xref: /linux/samples/bpf/task_fd_query_user.c (revision b8d312aa075f33282565467662c4628dae0a2aff)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <stdio.h>
4 #include <stdlib.h>
5 #include <signal.h>
6 #include <unistd.h>
7 #include <stdbool.h>
8 #include <string.h>
9 #include <stdint.h>
10 #include <fcntl.h>
11 #include <linux/bpf.h>
12 #include <sys/ioctl.h>
13 #include <sys/resource.h>
14 #include <sys/types.h>
15 #include <sys/stat.h>
16 
17 #include "libbpf.h"
18 #include "bpf_load.h"
19 #include "bpf_util.h"
20 #include "perf-sys.h"
21 #include "trace_helpers.h"
22 
23 #define CHECK_PERROR_RET(condition) ({			\
24 	int __ret = !!(condition);			\
25 	if (__ret) {					\
26 		printf("FAIL: %s:\n", __func__);	\
27 		perror("    ");			\
28 		return -1;				\
29 	}						\
30 })
31 
32 #define CHECK_AND_RET(condition) ({			\
33 	int __ret = !!(condition);			\
34 	if (__ret)					\
35 		return -1;				\
36 })
37 
38 static __u64 ptr_to_u64(void *ptr)
39 {
40 	return (__u64) (unsigned long) ptr;
41 }
42 
43 #define PMU_TYPE_FILE "/sys/bus/event_source/devices/%s/type"
44 static int bpf_find_probe_type(const char *event_type)
45 {
46 	char buf[256];
47 	int fd, ret;
48 
49 	ret = snprintf(buf, sizeof(buf), PMU_TYPE_FILE, event_type);
50 	CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
51 
52 	fd = open(buf, O_RDONLY);
53 	CHECK_PERROR_RET(fd < 0);
54 
55 	ret = read(fd, buf, sizeof(buf));
56 	close(fd);
57 	CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
58 
59 	errno = 0;
60 	ret = (int)strtol(buf, NULL, 10);
61 	CHECK_PERROR_RET(errno);
62 	return ret;
63 }
64 
65 #define PMU_RETPROBE_FILE "/sys/bus/event_source/devices/%s/format/retprobe"
66 static int bpf_get_retprobe_bit(const char *event_type)
67 {
68 	char buf[256];
69 	int fd, ret;
70 
71 	ret = snprintf(buf, sizeof(buf), PMU_RETPROBE_FILE, event_type);
72 	CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
73 
74 	fd = open(buf, O_RDONLY);
75 	CHECK_PERROR_RET(fd < 0);
76 
77 	ret = read(fd, buf, sizeof(buf));
78 	close(fd);
79 	CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
80 	CHECK_PERROR_RET(strlen(buf) < strlen("config:"));
81 
82 	errno = 0;
83 	ret = (int)strtol(buf + strlen("config:"), NULL, 10);
84 	CHECK_PERROR_RET(errno);
85 	return ret;
86 }
87 
88 static int test_debug_fs_kprobe(int prog_fd_idx, const char *fn_name,
89 				__u32 expected_fd_type)
90 {
91 	__u64 probe_offset, probe_addr;
92 	__u32 len, prog_id, fd_type;
93 	char buf[256];
94 	int err;
95 
96 	len = sizeof(buf);
97 	err = bpf_task_fd_query(getpid(), event_fd[prog_fd_idx], 0, buf, &len,
98 				&prog_id, &fd_type, &probe_offset,
99 				&probe_addr);
100 	if (err < 0) {
101 		printf("FAIL: %s, for event_fd idx %d, fn_name %s\n",
102 		       __func__, prog_fd_idx, fn_name);
103 		perror("    :");
104 		return -1;
105 	}
106 	if (strcmp(buf, fn_name) != 0 ||
107 	    fd_type != expected_fd_type ||
108 	    probe_offset != 0x0 || probe_addr != 0x0) {
109 		printf("FAIL: bpf_trace_event_query(event_fd[%d]):\n",
110 		       prog_fd_idx);
111 		printf("buf: %s, fd_type: %u, probe_offset: 0x%llx,"
112 		       " probe_addr: 0x%llx\n",
113 		       buf, fd_type, probe_offset, probe_addr);
114 		return -1;
115 	}
116 	return 0;
117 }
118 
119 static int test_nondebug_fs_kuprobe_common(const char *event_type,
120 	const char *name, __u64 offset, __u64 addr, bool is_return,
121 	char *buf, __u32 *buf_len, __u32 *prog_id, __u32 *fd_type,
122 	__u64 *probe_offset, __u64 *probe_addr)
123 {
124 	int is_return_bit = bpf_get_retprobe_bit(event_type);
125 	int type = bpf_find_probe_type(event_type);
126 	struct perf_event_attr attr = {};
127 	int fd;
128 
129 	if (type < 0 || is_return_bit < 0) {
130 		printf("FAIL: %s incorrect type (%d) or is_return_bit (%d)\n",
131 			__func__, type, is_return_bit);
132 		return -1;
133 	}
134 
135 	attr.sample_period = 1;
136 	attr.wakeup_events = 1;
137 	if (is_return)
138 		attr.config |= 1 << is_return_bit;
139 
140 	if (name) {
141 		attr.config1 = ptr_to_u64((void *)name);
142 		attr.config2 = offset;
143 	} else {
144 		attr.config1 = 0;
145 		attr.config2 = addr;
146 	}
147 	attr.size = sizeof(attr);
148 	attr.type = type;
149 
150 	fd = sys_perf_event_open(&attr, -1, 0, -1, 0);
151 	CHECK_PERROR_RET(fd < 0);
152 
153 	CHECK_PERROR_RET(ioctl(fd, PERF_EVENT_IOC_ENABLE, 0) < 0);
154 	CHECK_PERROR_RET(ioctl(fd, PERF_EVENT_IOC_SET_BPF, prog_fd[0]) < 0);
155 	CHECK_PERROR_RET(bpf_task_fd_query(getpid(), fd, 0, buf, buf_len,
156 			 prog_id, fd_type, probe_offset, probe_addr) < 0);
157 
158 	return 0;
159 }
160 
161 static int test_nondebug_fs_probe(const char *event_type, const char *name,
162 				  __u64 offset, __u64 addr, bool is_return,
163 				  __u32 expected_fd_type,
164 				  __u32 expected_ret_fd_type,
165 				  char *buf, __u32 buf_len)
166 {
167 	__u64 probe_offset, probe_addr;
168 	__u32 prog_id, fd_type;
169 	int err;
170 
171 	err = test_nondebug_fs_kuprobe_common(event_type, name,
172 					      offset, addr, is_return,
173 					      buf, &buf_len, &prog_id,
174 					      &fd_type, &probe_offset,
175 					      &probe_addr);
176 	if (err < 0) {
177 		printf("FAIL: %s, "
178 		       "for name %s, offset 0x%llx, addr 0x%llx, is_return %d\n",
179 		       __func__, name ? name : "", offset, addr, is_return);
180 		perror("    :");
181 		return -1;
182 	}
183 	if ((is_return && fd_type != expected_ret_fd_type) ||
184 	    (!is_return && fd_type != expected_fd_type)) {
185 		printf("FAIL: %s, incorrect fd_type %u\n",
186 		       __func__, fd_type);
187 		return -1;
188 	}
189 	if (name) {
190 		if (strcmp(name, buf) != 0) {
191 			printf("FAIL: %s, incorrect buf %s\n", __func__, buf);
192 			return -1;
193 		}
194 		if (probe_offset != offset) {
195 			printf("FAIL: %s, incorrect probe_offset 0x%llx\n",
196 			       __func__, probe_offset);
197 			return -1;
198 		}
199 	} else {
200 		if (buf_len != 0) {
201 			printf("FAIL: %s, incorrect buf %p\n",
202 			       __func__, buf);
203 			return -1;
204 		}
205 
206 		if (probe_addr != addr) {
207 			printf("FAIL: %s, incorrect probe_addr 0x%llx\n",
208 			       __func__, probe_addr);
209 			return -1;
210 		}
211 	}
212 	return 0;
213 }
214 
215 static int test_debug_fs_uprobe(char *binary_path, long offset, bool is_return)
216 {
217 	const char *event_type = "uprobe";
218 	struct perf_event_attr attr = {};
219 	char buf[256], event_alias[sizeof("test_1234567890")];
220 	__u64 probe_offset, probe_addr;
221 	__u32 len, prog_id, fd_type;
222 	int err, res, kfd, efd;
223 	ssize_t bytes;
224 
225 	snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/%s_events",
226 		 event_type);
227 	kfd = open(buf, O_WRONLY | O_APPEND, 0);
228 	CHECK_PERROR_RET(kfd < 0);
229 
230 	res = snprintf(event_alias, sizeof(event_alias), "test_%d", getpid());
231 	CHECK_PERROR_RET(res < 0 || res >= sizeof(event_alias));
232 
233 	res = snprintf(buf, sizeof(buf), "%c:%ss/%s %s:0x%lx",
234 		       is_return ? 'r' : 'p', event_type, event_alias,
235 		       binary_path, offset);
236 	CHECK_PERROR_RET(res < 0 || res >= sizeof(buf));
237 	CHECK_PERROR_RET(write(kfd, buf, strlen(buf)) < 0);
238 
239 	close(kfd);
240 	kfd = -1;
241 
242 	snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/events/%ss/%s/id",
243 		 event_type, event_alias);
244 	efd = open(buf, O_RDONLY, 0);
245 	CHECK_PERROR_RET(efd < 0);
246 
247 	bytes = read(efd, buf, sizeof(buf));
248 	CHECK_PERROR_RET(bytes <= 0 || bytes >= sizeof(buf));
249 	close(efd);
250 	buf[bytes] = '\0';
251 
252 	attr.config = strtol(buf, NULL, 0);
253 	attr.type = PERF_TYPE_TRACEPOINT;
254 	attr.sample_period = 1;
255 	attr.wakeup_events = 1;
256 	kfd = sys_perf_event_open(&attr, -1, 0, -1, PERF_FLAG_FD_CLOEXEC);
257 	CHECK_PERROR_RET(kfd < 0);
258 	CHECK_PERROR_RET(ioctl(kfd, PERF_EVENT_IOC_SET_BPF, prog_fd[0]) < 0);
259 	CHECK_PERROR_RET(ioctl(kfd, PERF_EVENT_IOC_ENABLE, 0) < 0);
260 
261 	len = sizeof(buf);
262 	err = bpf_task_fd_query(getpid(), kfd, 0, buf, &len,
263 				&prog_id, &fd_type, &probe_offset,
264 				&probe_addr);
265 	if (err < 0) {
266 		printf("FAIL: %s, binary_path %s\n", __func__, binary_path);
267 		perror("    :");
268 		return -1;
269 	}
270 	if ((is_return && fd_type != BPF_FD_TYPE_URETPROBE) ||
271 	    (!is_return && fd_type != BPF_FD_TYPE_UPROBE)) {
272 		printf("FAIL: %s, incorrect fd_type %u\n", __func__,
273 		       fd_type);
274 		return -1;
275 	}
276 	if (strcmp(binary_path, buf) != 0) {
277 		printf("FAIL: %s, incorrect buf %s\n", __func__, buf);
278 		return -1;
279 	}
280 	if (probe_offset != offset) {
281 		printf("FAIL: %s, incorrect probe_offset 0x%llx\n", __func__,
282 		       probe_offset);
283 		return -1;
284 	}
285 
286 	close(kfd);
287 	return 0;
288 }
289 
290 int main(int argc, char **argv)
291 {
292 	struct rlimit r = {1024*1024, RLIM_INFINITY};
293 	extern char __executable_start;
294 	char filename[256], buf[256];
295 	__u64 uprobe_file_offset;
296 
297 	snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
298 	if (setrlimit(RLIMIT_MEMLOCK, &r)) {
299 		perror("setrlimit(RLIMIT_MEMLOCK)");
300 		return 1;
301 	}
302 
303 	if (load_kallsyms()) {
304 		printf("failed to process /proc/kallsyms\n");
305 		return 1;
306 	}
307 
308 	if (load_bpf_file(filename)) {
309 		printf("%s", bpf_log_buf);
310 		return 1;
311 	}
312 
313 	/* test two functions in the corresponding *_kern.c file */
314 	CHECK_AND_RET(test_debug_fs_kprobe(0, "blk_mq_start_request",
315 					   BPF_FD_TYPE_KPROBE));
316 	CHECK_AND_RET(test_debug_fs_kprobe(1, "blk_account_io_completion",
317 					   BPF_FD_TYPE_KRETPROBE));
318 
319 	/* test nondebug fs kprobe */
320 	CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x0, 0x0,
321 					     false, BPF_FD_TYPE_KPROBE,
322 					     BPF_FD_TYPE_KRETPROBE,
323 					     buf, sizeof(buf)));
324 #ifdef __x86_64__
325 	/* set a kprobe on "bpf_check + 0x5", which is x64 specific */
326 	CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x5, 0x0,
327 					     false, BPF_FD_TYPE_KPROBE,
328 					     BPF_FD_TYPE_KRETPROBE,
329 					     buf, sizeof(buf)));
330 #endif
331 	CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x0, 0x0,
332 					     true, BPF_FD_TYPE_KPROBE,
333 					     BPF_FD_TYPE_KRETPROBE,
334 					     buf, sizeof(buf)));
335 	CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
336 					     ksym_get_addr("bpf_check"), false,
337 					     BPF_FD_TYPE_KPROBE,
338 					     BPF_FD_TYPE_KRETPROBE,
339 					     buf, sizeof(buf)));
340 	CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
341 					     ksym_get_addr("bpf_check"), false,
342 					     BPF_FD_TYPE_KPROBE,
343 					     BPF_FD_TYPE_KRETPROBE,
344 					     NULL, 0));
345 	CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
346 					     ksym_get_addr("bpf_check"), true,
347 					     BPF_FD_TYPE_KPROBE,
348 					     BPF_FD_TYPE_KRETPROBE,
349 					     buf, sizeof(buf)));
350 	CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
351 					     ksym_get_addr("bpf_check"), true,
352 					     BPF_FD_TYPE_KPROBE,
353 					     BPF_FD_TYPE_KRETPROBE,
354 					     0, 0));
355 
356 	/* test nondebug fs uprobe */
357 	/* the calculation of uprobe file offset is based on gcc 7.3.1 on x64
358 	 * and the default linker script, which defines __executable_start as
359 	 * the start of the .text section. The calculation could be different
360 	 * on different systems with different compilers. The right way is
361 	 * to parse the ELF file. We took a shortcut here.
362 	 */
363 	uprobe_file_offset = (__u64)main - (__u64)&__executable_start;
364 	CHECK_AND_RET(test_nondebug_fs_probe("uprobe", (char *)argv[0],
365 					     uprobe_file_offset, 0x0, false,
366 					     BPF_FD_TYPE_UPROBE,
367 					     BPF_FD_TYPE_URETPROBE,
368 					     buf, sizeof(buf)));
369 	CHECK_AND_RET(test_nondebug_fs_probe("uprobe", (char *)argv[0],
370 					     uprobe_file_offset, 0x0, true,
371 					     BPF_FD_TYPE_UPROBE,
372 					     BPF_FD_TYPE_URETPROBE,
373 					     buf, sizeof(buf)));
374 
375 	/* test debug fs uprobe */
376 	CHECK_AND_RET(test_debug_fs_uprobe((char *)argv[0], uprobe_file_offset,
377 					   false));
378 	CHECK_AND_RET(test_debug_fs_uprobe((char *)argv[0], uprobe_file_offset,
379 					   true));
380 
381 	return 0;
382 }
383