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/if.h>
54 #include <net/ethernet.h>
55 #include <netinet/ip.h>
56 #include <netinet/udp.h>
57 #include <poll.h>
58 #include <sched.h>
59 #include <stdint.h>
60 #include <stdio.h>
61 #include <stdlib.h>
62 #include <string.h>
63 #include <sys/mman.h>
64 #include <sys/socket.h>
65 #include <sys/stat.h>
66 #include <sys/types.h>
67 #include <unistd.h>
68 
69 #include "psock_lib.h"
70 
71 #define RING_NUM_FRAMES			20
72 
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)
76 {
77 	struct sockaddr_ll addr = {0};
78 	int fd, val;
79 
80 	fd = socket(PF_PACKET, SOCK_RAW, 0);
81 	if (fd < 0) {
82 		perror("socket packet");
83 		exit(1);
84 	}
85 
86 	pair_udp_setfilter(fd);
87 
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");
93 		exit(1);
94 	}
95 	if (bind(fd, (void *) &addr, sizeof(addr))) {
96 		perror("bind packet");
97 		exit(1);
98 	}
99 
100 	val = (((int) typeflags) << 16) | group_id;
101 	if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT, &val, sizeof(val))) {
102 		if (close(fd)) {
103 			perror("close packet");
104 			exit(1);
105 		}
106 		return -1;
107 	}
108 
109 	return fd;
110 }
111 
112 static void sock_fanout_set_cbpf(int fd)
113 {
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 */
117 	};
118 	struct sock_fprog bpf_prog;
119 
120 	bpf_prog.filter = bpf_filter;
121 	bpf_prog.len = sizeof(bpf_filter) / sizeof(struct sock_filter);
122 
123 	if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &bpf_prog,
124 		       sizeof(bpf_prog))) {
125 		perror("fanout data cbpf");
126 		exit(1);
127 	}
128 }
129 
130 static void sock_fanout_getopts(int fd, uint16_t *typeflags, uint16_t *group_id)
131 {
132 	int sockopt;
133 	socklen_t sockopt_len = sizeof(sockopt);
134 
135 	if (getsockopt(fd, SOL_PACKET, PACKET_FANOUT,
136 		       &sockopt, &sockopt_len)) {
137 		perror("failed to getsockopt");
138 		exit(1);
139 	}
140 	*typeflags = sockopt >> 16;
141 	*group_id = sockopt & 0xfffff;
142 }
143 
144 static void sock_fanout_set_ebpf(int fd)
145 {
146 	static char log_buf[65536];
147 
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 }
159 	};
160 	union bpf_attr attr;
161 	int pfd;
162 
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),
170 	attr.log_level = 1,
171 
172 	pfd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr));
173 	if (pfd < 0) {
174 		perror("bpf");
175 		fprintf(stderr, "bpf verifier:\n%s\n", log_buf);
176 		exit(1);
177 	}
178 
179 	if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &pfd, sizeof(pfd))) {
180 		perror("fanout data ebpf");
181 		exit(1);
182 	}
183 
184 	if (close(pfd)) {
185 		perror("close ebpf");
186 		exit(1);
187 	}
188 }
189 
190 static char *sock_fanout_open_ring(int fd)
191 {
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,
197 	};
198 	char *ring;
199 	int val = TPACKET_V2;
200 
201 	if (setsockopt(fd, SOL_PACKET, PACKET_VERSION, (void *) &val,
202 		       sizeof(val))) {
203 		perror("packetsock ring setsockopt version");
204 		exit(1);
205 	}
206 	if (setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *) &req,
207 		       sizeof(req))) {
208 		perror("packetsock ring setsockopt");
209 		exit(1);
210 	}
211 
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");
216 		exit(1);
217 	}
218 
219 	return ring;
220 }
221 
222 static int sock_fanout_read_ring(int fd, void *ring)
223 {
224 	struct tpacket2_hdr *header = ring;
225 	int count = 0;
226 
227 	while (count < RING_NUM_FRAMES && header->tp_status & TP_STATUS_USER) {
228 		count++;
229 		header = ring + (count * getpagesize());
230 	}
231 
232 	return count;
233 }
234 
235 static int sock_fanout_read(int fds[], char *rings[], const int expect[])
236 {
237 	int ret[2];
238 
239 	ret[0] = sock_fanout_read_ring(fds[0], rings[0]);
240 	ret[1] = sock_fanout_read_ring(fds[1], rings[1]);
241 
242 	fprintf(stderr, "info: count=%d,%d, expect=%d,%d\n",
243 			ret[0], ret[1], expect[0], expect[1]);
244 
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");
248 		return 1;
249 	}
250 
251 	return 0;
252 }
253 
254 /* Test illegal mode + flag combination */
255 static void test_control_single(void)
256 {
257 	fprintf(stderr, "test: control single socket\n");
258 
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");
262 		exit(1);
263 	}
264 }
265 
266 /* Test illegal group with different modes or flags */
267 static void test_control_group(void)
268 {
269 	int fds[2];
270 
271 	fprintf(stderr, "test: control multiple sockets\n");
272 
273 	fds[0] = sock_fanout_open(PACKET_FANOUT_HASH, 0);
274 	if (fds[0] == -1) {
275 		fprintf(stderr, "ERROR: failed to open HASH socket\n");
276 		exit(1);
277 	}
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");
281 		exit(1);
282 	}
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");
286 		exit(1);
287 	}
288 	if (sock_fanout_open(PACKET_FANOUT_CPU, 0) != -1) {
289 		fprintf(stderr, "ERROR: joined group with wrong mode\n");
290 		exit(1);
291 	}
292 	fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, 0);
293 	if (fds[1] == -1) {
294 		fprintf(stderr, "ERROR: failed to join group\n");
295 		exit(1);
296 	}
297 	if (close(fds[1]) || close(fds[0])) {
298 		fprintf(stderr, "ERROR: closing sockets\n");
299 		exit(1);
300 	}
301 }
302 
303 /* Test creating a unique fanout group ids */
304 static void test_unique_fanout_group_ids(void)
305 {
306 	int fds[3];
307 	uint16_t typeflags, first_group_id, second_group_id;
308 
309 	fprintf(stderr, "test: unique ids\n");
310 
311 	fds[0] = sock_fanout_open(PACKET_FANOUT_HASH |
312 				  PACKET_FANOUT_FLAG_UNIQUEID, 0);
313 	if (fds[0] == -1) {
314 		fprintf(stderr, "ERROR: failed to create a unique id group.\n");
315 		exit(1);
316 	}
317 
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);
321 		exit(1);
322 	}
323 
324 	if (sock_fanout_open(PACKET_FANOUT_CPU, first_group_id) != -1) {
325 		fprintf(stderr, "ERROR: joined group with wrong type.\n");
326 		exit(1);
327 	}
328 
329 	fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, first_group_id);
330 	if (fds[1] == -1) {
331 		fprintf(stderr,
332 			"ERROR: failed to join previously created group.\n");
333 		exit(1);
334 	}
335 
336 	fds[2] = sock_fanout_open(PACKET_FANOUT_HASH |
337 				  PACKET_FANOUT_FLAG_UNIQUEID, 0);
338 	if (fds[2] == -1) {
339 		fprintf(stderr,
340 			"ERROR: failed to create a second unique id group.\n");
341 		exit(1);
342 	}
343 
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) {
347 		fprintf(stderr,
348 			"ERROR: specified a group id when requesting unique id\n");
349 		exit(1);
350 	}
351 
352 	if (close(fds[0]) || close(fds[1]) || close(fds[2])) {
353 		fprintf(stderr, "ERROR: closing sockets\n");
354 		exit(1);
355 	}
356 }
357 
358 static int test_datapath(uint16_t typeflags, int port_off,
359 			 const int expect1[], const int expect2[])
360 {
361 	const int expect0[] = { 0, 0 };
362 	char *rings[2];
363 	uint8_t type = typeflags & 0xFF;
364 	int fds[2], fds_udp[2][2], ret;
365 
366 	fprintf(stderr, "\ntest: datapath 0x%hx ports %hu,%hu\n",
367 		typeflags, PORT_BASE, PORT_BASE + port_off);
368 
369 	fds[0] = sock_fanout_open(typeflags, 0);
370 	fds[1] = sock_fanout_open(typeflags, 0);
371 	if (fds[0] == -1 || fds[1] == -1) {
372 		fprintf(stderr, "ERROR: failed open\n");
373 		exit(1);
374 	}
375 	if (type == PACKET_FANOUT_CBPF)
376 		sock_fanout_set_cbpf(fds[0]);
377 	else if (type == PACKET_FANOUT_EBPF)
378 		sock_fanout_set_ebpf(fds[0]);
379 
380 	rings[0] = sock_fanout_open_ring(fds[0]);
381 	rings[1] = sock_fanout_open_ring(fds[1]);
382 	pair_udp_open(fds_udp[0], PORT_BASE);
383 	pair_udp_open(fds_udp[1], PORT_BASE + port_off);
384 	sock_fanout_read(fds, rings, expect0);
385 
386 	/* Send data, but not enough to overflow a queue */
387 	pair_udp_send(fds_udp[0], 15);
388 	pair_udp_send_char(fds_udp[1], 5, DATA_CHAR_1);
389 	ret = sock_fanout_read(fds, rings, expect1);
390 
391 	/* Send more data, overflow the queue */
392 	pair_udp_send_char(fds_udp[0], 15, DATA_CHAR_1);
393 	/* TODO: ensure consistent order between expect1 and expect2 */
394 	ret |= sock_fanout_read(fds, rings, expect2);
395 
396 	if (munmap(rings[1], RING_NUM_FRAMES * getpagesize()) ||
397 	    munmap(rings[0], RING_NUM_FRAMES * getpagesize())) {
398 		fprintf(stderr, "close rings\n");
399 		exit(1);
400 	}
401 	if (close(fds_udp[1][1]) || close(fds_udp[1][0]) ||
402 	    close(fds_udp[0][1]) || close(fds_udp[0][0]) ||
403 	    close(fds[1]) || close(fds[0])) {
404 		fprintf(stderr, "close datapath\n");
405 		exit(1);
406 	}
407 
408 	return ret;
409 }
410 
411 static int set_cpuaffinity(int cpuid)
412 {
413 	cpu_set_t mask;
414 
415 	CPU_ZERO(&mask);
416 	CPU_SET(cpuid, &mask);
417 	if (sched_setaffinity(0, sizeof(mask), &mask)) {
418 		if (errno != EINVAL) {
419 			fprintf(stderr, "setaffinity %d\n", cpuid);
420 			exit(1);
421 		}
422 		return 1;
423 	}
424 
425 	return 0;
426 }
427 
428 int main(int argc, char **argv)
429 {
430 	const int expect_hash[2][2]	= { { 15, 5 },  { 20, 5 } };
431 	const int expect_hash_rb[2][2]	= { { 15, 5 },  { 20, 15 } };
432 	const int expect_lb[2][2]	= { { 10, 10 }, { 18, 17 } };
433 	const int expect_rb[2][2]	= { { 15, 5 },  { 20, 15 } };
434 	const int expect_cpu0[2][2]	= { { 20, 0 },  { 20, 0 } };
435 	const int expect_cpu1[2][2]	= { { 0, 20 },  { 0, 20 } };
436 	const int expect_bpf[2][2]	= { { 15, 5 },  { 15, 20 } };
437 	const int expect_uniqueid[2][2] = { { 20, 20},  { 20, 20 } };
438 	int port_off = 2, tries = 20, ret;
439 
440 	test_control_single();
441 	test_control_group();
442 	test_unique_fanout_group_ids();
443 
444 	/* find a set of ports that do not collide onto the same socket */
445 	ret = test_datapath(PACKET_FANOUT_HASH, port_off,
446 			    expect_hash[0], expect_hash[1]);
447 	while (ret) {
448 		fprintf(stderr, "info: trying alternate ports (%d)\n", tries);
449 		ret = test_datapath(PACKET_FANOUT_HASH, ++port_off,
450 				    expect_hash[0], expect_hash[1]);
451 		if (!--tries) {
452 			fprintf(stderr, "too many collisions\n");
453 			return 1;
454 		}
455 	}
456 
457 	ret |= test_datapath(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_ROLLOVER,
458 			     port_off, expect_hash_rb[0], expect_hash_rb[1]);
459 	ret |= test_datapath(PACKET_FANOUT_LB,
460 			     port_off, expect_lb[0], expect_lb[1]);
461 	ret |= test_datapath(PACKET_FANOUT_ROLLOVER,
462 			     port_off, expect_rb[0], expect_rb[1]);
463 
464 	ret |= test_datapath(PACKET_FANOUT_CBPF,
465 			     port_off, expect_bpf[0], expect_bpf[1]);
466 	ret |= test_datapath(PACKET_FANOUT_EBPF,
467 			     port_off, expect_bpf[0], expect_bpf[1]);
468 
469 	set_cpuaffinity(0);
470 	ret |= test_datapath(PACKET_FANOUT_CPU, port_off,
471 			     expect_cpu0[0], expect_cpu0[1]);
472 	if (!set_cpuaffinity(1))
473 		/* TODO: test that choice alternates with previous */
474 		ret |= test_datapath(PACKET_FANOUT_CPU, port_off,
475 				     expect_cpu1[0], expect_cpu1[1]);
476 
477 	ret |= test_datapath(PACKET_FANOUT_FLAG_UNIQUEID, port_off,
478 			     expect_uniqueid[0], expect_uniqueid[1]);
479 
480 	if (ret)
481 		return 1;
482 
483 	printf("OK. All tests passed\n");
484 	return 0;
485 }
486