]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - tools/testing/selftests/net/psock_fanout.c
226e5e33105aaa68d788cc37c53435807822a1a0
[mirror_ubuntu-zesty-kernel.git] / tools / testing / selftests / net / psock_fanout.c
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
186 if (setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *) &req,
187 sizeof(req))) {
188 perror("packetsock ring setsockopt");
189 exit(1);
190 }
191
192 ring = mmap(0, req.tp_block_size * req.tp_block_nr,
193 PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
194 if (!ring) {
195 fprintf(stderr, "packetsock ring mmap\n");
196 exit(1);
197 }
198
199 return ring;
200 }
201
202 static int sock_fanout_read_ring(int fd, void *ring)
203 {
204 struct tpacket_hdr *header = ring;
205 int count = 0;
206
207 while (header->tp_status & TP_STATUS_USER && count < RING_NUM_FRAMES) {
208 count++;
209 header = ring + (count * getpagesize());
210 }
211
212 return count;
213 }
214
215 static int sock_fanout_read(int fds[], char *rings[], const int expect[])
216 {
217 int ret[2];
218
219 ret[0] = sock_fanout_read_ring(fds[0], rings[0]);
220 ret[1] = sock_fanout_read_ring(fds[1], rings[1]);
221
222 fprintf(stderr, "info: count=%d,%d, expect=%d,%d\n",
223 ret[0], ret[1], expect[0], expect[1]);
224
225 if ((!(ret[0] == expect[0] && ret[1] == expect[1])) &&
226 (!(ret[0] == expect[1] && ret[1] == expect[0]))) {
227 fprintf(stderr, "ERROR: incorrect queue lengths\n");
228 return 1;
229 }
230
231 return 0;
232 }
233
234 /* Test illegal mode + flag combination */
235 static void test_control_single(void)
236 {
237 fprintf(stderr, "test: control single socket\n");
238
239 if (sock_fanout_open(PACKET_FANOUT_ROLLOVER |
240 PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) {
241 fprintf(stderr, "ERROR: opened socket with dual rollover\n");
242 exit(1);
243 }
244 }
245
246 /* Test illegal group with different modes or flags */
247 static void test_control_group(void)
248 {
249 int fds[2];
250
251 fprintf(stderr, "test: control multiple sockets\n");
252
253 fds[0] = sock_fanout_open(PACKET_FANOUT_HASH, 20);
254 if (fds[0] == -1) {
255 fprintf(stderr, "ERROR: failed to open HASH socket\n");
256 exit(1);
257 }
258 if (sock_fanout_open(PACKET_FANOUT_HASH |
259 PACKET_FANOUT_FLAG_DEFRAG, 10) != -1) {
260 fprintf(stderr, "ERROR: joined group with wrong flag defrag\n");
261 exit(1);
262 }
263 if (sock_fanout_open(PACKET_FANOUT_HASH |
264 PACKET_FANOUT_FLAG_ROLLOVER, 10) != -1) {
265 fprintf(stderr, "ERROR: joined group with wrong flag ro\n");
266 exit(1);
267 }
268 if (sock_fanout_open(PACKET_FANOUT_CPU, 10) != -1) {
269 fprintf(stderr, "ERROR: joined group with wrong mode\n");
270 exit(1);
271 }
272 fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, 20);
273 if (fds[1] == -1) {
274 fprintf(stderr, "ERROR: failed to join group\n");
275 exit(1);
276 }
277 if (close(fds[1]) || close(fds[0])) {
278 fprintf(stderr, "ERROR: closing sockets\n");
279 exit(1);
280 }
281 }
282
283 static int test_datapath(uint16_t typeflags, int port_off,
284 const int expect1[], const int expect2[])
285 {
286 const int expect0[] = { 0, 0 };
287 char *rings[2];
288 int fds[2], fds_udp[2][2], ret;
289
290 fprintf(stderr, "test: datapath 0x%hx\n", typeflags);
291
292 fds[0] = sock_fanout_open(typeflags, 20);
293 fds[1] = sock_fanout_open(typeflags, 20);
294 if (fds[0] == -1 || fds[1] == -1) {
295 fprintf(stderr, "ERROR: failed open\n");
296 exit(1);
297 }
298 rings[0] = sock_fanout_open_ring(fds[0]);
299 rings[1] = sock_fanout_open_ring(fds[1]);
300 pair_udp_open(fds_udp[0], PORT_BASE);
301 pair_udp_open(fds_udp[1], PORT_BASE + port_off);
302 sock_fanout_read(fds, rings, expect0);
303
304 /* Send data, but not enough to overflow a queue */
305 pair_udp_send(fds_udp[0], 15);
306 pair_udp_send(fds_udp[1], 5);
307 ret = sock_fanout_read(fds, rings, expect1);
308
309 /* Send more data, overflow the queue */
310 pair_udp_send(fds_udp[0], 15);
311 /* TODO: ensure consistent order between expect1 and expect2 */
312 ret |= sock_fanout_read(fds, rings, expect2);
313
314 if (munmap(rings[1], RING_NUM_FRAMES * getpagesize()) ||
315 munmap(rings[0], RING_NUM_FRAMES * getpagesize())) {
316 fprintf(stderr, "close rings\n");
317 exit(1);
318 }
319 if (close(fds_udp[1][1]) || close(fds_udp[1][0]) ||
320 close(fds_udp[0][1]) || close(fds_udp[0][0]) ||
321 close(fds[1]) || close(fds[0])) {
322 fprintf(stderr, "close datapath\n");
323 exit(1);
324 }
325
326 return ret;
327 }
328
329 static int set_cpuaffinity(int cpuid)
330 {
331 cpu_set_t mask;
332
333 CPU_ZERO(&mask);
334 CPU_SET(cpuid, &mask);
335 if (sched_setaffinity(0, sizeof(mask), &mask)) {
336 if (errno != EINVAL) {
337 fprintf(stderr, "setaffinity %d\n", cpuid);
338 exit(1);
339 }
340 return 1;
341 }
342
343 return 0;
344 }
345
346 int main(int argc, char **argv)
347 {
348 const int expect_hash[2][2] = { { 15, 5 }, { 20, 5 } };
349 const int expect_hash_rb[2][2] = { { 15, 5 }, { 20, 15 } };
350 const int expect_lb[2][2] = { { 10, 10 }, { 18, 17 } };
351 const int expect_rb[2][2] = { { 20, 0 }, { 20, 15 } };
352 const int expect_cpu0[2][2] = { { 20, 0 }, { 20, 0 } };
353 const int expect_cpu1[2][2] = { { 0, 20 }, { 0, 20 } };
354 int port_off = 2, tries = 5, ret;
355
356 test_control_single();
357 test_control_group();
358
359 /* find a set of ports that do not collide onto the same socket */
360 ret = test_datapath(PACKET_FANOUT_HASH, port_off,
361 expect_hash[0], expect_hash[1]);
362 while (ret && tries--) {
363 fprintf(stderr, "info: trying alternate ports (%d)\n", tries);
364 ret = test_datapath(PACKET_FANOUT_HASH, ++port_off,
365 expect_hash[0], expect_hash[1]);
366 }
367
368 ret |= test_datapath(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_ROLLOVER,
369 port_off, expect_hash_rb[0], expect_hash_rb[1]);
370 ret |= test_datapath(PACKET_FANOUT_LB,
371 port_off, expect_lb[0], expect_lb[1]);
372 ret |= test_datapath(PACKET_FANOUT_ROLLOVER,
373 port_off, expect_rb[0], expect_rb[1]);
374
375 set_cpuaffinity(0);
376 ret |= test_datapath(PACKET_FANOUT_CPU, port_off,
377 expect_cpu0[0], expect_cpu0[1]);
378 if (!set_cpuaffinity(1))
379 /* TODO: test that choice alternates with previous */
380 ret |= test_datapath(PACKET_FANOUT_CPU, port_off,
381 expect_cpu1[0], expect_cpu1[1]);
382
383 if (ret)
384 return 1;
385
386 printf("OK. All tests passed\n");
387 return 0;
388 }