uint16_t typeflags, uint16_t group_id)
78 fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_IP));
80 perror("socket packet");
84 val = (((int) typeflags) << 16) | group_id;
85 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT, &val, sizeof(val))) {
87 perror("close packet");
93 pair_udp_setfilter(fd);
97 static void sock_fanout_set_cbpf(int fd)
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 */
103 struct sock_fprog bpf_prog;
105 bpf_prog.filter = bpf_filter;
106 bpf_prog.len = sizeof(bpf_filter) / sizeof(struct sock_filter);
108 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &bpf_prog,
110 perror("fanout data cbpf");
115 static void sock_fanout_getopts(int fd, uint16_t *typeflags, uint16_t *group_id)
118 socklen_t sockopt_len = sizeof(sockopt);
120 if (getsockopt(fd, SOL_PACKET, PACKET_FANOUT,
121 &sockopt, &sockopt_len)) {
122 perror("failed to getsockopt");
125 *typeflags = sockopt >> 16;
126 *group_id = sockopt & 0xfffff;
129 static void sock_fanout_set_ebpf(int fd)
131 static char log_buf[65536];
133 const int len_off = __builtin_offsetof(struct __sk_buff, len);
134 struct bpf_insn prog[] = {
135 { BPF_ALU64 | BPF_MOV | BPF_X, 6, 1, 0, 0 },
136 { BPF_LDX | BPF_W | BPF_MEM, 0, 6, len_off, 0 },
137 { BPF_JMP | BPF_JGE | BPF_K, 0, 0, 1, DATA_LEN },
138 { BPF_JMP | BPF_JA | BPF_K, 0, 0, 4, 0 },
139 { BPF_LD | BPF_B | BPF_ABS, 0, 0, 0, 0x50 },
140 { BPF_JMP | BPF_JEQ | BPF_K, 0, 0, 2, DATA_CHAR },
141 { BPF_JMP | BPF_JEQ | BPF_K, 0, 0, 1, DATA_CHAR_1 },
142 { BPF_ALU | BPF_MOV | BPF_K, 0, 0, 0, 0 },
143 { BPF_JMP | BPF_EXIT, 0, 0, 0, 0 }
148 memset(&attr, 0, sizeof(attr));
149 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
150 attr.insns = (unsigned long) prog;
151 attr.insn_cnt = sizeof(prog) / sizeof(prog[0]);
152 attr.license = (unsigned long) "GPL";
153 attr.log_buf = (unsigned long) log_buf,
154 attr.log_size = sizeof(log_buf),
157 pfd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr));
160 fprintf(stderr, "bpf verifier:\n%s\n", log_buf);
164 if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &pfd, sizeof(pfd))) {
165 perror("fanout data ebpf");
170 perror("close ebpf");
175 static char *sock_fanout_open_ring(int fd)
177 struct tpacket_req req = {
178 .tp_block_size = getpagesize(),
179 .tp_frame_size = getpagesize(),
180 .tp_block_nr = RING_NUM_FRAMES,
181 .tp_frame_nr = RING_NUM_FRAMES,
184 int val = TPACKET_V2;
186 if (setsockopt(fd, SOL_PACKET, PACKET_VERSION, (void *) &val,
188 perror("packetsock ring setsockopt version");
191 if (setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *) &req,
193 perror("packetsock ring setsockopt");
197 ring = mmap(0, req.tp_block_size * req.tp_block_nr,
198 PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
199 if (ring == MAP_FAILED) {
200 perror("packetsock ring mmap");
207 static int sock_fanout_read_ring(int fd, void *ring)
209 struct tpacket2_hdr *header = ring;
212 while (count < RING_NUM_FRAMES && header->tp_status & TP_STATUS_USER) {
214 header = ring + (count * getpagesize());
220 static int sock_fanout_read(int fds[], char *rings[], const int expect[])
224 ret[0] = sock_fanout_read_ring(fds[0], rings[0]);
225 ret[1] = sock_fanout_read_ring(fds[1], rings[1]);
227 fprintf(stderr, "info: count=%d,%d, expect=%d,%d\n",
228 ret[0], ret[1], expect[0], expect[1]);
230 if ((!(ret[0] == expect[0] && ret[1] == expect[1])) &&
231 (!(ret[0] == expect[1] && ret[1] == expect[0]))) {
232 fprintf(stderr, "ERROR: incorrect queue lengths\n");
239 /* Test illegal mode + flag combination */
240 static void test_control_single(void)
242 fprintf(stderr, "test: control single socket\n");
244 if (sock_fanout_open(PACKET_FANOUT_ROLLOVER |
245 PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) {
246 fprintf(stderr, "ERROR: opened socket with dual rollover\n");
251 /* Test illegal group with different modes or flags */
252 static void test_control_group(void)
256 fprintf(stderr, "test: control multiple sockets\n");
258 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH, 0);
260 fprintf(stderr, "ERROR: failed to open HASH socket\n");
263 if (sock_fanout_open(PACKET_FANOUT_HASH |
264 PACKET_FANOUT_FLAG_DEFRAG, 0) != -1) {
265 fprintf(stderr, "ERROR: joined group with wrong flag defrag\n");
268 if (sock_fanout_open(PACKET_FANOUT_HASH |
269 PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) {
270 fprintf(stderr, "ERROR: joined group with wrong flag ro\n");
273 if (sock_fanout_open(PACKET_FANOUT_CPU, 0) != -1) {
274 fprintf(stderr, "ERROR: joined group with wrong mode\n");
277 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, 0);
279 fprintf(stderr, "ERROR: failed to join group\n");
282 if (close(fds[1]) || close(fds[0])) {
283 fprintf(stderr, "ERROR: closing sockets\n");
288 /* Test creating a unique fanout group ids */
289 static void test_unique_fanout_group_ids(void)
292 uint16_t typeflags, first_group_id, second_group_id;
294 fprintf(stderr, "test: unique ids\n");
296 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH |
297 PACKET_FANOUT_FLAG_UNIQUEID, 0);
299 fprintf(stderr, "ERROR: failed to create a unique id group.\n");
303 sock_fanout_getopts(fds[0], &typeflags, &first_group_id);
304 if (typeflags != PACKET_FANOUT_HASH) {
305 fprintf(stderr, "ERROR: unexpected typeflags %x\n", typeflags);
309 if (sock_fanout_open(PACKET_FANOUT_CPU, first_group_id) != -1) {
310 fprintf(stderr, "ERROR: joined group with wrong type.\n");
314 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, first_group_id);
317 "ERROR: failed to join previously created group.\n");
321 fds[2] = sock_fanout_open(PACKET_FANOUT_HASH |
322 PACKET_FANOUT_FLAG_UNIQUEID, 0);
325 "ERROR: failed to create a second unique id group.\n");
329 sock_fanout_getopts(fds[2], &typeflags, &second_group_id);
330 if (sock_fanout_open(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_UNIQUEID,
331 second_group_id) != -1) {
333 "ERROR: specified a group id when requesting unique id\n");
337 if (close(fds[0]) || close(fds[1]) || close(fds[2])) {
338 fprintf(stderr, "ERROR: closing sockets\n");
343 static int test_datapath(uint16_t typeflags, int port_off,
344 const int expect1[], const int expect2[])
346 const int expect0[] = { 0, 0 };
348 uint8_t type = typeflags & 0xFF;
349 int fds[2], fds_udp[2][2], ret;
351 fprintf(stderr, "test: datapath 0x%hx\n", typeflags);
353 fds[0] = sock_fanout_open(typeflags, 0);
354 fds[1] = sock_fanout_open(typeflags, 0);
355 if (fds[0] == -1 || fds[1] == -1) {
356 fprintf(stderr, "ERROR: failed open\n");
359 if (type == PACKET_FANOUT_CBPF)
360 sock_fanout_set_cbpf(fds[0]);
361 else if (type == PACKET_FANOUT_EBPF)
362 sock_fanout_set_ebpf(fds[0]);
364 rings[0] = sock_fanout_open_ring(fds[0]);
365 rings[1] = sock_fanout_open_ring(fds[1]);
366 pair_udp_open(fds_udp[0], PORT_BASE);
367 pair_udp_open(fds_udp[1], PORT_BASE + port_off);
368 sock_fanout_read(fds, rings, expect0);
370 /* Send data, but not enough to overflow a queue */
371 pair_udp_send(fds_udp[0], 15);
372 pair_udp_send_char(fds_udp[1], 5, DATA_CHAR_1);
373 ret = sock_fanout_read(fds, rings, expect1);
375 /* Send more data, overflow the queue */
376 pair_udp_send_char(fds_udp[0], 15, DATA_CHAR_1);
377 /* TODO: ensure consistent order between expect1 and expect2 */
378 ret |= sock_fanout_read(fds, rings, expect2);
380 if (munmap(rings[1], RING_NUM_FRAMES * getpagesize()) ||
381 munmap(rings[0], RING_NUM_FRAMES * getpagesize())) {
382 fprintf(stderr, "close rings\n");
385 if (close(fds_udp[1][1]) || close(fds_udp[1][0]) ||
386 close(fds_udp[0][1]) || close(fds_udp[0][0]) ||
387 close(fds[1]) || close(fds[0])) {
388 fprintf(stderr, "close datapath\n");
395 static int set_cpuaffinity(int cpuid)
400 CPU_SET(cpuid, &mask);
401 if (sched_setaffinity(0, sizeof(mask), &mask)) {
402 if (errno != EINVAL) {
403 fprintf(stderr, "setaffinity %d\n", cpuid);
412 int main(int argc, char **argv)
414 const int expect_hash[2][2] = { { 15, 5 }, { 20, 5 } };
415 const int expect_hash_rb[2][2] = { { 15, 5 }, { 20, 15 } };
416 const int expect_lb[2][2] = { { 10, 10 }, { 18, 17 } };
417 const int expect_rb[2][2] = { { 15, 5 }, { 20, 15 } };
418 const int expect_cpu0[2][2] = { { 20, 0 }, { 20, 0 } };
419 const int expect_cpu1[2][2] = { { 0, 20 }, { 0, 20 } };
420 const int expect_bpf[2][2] = { { 15, 5 }, { 15, 20 } };
421 const int expect_uniqueid[2][2] = { { 20, 20}, { 20, 20 } };
422 int port_off = 2, tries = 5, ret;
424 test_control_single();
425 test_control_group();
426 test_unique_fanout_group_ids();
428 /* find a set of ports that do not collide onto the same socket */
429 ret = test_datapath(PACKET_FANOUT_HASH, port_off,
430 expect_hash[0], expect_hash[1]);
431 while (ret && tries--) {
432 fprintf(stderr, "info: trying alternate ports (%d)\n", tries);
433 ret = test_datapath(PACKET_FANOUT_HASH, ++port_off,
434 expect_hash[0], expect_hash[1]);
437 ret |= test_datapath(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_ROLLOVER,
438 port_off, expect_hash_rb[0], expect_hash_rb[1]);
439 ret |= test_datapath(PACKET_FANOUT_LB,
440 port_off, expect_lb[0], expect_lb[1]);
441 ret |= test_datapath(PACKET_FANOUT_ROLLOVER,
442 port_off, expect_rb[0], expect_rb[1]);
444 ret |= test_datapath(PACKET_FANOUT_CBPF,
445 port_off, expect_bpf[0], expect_bpf[1]);
446 ret |= test_datapath(PACKET_FANOUT_EBPF,
447 port_off, expect_bpf[0], expect_bpf[1]);
450 ret |= test_datapath(PACKET_FANOUT_CPU, port_off,
451 expect_cpu0[0], expect_cpu0[1]);
452 if (!set_cpuaffinity(1))
453 /* TODO: test that choice alternates with previous */
454 ret |= test_datapath(PACKET_FANOUT_CPU, port_off,
455 expect_cpu1[0], expect_cpu1[1]);
457 ret |= test_datapath(PACKET_FANOUT_FLAG_UNIQUEID, port_off,
458 expect_uniqueid[0], expect_uniqueid[1]);
463 printf("OK. All tests passed\n");