xref: /linux/tools/testing/selftests/bpf/prog_tests/ringbuf.c (revision 24bce201d79807b668bf9d9e0aca801c5c0d5f78)
1 // SPDX-License-Identifier: GPL-2.0
2 #define _GNU_SOURCE
3 #include <linux/compiler.h>
4 #include <asm/barrier.h>
5 #include <test_progs.h>
6 #include <sys/mman.h>
7 #include <sys/epoll.h>
8 #include <time.h>
9 #include <sched.h>
10 #include <signal.h>
11 #include <pthread.h>
12 #include <sys/sysinfo.h>
13 #include <linux/perf_event.h>
14 #include <linux/ring_buffer.h>
15 #include "test_ringbuf.lskel.h"
16 
17 #define EDONE 7777
18 
19 static int duration = 0;
20 
21 struct sample {
22 	int pid;
23 	int seq;
24 	long value;
25 	char comm[16];
26 };
27 
28 static int sample_cnt;
29 
30 static void atomic_inc(int *cnt)
31 {
32 	__atomic_add_fetch(cnt, 1, __ATOMIC_SEQ_CST);
33 }
34 
35 static int atomic_xchg(int *cnt, int val)
36 {
37 	return __atomic_exchange_n(cnt, val, __ATOMIC_SEQ_CST);
38 }
39 
40 static int process_sample(void *ctx, void *data, size_t len)
41 {
42 	struct sample *s = data;
43 
44 	atomic_inc(&sample_cnt);
45 
46 	switch (s->seq) {
47 	case 0:
48 		CHECK(s->value != 333, "sample1_value", "exp %ld, got %ld\n",
49 		      333L, s->value);
50 		return 0;
51 	case 1:
52 		CHECK(s->value != 777, "sample2_value", "exp %ld, got %ld\n",
53 		      777L, s->value);
54 		return -EDONE;
55 	default:
56 		/* we don't care about the rest */
57 		return 0;
58 	}
59 }
60 
61 static struct test_ringbuf_lskel *skel;
62 static struct ring_buffer *ringbuf;
63 
64 static void trigger_samples()
65 {
66 	skel->bss->dropped = 0;
67 	skel->bss->total = 0;
68 	skel->bss->discarded = 0;
69 
70 	/* trigger exactly two samples */
71 	skel->bss->value = 333;
72 	syscall(__NR_getpgid);
73 	skel->bss->value = 777;
74 	syscall(__NR_getpgid);
75 }
76 
77 static void *poll_thread(void *input)
78 {
79 	long timeout = (long)input;
80 
81 	return (void *)(long)ring_buffer__poll(ringbuf, timeout);
82 }
83 
84 void test_ringbuf(void)
85 {
86 	const size_t rec_sz = BPF_RINGBUF_HDR_SZ + sizeof(struct sample);
87 	pthread_t thread;
88 	long bg_ret = -1;
89 	int err, cnt, rb_fd;
90 	int page_size = getpagesize();
91 	void *mmap_ptr, *tmp_ptr;
92 
93 	skel = test_ringbuf_lskel__open();
94 	if (CHECK(!skel, "skel_open", "skeleton open failed\n"))
95 		return;
96 
97 	skel->maps.ringbuf.max_entries = page_size;
98 
99 	err = test_ringbuf_lskel__load(skel);
100 	if (CHECK(err != 0, "skel_load", "skeleton load failed\n"))
101 		goto cleanup;
102 
103 	rb_fd = skel->maps.ringbuf.map_fd;
104 	/* good read/write cons_pos */
105 	mmap_ptr = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED, rb_fd, 0);
106 	ASSERT_OK_PTR(mmap_ptr, "rw_cons_pos");
107 	tmp_ptr = mremap(mmap_ptr, page_size, 2 * page_size, MREMAP_MAYMOVE);
108 	if (!ASSERT_ERR_PTR(tmp_ptr, "rw_extend"))
109 		goto cleanup;
110 	ASSERT_ERR(mprotect(mmap_ptr, page_size, PROT_EXEC), "exec_cons_pos_protect");
111 	ASSERT_OK(munmap(mmap_ptr, page_size), "unmap_rw");
112 
113 	/* bad writeable prod_pos */
114 	mmap_ptr = mmap(NULL, page_size, PROT_WRITE, MAP_SHARED, rb_fd, page_size);
115 	err = -errno;
116 	ASSERT_ERR_PTR(mmap_ptr, "wr_prod_pos");
117 	ASSERT_EQ(err, -EPERM, "wr_prod_pos_err");
118 
119 	/* bad writeable data pages */
120 	mmap_ptr = mmap(NULL, page_size, PROT_WRITE, MAP_SHARED, rb_fd, 2 * page_size);
121 	err = -errno;
122 	ASSERT_ERR_PTR(mmap_ptr, "wr_data_page_one");
123 	ASSERT_EQ(err, -EPERM, "wr_data_page_one_err");
124 	mmap_ptr = mmap(NULL, page_size, PROT_WRITE, MAP_SHARED, rb_fd, 3 * page_size);
125 	ASSERT_ERR_PTR(mmap_ptr, "wr_data_page_two");
126 	mmap_ptr = mmap(NULL, 2 * page_size, PROT_WRITE, MAP_SHARED, rb_fd, 2 * page_size);
127 	ASSERT_ERR_PTR(mmap_ptr, "wr_data_page_all");
128 
129 	/* good read-only pages */
130 	mmap_ptr = mmap(NULL, 4 * page_size, PROT_READ, MAP_SHARED, rb_fd, 0);
131 	if (!ASSERT_OK_PTR(mmap_ptr, "ro_prod_pos"))
132 		goto cleanup;
133 
134 	ASSERT_ERR(mprotect(mmap_ptr, 4 * page_size, PROT_WRITE), "write_protect");
135 	ASSERT_ERR(mprotect(mmap_ptr, 4 * page_size, PROT_EXEC), "exec_protect");
136 	ASSERT_ERR_PTR(mremap(mmap_ptr, 0, 4 * page_size, MREMAP_MAYMOVE), "ro_remap");
137 	ASSERT_OK(munmap(mmap_ptr, 4 * page_size), "unmap_ro");
138 
139 	/* good read-only pages with initial offset */
140 	mmap_ptr = mmap(NULL, page_size, PROT_READ, MAP_SHARED, rb_fd, page_size);
141 	if (!ASSERT_OK_PTR(mmap_ptr, "ro_prod_pos"))
142 		goto cleanup;
143 
144 	ASSERT_ERR(mprotect(mmap_ptr, page_size, PROT_WRITE), "write_protect");
145 	ASSERT_ERR(mprotect(mmap_ptr, page_size, PROT_EXEC), "exec_protect");
146 	ASSERT_ERR_PTR(mremap(mmap_ptr, 0, 3 * page_size, MREMAP_MAYMOVE), "ro_remap");
147 	ASSERT_OK(munmap(mmap_ptr, page_size), "unmap_ro");
148 
149 	/* only trigger BPF program for current process */
150 	skel->bss->pid = getpid();
151 
152 	ringbuf = ring_buffer__new(skel->maps.ringbuf.map_fd,
153 				   process_sample, NULL, NULL);
154 	if (CHECK(!ringbuf, "ringbuf_create", "failed to create ringbuf\n"))
155 		goto cleanup;
156 
157 	err = test_ringbuf_lskel__attach(skel);
158 	if (CHECK(err, "skel_attach", "skeleton attachment failed: %d\n", err))
159 		goto cleanup;
160 
161 	trigger_samples();
162 
163 	/* 2 submitted + 1 discarded records */
164 	CHECK(skel->bss->avail_data != 3 * rec_sz,
165 	      "err_avail_size", "exp %ld, got %ld\n",
166 	      3L * rec_sz, skel->bss->avail_data);
167 	CHECK(skel->bss->ring_size != page_size,
168 	      "err_ring_size", "exp %ld, got %ld\n",
169 	      (long)page_size, skel->bss->ring_size);
170 	CHECK(skel->bss->cons_pos != 0,
171 	      "err_cons_pos", "exp %ld, got %ld\n",
172 	      0L, skel->bss->cons_pos);
173 	CHECK(skel->bss->prod_pos != 3 * rec_sz,
174 	      "err_prod_pos", "exp %ld, got %ld\n",
175 	      3L * rec_sz, skel->bss->prod_pos);
176 
177 	/* poll for samples */
178 	err = ring_buffer__poll(ringbuf, -1);
179 
180 	/* -EDONE is used as an indicator that we are done */
181 	if (CHECK(err != -EDONE, "err_done", "done err: %d\n", err))
182 		goto cleanup;
183 	cnt = atomic_xchg(&sample_cnt, 0);
184 	CHECK(cnt != 2, "cnt", "exp %d samples, got %d\n", 2, cnt);
185 
186 	/* we expect extra polling to return nothing */
187 	err = ring_buffer__poll(ringbuf, 0);
188 	if (CHECK(err != 0, "extra_samples", "poll result: %d\n", err))
189 		goto cleanup;
190 	cnt = atomic_xchg(&sample_cnt, 0);
191 	CHECK(cnt != 0, "cnt", "exp %d samples, got %d\n", 0, cnt);
192 
193 	CHECK(skel->bss->dropped != 0, "err_dropped", "exp %ld, got %ld\n",
194 	      0L, skel->bss->dropped);
195 	CHECK(skel->bss->total != 2, "err_total", "exp %ld, got %ld\n",
196 	      2L, skel->bss->total);
197 	CHECK(skel->bss->discarded != 1, "err_discarded", "exp %ld, got %ld\n",
198 	      1L, skel->bss->discarded);
199 
200 	/* now validate consumer position is updated and returned */
201 	trigger_samples();
202 	CHECK(skel->bss->cons_pos != 3 * rec_sz,
203 	      "err_cons_pos", "exp %ld, got %ld\n",
204 	      3L * rec_sz, skel->bss->cons_pos);
205 	err = ring_buffer__poll(ringbuf, -1);
206 	CHECK(err <= 0, "poll_err", "err %d\n", err);
207 	cnt = atomic_xchg(&sample_cnt, 0);
208 	CHECK(cnt != 2, "cnt", "exp %d samples, got %d\n", 2, cnt);
209 
210 	/* start poll in background w/ long timeout */
211 	err = pthread_create(&thread, NULL, poll_thread, (void *)(long)10000);
212 	if (CHECK(err, "bg_poll", "pthread_create failed: %d\n", err))
213 		goto cleanup;
214 
215 	/* turn off notifications now */
216 	skel->bss->flags = BPF_RB_NO_WAKEUP;
217 
218 	/* give background thread a bit of a time */
219 	usleep(50000);
220 	trigger_samples();
221 	/* sleeping arbitrarily is bad, but no better way to know that
222 	 * epoll_wait() **DID NOT** unblock in background thread
223 	 */
224 	usleep(50000);
225 	/* background poll should still be blocked */
226 	err = pthread_tryjoin_np(thread, (void **)&bg_ret);
227 	if (CHECK(err != EBUSY, "try_join", "err %d\n", err))
228 		goto cleanup;
229 
230 	/* BPF side did everything right */
231 	CHECK(skel->bss->dropped != 0, "err_dropped", "exp %ld, got %ld\n",
232 	      0L, skel->bss->dropped);
233 	CHECK(skel->bss->total != 2, "err_total", "exp %ld, got %ld\n",
234 	      2L, skel->bss->total);
235 	CHECK(skel->bss->discarded != 1, "err_discarded", "exp %ld, got %ld\n",
236 	      1L, skel->bss->discarded);
237 	cnt = atomic_xchg(&sample_cnt, 0);
238 	CHECK(cnt != 0, "cnt", "exp %d samples, got %d\n", 0, cnt);
239 
240 	/* clear flags to return to "adaptive" notification mode */
241 	skel->bss->flags = 0;
242 
243 	/* produce new samples, no notification should be triggered, because
244 	 * consumer is now behind
245 	 */
246 	trigger_samples();
247 
248 	/* background poll should still be blocked */
249 	err = pthread_tryjoin_np(thread, (void **)&bg_ret);
250 	if (CHECK(err != EBUSY, "try_join", "err %d\n", err))
251 		goto cleanup;
252 
253 	/* still no samples, because consumer is behind */
254 	cnt = atomic_xchg(&sample_cnt, 0);
255 	CHECK(cnt != 0, "cnt", "exp %d samples, got %d\n", 0, cnt);
256 
257 	skel->bss->dropped = 0;
258 	skel->bss->total = 0;
259 	skel->bss->discarded = 0;
260 
261 	skel->bss->value = 333;
262 	syscall(__NR_getpgid);
263 	/* now force notifications */
264 	skel->bss->flags = BPF_RB_FORCE_WAKEUP;
265 	skel->bss->value = 777;
266 	syscall(__NR_getpgid);
267 
268 	/* now we should get a pending notification */
269 	usleep(50000);
270 	err = pthread_tryjoin_np(thread, (void **)&bg_ret);
271 	if (CHECK(err, "join_bg", "err %d\n", err))
272 		goto cleanup;
273 
274 	if (CHECK(bg_ret <= 0, "bg_ret", "epoll_wait result: %ld", bg_ret))
275 		goto cleanup;
276 
277 	/* due to timing variations, there could still be non-notified
278 	 * samples, so consume them here to collect all the samples
279 	 */
280 	err = ring_buffer__consume(ringbuf);
281 	CHECK(err < 0, "rb_consume", "failed: %d\b", err);
282 
283 	/* 3 rounds, 2 samples each */
284 	cnt = atomic_xchg(&sample_cnt, 0);
285 	CHECK(cnt != 6, "cnt", "exp %d samples, got %d\n", 6, cnt);
286 
287 	/* BPF side did everything right */
288 	CHECK(skel->bss->dropped != 0, "err_dropped", "exp %ld, got %ld\n",
289 	      0L, skel->bss->dropped);
290 	CHECK(skel->bss->total != 2, "err_total", "exp %ld, got %ld\n",
291 	      2L, skel->bss->total);
292 	CHECK(skel->bss->discarded != 1, "err_discarded", "exp %ld, got %ld\n",
293 	      1L, skel->bss->discarded);
294 
295 	test_ringbuf_lskel__detach(skel);
296 cleanup:
297 	ring_buffer__free(ringbuf);
298 	test_ringbuf_lskel__destroy(skel);
299 }
300