xref: /openbmc/linux/tools/lib/bpf/libbpf_probes.c (revision c4e78957)
1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 /* Copyright (c) 2019 Netronome Systems, Inc. */
3 
4 #include <errno.h>
5 #include <fcntl.h>
6 #include <string.h>
7 #include <stdlib.h>
8 #include <unistd.h>
9 #include <net/if.h>
10 #include <sys/utsname.h>
11 
12 #include <linux/btf.h>
13 #include <linux/filter.h>
14 #include <linux/kernel.h>
15 
16 #include "bpf.h"
17 #include "libbpf.h"
18 #include "libbpf_internal.h"
19 
20 static bool grep(const char *buffer, const char *pattern)
21 {
22 	return !!strstr(buffer, pattern);
23 }
24 
25 static int get_vendor_id(int ifindex)
26 {
27 	char ifname[IF_NAMESIZE], path[64], buf[8];
28 	ssize_t len;
29 	int fd;
30 
31 	if (!if_indextoname(ifindex, ifname))
32 		return -1;
33 
34 	snprintf(path, sizeof(path), "/sys/class/net/%s/device/vendor", ifname);
35 
36 	fd = open(path, O_RDONLY | O_CLOEXEC);
37 	if (fd < 0)
38 		return -1;
39 
40 	len = read(fd, buf, sizeof(buf));
41 	close(fd);
42 	if (len < 0)
43 		return -1;
44 	if (len >= (ssize_t)sizeof(buf))
45 		return -1;
46 	buf[len] = '\0';
47 
48 	return strtol(buf, NULL, 0);
49 }
50 
51 static int probe_prog_load(enum bpf_prog_type prog_type,
52 			   const struct bpf_insn *insns, size_t insns_cnt,
53 			   char *log_buf, size_t log_buf_sz,
54 			   __u32 ifindex)
55 {
56 	LIBBPF_OPTS(bpf_prog_load_opts, opts,
57 		.log_buf = log_buf,
58 		.log_size = log_buf_sz,
59 		.log_level = log_buf ? 1 : 0,
60 		.prog_ifindex = ifindex,
61 	);
62 	int fd, err, exp_err = 0;
63 	const char *exp_msg = NULL;
64 	char buf[4096];
65 
66 	switch (prog_type) {
67 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
68 		opts.expected_attach_type = BPF_CGROUP_INET4_CONNECT;
69 		break;
70 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
71 		opts.expected_attach_type = BPF_CGROUP_GETSOCKOPT;
72 		break;
73 	case BPF_PROG_TYPE_SK_LOOKUP:
74 		opts.expected_attach_type = BPF_SK_LOOKUP;
75 		break;
76 	case BPF_PROG_TYPE_KPROBE:
77 		opts.kern_version = get_kernel_version();
78 		break;
79 	case BPF_PROG_TYPE_LIRC_MODE2:
80 		opts.expected_attach_type = BPF_LIRC_MODE2;
81 		break;
82 	case BPF_PROG_TYPE_TRACING:
83 	case BPF_PROG_TYPE_LSM:
84 		opts.log_buf = buf;
85 		opts.log_size = sizeof(buf);
86 		opts.log_level = 1;
87 		if (prog_type == BPF_PROG_TYPE_TRACING)
88 			opts.expected_attach_type = BPF_TRACE_FENTRY;
89 		else
90 			opts.expected_attach_type = BPF_MODIFY_RETURN;
91 		opts.attach_btf_id = 1;
92 
93 		exp_err = -EINVAL;
94 		exp_msg = "attach_btf_id 1 is not a function";
95 		break;
96 	case BPF_PROG_TYPE_EXT:
97 		opts.log_buf = buf;
98 		opts.log_size = sizeof(buf);
99 		opts.log_level = 1;
100 		opts.attach_btf_id = 1;
101 
102 		exp_err = -EINVAL;
103 		exp_msg = "Cannot replace kernel functions";
104 		break;
105 	case BPF_PROG_TYPE_SYSCALL:
106 		opts.prog_flags = BPF_F_SLEEPABLE;
107 		break;
108 	case BPF_PROG_TYPE_STRUCT_OPS:
109 		exp_err = -524; /* -ENOTSUPP */
110 		break;
111 	case BPF_PROG_TYPE_UNSPEC:
112 	case BPF_PROG_TYPE_SOCKET_FILTER:
113 	case BPF_PROG_TYPE_SCHED_CLS:
114 	case BPF_PROG_TYPE_SCHED_ACT:
115 	case BPF_PROG_TYPE_TRACEPOINT:
116 	case BPF_PROG_TYPE_XDP:
117 	case BPF_PROG_TYPE_PERF_EVENT:
118 	case BPF_PROG_TYPE_CGROUP_SKB:
119 	case BPF_PROG_TYPE_CGROUP_SOCK:
120 	case BPF_PROG_TYPE_LWT_IN:
121 	case BPF_PROG_TYPE_LWT_OUT:
122 	case BPF_PROG_TYPE_LWT_XMIT:
123 	case BPF_PROG_TYPE_SOCK_OPS:
124 	case BPF_PROG_TYPE_SK_SKB:
125 	case BPF_PROG_TYPE_CGROUP_DEVICE:
126 	case BPF_PROG_TYPE_SK_MSG:
127 	case BPF_PROG_TYPE_RAW_TRACEPOINT:
128 	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
129 	case BPF_PROG_TYPE_LWT_SEG6LOCAL:
130 	case BPF_PROG_TYPE_SK_REUSEPORT:
131 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
132 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
133 		break;
134 	default:
135 		return -EOPNOTSUPP;
136 	}
137 
138 	fd = bpf_prog_load(prog_type, NULL, "GPL", insns, insns_cnt, &opts);
139 	err = -errno;
140 	if (fd >= 0)
141 		close(fd);
142 	if (exp_err) {
143 		if (fd >= 0 || err != exp_err)
144 			return 0;
145 		if (exp_msg && !strstr(buf, exp_msg))
146 			return 0;
147 		return 1;
148 	}
149 	return fd >= 0 ? 1 : 0;
150 }
151 
152 int libbpf_probe_bpf_prog_type(enum bpf_prog_type prog_type, const void *opts)
153 {
154 	struct bpf_insn insns[] = {
155 		BPF_MOV64_IMM(BPF_REG_0, 0),
156 		BPF_EXIT_INSN()
157 	};
158 	const size_t insn_cnt = ARRAY_SIZE(insns);
159 	int ret;
160 
161 	if (opts)
162 		return libbpf_err(-EINVAL);
163 
164 	ret = probe_prog_load(prog_type, insns, insn_cnt, NULL, 0, 0);
165 	return libbpf_err(ret);
166 }
167 
168 bool bpf_probe_prog_type(enum bpf_prog_type prog_type, __u32 ifindex)
169 {
170 	struct bpf_insn insns[2] = {
171 		BPF_MOV64_IMM(BPF_REG_0, 0),
172 		BPF_EXIT_INSN()
173 	};
174 
175 	/* prefer libbpf_probe_bpf_prog_type() unless offload is requested */
176 	if (ifindex == 0)
177 		return libbpf_probe_bpf_prog_type(prog_type, NULL) == 1;
178 
179 	if (ifindex && prog_type == BPF_PROG_TYPE_SCHED_CLS)
180 		/* nfp returns -EINVAL on exit(0) with TC offload */
181 		insns[0].imm = 2;
182 
183 	errno = 0;
184 	probe_prog_load(prog_type, insns, ARRAY_SIZE(insns), NULL, 0, ifindex);
185 
186 	return errno != EINVAL && errno != EOPNOTSUPP;
187 }
188 
189 int libbpf__load_raw_btf(const char *raw_types, size_t types_len,
190 			 const char *str_sec, size_t str_len)
191 {
192 	struct btf_header hdr = {
193 		.magic = BTF_MAGIC,
194 		.version = BTF_VERSION,
195 		.hdr_len = sizeof(struct btf_header),
196 		.type_len = types_len,
197 		.str_off = types_len,
198 		.str_len = str_len,
199 	};
200 	int btf_fd, btf_len;
201 	__u8 *raw_btf;
202 
203 	btf_len = hdr.hdr_len + hdr.type_len + hdr.str_len;
204 	raw_btf = malloc(btf_len);
205 	if (!raw_btf)
206 		return -ENOMEM;
207 
208 	memcpy(raw_btf, &hdr, sizeof(hdr));
209 	memcpy(raw_btf + hdr.hdr_len, raw_types, hdr.type_len);
210 	memcpy(raw_btf + hdr.hdr_len + hdr.type_len, str_sec, hdr.str_len);
211 
212 	btf_fd = bpf_btf_load(raw_btf, btf_len, NULL);
213 
214 	free(raw_btf);
215 	return btf_fd;
216 }
217 
218 static int load_local_storage_btf(void)
219 {
220 	const char strs[] = "\0bpf_spin_lock\0val\0cnt\0l";
221 	/* struct bpf_spin_lock {
222 	 *   int val;
223 	 * };
224 	 * struct val {
225 	 *   int cnt;
226 	 *   struct bpf_spin_lock l;
227 	 * };
228 	 */
229 	__u32 types[] = {
230 		/* int */
231 		BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
232 		/* struct bpf_spin_lock */                      /* [2] */
233 		BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 1), 4),
234 		BTF_MEMBER_ENC(15, 1, 0), /* int val; */
235 		/* struct val */                                /* [3] */
236 		BTF_TYPE_ENC(15, BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 2), 8),
237 		BTF_MEMBER_ENC(19, 1, 0), /* int cnt; */
238 		BTF_MEMBER_ENC(23, 2, 32),/* struct bpf_spin_lock l; */
239 	};
240 
241 	return libbpf__load_raw_btf((char *)types, sizeof(types),
242 				     strs, sizeof(strs));
243 }
244 
245 static int probe_map_create(enum bpf_map_type map_type, __u32 ifindex)
246 {
247 	LIBBPF_OPTS(bpf_map_create_opts, opts);
248 	int key_size, value_size, max_entries;
249 	__u32 btf_key_type_id = 0, btf_value_type_id = 0;
250 	int fd = -1, btf_fd = -1, fd_inner = -1, exp_err = 0, err;
251 
252 	opts.map_ifindex = ifindex;
253 
254 	key_size	= sizeof(__u32);
255 	value_size	= sizeof(__u32);
256 	max_entries	= 1;
257 
258 	switch (map_type) {
259 	case BPF_MAP_TYPE_STACK_TRACE:
260 		value_size	= sizeof(__u64);
261 		break;
262 	case BPF_MAP_TYPE_LPM_TRIE:
263 		key_size	= sizeof(__u64);
264 		value_size	= sizeof(__u64);
265 		opts.map_flags	= BPF_F_NO_PREALLOC;
266 		break;
267 	case BPF_MAP_TYPE_CGROUP_STORAGE:
268 	case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
269 		key_size	= sizeof(struct bpf_cgroup_storage_key);
270 		value_size	= sizeof(__u64);
271 		max_entries	= 0;
272 		break;
273 	case BPF_MAP_TYPE_QUEUE:
274 	case BPF_MAP_TYPE_STACK:
275 		key_size	= 0;
276 		break;
277 	case BPF_MAP_TYPE_SK_STORAGE:
278 	case BPF_MAP_TYPE_INODE_STORAGE:
279 	case BPF_MAP_TYPE_TASK_STORAGE:
280 		btf_key_type_id = 1;
281 		btf_value_type_id = 3;
282 		value_size = 8;
283 		max_entries = 0;
284 		opts.map_flags = BPF_F_NO_PREALLOC;
285 		btf_fd = load_local_storage_btf();
286 		if (btf_fd < 0)
287 			return btf_fd;
288 		break;
289 	case BPF_MAP_TYPE_RINGBUF:
290 		key_size = 0;
291 		value_size = 0;
292 		max_entries = 4096;
293 		break;
294 	case BPF_MAP_TYPE_STRUCT_OPS:
295 		/* we'll get -ENOTSUPP for invalid BTF type ID for struct_ops */
296 		opts.btf_vmlinux_value_type_id = 1;
297 		exp_err = -524; /* -ENOTSUPP */
298 		break;
299 	case BPF_MAP_TYPE_BLOOM_FILTER:
300 		key_size = 0;
301 		max_entries = 1;
302 		break;
303 	case BPF_MAP_TYPE_HASH:
304 	case BPF_MAP_TYPE_ARRAY:
305 	case BPF_MAP_TYPE_PROG_ARRAY:
306 	case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
307 	case BPF_MAP_TYPE_PERCPU_HASH:
308 	case BPF_MAP_TYPE_PERCPU_ARRAY:
309 	case BPF_MAP_TYPE_CGROUP_ARRAY:
310 	case BPF_MAP_TYPE_LRU_HASH:
311 	case BPF_MAP_TYPE_LRU_PERCPU_HASH:
312 	case BPF_MAP_TYPE_ARRAY_OF_MAPS:
313 	case BPF_MAP_TYPE_HASH_OF_MAPS:
314 	case BPF_MAP_TYPE_DEVMAP:
315 	case BPF_MAP_TYPE_DEVMAP_HASH:
316 	case BPF_MAP_TYPE_SOCKMAP:
317 	case BPF_MAP_TYPE_CPUMAP:
318 	case BPF_MAP_TYPE_XSKMAP:
319 	case BPF_MAP_TYPE_SOCKHASH:
320 	case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
321 		break;
322 	case BPF_MAP_TYPE_UNSPEC:
323 	default:
324 		return -EOPNOTSUPP;
325 	}
326 
327 	if (map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
328 	    map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
329 		/* TODO: probe for device, once libbpf has a function to create
330 		 * map-in-map for offload
331 		 */
332 		if (ifindex)
333 			goto cleanup;
334 
335 		fd_inner = bpf_map_create(BPF_MAP_TYPE_HASH, NULL,
336 					  sizeof(__u32), sizeof(__u32), 1, NULL);
337 		if (fd_inner < 0)
338 			goto cleanup;
339 
340 		opts.inner_map_fd = fd_inner;
341 	}
342 
343 	if (btf_fd >= 0) {
344 		opts.btf_fd = btf_fd;
345 		opts.btf_key_type_id = btf_key_type_id;
346 		opts.btf_value_type_id = btf_value_type_id;
347 	}
348 
349 	fd = bpf_map_create(map_type, NULL, key_size, value_size, max_entries, &opts);
350 	err = -errno;
351 
352 cleanup:
353 	if (fd >= 0)
354 		close(fd);
355 	if (fd_inner >= 0)
356 		close(fd_inner);
357 	if (btf_fd >= 0)
358 		close(btf_fd);
359 
360 	if (exp_err)
361 		return fd < 0 && err == exp_err ? 1 : 0;
362 	else
363 		return fd >= 0 ? 1 : 0;
364 }
365 
366 int libbpf_probe_bpf_map_type(enum bpf_map_type map_type, const void *opts)
367 {
368 	int ret;
369 
370 	if (opts)
371 		return libbpf_err(-EINVAL);
372 
373 	ret = probe_map_create(map_type, 0);
374 	return libbpf_err(ret);
375 }
376 
377 bool bpf_probe_map_type(enum bpf_map_type map_type, __u32 ifindex)
378 {
379 	return probe_map_create(map_type, ifindex) == 1;
380 }
381 
382 int libbpf_probe_bpf_helper(enum bpf_prog_type prog_type, enum bpf_func_id helper_id,
383 			    const void *opts)
384 {
385 	struct bpf_insn insns[] = {
386 		BPF_EMIT_CALL((__u32)helper_id),
387 		BPF_EXIT_INSN(),
388 	};
389 	const size_t insn_cnt = ARRAY_SIZE(insns);
390 	char buf[4096];
391 	int ret;
392 
393 	if (opts)
394 		return libbpf_err(-EINVAL);
395 
396 	/* we can't successfully load all prog types to check for BPF helper
397 	 * support, so bail out with -EOPNOTSUPP error
398 	 */
399 	switch (prog_type) {
400 	case BPF_PROG_TYPE_TRACING:
401 	case BPF_PROG_TYPE_EXT:
402 	case BPF_PROG_TYPE_LSM:
403 	case BPF_PROG_TYPE_STRUCT_OPS:
404 		return -EOPNOTSUPP;
405 	default:
406 		break;
407 	}
408 
409 	buf[0] = '\0';
410 	ret = probe_prog_load(prog_type, insns, insn_cnt, buf, sizeof(buf), 0);
411 	if (ret < 0)
412 		return libbpf_err(ret);
413 
414 	/* If BPF verifier doesn't recognize BPF helper ID (enum bpf_func_id)
415 	 * at all, it will emit something like "invalid func unknown#181".
416 	 * If BPF verifier recognizes BPF helper but it's not supported for
417 	 * given BPF program type, it will emit "unknown func bpf_sys_bpf#166".
418 	 * In both cases, provided combination of BPF program type and BPF
419 	 * helper is not supported by the kernel.
420 	 * In all other cases, probe_prog_load() above will either succeed (e.g.,
421 	 * because BPF helper happens to accept no input arguments or it
422 	 * accepts one input argument and initial PTR_TO_CTX is fine for
423 	 * that), or we'll get some more specific BPF verifier error about
424 	 * some unsatisfied conditions.
425 	 */
426 	if (ret == 0 && (strstr(buf, "invalid func ") || strstr(buf, "unknown func ")))
427 		return 0;
428 	return 1; /* assume supported */
429 }
430 
431 bool bpf_probe_helper(enum bpf_func_id id, enum bpf_prog_type prog_type,
432 		      __u32 ifindex)
433 {
434 	struct bpf_insn insns[2] = {
435 		BPF_EMIT_CALL(id),
436 		BPF_EXIT_INSN()
437 	};
438 	char buf[4096] = {};
439 	bool res;
440 
441 	probe_prog_load(prog_type, insns, ARRAY_SIZE(insns), buf, sizeof(buf), ifindex);
442 	res = !grep(buf, "invalid func ") && !grep(buf, "unknown func ");
443 
444 	if (ifindex) {
445 		switch (get_vendor_id(ifindex)) {
446 		case 0x19ee: /* Netronome specific */
447 			res = res && !grep(buf, "not supported by FW") &&
448 				!grep(buf, "unsupported function id");
449 			break;
450 		default:
451 			break;
452 		}
453 	}
454 
455 	return res;
456 }
457 
458 /*
459  * Probe for availability of kernel commit (5.3):
460  *
461  * c04c0d2b968a ("bpf: increase complexity limit and maximum program size")
462  */
463 bool bpf_probe_large_insn_limit(__u32 ifindex)
464 {
465 	struct bpf_insn insns[BPF_MAXINSNS + 1];
466 	int i;
467 
468 	for (i = 0; i < BPF_MAXINSNS; i++)
469 		insns[i] = BPF_MOV64_IMM(BPF_REG_0, 1);
470 	insns[BPF_MAXINSNS] = BPF_EXIT_INSN();
471 
472 	errno = 0;
473 	probe_prog_load(BPF_PROG_TYPE_SCHED_CLS, insns, ARRAY_SIZE(insns), NULL, 0,
474 			ifindex);
475 
476 	return errno != E2BIG && errno != EINVAL;
477 }
478