1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (C) 2017-2018 Netronome Systems, Inc. */
3 
4 #define _GNU_SOURCE
5 #include <errno.h>
6 #include <fcntl.h>
7 #include <signal.h>
8 #include <stdarg.h>
9 #include <stdio.h>
10 #include <stdlib.h>
11 #include <string.h>
12 #include <time.h>
13 #include <unistd.h>
14 #include <net/if.h>
15 #include <sys/ioctl.h>
16 #include <sys/types.h>
17 #include <sys/stat.h>
18 #include <sys/syscall.h>
19 #include <dirent.h>
20 
21 #include <linux/err.h>
22 #include <linux/perf_event.h>
23 #include <linux/sizes.h>
24 
25 #include <bpf/bpf.h>
26 #include <bpf/btf.h>
27 #include <bpf/hashmap.h>
28 #include <bpf/libbpf.h>
29 #include <bpf/libbpf_internal.h>
30 #include <bpf/skel_internal.h>
31 
32 #include "cfg.h"
33 #include "main.h"
34 #include "xlated_dumper.h"
35 
36 #define BPF_METADATA_PREFIX "bpf_metadata_"
37 #define BPF_METADATA_PREFIX_LEN (sizeof(BPF_METADATA_PREFIX) - 1)
38 
39 enum dump_mode {
40 	DUMP_JITED,
41 	DUMP_XLATED,
42 };
43 
44 static const bool attach_types[] = {
45 	[BPF_SK_SKB_STREAM_PARSER] = true,
46 	[BPF_SK_SKB_STREAM_VERDICT] = true,
47 	[BPF_SK_SKB_VERDICT] = true,
48 	[BPF_SK_MSG_VERDICT] = true,
49 	[BPF_FLOW_DISSECTOR] = true,
50 	[__MAX_BPF_ATTACH_TYPE] = false,
51 };
52 
53 /* Textual representations traditionally used by the program and kept around
54  * for the sake of backwards compatibility.
55  */
56 static const char * const attach_type_strings[] = {
57 	[BPF_SK_SKB_STREAM_PARSER] = "stream_parser",
58 	[BPF_SK_SKB_STREAM_VERDICT] = "stream_verdict",
59 	[BPF_SK_SKB_VERDICT] = "skb_verdict",
60 	[BPF_SK_MSG_VERDICT] = "msg_verdict",
61 	[__MAX_BPF_ATTACH_TYPE] = NULL,
62 };
63 
64 static struct hashmap *prog_table;
65 
parse_attach_type(const char * str)66 static enum bpf_attach_type parse_attach_type(const char *str)
67 {
68 	enum bpf_attach_type type;
69 
70 	for (type = 0; type < __MAX_BPF_ATTACH_TYPE; type++) {
71 		if (attach_types[type]) {
72 			const char *attach_type_str;
73 
74 			attach_type_str = libbpf_bpf_attach_type_str(type);
75 			if (!strcmp(str, attach_type_str))
76 				return type;
77 		}
78 
79 		if (attach_type_strings[type] &&
80 		    is_prefix(str, attach_type_strings[type]))
81 			return type;
82 	}
83 
84 	return __MAX_BPF_ATTACH_TYPE;
85 }
86 
prep_prog_info(struct bpf_prog_info * const info,enum dump_mode mode,void ** info_data,size_t * const info_data_sz)87 static int prep_prog_info(struct bpf_prog_info *const info, enum dump_mode mode,
88 			  void **info_data, size_t *const info_data_sz)
89 {
90 	struct bpf_prog_info holder = {};
91 	size_t needed = 0;
92 	void *ptr;
93 
94 	if (mode == DUMP_JITED) {
95 		holder.jited_prog_len = info->jited_prog_len;
96 		needed += info->jited_prog_len;
97 	} else {
98 		holder.xlated_prog_len = info->xlated_prog_len;
99 		needed += info->xlated_prog_len;
100 	}
101 
102 	holder.nr_jited_ksyms = info->nr_jited_ksyms;
103 	needed += info->nr_jited_ksyms * sizeof(__u64);
104 
105 	holder.nr_jited_func_lens = info->nr_jited_func_lens;
106 	needed += info->nr_jited_func_lens * sizeof(__u32);
107 
108 	holder.nr_func_info = info->nr_func_info;
109 	holder.func_info_rec_size = info->func_info_rec_size;
110 	needed += info->nr_func_info * info->func_info_rec_size;
111 
112 	holder.nr_line_info = info->nr_line_info;
113 	holder.line_info_rec_size = info->line_info_rec_size;
114 	needed += info->nr_line_info * info->line_info_rec_size;
115 
116 	holder.nr_jited_line_info = info->nr_jited_line_info;
117 	holder.jited_line_info_rec_size = info->jited_line_info_rec_size;
118 	needed += info->nr_jited_line_info * info->jited_line_info_rec_size;
119 
120 	if (needed > *info_data_sz) {
121 		ptr = realloc(*info_data, needed);
122 		if (!ptr)
123 			return -1;
124 
125 		*info_data = ptr;
126 		*info_data_sz = needed;
127 	}
128 	ptr = *info_data;
129 
130 	if (mode == DUMP_JITED) {
131 		holder.jited_prog_insns = ptr_to_u64(ptr);
132 		ptr += holder.jited_prog_len;
133 	} else {
134 		holder.xlated_prog_insns = ptr_to_u64(ptr);
135 		ptr += holder.xlated_prog_len;
136 	}
137 
138 	holder.jited_ksyms = ptr_to_u64(ptr);
139 	ptr += holder.nr_jited_ksyms * sizeof(__u64);
140 
141 	holder.jited_func_lens = ptr_to_u64(ptr);
142 	ptr += holder.nr_jited_func_lens * sizeof(__u32);
143 
144 	holder.func_info = ptr_to_u64(ptr);
145 	ptr += holder.nr_func_info * holder.func_info_rec_size;
146 
147 	holder.line_info = ptr_to_u64(ptr);
148 	ptr += holder.nr_line_info * holder.line_info_rec_size;
149 
150 	holder.jited_line_info = ptr_to_u64(ptr);
151 	ptr += holder.nr_jited_line_info * holder.jited_line_info_rec_size;
152 
153 	*info = holder;
154 	return 0;
155 }
156 
print_boot_time(__u64 nsecs,char * buf,unsigned int size)157 static void print_boot_time(__u64 nsecs, char *buf, unsigned int size)
158 {
159 	struct timespec real_time_ts, boot_time_ts;
160 	time_t wallclock_secs;
161 	struct tm load_tm;
162 
163 	buf[--size] = '\0';
164 
165 	if (clock_gettime(CLOCK_REALTIME, &real_time_ts) ||
166 	    clock_gettime(CLOCK_BOOTTIME, &boot_time_ts)) {
167 		perror("Can't read clocks");
168 		snprintf(buf, size, "%llu", nsecs / 1000000000);
169 		return;
170 	}
171 
172 	wallclock_secs = (real_time_ts.tv_sec - boot_time_ts.tv_sec) +
173 		(real_time_ts.tv_nsec - boot_time_ts.tv_nsec + nsecs) /
174 		1000000000;
175 
176 
177 	if (!localtime_r(&wallclock_secs, &load_tm)) {
178 		snprintf(buf, size, "%llu", nsecs / 1000000000);
179 		return;
180 	}
181 
182 	if (json_output)
183 		strftime(buf, size, "%s", &load_tm);
184 	else
185 		strftime(buf, size, "%FT%T%z", &load_tm);
186 }
187 
show_prog_maps(int fd,__u32 num_maps)188 static void show_prog_maps(int fd, __u32 num_maps)
189 {
190 	struct bpf_prog_info info = {};
191 	__u32 len = sizeof(info);
192 	__u32 map_ids[num_maps];
193 	unsigned int i;
194 	int err;
195 
196 	info.nr_map_ids = num_maps;
197 	info.map_ids = ptr_to_u64(map_ids);
198 
199 	err = bpf_obj_get_info_by_fd(fd, &info, &len);
200 	if (err || !info.nr_map_ids)
201 		return;
202 
203 	if (json_output) {
204 		jsonw_name(json_wtr, "map_ids");
205 		jsonw_start_array(json_wtr);
206 		for (i = 0; i < info.nr_map_ids; i++)
207 			jsonw_uint(json_wtr, map_ids[i]);
208 		jsonw_end_array(json_wtr);
209 	} else {
210 		printf("  map_ids ");
211 		for (i = 0; i < info.nr_map_ids; i++)
212 			printf("%u%s", map_ids[i],
213 			       i == info.nr_map_ids - 1 ? "" : ",");
214 	}
215 }
216 
find_metadata(int prog_fd,struct bpf_map_info * map_info)217 static void *find_metadata(int prog_fd, struct bpf_map_info *map_info)
218 {
219 	struct bpf_prog_info prog_info;
220 	__u32 prog_info_len;
221 	__u32 map_info_len;
222 	void *value = NULL;
223 	__u32 *map_ids;
224 	int nr_maps;
225 	int key = 0;
226 	int map_fd;
227 	int ret;
228 	__u32 i;
229 
230 	memset(&prog_info, 0, sizeof(prog_info));
231 	prog_info_len = sizeof(prog_info);
232 	ret = bpf_obj_get_info_by_fd(prog_fd, &prog_info, &prog_info_len);
233 	if (ret)
234 		return NULL;
235 
236 	if (!prog_info.nr_map_ids)
237 		return NULL;
238 
239 	map_ids = calloc(prog_info.nr_map_ids, sizeof(__u32));
240 	if (!map_ids)
241 		return NULL;
242 
243 	nr_maps = prog_info.nr_map_ids;
244 	memset(&prog_info, 0, sizeof(prog_info));
245 	prog_info.nr_map_ids = nr_maps;
246 	prog_info.map_ids = ptr_to_u64(map_ids);
247 	prog_info_len = sizeof(prog_info);
248 
249 	ret = bpf_obj_get_info_by_fd(prog_fd, &prog_info, &prog_info_len);
250 	if (ret)
251 		goto free_map_ids;
252 
253 	for (i = 0; i < prog_info.nr_map_ids; i++) {
254 		map_fd = bpf_map_get_fd_by_id(map_ids[i]);
255 		if (map_fd < 0)
256 			goto free_map_ids;
257 
258 		memset(map_info, 0, sizeof(*map_info));
259 		map_info_len = sizeof(*map_info);
260 		ret = bpf_obj_get_info_by_fd(map_fd, map_info, &map_info_len);
261 		if (ret < 0) {
262 			close(map_fd);
263 			goto free_map_ids;
264 		}
265 
266 		if (map_info->type != BPF_MAP_TYPE_ARRAY ||
267 		    map_info->key_size != sizeof(int) ||
268 		    map_info->max_entries != 1 ||
269 		    !map_info->btf_value_type_id ||
270 		    !strstr(map_info->name, ".rodata")) {
271 			close(map_fd);
272 			continue;
273 		}
274 
275 		value = malloc(map_info->value_size);
276 		if (!value) {
277 			close(map_fd);
278 			goto free_map_ids;
279 		}
280 
281 		if (bpf_map_lookup_elem(map_fd, &key, value)) {
282 			close(map_fd);
283 			free(value);
284 			value = NULL;
285 			goto free_map_ids;
286 		}
287 
288 		close(map_fd);
289 		break;
290 	}
291 
292 free_map_ids:
293 	free(map_ids);
294 	return value;
295 }
296 
has_metadata_prefix(const char * s)297 static bool has_metadata_prefix(const char *s)
298 {
299 	return strncmp(s, BPF_METADATA_PREFIX, BPF_METADATA_PREFIX_LEN) == 0;
300 }
301 
show_prog_metadata(int fd,__u32 num_maps)302 static void show_prog_metadata(int fd, __u32 num_maps)
303 {
304 	const struct btf_type *t_datasec, *t_var;
305 	struct bpf_map_info map_info;
306 	struct btf_var_secinfo *vsi;
307 	bool printed_header = false;
308 	unsigned int i, vlen;
309 	void *value = NULL;
310 	const char *name;
311 	struct btf *btf;
312 	int err;
313 
314 	if (!num_maps)
315 		return;
316 
317 	memset(&map_info, 0, sizeof(map_info));
318 	value = find_metadata(fd, &map_info);
319 	if (!value)
320 		return;
321 
322 	btf = btf__load_from_kernel_by_id(map_info.btf_id);
323 	if (libbpf_get_error(btf))
324 		goto out_free;
325 
326 	t_datasec = btf__type_by_id(btf, map_info.btf_value_type_id);
327 	if (!btf_is_datasec(t_datasec))
328 		goto out_free;
329 
330 	vlen = btf_vlen(t_datasec);
331 	vsi = btf_var_secinfos(t_datasec);
332 
333 	/* We don't proceed to check the kinds of the elements of the DATASEC.
334 	 * The verifier enforces them to be BTF_KIND_VAR.
335 	 */
336 
337 	if (json_output) {
338 		struct btf_dumper d = {
339 			.btf = btf,
340 			.jw = json_wtr,
341 			.is_plain_text = false,
342 		};
343 
344 		for (i = 0; i < vlen; i++, vsi++) {
345 			t_var = btf__type_by_id(btf, vsi->type);
346 			name = btf__name_by_offset(btf, t_var->name_off);
347 
348 			if (!has_metadata_prefix(name))
349 				continue;
350 
351 			if (!printed_header) {
352 				jsonw_name(json_wtr, "metadata");
353 				jsonw_start_object(json_wtr);
354 				printed_header = true;
355 			}
356 
357 			jsonw_name(json_wtr, name + BPF_METADATA_PREFIX_LEN);
358 			err = btf_dumper_type(&d, t_var->type, value + vsi->offset);
359 			if (err) {
360 				p_err("btf dump failed: %d", err);
361 				break;
362 			}
363 		}
364 		if (printed_header)
365 			jsonw_end_object(json_wtr);
366 	} else {
367 		json_writer_t *btf_wtr;
368 		struct btf_dumper d = {
369 			.btf = btf,
370 			.is_plain_text = true,
371 		};
372 
373 		for (i = 0; i < vlen; i++, vsi++) {
374 			t_var = btf__type_by_id(btf, vsi->type);
375 			name = btf__name_by_offset(btf, t_var->name_off);
376 
377 			if (!has_metadata_prefix(name))
378 				continue;
379 
380 			if (!printed_header) {
381 				printf("\tmetadata:");
382 
383 				btf_wtr = jsonw_new(stdout);
384 				if (!btf_wtr) {
385 					p_err("jsonw alloc failed");
386 					goto out_free;
387 				}
388 				d.jw = btf_wtr,
389 
390 				printed_header = true;
391 			}
392 
393 			printf("\n\t\t%s = ", name + BPF_METADATA_PREFIX_LEN);
394 
395 			jsonw_reset(btf_wtr);
396 			err = btf_dumper_type(&d, t_var->type, value + vsi->offset);
397 			if (err) {
398 				p_err("btf dump failed: %d", err);
399 				break;
400 			}
401 		}
402 		if (printed_header)
403 			jsonw_destroy(&btf_wtr);
404 	}
405 
406 out_free:
407 	btf__free(btf);
408 	free(value);
409 }
410 
print_prog_header_json(struct bpf_prog_info * info,int fd)411 static void print_prog_header_json(struct bpf_prog_info *info, int fd)
412 {
413 	const char *prog_type_str;
414 	char prog_name[MAX_PROG_FULL_NAME];
415 
416 	jsonw_uint_field(json_wtr, "id", info->id);
417 	prog_type_str = libbpf_bpf_prog_type_str(info->type);
418 
419 	if (prog_type_str)
420 		jsonw_string_field(json_wtr, "type", prog_type_str);
421 	else
422 		jsonw_uint_field(json_wtr, "type", info->type);
423 
424 	if (*info->name) {
425 		get_prog_full_name(info, fd, prog_name, sizeof(prog_name));
426 		jsonw_string_field(json_wtr, "name", prog_name);
427 	}
428 
429 	jsonw_name(json_wtr, "tag");
430 	jsonw_printf(json_wtr, "\"" BPF_TAG_FMT "\"",
431 		     info->tag[0], info->tag[1], info->tag[2], info->tag[3],
432 		     info->tag[4], info->tag[5], info->tag[6], info->tag[7]);
433 
434 	jsonw_bool_field(json_wtr, "gpl_compatible", info->gpl_compatible);
435 	if (info->run_time_ns) {
436 		jsonw_uint_field(json_wtr, "run_time_ns", info->run_time_ns);
437 		jsonw_uint_field(json_wtr, "run_cnt", info->run_cnt);
438 	}
439 	if (info->recursion_misses)
440 		jsonw_uint_field(json_wtr, "recursion_misses", info->recursion_misses);
441 }
442 
print_prog_json(struct bpf_prog_info * info,int fd)443 static void print_prog_json(struct bpf_prog_info *info, int fd)
444 {
445 	char *memlock;
446 
447 	jsonw_start_object(json_wtr);
448 	print_prog_header_json(info, fd);
449 	print_dev_json(info->ifindex, info->netns_dev, info->netns_ino);
450 
451 	if (info->load_time) {
452 		char buf[32];
453 
454 		print_boot_time(info->load_time, buf, sizeof(buf));
455 
456 		/* Piggy back on load_time, since 0 uid is a valid one */
457 		jsonw_name(json_wtr, "loaded_at");
458 		jsonw_printf(json_wtr, "%s", buf);
459 		jsonw_uint_field(json_wtr, "uid", info->created_by_uid);
460 	}
461 
462 	jsonw_uint_field(json_wtr, "bytes_xlated", info->xlated_prog_len);
463 
464 	if (info->jited_prog_len) {
465 		jsonw_bool_field(json_wtr, "jited", true);
466 		jsonw_uint_field(json_wtr, "bytes_jited", info->jited_prog_len);
467 	} else {
468 		jsonw_bool_field(json_wtr, "jited", false);
469 	}
470 
471 	memlock = get_fdinfo(fd, "memlock");
472 	if (memlock)
473 		jsonw_int_field(json_wtr, "bytes_memlock", atoll(memlock));
474 	free(memlock);
475 
476 	if (info->nr_map_ids)
477 		show_prog_maps(fd, info->nr_map_ids);
478 
479 	if (info->btf_id)
480 		jsonw_int_field(json_wtr, "btf_id", info->btf_id);
481 
482 	if (!hashmap__empty(prog_table)) {
483 		struct hashmap_entry *entry;
484 
485 		jsonw_name(json_wtr, "pinned");
486 		jsonw_start_array(json_wtr);
487 		hashmap__for_each_key_entry(prog_table, entry,
488 					    u32_as_hash_field(info->id))
489 			jsonw_string(json_wtr, entry->value);
490 		jsonw_end_array(json_wtr);
491 	}
492 
493 	emit_obj_refs_json(refs_table, info->id, json_wtr);
494 
495 	show_prog_metadata(fd, info->nr_map_ids);
496 
497 	jsonw_end_object(json_wtr);
498 }
499 
print_prog_header_plain(struct bpf_prog_info * info,int fd)500 static void print_prog_header_plain(struct bpf_prog_info *info, int fd)
501 {
502 	const char *prog_type_str;
503 	char prog_name[MAX_PROG_FULL_NAME];
504 
505 	printf("%u: ", info->id);
506 	prog_type_str = libbpf_bpf_prog_type_str(info->type);
507 	if (prog_type_str)
508 		printf("%s  ", prog_type_str);
509 	else
510 		printf("type %u  ", info->type);
511 
512 	if (*info->name) {
513 		get_prog_full_name(info, fd, prog_name, sizeof(prog_name));
514 		printf("name %s  ", prog_name);
515 	}
516 
517 	printf("tag ");
518 	fprint_hex(stdout, info->tag, BPF_TAG_SIZE, "");
519 	print_dev_plain(info->ifindex, info->netns_dev, info->netns_ino);
520 	printf("%s", info->gpl_compatible ? "  gpl" : "");
521 	if (info->run_time_ns)
522 		printf(" run_time_ns %lld run_cnt %lld",
523 		       info->run_time_ns, info->run_cnt);
524 	if (info->recursion_misses)
525 		printf(" recursion_misses %lld", info->recursion_misses);
526 	printf("\n");
527 }
528 
print_prog_plain(struct bpf_prog_info * info,int fd)529 static void print_prog_plain(struct bpf_prog_info *info, int fd)
530 {
531 	char *memlock;
532 
533 	print_prog_header_plain(info, fd);
534 
535 	if (info->load_time) {
536 		char buf[32];
537 
538 		print_boot_time(info->load_time, buf, sizeof(buf));
539 
540 		/* Piggy back on load_time, since 0 uid is a valid one */
541 		printf("\tloaded_at %s  uid %u\n", buf, info->created_by_uid);
542 	}
543 
544 	printf("\txlated %uB", info->xlated_prog_len);
545 
546 	if (info->jited_prog_len)
547 		printf("  jited %uB", info->jited_prog_len);
548 	else
549 		printf("  not jited");
550 
551 	memlock = get_fdinfo(fd, "memlock");
552 	if (memlock)
553 		printf("  memlock %sB", memlock);
554 	free(memlock);
555 
556 	if (info->nr_map_ids)
557 		show_prog_maps(fd, info->nr_map_ids);
558 
559 	if (!hashmap__empty(prog_table)) {
560 		struct hashmap_entry *entry;
561 
562 		hashmap__for_each_key_entry(prog_table, entry,
563 					    u32_as_hash_field(info->id))
564 			printf("\n\tpinned %s", (char *)entry->value);
565 	}
566 
567 	if (info->btf_id)
568 		printf("\n\tbtf_id %d", info->btf_id);
569 
570 	emit_obj_refs_plain(refs_table, info->id, "\n\tpids ");
571 
572 	printf("\n");
573 
574 	show_prog_metadata(fd, info->nr_map_ids);
575 }
576 
show_prog(int fd)577 static int show_prog(int fd)
578 {
579 	struct bpf_prog_info info = {};
580 	__u32 len = sizeof(info);
581 	int err;
582 
583 	err = bpf_obj_get_info_by_fd(fd, &info, &len);
584 	if (err) {
585 		p_err("can't get prog info: %s", strerror(errno));
586 		return -1;
587 	}
588 
589 	if (json_output)
590 		print_prog_json(&info, fd);
591 	else
592 		print_prog_plain(&info, fd);
593 
594 	return 0;
595 }
596 
do_show_subset(int argc,char ** argv)597 static int do_show_subset(int argc, char **argv)
598 {
599 	int *fds = NULL;
600 	int nb_fds, i;
601 	int err = -1;
602 
603 	fds = malloc(sizeof(int));
604 	if (!fds) {
605 		p_err("mem alloc failed");
606 		return -1;
607 	}
608 	nb_fds = prog_parse_fds(&argc, &argv, &fds);
609 	if (nb_fds < 1)
610 		goto exit_free;
611 
612 	if (json_output && nb_fds > 1)
613 		jsonw_start_array(json_wtr);	/* root array */
614 	for (i = 0; i < nb_fds; i++) {
615 		err = show_prog(fds[i]);
616 		if (err) {
617 			for (; i < nb_fds; i++)
618 				close(fds[i]);
619 			break;
620 		}
621 		close(fds[i]);
622 	}
623 	if (json_output && nb_fds > 1)
624 		jsonw_end_array(json_wtr);	/* root array */
625 
626 exit_free:
627 	free(fds);
628 	return err;
629 }
630 
do_show(int argc,char ** argv)631 static int do_show(int argc, char **argv)
632 {
633 	__u32 id = 0;
634 	int err;
635 	int fd;
636 
637 	if (show_pinned) {
638 		prog_table = hashmap__new(hash_fn_for_key_as_id,
639 					  equal_fn_for_key_as_id, NULL);
640 		if (IS_ERR(prog_table)) {
641 			p_err("failed to create hashmap for pinned paths");
642 			return -1;
643 		}
644 		build_pinned_obj_table(prog_table, BPF_OBJ_PROG);
645 	}
646 	build_obj_refs_table(&refs_table, BPF_OBJ_PROG);
647 
648 	if (argc == 2)
649 		return do_show_subset(argc, argv);
650 
651 	if (argc)
652 		return BAD_ARG();
653 
654 	if (json_output)
655 		jsonw_start_array(json_wtr);
656 	while (true) {
657 		err = bpf_prog_get_next_id(id, &id);
658 		if (err) {
659 			if (errno == ENOENT) {
660 				err = 0;
661 				break;
662 			}
663 			p_err("can't get next program: %s%s", strerror(errno),
664 			      errno == EINVAL ? " -- kernel too old?" : "");
665 			err = -1;
666 			break;
667 		}
668 
669 		fd = bpf_prog_get_fd_by_id(id);
670 		if (fd < 0) {
671 			if (errno == ENOENT)
672 				continue;
673 			p_err("can't get prog by id (%u): %s",
674 			      id, strerror(errno));
675 			err = -1;
676 			break;
677 		}
678 
679 		err = show_prog(fd);
680 		close(fd);
681 		if (err)
682 			break;
683 	}
684 
685 	if (json_output)
686 		jsonw_end_array(json_wtr);
687 
688 	delete_obj_refs_table(refs_table);
689 
690 	if (show_pinned)
691 		delete_pinned_obj_table(prog_table);
692 
693 	return err;
694 }
695 
696 static int
prog_dump(struct bpf_prog_info * info,enum dump_mode mode,char * filepath,bool opcodes,bool visual,bool linum)697 prog_dump(struct bpf_prog_info *info, enum dump_mode mode,
698 	  char *filepath, bool opcodes, bool visual, bool linum)
699 {
700 	struct bpf_prog_linfo *prog_linfo = NULL;
701 	const char *disasm_opt = NULL;
702 	struct dump_data dd = {};
703 	void *func_info = NULL;
704 	struct btf *btf = NULL;
705 	char func_sig[1024];
706 	unsigned char *buf;
707 	__u32 member_len;
708 	int fd, err = -1;
709 	ssize_t n;
710 
711 	if (mode == DUMP_JITED) {
712 		if (info->jited_prog_len == 0 || !info->jited_prog_insns) {
713 			p_info("no instructions returned");
714 			return -1;
715 		}
716 		buf = u64_to_ptr(info->jited_prog_insns);
717 		member_len = info->jited_prog_len;
718 	} else {	/* DUMP_XLATED */
719 		if (info->xlated_prog_len == 0 || !info->xlated_prog_insns) {
720 			p_err("error retrieving insn dump: kernel.kptr_restrict set?");
721 			return -1;
722 		}
723 		buf = u64_to_ptr(info->xlated_prog_insns);
724 		member_len = info->xlated_prog_len;
725 	}
726 
727 	if (info->btf_id) {
728 		btf = btf__load_from_kernel_by_id(info->btf_id);
729 		if (libbpf_get_error(btf)) {
730 			p_err("failed to get btf");
731 			return -1;
732 		}
733 	}
734 
735 	func_info = u64_to_ptr(info->func_info);
736 
737 	if (info->nr_line_info) {
738 		prog_linfo = bpf_prog_linfo__new(info);
739 		if (!prog_linfo)
740 			p_info("error in processing bpf_line_info.  continue without it.");
741 	}
742 
743 	if (filepath) {
744 		fd = open(filepath, O_WRONLY | O_CREAT | O_TRUNC, 0600);
745 		if (fd < 0) {
746 			p_err("can't open file %s: %s", filepath,
747 			      strerror(errno));
748 			goto exit_free;
749 		}
750 
751 		n = write(fd, buf, member_len);
752 		close(fd);
753 		if (n != (ssize_t)member_len) {
754 			p_err("error writing output file: %s",
755 			      n < 0 ? strerror(errno) : "short write");
756 			goto exit_free;
757 		}
758 
759 		if (json_output)
760 			jsonw_null(json_wtr);
761 	} else if (mode == DUMP_JITED) {
762 		const char *name = NULL;
763 
764 		if (info->ifindex) {
765 			name = ifindex_to_bfd_params(info->ifindex,
766 						     info->netns_dev,
767 						     info->netns_ino,
768 						     &disasm_opt);
769 			if (!name)
770 				goto exit_free;
771 		}
772 
773 		if (info->nr_jited_func_lens && info->jited_func_lens) {
774 			struct kernel_sym *sym = NULL;
775 			struct bpf_func_info *record;
776 			char sym_name[SYM_MAX_NAME];
777 			unsigned char *img = buf;
778 			__u64 *ksyms = NULL;
779 			__u32 *lens;
780 			__u32 i;
781 			if (info->nr_jited_ksyms) {
782 				kernel_syms_load(&dd);
783 				ksyms = u64_to_ptr(info->jited_ksyms);
784 			}
785 
786 			if (json_output)
787 				jsonw_start_array(json_wtr);
788 
789 			lens = u64_to_ptr(info->jited_func_lens);
790 			for (i = 0; i < info->nr_jited_func_lens; i++) {
791 				if (ksyms) {
792 					sym = kernel_syms_search(&dd, ksyms[i]);
793 					if (sym)
794 						sprintf(sym_name, "%s", sym->name);
795 					else
796 						sprintf(sym_name, "0x%016llx", ksyms[i]);
797 				} else {
798 					strcpy(sym_name, "unknown");
799 				}
800 
801 				if (func_info) {
802 					record = func_info + i * info->func_info_rec_size;
803 					btf_dumper_type_only(btf, record->type_id,
804 							     func_sig,
805 							     sizeof(func_sig));
806 				}
807 
808 				if (json_output) {
809 					jsonw_start_object(json_wtr);
810 					if (func_info && func_sig[0] != '\0') {
811 						jsonw_name(json_wtr, "proto");
812 						jsonw_string(json_wtr, func_sig);
813 					}
814 					jsonw_name(json_wtr, "name");
815 					jsonw_string(json_wtr, sym_name);
816 					jsonw_name(json_wtr, "insns");
817 				} else {
818 					if (func_info && func_sig[0] != '\0')
819 						printf("%s:\n", func_sig);
820 					printf("%s:\n", sym_name);
821 				}
822 
823 				disasm_print_insn(img, lens[i], opcodes,
824 						  name, disasm_opt, btf,
825 						  prog_linfo, ksyms[i], i,
826 						  linum);
827 
828 				img += lens[i];
829 
830 				if (json_output)
831 					jsonw_end_object(json_wtr);
832 				else
833 					printf("\n");
834 			}
835 
836 			if (json_output)
837 				jsonw_end_array(json_wtr);
838 		} else {
839 			disasm_print_insn(buf, member_len, opcodes, name,
840 					  disasm_opt, btf, NULL, 0, 0, false);
841 		}
842 	} else if (visual) {
843 		if (json_output)
844 			jsonw_null(json_wtr);
845 		else
846 			dump_xlated_cfg(buf, member_len);
847 	} else {
848 		kernel_syms_load(&dd);
849 		dd.nr_jited_ksyms = info->nr_jited_ksyms;
850 		dd.jited_ksyms = u64_to_ptr(info->jited_ksyms);
851 		dd.btf = btf;
852 		dd.func_info = func_info;
853 		dd.finfo_rec_size = info->func_info_rec_size;
854 		dd.prog_linfo = prog_linfo;
855 
856 		if (json_output)
857 			dump_xlated_json(&dd, buf, member_len, opcodes,
858 					 linum);
859 		else
860 			dump_xlated_plain(&dd, buf, member_len, opcodes,
861 					  linum);
862 		kernel_syms_destroy(&dd);
863 	}
864 
865 	err = 0;
866 
867 exit_free:
868 	btf__free(btf);
869 	bpf_prog_linfo__free(prog_linfo);
870 	return err;
871 }
872 
do_dump(int argc,char ** argv)873 static int do_dump(int argc, char **argv)
874 {
875 	struct bpf_prog_info info;
876 	__u32 info_len = sizeof(info);
877 	size_t info_data_sz = 0;
878 	void *info_data = NULL;
879 	char *filepath = NULL;
880 	bool opcodes = false;
881 	bool visual = false;
882 	enum dump_mode mode;
883 	bool linum = false;
884 	int nb_fds, i = 0;
885 	int *fds = NULL;
886 	int err = -1;
887 
888 	if (is_prefix(*argv, "jited")) {
889 		if (disasm_init())
890 			return -1;
891 		mode = DUMP_JITED;
892 	} else if (is_prefix(*argv, "xlated")) {
893 		mode = DUMP_XLATED;
894 	} else {
895 		p_err("expected 'xlated' or 'jited', got: %s", *argv);
896 		return -1;
897 	}
898 	NEXT_ARG();
899 
900 	if (argc < 2)
901 		usage();
902 
903 	fds = malloc(sizeof(int));
904 	if (!fds) {
905 		p_err("mem alloc failed");
906 		return -1;
907 	}
908 	nb_fds = prog_parse_fds(&argc, &argv, &fds);
909 	if (nb_fds < 1)
910 		goto exit_free;
911 
912 	if (is_prefix(*argv, "file")) {
913 		NEXT_ARG();
914 		if (!argc) {
915 			p_err("expected file path");
916 			goto exit_close;
917 		}
918 		if (nb_fds > 1) {
919 			p_err("several programs matched");
920 			goto exit_close;
921 		}
922 
923 		filepath = *argv;
924 		NEXT_ARG();
925 	} else if (is_prefix(*argv, "opcodes")) {
926 		opcodes = true;
927 		NEXT_ARG();
928 	} else if (is_prefix(*argv, "visual")) {
929 		if (nb_fds > 1) {
930 			p_err("several programs matched");
931 			goto exit_close;
932 		}
933 
934 		visual = true;
935 		NEXT_ARG();
936 	} else if (is_prefix(*argv, "linum")) {
937 		linum = true;
938 		NEXT_ARG();
939 	}
940 
941 	if (argc) {
942 		usage();
943 		goto exit_close;
944 	}
945 
946 	if (json_output && nb_fds > 1)
947 		jsonw_start_array(json_wtr);	/* root array */
948 	for (i = 0; i < nb_fds; i++) {
949 		memset(&info, 0, sizeof(info));
950 
951 		err = bpf_obj_get_info_by_fd(fds[i], &info, &info_len);
952 		if (err) {
953 			p_err("can't get prog info: %s", strerror(errno));
954 			break;
955 		}
956 
957 		err = prep_prog_info(&info, mode, &info_data, &info_data_sz);
958 		if (err) {
959 			p_err("can't grow prog info_data");
960 			break;
961 		}
962 
963 		err = bpf_obj_get_info_by_fd(fds[i], &info, &info_len);
964 		if (err) {
965 			p_err("can't get prog info: %s", strerror(errno));
966 			break;
967 		}
968 
969 		if (json_output && nb_fds > 1) {
970 			jsonw_start_object(json_wtr);	/* prog object */
971 			print_prog_header_json(&info, fds[i]);
972 			jsonw_name(json_wtr, "insns");
973 		} else if (nb_fds > 1) {
974 			print_prog_header_plain(&info, fds[i]);
975 		}
976 
977 		err = prog_dump(&info, mode, filepath, opcodes, visual, linum);
978 
979 		if (json_output && nb_fds > 1)
980 			jsonw_end_object(json_wtr);	/* prog object */
981 		else if (i != nb_fds - 1 && nb_fds > 1)
982 			printf("\n");
983 
984 		if (err)
985 			break;
986 		close(fds[i]);
987 	}
988 	if (json_output && nb_fds > 1)
989 		jsonw_end_array(json_wtr);	/* root array */
990 
991 exit_close:
992 	for (; i < nb_fds; i++)
993 		close(fds[i]);
994 exit_free:
995 	free(info_data);
996 	free(fds);
997 	return err;
998 }
999 
do_pin(int argc,char ** argv)1000 static int do_pin(int argc, char **argv)
1001 {
1002 	int err;
1003 
1004 	err = do_pin_any(argc, argv, prog_parse_fd);
1005 	if (!err && json_output)
1006 		jsonw_null(json_wtr);
1007 	return err;
1008 }
1009 
1010 struct map_replace {
1011 	int idx;
1012 	int fd;
1013 	char *name;
1014 };
1015 
map_replace_compar(const void * p1,const void * p2)1016 static int map_replace_compar(const void *p1, const void *p2)
1017 {
1018 	const struct map_replace *a = p1, *b = p2;
1019 
1020 	return a->idx - b->idx;
1021 }
1022 
parse_attach_detach_args(int argc,char ** argv,int * progfd,enum bpf_attach_type * attach_type,int * mapfd)1023 static int parse_attach_detach_args(int argc, char **argv, int *progfd,
1024 				    enum bpf_attach_type *attach_type,
1025 				    int *mapfd)
1026 {
1027 	if (!REQ_ARGS(3))
1028 		return -EINVAL;
1029 
1030 	*progfd = prog_parse_fd(&argc, &argv);
1031 	if (*progfd < 0)
1032 		return *progfd;
1033 
1034 	*attach_type = parse_attach_type(*argv);
1035 	if (*attach_type == __MAX_BPF_ATTACH_TYPE) {
1036 		p_err("invalid attach/detach type");
1037 		return -EINVAL;
1038 	}
1039 
1040 	if (*attach_type == BPF_FLOW_DISSECTOR) {
1041 		*mapfd = 0;
1042 		return 0;
1043 	}
1044 
1045 	NEXT_ARG();
1046 	if (!REQ_ARGS(2))
1047 		return -EINVAL;
1048 
1049 	*mapfd = map_parse_fd(&argc, &argv);
1050 	if (*mapfd < 0)
1051 		return *mapfd;
1052 
1053 	return 0;
1054 }
1055 
do_attach(int argc,char ** argv)1056 static int do_attach(int argc, char **argv)
1057 {
1058 	enum bpf_attach_type attach_type;
1059 	int err, progfd;
1060 	int mapfd;
1061 
1062 	err = parse_attach_detach_args(argc, argv,
1063 				       &progfd, &attach_type, &mapfd);
1064 	if (err)
1065 		return err;
1066 
1067 	err = bpf_prog_attach(progfd, mapfd, attach_type, 0);
1068 	if (err) {
1069 		p_err("failed prog attach to map");
1070 		return -EINVAL;
1071 	}
1072 
1073 	if (json_output)
1074 		jsonw_null(json_wtr);
1075 	return 0;
1076 }
1077 
do_detach(int argc,char ** argv)1078 static int do_detach(int argc, char **argv)
1079 {
1080 	enum bpf_attach_type attach_type;
1081 	int err, progfd;
1082 	int mapfd;
1083 
1084 	err = parse_attach_detach_args(argc, argv,
1085 				       &progfd, &attach_type, &mapfd);
1086 	if (err)
1087 		return err;
1088 
1089 	err = bpf_prog_detach2(progfd, mapfd, attach_type);
1090 	if (err) {
1091 		p_err("failed prog detach from map");
1092 		return -EINVAL;
1093 	}
1094 
1095 	if (json_output)
1096 		jsonw_null(json_wtr);
1097 	return 0;
1098 }
1099 
check_single_stdin(char * file_data_in,char * file_ctx_in)1100 static int check_single_stdin(char *file_data_in, char *file_ctx_in)
1101 {
1102 	if (file_data_in && file_ctx_in &&
1103 	    !strcmp(file_data_in, "-") && !strcmp(file_ctx_in, "-")) {
1104 		p_err("cannot use standard input for both data_in and ctx_in");
1105 		return -1;
1106 	}
1107 
1108 	return 0;
1109 }
1110 
get_run_data(const char * fname,void ** data_ptr,unsigned int * size)1111 static int get_run_data(const char *fname, void **data_ptr, unsigned int *size)
1112 {
1113 	size_t block_size = 256;
1114 	size_t buf_size = block_size;
1115 	size_t nb_read = 0;
1116 	void *tmp;
1117 	FILE *f;
1118 
1119 	if (!fname) {
1120 		*data_ptr = NULL;
1121 		*size = 0;
1122 		return 0;
1123 	}
1124 
1125 	if (!strcmp(fname, "-"))
1126 		f = stdin;
1127 	else
1128 		f = fopen(fname, "r");
1129 	if (!f) {
1130 		p_err("failed to open %s: %s", fname, strerror(errno));
1131 		return -1;
1132 	}
1133 
1134 	*data_ptr = malloc(block_size);
1135 	if (!*data_ptr) {
1136 		p_err("failed to allocate memory for data_in/ctx_in: %s",
1137 		      strerror(errno));
1138 		goto err_fclose;
1139 	}
1140 
1141 	while ((nb_read += fread(*data_ptr + nb_read, 1, block_size, f))) {
1142 		if (feof(f))
1143 			break;
1144 		if (ferror(f)) {
1145 			p_err("failed to read data_in/ctx_in from %s: %s",
1146 			      fname, strerror(errno));
1147 			goto err_free;
1148 		}
1149 		if (nb_read > buf_size - block_size) {
1150 			if (buf_size == UINT32_MAX) {
1151 				p_err("data_in/ctx_in is too long (max: %d)",
1152 				      UINT32_MAX);
1153 				goto err_free;
1154 			}
1155 			/* No space for fread()-ing next chunk; realloc() */
1156 			buf_size *= 2;
1157 			tmp = realloc(*data_ptr, buf_size);
1158 			if (!tmp) {
1159 				p_err("failed to reallocate data_in/ctx_in: %s",
1160 				      strerror(errno));
1161 				goto err_free;
1162 			}
1163 			*data_ptr = tmp;
1164 		}
1165 	}
1166 	if (f != stdin)
1167 		fclose(f);
1168 
1169 	*size = nb_read;
1170 	return 0;
1171 
1172 err_free:
1173 	free(*data_ptr);
1174 	*data_ptr = NULL;
1175 err_fclose:
1176 	if (f != stdin)
1177 		fclose(f);
1178 	return -1;
1179 }
1180 
hex_print(void * data,unsigned int size,FILE * f)1181 static void hex_print(void *data, unsigned int size, FILE *f)
1182 {
1183 	size_t i, j;
1184 	char c;
1185 
1186 	for (i = 0; i < size; i += 16) {
1187 		/* Row offset */
1188 		fprintf(f, "%07zx\t", i);
1189 
1190 		/* Hexadecimal values */
1191 		for (j = i; j < i + 16 && j < size; j++)
1192 			fprintf(f, "%02x%s", *(uint8_t *)(data + j),
1193 				j % 2 ? " " : "");
1194 		for (; j < i + 16; j++)
1195 			fprintf(f, "  %s", j % 2 ? " " : "");
1196 
1197 		/* ASCII values (if relevant), '.' otherwise */
1198 		fprintf(f, "| ");
1199 		for (j = i; j < i + 16 && j < size; j++) {
1200 			c = *(char *)(data + j);
1201 			if (c < ' ' || c > '~')
1202 				c = '.';
1203 			fprintf(f, "%c%s", c, j == i + 7 ? " " : "");
1204 		}
1205 
1206 		fprintf(f, "\n");
1207 	}
1208 }
1209 
1210 static int
print_run_output(void * data,unsigned int size,const char * fname,const char * json_key)1211 print_run_output(void *data, unsigned int size, const char *fname,
1212 		 const char *json_key)
1213 {
1214 	size_t nb_written;
1215 	FILE *f;
1216 
1217 	if (!fname)
1218 		return 0;
1219 
1220 	if (!strcmp(fname, "-")) {
1221 		f = stdout;
1222 		if (json_output) {
1223 			jsonw_name(json_wtr, json_key);
1224 			print_data_json(data, size);
1225 		} else {
1226 			hex_print(data, size, f);
1227 		}
1228 		return 0;
1229 	}
1230 
1231 	f = fopen(fname, "w");
1232 	if (!f) {
1233 		p_err("failed to open %s: %s", fname, strerror(errno));
1234 		return -1;
1235 	}
1236 
1237 	nb_written = fwrite(data, 1, size, f);
1238 	fclose(f);
1239 	if (nb_written != size) {
1240 		p_err("failed to write output data/ctx: %s", strerror(errno));
1241 		return -1;
1242 	}
1243 
1244 	return 0;
1245 }
1246 
alloc_run_data(void ** data_ptr,unsigned int size_out)1247 static int alloc_run_data(void **data_ptr, unsigned int size_out)
1248 {
1249 	*data_ptr = calloc(size_out, 1);
1250 	if (!*data_ptr) {
1251 		p_err("failed to allocate memory for output data/ctx: %s",
1252 		      strerror(errno));
1253 		return -1;
1254 	}
1255 
1256 	return 0;
1257 }
1258 
do_run(int argc,char ** argv)1259 static int do_run(int argc, char **argv)
1260 {
1261 	char *data_fname_in = NULL, *data_fname_out = NULL;
1262 	char *ctx_fname_in = NULL, *ctx_fname_out = NULL;
1263 	const unsigned int default_size = SZ_32K;
1264 	void *data_in = NULL, *data_out = NULL;
1265 	void *ctx_in = NULL, *ctx_out = NULL;
1266 	unsigned int repeat = 1;
1267 	int fd, err;
1268 	LIBBPF_OPTS(bpf_test_run_opts, test_attr);
1269 
1270 	if (!REQ_ARGS(4))
1271 		return -1;
1272 
1273 	fd = prog_parse_fd(&argc, &argv);
1274 	if (fd < 0)
1275 		return -1;
1276 
1277 	while (argc) {
1278 		if (detect_common_prefix(*argv, "data_in", "data_out",
1279 					 "data_size_out", NULL))
1280 			return -1;
1281 		if (detect_common_prefix(*argv, "ctx_in", "ctx_out",
1282 					 "ctx_size_out", NULL))
1283 			return -1;
1284 
1285 		if (is_prefix(*argv, "data_in")) {
1286 			NEXT_ARG();
1287 			if (!REQ_ARGS(1))
1288 				return -1;
1289 
1290 			data_fname_in = GET_ARG();
1291 			if (check_single_stdin(data_fname_in, ctx_fname_in))
1292 				return -1;
1293 		} else if (is_prefix(*argv, "data_out")) {
1294 			NEXT_ARG();
1295 			if (!REQ_ARGS(1))
1296 				return -1;
1297 
1298 			data_fname_out = GET_ARG();
1299 		} else if (is_prefix(*argv, "data_size_out")) {
1300 			char *endptr;
1301 
1302 			NEXT_ARG();
1303 			if (!REQ_ARGS(1))
1304 				return -1;
1305 
1306 			test_attr.data_size_out = strtoul(*argv, &endptr, 0);
1307 			if (*endptr) {
1308 				p_err("can't parse %s as output data size",
1309 				      *argv);
1310 				return -1;
1311 			}
1312 			NEXT_ARG();
1313 		} else if (is_prefix(*argv, "ctx_in")) {
1314 			NEXT_ARG();
1315 			if (!REQ_ARGS(1))
1316 				return -1;
1317 
1318 			ctx_fname_in = GET_ARG();
1319 			if (check_single_stdin(data_fname_in, ctx_fname_in))
1320 				return -1;
1321 		} else if (is_prefix(*argv, "ctx_out")) {
1322 			NEXT_ARG();
1323 			if (!REQ_ARGS(1))
1324 				return -1;
1325 
1326 			ctx_fname_out = GET_ARG();
1327 		} else if (is_prefix(*argv, "ctx_size_out")) {
1328 			char *endptr;
1329 
1330 			NEXT_ARG();
1331 			if (!REQ_ARGS(1))
1332 				return -1;
1333 
1334 			test_attr.ctx_size_out = strtoul(*argv, &endptr, 0);
1335 			if (*endptr) {
1336 				p_err("can't parse %s as output context size",
1337 				      *argv);
1338 				return -1;
1339 			}
1340 			NEXT_ARG();
1341 		} else if (is_prefix(*argv, "repeat")) {
1342 			char *endptr;
1343 
1344 			NEXT_ARG();
1345 			if (!REQ_ARGS(1))
1346 				return -1;
1347 
1348 			repeat = strtoul(*argv, &endptr, 0);
1349 			if (*endptr) {
1350 				p_err("can't parse %s as repeat number",
1351 				      *argv);
1352 				return -1;
1353 			}
1354 			NEXT_ARG();
1355 		} else {
1356 			p_err("expected no more arguments, 'data_in', 'data_out', 'data_size_out', 'ctx_in', 'ctx_out', 'ctx_size_out' or 'repeat', got: '%s'?",
1357 			      *argv);
1358 			return -1;
1359 		}
1360 	}
1361 
1362 	err = get_run_data(data_fname_in, &data_in, &test_attr.data_size_in);
1363 	if (err)
1364 		return -1;
1365 
1366 	if (data_in) {
1367 		if (!test_attr.data_size_out)
1368 			test_attr.data_size_out = default_size;
1369 		err = alloc_run_data(&data_out, test_attr.data_size_out);
1370 		if (err)
1371 			goto free_data_in;
1372 	}
1373 
1374 	err = get_run_data(ctx_fname_in, &ctx_in, &test_attr.ctx_size_in);
1375 	if (err)
1376 		goto free_data_out;
1377 
1378 	if (ctx_in) {
1379 		if (!test_attr.ctx_size_out)
1380 			test_attr.ctx_size_out = default_size;
1381 		err = alloc_run_data(&ctx_out, test_attr.ctx_size_out);
1382 		if (err)
1383 			goto free_ctx_in;
1384 	}
1385 
1386 	test_attr.repeat	= repeat;
1387 	test_attr.data_in	= data_in;
1388 	test_attr.data_out	= data_out;
1389 	test_attr.ctx_in	= ctx_in;
1390 	test_attr.ctx_out	= ctx_out;
1391 
1392 	err = bpf_prog_test_run_opts(fd, &test_attr);
1393 	if (err) {
1394 		p_err("failed to run program: %s", strerror(errno));
1395 		goto free_ctx_out;
1396 	}
1397 
1398 	err = 0;
1399 
1400 	if (json_output)
1401 		jsonw_start_object(json_wtr);	/* root */
1402 
1403 	/* Do not exit on errors occurring when printing output data/context,
1404 	 * we still want to print return value and duration for program run.
1405 	 */
1406 	if (test_attr.data_size_out)
1407 		err += print_run_output(test_attr.data_out,
1408 					test_attr.data_size_out,
1409 					data_fname_out, "data_out");
1410 	if (test_attr.ctx_size_out)
1411 		err += print_run_output(test_attr.ctx_out,
1412 					test_attr.ctx_size_out,
1413 					ctx_fname_out, "ctx_out");
1414 
1415 	if (json_output) {
1416 		jsonw_uint_field(json_wtr, "retval", test_attr.retval);
1417 		jsonw_uint_field(json_wtr, "duration", test_attr.duration);
1418 		jsonw_end_object(json_wtr);	/* root */
1419 	} else {
1420 		fprintf(stdout, "Return value: %u, duration%s: %uns\n",
1421 			test_attr.retval,
1422 			repeat > 1 ? " (average)" : "", test_attr.duration);
1423 	}
1424 
1425 free_ctx_out:
1426 	free(ctx_out);
1427 free_ctx_in:
1428 	free(ctx_in);
1429 free_data_out:
1430 	free(data_out);
1431 free_data_in:
1432 	free(data_in);
1433 
1434 	return err;
1435 }
1436 
1437 static int
get_prog_type_by_name(const char * name,enum bpf_prog_type * prog_type,enum bpf_attach_type * expected_attach_type)1438 get_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
1439 		      enum bpf_attach_type *expected_attach_type)
1440 {
1441 	libbpf_print_fn_t print_backup;
1442 	int ret;
1443 
1444 	ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type);
1445 	if (!ret)
1446 		return ret;
1447 
1448 	/* libbpf_prog_type_by_name() failed, let's re-run with debug level */
1449 	print_backup = libbpf_set_print(print_all_levels);
1450 	ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type);
1451 	libbpf_set_print(print_backup);
1452 
1453 	return ret;
1454 }
1455 
load_with_options(int argc,char ** argv,bool first_prog_only)1456 static int load_with_options(int argc, char **argv, bool first_prog_only)
1457 {
1458 	enum bpf_prog_type common_prog_type = BPF_PROG_TYPE_UNSPEC;
1459 	DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts,
1460 		.relaxed_maps = relaxed_maps,
1461 	);
1462 	enum bpf_attach_type expected_attach_type;
1463 	struct map_replace *map_replace = NULL;
1464 	struct bpf_program *prog = NULL, *pos;
1465 	unsigned int old_map_fds = 0;
1466 	const char *pinmaps = NULL;
1467 	struct bpf_object *obj;
1468 	struct bpf_map *map;
1469 	const char *pinfile;
1470 	unsigned int i, j;
1471 	__u32 ifindex = 0;
1472 	const char *file;
1473 	int idx, err;
1474 
1475 
1476 	if (!REQ_ARGS(2))
1477 		return -1;
1478 	file = GET_ARG();
1479 	pinfile = GET_ARG();
1480 
1481 	while (argc) {
1482 		if (is_prefix(*argv, "type")) {
1483 			NEXT_ARG();
1484 
1485 			if (common_prog_type != BPF_PROG_TYPE_UNSPEC) {
1486 				p_err("program type already specified");
1487 				goto err_free_reuse_maps;
1488 			}
1489 			if (!REQ_ARGS(1))
1490 				goto err_free_reuse_maps;
1491 
1492 			err = libbpf_prog_type_by_name(*argv, &common_prog_type,
1493 						       &expected_attach_type);
1494 			if (err < 0) {
1495 				/* Put a '/' at the end of type to appease libbpf */
1496 				char *type = malloc(strlen(*argv) + 2);
1497 
1498 				if (!type) {
1499 					p_err("mem alloc failed");
1500 					goto err_free_reuse_maps;
1501 				}
1502 				*type = 0;
1503 				strcat(type, *argv);
1504 				strcat(type, "/");
1505 
1506 				err = get_prog_type_by_name(type, &common_prog_type,
1507 							    &expected_attach_type);
1508 				free(type);
1509 				if (err < 0)
1510 					goto err_free_reuse_maps;
1511 			}
1512 
1513 			NEXT_ARG();
1514 		} else if (is_prefix(*argv, "map")) {
1515 			void *new_map_replace;
1516 			char *endptr, *name;
1517 			int fd;
1518 
1519 			NEXT_ARG();
1520 
1521 			if (!REQ_ARGS(4))
1522 				goto err_free_reuse_maps;
1523 
1524 			if (is_prefix(*argv, "idx")) {
1525 				NEXT_ARG();
1526 
1527 				idx = strtoul(*argv, &endptr, 0);
1528 				if (*endptr) {
1529 					p_err("can't parse %s as IDX", *argv);
1530 					goto err_free_reuse_maps;
1531 				}
1532 				name = NULL;
1533 			} else if (is_prefix(*argv, "name")) {
1534 				NEXT_ARG();
1535 
1536 				name = *argv;
1537 				idx = -1;
1538 			} else {
1539 				p_err("expected 'idx' or 'name', got: '%s'?",
1540 				      *argv);
1541 				goto err_free_reuse_maps;
1542 			}
1543 			NEXT_ARG();
1544 
1545 			fd = map_parse_fd(&argc, &argv);
1546 			if (fd < 0)
1547 				goto err_free_reuse_maps;
1548 
1549 			new_map_replace = libbpf_reallocarray(map_replace,
1550 							      old_map_fds + 1,
1551 							      sizeof(*map_replace));
1552 			if (!new_map_replace) {
1553 				p_err("mem alloc failed");
1554 				goto err_free_reuse_maps;
1555 			}
1556 			map_replace = new_map_replace;
1557 
1558 			map_replace[old_map_fds].idx = idx;
1559 			map_replace[old_map_fds].name = name;
1560 			map_replace[old_map_fds].fd = fd;
1561 			old_map_fds++;
1562 		} else if (is_prefix(*argv, "dev")) {
1563 			NEXT_ARG();
1564 
1565 			if (ifindex) {
1566 				p_err("offload device already specified");
1567 				goto err_free_reuse_maps;
1568 			}
1569 			if (!REQ_ARGS(1))
1570 				goto err_free_reuse_maps;
1571 
1572 			ifindex = if_nametoindex(*argv);
1573 			if (!ifindex) {
1574 				p_err("unrecognized netdevice '%s': %s",
1575 				      *argv, strerror(errno));
1576 				goto err_free_reuse_maps;
1577 			}
1578 			NEXT_ARG();
1579 		} else if (is_prefix(*argv, "pinmaps")) {
1580 			NEXT_ARG();
1581 
1582 			if (!REQ_ARGS(1))
1583 				goto err_free_reuse_maps;
1584 
1585 			pinmaps = GET_ARG();
1586 		} else {
1587 			p_err("expected no more arguments, 'type', 'map' or 'dev', got: '%s'?",
1588 			      *argv);
1589 			goto err_free_reuse_maps;
1590 		}
1591 	}
1592 
1593 	set_max_rlimit();
1594 
1595 	if (verifier_logs)
1596 		/* log_level1 + log_level2 + stats, but not stable UAPI */
1597 		open_opts.kernel_log_level = 1 + 2 + 4;
1598 
1599 	obj = bpf_object__open_file(file, &open_opts);
1600 	if (libbpf_get_error(obj)) {
1601 		p_err("failed to open object file");
1602 		goto err_free_reuse_maps;
1603 	}
1604 
1605 	bpf_object__for_each_program(pos, obj) {
1606 		enum bpf_prog_type prog_type = common_prog_type;
1607 
1608 		if (prog_type == BPF_PROG_TYPE_UNSPEC) {
1609 			const char *sec_name = bpf_program__section_name(pos);
1610 
1611 			err = get_prog_type_by_name(sec_name, &prog_type,
1612 						    &expected_attach_type);
1613 			if (err < 0)
1614 				goto err_close_obj;
1615 		}
1616 
1617 		bpf_program__set_ifindex(pos, ifindex);
1618 		bpf_program__set_type(pos, prog_type);
1619 		bpf_program__set_expected_attach_type(pos, expected_attach_type);
1620 	}
1621 
1622 	qsort(map_replace, old_map_fds, sizeof(*map_replace),
1623 	      map_replace_compar);
1624 
1625 	/* After the sort maps by name will be first on the list, because they
1626 	 * have idx == -1.  Resolve them.
1627 	 */
1628 	j = 0;
1629 	while (j < old_map_fds && map_replace[j].name) {
1630 		i = 0;
1631 		bpf_object__for_each_map(map, obj) {
1632 			if (!strcmp(bpf_map__name(map), map_replace[j].name)) {
1633 				map_replace[j].idx = i;
1634 				break;
1635 			}
1636 			i++;
1637 		}
1638 		if (map_replace[j].idx == -1) {
1639 			p_err("unable to find map '%s'", map_replace[j].name);
1640 			goto err_close_obj;
1641 		}
1642 		j++;
1643 	}
1644 	/* Resort if any names were resolved */
1645 	if (j)
1646 		qsort(map_replace, old_map_fds, sizeof(*map_replace),
1647 		      map_replace_compar);
1648 
1649 	/* Set ifindex and name reuse */
1650 	j = 0;
1651 	idx = 0;
1652 	bpf_object__for_each_map(map, obj) {
1653 		if (bpf_map__type(map) != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1654 			bpf_map__set_ifindex(map, ifindex);
1655 
1656 		if (j < old_map_fds && idx == map_replace[j].idx) {
1657 			err = bpf_map__reuse_fd(map, map_replace[j++].fd);
1658 			if (err) {
1659 				p_err("unable to set up map reuse: %d", err);
1660 				goto err_close_obj;
1661 			}
1662 
1663 			/* Next reuse wants to apply to the same map */
1664 			if (j < old_map_fds && map_replace[j].idx == idx) {
1665 				p_err("replacement for map idx %d specified more than once",
1666 				      idx);
1667 				goto err_close_obj;
1668 			}
1669 		}
1670 
1671 		idx++;
1672 	}
1673 	if (j < old_map_fds) {
1674 		p_err("map idx '%d' not used", map_replace[j].idx);
1675 		goto err_close_obj;
1676 	}
1677 
1678 	err = bpf_object__load(obj);
1679 	if (err) {
1680 		p_err("failed to load object file");
1681 		goto err_close_obj;
1682 	}
1683 
1684 	err = mount_bpffs_for_pin(pinfile);
1685 	if (err)
1686 		goto err_close_obj;
1687 
1688 	if (first_prog_only) {
1689 		prog = bpf_object__next_program(obj, NULL);
1690 		if (!prog) {
1691 			p_err("object file doesn't contain any bpf program");
1692 			goto err_close_obj;
1693 		}
1694 
1695 		err = bpf_obj_pin(bpf_program__fd(prog), pinfile);
1696 		if (err) {
1697 			p_err("failed to pin program %s",
1698 			      bpf_program__section_name(prog));
1699 			goto err_close_obj;
1700 		}
1701 	} else {
1702 		err = bpf_object__pin_programs(obj, pinfile);
1703 		if (err) {
1704 			p_err("failed to pin all programs");
1705 			goto err_close_obj;
1706 		}
1707 	}
1708 
1709 	if (pinmaps) {
1710 		err = bpf_object__pin_maps(obj, pinmaps);
1711 		if (err) {
1712 			p_err("failed to pin all maps");
1713 			goto err_unpin;
1714 		}
1715 	}
1716 
1717 	if (json_output)
1718 		jsonw_null(json_wtr);
1719 
1720 	bpf_object__close(obj);
1721 	for (i = 0; i < old_map_fds; i++)
1722 		close(map_replace[i].fd);
1723 	free(map_replace);
1724 
1725 	return 0;
1726 
1727 err_unpin:
1728 	if (first_prog_only)
1729 		unlink(pinfile);
1730 	else
1731 		bpf_object__unpin_programs(obj, pinfile);
1732 err_close_obj:
1733 	if (!legacy_libbpf) {
1734 		p_info("Warning: bpftool is now running in libbpf strict mode and has more stringent requirements about BPF programs.\n"
1735 		       "If it used to work for this object file but now doesn't, see --legacy option for more details.\n");
1736 	}
1737 
1738 	bpf_object__close(obj);
1739 err_free_reuse_maps:
1740 	for (i = 0; i < old_map_fds; i++)
1741 		close(map_replace[i].fd);
1742 	free(map_replace);
1743 	return -1;
1744 }
1745 
count_open_fds(void)1746 static int count_open_fds(void)
1747 {
1748 	DIR *dp = opendir("/proc/self/fd");
1749 	struct dirent *de;
1750 	int cnt = -3;
1751 
1752 	if (!dp)
1753 		return -1;
1754 
1755 	while ((de = readdir(dp)))
1756 		cnt++;
1757 
1758 	closedir(dp);
1759 	return cnt;
1760 }
1761 
try_loader(struct gen_loader_opts * gen)1762 static int try_loader(struct gen_loader_opts *gen)
1763 {
1764 	struct bpf_load_and_run_opts opts = {};
1765 	struct bpf_loader_ctx *ctx;
1766 	int ctx_sz = sizeof(*ctx) + 64 * max(sizeof(struct bpf_map_desc),
1767 					     sizeof(struct bpf_prog_desc));
1768 	int log_buf_sz = (1u << 24) - 1;
1769 	int err, fds_before, fd_delta;
1770 	char *log_buf = NULL;
1771 
1772 	ctx = alloca(ctx_sz);
1773 	memset(ctx, 0, ctx_sz);
1774 	ctx->sz = ctx_sz;
1775 	if (verifier_logs) {
1776 		ctx->log_level = 1 + 2 + 4;
1777 		ctx->log_size = log_buf_sz;
1778 		log_buf = malloc(log_buf_sz);
1779 		if (!log_buf)
1780 			return -ENOMEM;
1781 		ctx->log_buf = (long) log_buf;
1782 	}
1783 	opts.ctx = ctx;
1784 	opts.data = gen->data;
1785 	opts.data_sz = gen->data_sz;
1786 	opts.insns = gen->insns;
1787 	opts.insns_sz = gen->insns_sz;
1788 	fds_before = count_open_fds();
1789 	err = bpf_load_and_run(&opts);
1790 	fd_delta = count_open_fds() - fds_before;
1791 	if (err < 0 || verifier_logs) {
1792 		fprintf(stderr, "err %d\n%s\n%s", err, opts.errstr, log_buf);
1793 		if (fd_delta && err < 0)
1794 			fprintf(stderr, "loader prog leaked %d FDs\n",
1795 				fd_delta);
1796 	}
1797 	free(log_buf);
1798 	return err;
1799 }
1800 
do_loader(int argc,char ** argv)1801 static int do_loader(int argc, char **argv)
1802 {
1803 	DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts);
1804 	DECLARE_LIBBPF_OPTS(gen_loader_opts, gen);
1805 	struct bpf_object *obj;
1806 	const char *file;
1807 	int err = 0;
1808 
1809 	if (!REQ_ARGS(1))
1810 		return -1;
1811 	file = GET_ARG();
1812 
1813 	if (verifier_logs)
1814 		/* log_level1 + log_level2 + stats, but not stable UAPI */
1815 		open_opts.kernel_log_level = 1 + 2 + 4;
1816 
1817 	obj = bpf_object__open_file(file, &open_opts);
1818 	if (libbpf_get_error(obj)) {
1819 		p_err("failed to open object file");
1820 		goto err_close_obj;
1821 	}
1822 
1823 	err = bpf_object__gen_loader(obj, &gen);
1824 	if (err)
1825 		goto err_close_obj;
1826 
1827 	err = bpf_object__load(obj);
1828 	if (err) {
1829 		p_err("failed to load object file");
1830 		goto err_close_obj;
1831 	}
1832 
1833 	if (verifier_logs) {
1834 		struct dump_data dd = {};
1835 
1836 		kernel_syms_load(&dd);
1837 		dump_xlated_plain(&dd, (void *)gen.insns, gen.insns_sz, false, false);
1838 		kernel_syms_destroy(&dd);
1839 	}
1840 	err = try_loader(&gen);
1841 err_close_obj:
1842 	bpf_object__close(obj);
1843 	return err;
1844 }
1845 
do_load(int argc,char ** argv)1846 static int do_load(int argc, char **argv)
1847 {
1848 	if (use_loader)
1849 		return do_loader(argc, argv);
1850 	return load_with_options(argc, argv, true);
1851 }
1852 
do_loadall(int argc,char ** argv)1853 static int do_loadall(int argc, char **argv)
1854 {
1855 	return load_with_options(argc, argv, false);
1856 }
1857 
1858 #ifdef BPFTOOL_WITHOUT_SKELETONS
1859 
do_profile(int argc,char ** argv)1860 static int do_profile(int argc, char **argv)
1861 {
1862 	p_err("bpftool prog profile command is not supported. Please build bpftool with clang >= 10.0.0");
1863 	return 0;
1864 }
1865 
1866 #else /* BPFTOOL_WITHOUT_SKELETONS */
1867 
1868 #include "profiler.skel.h"
1869 
1870 struct profile_metric {
1871 	const char *name;
1872 	struct bpf_perf_event_value val;
1873 	struct perf_event_attr attr;
1874 	bool selected;
1875 
1876 	/* calculate ratios like instructions per cycle */
1877 	const int ratio_metric; /* 0 for N/A, 1 for index 0 (cycles) */
1878 	const char *ratio_desc;
1879 	const float ratio_mul;
1880 } metrics[] = {
1881 	{
1882 		.name = "cycles",
1883 		.attr = {
1884 			.type = PERF_TYPE_HARDWARE,
1885 			.config = PERF_COUNT_HW_CPU_CYCLES,
1886 			.exclude_user = 1,
1887 		},
1888 	},
1889 	{
1890 		.name = "instructions",
1891 		.attr = {
1892 			.type = PERF_TYPE_HARDWARE,
1893 			.config = PERF_COUNT_HW_INSTRUCTIONS,
1894 			.exclude_user = 1,
1895 		},
1896 		.ratio_metric = 1,
1897 		.ratio_desc = "insns per cycle",
1898 		.ratio_mul = 1.0,
1899 	},
1900 	{
1901 		.name = "l1d_loads",
1902 		.attr = {
1903 			.type = PERF_TYPE_HW_CACHE,
1904 			.config =
1905 				PERF_COUNT_HW_CACHE_L1D |
1906 				(PERF_COUNT_HW_CACHE_OP_READ << 8) |
1907 				(PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16),
1908 			.exclude_user = 1,
1909 		},
1910 	},
1911 	{
1912 		.name = "llc_misses",
1913 		.attr = {
1914 			.type = PERF_TYPE_HW_CACHE,
1915 			.config =
1916 				PERF_COUNT_HW_CACHE_LL |
1917 				(PERF_COUNT_HW_CACHE_OP_READ << 8) |
1918 				(PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
1919 			.exclude_user = 1
1920 		},
1921 		.ratio_metric = 2,
1922 		.ratio_desc = "LLC misses per million insns",
1923 		.ratio_mul = 1e6,
1924 	},
1925 	{
1926 		.name = "itlb_misses",
1927 		.attr = {
1928 			.type = PERF_TYPE_HW_CACHE,
1929 			.config =
1930 				PERF_COUNT_HW_CACHE_ITLB |
1931 				(PERF_COUNT_HW_CACHE_OP_READ << 8) |
1932 				(PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
1933 			.exclude_user = 1
1934 		},
1935 		.ratio_metric = 2,
1936 		.ratio_desc = "itlb misses per million insns",
1937 		.ratio_mul = 1e6,
1938 	},
1939 	{
1940 		.name = "dtlb_misses",
1941 		.attr = {
1942 			.type = PERF_TYPE_HW_CACHE,
1943 			.config =
1944 				PERF_COUNT_HW_CACHE_DTLB |
1945 				(PERF_COUNT_HW_CACHE_OP_READ << 8) |
1946 				(PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
1947 			.exclude_user = 1
1948 		},
1949 		.ratio_metric = 2,
1950 		.ratio_desc = "dtlb misses per million insns",
1951 		.ratio_mul = 1e6,
1952 	},
1953 };
1954 
1955 static __u64 profile_total_count;
1956 
1957 #define MAX_NUM_PROFILE_METRICS 4
1958 
profile_parse_metrics(int argc,char ** argv)1959 static int profile_parse_metrics(int argc, char **argv)
1960 {
1961 	unsigned int metric_cnt;
1962 	int selected_cnt = 0;
1963 	unsigned int i;
1964 
1965 	metric_cnt = ARRAY_SIZE(metrics);
1966 
1967 	while (argc > 0) {
1968 		for (i = 0; i < metric_cnt; i++) {
1969 			if (is_prefix(argv[0], metrics[i].name)) {
1970 				if (!metrics[i].selected)
1971 					selected_cnt++;
1972 				metrics[i].selected = true;
1973 				break;
1974 			}
1975 		}
1976 		if (i == metric_cnt) {
1977 			p_err("unknown metric %s", argv[0]);
1978 			return -1;
1979 		}
1980 		NEXT_ARG();
1981 	}
1982 	if (selected_cnt > MAX_NUM_PROFILE_METRICS) {
1983 		p_err("too many (%d) metrics, please specify no more than %d metrics at at time",
1984 		      selected_cnt, MAX_NUM_PROFILE_METRICS);
1985 		return -1;
1986 	}
1987 	return selected_cnt;
1988 }
1989 
profile_read_values(struct profiler_bpf * obj)1990 static void profile_read_values(struct profiler_bpf *obj)
1991 {
1992 	__u32 m, cpu, num_cpu = obj->rodata->num_cpu;
1993 	int reading_map_fd, count_map_fd;
1994 	__u64 counts[num_cpu];
1995 	__u32 key = 0;
1996 	int err;
1997 
1998 	reading_map_fd = bpf_map__fd(obj->maps.accum_readings);
1999 	count_map_fd = bpf_map__fd(obj->maps.counts);
2000 	if (reading_map_fd < 0 || count_map_fd < 0) {
2001 		p_err("failed to get fd for map");
2002 		return;
2003 	}
2004 
2005 	err = bpf_map_lookup_elem(count_map_fd, &key, counts);
2006 	if (err) {
2007 		p_err("failed to read count_map: %s", strerror(errno));
2008 		return;
2009 	}
2010 
2011 	profile_total_count = 0;
2012 	for (cpu = 0; cpu < num_cpu; cpu++)
2013 		profile_total_count += counts[cpu];
2014 
2015 	for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2016 		struct bpf_perf_event_value values[num_cpu];
2017 
2018 		if (!metrics[m].selected)
2019 			continue;
2020 
2021 		err = bpf_map_lookup_elem(reading_map_fd, &key, values);
2022 		if (err) {
2023 			p_err("failed to read reading_map: %s",
2024 			      strerror(errno));
2025 			return;
2026 		}
2027 		for (cpu = 0; cpu < num_cpu; cpu++) {
2028 			metrics[m].val.counter += values[cpu].counter;
2029 			metrics[m].val.enabled += values[cpu].enabled;
2030 			metrics[m].val.running += values[cpu].running;
2031 		}
2032 		key++;
2033 	}
2034 }
2035 
profile_print_readings_json(void)2036 static void profile_print_readings_json(void)
2037 {
2038 	__u32 m;
2039 
2040 	jsonw_start_array(json_wtr);
2041 	for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2042 		if (!metrics[m].selected)
2043 			continue;
2044 		jsonw_start_object(json_wtr);
2045 		jsonw_string_field(json_wtr, "metric", metrics[m].name);
2046 		jsonw_lluint_field(json_wtr, "run_cnt", profile_total_count);
2047 		jsonw_lluint_field(json_wtr, "value", metrics[m].val.counter);
2048 		jsonw_lluint_field(json_wtr, "enabled", metrics[m].val.enabled);
2049 		jsonw_lluint_field(json_wtr, "running", metrics[m].val.running);
2050 
2051 		jsonw_end_object(json_wtr);
2052 	}
2053 	jsonw_end_array(json_wtr);
2054 }
2055 
profile_print_readings_plain(void)2056 static void profile_print_readings_plain(void)
2057 {
2058 	__u32 m;
2059 
2060 	printf("\n%18llu %-20s\n", profile_total_count, "run_cnt");
2061 	for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2062 		struct bpf_perf_event_value *val = &metrics[m].val;
2063 		int r;
2064 
2065 		if (!metrics[m].selected)
2066 			continue;
2067 		printf("%18llu %-20s", val->counter, metrics[m].name);
2068 
2069 		r = metrics[m].ratio_metric - 1;
2070 		if (r >= 0 && metrics[r].selected &&
2071 		    metrics[r].val.counter > 0) {
2072 			printf("# %8.2f %-30s",
2073 			       val->counter * metrics[m].ratio_mul /
2074 			       metrics[r].val.counter,
2075 			       metrics[m].ratio_desc);
2076 		} else {
2077 			printf("%-41s", "");
2078 		}
2079 
2080 		if (val->enabled > val->running)
2081 			printf("(%4.2f%%)",
2082 			       val->running * 100.0 / val->enabled);
2083 		printf("\n");
2084 	}
2085 }
2086 
profile_print_readings(void)2087 static void profile_print_readings(void)
2088 {
2089 	if (json_output)
2090 		profile_print_readings_json();
2091 	else
2092 		profile_print_readings_plain();
2093 }
2094 
profile_target_name(int tgt_fd)2095 static char *profile_target_name(int tgt_fd)
2096 {
2097 	struct bpf_func_info func_info;
2098 	struct bpf_prog_info info = {};
2099 	__u32 info_len = sizeof(info);
2100 	const struct btf_type *t;
2101 	__u32 func_info_rec_size;
2102 	struct btf *btf = NULL;
2103 	char *name = NULL;
2104 	int err;
2105 
2106 	err = bpf_obj_get_info_by_fd(tgt_fd, &info, &info_len);
2107 	if (err) {
2108 		p_err("failed to bpf_obj_get_info_by_fd for prog FD %d", tgt_fd);
2109 		goto out;
2110 	}
2111 
2112 	if (info.btf_id == 0) {
2113 		p_err("prog FD %d doesn't have valid btf", tgt_fd);
2114 		goto out;
2115 	}
2116 
2117 	func_info_rec_size = info.func_info_rec_size;
2118 	if (info.nr_func_info == 0) {
2119 		p_err("bpf_obj_get_info_by_fd for prog FD %d found 0 func_info", tgt_fd);
2120 		goto out;
2121 	}
2122 
2123 	memset(&info, 0, sizeof(info));
2124 	info.nr_func_info = 1;
2125 	info.func_info_rec_size = func_info_rec_size;
2126 	info.func_info = ptr_to_u64(&func_info);
2127 
2128 	err = bpf_obj_get_info_by_fd(tgt_fd, &info, &info_len);
2129 	if (err) {
2130 		p_err("failed to get func_info for prog FD %d", tgt_fd);
2131 		goto out;
2132 	}
2133 
2134 	btf = btf__load_from_kernel_by_id(info.btf_id);
2135 	if (libbpf_get_error(btf)) {
2136 		p_err("failed to load btf for prog FD %d", tgt_fd);
2137 		goto out;
2138 	}
2139 
2140 	t = btf__type_by_id(btf, func_info.type_id);
2141 	if (!t) {
2142 		p_err("btf %d doesn't have type %d",
2143 		      info.btf_id, func_info.type_id);
2144 		goto out;
2145 	}
2146 	name = strdup(btf__name_by_offset(btf, t->name_off));
2147 out:
2148 	btf__free(btf);
2149 	return name;
2150 }
2151 
2152 static struct profiler_bpf *profile_obj;
2153 static int profile_tgt_fd = -1;
2154 static char *profile_tgt_name;
2155 static int *profile_perf_events;
2156 static int profile_perf_event_cnt;
2157 
profile_close_perf_events(struct profiler_bpf * obj)2158 static void profile_close_perf_events(struct profiler_bpf *obj)
2159 {
2160 	int i;
2161 
2162 	for (i = profile_perf_event_cnt - 1; i >= 0; i--)
2163 		close(profile_perf_events[i]);
2164 
2165 	free(profile_perf_events);
2166 	profile_perf_event_cnt = 0;
2167 }
2168 
profile_open_perf_events(struct profiler_bpf * obj)2169 static int profile_open_perf_events(struct profiler_bpf *obj)
2170 {
2171 	unsigned int cpu, m;
2172 	int map_fd, pmu_fd;
2173 
2174 	profile_perf_events = calloc(
2175 		sizeof(int), obj->rodata->num_cpu * obj->rodata->num_metric);
2176 	if (!profile_perf_events) {
2177 		p_err("failed to allocate memory for perf_event array: %s",
2178 		      strerror(errno));
2179 		return -1;
2180 	}
2181 	map_fd = bpf_map__fd(obj->maps.events);
2182 	if (map_fd < 0) {
2183 		p_err("failed to get fd for events map");
2184 		return -1;
2185 	}
2186 
2187 	for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2188 		if (!metrics[m].selected)
2189 			continue;
2190 		for (cpu = 0; cpu < obj->rodata->num_cpu; cpu++) {
2191 			pmu_fd = syscall(__NR_perf_event_open, &metrics[m].attr,
2192 					 -1/*pid*/, cpu, -1/*group_fd*/, 0);
2193 			if (pmu_fd < 0 ||
2194 			    bpf_map_update_elem(map_fd, &profile_perf_event_cnt,
2195 						&pmu_fd, BPF_ANY) ||
2196 			    ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0)) {
2197 				p_err("failed to create event %s on cpu %d",
2198 				      metrics[m].name, cpu);
2199 				return -1;
2200 			}
2201 			profile_perf_events[profile_perf_event_cnt++] = pmu_fd;
2202 		}
2203 	}
2204 	return 0;
2205 }
2206 
profile_print_and_cleanup(void)2207 static void profile_print_and_cleanup(void)
2208 {
2209 	profile_close_perf_events(profile_obj);
2210 	profile_read_values(profile_obj);
2211 	profile_print_readings();
2212 	profiler_bpf__destroy(profile_obj);
2213 
2214 	close(profile_tgt_fd);
2215 	free(profile_tgt_name);
2216 }
2217 
int_exit(int signo)2218 static void int_exit(int signo)
2219 {
2220 	profile_print_and_cleanup();
2221 	exit(0);
2222 }
2223 
do_profile(int argc,char ** argv)2224 static int do_profile(int argc, char **argv)
2225 {
2226 	int num_metric, num_cpu, err = -1;
2227 	struct bpf_program *prog;
2228 	unsigned long duration;
2229 	char *endptr;
2230 
2231 	/* we at least need two args for the prog and one metric */
2232 	if (!REQ_ARGS(3))
2233 		return -EINVAL;
2234 
2235 	/* parse target fd */
2236 	profile_tgt_fd = prog_parse_fd(&argc, &argv);
2237 	if (profile_tgt_fd < 0) {
2238 		p_err("failed to parse fd");
2239 		return -1;
2240 	}
2241 
2242 	/* parse profiling optional duration */
2243 	if (argc > 2 && is_prefix(argv[0], "duration")) {
2244 		NEXT_ARG();
2245 		duration = strtoul(*argv, &endptr, 0);
2246 		if (*endptr)
2247 			usage();
2248 		NEXT_ARG();
2249 	} else {
2250 		duration = UINT_MAX;
2251 	}
2252 
2253 	num_metric = profile_parse_metrics(argc, argv);
2254 	if (num_metric <= 0)
2255 		goto out;
2256 
2257 	num_cpu = libbpf_num_possible_cpus();
2258 	if (num_cpu <= 0) {
2259 		p_err("failed to identify number of CPUs");
2260 		goto out;
2261 	}
2262 
2263 	profile_obj = profiler_bpf__open();
2264 	if (!profile_obj) {
2265 		p_err("failed to open and/or load BPF object");
2266 		goto out;
2267 	}
2268 
2269 	profile_obj->rodata->num_cpu = num_cpu;
2270 	profile_obj->rodata->num_metric = num_metric;
2271 
2272 	/* adjust map sizes */
2273 	bpf_map__set_max_entries(profile_obj->maps.events, num_metric * num_cpu);
2274 	bpf_map__set_max_entries(profile_obj->maps.fentry_readings, num_metric);
2275 	bpf_map__set_max_entries(profile_obj->maps.accum_readings, num_metric);
2276 	bpf_map__set_max_entries(profile_obj->maps.counts, 1);
2277 
2278 	/* change target name */
2279 	profile_tgt_name = profile_target_name(profile_tgt_fd);
2280 	if (!profile_tgt_name)
2281 		goto out;
2282 
2283 	bpf_object__for_each_program(prog, profile_obj->obj) {
2284 		err = bpf_program__set_attach_target(prog, profile_tgt_fd,
2285 						     profile_tgt_name);
2286 		if (err) {
2287 			p_err("failed to set attach target\n");
2288 			goto out;
2289 		}
2290 	}
2291 
2292 	set_max_rlimit();
2293 	err = profiler_bpf__load(profile_obj);
2294 	if (err) {
2295 		p_err("failed to load profile_obj");
2296 		goto out;
2297 	}
2298 
2299 	err = profile_open_perf_events(profile_obj);
2300 	if (err)
2301 		goto out;
2302 
2303 	err = profiler_bpf__attach(profile_obj);
2304 	if (err) {
2305 		p_err("failed to attach profile_obj");
2306 		goto out;
2307 	}
2308 	signal(SIGINT, int_exit);
2309 
2310 	sleep(duration);
2311 	profile_print_and_cleanup();
2312 	return 0;
2313 
2314 out:
2315 	profile_close_perf_events(profile_obj);
2316 	if (profile_obj)
2317 		profiler_bpf__destroy(profile_obj);
2318 	close(profile_tgt_fd);
2319 	free(profile_tgt_name);
2320 	return err;
2321 }
2322 
2323 #endif /* BPFTOOL_WITHOUT_SKELETONS */
2324 
do_help(int argc,char ** argv)2325 static int do_help(int argc, char **argv)
2326 {
2327 	if (json_output) {
2328 		jsonw_null(json_wtr);
2329 		return 0;
2330 	}
2331 
2332 	fprintf(stderr,
2333 		"Usage: %1$s %2$s { show | list } [PROG]\n"
2334 		"       %1$s %2$s dump xlated PROG [{ file FILE | opcodes | visual | linum }]\n"
2335 		"       %1$s %2$s dump jited  PROG [{ file FILE | opcodes | linum }]\n"
2336 		"       %1$s %2$s pin   PROG FILE\n"
2337 		"       %1$s %2$s { load | loadall } OBJ  PATH \\\n"
2338 		"                         [type TYPE] [dev NAME] \\\n"
2339 		"                         [map { idx IDX | name NAME } MAP]\\\n"
2340 		"                         [pinmaps MAP_DIR]\n"
2341 		"       %1$s %2$s attach PROG ATTACH_TYPE [MAP]\n"
2342 		"       %1$s %2$s detach PROG ATTACH_TYPE [MAP]\n"
2343 		"       %1$s %2$s run PROG \\\n"
2344 		"                         data_in FILE \\\n"
2345 		"                         [data_out FILE [data_size_out L]] \\\n"
2346 		"                         [ctx_in FILE [ctx_out FILE [ctx_size_out M]]] \\\n"
2347 		"                         [repeat N]\n"
2348 		"       %1$s %2$s profile PROG [duration DURATION] METRICs\n"
2349 		"       %1$s %2$s tracelog\n"
2350 		"       %1$s %2$s help\n"
2351 		"\n"
2352 		"       " HELP_SPEC_MAP "\n"
2353 		"       " HELP_SPEC_PROGRAM "\n"
2354 		"       TYPE := { socket | kprobe | kretprobe | classifier | action |\n"
2355 		"                 tracepoint | raw_tracepoint | xdp | perf_event | cgroup/skb |\n"
2356 		"                 cgroup/sock | cgroup/dev | lwt_in | lwt_out | lwt_xmit |\n"
2357 		"                 lwt_seg6local | sockops | sk_skb | sk_msg | lirc_mode2 |\n"
2358 		"                 sk_reuseport | flow_dissector | cgroup/sysctl |\n"
2359 		"                 cgroup/bind4 | cgroup/bind6 | cgroup/post_bind4 |\n"
2360 		"                 cgroup/post_bind6 | cgroup/connect4 | cgroup/connect6 |\n"
2361 		"                 cgroup/getpeername4 | cgroup/getpeername6 |\n"
2362 		"                 cgroup/getsockname4 | cgroup/getsockname6 | cgroup/sendmsg4 |\n"
2363 		"                 cgroup/sendmsg6 | cgroup/recvmsg4 | cgroup/recvmsg6 |\n"
2364 		"                 cgroup/getsockopt | cgroup/setsockopt | cgroup/sock_release |\n"
2365 		"                 struct_ops | fentry | fexit | freplace | sk_lookup }\n"
2366 		"       ATTACH_TYPE := { sk_msg_verdict | sk_skb_verdict | sk_skb_stream_verdict |\n"
2367 		"                        sk_skb_stream_parser | flow_dissector }\n"
2368 		"       METRIC := { cycles | instructions | l1d_loads | llc_misses | itlb_misses | dtlb_misses }\n"
2369 		"       " HELP_SPEC_OPTIONS " |\n"
2370 		"                    {-f|--bpffs} | {-m|--mapcompat} | {-n|--nomount} |\n"
2371 		"                    {-L|--use-loader} }\n"
2372 		"",
2373 		bin_name, argv[-2]);
2374 
2375 	return 0;
2376 }
2377 
2378 static const struct cmd cmds[] = {
2379 	{ "show",	do_show },
2380 	{ "list",	do_show },
2381 	{ "help",	do_help },
2382 	{ "dump",	do_dump },
2383 	{ "pin",	do_pin },
2384 	{ "load",	do_load },
2385 	{ "loadall",	do_loadall },
2386 	{ "attach",	do_attach },
2387 	{ "detach",	do_detach },
2388 	{ "tracelog",	do_tracelog },
2389 	{ "run",	do_run },
2390 	{ "profile",	do_profile },
2391 	{ 0 }
2392 };
2393 
do_prog(int argc,char ** argv)2394 int do_prog(int argc, char **argv)
2395 {
2396 	return cmd_select(cmds, argc, argv, do_help);
2397 }
2398