xref: /linux/tools/bpf/bpftool/feature.c (revision 9406b485dea5e25bed7c81cd822747d494cc8bde)
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 <string.h>
7 #include <unistd.h>
8 #include <net/if.h>
9 #ifdef USE_LIBCAP
10 #include <sys/capability.h>
11 #endif
12 #include <sys/utsname.h>
13 #include <sys/vfs.h>
14 
15 #include <linux/filter.h>
16 #include <linux/limits.h>
17 
18 #include <bpf/bpf.h>
19 #include <bpf/libbpf.h>
20 #include <zlib.h>
21 
22 #include "main.h"
23 
24 #ifndef PROC_SUPER_MAGIC
25 # define PROC_SUPER_MAGIC	0x9fa0
26 #endif
27 
28 enum probe_component {
29 	COMPONENT_UNSPEC,
30 	COMPONENT_KERNEL,
31 	COMPONENT_DEVICE,
32 };
33 
34 #define BPF_HELPER_MAKE_ENTRY(name)	[BPF_FUNC_ ## name] = "bpf_" # name
35 static const char * const helper_name[] = {
36 	__BPF_FUNC_MAPPER(BPF_HELPER_MAKE_ENTRY)
37 };
38 
39 #undef BPF_HELPER_MAKE_ENTRY
40 
41 static bool full_mode;
42 #ifdef USE_LIBCAP
43 static bool run_as_unprivileged;
44 #endif
45 
46 /* Miscellaneous utility functions */
47 
48 static bool check_procfs(void)
49 {
50 	struct statfs st_fs;
51 
52 	if (statfs("/proc", &st_fs) < 0)
53 		return false;
54 	if ((unsigned long)st_fs.f_type != PROC_SUPER_MAGIC)
55 		return false;
56 
57 	return true;
58 }
59 
60 static void uppercase(char *str, size_t len)
61 {
62 	size_t i;
63 
64 	for (i = 0; i < len && str[i] != '\0'; i++)
65 		str[i] = toupper(str[i]);
66 }
67 
68 /* Printing utility functions */
69 
70 static void
71 print_bool_feature(const char *feat_name, const char *plain_name,
72 		   const char *define_name, bool res, const char *define_prefix)
73 {
74 	if (json_output)
75 		jsonw_bool_field(json_wtr, feat_name, res);
76 	else if (define_prefix)
77 		printf("#define %s%sHAVE_%s\n", define_prefix,
78 		       res ? "" : "NO_", define_name);
79 	else
80 		printf("%s is %savailable\n", plain_name, res ? "" : "NOT ");
81 }
82 
83 static void print_kernel_option(const char *name, const char *value)
84 {
85 	char *endptr;
86 	int res;
87 
88 	/* No support for C-style ouptut */
89 
90 	if (json_output) {
91 		if (!value) {
92 			jsonw_null_field(json_wtr, name);
93 			return;
94 		}
95 		errno = 0;
96 		res = strtol(value, &endptr, 0);
97 		if (!errno && *endptr == '\n')
98 			jsonw_int_field(json_wtr, name, res);
99 		else
100 			jsonw_string_field(json_wtr, name, value);
101 	} else {
102 		if (value)
103 			printf("%s is set to %s\n", name, value);
104 		else
105 			printf("%s is not set\n", name);
106 	}
107 }
108 
109 static void
110 print_start_section(const char *json_title, const char *plain_title,
111 		    const char *define_comment, const char *define_prefix)
112 {
113 	if (json_output) {
114 		jsonw_name(json_wtr, json_title);
115 		jsonw_start_object(json_wtr);
116 	} else if (define_prefix) {
117 		printf("%s\n", define_comment);
118 	} else {
119 		printf("%s\n", plain_title);
120 	}
121 }
122 
123 static void print_end_section(void)
124 {
125 	if (json_output)
126 		jsonw_end_object(json_wtr);
127 	else
128 		printf("\n");
129 }
130 
131 /* Probing functions */
132 
133 static int read_procfs(const char *path)
134 {
135 	char *endptr, *line = NULL;
136 	size_t len = 0;
137 	FILE *fd;
138 	int res;
139 
140 	fd = fopen(path, "r");
141 	if (!fd)
142 		return -1;
143 
144 	res = getline(&line, &len, fd);
145 	fclose(fd);
146 	if (res < 0)
147 		return -1;
148 
149 	errno = 0;
150 	res = strtol(line, &endptr, 10);
151 	if (errno || *line == '\0' || *endptr != '\n')
152 		res = -1;
153 	free(line);
154 
155 	return res;
156 }
157 
158 static void probe_unprivileged_disabled(void)
159 {
160 	int res;
161 
162 	/* No support for C-style ouptut */
163 
164 	res = read_procfs("/proc/sys/kernel/unprivileged_bpf_disabled");
165 	if (json_output) {
166 		jsonw_int_field(json_wtr, "unprivileged_bpf_disabled", res);
167 	} else {
168 		switch (res) {
169 		case 0:
170 			printf("bpf() syscall for unprivileged users is enabled\n");
171 			break;
172 		case 1:
173 			printf("bpf() syscall restricted to privileged users\n");
174 			break;
175 		case -1:
176 			printf("Unable to retrieve required privileges for bpf() syscall\n");
177 			break;
178 		default:
179 			printf("bpf() syscall restriction has unknown value %d\n", res);
180 		}
181 	}
182 }
183 
184 static void probe_jit_enable(void)
185 {
186 	int res;
187 
188 	/* No support for C-style ouptut */
189 
190 	res = read_procfs("/proc/sys/net/core/bpf_jit_enable");
191 	if (json_output) {
192 		jsonw_int_field(json_wtr, "bpf_jit_enable", res);
193 	} else {
194 		switch (res) {
195 		case 0:
196 			printf("JIT compiler is disabled\n");
197 			break;
198 		case 1:
199 			printf("JIT compiler is enabled\n");
200 			break;
201 		case 2:
202 			printf("JIT compiler is enabled with debugging traces in kernel logs\n");
203 			break;
204 		case -1:
205 			printf("Unable to retrieve JIT-compiler status\n");
206 			break;
207 		default:
208 			printf("JIT-compiler status has unknown value %d\n",
209 			       res);
210 		}
211 	}
212 }
213 
214 static void probe_jit_harden(void)
215 {
216 	int res;
217 
218 	/* No support for C-style ouptut */
219 
220 	res = read_procfs("/proc/sys/net/core/bpf_jit_harden");
221 	if (json_output) {
222 		jsonw_int_field(json_wtr, "bpf_jit_harden", res);
223 	} else {
224 		switch (res) {
225 		case 0:
226 			printf("JIT compiler hardening is disabled\n");
227 			break;
228 		case 1:
229 			printf("JIT compiler hardening is enabled for unprivileged users\n");
230 			break;
231 		case 2:
232 			printf("JIT compiler hardening is enabled for all users\n");
233 			break;
234 		case -1:
235 			printf("Unable to retrieve JIT hardening status\n");
236 			break;
237 		default:
238 			printf("JIT hardening status has unknown value %d\n",
239 			       res);
240 		}
241 	}
242 }
243 
244 static void probe_jit_kallsyms(void)
245 {
246 	int res;
247 
248 	/* No support for C-style ouptut */
249 
250 	res = read_procfs("/proc/sys/net/core/bpf_jit_kallsyms");
251 	if (json_output) {
252 		jsonw_int_field(json_wtr, "bpf_jit_kallsyms", res);
253 	} else {
254 		switch (res) {
255 		case 0:
256 			printf("JIT compiler kallsyms exports are disabled\n");
257 			break;
258 		case 1:
259 			printf("JIT compiler kallsyms exports are enabled for root\n");
260 			break;
261 		case -1:
262 			printf("Unable to retrieve JIT kallsyms export status\n");
263 			break;
264 		default:
265 			printf("JIT kallsyms exports status has unknown value %d\n", res);
266 		}
267 	}
268 }
269 
270 static void probe_jit_limit(void)
271 {
272 	int res;
273 
274 	/* No support for C-style ouptut */
275 
276 	res = read_procfs("/proc/sys/net/core/bpf_jit_limit");
277 	if (json_output) {
278 		jsonw_int_field(json_wtr, "bpf_jit_limit", res);
279 	} else {
280 		switch (res) {
281 		case -1:
282 			printf("Unable to retrieve global memory limit for JIT compiler for unprivileged users\n");
283 			break;
284 		default:
285 			printf("Global memory limit for JIT compiler for unprivileged users is %d bytes\n", res);
286 		}
287 	}
288 }
289 
290 static bool read_next_kernel_config_option(gzFile file, char *buf, size_t n,
291 					   char **value)
292 {
293 	char *sep;
294 
295 	while (gzgets(file, buf, n)) {
296 		if (strncmp(buf, "CONFIG_", 7))
297 			continue;
298 
299 		sep = strchr(buf, '=');
300 		if (!sep)
301 			continue;
302 
303 		/* Trim ending '\n' */
304 		buf[strlen(buf) - 1] = '\0';
305 
306 		/* Split on '=' and ensure that a value is present. */
307 		*sep = '\0';
308 		if (!sep[1])
309 			continue;
310 
311 		*value = sep + 1;
312 		return true;
313 	}
314 
315 	return false;
316 }
317 
318 static void probe_kernel_image_config(void)
319 {
320 	static const char * const options[] = {
321 		/* Enable BPF */
322 		"CONFIG_BPF",
323 		/* Enable bpf() syscall */
324 		"CONFIG_BPF_SYSCALL",
325 		/* Does selected architecture support eBPF JIT compiler */
326 		"CONFIG_HAVE_EBPF_JIT",
327 		/* Compile eBPF JIT compiler */
328 		"CONFIG_BPF_JIT",
329 		/* Avoid compiling eBPF interpreter (use JIT only) */
330 		"CONFIG_BPF_JIT_ALWAYS_ON",
331 
332 		/* cgroups */
333 		"CONFIG_CGROUPS",
334 		/* BPF programs attached to cgroups */
335 		"CONFIG_CGROUP_BPF",
336 		/* bpf_get_cgroup_classid() helper */
337 		"CONFIG_CGROUP_NET_CLASSID",
338 		/* bpf_skb_{,ancestor_}cgroup_id() helpers */
339 		"CONFIG_SOCK_CGROUP_DATA",
340 
341 		/* Tracing: attach BPF to kprobes, tracepoints, etc. */
342 		"CONFIG_BPF_EVENTS",
343 		/* Kprobes */
344 		"CONFIG_KPROBE_EVENTS",
345 		/* Uprobes */
346 		"CONFIG_UPROBE_EVENTS",
347 		/* Tracepoints */
348 		"CONFIG_TRACING",
349 		/* Syscall tracepoints */
350 		"CONFIG_FTRACE_SYSCALLS",
351 		/* bpf_override_return() helper support for selected arch */
352 		"CONFIG_FUNCTION_ERROR_INJECTION",
353 		/* bpf_override_return() helper */
354 		"CONFIG_BPF_KPROBE_OVERRIDE",
355 
356 		/* Network */
357 		"CONFIG_NET",
358 		/* AF_XDP sockets */
359 		"CONFIG_XDP_SOCKETS",
360 		/* BPF_PROG_TYPE_LWT_* and related helpers */
361 		"CONFIG_LWTUNNEL_BPF",
362 		/* BPF_PROG_TYPE_SCHED_ACT, TC (traffic control) actions */
363 		"CONFIG_NET_ACT_BPF",
364 		/* BPF_PROG_TYPE_SCHED_CLS, TC filters */
365 		"CONFIG_NET_CLS_BPF",
366 		/* TC clsact qdisc */
367 		"CONFIG_NET_CLS_ACT",
368 		/* Ingress filtering with TC */
369 		"CONFIG_NET_SCH_INGRESS",
370 		/* bpf_skb_get_xfrm_state() helper */
371 		"CONFIG_XFRM",
372 		/* bpf_get_route_realm() helper */
373 		"CONFIG_IP_ROUTE_CLASSID",
374 		/* BPF_PROG_TYPE_LWT_SEG6_LOCAL and related helpers */
375 		"CONFIG_IPV6_SEG6_BPF",
376 		/* BPF_PROG_TYPE_LIRC_MODE2 and related helpers */
377 		"CONFIG_BPF_LIRC_MODE2",
378 		/* BPF stream parser and BPF socket maps */
379 		"CONFIG_BPF_STREAM_PARSER",
380 		/* xt_bpf module for passing BPF programs to netfilter  */
381 		"CONFIG_NETFILTER_XT_MATCH_BPF",
382 		/* bpfilter back-end for iptables */
383 		"CONFIG_BPFILTER",
384 		/* bpftilter module with "user mode helper" */
385 		"CONFIG_BPFILTER_UMH",
386 
387 		/* test_bpf module for BPF tests */
388 		"CONFIG_TEST_BPF",
389 	};
390 	char *values[ARRAY_SIZE(options)] = { };
391 	struct utsname utsn;
392 	char path[PATH_MAX];
393 	gzFile file = NULL;
394 	char buf[4096];
395 	char *value;
396 	size_t i;
397 
398 	if (!uname(&utsn)) {
399 		snprintf(path, sizeof(path), "/boot/config-%s", utsn.release);
400 
401 		/* gzopen also accepts uncompressed files. */
402 		file = gzopen(path, "r");
403 	}
404 
405 	if (!file) {
406 		/* Some distributions build with CONFIG_IKCONFIG=y and put the
407 		 * config file at /proc/config.gz.
408 		 */
409 		file = gzopen("/proc/config.gz", "r");
410 	}
411 	if (!file) {
412 		p_info("skipping kernel config, can't open file: %s",
413 		       strerror(errno));
414 		goto end_parse;
415 	}
416 	/* Sanity checks */
417 	if (!gzgets(file, buf, sizeof(buf)) ||
418 	    !gzgets(file, buf, sizeof(buf))) {
419 		p_info("skipping kernel config, can't read from file: %s",
420 		       strerror(errno));
421 		goto end_parse;
422 	}
423 	if (strcmp(buf, "# Automatically generated file; DO NOT EDIT.\n")) {
424 		p_info("skipping kernel config, can't find correct file");
425 		goto end_parse;
426 	}
427 
428 	while (read_next_kernel_config_option(file, buf, sizeof(buf), &value)) {
429 		for (i = 0; i < ARRAY_SIZE(options); i++) {
430 			if (values[i] || strcmp(buf, options[i]))
431 				continue;
432 
433 			values[i] = strdup(value);
434 		}
435 	}
436 
437 end_parse:
438 	if (file)
439 		gzclose(file);
440 
441 	for (i = 0; i < ARRAY_SIZE(options); i++) {
442 		print_kernel_option(options[i], values[i]);
443 		free(values[i]);
444 	}
445 }
446 
447 static bool probe_bpf_syscall(const char *define_prefix)
448 {
449 	bool res;
450 
451 	bpf_load_program(BPF_PROG_TYPE_UNSPEC, NULL, 0, NULL, 0, NULL, 0);
452 	res = (errno != ENOSYS);
453 
454 	print_bool_feature("have_bpf_syscall",
455 			   "bpf() syscall",
456 			   "BPF_SYSCALL",
457 			   res, define_prefix);
458 
459 	return res;
460 }
461 
462 static void
463 probe_prog_type(enum bpf_prog_type prog_type, bool *supported_types,
464 		const char *define_prefix, __u32 ifindex)
465 {
466 	char feat_name[128], plain_desc[128], define_name[128];
467 	const char *plain_comment = "eBPF program_type ";
468 	size_t maxlen;
469 	bool res;
470 
471 	if (ifindex)
472 		/* Only test offload-able program types */
473 		switch (prog_type) {
474 		case BPF_PROG_TYPE_SCHED_CLS:
475 		case BPF_PROG_TYPE_XDP:
476 			break;
477 		default:
478 			return;
479 		}
480 
481 	res = bpf_probe_prog_type(prog_type, ifindex);
482 #ifdef USE_LIBCAP
483 	/* Probe may succeed even if program load fails, for unprivileged users
484 	 * check that we did not fail because of insufficient permissions
485 	 */
486 	if (run_as_unprivileged && errno == EPERM)
487 		res = false;
488 #endif
489 
490 	supported_types[prog_type] |= res;
491 
492 	maxlen = sizeof(plain_desc) - strlen(plain_comment) - 1;
493 	if (strlen(prog_type_name[prog_type]) > maxlen) {
494 		p_info("program type name too long");
495 		return;
496 	}
497 
498 	sprintf(feat_name, "have_%s_prog_type", prog_type_name[prog_type]);
499 	sprintf(define_name, "%s_prog_type", prog_type_name[prog_type]);
500 	uppercase(define_name, sizeof(define_name));
501 	sprintf(plain_desc, "%s%s", plain_comment, prog_type_name[prog_type]);
502 	print_bool_feature(feat_name, plain_desc, define_name, res,
503 			   define_prefix);
504 }
505 
506 static void
507 probe_map_type(enum bpf_map_type map_type, const char *define_prefix,
508 	       __u32 ifindex)
509 {
510 	char feat_name[128], plain_desc[128], define_name[128];
511 	const char *plain_comment = "eBPF map_type ";
512 	size_t maxlen;
513 	bool res;
514 
515 	res = bpf_probe_map_type(map_type, ifindex);
516 
517 	/* Probe result depends on the success of map creation, no additional
518 	 * check required for unprivileged users
519 	 */
520 
521 	maxlen = sizeof(plain_desc) - strlen(plain_comment) - 1;
522 	if (strlen(map_type_name[map_type]) > maxlen) {
523 		p_info("map type name too long");
524 		return;
525 	}
526 
527 	sprintf(feat_name, "have_%s_map_type", map_type_name[map_type]);
528 	sprintf(define_name, "%s_map_type", map_type_name[map_type]);
529 	uppercase(define_name, sizeof(define_name));
530 	sprintf(plain_desc, "%s%s", plain_comment, map_type_name[map_type]);
531 	print_bool_feature(feat_name, plain_desc, define_name, res,
532 			   define_prefix);
533 }
534 
535 static void
536 probe_helper_for_progtype(enum bpf_prog_type prog_type, bool supported_type,
537 			  const char *define_prefix, unsigned int id,
538 			  const char *ptype_name, __u32 ifindex)
539 {
540 	bool res = false;
541 
542 	if (supported_type) {
543 		res = bpf_probe_helper(id, prog_type, ifindex);
544 #ifdef USE_LIBCAP
545 		/* Probe may succeed even if program load fails, for
546 		 * unprivileged users check that we did not fail because of
547 		 * insufficient permissions
548 		 */
549 		if (run_as_unprivileged && errno == EPERM)
550 			res = false;
551 #endif
552 	}
553 
554 	if (json_output) {
555 		if (res)
556 			jsonw_string(json_wtr, helper_name[id]);
557 	} else if (define_prefix) {
558 		printf("#define %sBPF__PROG_TYPE_%s__HELPER_%s %s\n",
559 		       define_prefix, ptype_name, helper_name[id],
560 		       res ? "1" : "0");
561 	} else {
562 		if (res)
563 			printf("\n\t- %s", helper_name[id]);
564 	}
565 }
566 
567 static void
568 probe_helpers_for_progtype(enum bpf_prog_type prog_type, bool supported_type,
569 			   const char *define_prefix, __u32 ifindex)
570 {
571 	const char *ptype_name = prog_type_name[prog_type];
572 	char feat_name[128];
573 	unsigned int id;
574 
575 	if (ifindex)
576 		/* Only test helpers for offload-able program types */
577 		switch (prog_type) {
578 		case BPF_PROG_TYPE_SCHED_CLS:
579 		case BPF_PROG_TYPE_XDP:
580 			break;
581 		default:
582 			return;
583 		}
584 
585 	if (json_output) {
586 		sprintf(feat_name, "%s_available_helpers", ptype_name);
587 		jsonw_name(json_wtr, feat_name);
588 		jsonw_start_array(json_wtr);
589 	} else if (!define_prefix) {
590 		printf("eBPF helpers supported for program type %s:",
591 		       ptype_name);
592 	}
593 
594 	for (id = 1; id < ARRAY_SIZE(helper_name); id++) {
595 		/* Skip helper functions which emit dmesg messages when not in
596 		 * the full mode.
597 		 */
598 		switch (id) {
599 		case BPF_FUNC_trace_printk:
600 		case BPF_FUNC_probe_write_user:
601 			if (!full_mode)
602 				continue;
603 			/* fallthrough */
604 		default:
605 			probe_helper_for_progtype(prog_type, supported_type,
606 						  define_prefix, id, ptype_name,
607 						  ifindex);
608 		}
609 	}
610 
611 	if (json_output)
612 		jsonw_end_array(json_wtr);
613 	else if (!define_prefix)
614 		printf("\n");
615 }
616 
617 static void
618 probe_large_insn_limit(const char *define_prefix, __u32 ifindex)
619 {
620 	bool res;
621 
622 	res = bpf_probe_large_insn_limit(ifindex);
623 	print_bool_feature("have_large_insn_limit",
624 			   "Large program size limit",
625 			   "LARGE_INSN_LIMIT",
626 			   res, define_prefix);
627 }
628 
629 static void
630 section_system_config(enum probe_component target, const char *define_prefix)
631 {
632 	switch (target) {
633 	case COMPONENT_KERNEL:
634 	case COMPONENT_UNSPEC:
635 		if (define_prefix)
636 			break;
637 
638 		print_start_section("system_config",
639 				    "Scanning system configuration...",
640 				    NULL, /* define_comment never used here */
641 				    NULL); /* define_prefix always NULL here */
642 		if (check_procfs()) {
643 			probe_unprivileged_disabled();
644 			probe_jit_enable();
645 			probe_jit_harden();
646 			probe_jit_kallsyms();
647 			probe_jit_limit();
648 		} else {
649 			p_info("/* procfs not mounted, skipping related probes */");
650 		}
651 		probe_kernel_image_config();
652 		print_end_section();
653 		break;
654 	default:
655 		break;
656 	}
657 }
658 
659 static bool section_syscall_config(const char *define_prefix)
660 {
661 	bool res;
662 
663 	print_start_section("syscall_config",
664 			    "Scanning system call availability...",
665 			    "/*** System call availability ***/",
666 			    define_prefix);
667 	res = probe_bpf_syscall(define_prefix);
668 	print_end_section();
669 
670 	return res;
671 }
672 
673 static void
674 section_program_types(bool *supported_types, const char *define_prefix,
675 		      __u32 ifindex)
676 {
677 	unsigned int i;
678 
679 	print_start_section("program_types",
680 			    "Scanning eBPF program types...",
681 			    "/*** eBPF program types ***/",
682 			    define_prefix);
683 
684 	for (i = BPF_PROG_TYPE_UNSPEC + 1; i < ARRAY_SIZE(prog_type_name); i++)
685 		probe_prog_type(i, supported_types, define_prefix, ifindex);
686 
687 	print_end_section();
688 }
689 
690 static void section_map_types(const char *define_prefix, __u32 ifindex)
691 {
692 	unsigned int i;
693 
694 	print_start_section("map_types",
695 			    "Scanning eBPF map types...",
696 			    "/*** eBPF map types ***/",
697 			    define_prefix);
698 
699 	for (i = BPF_MAP_TYPE_UNSPEC + 1; i < map_type_name_size; i++)
700 		probe_map_type(i, define_prefix, ifindex);
701 
702 	print_end_section();
703 }
704 
705 static void
706 section_helpers(bool *supported_types, const char *define_prefix, __u32 ifindex)
707 {
708 	unsigned int i;
709 
710 	print_start_section("helpers",
711 			    "Scanning eBPF helper functions...",
712 			    "/*** eBPF helper functions ***/",
713 			    define_prefix);
714 
715 	if (define_prefix)
716 		printf("/*\n"
717 		       " * Use %sHAVE_PROG_TYPE_HELPER(prog_type_name, helper_name)\n"
718 		       " * to determine if <helper_name> is available for <prog_type_name>,\n"
719 		       " * e.g.\n"
720 		       " *	#if %sHAVE_PROG_TYPE_HELPER(xdp, bpf_redirect)\n"
721 		       " *		// do stuff with this helper\n"
722 		       " *	#elif\n"
723 		       " *		// use a workaround\n"
724 		       " *	#endif\n"
725 		       " */\n"
726 		       "#define %sHAVE_PROG_TYPE_HELPER(prog_type, helper)	\\\n"
727 		       "	%sBPF__PROG_TYPE_ ## prog_type ## __HELPER_ ## helper\n",
728 		       define_prefix, define_prefix, define_prefix,
729 		       define_prefix);
730 	for (i = BPF_PROG_TYPE_UNSPEC + 1; i < ARRAY_SIZE(prog_type_name); i++)
731 		probe_helpers_for_progtype(i, supported_types[i], define_prefix,
732 					   ifindex);
733 
734 	print_end_section();
735 }
736 
737 static void section_misc(const char *define_prefix, __u32 ifindex)
738 {
739 	print_start_section("misc",
740 			    "Scanning miscellaneous eBPF features...",
741 			    "/*** eBPF misc features ***/",
742 			    define_prefix);
743 	probe_large_insn_limit(define_prefix, ifindex);
744 	print_end_section();
745 }
746 
747 static int handle_perms(void)
748 {
749 #ifdef USE_LIBCAP
750 	cap_value_t cap_list[1] = { CAP_SYS_ADMIN };
751 	bool has_sys_admin_cap = false;
752 	cap_flag_value_t val;
753 	int res = -1;
754 	cap_t caps;
755 
756 	caps = cap_get_proc();
757 	if (!caps) {
758 		p_err("failed to get capabilities for process: %s",
759 		      strerror(errno));
760 		return -1;
761 	}
762 
763 	if (cap_get_flag(caps, CAP_SYS_ADMIN, CAP_EFFECTIVE, &val)) {
764 		p_err("bug: failed to retrieve CAP_SYS_ADMIN status");
765 		goto exit_free;
766 	}
767 	if (val == CAP_SET)
768 		has_sys_admin_cap = true;
769 
770 	if (!run_as_unprivileged && !has_sys_admin_cap) {
771 		p_err("full feature probing requires CAP_SYS_ADMIN, run as root or use 'unprivileged'");
772 		goto exit_free;
773 	}
774 
775 	if ((run_as_unprivileged && !has_sys_admin_cap) ||
776 	    (!run_as_unprivileged && has_sys_admin_cap)) {
777 		/* We are all good, exit now */
778 		res = 0;
779 		goto exit_free;
780 	}
781 
782 	/* if (run_as_unprivileged && has_sys_admin_cap), drop CAP_SYS_ADMIN */
783 
784 	if (cap_set_flag(caps, CAP_EFFECTIVE, ARRAY_SIZE(cap_list), cap_list,
785 			 CAP_CLEAR)) {
786 		p_err("bug: failed to clear CAP_SYS_ADMIN from capabilities");
787 		goto exit_free;
788 	}
789 
790 	if (cap_set_proc(caps)) {
791 		p_err("failed to drop CAP_SYS_ADMIN: %s", strerror(errno));
792 		goto exit_free;
793 	}
794 
795 	res = 0;
796 
797 exit_free:
798 	if (cap_free(caps) && !res) {
799 		p_err("failed to clear storage object for capabilities: %s",
800 		      strerror(errno));
801 		res = -1;
802 	}
803 
804 	return res;
805 #else
806 	/* Detection assumes user has sufficient privileges (CAP_SYS_ADMIN).
807 	 * We do not use libpcap so let's approximate, and restrict usage to
808 	 * root user only.
809 	 */
810 	if (geteuid()) {
811 		p_err("full feature probing requires root privileges");
812 		return -1;
813 	}
814 
815 	return 0;
816 #endif /* USE_LIBCAP */
817 }
818 
819 static int do_probe(int argc, char **argv)
820 {
821 	enum probe_component target = COMPONENT_UNSPEC;
822 	const char *define_prefix = NULL;
823 	bool supported_types[128] = {};
824 	__u32 ifindex = 0;
825 	char *ifname;
826 
827 	set_max_rlimit();
828 
829 	while (argc) {
830 		if (is_prefix(*argv, "kernel")) {
831 			if (target != COMPONENT_UNSPEC) {
832 				p_err("component to probe already specified");
833 				return -1;
834 			}
835 			target = COMPONENT_KERNEL;
836 			NEXT_ARG();
837 		} else if (is_prefix(*argv, "dev")) {
838 			NEXT_ARG();
839 
840 			if (target != COMPONENT_UNSPEC || ifindex) {
841 				p_err("component to probe already specified");
842 				return -1;
843 			}
844 			if (!REQ_ARGS(1))
845 				return -1;
846 
847 			target = COMPONENT_DEVICE;
848 			ifname = GET_ARG();
849 			ifindex = if_nametoindex(ifname);
850 			if (!ifindex) {
851 				p_err("unrecognized netdevice '%s': %s", ifname,
852 				      strerror(errno));
853 				return -1;
854 			}
855 		} else if (is_prefix(*argv, "full")) {
856 			full_mode = true;
857 			NEXT_ARG();
858 		} else if (is_prefix(*argv, "macros") && !define_prefix) {
859 			define_prefix = "";
860 			NEXT_ARG();
861 		} else if (is_prefix(*argv, "prefix")) {
862 			if (!define_prefix) {
863 				p_err("'prefix' argument can only be use after 'macros'");
864 				return -1;
865 			}
866 			if (strcmp(define_prefix, "")) {
867 				p_err("'prefix' already defined");
868 				return -1;
869 			}
870 			NEXT_ARG();
871 
872 			if (!REQ_ARGS(1))
873 				return -1;
874 			define_prefix = GET_ARG();
875 		} else if (is_prefix(*argv, "unprivileged")) {
876 #ifdef USE_LIBCAP
877 			run_as_unprivileged = true;
878 			NEXT_ARG();
879 #else
880 			p_err("unprivileged run not supported, recompile bpftool with libcap");
881 			return -1;
882 #endif
883 		} else {
884 			p_err("expected no more arguments, 'kernel', 'dev', 'macros' or 'prefix', got: '%s'?",
885 			      *argv);
886 			return -1;
887 		}
888 	}
889 
890 	/* Full feature detection requires CAP_SYS_ADMIN privilege.
891 	 * Let's approximate, and warn if user is not root.
892 	 */
893 	if (handle_perms())
894 		return -1;
895 
896 	if (json_output) {
897 		define_prefix = NULL;
898 		jsonw_start_object(json_wtr);
899 	}
900 
901 	section_system_config(target, define_prefix);
902 	if (!section_syscall_config(define_prefix))
903 		/* bpf() syscall unavailable, don't probe other BPF features */
904 		goto exit_close_json;
905 	section_program_types(supported_types, define_prefix, ifindex);
906 	section_map_types(define_prefix, ifindex);
907 	section_helpers(supported_types, define_prefix, ifindex);
908 	section_misc(define_prefix, ifindex);
909 
910 exit_close_json:
911 	if (json_output)
912 		/* End root object */
913 		jsonw_end_object(json_wtr);
914 
915 	return 0;
916 }
917 
918 static int do_help(int argc, char **argv)
919 {
920 	if (json_output) {
921 		jsonw_null(json_wtr);
922 		return 0;
923 	}
924 
925 	fprintf(stderr,
926 		"Usage: %s %s probe [COMPONENT] [full] [unprivileged] [macros [prefix PREFIX]]\n"
927 		"       %s %s help\n"
928 		"\n"
929 		"       COMPONENT := { kernel | dev NAME }\n"
930 		"",
931 		bin_name, argv[-2], bin_name, argv[-2]);
932 
933 	return 0;
934 }
935 
936 static const struct cmd cmds[] = {
937 	{ "probe",	do_probe },
938 	{ "help",	do_help },
939 	{ 0 }
940 };
941 
942 int do_feature(int argc, char **argv)
943 {
944 	return cmd_select(cmds, argc, argv, do_help);
945 }
946