xref: /openbmc/linux/tools/bpf/bpftool/feature.c (revision 22a41e9a5044bf3519f05b4a00e99af34bfeb40c)
1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (c) 2019 Netronome Systems, Inc. */
3 
4 #include <ctype.h>
5 #include <errno.h>
6 #include <fcntl.h>
7 #include <string.h>
8 #include <unistd.h>
9 #include <net/if.h>
10 #ifdef USE_LIBCAP
11 #include <sys/capability.h>
12 #endif
13 #include <sys/utsname.h>
14 #include <sys/vfs.h>
15 
16 #include <linux/filter.h>
17 #include <linux/limits.h>
18 
19 #include <bpf/bpf.h>
20 #include <bpf/libbpf.h>
21 #include <zlib.h>
22 
23 #include "main.h"
24 
25 #ifndef PROC_SUPER_MAGIC
26 # define PROC_SUPER_MAGIC	0x9fa0
27 #endif
28 
29 enum probe_component {
30 	COMPONENT_UNSPEC,
31 	COMPONENT_KERNEL,
32 	COMPONENT_DEVICE,
33 };
34 
35 #define BPF_HELPER_MAKE_ENTRY(name)	[BPF_FUNC_ ## name] = "bpf_" # name
36 static const char * const helper_name[] = {
37 	__BPF_FUNC_MAPPER(BPF_HELPER_MAKE_ENTRY)
38 };
39 
40 #undef BPF_HELPER_MAKE_ENTRY
41 
42 static bool full_mode;
43 #ifdef USE_LIBCAP
44 static bool run_as_unprivileged;
45 #endif
46 
47 /* Miscellaneous utility functions */
48 
49 static bool grep(const char *buffer, const char *pattern)
50 {
51 	return !!strstr(buffer, pattern);
52 }
53 
54 static bool check_procfs(void)
55 {
56 	struct statfs st_fs;
57 
58 	if (statfs("/proc", &st_fs) < 0)
59 		return false;
60 	if ((unsigned long)st_fs.f_type != PROC_SUPER_MAGIC)
61 		return false;
62 
63 	return true;
64 }
65 
66 static void uppercase(char *str, size_t len)
67 {
68 	size_t i;
69 
70 	for (i = 0; i < len && str[i] != '\0'; i++)
71 		str[i] = toupper(str[i]);
72 }
73 
74 /* Printing utility functions */
75 
76 static void
77 print_bool_feature(const char *feat_name, const char *plain_name,
78 		   const char *define_name, bool res, const char *define_prefix)
79 {
80 	if (json_output)
81 		jsonw_bool_field(json_wtr, feat_name, res);
82 	else if (define_prefix)
83 		printf("#define %s%sHAVE_%s\n", define_prefix,
84 		       res ? "" : "NO_", define_name);
85 	else
86 		printf("%s is %savailable\n", plain_name, res ? "" : "NOT ");
87 }
88 
89 static void print_kernel_option(const char *name, const char *value,
90 				const char *define_prefix)
91 {
92 	char *endptr;
93 	int res;
94 
95 	if (json_output) {
96 		if (!value) {
97 			jsonw_null_field(json_wtr, name);
98 			return;
99 		}
100 		errno = 0;
101 		res = strtol(value, &endptr, 0);
102 		if (!errno && *endptr == '\n')
103 			jsonw_int_field(json_wtr, name, res);
104 		else
105 			jsonw_string_field(json_wtr, name, value);
106 	} else if (define_prefix) {
107 		if (value)
108 			printf("#define %s%s %s\n", define_prefix,
109 			       name, value);
110 		else
111 			printf("/* %s%s is not set */\n", define_prefix, name);
112 	} else {
113 		if (value)
114 			printf("%s is set to %s\n", name, value);
115 		else
116 			printf("%s is not set\n", name);
117 	}
118 }
119 
120 static void
121 print_start_section(const char *json_title, const char *plain_title,
122 		    const char *define_comment, const char *define_prefix)
123 {
124 	if (json_output) {
125 		jsonw_name(json_wtr, json_title);
126 		jsonw_start_object(json_wtr);
127 	} else if (define_prefix) {
128 		printf("%s\n", define_comment);
129 	} else {
130 		printf("%s\n", plain_title);
131 	}
132 }
133 
134 static void print_end_section(void)
135 {
136 	if (json_output)
137 		jsonw_end_object(json_wtr);
138 	else
139 		printf("\n");
140 }
141 
142 /* Probing functions */
143 
144 static int get_vendor_id(int ifindex)
145 {
146 	char ifname[IF_NAMESIZE], path[64], buf[8];
147 	ssize_t len;
148 	int fd;
149 
150 	if (!if_indextoname(ifindex, ifname))
151 		return -1;
152 
153 	snprintf(path, sizeof(path), "/sys/class/net/%s/device/vendor", ifname);
154 
155 	fd = open(path, O_RDONLY | O_CLOEXEC);
156 	if (fd < 0)
157 		return -1;
158 
159 	len = read(fd, buf, sizeof(buf));
160 	close(fd);
161 	if (len < 0)
162 		return -1;
163 	if (len >= (ssize_t)sizeof(buf))
164 		return -1;
165 	buf[len] = '\0';
166 
167 	return strtol(buf, NULL, 0);
168 }
169 
170 static int read_procfs(const char *path)
171 {
172 	char *endptr, *line = NULL;
173 	size_t len = 0;
174 	FILE *fd;
175 	int res;
176 
177 	fd = fopen(path, "r");
178 	if (!fd)
179 		return -1;
180 
181 	res = getline(&line, &len, fd);
182 	fclose(fd);
183 	if (res < 0)
184 		return -1;
185 
186 	errno = 0;
187 	res = strtol(line, &endptr, 10);
188 	if (errno || *line == '\0' || *endptr != '\n')
189 		res = -1;
190 	free(line);
191 
192 	return res;
193 }
194 
195 static void probe_unprivileged_disabled(void)
196 {
197 	int res;
198 
199 	/* No support for C-style ouptut */
200 
201 	res = read_procfs("/proc/sys/kernel/unprivileged_bpf_disabled");
202 	if (json_output) {
203 		jsonw_int_field(json_wtr, "unprivileged_bpf_disabled", res);
204 	} else {
205 		switch (res) {
206 		case 0:
207 			printf("bpf() syscall for unprivileged users is enabled\n");
208 			break;
209 		case 1:
210 			printf("bpf() syscall restricted to privileged users\n");
211 			break;
212 		case -1:
213 			printf("Unable to retrieve required privileges for bpf() syscall\n");
214 			break;
215 		default:
216 			printf("bpf() syscall restriction has unknown value %d\n", res);
217 		}
218 	}
219 }
220 
221 static void probe_jit_enable(void)
222 {
223 	int res;
224 
225 	/* No support for C-style ouptut */
226 
227 	res = read_procfs("/proc/sys/net/core/bpf_jit_enable");
228 	if (json_output) {
229 		jsonw_int_field(json_wtr, "bpf_jit_enable", res);
230 	} else {
231 		switch (res) {
232 		case 0:
233 			printf("JIT compiler is disabled\n");
234 			break;
235 		case 1:
236 			printf("JIT compiler is enabled\n");
237 			break;
238 		case 2:
239 			printf("JIT compiler is enabled with debugging traces in kernel logs\n");
240 			break;
241 		case -1:
242 			printf("Unable to retrieve JIT-compiler status\n");
243 			break;
244 		default:
245 			printf("JIT-compiler status has unknown value %d\n",
246 			       res);
247 		}
248 	}
249 }
250 
251 static void probe_jit_harden(void)
252 {
253 	int res;
254 
255 	/* No support for C-style ouptut */
256 
257 	res = read_procfs("/proc/sys/net/core/bpf_jit_harden");
258 	if (json_output) {
259 		jsonw_int_field(json_wtr, "bpf_jit_harden", res);
260 	} else {
261 		switch (res) {
262 		case 0:
263 			printf("JIT compiler hardening is disabled\n");
264 			break;
265 		case 1:
266 			printf("JIT compiler hardening is enabled for unprivileged users\n");
267 			break;
268 		case 2:
269 			printf("JIT compiler hardening is enabled for all users\n");
270 			break;
271 		case -1:
272 			printf("Unable to retrieve JIT hardening status\n");
273 			break;
274 		default:
275 			printf("JIT hardening status has unknown value %d\n",
276 			       res);
277 		}
278 	}
279 }
280 
281 static void probe_jit_kallsyms(void)
282 {
283 	int res;
284 
285 	/* No support for C-style ouptut */
286 
287 	res = read_procfs("/proc/sys/net/core/bpf_jit_kallsyms");
288 	if (json_output) {
289 		jsonw_int_field(json_wtr, "bpf_jit_kallsyms", res);
290 	} else {
291 		switch (res) {
292 		case 0:
293 			printf("JIT compiler kallsyms exports are disabled\n");
294 			break;
295 		case 1:
296 			printf("JIT compiler kallsyms exports are enabled for root\n");
297 			break;
298 		case -1:
299 			printf("Unable to retrieve JIT kallsyms export status\n");
300 			break;
301 		default:
302 			printf("JIT kallsyms exports status has unknown value %d\n", res);
303 		}
304 	}
305 }
306 
307 static void probe_jit_limit(void)
308 {
309 	int res;
310 
311 	/* No support for C-style ouptut */
312 
313 	res = read_procfs("/proc/sys/net/core/bpf_jit_limit");
314 	if (json_output) {
315 		jsonw_int_field(json_wtr, "bpf_jit_limit", res);
316 	} else {
317 		switch (res) {
318 		case -1:
319 			printf("Unable to retrieve global memory limit for JIT compiler for unprivileged users\n");
320 			break;
321 		default:
322 			printf("Global memory limit for JIT compiler for unprivileged users is %d bytes\n", res);
323 		}
324 	}
325 }
326 
327 static bool read_next_kernel_config_option(gzFile file, char *buf, size_t n,
328 					   char **value)
329 {
330 	char *sep;
331 
332 	while (gzgets(file, buf, n)) {
333 		if (strncmp(buf, "CONFIG_", 7))
334 			continue;
335 
336 		sep = strchr(buf, '=');
337 		if (!sep)
338 			continue;
339 
340 		/* Trim ending '\n' */
341 		buf[strlen(buf) - 1] = '\0';
342 
343 		/* Split on '=' and ensure that a value is present. */
344 		*sep = '\0';
345 		if (!sep[1])
346 			continue;
347 
348 		*value = sep + 1;
349 		return true;
350 	}
351 
352 	return false;
353 }
354 
355 static void probe_kernel_image_config(const char *define_prefix)
356 {
357 	static const struct {
358 		const char * const name;
359 		bool macro_dump;
360 	} options[] = {
361 		/* Enable BPF */
362 		{ "CONFIG_BPF", },
363 		/* Enable bpf() syscall */
364 		{ "CONFIG_BPF_SYSCALL", },
365 		/* Does selected architecture support eBPF JIT compiler */
366 		{ "CONFIG_HAVE_EBPF_JIT", },
367 		/* Compile eBPF JIT compiler */
368 		{ "CONFIG_BPF_JIT", },
369 		/* Avoid compiling eBPF interpreter (use JIT only) */
370 		{ "CONFIG_BPF_JIT_ALWAYS_ON", },
371 		/* Kernel BTF debug information available */
372 		{ "CONFIG_DEBUG_INFO_BTF", },
373 		/* Kernel module BTF debug information available */
374 		{ "CONFIG_DEBUG_INFO_BTF_MODULES", },
375 
376 		/* cgroups */
377 		{ "CONFIG_CGROUPS", },
378 		/* BPF programs attached to cgroups */
379 		{ "CONFIG_CGROUP_BPF", },
380 		/* bpf_get_cgroup_classid() helper */
381 		{ "CONFIG_CGROUP_NET_CLASSID", },
382 		/* bpf_skb_{,ancestor_}cgroup_id() helpers */
383 		{ "CONFIG_SOCK_CGROUP_DATA", },
384 
385 		/* Tracing: attach BPF to kprobes, tracepoints, etc. */
386 		{ "CONFIG_BPF_EVENTS", },
387 		/* Kprobes */
388 		{ "CONFIG_KPROBE_EVENTS", },
389 		/* Uprobes */
390 		{ "CONFIG_UPROBE_EVENTS", },
391 		/* Tracepoints */
392 		{ "CONFIG_TRACING", },
393 		/* Syscall tracepoints */
394 		{ "CONFIG_FTRACE_SYSCALLS", },
395 		/* bpf_override_return() helper support for selected arch */
396 		{ "CONFIG_FUNCTION_ERROR_INJECTION", },
397 		/* bpf_override_return() helper */
398 		{ "CONFIG_BPF_KPROBE_OVERRIDE", },
399 
400 		/* Network */
401 		{ "CONFIG_NET", },
402 		/* AF_XDP sockets */
403 		{ "CONFIG_XDP_SOCKETS", },
404 		/* BPF_PROG_TYPE_LWT_* and related helpers */
405 		{ "CONFIG_LWTUNNEL_BPF", },
406 		/* BPF_PROG_TYPE_SCHED_ACT, TC (traffic control) actions */
407 		{ "CONFIG_NET_ACT_BPF", },
408 		/* BPF_PROG_TYPE_SCHED_CLS, TC filters */
409 		{ "CONFIG_NET_CLS_BPF", },
410 		/* TC clsact qdisc */
411 		{ "CONFIG_NET_CLS_ACT", },
412 		/* Ingress filtering with TC */
413 		{ "CONFIG_NET_SCH_INGRESS", },
414 		/* bpf_skb_get_xfrm_state() helper */
415 		{ "CONFIG_XFRM", },
416 		/* bpf_get_route_realm() helper */
417 		{ "CONFIG_IP_ROUTE_CLASSID", },
418 		/* BPF_PROG_TYPE_LWT_SEG6_LOCAL and related helpers */
419 		{ "CONFIG_IPV6_SEG6_BPF", },
420 		/* BPF_PROG_TYPE_LIRC_MODE2 and related helpers */
421 		{ "CONFIG_BPF_LIRC_MODE2", },
422 		/* BPF stream parser and BPF socket maps */
423 		{ "CONFIG_BPF_STREAM_PARSER", },
424 		/* xt_bpf module for passing BPF programs to netfilter  */
425 		{ "CONFIG_NETFILTER_XT_MATCH_BPF", },
426 		/* bpfilter back-end for iptables */
427 		{ "CONFIG_BPFILTER", },
428 		/* bpftilter module with "user mode helper" */
429 		{ "CONFIG_BPFILTER_UMH", },
430 
431 		/* test_bpf module for BPF tests */
432 		{ "CONFIG_TEST_BPF", },
433 
434 		/* Misc configs useful in BPF C programs */
435 		/* jiffies <-> sec conversion for bpf_jiffies64() helper */
436 		{ "CONFIG_HZ", true, }
437 	};
438 	char *values[ARRAY_SIZE(options)] = { };
439 	struct utsname utsn;
440 	char path[PATH_MAX];
441 	gzFile file = NULL;
442 	char buf[4096];
443 	char *value;
444 	size_t i;
445 
446 	if (!uname(&utsn)) {
447 		snprintf(path, sizeof(path), "/boot/config-%s", utsn.release);
448 
449 		/* gzopen also accepts uncompressed files. */
450 		file = gzopen(path, "r");
451 	}
452 
453 	if (!file) {
454 		/* Some distributions build with CONFIG_IKCONFIG=y and put the
455 		 * config file at /proc/config.gz.
456 		 */
457 		file = gzopen("/proc/config.gz", "r");
458 	}
459 	if (!file) {
460 		p_info("skipping kernel config, can't open file: %s",
461 		       strerror(errno));
462 		goto end_parse;
463 	}
464 	/* Sanity checks */
465 	if (!gzgets(file, buf, sizeof(buf)) ||
466 	    !gzgets(file, buf, sizeof(buf))) {
467 		p_info("skipping kernel config, can't read from file: %s",
468 		       strerror(errno));
469 		goto end_parse;
470 	}
471 	if (strcmp(buf, "# Automatically generated file; DO NOT EDIT.\n")) {
472 		p_info("skipping kernel config, can't find correct file");
473 		goto end_parse;
474 	}
475 
476 	while (read_next_kernel_config_option(file, buf, sizeof(buf), &value)) {
477 		for (i = 0; i < ARRAY_SIZE(options); i++) {
478 			if ((define_prefix && !options[i].macro_dump) ||
479 			    values[i] || strcmp(buf, options[i].name))
480 				continue;
481 
482 			values[i] = strdup(value);
483 		}
484 	}
485 
486 end_parse:
487 	if (file)
488 		gzclose(file);
489 
490 	for (i = 0; i < ARRAY_SIZE(options); i++) {
491 		if (define_prefix && !options[i].macro_dump)
492 			continue;
493 		print_kernel_option(options[i].name, values[i], define_prefix);
494 		free(values[i]);
495 	}
496 }
497 
498 static bool probe_bpf_syscall(const char *define_prefix)
499 {
500 	bool res;
501 
502 	bpf_prog_load(BPF_PROG_TYPE_UNSPEC, NULL, NULL, NULL, 0, NULL);
503 	res = (errno != ENOSYS);
504 
505 	print_bool_feature("have_bpf_syscall",
506 			   "bpf() syscall",
507 			   "BPF_SYSCALL",
508 			   res, define_prefix);
509 
510 	return res;
511 }
512 
513 static bool
514 probe_prog_load_ifindex(enum bpf_prog_type prog_type,
515 			const struct bpf_insn *insns, size_t insns_cnt,
516 			char *log_buf, size_t log_buf_sz,
517 			__u32 ifindex)
518 {
519 	LIBBPF_OPTS(bpf_prog_load_opts, opts,
520 		    .log_buf = log_buf,
521 		    .log_size = log_buf_sz,
522 		    .log_level = log_buf ? 1 : 0,
523 		    .prog_ifindex = ifindex,
524 		   );
525 	int fd;
526 
527 	errno = 0;
528 	fd = bpf_prog_load(prog_type, NULL, "GPL", insns, insns_cnt, &opts);
529 	if (fd >= 0)
530 		close(fd);
531 
532 	return fd >= 0 && errno != EINVAL && errno != EOPNOTSUPP;
533 }
534 
535 static bool probe_prog_type_ifindex(enum bpf_prog_type prog_type, __u32 ifindex)
536 {
537 	/* nfp returns -EINVAL on exit(0) with TC offload */
538 	struct bpf_insn insns[2] = {
539 		BPF_MOV64_IMM(BPF_REG_0, 2),
540 		BPF_EXIT_INSN()
541 	};
542 
543 	return probe_prog_load_ifindex(prog_type, insns, ARRAY_SIZE(insns),
544 				       NULL, 0, ifindex);
545 }
546 
547 static void
548 probe_prog_type(enum bpf_prog_type prog_type, bool *supported_types,
549 		const char *define_prefix, __u32 ifindex)
550 {
551 	char feat_name[128], plain_desc[128], define_name[128];
552 	const char *plain_comment = "eBPF program_type ";
553 	size_t maxlen;
554 	bool res;
555 
556 	if (ifindex) {
557 		switch (prog_type) {
558 		case BPF_PROG_TYPE_SCHED_CLS:
559 		case BPF_PROG_TYPE_XDP:
560 			break;
561 		default:
562 			return;
563 		}
564 
565 		res = probe_prog_type_ifindex(prog_type, ifindex);
566 	} else {
567 		res = libbpf_probe_bpf_prog_type(prog_type, NULL);
568 	}
569 
570 #ifdef USE_LIBCAP
571 	/* Probe may succeed even if program load fails, for unprivileged users
572 	 * check that we did not fail because of insufficient permissions
573 	 */
574 	if (run_as_unprivileged && errno == EPERM)
575 		res = false;
576 #endif
577 
578 	supported_types[prog_type] |= res;
579 
580 	if (!prog_type_name[prog_type]) {
581 		p_info("program type name not found (type %d)", prog_type);
582 		return;
583 	}
584 	maxlen = sizeof(plain_desc) - strlen(plain_comment) - 1;
585 	if (strlen(prog_type_name[prog_type]) > maxlen) {
586 		p_info("program type name too long");
587 		return;
588 	}
589 
590 	sprintf(feat_name, "have_%s_prog_type", prog_type_name[prog_type]);
591 	sprintf(define_name, "%s_prog_type", prog_type_name[prog_type]);
592 	uppercase(define_name, sizeof(define_name));
593 	sprintf(plain_desc, "%s%s", plain_comment, prog_type_name[prog_type]);
594 	print_bool_feature(feat_name, plain_desc, define_name, res,
595 			   define_prefix);
596 }
597 
598 static bool probe_map_type_ifindex(enum bpf_map_type map_type, __u32 ifindex)
599 {
600 	LIBBPF_OPTS(bpf_map_create_opts, opts);
601 	int key_size, value_size, max_entries;
602 	int fd;
603 
604 	opts.map_ifindex = ifindex;
605 
606 	key_size = sizeof(__u32);
607 	value_size = sizeof(__u32);
608 	max_entries = 1;
609 
610 	fd = bpf_map_create(map_type, NULL, key_size, value_size, max_entries,
611 			    &opts);
612 	if (fd >= 0)
613 		close(fd);
614 
615 	return fd >= 0;
616 }
617 
618 static void
619 probe_map_type(enum bpf_map_type map_type, const char *define_prefix,
620 	       __u32 ifindex)
621 {
622 	char feat_name[128], plain_desc[128], define_name[128];
623 	const char *plain_comment = "eBPF map_type ";
624 	size_t maxlen;
625 	bool res;
626 
627 	if (ifindex) {
628 		switch (map_type) {
629 		case BPF_MAP_TYPE_HASH:
630 		case BPF_MAP_TYPE_ARRAY:
631 			break;
632 		default:
633 			return;
634 		}
635 
636 		res = probe_map_type_ifindex(map_type, ifindex);
637 	} else {
638 		res = libbpf_probe_bpf_map_type(map_type, NULL);
639 	}
640 
641 	/* Probe result depends on the success of map creation, no additional
642 	 * check required for unprivileged users
643 	 */
644 
645 	if (!map_type_name[map_type]) {
646 		p_info("map type name not found (type %d)", map_type);
647 		return;
648 	}
649 	maxlen = sizeof(plain_desc) - strlen(plain_comment) - 1;
650 	if (strlen(map_type_name[map_type]) > maxlen) {
651 		p_info("map type name too long");
652 		return;
653 	}
654 
655 	sprintf(feat_name, "have_%s_map_type", map_type_name[map_type]);
656 	sprintf(define_name, "%s_map_type", map_type_name[map_type]);
657 	uppercase(define_name, sizeof(define_name));
658 	sprintf(plain_desc, "%s%s", plain_comment, map_type_name[map_type]);
659 	print_bool_feature(feat_name, plain_desc, define_name, res,
660 			   define_prefix);
661 }
662 
663 static bool
664 probe_helper_ifindex(enum bpf_func_id id, enum bpf_prog_type prog_type,
665 		     __u32 ifindex)
666 {
667 	struct bpf_insn insns[2] = {
668 		BPF_EMIT_CALL(id),
669 		BPF_EXIT_INSN()
670 	};
671 	char buf[4096] = {};
672 	bool res;
673 
674 	probe_prog_load_ifindex(prog_type, insns, ARRAY_SIZE(insns), buf,
675 				sizeof(buf), ifindex);
676 	res = !grep(buf, "invalid func ") && !grep(buf, "unknown func ");
677 
678 	switch (get_vendor_id(ifindex)) {
679 	case 0x19ee: /* Netronome specific */
680 		res = res && !grep(buf, "not supported by FW") &&
681 			!grep(buf, "unsupported function id");
682 		break;
683 	default:
684 		break;
685 	}
686 
687 	return res;
688 }
689 
690 static void
691 probe_helper_for_progtype(enum bpf_prog_type prog_type, bool supported_type,
692 			  const char *define_prefix, unsigned int id,
693 			  const char *ptype_name, __u32 ifindex)
694 {
695 	bool res = false;
696 
697 	if (supported_type) {
698 		if (ifindex)
699 			res = probe_helper_ifindex(id, prog_type, ifindex);
700 		else
701 			res = libbpf_probe_bpf_helper(prog_type, id, NULL);
702 #ifdef USE_LIBCAP
703 		/* Probe may succeed even if program load fails, for
704 		 * unprivileged users check that we did not fail because of
705 		 * insufficient permissions
706 		 */
707 		if (run_as_unprivileged && errno == EPERM)
708 			res = false;
709 #endif
710 	}
711 
712 	if (json_output) {
713 		if (res)
714 			jsonw_string(json_wtr, helper_name[id]);
715 	} else if (define_prefix) {
716 		printf("#define %sBPF__PROG_TYPE_%s__HELPER_%s %s\n",
717 		       define_prefix, ptype_name, helper_name[id],
718 		       res ? "1" : "0");
719 	} else {
720 		if (res)
721 			printf("\n\t- %s", helper_name[id]);
722 	}
723 }
724 
725 static void
726 probe_helpers_for_progtype(enum bpf_prog_type prog_type, bool supported_type,
727 			   const char *define_prefix, __u32 ifindex)
728 {
729 	const char *ptype_name = prog_type_name[prog_type];
730 	char feat_name[128];
731 	unsigned int id;
732 
733 	if (ifindex)
734 		/* Only test helpers for offload-able program types */
735 		switch (prog_type) {
736 		case BPF_PROG_TYPE_SCHED_CLS:
737 		case BPF_PROG_TYPE_XDP:
738 			break;
739 		default:
740 			return;
741 		}
742 
743 	if (json_output) {
744 		sprintf(feat_name, "%s_available_helpers", ptype_name);
745 		jsonw_name(json_wtr, feat_name);
746 		jsonw_start_array(json_wtr);
747 	} else if (!define_prefix) {
748 		printf("eBPF helpers supported for program type %s:",
749 		       ptype_name);
750 	}
751 
752 	for (id = 1; id < ARRAY_SIZE(helper_name); id++) {
753 		/* Skip helper functions which emit dmesg messages when not in
754 		 * the full mode.
755 		 */
756 		switch (id) {
757 		case BPF_FUNC_trace_printk:
758 		case BPF_FUNC_trace_vprintk:
759 		case BPF_FUNC_probe_write_user:
760 			if (!full_mode)
761 				continue;
762 			/* fallthrough */
763 		default:
764 			probe_helper_for_progtype(prog_type, supported_type,
765 						  define_prefix, id, ptype_name,
766 						  ifindex);
767 		}
768 	}
769 
770 	if (json_output)
771 		jsonw_end_array(json_wtr);
772 	else if (!define_prefix)
773 		printf("\n");
774 }
775 
776 static void
777 probe_misc_feature(struct bpf_insn *insns, size_t len,
778 		   const char *define_prefix, __u32 ifindex,
779 		   const char *feat_name, const char *plain_name,
780 		   const char *define_name)
781 {
782 	LIBBPF_OPTS(bpf_prog_load_opts, opts,
783 		.prog_ifindex = ifindex,
784 	);
785 	bool res;
786 	int fd;
787 
788 	errno = 0;
789 	fd = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL",
790 			   insns, len, &opts);
791 	res = fd >= 0 || !errno;
792 
793 	if (fd >= 0)
794 		close(fd);
795 
796 	print_bool_feature(feat_name, plain_name, define_name, res,
797 			   define_prefix);
798 }
799 
800 /*
801  * Probe for availability of kernel commit (5.3):
802  *
803  * c04c0d2b968a ("bpf: increase complexity limit and maximum program size")
804  */
805 static void probe_large_insn_limit(const char *define_prefix, __u32 ifindex)
806 {
807 	struct bpf_insn insns[BPF_MAXINSNS + 1];
808 	int i;
809 
810 	for (i = 0; i < BPF_MAXINSNS; i++)
811 		insns[i] = BPF_MOV64_IMM(BPF_REG_0, 1);
812 	insns[BPF_MAXINSNS] = BPF_EXIT_INSN();
813 
814 	probe_misc_feature(insns, ARRAY_SIZE(insns),
815 			   define_prefix, ifindex,
816 			   "have_large_insn_limit",
817 			   "Large program size limit",
818 			   "LARGE_INSN_LIMIT");
819 }
820 
821 /*
822  * Probe for bounded loop support introduced in commit 2589726d12a1
823  * ("bpf: introduce bounded loops").
824  */
825 static void
826 probe_bounded_loops(const char *define_prefix, __u32 ifindex)
827 {
828 	struct bpf_insn insns[4] = {
829 		BPF_MOV64_IMM(BPF_REG_0, 10),
830 		BPF_ALU64_IMM(BPF_SUB, BPF_REG_0, 1),
831 		BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, -2),
832 		BPF_EXIT_INSN()
833 	};
834 
835 	probe_misc_feature(insns, ARRAY_SIZE(insns),
836 			   define_prefix, ifindex,
837 			   "have_bounded_loops",
838 			   "Bounded loop support",
839 			   "BOUNDED_LOOPS");
840 }
841 
842 /*
843  * Probe for the v2 instruction set extension introduced in commit 92b31a9af73b
844  * ("bpf: add BPF_J{LT,LE,SLT,SLE} instructions").
845  */
846 static void
847 probe_v2_isa_extension(const char *define_prefix, __u32 ifindex)
848 {
849 	struct bpf_insn insns[4] = {
850 		BPF_MOV64_IMM(BPF_REG_0, 0),
851 		BPF_JMP_IMM(BPF_JLT, BPF_REG_0, 0, 1),
852 		BPF_MOV64_IMM(BPF_REG_0, 1),
853 		BPF_EXIT_INSN()
854 	};
855 
856 	probe_misc_feature(insns, ARRAY_SIZE(insns),
857 			   define_prefix, ifindex,
858 			   "have_v2_isa_extension",
859 			   "ISA extension v2",
860 			   "V2_ISA_EXTENSION");
861 }
862 
863 /*
864  * Probe for the v3 instruction set extension introduced in commit 092ed0968bb6
865  * ("bpf: verifier support JMP32").
866  */
867 static void
868 probe_v3_isa_extension(const char *define_prefix, __u32 ifindex)
869 {
870 	struct bpf_insn insns[4] = {
871 		BPF_MOV64_IMM(BPF_REG_0, 0),
872 		BPF_JMP32_IMM(BPF_JLT, BPF_REG_0, 0, 1),
873 		BPF_MOV64_IMM(BPF_REG_0, 1),
874 		BPF_EXIT_INSN()
875 	};
876 
877 	probe_misc_feature(insns, ARRAY_SIZE(insns),
878 			   define_prefix, ifindex,
879 			   "have_v3_isa_extension",
880 			   "ISA extension v3",
881 			   "V3_ISA_EXTENSION");
882 }
883 
884 static void
885 section_system_config(enum probe_component target, const char *define_prefix)
886 {
887 	switch (target) {
888 	case COMPONENT_KERNEL:
889 	case COMPONENT_UNSPEC:
890 		print_start_section("system_config",
891 				    "Scanning system configuration...",
892 				    "/*** Misc kernel config items ***/",
893 				    define_prefix);
894 		if (!define_prefix) {
895 			if (check_procfs()) {
896 				probe_unprivileged_disabled();
897 				probe_jit_enable();
898 				probe_jit_harden();
899 				probe_jit_kallsyms();
900 				probe_jit_limit();
901 			} else {
902 				p_info("/* procfs not mounted, skipping related probes */");
903 			}
904 		}
905 		probe_kernel_image_config(define_prefix);
906 		print_end_section();
907 		break;
908 	default:
909 		break;
910 	}
911 }
912 
913 static bool section_syscall_config(const char *define_prefix)
914 {
915 	bool res;
916 
917 	print_start_section("syscall_config",
918 			    "Scanning system call availability...",
919 			    "/*** System call availability ***/",
920 			    define_prefix);
921 	res = probe_bpf_syscall(define_prefix);
922 	print_end_section();
923 
924 	return res;
925 }
926 
927 static void
928 section_program_types(bool *supported_types, const char *define_prefix,
929 		      __u32 ifindex)
930 {
931 	unsigned int i;
932 
933 	print_start_section("program_types",
934 			    "Scanning eBPF program types...",
935 			    "/*** eBPF program types ***/",
936 			    define_prefix);
937 
938 	for (i = BPF_PROG_TYPE_UNSPEC + 1; i < prog_type_name_size; i++)
939 		probe_prog_type(i, supported_types, define_prefix, ifindex);
940 
941 	print_end_section();
942 }
943 
944 static void section_map_types(const char *define_prefix, __u32 ifindex)
945 {
946 	unsigned int i;
947 
948 	print_start_section("map_types",
949 			    "Scanning eBPF map types...",
950 			    "/*** eBPF map types ***/",
951 			    define_prefix);
952 
953 	for (i = BPF_MAP_TYPE_UNSPEC + 1; i < map_type_name_size; i++)
954 		probe_map_type(i, define_prefix, ifindex);
955 
956 	print_end_section();
957 }
958 
959 static void
960 section_helpers(bool *supported_types, const char *define_prefix, __u32 ifindex)
961 {
962 	unsigned int i;
963 
964 	print_start_section("helpers",
965 			    "Scanning eBPF helper functions...",
966 			    "/*** eBPF helper functions ***/",
967 			    define_prefix);
968 
969 	if (define_prefix)
970 		printf("/*\n"
971 		       " * Use %sHAVE_PROG_TYPE_HELPER(prog_type_name, helper_name)\n"
972 		       " * to determine if <helper_name> is available for <prog_type_name>,\n"
973 		       " * e.g.\n"
974 		       " *	#if %sHAVE_PROG_TYPE_HELPER(xdp, bpf_redirect)\n"
975 		       " *		// do stuff with this helper\n"
976 		       " *	#elif\n"
977 		       " *		// use a workaround\n"
978 		       " *	#endif\n"
979 		       " */\n"
980 		       "#define %sHAVE_PROG_TYPE_HELPER(prog_type, helper)	\\\n"
981 		       "	%sBPF__PROG_TYPE_ ## prog_type ## __HELPER_ ## helper\n",
982 		       define_prefix, define_prefix, define_prefix,
983 		       define_prefix);
984 	for (i = BPF_PROG_TYPE_UNSPEC + 1; i < prog_type_name_size; i++)
985 		probe_helpers_for_progtype(i, supported_types[i], define_prefix,
986 					   ifindex);
987 
988 	print_end_section();
989 }
990 
991 static void section_misc(const char *define_prefix, __u32 ifindex)
992 {
993 	print_start_section("misc",
994 			    "Scanning miscellaneous eBPF features...",
995 			    "/*** eBPF misc features ***/",
996 			    define_prefix);
997 	probe_large_insn_limit(define_prefix, ifindex);
998 	probe_bounded_loops(define_prefix, ifindex);
999 	probe_v2_isa_extension(define_prefix, ifindex);
1000 	probe_v3_isa_extension(define_prefix, ifindex);
1001 	print_end_section();
1002 }
1003 
1004 #ifdef USE_LIBCAP
1005 #define capability(c) { c, false, #c }
1006 #define capability_msg(a, i) a[i].set ? "" : a[i].name, a[i].set ? "" : ", "
1007 #endif
1008 
1009 static int handle_perms(void)
1010 {
1011 #ifdef USE_LIBCAP
1012 	struct {
1013 		cap_value_t cap;
1014 		bool set;
1015 		char name[14];	/* strlen("CAP_SYS_ADMIN") */
1016 	} bpf_caps[] = {
1017 		capability(CAP_SYS_ADMIN),
1018 #ifdef CAP_BPF
1019 		capability(CAP_BPF),
1020 		capability(CAP_NET_ADMIN),
1021 		capability(CAP_PERFMON),
1022 #endif
1023 	};
1024 	cap_value_t cap_list[ARRAY_SIZE(bpf_caps)];
1025 	unsigned int i, nb_bpf_caps = 0;
1026 	bool cap_sys_admin_only = true;
1027 	cap_flag_value_t val;
1028 	int res = -1;
1029 	cap_t caps;
1030 
1031 	caps = cap_get_proc();
1032 	if (!caps) {
1033 		p_err("failed to get capabilities for process: %s",
1034 		      strerror(errno));
1035 		return -1;
1036 	}
1037 
1038 #ifdef CAP_BPF
1039 	if (CAP_IS_SUPPORTED(CAP_BPF))
1040 		cap_sys_admin_only = false;
1041 #endif
1042 
1043 	for (i = 0; i < ARRAY_SIZE(bpf_caps); i++) {
1044 		const char *cap_name = bpf_caps[i].name;
1045 		cap_value_t cap = bpf_caps[i].cap;
1046 
1047 		if (cap_get_flag(caps, cap, CAP_EFFECTIVE, &val)) {
1048 			p_err("bug: failed to retrieve %s status: %s", cap_name,
1049 			      strerror(errno));
1050 			goto exit_free;
1051 		}
1052 
1053 		if (val == CAP_SET) {
1054 			bpf_caps[i].set = true;
1055 			cap_list[nb_bpf_caps++] = cap;
1056 		}
1057 
1058 		if (cap_sys_admin_only)
1059 			/* System does not know about CAP_BPF, meaning that
1060 			 * CAP_SYS_ADMIN is the only capability required. We
1061 			 * just checked it, break.
1062 			 */
1063 			break;
1064 	}
1065 
1066 	if ((run_as_unprivileged && !nb_bpf_caps) ||
1067 	    (!run_as_unprivileged && nb_bpf_caps == ARRAY_SIZE(bpf_caps)) ||
1068 	    (!run_as_unprivileged && cap_sys_admin_only && nb_bpf_caps)) {
1069 		/* We are all good, exit now */
1070 		res = 0;
1071 		goto exit_free;
1072 	}
1073 
1074 	if (!run_as_unprivileged) {
1075 		if (cap_sys_admin_only)
1076 			p_err("missing %s, required for full feature probing; run as root or use 'unprivileged'",
1077 			      bpf_caps[0].name);
1078 		else
1079 			p_err("missing %s%s%s%s%s%s%s%srequired for full feature probing; run as root or use 'unprivileged'",
1080 			      capability_msg(bpf_caps, 0),
1081 #ifdef CAP_BPF
1082 			      capability_msg(bpf_caps, 1),
1083 			      capability_msg(bpf_caps, 2),
1084 			      capability_msg(bpf_caps, 3)
1085 #else
1086 				"", "", "", "", "", ""
1087 #endif /* CAP_BPF */
1088 				);
1089 		goto exit_free;
1090 	}
1091 
1092 	/* if (run_as_unprivileged && nb_bpf_caps > 0), drop capabilities. */
1093 	if (cap_set_flag(caps, CAP_EFFECTIVE, nb_bpf_caps, cap_list,
1094 			 CAP_CLEAR)) {
1095 		p_err("bug: failed to clear capabilities: %s", strerror(errno));
1096 		goto exit_free;
1097 	}
1098 
1099 	if (cap_set_proc(caps)) {
1100 		p_err("failed to drop capabilities: %s", strerror(errno));
1101 		goto exit_free;
1102 	}
1103 
1104 	res = 0;
1105 
1106 exit_free:
1107 	if (cap_free(caps) && !res) {
1108 		p_err("failed to clear storage object for capabilities: %s",
1109 		      strerror(errno));
1110 		res = -1;
1111 	}
1112 
1113 	return res;
1114 #else
1115 	/* Detection assumes user has specific privileges.
1116 	 * We do not use libpcap so let's approximate, and restrict usage to
1117 	 * root user only.
1118 	 */
1119 	if (geteuid()) {
1120 		p_err("full feature probing requires root privileges");
1121 		return -1;
1122 	}
1123 
1124 	return 0;
1125 #endif /* USE_LIBCAP */
1126 }
1127 
1128 static int do_probe(int argc, char **argv)
1129 {
1130 	enum probe_component target = COMPONENT_UNSPEC;
1131 	const char *define_prefix = NULL;
1132 	bool supported_types[128] = {};
1133 	__u32 ifindex = 0;
1134 	char *ifname;
1135 
1136 	set_max_rlimit();
1137 
1138 	while (argc) {
1139 		if (is_prefix(*argv, "kernel")) {
1140 			if (target != COMPONENT_UNSPEC) {
1141 				p_err("component to probe already specified");
1142 				return -1;
1143 			}
1144 			target = COMPONENT_KERNEL;
1145 			NEXT_ARG();
1146 		} else if (is_prefix(*argv, "dev")) {
1147 			NEXT_ARG();
1148 
1149 			if (target != COMPONENT_UNSPEC || ifindex) {
1150 				p_err("component to probe already specified");
1151 				return -1;
1152 			}
1153 			if (!REQ_ARGS(1))
1154 				return -1;
1155 
1156 			target = COMPONENT_DEVICE;
1157 			ifname = GET_ARG();
1158 			ifindex = if_nametoindex(ifname);
1159 			if (!ifindex) {
1160 				p_err("unrecognized netdevice '%s': %s", ifname,
1161 				      strerror(errno));
1162 				return -1;
1163 			}
1164 		} else if (is_prefix(*argv, "full")) {
1165 			full_mode = true;
1166 			NEXT_ARG();
1167 		} else if (is_prefix(*argv, "macros") && !define_prefix) {
1168 			define_prefix = "";
1169 			NEXT_ARG();
1170 		} else if (is_prefix(*argv, "prefix")) {
1171 			if (!define_prefix) {
1172 				p_err("'prefix' argument can only be use after 'macros'");
1173 				return -1;
1174 			}
1175 			if (strcmp(define_prefix, "")) {
1176 				p_err("'prefix' already defined");
1177 				return -1;
1178 			}
1179 			NEXT_ARG();
1180 
1181 			if (!REQ_ARGS(1))
1182 				return -1;
1183 			define_prefix = GET_ARG();
1184 		} else if (is_prefix(*argv, "unprivileged")) {
1185 #ifdef USE_LIBCAP
1186 			run_as_unprivileged = true;
1187 			NEXT_ARG();
1188 #else
1189 			p_err("unprivileged run not supported, recompile bpftool with libcap");
1190 			return -1;
1191 #endif
1192 		} else {
1193 			p_err("expected no more arguments, 'kernel', 'dev', 'macros' or 'prefix', got: '%s'?",
1194 			      *argv);
1195 			return -1;
1196 		}
1197 	}
1198 
1199 	/* Full feature detection requires specific privileges.
1200 	 * Let's approximate, and warn if user is not root.
1201 	 */
1202 	if (handle_perms())
1203 		return -1;
1204 
1205 	if (json_output) {
1206 		define_prefix = NULL;
1207 		jsonw_start_object(json_wtr);
1208 	}
1209 
1210 	section_system_config(target, define_prefix);
1211 	if (!section_syscall_config(define_prefix))
1212 		/* bpf() syscall unavailable, don't probe other BPF features */
1213 		goto exit_close_json;
1214 	section_program_types(supported_types, define_prefix, ifindex);
1215 	section_map_types(define_prefix, ifindex);
1216 	section_helpers(supported_types, define_prefix, ifindex);
1217 	section_misc(define_prefix, ifindex);
1218 
1219 exit_close_json:
1220 	if (json_output)
1221 		/* End root object */
1222 		jsonw_end_object(json_wtr);
1223 
1224 	return 0;
1225 }
1226 
1227 static int do_help(int argc, char **argv)
1228 {
1229 	if (json_output) {
1230 		jsonw_null(json_wtr);
1231 		return 0;
1232 	}
1233 
1234 	fprintf(stderr,
1235 		"Usage: %1$s %2$s probe [COMPONENT] [full] [unprivileged] [macros [prefix PREFIX]]\n"
1236 		"       %1$s %2$s help\n"
1237 		"\n"
1238 		"       COMPONENT := { kernel | dev NAME }\n"
1239 		"       " HELP_SPEC_OPTIONS " }\n"
1240 		"",
1241 		bin_name, argv[-2]);
1242 
1243 	return 0;
1244 }
1245 
1246 static const struct cmd cmds[] = {
1247 	{ "probe",	do_probe },
1248 	{ "help",	do_help },
1249 	{ 0 }
1250 };
1251 
1252 int do_feature(int argc, char **argv)
1253 {
1254 	return cmd_select(cmds, argc, argv, do_help);
1255 }
1256