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, int num_packets)
75{
76	int fd, val;
77
78	fd = socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_IP));
79	if (fd < 0) {
80		perror("socket packet");
81		exit(1);
82	}
83
84	/* fanout group ID is always 0: tests whether old groups are deleted */
85	val = ((int) typeflags) << 16;
86	if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT, &val, sizeof(val))) {
87		if (close(fd)) {
88			perror("close packet");
89			exit(1);
90		}
91		return -1;
92	}
93
94	pair_udp_setfilter(fd);
95	return fd;
96}
97
98static void sock_fanout_set_ebpf(int fd)
99{
100	const int len_off = __builtin_offsetof(struct __sk_buff, len);
101	struct bpf_insn prog[] = {
102		{ BPF_ALU64 | BPF_MOV | BPF_X,   6, 1, 0, 0 },
103		{ BPF_LDX   | BPF_W   | BPF_MEM, 0, 6, len_off, 0 },
104		{ BPF_JMP   | BPF_JGE | BPF_K,   0, 0, 1, DATA_LEN },
105		{ BPF_JMP   | BPF_JA  | BPF_K,   0, 0, 4, 0 },
106		{ BPF_LD    | BPF_B   | BPF_ABS, 0, 0, 0, 0x50 },
107		{ BPF_JMP   | BPF_JEQ | BPF_K,   0, 0, 2, DATA_CHAR },
108		{ BPF_JMP   | BPF_JEQ | BPF_K,   0, 0, 1, DATA_CHAR_1 },
109		{ BPF_ALU   | BPF_MOV | BPF_K,   0, 0, 0, 0 },
110		{ BPF_JMP   | BPF_EXIT,          0, 0, 0, 0 }
111	};
112	char log_buf[512];
113	union bpf_attr attr;
114	int pfd;
115
116	memset(&attr, 0, sizeof(attr));
117	attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
118	attr.insns = (unsigned long) prog;
119	attr.insn_cnt = sizeof(prog) / sizeof(prog[0]);
120	attr.license = (unsigned long) "GPL";
121	attr.log_buf = (unsigned long) log_buf,
122	attr.log_size = sizeof(log_buf),
123	attr.log_level = 1,
124
125	pfd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr));
126	if (pfd < 0) {
127		perror("bpf");
128		fprintf(stderr, "bpf verifier:\n%s\n", log_buf);
129		exit(1);
130	}
131
132	if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT_DATA, &pfd, sizeof(pfd))) {
133		perror("fanout data ebpf");
134		exit(1);
135	}
136
137	if (close(pfd)) {
138		perror("close ebpf");
139		exit(1);
140	}
141}
142
143static char *sock_fanout_open_ring(int fd)
144{
145	struct tpacket_req req = {
146		.tp_block_size = getpagesize(),
147		.tp_frame_size = getpagesize(),
148		.tp_block_nr   = RING_NUM_FRAMES,
149		.tp_frame_nr   = RING_NUM_FRAMES,
150	};
151	char *ring;
152	int val = TPACKET_V2;
153
154	if (setsockopt(fd, SOL_PACKET, PACKET_VERSION, (void *) &val,
155		       sizeof(val))) {
156		perror("packetsock ring setsockopt version");
157		exit(1);
158	}
159	if (setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *) &req,
160		       sizeof(req))) {
161		perror("packetsock ring setsockopt");
162		exit(1);
163	}
164
165	ring = mmap(0, req.tp_block_size * req.tp_block_nr,
166		    PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
167	if (ring == MAP_FAILED) {
168		perror("packetsock ring mmap");
169		exit(1);
170	}
171
172	return ring;
173}
174
175static int sock_fanout_read_ring(int fd, void *ring)
176{
177	struct tpacket2_hdr *header = ring;
178	int count = 0;
179
180	while (count < RING_NUM_FRAMES && header->tp_status & TP_STATUS_USER) {
181		count++;
182		header = ring + (count * getpagesize());
183	}
184
185	return count;
186}
187
188static int sock_fanout_read(int fds[], char *rings[], const int expect[])
189{
190	int ret[2];
191
192	ret[0] = sock_fanout_read_ring(fds[0], rings[0]);
193	ret[1] = sock_fanout_read_ring(fds[1], rings[1]);
194
195	fprintf(stderr, "info: count=%d,%d, expect=%d,%d\n",
196			ret[0], ret[1], expect[0], expect[1]);
197
198	if ((!(ret[0] == expect[0] && ret[1] == expect[1])) &&
199	    (!(ret[0] == expect[1] && ret[1] == expect[0]))) {
200		fprintf(stderr, "ERROR: incorrect queue lengths\n");
201		return 1;
202	}
203
204	return 0;
205}
206
207/* Test illegal mode + flag combination */
208static void test_control_single(void)
209{
210	fprintf(stderr, "test: control single socket\n");
211
212	if (sock_fanout_open(PACKET_FANOUT_ROLLOVER |
213			       PACKET_FANOUT_FLAG_ROLLOVER, 0) != -1) {
214		fprintf(stderr, "ERROR: opened socket with dual rollover\n");
215		exit(1);
216	}
217}
218
219/* Test illegal group with different modes or flags */
220static void test_control_group(void)
221{
222	int fds[2];
223
224	fprintf(stderr, "test: control multiple sockets\n");
225
226	fds[0] = sock_fanout_open(PACKET_FANOUT_HASH, 20);
227	if (fds[0] == -1) {
228		fprintf(stderr, "ERROR: failed to open HASH socket\n");
229		exit(1);
230	}
231	if (sock_fanout_open(PACKET_FANOUT_HASH |
232			       PACKET_FANOUT_FLAG_DEFRAG, 10) != -1) {
233		fprintf(stderr, "ERROR: joined group with wrong flag defrag\n");
234		exit(1);
235	}
236	if (sock_fanout_open(PACKET_FANOUT_HASH |
237			       PACKET_FANOUT_FLAG_ROLLOVER, 10) != -1) {
238		fprintf(stderr, "ERROR: joined group with wrong flag ro\n");
239		exit(1);
240	}
241	if (sock_fanout_open(PACKET_FANOUT_CPU, 10) != -1) {
242		fprintf(stderr, "ERROR: joined group with wrong mode\n");
243		exit(1);
244	}
245	fds[1] = sock_fanout_open(PACKET_FANOUT_HASH, 20);
246	if (fds[1] == -1) {
247		fprintf(stderr, "ERROR: failed to join group\n");
248		exit(1);
249	}
250	if (close(fds[1]) || close(fds[0])) {
251		fprintf(stderr, "ERROR: closing sockets\n");
252		exit(1);
253	}
254}
255
256static int test_datapath(uint16_t typeflags, int port_off,
257			 const int expect1[], const int expect2[])
258{
259	const int expect0[] = { 0, 0 };
260	char *rings[2];
261	uint8_t type = typeflags & 0xFF;
262	int fds[2], fds_udp[2][2], ret;
263
264	fprintf(stderr, "test: datapath 0x%hx\n", typeflags);
265
266	fds[0] = sock_fanout_open(typeflags, 20);
267	fds[1] = sock_fanout_open(typeflags, 20);
268	if (fds[0] == -1 || fds[1] == -1) {
269		fprintf(stderr, "ERROR: failed open\n");
270		exit(1);
271	}
272	if (type == PACKET_FANOUT_CBPF)
273		sock_setfilter(fds[0], SOL_PACKET, PACKET_FANOUT_DATA);
274	else if (type == PACKET_FANOUT_EBPF)
275		sock_fanout_set_ebpf(fds[0]);
276
277	rings[0] = sock_fanout_open_ring(fds[0]);
278	rings[1] = sock_fanout_open_ring(fds[1]);
279	pair_udp_open(fds_udp[0], PORT_BASE);
280	pair_udp_open(fds_udp[1], PORT_BASE + port_off);
281	sock_fanout_read(fds, rings, expect0);
282
283	/* Send data, but not enough to overflow a queue */
284	pair_udp_send(fds_udp[0], 15);
285	pair_udp_send_char(fds_udp[1], 5, DATA_CHAR_1);
286	ret = sock_fanout_read(fds, rings, expect1);
287
288	/* Send more data, overflow the queue */
289	pair_udp_send_char(fds_udp[0], 15, DATA_CHAR_1);
290	/* TODO: ensure consistent order between expect1 and expect2 */
291	ret |= sock_fanout_read(fds, rings, expect2);
292
293	if (munmap(rings[1], RING_NUM_FRAMES * getpagesize()) ||
294	    munmap(rings[0], RING_NUM_FRAMES * getpagesize())) {
295		fprintf(stderr, "close rings\n");
296		exit(1);
297	}
298	if (close(fds_udp[1][1]) || close(fds_udp[1][0]) ||
299	    close(fds_udp[0][1]) || close(fds_udp[0][0]) ||
300	    close(fds[1]) || close(fds[0])) {
301		fprintf(stderr, "close datapath\n");
302		exit(1);
303	}
304
305	return ret;
306}
307
308static int set_cpuaffinity(int cpuid)
309{
310	cpu_set_t mask;
311
312	CPU_ZERO(&mask);
313	CPU_SET(cpuid, &mask);
314	if (sched_setaffinity(0, sizeof(mask), &mask)) {
315		if (errno != EINVAL) {
316			fprintf(stderr, "setaffinity %d\n", cpuid);
317			exit(1);
318		}
319		return 1;
320	}
321
322	return 0;
323}
324
325int main(int argc, char **argv)
326{
327	const int expect_hash[2][2]	= { { 15, 5 },  { 20, 5 } };
328	const int expect_hash_rb[2][2]	= { { 15, 5 },  { 20, 15 } };
329	const int expect_lb[2][2]	= { { 10, 10 }, { 18, 17 } };
330	const int expect_rb[2][2]	= { { 15, 5 },  { 20, 15 } };
331	const int expect_cpu0[2][2]	= { { 20, 0 },  { 20, 0 } };
332	const int expect_cpu1[2][2]	= { { 0, 20 },  { 0, 20 } };
333	const int expect_bpf[2][2]	= { { 15, 5 },  { 15, 20 } };
334	int port_off = 2, tries = 5, ret;
335
336	test_control_single();
337	test_control_group();
338
339	/* find a set of ports that do not collide onto the same socket */
340	ret = test_datapath(PACKET_FANOUT_HASH, port_off,
341			    expect_hash[0], expect_hash[1]);
342	while (ret && tries--) {
343		fprintf(stderr, "info: trying alternate ports (%d)\n", tries);
344		ret = test_datapath(PACKET_FANOUT_HASH, ++port_off,
345				    expect_hash[0], expect_hash[1]);
346	}
347
348	ret |= test_datapath(PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_ROLLOVER,
349			     port_off, expect_hash_rb[0], expect_hash_rb[1]);
350	ret |= test_datapath(PACKET_FANOUT_LB,
351			     port_off, expect_lb[0], expect_lb[1]);
352	ret |= test_datapath(PACKET_FANOUT_ROLLOVER,
353			     port_off, expect_rb[0], expect_rb[1]);
354
355	ret |= test_datapath(PACKET_FANOUT_CBPF,
356			     port_off, expect_bpf[0], expect_bpf[1]);
357	ret |= test_datapath(PACKET_FANOUT_EBPF,
358			     port_off, expect_bpf[0], expect_bpf[1]);
359
360	set_cpuaffinity(0);
361	ret |= test_datapath(PACKET_FANOUT_CPU, port_off,
362			     expect_cpu0[0], expect_cpu0[1]);
363	if (!set_cpuaffinity(1))
364		/* TODO: test that choice alternates with previous */
365		ret |= test_datapath(PACKET_FANOUT_CPU, port_off,
366				     expect_cpu1[0], expect_cpu1[1]);
367
368	if (ret)
369		return 1;
370
371	printf("OK. All tests passed\n");
372	return 0;
373}
374