1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (C) 2017-2018 Netronome Systems, Inc. */
3
4 #ifndef _GNU_SOURCE
5 #define _GNU_SOURCE
6 #endif
7 #include <errno.h>
8 #include <fcntl.h>
9 #include <signal.h>
10 #include <stdarg.h>
11 #include <stdio.h>
12 #include <stdlib.h>
13 #include <string.h>
14 #include <time.h>
15 #include <unistd.h>
16 #include <net/if.h>
17 #include <sys/ioctl.h>
18 #include <sys/types.h>
19 #include <sys/stat.h>
20 #include <sys/syscall.h>
21 #include <dirent.h>
22
23 #include <linux/err.h>
24 #include <linux/perf_event.h>
25 #include <linux/sizes.h>
26
27 #include <bpf/bpf.h>
28 #include <bpf/btf.h>
29 #include <bpf/hashmap.h>
30 #include <bpf/libbpf.h>
31 #include <bpf/libbpf_internal.h>
32 #include <bpf/skel_internal.h>
33
34 #include "cfg.h"
35 #include "main.h"
36 #include "xlated_dumper.h"
37
38 #define BPF_METADATA_PREFIX "bpf_metadata_"
39 #define BPF_METADATA_PREFIX_LEN (sizeof(BPF_METADATA_PREFIX) - 1)
40
41 enum dump_mode {
42 DUMP_JITED,
43 DUMP_XLATED,
44 };
45
46 static const bool attach_types[] = {
47 [BPF_SK_SKB_STREAM_PARSER] = true,
48 [BPF_SK_SKB_STREAM_VERDICT] = true,
49 [BPF_SK_SKB_VERDICT] = true,
50 [BPF_SK_MSG_VERDICT] = true,
51 [BPF_FLOW_DISSECTOR] = true,
52 [__MAX_BPF_ATTACH_TYPE] = false,
53 };
54
55 /* Textual representations traditionally used by the program and kept around
56 * for the sake of backwards compatibility.
57 */
58 static const char * const attach_type_strings[] = {
59 [BPF_SK_SKB_STREAM_PARSER] = "stream_parser",
60 [BPF_SK_SKB_STREAM_VERDICT] = "stream_verdict",
61 [BPF_SK_SKB_VERDICT] = "skb_verdict",
62 [BPF_SK_MSG_VERDICT] = "msg_verdict",
63 [__MAX_BPF_ATTACH_TYPE] = NULL,
64 };
65
66 static struct hashmap *prog_table;
67
parse_attach_type(const char * str)68 static enum bpf_attach_type parse_attach_type(const char *str)
69 {
70 enum bpf_attach_type type;
71
72 for (type = 0; type < __MAX_BPF_ATTACH_TYPE; type++) {
73 if (attach_types[type]) {
74 const char *attach_type_str;
75
76 attach_type_str = libbpf_bpf_attach_type_str(type);
77 if (!strcmp(str, attach_type_str))
78 return type;
79 }
80
81 if (attach_type_strings[type] &&
82 is_prefix(str, attach_type_strings[type]))
83 return type;
84 }
85
86 return __MAX_BPF_ATTACH_TYPE;
87 }
88
prep_prog_info(struct bpf_prog_info * const info,enum dump_mode mode,void ** info_data,size_t * const info_data_sz)89 static int prep_prog_info(struct bpf_prog_info *const info, enum dump_mode mode,
90 void **info_data, size_t *const info_data_sz)
91 {
92 struct bpf_prog_info holder = {};
93 size_t needed = 0;
94 void *ptr;
95
96 if (mode == DUMP_JITED) {
97 holder.jited_prog_len = info->jited_prog_len;
98 needed += info->jited_prog_len;
99 } else {
100 holder.xlated_prog_len = info->xlated_prog_len;
101 needed += info->xlated_prog_len;
102 }
103
104 holder.nr_jited_ksyms = info->nr_jited_ksyms;
105 needed += info->nr_jited_ksyms * sizeof(__u64);
106
107 holder.nr_jited_func_lens = info->nr_jited_func_lens;
108 needed += info->nr_jited_func_lens * sizeof(__u32);
109
110 holder.nr_func_info = info->nr_func_info;
111 holder.func_info_rec_size = info->func_info_rec_size;
112 needed += info->nr_func_info * info->func_info_rec_size;
113
114 holder.nr_line_info = info->nr_line_info;
115 holder.line_info_rec_size = info->line_info_rec_size;
116 needed += info->nr_line_info * info->line_info_rec_size;
117
118 holder.nr_jited_line_info = info->nr_jited_line_info;
119 holder.jited_line_info_rec_size = info->jited_line_info_rec_size;
120 needed += info->nr_jited_line_info * info->jited_line_info_rec_size;
121
122 if (needed > *info_data_sz) {
123 ptr = realloc(*info_data, needed);
124 if (!ptr)
125 return -1;
126
127 *info_data = ptr;
128 *info_data_sz = needed;
129 }
130 ptr = *info_data;
131
132 if (mode == DUMP_JITED) {
133 holder.jited_prog_insns = ptr_to_u64(ptr);
134 ptr += holder.jited_prog_len;
135 } else {
136 holder.xlated_prog_insns = ptr_to_u64(ptr);
137 ptr += holder.xlated_prog_len;
138 }
139
140 holder.jited_ksyms = ptr_to_u64(ptr);
141 ptr += holder.nr_jited_ksyms * sizeof(__u64);
142
143 holder.jited_func_lens = ptr_to_u64(ptr);
144 ptr += holder.nr_jited_func_lens * sizeof(__u32);
145
146 holder.func_info = ptr_to_u64(ptr);
147 ptr += holder.nr_func_info * holder.func_info_rec_size;
148
149 holder.line_info = ptr_to_u64(ptr);
150 ptr += holder.nr_line_info * holder.line_info_rec_size;
151
152 holder.jited_line_info = ptr_to_u64(ptr);
153 ptr += holder.nr_jited_line_info * holder.jited_line_info_rec_size;
154
155 *info = holder;
156 return 0;
157 }
158
print_boot_time(__u64 nsecs,char * buf,unsigned int size)159 static void print_boot_time(__u64 nsecs, char *buf, unsigned int size)
160 {
161 struct timespec real_time_ts, boot_time_ts;
162 time_t wallclock_secs;
163 struct tm load_tm;
164
165 buf[--size] = '\0';
166
167 if (clock_gettime(CLOCK_REALTIME, &real_time_ts) ||
168 clock_gettime(CLOCK_BOOTTIME, &boot_time_ts)) {
169 perror("Can't read clocks");
170 snprintf(buf, size, "%llu", nsecs / 1000000000);
171 return;
172 }
173
174 wallclock_secs = (real_time_ts.tv_sec - boot_time_ts.tv_sec) +
175 (real_time_ts.tv_nsec - boot_time_ts.tv_nsec + nsecs) /
176 1000000000;
177
178
179 if (!localtime_r(&wallclock_secs, &load_tm)) {
180 snprintf(buf, size, "%llu", nsecs / 1000000000);
181 return;
182 }
183
184 if (json_output)
185 strftime(buf, size, "%s", &load_tm);
186 else
187 strftime(buf, size, "%FT%T%z", &load_tm);
188 }
189
show_prog_maps(int fd,__u32 num_maps)190 static void show_prog_maps(int fd, __u32 num_maps)
191 {
192 struct bpf_prog_info info = {};
193 __u32 len = sizeof(info);
194 __u32 map_ids[num_maps];
195 unsigned int i;
196 int err;
197
198 info.nr_map_ids = num_maps;
199 info.map_ids = ptr_to_u64(map_ids);
200
201 err = bpf_prog_get_info_by_fd(fd, &info, &len);
202 if (err || !info.nr_map_ids)
203 return;
204
205 if (json_output) {
206 jsonw_name(json_wtr, "map_ids");
207 jsonw_start_array(json_wtr);
208 for (i = 0; i < info.nr_map_ids; i++)
209 jsonw_uint(json_wtr, map_ids[i]);
210 jsonw_end_array(json_wtr);
211 } else {
212 printf(" map_ids ");
213 for (i = 0; i < info.nr_map_ids; i++)
214 printf("%u%s", map_ids[i],
215 i == info.nr_map_ids - 1 ? "" : ",");
216 }
217 }
218
find_metadata(int prog_fd,struct bpf_map_info * map_info)219 static void *find_metadata(int prog_fd, struct bpf_map_info *map_info)
220 {
221 struct bpf_prog_info prog_info;
222 __u32 prog_info_len;
223 __u32 map_info_len;
224 void *value = NULL;
225 __u32 *map_ids;
226 int nr_maps;
227 int key = 0;
228 int map_fd;
229 int ret;
230 __u32 i;
231
232 memset(&prog_info, 0, sizeof(prog_info));
233 prog_info_len = sizeof(prog_info);
234 ret = bpf_prog_get_info_by_fd(prog_fd, &prog_info, &prog_info_len);
235 if (ret)
236 return NULL;
237
238 if (!prog_info.nr_map_ids)
239 return NULL;
240
241 map_ids = calloc(prog_info.nr_map_ids, sizeof(__u32));
242 if (!map_ids)
243 return NULL;
244
245 nr_maps = prog_info.nr_map_ids;
246 memset(&prog_info, 0, sizeof(prog_info));
247 prog_info.nr_map_ids = nr_maps;
248 prog_info.map_ids = ptr_to_u64(map_ids);
249 prog_info_len = sizeof(prog_info);
250
251 ret = bpf_prog_get_info_by_fd(prog_fd, &prog_info, &prog_info_len);
252 if (ret)
253 goto free_map_ids;
254
255 for (i = 0; i < prog_info.nr_map_ids; i++) {
256 map_fd = bpf_map_get_fd_by_id(map_ids[i]);
257 if (map_fd < 0)
258 goto free_map_ids;
259
260 memset(map_info, 0, sizeof(*map_info));
261 map_info_len = sizeof(*map_info);
262 ret = bpf_map_get_info_by_fd(map_fd, map_info, &map_info_len);
263 if (ret < 0) {
264 close(map_fd);
265 goto free_map_ids;
266 }
267
268 if (map_info->type != BPF_MAP_TYPE_ARRAY ||
269 map_info->key_size != sizeof(int) ||
270 map_info->max_entries != 1 ||
271 !map_info->btf_value_type_id ||
272 !strstr(map_info->name, ".rodata")) {
273 close(map_fd);
274 continue;
275 }
276
277 value = malloc(map_info->value_size);
278 if (!value) {
279 close(map_fd);
280 goto free_map_ids;
281 }
282
283 if (bpf_map_lookup_elem(map_fd, &key, value)) {
284 close(map_fd);
285 free(value);
286 value = NULL;
287 goto free_map_ids;
288 }
289
290 close(map_fd);
291 break;
292 }
293
294 free_map_ids:
295 free(map_ids);
296 return value;
297 }
298
has_metadata_prefix(const char * s)299 static bool has_metadata_prefix(const char *s)
300 {
301 return strncmp(s, BPF_METADATA_PREFIX, BPF_METADATA_PREFIX_LEN) == 0;
302 }
303
show_prog_metadata(int fd,__u32 num_maps)304 static void show_prog_metadata(int fd, __u32 num_maps)
305 {
306 const struct btf_type *t_datasec, *t_var;
307 struct bpf_map_info map_info;
308 struct btf_var_secinfo *vsi;
309 bool printed_header = false;
310 unsigned int i, vlen;
311 void *value = NULL;
312 const char *name;
313 struct btf *btf;
314 int err;
315
316 if (!num_maps)
317 return;
318
319 memset(&map_info, 0, sizeof(map_info));
320 value = find_metadata(fd, &map_info);
321 if (!value)
322 return;
323
324 btf = btf__load_from_kernel_by_id(map_info.btf_id);
325 if (!btf)
326 goto out_free;
327
328 t_datasec = btf__type_by_id(btf, map_info.btf_value_type_id);
329 if (!btf_is_datasec(t_datasec))
330 goto out_free;
331
332 vlen = btf_vlen(t_datasec);
333 vsi = btf_var_secinfos(t_datasec);
334
335 /* We don't proceed to check the kinds of the elements of the DATASEC.
336 * The verifier enforces them to be BTF_KIND_VAR.
337 */
338
339 if (json_output) {
340 struct btf_dumper d = {
341 .btf = btf,
342 .jw = json_wtr,
343 .is_plain_text = false,
344 };
345
346 for (i = 0; i < vlen; i++, vsi++) {
347 t_var = btf__type_by_id(btf, vsi->type);
348 name = btf__name_by_offset(btf, t_var->name_off);
349
350 if (!has_metadata_prefix(name))
351 continue;
352
353 if (!printed_header) {
354 jsonw_name(json_wtr, "metadata");
355 jsonw_start_object(json_wtr);
356 printed_header = true;
357 }
358
359 jsonw_name(json_wtr, name + BPF_METADATA_PREFIX_LEN);
360 err = btf_dumper_type(&d, t_var->type, value + vsi->offset);
361 if (err) {
362 p_err("btf dump failed: %d", err);
363 break;
364 }
365 }
366 if (printed_header)
367 jsonw_end_object(json_wtr);
368 } else {
369 json_writer_t *btf_wtr;
370 struct btf_dumper d = {
371 .btf = btf,
372 .is_plain_text = true,
373 };
374
375 for (i = 0; i < vlen; i++, vsi++) {
376 t_var = btf__type_by_id(btf, vsi->type);
377 name = btf__name_by_offset(btf, t_var->name_off);
378
379 if (!has_metadata_prefix(name))
380 continue;
381
382 if (!printed_header) {
383 printf("\tmetadata:");
384
385 btf_wtr = jsonw_new(stdout);
386 if (!btf_wtr) {
387 p_err("jsonw alloc failed");
388 goto out_free;
389 }
390 d.jw = btf_wtr,
391
392 printed_header = true;
393 }
394
395 printf("\n\t\t%s = ", name + BPF_METADATA_PREFIX_LEN);
396
397 jsonw_reset(btf_wtr);
398 err = btf_dumper_type(&d, t_var->type, value + vsi->offset);
399 if (err) {
400 p_err("btf dump failed: %d", err);
401 break;
402 }
403 }
404 if (printed_header)
405 jsonw_destroy(&btf_wtr);
406 }
407
408 out_free:
409 btf__free(btf);
410 free(value);
411 }
412
print_prog_header_json(struct bpf_prog_info * info,int fd)413 static void print_prog_header_json(struct bpf_prog_info *info, int fd)
414 {
415 const char *prog_type_str;
416 char prog_name[MAX_PROG_FULL_NAME];
417
418 jsonw_uint_field(json_wtr, "id", info->id);
419 prog_type_str = libbpf_bpf_prog_type_str(info->type);
420
421 if (prog_type_str)
422 jsonw_string_field(json_wtr, "type", prog_type_str);
423 else
424 jsonw_uint_field(json_wtr, "type", info->type);
425
426 if (*info->name) {
427 get_prog_full_name(info, fd, prog_name, sizeof(prog_name));
428 jsonw_string_field(json_wtr, "name", prog_name);
429 }
430
431 jsonw_name(json_wtr, "tag");
432 jsonw_printf(json_wtr, "\"" BPF_TAG_FMT "\"",
433 info->tag[0], info->tag[1], info->tag[2], info->tag[3],
434 info->tag[4], info->tag[5], info->tag[6], info->tag[7]);
435
436 jsonw_bool_field(json_wtr, "gpl_compatible", info->gpl_compatible);
437 if (info->run_time_ns) {
438 jsonw_uint_field(json_wtr, "run_time_ns", info->run_time_ns);
439 jsonw_uint_field(json_wtr, "run_cnt", info->run_cnt);
440 }
441 if (info->recursion_misses)
442 jsonw_uint_field(json_wtr, "recursion_misses", info->recursion_misses);
443 }
444
print_prog_json(struct bpf_prog_info * info,int fd)445 static void print_prog_json(struct bpf_prog_info *info, int fd)
446 {
447 char *memlock;
448
449 jsonw_start_object(json_wtr);
450 print_prog_header_json(info, fd);
451 print_dev_json(info->ifindex, info->netns_dev, info->netns_ino);
452
453 if (info->load_time) {
454 char buf[32];
455
456 print_boot_time(info->load_time, buf, sizeof(buf));
457
458 /* Piggy back on load_time, since 0 uid is a valid one */
459 jsonw_name(json_wtr, "loaded_at");
460 jsonw_printf(json_wtr, "%s", buf);
461 jsonw_uint_field(json_wtr, "uid", info->created_by_uid);
462 }
463
464 jsonw_uint_field(json_wtr, "bytes_xlated", info->xlated_prog_len);
465
466 if (info->jited_prog_len) {
467 jsonw_bool_field(json_wtr, "jited", true);
468 jsonw_uint_field(json_wtr, "bytes_jited", info->jited_prog_len);
469 } else {
470 jsonw_bool_field(json_wtr, "jited", false);
471 }
472
473 memlock = get_fdinfo(fd, "memlock");
474 if (memlock)
475 jsonw_int_field(json_wtr, "bytes_memlock", atoll(memlock));
476 free(memlock);
477
478 if (info->nr_map_ids)
479 show_prog_maps(fd, info->nr_map_ids);
480
481 if (info->btf_id)
482 jsonw_int_field(json_wtr, "btf_id", info->btf_id);
483
484 if (!hashmap__empty(prog_table)) {
485 struct hashmap_entry *entry;
486
487 jsonw_name(json_wtr, "pinned");
488 jsonw_start_array(json_wtr);
489 hashmap__for_each_key_entry(prog_table, entry, info->id)
490 jsonw_string(json_wtr, entry->pvalue);
491 jsonw_end_array(json_wtr);
492 }
493
494 emit_obj_refs_json(refs_table, info->id, json_wtr);
495
496 show_prog_metadata(fd, info->nr_map_ids);
497
498 jsonw_end_object(json_wtr);
499 }
500
print_prog_header_plain(struct bpf_prog_info * info,int fd)501 static void print_prog_header_plain(struct bpf_prog_info *info, int fd)
502 {
503 const char *prog_type_str;
504 char prog_name[MAX_PROG_FULL_NAME];
505
506 printf("%u: ", info->id);
507 prog_type_str = libbpf_bpf_prog_type_str(info->type);
508 if (prog_type_str)
509 printf("%s ", prog_type_str);
510 else
511 printf("type %u ", info->type);
512
513 if (*info->name) {
514 get_prog_full_name(info, fd, prog_name, sizeof(prog_name));
515 printf("name %s ", prog_name);
516 }
517
518 printf("tag ");
519 fprint_hex(stdout, info->tag, BPF_TAG_SIZE, "");
520 print_dev_plain(info->ifindex, info->netns_dev, info->netns_ino);
521 printf("%s", info->gpl_compatible ? " gpl" : "");
522 if (info->run_time_ns)
523 printf(" run_time_ns %lld run_cnt %lld",
524 info->run_time_ns, info->run_cnt);
525 if (info->recursion_misses)
526 printf(" recursion_misses %lld", info->recursion_misses);
527 printf("\n");
528 }
529
print_prog_plain(struct bpf_prog_info * info,int fd)530 static void print_prog_plain(struct bpf_prog_info *info, int fd)
531 {
532 char *memlock;
533
534 print_prog_header_plain(info, fd);
535
536 if (info->load_time) {
537 char buf[32];
538
539 print_boot_time(info->load_time, buf, sizeof(buf));
540
541 /* Piggy back on load_time, since 0 uid is a valid one */
542 printf("\tloaded_at %s uid %u\n", buf, info->created_by_uid);
543 }
544
545 printf("\txlated %uB", info->xlated_prog_len);
546
547 if (info->jited_prog_len)
548 printf(" jited %uB", info->jited_prog_len);
549 else
550 printf(" not jited");
551
552 memlock = get_fdinfo(fd, "memlock");
553 if (memlock)
554 printf(" memlock %sB", memlock);
555 free(memlock);
556
557 if (info->nr_map_ids)
558 show_prog_maps(fd, info->nr_map_ids);
559
560 if (!hashmap__empty(prog_table)) {
561 struct hashmap_entry *entry;
562
563 hashmap__for_each_key_entry(prog_table, entry, info->id)
564 printf("\n\tpinned %s", (char *)entry->pvalue);
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_prog_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 (!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_arch(info->ifindex, info->netns_dev,
766 info->netns_ino, &disasm_opt);
767 if (!name)
768 goto exit_free;
769 }
770
771 if (info->nr_jited_func_lens && info->jited_func_lens) {
772 struct kernel_sym *sym = NULL;
773 struct bpf_func_info *record;
774 char sym_name[SYM_MAX_NAME];
775 unsigned char *img = buf;
776 __u64 *ksyms = NULL;
777 __u32 *lens;
778 __u32 i;
779 if (info->nr_jited_ksyms) {
780 kernel_syms_load(&dd);
781 ksyms = u64_to_ptr(info->jited_ksyms);
782 }
783
784 if (json_output)
785 jsonw_start_array(json_wtr);
786
787 lens = u64_to_ptr(info->jited_func_lens);
788 for (i = 0; i < info->nr_jited_func_lens; i++) {
789 if (ksyms) {
790 sym = kernel_syms_search(&dd, ksyms[i]);
791 if (sym)
792 sprintf(sym_name, "%s", sym->name);
793 else
794 sprintf(sym_name, "0x%016llx", ksyms[i]);
795 } else {
796 strcpy(sym_name, "unknown");
797 }
798
799 if (func_info) {
800 record = func_info + i * info->func_info_rec_size;
801 btf_dumper_type_only(btf, record->type_id,
802 func_sig,
803 sizeof(func_sig));
804 }
805
806 if (json_output) {
807 jsonw_start_object(json_wtr);
808 if (func_info && func_sig[0] != '\0') {
809 jsonw_name(json_wtr, "proto");
810 jsonw_string(json_wtr, func_sig);
811 }
812 jsonw_name(json_wtr, "name");
813 jsonw_string(json_wtr, sym_name);
814 jsonw_name(json_wtr, "insns");
815 } else {
816 if (func_info && func_sig[0] != '\0')
817 printf("%s:\n", func_sig);
818 printf("%s:\n", sym_name);
819 }
820
821 if (ksyms) {
822 if (disasm_print_insn(img, lens[i], opcodes,
823 name, disasm_opt, btf,
824 prog_linfo, ksyms[i], i,
825 linum))
826 goto exit_free;
827 } else {
828 if (disasm_print_insn(img, lens[i], opcodes,
829 name, disasm_opt, btf,
830 NULL, 0, 0, false))
831 goto exit_free;
832 }
833
834 img += lens[i];
835
836 if (json_output)
837 jsonw_end_object(json_wtr);
838 else
839 printf("\n");
840 }
841
842 if (json_output)
843 jsonw_end_array(json_wtr);
844 } else {
845 if (disasm_print_insn(buf, member_len, opcodes, name,
846 disasm_opt, btf, NULL, 0, 0,
847 false))
848 goto exit_free;
849 }
850 } else {
851 kernel_syms_load(&dd);
852 dd.nr_jited_ksyms = info->nr_jited_ksyms;
853 dd.jited_ksyms = u64_to_ptr(info->jited_ksyms);
854 dd.btf = btf;
855 dd.func_info = func_info;
856 dd.finfo_rec_size = info->func_info_rec_size;
857 dd.prog_linfo = prog_linfo;
858
859 if (json_output)
860 dump_xlated_json(&dd, buf, member_len, opcodes, linum);
861 else if (visual)
862 dump_xlated_cfg(&dd, buf, member_len, opcodes, linum);
863 else
864 dump_xlated_plain(&dd, buf, member_len, opcodes, linum);
865 kernel_syms_destroy(&dd);
866 }
867
868 err = 0;
869
870 exit_free:
871 btf__free(btf);
872 bpf_prog_linfo__free(prog_linfo);
873 return err;
874 }
875
do_dump(int argc,char ** argv)876 static int do_dump(int argc, char **argv)
877 {
878 struct bpf_prog_info info;
879 __u32 info_len = sizeof(info);
880 size_t info_data_sz = 0;
881 void *info_data = NULL;
882 char *filepath = NULL;
883 bool opcodes = false;
884 bool visual = false;
885 enum dump_mode mode;
886 bool linum = false;
887 int nb_fds, i = 0;
888 int *fds = NULL;
889 int err = -1;
890
891 if (is_prefix(*argv, "jited")) {
892 if (disasm_init())
893 return -1;
894 mode = DUMP_JITED;
895 } else if (is_prefix(*argv, "xlated")) {
896 mode = DUMP_XLATED;
897 } else {
898 p_err("expected 'xlated' or 'jited', got: %s", *argv);
899 return -1;
900 }
901 NEXT_ARG();
902
903 if (argc < 2)
904 usage();
905
906 fds = malloc(sizeof(int));
907 if (!fds) {
908 p_err("mem alloc failed");
909 return -1;
910 }
911 nb_fds = prog_parse_fds(&argc, &argv, &fds);
912 if (nb_fds < 1)
913 goto exit_free;
914
915 while (argc) {
916 if (is_prefix(*argv, "file")) {
917 NEXT_ARG();
918 if (!argc) {
919 p_err("expected file path");
920 goto exit_close;
921 }
922 if (nb_fds > 1) {
923 p_err("several programs matched");
924 goto exit_close;
925 }
926
927 filepath = *argv;
928 NEXT_ARG();
929 } else if (is_prefix(*argv, "opcodes")) {
930 opcodes = true;
931 NEXT_ARG();
932 } else if (is_prefix(*argv, "visual")) {
933 if (nb_fds > 1) {
934 p_err("several programs matched");
935 goto exit_close;
936 }
937
938 visual = true;
939 NEXT_ARG();
940 } else if (is_prefix(*argv, "linum")) {
941 linum = true;
942 NEXT_ARG();
943 } else {
944 usage();
945 goto exit_close;
946 }
947 }
948
949 if (filepath && (opcodes || visual || linum)) {
950 p_err("'file' is not compatible with 'opcodes', 'visual', or 'linum'");
951 goto exit_close;
952 }
953 if (json_output && visual) {
954 p_err("'visual' is not compatible with JSON output");
955 goto exit_close;
956 }
957
958 if (json_output && nb_fds > 1)
959 jsonw_start_array(json_wtr); /* root array */
960 for (i = 0; i < nb_fds; i++) {
961 memset(&info, 0, sizeof(info));
962
963 err = bpf_prog_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 err = prep_prog_info(&info, mode, &info_data, &info_data_sz);
970 if (err) {
971 p_err("can't grow prog info_data");
972 break;
973 }
974
975 err = bpf_prog_get_info_by_fd(fds[i], &info, &info_len);
976 if (err) {
977 p_err("can't get prog info: %s", strerror(errno));
978 break;
979 }
980
981 if (json_output && nb_fds > 1) {
982 jsonw_start_object(json_wtr); /* prog object */
983 print_prog_header_json(&info, fds[i]);
984 jsonw_name(json_wtr, "insns");
985 } else if (nb_fds > 1) {
986 print_prog_header_plain(&info, fds[i]);
987 }
988
989 err = prog_dump(&info, mode, filepath, opcodes, visual, linum);
990
991 if (json_output && nb_fds > 1)
992 jsonw_end_object(json_wtr); /* prog object */
993 else if (i != nb_fds - 1 && nb_fds > 1)
994 printf("\n");
995
996 if (err)
997 break;
998 close(fds[i]);
999 }
1000 if (json_output && nb_fds > 1)
1001 jsonw_end_array(json_wtr); /* root array */
1002
1003 exit_close:
1004 for (; i < nb_fds; i++)
1005 close(fds[i]);
1006 exit_free:
1007 free(info_data);
1008 free(fds);
1009 return err;
1010 }
1011
do_pin(int argc,char ** argv)1012 static int do_pin(int argc, char **argv)
1013 {
1014 int err;
1015
1016 err = do_pin_any(argc, argv, prog_parse_fd);
1017 if (!err && json_output)
1018 jsonw_null(json_wtr);
1019 return err;
1020 }
1021
1022 struct map_replace {
1023 int idx;
1024 int fd;
1025 char *name;
1026 };
1027
map_replace_compar(const void * p1,const void * p2)1028 static int map_replace_compar(const void *p1, const void *p2)
1029 {
1030 const struct map_replace *a = p1, *b = p2;
1031
1032 return a->idx - b->idx;
1033 }
1034
parse_attach_detach_args(int argc,char ** argv,int * progfd,enum bpf_attach_type * attach_type,int * mapfd)1035 static int parse_attach_detach_args(int argc, char **argv, int *progfd,
1036 enum bpf_attach_type *attach_type,
1037 int *mapfd)
1038 {
1039 if (!REQ_ARGS(3))
1040 return -EINVAL;
1041
1042 *progfd = prog_parse_fd(&argc, &argv);
1043 if (*progfd < 0)
1044 return *progfd;
1045
1046 *attach_type = parse_attach_type(*argv);
1047 if (*attach_type == __MAX_BPF_ATTACH_TYPE) {
1048 p_err("invalid attach/detach type");
1049 return -EINVAL;
1050 }
1051
1052 if (*attach_type == BPF_FLOW_DISSECTOR) {
1053 *mapfd = 0;
1054 return 0;
1055 }
1056
1057 NEXT_ARG();
1058 if (!REQ_ARGS(2))
1059 return -EINVAL;
1060
1061 *mapfd = map_parse_fd(&argc, &argv);
1062 if (*mapfd < 0)
1063 return *mapfd;
1064
1065 return 0;
1066 }
1067
do_attach(int argc,char ** argv)1068 static int do_attach(int argc, char **argv)
1069 {
1070 enum bpf_attach_type attach_type;
1071 int err, progfd;
1072 int mapfd;
1073
1074 err = parse_attach_detach_args(argc, argv,
1075 &progfd, &attach_type, &mapfd);
1076 if (err)
1077 return err;
1078
1079 err = bpf_prog_attach(progfd, mapfd, attach_type, 0);
1080 if (err) {
1081 p_err("failed prog attach to map");
1082 return -EINVAL;
1083 }
1084
1085 if (json_output)
1086 jsonw_null(json_wtr);
1087 return 0;
1088 }
1089
do_detach(int argc,char ** argv)1090 static int do_detach(int argc, char **argv)
1091 {
1092 enum bpf_attach_type attach_type;
1093 int err, progfd;
1094 int mapfd;
1095
1096 err = parse_attach_detach_args(argc, argv,
1097 &progfd, &attach_type, &mapfd);
1098 if (err)
1099 return err;
1100
1101 err = bpf_prog_detach2(progfd, mapfd, attach_type);
1102 if (err) {
1103 p_err("failed prog detach from map");
1104 return -EINVAL;
1105 }
1106
1107 if (json_output)
1108 jsonw_null(json_wtr);
1109 return 0;
1110 }
1111
check_single_stdin(char * file_data_in,char * file_ctx_in)1112 static int check_single_stdin(char *file_data_in, char *file_ctx_in)
1113 {
1114 if (file_data_in && file_ctx_in &&
1115 !strcmp(file_data_in, "-") && !strcmp(file_ctx_in, "-")) {
1116 p_err("cannot use standard input for both data_in and ctx_in");
1117 return -1;
1118 }
1119
1120 return 0;
1121 }
1122
get_run_data(const char * fname,void ** data_ptr,unsigned int * size)1123 static int get_run_data(const char *fname, void **data_ptr, unsigned int *size)
1124 {
1125 size_t block_size = 256;
1126 size_t buf_size = block_size;
1127 size_t nb_read = 0;
1128 void *tmp;
1129 FILE *f;
1130
1131 if (!fname) {
1132 *data_ptr = NULL;
1133 *size = 0;
1134 return 0;
1135 }
1136
1137 if (!strcmp(fname, "-"))
1138 f = stdin;
1139 else
1140 f = fopen(fname, "r");
1141 if (!f) {
1142 p_err("failed to open %s: %s", fname, strerror(errno));
1143 return -1;
1144 }
1145
1146 *data_ptr = malloc(block_size);
1147 if (!*data_ptr) {
1148 p_err("failed to allocate memory for data_in/ctx_in: %s",
1149 strerror(errno));
1150 goto err_fclose;
1151 }
1152
1153 while ((nb_read += fread(*data_ptr + nb_read, 1, block_size, f))) {
1154 if (feof(f))
1155 break;
1156 if (ferror(f)) {
1157 p_err("failed to read data_in/ctx_in from %s: %s",
1158 fname, strerror(errno));
1159 goto err_free;
1160 }
1161 if (nb_read > buf_size - block_size) {
1162 if (buf_size == UINT32_MAX) {
1163 p_err("data_in/ctx_in is too long (max: %d)",
1164 UINT32_MAX);
1165 goto err_free;
1166 }
1167 /* No space for fread()-ing next chunk; realloc() */
1168 buf_size *= 2;
1169 tmp = realloc(*data_ptr, buf_size);
1170 if (!tmp) {
1171 p_err("failed to reallocate data_in/ctx_in: %s",
1172 strerror(errno));
1173 goto err_free;
1174 }
1175 *data_ptr = tmp;
1176 }
1177 }
1178 if (f != stdin)
1179 fclose(f);
1180
1181 *size = nb_read;
1182 return 0;
1183
1184 err_free:
1185 free(*data_ptr);
1186 *data_ptr = NULL;
1187 err_fclose:
1188 if (f != stdin)
1189 fclose(f);
1190 return -1;
1191 }
1192
hex_print(void * data,unsigned int size,FILE * f)1193 static void hex_print(void *data, unsigned int size, FILE *f)
1194 {
1195 size_t i, j;
1196 char c;
1197
1198 for (i = 0; i < size; i += 16) {
1199 /* Row offset */
1200 fprintf(f, "%07zx\t", i);
1201
1202 /* Hexadecimal values */
1203 for (j = i; j < i + 16 && j < size; j++)
1204 fprintf(f, "%02x%s", *(uint8_t *)(data + j),
1205 j % 2 ? " " : "");
1206 for (; j < i + 16; j++)
1207 fprintf(f, " %s", j % 2 ? " " : "");
1208
1209 /* ASCII values (if relevant), '.' otherwise */
1210 fprintf(f, "| ");
1211 for (j = i; j < i + 16 && j < size; j++) {
1212 c = *(char *)(data + j);
1213 if (c < ' ' || c > '~')
1214 c = '.';
1215 fprintf(f, "%c%s", c, j == i + 7 ? " " : "");
1216 }
1217
1218 fprintf(f, "\n");
1219 }
1220 }
1221
1222 static int
print_run_output(void * data,unsigned int size,const char * fname,const char * json_key)1223 print_run_output(void *data, unsigned int size, const char *fname,
1224 const char *json_key)
1225 {
1226 size_t nb_written;
1227 FILE *f;
1228
1229 if (!fname)
1230 return 0;
1231
1232 if (!strcmp(fname, "-")) {
1233 f = stdout;
1234 if (json_output) {
1235 jsonw_name(json_wtr, json_key);
1236 print_data_json(data, size);
1237 } else {
1238 hex_print(data, size, f);
1239 }
1240 return 0;
1241 }
1242
1243 f = fopen(fname, "w");
1244 if (!f) {
1245 p_err("failed to open %s: %s", fname, strerror(errno));
1246 return -1;
1247 }
1248
1249 nb_written = fwrite(data, 1, size, f);
1250 fclose(f);
1251 if (nb_written != size) {
1252 p_err("failed to write output data/ctx: %s", strerror(errno));
1253 return -1;
1254 }
1255
1256 return 0;
1257 }
1258
alloc_run_data(void ** data_ptr,unsigned int size_out)1259 static int alloc_run_data(void **data_ptr, unsigned int size_out)
1260 {
1261 *data_ptr = calloc(size_out, 1);
1262 if (!*data_ptr) {
1263 p_err("failed to allocate memory for output data/ctx: %s",
1264 strerror(errno));
1265 return -1;
1266 }
1267
1268 return 0;
1269 }
1270
do_run(int argc,char ** argv)1271 static int do_run(int argc, char **argv)
1272 {
1273 char *data_fname_in = NULL, *data_fname_out = NULL;
1274 char *ctx_fname_in = NULL, *ctx_fname_out = NULL;
1275 const unsigned int default_size = SZ_32K;
1276 void *data_in = NULL, *data_out = NULL;
1277 void *ctx_in = NULL, *ctx_out = NULL;
1278 unsigned int repeat = 1;
1279 int fd, err;
1280 LIBBPF_OPTS(bpf_test_run_opts, test_attr);
1281
1282 if (!REQ_ARGS(4))
1283 return -1;
1284
1285 fd = prog_parse_fd(&argc, &argv);
1286 if (fd < 0)
1287 return -1;
1288
1289 while (argc) {
1290 if (detect_common_prefix(*argv, "data_in", "data_out",
1291 "data_size_out", NULL))
1292 return -1;
1293 if (detect_common_prefix(*argv, "ctx_in", "ctx_out",
1294 "ctx_size_out", NULL))
1295 return -1;
1296
1297 if (is_prefix(*argv, "data_in")) {
1298 NEXT_ARG();
1299 if (!REQ_ARGS(1))
1300 return -1;
1301
1302 data_fname_in = GET_ARG();
1303 if (check_single_stdin(data_fname_in, ctx_fname_in))
1304 return -1;
1305 } else if (is_prefix(*argv, "data_out")) {
1306 NEXT_ARG();
1307 if (!REQ_ARGS(1))
1308 return -1;
1309
1310 data_fname_out = GET_ARG();
1311 } else if (is_prefix(*argv, "data_size_out")) {
1312 char *endptr;
1313
1314 NEXT_ARG();
1315 if (!REQ_ARGS(1))
1316 return -1;
1317
1318 test_attr.data_size_out = strtoul(*argv, &endptr, 0);
1319 if (*endptr) {
1320 p_err("can't parse %s as output data size",
1321 *argv);
1322 return -1;
1323 }
1324 NEXT_ARG();
1325 } else if (is_prefix(*argv, "ctx_in")) {
1326 NEXT_ARG();
1327 if (!REQ_ARGS(1))
1328 return -1;
1329
1330 ctx_fname_in = GET_ARG();
1331 if (check_single_stdin(data_fname_in, ctx_fname_in))
1332 return -1;
1333 } else if (is_prefix(*argv, "ctx_out")) {
1334 NEXT_ARG();
1335 if (!REQ_ARGS(1))
1336 return -1;
1337
1338 ctx_fname_out = GET_ARG();
1339 } else if (is_prefix(*argv, "ctx_size_out")) {
1340 char *endptr;
1341
1342 NEXT_ARG();
1343 if (!REQ_ARGS(1))
1344 return -1;
1345
1346 test_attr.ctx_size_out = strtoul(*argv, &endptr, 0);
1347 if (*endptr) {
1348 p_err("can't parse %s as output context size",
1349 *argv);
1350 return -1;
1351 }
1352 NEXT_ARG();
1353 } else if (is_prefix(*argv, "repeat")) {
1354 char *endptr;
1355
1356 NEXT_ARG();
1357 if (!REQ_ARGS(1))
1358 return -1;
1359
1360 repeat = strtoul(*argv, &endptr, 0);
1361 if (*endptr) {
1362 p_err("can't parse %s as repeat number",
1363 *argv);
1364 return -1;
1365 }
1366 NEXT_ARG();
1367 } else {
1368 p_err("expected no more arguments, 'data_in', 'data_out', 'data_size_out', 'ctx_in', 'ctx_out', 'ctx_size_out' or 'repeat', got: '%s'?",
1369 *argv);
1370 return -1;
1371 }
1372 }
1373
1374 err = get_run_data(data_fname_in, &data_in, &test_attr.data_size_in);
1375 if (err)
1376 return -1;
1377
1378 if (data_in) {
1379 if (!test_attr.data_size_out)
1380 test_attr.data_size_out = default_size;
1381 err = alloc_run_data(&data_out, test_attr.data_size_out);
1382 if (err)
1383 goto free_data_in;
1384 }
1385
1386 err = get_run_data(ctx_fname_in, &ctx_in, &test_attr.ctx_size_in);
1387 if (err)
1388 goto free_data_out;
1389
1390 if (ctx_in) {
1391 if (!test_attr.ctx_size_out)
1392 test_attr.ctx_size_out = default_size;
1393 err = alloc_run_data(&ctx_out, test_attr.ctx_size_out);
1394 if (err)
1395 goto free_ctx_in;
1396 }
1397
1398 test_attr.repeat = repeat;
1399 test_attr.data_in = data_in;
1400 test_attr.data_out = data_out;
1401 test_attr.ctx_in = ctx_in;
1402 test_attr.ctx_out = ctx_out;
1403
1404 err = bpf_prog_test_run_opts(fd, &test_attr);
1405 if (err) {
1406 p_err("failed to run program: %s", strerror(errno));
1407 goto free_ctx_out;
1408 }
1409
1410 err = 0;
1411
1412 if (json_output)
1413 jsonw_start_object(json_wtr); /* root */
1414
1415 /* Do not exit on errors occurring when printing output data/context,
1416 * we still want to print return value and duration for program run.
1417 */
1418 if (test_attr.data_size_out)
1419 err += print_run_output(test_attr.data_out,
1420 test_attr.data_size_out,
1421 data_fname_out, "data_out");
1422 if (test_attr.ctx_size_out)
1423 err += print_run_output(test_attr.ctx_out,
1424 test_attr.ctx_size_out,
1425 ctx_fname_out, "ctx_out");
1426
1427 if (json_output) {
1428 jsonw_uint_field(json_wtr, "retval", test_attr.retval);
1429 jsonw_uint_field(json_wtr, "duration", test_attr.duration);
1430 jsonw_end_object(json_wtr); /* root */
1431 } else {
1432 fprintf(stdout, "Return value: %u, duration%s: %uns\n",
1433 test_attr.retval,
1434 repeat > 1 ? " (average)" : "", test_attr.duration);
1435 }
1436
1437 free_ctx_out:
1438 free(ctx_out);
1439 free_ctx_in:
1440 free(ctx_in);
1441 free_data_out:
1442 free(data_out);
1443 free_data_in:
1444 free(data_in);
1445
1446 return err;
1447 }
1448
1449 static int
get_prog_type_by_name(const char * name,enum bpf_prog_type * prog_type,enum bpf_attach_type * expected_attach_type)1450 get_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
1451 enum bpf_attach_type *expected_attach_type)
1452 {
1453 libbpf_print_fn_t print_backup;
1454 int ret;
1455
1456 ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type);
1457 if (!ret)
1458 return ret;
1459
1460 /* libbpf_prog_type_by_name() failed, let's re-run with debug level */
1461 print_backup = libbpf_set_print(print_all_levels);
1462 ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type);
1463 libbpf_set_print(print_backup);
1464
1465 return ret;
1466 }
1467
1468 static int
auto_attach_program(struct bpf_program * prog,const char * path)1469 auto_attach_program(struct bpf_program *prog, const char *path)
1470 {
1471 struct bpf_link *link;
1472 int err;
1473
1474 link = bpf_program__attach(prog);
1475 if (!link) {
1476 p_info("Program %s does not support autoattach, falling back to pinning",
1477 bpf_program__name(prog));
1478 return bpf_obj_pin(bpf_program__fd(prog), path);
1479 }
1480
1481 err = bpf_link__pin(link, path);
1482 bpf_link__destroy(link);
1483 return err;
1484 }
1485
1486 static int
auto_attach_programs(struct bpf_object * obj,const char * path)1487 auto_attach_programs(struct bpf_object *obj, const char *path)
1488 {
1489 struct bpf_program *prog;
1490 char buf[PATH_MAX];
1491 int err;
1492
1493 bpf_object__for_each_program(prog, obj) {
1494 err = pathname_concat(buf, sizeof(buf), path, bpf_program__name(prog));
1495 if (err)
1496 goto err_unpin_programs;
1497
1498 err = auto_attach_program(prog, buf);
1499 if (err)
1500 goto err_unpin_programs;
1501 }
1502
1503 return 0;
1504
1505 err_unpin_programs:
1506 while ((prog = bpf_object__prev_program(obj, prog))) {
1507 if (pathname_concat(buf, sizeof(buf), path, bpf_program__name(prog)))
1508 continue;
1509
1510 bpf_program__unpin(prog, buf);
1511 }
1512
1513 return err;
1514 }
1515
load_with_options(int argc,char ** argv,bool first_prog_only)1516 static int load_with_options(int argc, char **argv, bool first_prog_only)
1517 {
1518 enum bpf_prog_type common_prog_type = BPF_PROG_TYPE_UNSPEC;
1519 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts,
1520 .relaxed_maps = relaxed_maps,
1521 );
1522 enum bpf_attach_type expected_attach_type;
1523 struct map_replace *map_replace = NULL;
1524 struct bpf_program *prog = NULL, *pos;
1525 unsigned int old_map_fds = 0;
1526 const char *pinmaps = NULL;
1527 __u32 xdpmeta_ifindex = 0;
1528 __u32 offload_ifindex = 0;
1529 bool auto_attach = false;
1530 struct bpf_object *obj;
1531 struct bpf_map *map;
1532 const char *pinfile;
1533 unsigned int i, j;
1534 const char *file;
1535 int idx, err;
1536
1537
1538 if (!REQ_ARGS(2))
1539 return -1;
1540 file = GET_ARG();
1541 pinfile = GET_ARG();
1542
1543 while (argc) {
1544 if (is_prefix(*argv, "type")) {
1545 NEXT_ARG();
1546
1547 if (common_prog_type != BPF_PROG_TYPE_UNSPEC) {
1548 p_err("program type already specified");
1549 goto err_free_reuse_maps;
1550 }
1551 if (!REQ_ARGS(1))
1552 goto err_free_reuse_maps;
1553
1554 err = libbpf_prog_type_by_name(*argv, &common_prog_type,
1555 &expected_attach_type);
1556 if (err < 0) {
1557 /* Put a '/' at the end of type to appease libbpf */
1558 char *type = malloc(strlen(*argv) + 2);
1559
1560 if (!type) {
1561 p_err("mem alloc failed");
1562 goto err_free_reuse_maps;
1563 }
1564 *type = 0;
1565 strcat(type, *argv);
1566 strcat(type, "/");
1567
1568 err = get_prog_type_by_name(type, &common_prog_type,
1569 &expected_attach_type);
1570 free(type);
1571 if (err < 0)
1572 goto err_free_reuse_maps;
1573 }
1574
1575 NEXT_ARG();
1576 } else if (is_prefix(*argv, "map")) {
1577 void *new_map_replace;
1578 char *endptr, *name;
1579 int fd;
1580
1581 NEXT_ARG();
1582
1583 if (!REQ_ARGS(4))
1584 goto err_free_reuse_maps;
1585
1586 if (is_prefix(*argv, "idx")) {
1587 NEXT_ARG();
1588
1589 idx = strtoul(*argv, &endptr, 0);
1590 if (*endptr) {
1591 p_err("can't parse %s as IDX", *argv);
1592 goto err_free_reuse_maps;
1593 }
1594 name = NULL;
1595 } else if (is_prefix(*argv, "name")) {
1596 NEXT_ARG();
1597
1598 name = *argv;
1599 idx = -1;
1600 } else {
1601 p_err("expected 'idx' or 'name', got: '%s'?",
1602 *argv);
1603 goto err_free_reuse_maps;
1604 }
1605 NEXT_ARG();
1606
1607 fd = map_parse_fd(&argc, &argv);
1608 if (fd < 0)
1609 goto err_free_reuse_maps;
1610
1611 new_map_replace = libbpf_reallocarray(map_replace,
1612 old_map_fds + 1,
1613 sizeof(*map_replace));
1614 if (!new_map_replace) {
1615 p_err("mem alloc failed");
1616 goto err_free_reuse_maps;
1617 }
1618 map_replace = new_map_replace;
1619
1620 map_replace[old_map_fds].idx = idx;
1621 map_replace[old_map_fds].name = name;
1622 map_replace[old_map_fds].fd = fd;
1623 old_map_fds++;
1624 } else if (is_prefix(*argv, "dev")) {
1625 p_info("Warning: 'bpftool prog load [...] dev <ifname>' syntax is deprecated.\n"
1626 "Going further, please use 'offload_dev <ifname>' to offload program to device.\n"
1627 "For applications using XDP hints only, use 'xdpmeta_dev <ifname>'.");
1628 goto offload_dev;
1629 } else if (is_prefix(*argv, "offload_dev")) {
1630 offload_dev:
1631 NEXT_ARG();
1632
1633 if (offload_ifindex) {
1634 p_err("offload_dev already specified");
1635 goto err_free_reuse_maps;
1636 } else if (xdpmeta_ifindex) {
1637 p_err("xdpmeta_dev and offload_dev are mutually exclusive");
1638 goto err_free_reuse_maps;
1639 }
1640 if (!REQ_ARGS(1))
1641 goto err_free_reuse_maps;
1642
1643 offload_ifindex = if_nametoindex(*argv);
1644 if (!offload_ifindex) {
1645 p_err("unrecognized netdevice '%s': %s",
1646 *argv, strerror(errno));
1647 goto err_free_reuse_maps;
1648 }
1649 NEXT_ARG();
1650 } else if (is_prefix(*argv, "xdpmeta_dev")) {
1651 NEXT_ARG();
1652
1653 if (xdpmeta_ifindex) {
1654 p_err("xdpmeta_dev already specified");
1655 goto err_free_reuse_maps;
1656 } else if (offload_ifindex) {
1657 p_err("xdpmeta_dev and offload_dev are mutually exclusive");
1658 goto err_free_reuse_maps;
1659 }
1660 if (!REQ_ARGS(1))
1661 goto err_free_reuse_maps;
1662
1663 xdpmeta_ifindex = if_nametoindex(*argv);
1664 if (!xdpmeta_ifindex) {
1665 p_err("unrecognized netdevice '%s': %s",
1666 *argv, strerror(errno));
1667 goto err_free_reuse_maps;
1668 }
1669 NEXT_ARG();
1670 } else if (is_prefix(*argv, "pinmaps")) {
1671 NEXT_ARG();
1672
1673 if (!REQ_ARGS(1))
1674 goto err_free_reuse_maps;
1675
1676 pinmaps = GET_ARG();
1677 } else if (is_prefix(*argv, "autoattach")) {
1678 auto_attach = true;
1679 NEXT_ARG();
1680 } else {
1681 p_err("expected no more arguments, 'type', 'map' or 'dev', got: '%s'?",
1682 *argv);
1683 goto err_free_reuse_maps;
1684 }
1685 }
1686
1687 set_max_rlimit();
1688
1689 if (verifier_logs)
1690 /* log_level1 + log_level2 + stats, but not stable UAPI */
1691 open_opts.kernel_log_level = 1 + 2 + 4;
1692
1693 obj = bpf_object__open_file(file, &open_opts);
1694 if (!obj) {
1695 p_err("failed to open object file");
1696 goto err_free_reuse_maps;
1697 }
1698
1699 bpf_object__for_each_program(pos, obj) {
1700 enum bpf_prog_type prog_type = common_prog_type;
1701
1702 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
1703 const char *sec_name = bpf_program__section_name(pos);
1704
1705 err = get_prog_type_by_name(sec_name, &prog_type,
1706 &expected_attach_type);
1707 if (err < 0)
1708 goto err_close_obj;
1709 }
1710
1711 if (prog_type == BPF_PROG_TYPE_XDP && xdpmeta_ifindex) {
1712 bpf_program__set_flags(pos, BPF_F_XDP_DEV_BOUND_ONLY);
1713 bpf_program__set_ifindex(pos, xdpmeta_ifindex);
1714 } else {
1715 bpf_program__set_ifindex(pos, offload_ifindex);
1716 }
1717 if (bpf_program__type(pos) != prog_type)
1718 bpf_program__set_type(pos, prog_type);
1719 bpf_program__set_expected_attach_type(pos, expected_attach_type);
1720 }
1721
1722 qsort(map_replace, old_map_fds, sizeof(*map_replace),
1723 map_replace_compar);
1724
1725 /* After the sort maps by name will be first on the list, because they
1726 * have idx == -1. Resolve them.
1727 */
1728 j = 0;
1729 while (j < old_map_fds && map_replace[j].name) {
1730 i = 0;
1731 bpf_object__for_each_map(map, obj) {
1732 if (!strcmp(bpf_map__name(map), map_replace[j].name)) {
1733 map_replace[j].idx = i;
1734 break;
1735 }
1736 i++;
1737 }
1738 if (map_replace[j].idx == -1) {
1739 p_err("unable to find map '%s'", map_replace[j].name);
1740 goto err_close_obj;
1741 }
1742 j++;
1743 }
1744 /* Resort if any names were resolved */
1745 if (j)
1746 qsort(map_replace, old_map_fds, sizeof(*map_replace),
1747 map_replace_compar);
1748
1749 /* Set ifindex and name reuse */
1750 j = 0;
1751 idx = 0;
1752 bpf_object__for_each_map(map, obj) {
1753 if (bpf_map__type(map) != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1754 bpf_map__set_ifindex(map, offload_ifindex);
1755
1756 if (j < old_map_fds && idx == map_replace[j].idx) {
1757 err = bpf_map__reuse_fd(map, map_replace[j++].fd);
1758 if (err) {
1759 p_err("unable to set up map reuse: %d", err);
1760 goto err_close_obj;
1761 }
1762
1763 /* Next reuse wants to apply to the same map */
1764 if (j < old_map_fds && map_replace[j].idx == idx) {
1765 p_err("replacement for map idx %d specified more than once",
1766 idx);
1767 goto err_close_obj;
1768 }
1769 }
1770
1771 idx++;
1772 }
1773 if (j < old_map_fds) {
1774 p_err("map idx '%d' not used", map_replace[j].idx);
1775 goto err_close_obj;
1776 }
1777
1778 err = bpf_object__load(obj);
1779 if (err) {
1780 p_err("failed to load object file");
1781 goto err_close_obj;
1782 }
1783
1784 if (first_prog_only)
1785 err = mount_bpffs_for_file(pinfile);
1786 else
1787 err = create_and_mount_bpffs_dir(pinfile);
1788 if (err)
1789 goto err_close_obj;
1790
1791 if (first_prog_only) {
1792 prog = bpf_object__next_program(obj, NULL);
1793 if (!prog) {
1794 p_err("object file doesn't contain any bpf program");
1795 goto err_close_obj;
1796 }
1797
1798 if (auto_attach)
1799 err = auto_attach_program(prog, pinfile);
1800 else
1801 err = bpf_obj_pin(bpf_program__fd(prog), pinfile);
1802 if (err) {
1803 p_err("failed to pin program %s",
1804 bpf_program__section_name(prog));
1805 goto err_close_obj;
1806 }
1807 } else {
1808 if (auto_attach)
1809 err = auto_attach_programs(obj, pinfile);
1810 else
1811 err = bpf_object__pin_programs(obj, pinfile);
1812 if (err) {
1813 p_err("failed to pin all programs");
1814 goto err_close_obj;
1815 }
1816 }
1817
1818 if (pinmaps) {
1819 err = create_and_mount_bpffs_dir(pinmaps);
1820 if (err)
1821 goto err_unpin;
1822
1823 err = bpf_object__pin_maps(obj, pinmaps);
1824 if (err) {
1825 p_err("failed to pin all maps");
1826 goto err_unpin;
1827 }
1828 }
1829
1830 if (json_output)
1831 jsonw_null(json_wtr);
1832
1833 bpf_object__close(obj);
1834 for (i = 0; i < old_map_fds; i++)
1835 close(map_replace[i].fd);
1836 free(map_replace);
1837
1838 return 0;
1839
1840 err_unpin:
1841 if (first_prog_only)
1842 unlink(pinfile);
1843 else
1844 bpf_object__unpin_programs(obj, pinfile);
1845 err_close_obj:
1846 bpf_object__close(obj);
1847 err_free_reuse_maps:
1848 for (i = 0; i < old_map_fds; i++)
1849 close(map_replace[i].fd);
1850 free(map_replace);
1851 return -1;
1852 }
1853
count_open_fds(void)1854 static int count_open_fds(void)
1855 {
1856 DIR *dp = opendir("/proc/self/fd");
1857 struct dirent *de;
1858 int cnt = -3;
1859
1860 if (!dp)
1861 return -1;
1862
1863 while ((de = readdir(dp)))
1864 cnt++;
1865
1866 closedir(dp);
1867 return cnt;
1868 }
1869
try_loader(struct gen_loader_opts * gen)1870 static int try_loader(struct gen_loader_opts *gen)
1871 {
1872 struct bpf_load_and_run_opts opts = {};
1873 struct bpf_loader_ctx *ctx;
1874 int ctx_sz = sizeof(*ctx) + 64 * max(sizeof(struct bpf_map_desc),
1875 sizeof(struct bpf_prog_desc));
1876 int log_buf_sz = (1u << 24) - 1;
1877 int err, fds_before, fd_delta;
1878 char *log_buf = NULL;
1879
1880 ctx = alloca(ctx_sz);
1881 memset(ctx, 0, ctx_sz);
1882 ctx->sz = ctx_sz;
1883 if (verifier_logs) {
1884 ctx->log_level = 1 + 2 + 4;
1885 ctx->log_size = log_buf_sz;
1886 log_buf = malloc(log_buf_sz);
1887 if (!log_buf)
1888 return -ENOMEM;
1889 ctx->log_buf = (long) log_buf;
1890 }
1891 opts.ctx = ctx;
1892 opts.data = gen->data;
1893 opts.data_sz = gen->data_sz;
1894 opts.insns = gen->insns;
1895 opts.insns_sz = gen->insns_sz;
1896 fds_before = count_open_fds();
1897 err = bpf_load_and_run(&opts);
1898 fd_delta = count_open_fds() - fds_before;
1899 if (err < 0 || verifier_logs) {
1900 fprintf(stderr, "err %d\n%s\n%s", err, opts.errstr, log_buf);
1901 if (fd_delta && err < 0)
1902 fprintf(stderr, "loader prog leaked %d FDs\n",
1903 fd_delta);
1904 }
1905 free(log_buf);
1906 return err;
1907 }
1908
do_loader(int argc,char ** argv)1909 static int do_loader(int argc, char **argv)
1910 {
1911 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts);
1912 DECLARE_LIBBPF_OPTS(gen_loader_opts, gen);
1913 struct bpf_object *obj;
1914 const char *file;
1915 int err = 0;
1916
1917 if (!REQ_ARGS(1))
1918 return -1;
1919 file = GET_ARG();
1920
1921 if (verifier_logs)
1922 /* log_level1 + log_level2 + stats, but not stable UAPI */
1923 open_opts.kernel_log_level = 1 + 2 + 4;
1924
1925 obj = bpf_object__open_file(file, &open_opts);
1926 if (!obj) {
1927 p_err("failed to open object file");
1928 goto err_close_obj;
1929 }
1930
1931 err = bpf_object__gen_loader(obj, &gen);
1932 if (err)
1933 goto err_close_obj;
1934
1935 err = bpf_object__load(obj);
1936 if (err) {
1937 p_err("failed to load object file");
1938 goto err_close_obj;
1939 }
1940
1941 if (verifier_logs) {
1942 struct dump_data dd = {};
1943
1944 kernel_syms_load(&dd);
1945 dump_xlated_plain(&dd, (void *)gen.insns, gen.insns_sz, false, false);
1946 kernel_syms_destroy(&dd);
1947 }
1948 err = try_loader(&gen);
1949 err_close_obj:
1950 bpf_object__close(obj);
1951 return err;
1952 }
1953
do_load(int argc,char ** argv)1954 static int do_load(int argc, char **argv)
1955 {
1956 if (use_loader)
1957 return do_loader(argc, argv);
1958 return load_with_options(argc, argv, true);
1959 }
1960
do_loadall(int argc,char ** argv)1961 static int do_loadall(int argc, char **argv)
1962 {
1963 return load_with_options(argc, argv, false);
1964 }
1965
1966 #ifdef BPFTOOL_WITHOUT_SKELETONS
1967
do_profile(int argc,char ** argv)1968 static int do_profile(int argc, char **argv)
1969 {
1970 p_err("bpftool prog profile command is not supported. Please build bpftool with clang >= 10.0.0");
1971 return 0;
1972 }
1973
1974 #else /* BPFTOOL_WITHOUT_SKELETONS */
1975
1976 #include "profiler.skel.h"
1977
1978 struct profile_metric {
1979 const char *name;
1980 struct bpf_perf_event_value val;
1981 struct perf_event_attr attr;
1982 bool selected;
1983
1984 /* calculate ratios like instructions per cycle */
1985 const int ratio_metric; /* 0 for N/A, 1 for index 0 (cycles) */
1986 const char *ratio_desc;
1987 const float ratio_mul;
1988 } metrics[] = {
1989 {
1990 .name = "cycles",
1991 .attr = {
1992 .type = PERF_TYPE_HARDWARE,
1993 .config = PERF_COUNT_HW_CPU_CYCLES,
1994 .exclude_user = 1,
1995 },
1996 },
1997 {
1998 .name = "instructions",
1999 .attr = {
2000 .type = PERF_TYPE_HARDWARE,
2001 .config = PERF_COUNT_HW_INSTRUCTIONS,
2002 .exclude_user = 1,
2003 },
2004 .ratio_metric = 1,
2005 .ratio_desc = "insns per cycle",
2006 .ratio_mul = 1.0,
2007 },
2008 {
2009 .name = "l1d_loads",
2010 .attr = {
2011 .type = PERF_TYPE_HW_CACHE,
2012 .config =
2013 PERF_COUNT_HW_CACHE_L1D |
2014 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2015 (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16),
2016 .exclude_user = 1,
2017 },
2018 },
2019 {
2020 .name = "llc_misses",
2021 .attr = {
2022 .type = PERF_TYPE_HW_CACHE,
2023 .config =
2024 PERF_COUNT_HW_CACHE_LL |
2025 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2026 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2027 .exclude_user = 1
2028 },
2029 .ratio_metric = 2,
2030 .ratio_desc = "LLC misses per million insns",
2031 .ratio_mul = 1e6,
2032 },
2033 {
2034 .name = "itlb_misses",
2035 .attr = {
2036 .type = PERF_TYPE_HW_CACHE,
2037 .config =
2038 PERF_COUNT_HW_CACHE_ITLB |
2039 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2040 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2041 .exclude_user = 1
2042 },
2043 .ratio_metric = 2,
2044 .ratio_desc = "itlb misses per million insns",
2045 .ratio_mul = 1e6,
2046 },
2047 {
2048 .name = "dtlb_misses",
2049 .attr = {
2050 .type = PERF_TYPE_HW_CACHE,
2051 .config =
2052 PERF_COUNT_HW_CACHE_DTLB |
2053 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2054 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2055 .exclude_user = 1
2056 },
2057 .ratio_metric = 2,
2058 .ratio_desc = "dtlb misses per million insns",
2059 .ratio_mul = 1e6,
2060 },
2061 };
2062
2063 static __u64 profile_total_count;
2064
2065 #define MAX_NUM_PROFILE_METRICS 4
2066
profile_parse_metrics(int argc,char ** argv)2067 static int profile_parse_metrics(int argc, char **argv)
2068 {
2069 unsigned int metric_cnt;
2070 int selected_cnt = 0;
2071 unsigned int i;
2072
2073 metric_cnt = ARRAY_SIZE(metrics);
2074
2075 while (argc > 0) {
2076 for (i = 0; i < metric_cnt; i++) {
2077 if (is_prefix(argv[0], metrics[i].name)) {
2078 if (!metrics[i].selected)
2079 selected_cnt++;
2080 metrics[i].selected = true;
2081 break;
2082 }
2083 }
2084 if (i == metric_cnt) {
2085 p_err("unknown metric %s", argv[0]);
2086 return -1;
2087 }
2088 NEXT_ARG();
2089 }
2090 if (selected_cnt > MAX_NUM_PROFILE_METRICS) {
2091 p_err("too many (%d) metrics, please specify no more than %d metrics at at time",
2092 selected_cnt, MAX_NUM_PROFILE_METRICS);
2093 return -1;
2094 }
2095 return selected_cnt;
2096 }
2097
profile_read_values(struct profiler_bpf * obj)2098 static void profile_read_values(struct profiler_bpf *obj)
2099 {
2100 __u32 m, cpu, num_cpu = obj->rodata->num_cpu;
2101 int reading_map_fd, count_map_fd;
2102 __u64 counts[num_cpu];
2103 __u32 key = 0;
2104 int err;
2105
2106 reading_map_fd = bpf_map__fd(obj->maps.accum_readings);
2107 count_map_fd = bpf_map__fd(obj->maps.counts);
2108 if (reading_map_fd < 0 || count_map_fd < 0) {
2109 p_err("failed to get fd for map");
2110 return;
2111 }
2112
2113 err = bpf_map_lookup_elem(count_map_fd, &key, counts);
2114 if (err) {
2115 p_err("failed to read count_map: %s", strerror(errno));
2116 return;
2117 }
2118
2119 profile_total_count = 0;
2120 for (cpu = 0; cpu < num_cpu; cpu++)
2121 profile_total_count += counts[cpu];
2122
2123 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2124 struct bpf_perf_event_value values[num_cpu];
2125
2126 if (!metrics[m].selected)
2127 continue;
2128
2129 err = bpf_map_lookup_elem(reading_map_fd, &key, values);
2130 if (err) {
2131 p_err("failed to read reading_map: %s",
2132 strerror(errno));
2133 return;
2134 }
2135 for (cpu = 0; cpu < num_cpu; cpu++) {
2136 metrics[m].val.counter += values[cpu].counter;
2137 metrics[m].val.enabled += values[cpu].enabled;
2138 metrics[m].val.running += values[cpu].running;
2139 }
2140 key++;
2141 }
2142 }
2143
profile_print_readings_json(void)2144 static void profile_print_readings_json(void)
2145 {
2146 __u32 m;
2147
2148 jsonw_start_array(json_wtr);
2149 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2150 if (!metrics[m].selected)
2151 continue;
2152 jsonw_start_object(json_wtr);
2153 jsonw_string_field(json_wtr, "metric", metrics[m].name);
2154 jsonw_lluint_field(json_wtr, "run_cnt", profile_total_count);
2155 jsonw_lluint_field(json_wtr, "value", metrics[m].val.counter);
2156 jsonw_lluint_field(json_wtr, "enabled", metrics[m].val.enabled);
2157 jsonw_lluint_field(json_wtr, "running", metrics[m].val.running);
2158
2159 jsonw_end_object(json_wtr);
2160 }
2161 jsonw_end_array(json_wtr);
2162 }
2163
profile_print_readings_plain(void)2164 static void profile_print_readings_plain(void)
2165 {
2166 __u32 m;
2167
2168 printf("\n%18llu %-20s\n", profile_total_count, "run_cnt");
2169 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2170 struct bpf_perf_event_value *val = &metrics[m].val;
2171 int r;
2172
2173 if (!metrics[m].selected)
2174 continue;
2175 printf("%18llu %-20s", val->counter, metrics[m].name);
2176
2177 r = metrics[m].ratio_metric - 1;
2178 if (r >= 0 && metrics[r].selected &&
2179 metrics[r].val.counter > 0) {
2180 printf("# %8.2f %-30s",
2181 val->counter * metrics[m].ratio_mul /
2182 metrics[r].val.counter,
2183 metrics[m].ratio_desc);
2184 } else {
2185 printf("%-41s", "");
2186 }
2187
2188 if (val->enabled > val->running)
2189 printf("(%4.2f%%)",
2190 val->running * 100.0 / val->enabled);
2191 printf("\n");
2192 }
2193 }
2194
profile_print_readings(void)2195 static void profile_print_readings(void)
2196 {
2197 if (json_output)
2198 profile_print_readings_json();
2199 else
2200 profile_print_readings_plain();
2201 }
2202
profile_target_name(int tgt_fd)2203 static char *profile_target_name(int tgt_fd)
2204 {
2205 struct bpf_func_info func_info;
2206 struct bpf_prog_info info = {};
2207 __u32 info_len = sizeof(info);
2208 const struct btf_type *t;
2209 __u32 func_info_rec_size;
2210 struct btf *btf = NULL;
2211 char *name = NULL;
2212 int err;
2213
2214 err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len);
2215 if (err) {
2216 p_err("failed to get info for prog FD %d", tgt_fd);
2217 goto out;
2218 }
2219
2220 if (info.btf_id == 0) {
2221 p_err("prog FD %d doesn't have valid btf", tgt_fd);
2222 goto out;
2223 }
2224
2225 func_info_rec_size = info.func_info_rec_size;
2226 if (info.nr_func_info == 0) {
2227 p_err("found 0 func_info for prog FD %d", tgt_fd);
2228 goto out;
2229 }
2230
2231 memset(&info, 0, sizeof(info));
2232 info.nr_func_info = 1;
2233 info.func_info_rec_size = func_info_rec_size;
2234 info.func_info = ptr_to_u64(&func_info);
2235
2236 err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len);
2237 if (err) {
2238 p_err("failed to get func_info for prog FD %d", tgt_fd);
2239 goto out;
2240 }
2241
2242 btf = btf__load_from_kernel_by_id(info.btf_id);
2243 if (!btf) {
2244 p_err("failed to load btf for prog FD %d", tgt_fd);
2245 goto out;
2246 }
2247
2248 t = btf__type_by_id(btf, func_info.type_id);
2249 if (!t) {
2250 p_err("btf %d doesn't have type %d",
2251 info.btf_id, func_info.type_id);
2252 goto out;
2253 }
2254 name = strdup(btf__name_by_offset(btf, t->name_off));
2255 out:
2256 btf__free(btf);
2257 return name;
2258 }
2259
2260 static struct profiler_bpf *profile_obj;
2261 static int profile_tgt_fd = -1;
2262 static char *profile_tgt_name;
2263 static int *profile_perf_events;
2264 static int profile_perf_event_cnt;
2265
profile_close_perf_events(struct profiler_bpf * obj)2266 static void profile_close_perf_events(struct profiler_bpf *obj)
2267 {
2268 int i;
2269
2270 for (i = profile_perf_event_cnt - 1; i >= 0; i--)
2271 close(profile_perf_events[i]);
2272
2273 free(profile_perf_events);
2274 profile_perf_event_cnt = 0;
2275 }
2276
profile_open_perf_event(int mid,int cpu,int map_fd)2277 static int profile_open_perf_event(int mid, int cpu, int map_fd)
2278 {
2279 int pmu_fd;
2280
2281 pmu_fd = syscall(__NR_perf_event_open, &metrics[mid].attr,
2282 -1 /*pid*/, cpu, -1 /*group_fd*/, 0);
2283 if (pmu_fd < 0) {
2284 if (errno == ENODEV) {
2285 p_info("cpu %d may be offline, skip %s profiling.",
2286 cpu, metrics[mid].name);
2287 profile_perf_event_cnt++;
2288 return 0;
2289 }
2290 return -1;
2291 }
2292
2293 if (bpf_map_update_elem(map_fd,
2294 &profile_perf_event_cnt,
2295 &pmu_fd, BPF_ANY) ||
2296 ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0)) {
2297 close(pmu_fd);
2298 return -1;
2299 }
2300
2301 profile_perf_events[profile_perf_event_cnt++] = pmu_fd;
2302 return 0;
2303 }
2304
profile_open_perf_events(struct profiler_bpf * obj)2305 static int profile_open_perf_events(struct profiler_bpf *obj)
2306 {
2307 unsigned int cpu, m;
2308 int map_fd;
2309
2310 profile_perf_events = calloc(
2311 obj->rodata->num_cpu * obj->rodata->num_metric, sizeof(int));
2312 if (!profile_perf_events) {
2313 p_err("failed to allocate memory for perf_event array: %s",
2314 strerror(errno));
2315 return -1;
2316 }
2317 map_fd = bpf_map__fd(obj->maps.events);
2318 if (map_fd < 0) {
2319 p_err("failed to get fd for events map");
2320 return -1;
2321 }
2322
2323 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2324 if (!metrics[m].selected)
2325 continue;
2326 for (cpu = 0; cpu < obj->rodata->num_cpu; cpu++) {
2327 if (profile_open_perf_event(m, cpu, map_fd)) {
2328 p_err("failed to create event %s on cpu %d",
2329 metrics[m].name, cpu);
2330 return -1;
2331 }
2332 }
2333 }
2334 return 0;
2335 }
2336
profile_print_and_cleanup(void)2337 static void profile_print_and_cleanup(void)
2338 {
2339 profile_close_perf_events(profile_obj);
2340 profile_read_values(profile_obj);
2341 profile_print_readings();
2342 profiler_bpf__destroy(profile_obj);
2343
2344 close(profile_tgt_fd);
2345 free(profile_tgt_name);
2346 }
2347
int_exit(int signo)2348 static void int_exit(int signo)
2349 {
2350 profile_print_and_cleanup();
2351 exit(0);
2352 }
2353
do_profile(int argc,char ** argv)2354 static int do_profile(int argc, char **argv)
2355 {
2356 int num_metric, num_cpu, err = -1;
2357 struct bpf_program *prog;
2358 unsigned long duration;
2359 char *endptr;
2360
2361 /* we at least need two args for the prog and one metric */
2362 if (!REQ_ARGS(3))
2363 return -EINVAL;
2364
2365 /* parse target fd */
2366 profile_tgt_fd = prog_parse_fd(&argc, &argv);
2367 if (profile_tgt_fd < 0) {
2368 p_err("failed to parse fd");
2369 return -1;
2370 }
2371
2372 /* parse profiling optional duration */
2373 if (argc > 2 && is_prefix(argv[0], "duration")) {
2374 NEXT_ARG();
2375 duration = strtoul(*argv, &endptr, 0);
2376 if (*endptr)
2377 usage();
2378 NEXT_ARG();
2379 } else {
2380 duration = UINT_MAX;
2381 }
2382
2383 num_metric = profile_parse_metrics(argc, argv);
2384 if (num_metric <= 0)
2385 goto out;
2386
2387 num_cpu = libbpf_num_possible_cpus();
2388 if (num_cpu <= 0) {
2389 p_err("failed to identify number of CPUs");
2390 goto out;
2391 }
2392
2393 profile_obj = profiler_bpf__open();
2394 if (!profile_obj) {
2395 p_err("failed to open and/or load BPF object");
2396 goto out;
2397 }
2398
2399 profile_obj->rodata->num_cpu = num_cpu;
2400 profile_obj->rodata->num_metric = num_metric;
2401
2402 /* adjust map sizes */
2403 bpf_map__set_max_entries(profile_obj->maps.events, num_metric * num_cpu);
2404 bpf_map__set_max_entries(profile_obj->maps.fentry_readings, num_metric);
2405 bpf_map__set_max_entries(profile_obj->maps.accum_readings, num_metric);
2406 bpf_map__set_max_entries(profile_obj->maps.counts, 1);
2407
2408 /* change target name */
2409 profile_tgt_name = profile_target_name(profile_tgt_fd);
2410 if (!profile_tgt_name)
2411 goto out;
2412
2413 bpf_object__for_each_program(prog, profile_obj->obj) {
2414 err = bpf_program__set_attach_target(prog, profile_tgt_fd,
2415 profile_tgt_name);
2416 if (err) {
2417 p_err("failed to set attach target\n");
2418 goto out;
2419 }
2420 }
2421
2422 set_max_rlimit();
2423 err = profiler_bpf__load(profile_obj);
2424 if (err) {
2425 p_err("failed to load profile_obj");
2426 goto out;
2427 }
2428
2429 err = profile_open_perf_events(profile_obj);
2430 if (err)
2431 goto out;
2432
2433 err = profiler_bpf__attach(profile_obj);
2434 if (err) {
2435 p_err("failed to attach profile_obj");
2436 goto out;
2437 }
2438 signal(SIGINT, int_exit);
2439
2440 sleep(duration);
2441 profile_print_and_cleanup();
2442 return 0;
2443
2444 out:
2445 profile_close_perf_events(profile_obj);
2446 if (profile_obj)
2447 profiler_bpf__destroy(profile_obj);
2448 close(profile_tgt_fd);
2449 free(profile_tgt_name);
2450 return err;
2451 }
2452
2453 #endif /* BPFTOOL_WITHOUT_SKELETONS */
2454
do_help(int argc,char ** argv)2455 static int do_help(int argc, char **argv)
2456 {
2457 if (json_output) {
2458 jsonw_null(json_wtr);
2459 return 0;
2460 }
2461
2462 fprintf(stderr,
2463 "Usage: %1$s %2$s { show | list } [PROG]\n"
2464 " %1$s %2$s dump xlated PROG [{ file FILE | [opcodes] [linum] [visual] }]\n"
2465 " %1$s %2$s dump jited PROG [{ file FILE | [opcodes] [linum] }]\n"
2466 " %1$s %2$s pin PROG FILE\n"
2467 " %1$s %2$s { load | loadall } OBJ PATH \\\n"
2468 " [type TYPE] [{ offload_dev | xdpmeta_dev } NAME] \\\n"
2469 " [map { idx IDX | name NAME } MAP]\\\n"
2470 " [pinmaps MAP_DIR]\n"
2471 " [autoattach]\n"
2472 " %1$s %2$s attach PROG ATTACH_TYPE [MAP]\n"
2473 " %1$s %2$s detach PROG ATTACH_TYPE [MAP]\n"
2474 " %1$s %2$s run PROG \\\n"
2475 " data_in FILE \\\n"
2476 " [data_out FILE [data_size_out L]] \\\n"
2477 " [ctx_in FILE [ctx_out FILE [ctx_size_out M]]] \\\n"
2478 " [repeat N]\n"
2479 " %1$s %2$s profile PROG [duration DURATION] METRICs\n"
2480 " %1$s %2$s tracelog\n"
2481 " %1$s %2$s help\n"
2482 "\n"
2483 " " HELP_SPEC_MAP "\n"
2484 " " HELP_SPEC_PROGRAM "\n"
2485 " TYPE := { socket | kprobe | kretprobe | classifier | action |\n"
2486 " tracepoint | raw_tracepoint | xdp | perf_event | cgroup/skb |\n"
2487 " cgroup/sock | cgroup/dev | lwt_in | lwt_out | lwt_xmit |\n"
2488 " lwt_seg6local | sockops | sk_skb | sk_msg | lirc_mode2 |\n"
2489 " sk_reuseport | flow_dissector | cgroup/sysctl |\n"
2490 " cgroup/bind4 | cgroup/bind6 | cgroup/post_bind4 |\n"
2491 " cgroup/post_bind6 | cgroup/connect4 | cgroup/connect6 |\n"
2492 " cgroup/getpeername4 | cgroup/getpeername6 |\n"
2493 " cgroup/getsockname4 | cgroup/getsockname6 | cgroup/sendmsg4 |\n"
2494 " cgroup/sendmsg6 | cgroup/recvmsg4 | cgroup/recvmsg6 |\n"
2495 " cgroup/getsockopt | cgroup/setsockopt | cgroup/sock_release |\n"
2496 " struct_ops | fentry | fexit | freplace | sk_lookup }\n"
2497 " ATTACH_TYPE := { sk_msg_verdict | sk_skb_verdict | sk_skb_stream_verdict |\n"
2498 " sk_skb_stream_parser | flow_dissector }\n"
2499 " METRIC := { cycles | instructions | l1d_loads | llc_misses | itlb_misses | dtlb_misses }\n"
2500 " " HELP_SPEC_OPTIONS " |\n"
2501 " {-f|--bpffs} | {-m|--mapcompat} | {-n|--nomount} |\n"
2502 " {-L|--use-loader} }\n"
2503 "",
2504 bin_name, argv[-2]);
2505
2506 return 0;
2507 }
2508
2509 static const struct cmd cmds[] = {
2510 { "show", do_show },
2511 { "list", do_show },
2512 { "help", do_help },
2513 { "dump", do_dump },
2514 { "pin", do_pin },
2515 { "load", do_load },
2516 { "loadall", do_loadall },
2517 { "attach", do_attach },
2518 { "detach", do_detach },
2519 { "tracelog", do_tracelog },
2520 { "run", do_run },
2521 { "profile", do_profile },
2522 { 0 }
2523 };
2524
do_prog(int argc,char ** argv)2525 int do_prog(int argc, char **argv)
2526 {
2527 return cmd_select(cmds, argc, argv, do_help);
2528 }
2529