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 (disasm_print_insn(img, lens[i], opcodes,
822 name, disasm_opt, btf,
823 prog_linfo, ksyms[i], i,
824 linum))
825 goto exit_free;
826
827 img += lens[i];
828
829 if (json_output)
830 jsonw_end_object(json_wtr);
831 else
832 printf("\n");
833 }
834
835 if (json_output)
836 jsonw_end_array(json_wtr);
837 } else {
838 if (disasm_print_insn(buf, member_len, opcodes, name,
839 disasm_opt, btf, NULL, 0, 0,
840 false))
841 goto exit_free;
842 }
843 } else {
844 kernel_syms_load(&dd);
845 dd.nr_jited_ksyms = info->nr_jited_ksyms;
846 dd.jited_ksyms = u64_to_ptr(info->jited_ksyms);
847 dd.btf = btf;
848 dd.func_info = func_info;
849 dd.finfo_rec_size = info->func_info_rec_size;
850 dd.prog_linfo = prog_linfo;
851
852 if (json_output)
853 dump_xlated_json(&dd, buf, member_len, opcodes, linum);
854 else if (visual)
855 dump_xlated_cfg(&dd, buf, member_len, opcodes, linum);
856 else
857 dump_xlated_plain(&dd, buf, member_len, opcodes, linum);
858 kernel_syms_destroy(&dd);
859 }
860
861 err = 0;
862
863 exit_free:
864 btf__free(btf);
865 bpf_prog_linfo__free(prog_linfo);
866 return err;
867 }
868
do_dump(int argc,char ** argv)869 static int do_dump(int argc, char **argv)
870 {
871 struct bpf_prog_info info;
872 __u32 info_len = sizeof(info);
873 size_t info_data_sz = 0;
874 void *info_data = NULL;
875 char *filepath = NULL;
876 bool opcodes = false;
877 bool visual = false;
878 enum dump_mode mode;
879 bool linum = false;
880 int nb_fds, i = 0;
881 int *fds = NULL;
882 int err = -1;
883
884 if (is_prefix(*argv, "jited")) {
885 if (disasm_init())
886 return -1;
887 mode = DUMP_JITED;
888 } else if (is_prefix(*argv, "xlated")) {
889 mode = DUMP_XLATED;
890 } else {
891 p_err("expected 'xlated' or 'jited', got: %s", *argv);
892 return -1;
893 }
894 NEXT_ARG();
895
896 if (argc < 2)
897 usage();
898
899 fds = malloc(sizeof(int));
900 if (!fds) {
901 p_err("mem alloc failed");
902 return -1;
903 }
904 nb_fds = prog_parse_fds(&argc, &argv, &fds);
905 if (nb_fds < 1)
906 goto exit_free;
907
908 while (argc) {
909 if (is_prefix(*argv, "file")) {
910 NEXT_ARG();
911 if (!argc) {
912 p_err("expected file path");
913 goto exit_close;
914 }
915 if (nb_fds > 1) {
916 p_err("several programs matched");
917 goto exit_close;
918 }
919
920 filepath = *argv;
921 NEXT_ARG();
922 } else if (is_prefix(*argv, "opcodes")) {
923 opcodes = true;
924 NEXT_ARG();
925 } else if (is_prefix(*argv, "visual")) {
926 if (nb_fds > 1) {
927 p_err("several programs matched");
928 goto exit_close;
929 }
930
931 visual = true;
932 NEXT_ARG();
933 } else if (is_prefix(*argv, "linum")) {
934 linum = true;
935 NEXT_ARG();
936 } else {
937 usage();
938 goto exit_close;
939 }
940 }
941
942 if (filepath && (opcodes || visual || linum)) {
943 p_err("'file' is not compatible with 'opcodes', 'visual', or 'linum'");
944 goto exit_close;
945 }
946 if (json_output && visual) {
947 p_err("'visual' is not compatible with JSON output");
948 goto exit_close;
949 }
950
951 if (json_output && nb_fds > 1)
952 jsonw_start_array(json_wtr); /* root array */
953 for (i = 0; i < nb_fds; i++) {
954 memset(&info, 0, sizeof(info));
955
956 err = bpf_prog_get_info_by_fd(fds[i], &info, &info_len);
957 if (err) {
958 p_err("can't get prog info: %s", strerror(errno));
959 break;
960 }
961
962 err = prep_prog_info(&info, mode, &info_data, &info_data_sz);
963 if (err) {
964 p_err("can't grow prog info_data");
965 break;
966 }
967
968 err = bpf_prog_get_info_by_fd(fds[i], &info, &info_len);
969 if (err) {
970 p_err("can't get prog info: %s", strerror(errno));
971 break;
972 }
973
974 if (json_output && nb_fds > 1) {
975 jsonw_start_object(json_wtr); /* prog object */
976 print_prog_header_json(&info, fds[i]);
977 jsonw_name(json_wtr, "insns");
978 } else if (nb_fds > 1) {
979 print_prog_header_plain(&info, fds[i]);
980 }
981
982 err = prog_dump(&info, mode, filepath, opcodes, visual, linum);
983
984 if (json_output && nb_fds > 1)
985 jsonw_end_object(json_wtr); /* prog object */
986 else if (i != nb_fds - 1 && nb_fds > 1)
987 printf("\n");
988
989 if (err)
990 break;
991 close(fds[i]);
992 }
993 if (json_output && nb_fds > 1)
994 jsonw_end_array(json_wtr); /* root array */
995
996 exit_close:
997 for (; i < nb_fds; i++)
998 close(fds[i]);
999 exit_free:
1000 free(info_data);
1001 free(fds);
1002 return err;
1003 }
1004
do_pin(int argc,char ** argv)1005 static int do_pin(int argc, char **argv)
1006 {
1007 int err;
1008
1009 err = do_pin_any(argc, argv, prog_parse_fd);
1010 if (!err && json_output)
1011 jsonw_null(json_wtr);
1012 return err;
1013 }
1014
1015 struct map_replace {
1016 int idx;
1017 int fd;
1018 char *name;
1019 };
1020
map_replace_compar(const void * p1,const void * p2)1021 static int map_replace_compar(const void *p1, const void *p2)
1022 {
1023 const struct map_replace *a = p1, *b = p2;
1024
1025 return a->idx - b->idx;
1026 }
1027
parse_attach_detach_args(int argc,char ** argv,int * progfd,enum bpf_attach_type * attach_type,int * mapfd)1028 static int parse_attach_detach_args(int argc, char **argv, int *progfd,
1029 enum bpf_attach_type *attach_type,
1030 int *mapfd)
1031 {
1032 if (!REQ_ARGS(3))
1033 return -EINVAL;
1034
1035 *progfd = prog_parse_fd(&argc, &argv);
1036 if (*progfd < 0)
1037 return *progfd;
1038
1039 *attach_type = parse_attach_type(*argv);
1040 if (*attach_type == __MAX_BPF_ATTACH_TYPE) {
1041 p_err("invalid attach/detach type");
1042 return -EINVAL;
1043 }
1044
1045 if (*attach_type == BPF_FLOW_DISSECTOR) {
1046 *mapfd = 0;
1047 return 0;
1048 }
1049
1050 NEXT_ARG();
1051 if (!REQ_ARGS(2))
1052 return -EINVAL;
1053
1054 *mapfd = map_parse_fd(&argc, &argv);
1055 if (*mapfd < 0)
1056 return *mapfd;
1057
1058 return 0;
1059 }
1060
do_attach(int argc,char ** argv)1061 static int do_attach(int argc, char **argv)
1062 {
1063 enum bpf_attach_type attach_type;
1064 int err, progfd;
1065 int mapfd;
1066
1067 err = parse_attach_detach_args(argc, argv,
1068 &progfd, &attach_type, &mapfd);
1069 if (err)
1070 return err;
1071
1072 err = bpf_prog_attach(progfd, mapfd, attach_type, 0);
1073 if (err) {
1074 p_err("failed prog attach to map");
1075 return -EINVAL;
1076 }
1077
1078 if (json_output)
1079 jsonw_null(json_wtr);
1080 return 0;
1081 }
1082
do_detach(int argc,char ** argv)1083 static int do_detach(int argc, char **argv)
1084 {
1085 enum bpf_attach_type attach_type;
1086 int err, progfd;
1087 int mapfd;
1088
1089 err = parse_attach_detach_args(argc, argv,
1090 &progfd, &attach_type, &mapfd);
1091 if (err)
1092 return err;
1093
1094 err = bpf_prog_detach2(progfd, mapfd, attach_type);
1095 if (err) {
1096 p_err("failed prog detach from map");
1097 return -EINVAL;
1098 }
1099
1100 if (json_output)
1101 jsonw_null(json_wtr);
1102 return 0;
1103 }
1104
check_single_stdin(char * file_data_in,char * file_ctx_in)1105 static int check_single_stdin(char *file_data_in, char *file_ctx_in)
1106 {
1107 if (file_data_in && file_ctx_in &&
1108 !strcmp(file_data_in, "-") && !strcmp(file_ctx_in, "-")) {
1109 p_err("cannot use standard input for both data_in and ctx_in");
1110 return -1;
1111 }
1112
1113 return 0;
1114 }
1115
get_run_data(const char * fname,void ** data_ptr,unsigned int * size)1116 static int get_run_data(const char *fname, void **data_ptr, unsigned int *size)
1117 {
1118 size_t block_size = 256;
1119 size_t buf_size = block_size;
1120 size_t nb_read = 0;
1121 void *tmp;
1122 FILE *f;
1123
1124 if (!fname) {
1125 *data_ptr = NULL;
1126 *size = 0;
1127 return 0;
1128 }
1129
1130 if (!strcmp(fname, "-"))
1131 f = stdin;
1132 else
1133 f = fopen(fname, "r");
1134 if (!f) {
1135 p_err("failed to open %s: %s", fname, strerror(errno));
1136 return -1;
1137 }
1138
1139 *data_ptr = malloc(block_size);
1140 if (!*data_ptr) {
1141 p_err("failed to allocate memory for data_in/ctx_in: %s",
1142 strerror(errno));
1143 goto err_fclose;
1144 }
1145
1146 while ((nb_read += fread(*data_ptr + nb_read, 1, block_size, f))) {
1147 if (feof(f))
1148 break;
1149 if (ferror(f)) {
1150 p_err("failed to read data_in/ctx_in from %s: %s",
1151 fname, strerror(errno));
1152 goto err_free;
1153 }
1154 if (nb_read > buf_size - block_size) {
1155 if (buf_size == UINT32_MAX) {
1156 p_err("data_in/ctx_in is too long (max: %d)",
1157 UINT32_MAX);
1158 goto err_free;
1159 }
1160 /* No space for fread()-ing next chunk; realloc() */
1161 buf_size *= 2;
1162 tmp = realloc(*data_ptr, buf_size);
1163 if (!tmp) {
1164 p_err("failed to reallocate data_in/ctx_in: %s",
1165 strerror(errno));
1166 goto err_free;
1167 }
1168 *data_ptr = tmp;
1169 }
1170 }
1171 if (f != stdin)
1172 fclose(f);
1173
1174 *size = nb_read;
1175 return 0;
1176
1177 err_free:
1178 free(*data_ptr);
1179 *data_ptr = NULL;
1180 err_fclose:
1181 if (f != stdin)
1182 fclose(f);
1183 return -1;
1184 }
1185
hex_print(void * data,unsigned int size,FILE * f)1186 static void hex_print(void *data, unsigned int size, FILE *f)
1187 {
1188 size_t i, j;
1189 char c;
1190
1191 for (i = 0; i < size; i += 16) {
1192 /* Row offset */
1193 fprintf(f, "%07zx\t", i);
1194
1195 /* Hexadecimal values */
1196 for (j = i; j < i + 16 && j < size; j++)
1197 fprintf(f, "%02x%s", *(uint8_t *)(data + j),
1198 j % 2 ? " " : "");
1199 for (; j < i + 16; j++)
1200 fprintf(f, " %s", j % 2 ? " " : "");
1201
1202 /* ASCII values (if relevant), '.' otherwise */
1203 fprintf(f, "| ");
1204 for (j = i; j < i + 16 && j < size; j++) {
1205 c = *(char *)(data + j);
1206 if (c < ' ' || c > '~')
1207 c = '.';
1208 fprintf(f, "%c%s", c, j == i + 7 ? " " : "");
1209 }
1210
1211 fprintf(f, "\n");
1212 }
1213 }
1214
1215 static int
print_run_output(void * data,unsigned int size,const char * fname,const char * json_key)1216 print_run_output(void *data, unsigned int size, const char *fname,
1217 const char *json_key)
1218 {
1219 size_t nb_written;
1220 FILE *f;
1221
1222 if (!fname)
1223 return 0;
1224
1225 if (!strcmp(fname, "-")) {
1226 f = stdout;
1227 if (json_output) {
1228 jsonw_name(json_wtr, json_key);
1229 print_data_json(data, size);
1230 } else {
1231 hex_print(data, size, f);
1232 }
1233 return 0;
1234 }
1235
1236 f = fopen(fname, "w");
1237 if (!f) {
1238 p_err("failed to open %s: %s", fname, strerror(errno));
1239 return -1;
1240 }
1241
1242 nb_written = fwrite(data, 1, size, f);
1243 fclose(f);
1244 if (nb_written != size) {
1245 p_err("failed to write output data/ctx: %s", strerror(errno));
1246 return -1;
1247 }
1248
1249 return 0;
1250 }
1251
alloc_run_data(void ** data_ptr,unsigned int size_out)1252 static int alloc_run_data(void **data_ptr, unsigned int size_out)
1253 {
1254 *data_ptr = calloc(size_out, 1);
1255 if (!*data_ptr) {
1256 p_err("failed to allocate memory for output data/ctx: %s",
1257 strerror(errno));
1258 return -1;
1259 }
1260
1261 return 0;
1262 }
1263
do_run(int argc,char ** argv)1264 static int do_run(int argc, char **argv)
1265 {
1266 char *data_fname_in = NULL, *data_fname_out = NULL;
1267 char *ctx_fname_in = NULL, *ctx_fname_out = NULL;
1268 const unsigned int default_size = SZ_32K;
1269 void *data_in = NULL, *data_out = NULL;
1270 void *ctx_in = NULL, *ctx_out = NULL;
1271 unsigned int repeat = 1;
1272 int fd, err;
1273 LIBBPF_OPTS(bpf_test_run_opts, test_attr);
1274
1275 if (!REQ_ARGS(4))
1276 return -1;
1277
1278 fd = prog_parse_fd(&argc, &argv);
1279 if (fd < 0)
1280 return -1;
1281
1282 while (argc) {
1283 if (detect_common_prefix(*argv, "data_in", "data_out",
1284 "data_size_out", NULL))
1285 return -1;
1286 if (detect_common_prefix(*argv, "ctx_in", "ctx_out",
1287 "ctx_size_out", NULL))
1288 return -1;
1289
1290 if (is_prefix(*argv, "data_in")) {
1291 NEXT_ARG();
1292 if (!REQ_ARGS(1))
1293 return -1;
1294
1295 data_fname_in = GET_ARG();
1296 if (check_single_stdin(data_fname_in, ctx_fname_in))
1297 return -1;
1298 } else if (is_prefix(*argv, "data_out")) {
1299 NEXT_ARG();
1300 if (!REQ_ARGS(1))
1301 return -1;
1302
1303 data_fname_out = GET_ARG();
1304 } else if (is_prefix(*argv, "data_size_out")) {
1305 char *endptr;
1306
1307 NEXT_ARG();
1308 if (!REQ_ARGS(1))
1309 return -1;
1310
1311 test_attr.data_size_out = strtoul(*argv, &endptr, 0);
1312 if (*endptr) {
1313 p_err("can't parse %s as output data size",
1314 *argv);
1315 return -1;
1316 }
1317 NEXT_ARG();
1318 } else if (is_prefix(*argv, "ctx_in")) {
1319 NEXT_ARG();
1320 if (!REQ_ARGS(1))
1321 return -1;
1322
1323 ctx_fname_in = GET_ARG();
1324 if (check_single_stdin(data_fname_in, ctx_fname_in))
1325 return -1;
1326 } else if (is_prefix(*argv, "ctx_out")) {
1327 NEXT_ARG();
1328 if (!REQ_ARGS(1))
1329 return -1;
1330
1331 ctx_fname_out = GET_ARG();
1332 } else if (is_prefix(*argv, "ctx_size_out")) {
1333 char *endptr;
1334
1335 NEXT_ARG();
1336 if (!REQ_ARGS(1))
1337 return -1;
1338
1339 test_attr.ctx_size_out = strtoul(*argv, &endptr, 0);
1340 if (*endptr) {
1341 p_err("can't parse %s as output context size",
1342 *argv);
1343 return -1;
1344 }
1345 NEXT_ARG();
1346 } else if (is_prefix(*argv, "repeat")) {
1347 char *endptr;
1348
1349 NEXT_ARG();
1350 if (!REQ_ARGS(1))
1351 return -1;
1352
1353 repeat = strtoul(*argv, &endptr, 0);
1354 if (*endptr) {
1355 p_err("can't parse %s as repeat number",
1356 *argv);
1357 return -1;
1358 }
1359 NEXT_ARG();
1360 } else {
1361 p_err("expected no more arguments, 'data_in', 'data_out', 'data_size_out', 'ctx_in', 'ctx_out', 'ctx_size_out' or 'repeat', got: '%s'?",
1362 *argv);
1363 return -1;
1364 }
1365 }
1366
1367 err = get_run_data(data_fname_in, &data_in, &test_attr.data_size_in);
1368 if (err)
1369 return -1;
1370
1371 if (data_in) {
1372 if (!test_attr.data_size_out)
1373 test_attr.data_size_out = default_size;
1374 err = alloc_run_data(&data_out, test_attr.data_size_out);
1375 if (err)
1376 goto free_data_in;
1377 }
1378
1379 err = get_run_data(ctx_fname_in, &ctx_in, &test_attr.ctx_size_in);
1380 if (err)
1381 goto free_data_out;
1382
1383 if (ctx_in) {
1384 if (!test_attr.ctx_size_out)
1385 test_attr.ctx_size_out = default_size;
1386 err = alloc_run_data(&ctx_out, test_attr.ctx_size_out);
1387 if (err)
1388 goto free_ctx_in;
1389 }
1390
1391 test_attr.repeat = repeat;
1392 test_attr.data_in = data_in;
1393 test_attr.data_out = data_out;
1394 test_attr.ctx_in = ctx_in;
1395 test_attr.ctx_out = ctx_out;
1396
1397 err = bpf_prog_test_run_opts(fd, &test_attr);
1398 if (err) {
1399 p_err("failed to run program: %s", strerror(errno));
1400 goto free_ctx_out;
1401 }
1402
1403 err = 0;
1404
1405 if (json_output)
1406 jsonw_start_object(json_wtr); /* root */
1407
1408 /* Do not exit on errors occurring when printing output data/context,
1409 * we still want to print return value and duration for program run.
1410 */
1411 if (test_attr.data_size_out)
1412 err += print_run_output(test_attr.data_out,
1413 test_attr.data_size_out,
1414 data_fname_out, "data_out");
1415 if (test_attr.ctx_size_out)
1416 err += print_run_output(test_attr.ctx_out,
1417 test_attr.ctx_size_out,
1418 ctx_fname_out, "ctx_out");
1419
1420 if (json_output) {
1421 jsonw_uint_field(json_wtr, "retval", test_attr.retval);
1422 jsonw_uint_field(json_wtr, "duration", test_attr.duration);
1423 jsonw_end_object(json_wtr); /* root */
1424 } else {
1425 fprintf(stdout, "Return value: %u, duration%s: %uns\n",
1426 test_attr.retval,
1427 repeat > 1 ? " (average)" : "", test_attr.duration);
1428 }
1429
1430 free_ctx_out:
1431 free(ctx_out);
1432 free_ctx_in:
1433 free(ctx_in);
1434 free_data_out:
1435 free(data_out);
1436 free_data_in:
1437 free(data_in);
1438
1439 return err;
1440 }
1441
1442 static int
get_prog_type_by_name(const char * name,enum bpf_prog_type * prog_type,enum bpf_attach_type * expected_attach_type)1443 get_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
1444 enum bpf_attach_type *expected_attach_type)
1445 {
1446 libbpf_print_fn_t print_backup;
1447 int ret;
1448
1449 ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type);
1450 if (!ret)
1451 return ret;
1452
1453 /* libbpf_prog_type_by_name() failed, let's re-run with debug level */
1454 print_backup = libbpf_set_print(print_all_levels);
1455 ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type);
1456 libbpf_set_print(print_backup);
1457
1458 return ret;
1459 }
1460
1461 static int
auto_attach_program(struct bpf_program * prog,const char * path)1462 auto_attach_program(struct bpf_program *prog, const char *path)
1463 {
1464 struct bpf_link *link;
1465 int err;
1466
1467 link = bpf_program__attach(prog);
1468 if (!link) {
1469 p_info("Program %s does not support autoattach, falling back to pinning",
1470 bpf_program__name(prog));
1471 return bpf_obj_pin(bpf_program__fd(prog), path);
1472 }
1473
1474 err = bpf_link__pin(link, path);
1475 bpf_link__destroy(link);
1476 return err;
1477 }
1478
1479 static int
auto_attach_programs(struct bpf_object * obj,const char * path)1480 auto_attach_programs(struct bpf_object *obj, const char *path)
1481 {
1482 struct bpf_program *prog;
1483 char buf[PATH_MAX];
1484 int err;
1485
1486 bpf_object__for_each_program(prog, obj) {
1487 err = pathname_concat(buf, sizeof(buf), path, bpf_program__name(prog));
1488 if (err)
1489 goto err_unpin_programs;
1490
1491 err = auto_attach_program(prog, buf);
1492 if (err)
1493 goto err_unpin_programs;
1494 }
1495
1496 return 0;
1497
1498 err_unpin_programs:
1499 while ((prog = bpf_object__prev_program(obj, prog))) {
1500 if (pathname_concat(buf, sizeof(buf), path, bpf_program__name(prog)))
1501 continue;
1502
1503 bpf_program__unpin(prog, buf);
1504 }
1505
1506 return err;
1507 }
1508
load_with_options(int argc,char ** argv,bool first_prog_only)1509 static int load_with_options(int argc, char **argv, bool first_prog_only)
1510 {
1511 enum bpf_prog_type common_prog_type = BPF_PROG_TYPE_UNSPEC;
1512 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts,
1513 .relaxed_maps = relaxed_maps,
1514 );
1515 enum bpf_attach_type expected_attach_type;
1516 struct map_replace *map_replace = NULL;
1517 struct bpf_program *prog = NULL, *pos;
1518 unsigned int old_map_fds = 0;
1519 const char *pinmaps = NULL;
1520 __u32 xdpmeta_ifindex = 0;
1521 __u32 offload_ifindex = 0;
1522 bool auto_attach = false;
1523 struct bpf_object *obj;
1524 struct bpf_map *map;
1525 const char *pinfile;
1526 unsigned int i, j;
1527 const char *file;
1528 int idx, err;
1529
1530
1531 if (!REQ_ARGS(2))
1532 return -1;
1533 file = GET_ARG();
1534 pinfile = GET_ARG();
1535
1536 while (argc) {
1537 if (is_prefix(*argv, "type")) {
1538 NEXT_ARG();
1539
1540 if (common_prog_type != BPF_PROG_TYPE_UNSPEC) {
1541 p_err("program type already specified");
1542 goto err_free_reuse_maps;
1543 }
1544 if (!REQ_ARGS(1))
1545 goto err_free_reuse_maps;
1546
1547 err = libbpf_prog_type_by_name(*argv, &common_prog_type,
1548 &expected_attach_type);
1549 if (err < 0) {
1550 /* Put a '/' at the end of type to appease libbpf */
1551 char *type = malloc(strlen(*argv) + 2);
1552
1553 if (!type) {
1554 p_err("mem alloc failed");
1555 goto err_free_reuse_maps;
1556 }
1557 *type = 0;
1558 strcat(type, *argv);
1559 strcat(type, "/");
1560
1561 err = get_prog_type_by_name(type, &common_prog_type,
1562 &expected_attach_type);
1563 free(type);
1564 if (err < 0)
1565 goto err_free_reuse_maps;
1566 }
1567
1568 NEXT_ARG();
1569 } else if (is_prefix(*argv, "map")) {
1570 void *new_map_replace;
1571 char *endptr, *name;
1572 int fd;
1573
1574 NEXT_ARG();
1575
1576 if (!REQ_ARGS(4))
1577 goto err_free_reuse_maps;
1578
1579 if (is_prefix(*argv, "idx")) {
1580 NEXT_ARG();
1581
1582 idx = strtoul(*argv, &endptr, 0);
1583 if (*endptr) {
1584 p_err("can't parse %s as IDX", *argv);
1585 goto err_free_reuse_maps;
1586 }
1587 name = NULL;
1588 } else if (is_prefix(*argv, "name")) {
1589 NEXT_ARG();
1590
1591 name = *argv;
1592 idx = -1;
1593 } else {
1594 p_err("expected 'idx' or 'name', got: '%s'?",
1595 *argv);
1596 goto err_free_reuse_maps;
1597 }
1598 NEXT_ARG();
1599
1600 fd = map_parse_fd(&argc, &argv);
1601 if (fd < 0)
1602 goto err_free_reuse_maps;
1603
1604 new_map_replace = libbpf_reallocarray(map_replace,
1605 old_map_fds + 1,
1606 sizeof(*map_replace));
1607 if (!new_map_replace) {
1608 p_err("mem alloc failed");
1609 goto err_free_reuse_maps;
1610 }
1611 map_replace = new_map_replace;
1612
1613 map_replace[old_map_fds].idx = idx;
1614 map_replace[old_map_fds].name = name;
1615 map_replace[old_map_fds].fd = fd;
1616 old_map_fds++;
1617 } else if (is_prefix(*argv, "dev")) {
1618 p_info("Warning: 'bpftool prog load [...] dev <ifname>' syntax is deprecated.\n"
1619 "Going further, please use 'offload_dev <ifname>' to offload program to device.\n"
1620 "For applications using XDP hints only, use 'xdpmeta_dev <ifname>'.");
1621 goto offload_dev;
1622 } else if (is_prefix(*argv, "offload_dev")) {
1623 offload_dev:
1624 NEXT_ARG();
1625
1626 if (offload_ifindex) {
1627 p_err("offload_dev already specified");
1628 goto err_free_reuse_maps;
1629 } else if (xdpmeta_ifindex) {
1630 p_err("xdpmeta_dev and offload_dev are mutually exclusive");
1631 goto err_free_reuse_maps;
1632 }
1633 if (!REQ_ARGS(1))
1634 goto err_free_reuse_maps;
1635
1636 offload_ifindex = if_nametoindex(*argv);
1637 if (!offload_ifindex) {
1638 p_err("unrecognized netdevice '%s': %s",
1639 *argv, strerror(errno));
1640 goto err_free_reuse_maps;
1641 }
1642 NEXT_ARG();
1643 } else if (is_prefix(*argv, "xdpmeta_dev")) {
1644 NEXT_ARG();
1645
1646 if (xdpmeta_ifindex) {
1647 p_err("xdpmeta_dev already specified");
1648 goto err_free_reuse_maps;
1649 } else if (offload_ifindex) {
1650 p_err("xdpmeta_dev and offload_dev are mutually exclusive");
1651 goto err_free_reuse_maps;
1652 }
1653 if (!REQ_ARGS(1))
1654 goto err_free_reuse_maps;
1655
1656 xdpmeta_ifindex = if_nametoindex(*argv);
1657 if (!xdpmeta_ifindex) {
1658 p_err("unrecognized netdevice '%s': %s",
1659 *argv, strerror(errno));
1660 goto err_free_reuse_maps;
1661 }
1662 NEXT_ARG();
1663 } else if (is_prefix(*argv, "pinmaps")) {
1664 NEXT_ARG();
1665
1666 if (!REQ_ARGS(1))
1667 goto err_free_reuse_maps;
1668
1669 pinmaps = GET_ARG();
1670 } else if (is_prefix(*argv, "autoattach")) {
1671 auto_attach = true;
1672 NEXT_ARG();
1673 } else {
1674 p_err("expected no more arguments, 'type', 'map' or 'dev', got: '%s'?",
1675 *argv);
1676 goto err_free_reuse_maps;
1677 }
1678 }
1679
1680 set_max_rlimit();
1681
1682 if (verifier_logs)
1683 /* log_level1 + log_level2 + stats, but not stable UAPI */
1684 open_opts.kernel_log_level = 1 + 2 + 4;
1685
1686 obj = bpf_object__open_file(file, &open_opts);
1687 if (!obj) {
1688 p_err("failed to open object file");
1689 goto err_free_reuse_maps;
1690 }
1691
1692 bpf_object__for_each_program(pos, obj) {
1693 enum bpf_prog_type prog_type = common_prog_type;
1694
1695 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
1696 const char *sec_name = bpf_program__section_name(pos);
1697
1698 err = get_prog_type_by_name(sec_name, &prog_type,
1699 &expected_attach_type);
1700 if (err < 0)
1701 goto err_close_obj;
1702 }
1703
1704 if (prog_type == BPF_PROG_TYPE_XDP && xdpmeta_ifindex) {
1705 bpf_program__set_flags(pos, BPF_F_XDP_DEV_BOUND_ONLY);
1706 bpf_program__set_ifindex(pos, xdpmeta_ifindex);
1707 } else {
1708 bpf_program__set_ifindex(pos, offload_ifindex);
1709 }
1710 if (bpf_program__type(pos) != prog_type)
1711 bpf_program__set_type(pos, prog_type);
1712 bpf_program__set_expected_attach_type(pos, expected_attach_type);
1713 }
1714
1715 qsort(map_replace, old_map_fds, sizeof(*map_replace),
1716 map_replace_compar);
1717
1718 /* After the sort maps by name will be first on the list, because they
1719 * have idx == -1. Resolve them.
1720 */
1721 j = 0;
1722 while (j < old_map_fds && map_replace[j].name) {
1723 i = 0;
1724 bpf_object__for_each_map(map, obj) {
1725 if (!strcmp(bpf_map__name(map), map_replace[j].name)) {
1726 map_replace[j].idx = i;
1727 break;
1728 }
1729 i++;
1730 }
1731 if (map_replace[j].idx == -1) {
1732 p_err("unable to find map '%s'", map_replace[j].name);
1733 goto err_close_obj;
1734 }
1735 j++;
1736 }
1737 /* Resort if any names were resolved */
1738 if (j)
1739 qsort(map_replace, old_map_fds, sizeof(*map_replace),
1740 map_replace_compar);
1741
1742 /* Set ifindex and name reuse */
1743 j = 0;
1744 idx = 0;
1745 bpf_object__for_each_map(map, obj) {
1746 if (bpf_map__type(map) != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1747 bpf_map__set_ifindex(map, offload_ifindex);
1748
1749 if (j < old_map_fds && idx == map_replace[j].idx) {
1750 err = bpf_map__reuse_fd(map, map_replace[j++].fd);
1751 if (err) {
1752 p_err("unable to set up map reuse: %d", err);
1753 goto err_close_obj;
1754 }
1755
1756 /* Next reuse wants to apply to the same map */
1757 if (j < old_map_fds && map_replace[j].idx == idx) {
1758 p_err("replacement for map idx %d specified more than once",
1759 idx);
1760 goto err_close_obj;
1761 }
1762 }
1763
1764 idx++;
1765 }
1766 if (j < old_map_fds) {
1767 p_err("map idx '%d' not used", map_replace[j].idx);
1768 goto err_close_obj;
1769 }
1770
1771 err = bpf_object__load(obj);
1772 if (err) {
1773 p_err("failed to load object file");
1774 goto err_close_obj;
1775 }
1776
1777 if (first_prog_only)
1778 err = mount_bpffs_for_file(pinfile);
1779 else
1780 err = create_and_mount_bpffs_dir(pinfile);
1781 if (err)
1782 goto err_close_obj;
1783
1784 if (first_prog_only) {
1785 prog = bpf_object__next_program(obj, NULL);
1786 if (!prog) {
1787 p_err("object file doesn't contain any bpf program");
1788 goto err_close_obj;
1789 }
1790
1791 if (auto_attach)
1792 err = auto_attach_program(prog, pinfile);
1793 else
1794 err = bpf_obj_pin(bpf_program__fd(prog), pinfile);
1795 if (err) {
1796 p_err("failed to pin program %s",
1797 bpf_program__section_name(prog));
1798 goto err_close_obj;
1799 }
1800 } else {
1801 if (auto_attach)
1802 err = auto_attach_programs(obj, pinfile);
1803 else
1804 err = bpf_object__pin_programs(obj, pinfile);
1805 if (err) {
1806 p_err("failed to pin all programs");
1807 goto err_close_obj;
1808 }
1809 }
1810
1811 if (pinmaps) {
1812 err = bpf_object__pin_maps(obj, pinmaps);
1813 if (err) {
1814 p_err("failed to pin all maps");
1815 goto err_unpin;
1816 }
1817 }
1818
1819 if (json_output)
1820 jsonw_null(json_wtr);
1821
1822 bpf_object__close(obj);
1823 for (i = 0; i < old_map_fds; i++)
1824 close(map_replace[i].fd);
1825 free(map_replace);
1826
1827 return 0;
1828
1829 err_unpin:
1830 if (first_prog_only)
1831 unlink(pinfile);
1832 else
1833 bpf_object__unpin_programs(obj, pinfile);
1834 err_close_obj:
1835 bpf_object__close(obj);
1836 err_free_reuse_maps:
1837 for (i = 0; i < old_map_fds; i++)
1838 close(map_replace[i].fd);
1839 free(map_replace);
1840 return -1;
1841 }
1842
count_open_fds(void)1843 static int count_open_fds(void)
1844 {
1845 DIR *dp = opendir("/proc/self/fd");
1846 struct dirent *de;
1847 int cnt = -3;
1848
1849 if (!dp)
1850 return -1;
1851
1852 while ((de = readdir(dp)))
1853 cnt++;
1854
1855 closedir(dp);
1856 return cnt;
1857 }
1858
try_loader(struct gen_loader_opts * gen)1859 static int try_loader(struct gen_loader_opts *gen)
1860 {
1861 struct bpf_load_and_run_opts opts = {};
1862 struct bpf_loader_ctx *ctx;
1863 int ctx_sz = sizeof(*ctx) + 64 * max(sizeof(struct bpf_map_desc),
1864 sizeof(struct bpf_prog_desc));
1865 int log_buf_sz = (1u << 24) - 1;
1866 int err, fds_before, fd_delta;
1867 char *log_buf = NULL;
1868
1869 ctx = alloca(ctx_sz);
1870 memset(ctx, 0, ctx_sz);
1871 ctx->sz = ctx_sz;
1872 if (verifier_logs) {
1873 ctx->log_level = 1 + 2 + 4;
1874 ctx->log_size = log_buf_sz;
1875 log_buf = malloc(log_buf_sz);
1876 if (!log_buf)
1877 return -ENOMEM;
1878 ctx->log_buf = (long) log_buf;
1879 }
1880 opts.ctx = ctx;
1881 opts.data = gen->data;
1882 opts.data_sz = gen->data_sz;
1883 opts.insns = gen->insns;
1884 opts.insns_sz = gen->insns_sz;
1885 fds_before = count_open_fds();
1886 err = bpf_load_and_run(&opts);
1887 fd_delta = count_open_fds() - fds_before;
1888 if (err < 0 || verifier_logs) {
1889 fprintf(stderr, "err %d\n%s\n%s", err, opts.errstr, log_buf);
1890 if (fd_delta && err < 0)
1891 fprintf(stderr, "loader prog leaked %d FDs\n",
1892 fd_delta);
1893 }
1894 free(log_buf);
1895 return err;
1896 }
1897
do_loader(int argc,char ** argv)1898 static int do_loader(int argc, char **argv)
1899 {
1900 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts);
1901 DECLARE_LIBBPF_OPTS(gen_loader_opts, gen);
1902 struct bpf_object *obj;
1903 const char *file;
1904 int err = 0;
1905
1906 if (!REQ_ARGS(1))
1907 return -1;
1908 file = GET_ARG();
1909
1910 if (verifier_logs)
1911 /* log_level1 + log_level2 + stats, but not stable UAPI */
1912 open_opts.kernel_log_level = 1 + 2 + 4;
1913
1914 obj = bpf_object__open_file(file, &open_opts);
1915 if (!obj) {
1916 p_err("failed to open object file");
1917 goto err_close_obj;
1918 }
1919
1920 err = bpf_object__gen_loader(obj, &gen);
1921 if (err)
1922 goto err_close_obj;
1923
1924 err = bpf_object__load(obj);
1925 if (err) {
1926 p_err("failed to load object file");
1927 goto err_close_obj;
1928 }
1929
1930 if (verifier_logs) {
1931 struct dump_data dd = {};
1932
1933 kernel_syms_load(&dd);
1934 dump_xlated_plain(&dd, (void *)gen.insns, gen.insns_sz, false, false);
1935 kernel_syms_destroy(&dd);
1936 }
1937 err = try_loader(&gen);
1938 err_close_obj:
1939 bpf_object__close(obj);
1940 return err;
1941 }
1942
do_load(int argc,char ** argv)1943 static int do_load(int argc, char **argv)
1944 {
1945 if (use_loader)
1946 return do_loader(argc, argv);
1947 return load_with_options(argc, argv, true);
1948 }
1949
do_loadall(int argc,char ** argv)1950 static int do_loadall(int argc, char **argv)
1951 {
1952 return load_with_options(argc, argv, false);
1953 }
1954
1955 #ifdef BPFTOOL_WITHOUT_SKELETONS
1956
do_profile(int argc,char ** argv)1957 static int do_profile(int argc, char **argv)
1958 {
1959 p_err("bpftool prog profile command is not supported. Please build bpftool with clang >= 10.0.0");
1960 return 0;
1961 }
1962
1963 #else /* BPFTOOL_WITHOUT_SKELETONS */
1964
1965 #include "profiler.skel.h"
1966
1967 struct profile_metric {
1968 const char *name;
1969 struct bpf_perf_event_value val;
1970 struct perf_event_attr attr;
1971 bool selected;
1972
1973 /* calculate ratios like instructions per cycle */
1974 const int ratio_metric; /* 0 for N/A, 1 for index 0 (cycles) */
1975 const char *ratio_desc;
1976 const float ratio_mul;
1977 } metrics[] = {
1978 {
1979 .name = "cycles",
1980 .attr = {
1981 .type = PERF_TYPE_HARDWARE,
1982 .config = PERF_COUNT_HW_CPU_CYCLES,
1983 .exclude_user = 1,
1984 },
1985 },
1986 {
1987 .name = "instructions",
1988 .attr = {
1989 .type = PERF_TYPE_HARDWARE,
1990 .config = PERF_COUNT_HW_INSTRUCTIONS,
1991 .exclude_user = 1,
1992 },
1993 .ratio_metric = 1,
1994 .ratio_desc = "insns per cycle",
1995 .ratio_mul = 1.0,
1996 },
1997 {
1998 .name = "l1d_loads",
1999 .attr = {
2000 .type = PERF_TYPE_HW_CACHE,
2001 .config =
2002 PERF_COUNT_HW_CACHE_L1D |
2003 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2004 (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16),
2005 .exclude_user = 1,
2006 },
2007 },
2008 {
2009 .name = "llc_misses",
2010 .attr = {
2011 .type = PERF_TYPE_HW_CACHE,
2012 .config =
2013 PERF_COUNT_HW_CACHE_LL |
2014 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2015 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2016 .exclude_user = 1
2017 },
2018 .ratio_metric = 2,
2019 .ratio_desc = "LLC misses per million insns",
2020 .ratio_mul = 1e6,
2021 },
2022 {
2023 .name = "itlb_misses",
2024 .attr = {
2025 .type = PERF_TYPE_HW_CACHE,
2026 .config =
2027 PERF_COUNT_HW_CACHE_ITLB |
2028 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2029 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2030 .exclude_user = 1
2031 },
2032 .ratio_metric = 2,
2033 .ratio_desc = "itlb misses per million insns",
2034 .ratio_mul = 1e6,
2035 },
2036 {
2037 .name = "dtlb_misses",
2038 .attr = {
2039 .type = PERF_TYPE_HW_CACHE,
2040 .config =
2041 PERF_COUNT_HW_CACHE_DTLB |
2042 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2043 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2044 .exclude_user = 1
2045 },
2046 .ratio_metric = 2,
2047 .ratio_desc = "dtlb misses per million insns",
2048 .ratio_mul = 1e6,
2049 },
2050 };
2051
2052 static __u64 profile_total_count;
2053
2054 #define MAX_NUM_PROFILE_METRICS 4
2055
profile_parse_metrics(int argc,char ** argv)2056 static int profile_parse_metrics(int argc, char **argv)
2057 {
2058 unsigned int metric_cnt;
2059 int selected_cnt = 0;
2060 unsigned int i;
2061
2062 metric_cnt = ARRAY_SIZE(metrics);
2063
2064 while (argc > 0) {
2065 for (i = 0; i < metric_cnt; i++) {
2066 if (is_prefix(argv[0], metrics[i].name)) {
2067 if (!metrics[i].selected)
2068 selected_cnt++;
2069 metrics[i].selected = true;
2070 break;
2071 }
2072 }
2073 if (i == metric_cnt) {
2074 p_err("unknown metric %s", argv[0]);
2075 return -1;
2076 }
2077 NEXT_ARG();
2078 }
2079 if (selected_cnt > MAX_NUM_PROFILE_METRICS) {
2080 p_err("too many (%d) metrics, please specify no more than %d metrics at at time",
2081 selected_cnt, MAX_NUM_PROFILE_METRICS);
2082 return -1;
2083 }
2084 return selected_cnt;
2085 }
2086
profile_read_values(struct profiler_bpf * obj)2087 static void profile_read_values(struct profiler_bpf *obj)
2088 {
2089 __u32 m, cpu, num_cpu = obj->rodata->num_cpu;
2090 int reading_map_fd, count_map_fd;
2091 __u64 counts[num_cpu];
2092 __u32 key = 0;
2093 int err;
2094
2095 reading_map_fd = bpf_map__fd(obj->maps.accum_readings);
2096 count_map_fd = bpf_map__fd(obj->maps.counts);
2097 if (reading_map_fd < 0 || count_map_fd < 0) {
2098 p_err("failed to get fd for map");
2099 return;
2100 }
2101
2102 err = bpf_map_lookup_elem(count_map_fd, &key, counts);
2103 if (err) {
2104 p_err("failed to read count_map: %s", strerror(errno));
2105 return;
2106 }
2107
2108 profile_total_count = 0;
2109 for (cpu = 0; cpu < num_cpu; cpu++)
2110 profile_total_count += counts[cpu];
2111
2112 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2113 struct bpf_perf_event_value values[num_cpu];
2114
2115 if (!metrics[m].selected)
2116 continue;
2117
2118 err = bpf_map_lookup_elem(reading_map_fd, &key, values);
2119 if (err) {
2120 p_err("failed to read reading_map: %s",
2121 strerror(errno));
2122 return;
2123 }
2124 for (cpu = 0; cpu < num_cpu; cpu++) {
2125 metrics[m].val.counter += values[cpu].counter;
2126 metrics[m].val.enabled += values[cpu].enabled;
2127 metrics[m].val.running += values[cpu].running;
2128 }
2129 key++;
2130 }
2131 }
2132
profile_print_readings_json(void)2133 static void profile_print_readings_json(void)
2134 {
2135 __u32 m;
2136
2137 jsonw_start_array(json_wtr);
2138 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2139 if (!metrics[m].selected)
2140 continue;
2141 jsonw_start_object(json_wtr);
2142 jsonw_string_field(json_wtr, "metric", metrics[m].name);
2143 jsonw_lluint_field(json_wtr, "run_cnt", profile_total_count);
2144 jsonw_lluint_field(json_wtr, "value", metrics[m].val.counter);
2145 jsonw_lluint_field(json_wtr, "enabled", metrics[m].val.enabled);
2146 jsonw_lluint_field(json_wtr, "running", metrics[m].val.running);
2147
2148 jsonw_end_object(json_wtr);
2149 }
2150 jsonw_end_array(json_wtr);
2151 }
2152
profile_print_readings_plain(void)2153 static void profile_print_readings_plain(void)
2154 {
2155 __u32 m;
2156
2157 printf("\n%18llu %-20s\n", profile_total_count, "run_cnt");
2158 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2159 struct bpf_perf_event_value *val = &metrics[m].val;
2160 int r;
2161
2162 if (!metrics[m].selected)
2163 continue;
2164 printf("%18llu %-20s", val->counter, metrics[m].name);
2165
2166 r = metrics[m].ratio_metric - 1;
2167 if (r >= 0 && metrics[r].selected &&
2168 metrics[r].val.counter > 0) {
2169 printf("# %8.2f %-30s",
2170 val->counter * metrics[m].ratio_mul /
2171 metrics[r].val.counter,
2172 metrics[m].ratio_desc);
2173 } else {
2174 printf("%-41s", "");
2175 }
2176
2177 if (val->enabled > val->running)
2178 printf("(%4.2f%%)",
2179 val->running * 100.0 / val->enabled);
2180 printf("\n");
2181 }
2182 }
2183
profile_print_readings(void)2184 static void profile_print_readings(void)
2185 {
2186 if (json_output)
2187 profile_print_readings_json();
2188 else
2189 profile_print_readings_plain();
2190 }
2191
profile_target_name(int tgt_fd)2192 static char *profile_target_name(int tgt_fd)
2193 {
2194 struct bpf_func_info func_info;
2195 struct bpf_prog_info info = {};
2196 __u32 info_len = sizeof(info);
2197 const struct btf_type *t;
2198 __u32 func_info_rec_size;
2199 struct btf *btf = NULL;
2200 char *name = NULL;
2201 int err;
2202
2203 err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len);
2204 if (err) {
2205 p_err("failed to get info for prog FD %d", tgt_fd);
2206 goto out;
2207 }
2208
2209 if (info.btf_id == 0) {
2210 p_err("prog FD %d doesn't have valid btf", tgt_fd);
2211 goto out;
2212 }
2213
2214 func_info_rec_size = info.func_info_rec_size;
2215 if (info.nr_func_info == 0) {
2216 p_err("found 0 func_info for prog FD %d", tgt_fd);
2217 goto out;
2218 }
2219
2220 memset(&info, 0, sizeof(info));
2221 info.nr_func_info = 1;
2222 info.func_info_rec_size = func_info_rec_size;
2223 info.func_info = ptr_to_u64(&func_info);
2224
2225 err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len);
2226 if (err) {
2227 p_err("failed to get func_info for prog FD %d", tgt_fd);
2228 goto out;
2229 }
2230
2231 btf = btf__load_from_kernel_by_id(info.btf_id);
2232 if (!btf) {
2233 p_err("failed to load btf for prog FD %d", tgt_fd);
2234 goto out;
2235 }
2236
2237 t = btf__type_by_id(btf, func_info.type_id);
2238 if (!t) {
2239 p_err("btf %d doesn't have type %d",
2240 info.btf_id, func_info.type_id);
2241 goto out;
2242 }
2243 name = strdup(btf__name_by_offset(btf, t->name_off));
2244 out:
2245 btf__free(btf);
2246 return name;
2247 }
2248
2249 static struct profiler_bpf *profile_obj;
2250 static int profile_tgt_fd = -1;
2251 static char *profile_tgt_name;
2252 static int *profile_perf_events;
2253 static int profile_perf_event_cnt;
2254
profile_close_perf_events(struct profiler_bpf * obj)2255 static void profile_close_perf_events(struct profiler_bpf *obj)
2256 {
2257 int i;
2258
2259 for (i = profile_perf_event_cnt - 1; i >= 0; i--)
2260 close(profile_perf_events[i]);
2261
2262 free(profile_perf_events);
2263 profile_perf_event_cnt = 0;
2264 }
2265
profile_open_perf_event(int mid,int cpu,int map_fd)2266 static int profile_open_perf_event(int mid, int cpu, int map_fd)
2267 {
2268 int pmu_fd;
2269
2270 pmu_fd = syscall(__NR_perf_event_open, &metrics[mid].attr,
2271 -1 /*pid*/, cpu, -1 /*group_fd*/, 0);
2272 if (pmu_fd < 0) {
2273 if (errno == ENODEV) {
2274 p_info("cpu %d may be offline, skip %s profiling.",
2275 cpu, metrics[mid].name);
2276 profile_perf_event_cnt++;
2277 return 0;
2278 }
2279 return -1;
2280 }
2281
2282 if (bpf_map_update_elem(map_fd,
2283 &profile_perf_event_cnt,
2284 &pmu_fd, BPF_ANY) ||
2285 ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0)) {
2286 close(pmu_fd);
2287 return -1;
2288 }
2289
2290 profile_perf_events[profile_perf_event_cnt++] = pmu_fd;
2291 return 0;
2292 }
2293
profile_open_perf_events(struct profiler_bpf * obj)2294 static int profile_open_perf_events(struct profiler_bpf *obj)
2295 {
2296 unsigned int cpu, m;
2297 int map_fd;
2298
2299 profile_perf_events = calloc(
2300 obj->rodata->num_cpu * obj->rodata->num_metric, sizeof(int));
2301 if (!profile_perf_events) {
2302 p_err("failed to allocate memory for perf_event array: %s",
2303 strerror(errno));
2304 return -1;
2305 }
2306 map_fd = bpf_map__fd(obj->maps.events);
2307 if (map_fd < 0) {
2308 p_err("failed to get fd for events map");
2309 return -1;
2310 }
2311
2312 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2313 if (!metrics[m].selected)
2314 continue;
2315 for (cpu = 0; cpu < obj->rodata->num_cpu; cpu++) {
2316 if (profile_open_perf_event(m, cpu, map_fd)) {
2317 p_err("failed to create event %s on cpu %d",
2318 metrics[m].name, cpu);
2319 return -1;
2320 }
2321 }
2322 }
2323 return 0;
2324 }
2325
profile_print_and_cleanup(void)2326 static void profile_print_and_cleanup(void)
2327 {
2328 profile_close_perf_events(profile_obj);
2329 profile_read_values(profile_obj);
2330 profile_print_readings();
2331 profiler_bpf__destroy(profile_obj);
2332
2333 close(profile_tgt_fd);
2334 free(profile_tgt_name);
2335 }
2336
int_exit(int signo)2337 static void int_exit(int signo)
2338 {
2339 profile_print_and_cleanup();
2340 exit(0);
2341 }
2342
do_profile(int argc,char ** argv)2343 static int do_profile(int argc, char **argv)
2344 {
2345 int num_metric, num_cpu, err = -1;
2346 struct bpf_program *prog;
2347 unsigned long duration;
2348 char *endptr;
2349
2350 /* we at least need two args for the prog and one metric */
2351 if (!REQ_ARGS(3))
2352 return -EINVAL;
2353
2354 /* parse target fd */
2355 profile_tgt_fd = prog_parse_fd(&argc, &argv);
2356 if (profile_tgt_fd < 0) {
2357 p_err("failed to parse fd");
2358 return -1;
2359 }
2360
2361 /* parse profiling optional duration */
2362 if (argc > 2 && is_prefix(argv[0], "duration")) {
2363 NEXT_ARG();
2364 duration = strtoul(*argv, &endptr, 0);
2365 if (*endptr)
2366 usage();
2367 NEXT_ARG();
2368 } else {
2369 duration = UINT_MAX;
2370 }
2371
2372 num_metric = profile_parse_metrics(argc, argv);
2373 if (num_metric <= 0)
2374 goto out;
2375
2376 num_cpu = libbpf_num_possible_cpus();
2377 if (num_cpu <= 0) {
2378 p_err("failed to identify number of CPUs");
2379 goto out;
2380 }
2381
2382 profile_obj = profiler_bpf__open();
2383 if (!profile_obj) {
2384 p_err("failed to open and/or load BPF object");
2385 goto out;
2386 }
2387
2388 profile_obj->rodata->num_cpu = num_cpu;
2389 profile_obj->rodata->num_metric = num_metric;
2390
2391 /* adjust map sizes */
2392 bpf_map__set_max_entries(profile_obj->maps.events, num_metric * num_cpu);
2393 bpf_map__set_max_entries(profile_obj->maps.fentry_readings, num_metric);
2394 bpf_map__set_max_entries(profile_obj->maps.accum_readings, num_metric);
2395 bpf_map__set_max_entries(profile_obj->maps.counts, 1);
2396
2397 /* change target name */
2398 profile_tgt_name = profile_target_name(profile_tgt_fd);
2399 if (!profile_tgt_name)
2400 goto out;
2401
2402 bpf_object__for_each_program(prog, profile_obj->obj) {
2403 err = bpf_program__set_attach_target(prog, profile_tgt_fd,
2404 profile_tgt_name);
2405 if (err) {
2406 p_err("failed to set attach target\n");
2407 goto out;
2408 }
2409 }
2410
2411 set_max_rlimit();
2412 err = profiler_bpf__load(profile_obj);
2413 if (err) {
2414 p_err("failed to load profile_obj");
2415 goto out;
2416 }
2417
2418 err = profile_open_perf_events(profile_obj);
2419 if (err)
2420 goto out;
2421
2422 err = profiler_bpf__attach(profile_obj);
2423 if (err) {
2424 p_err("failed to attach profile_obj");
2425 goto out;
2426 }
2427 signal(SIGINT, int_exit);
2428
2429 sleep(duration);
2430 profile_print_and_cleanup();
2431 return 0;
2432
2433 out:
2434 profile_close_perf_events(profile_obj);
2435 if (profile_obj)
2436 profiler_bpf__destroy(profile_obj);
2437 close(profile_tgt_fd);
2438 free(profile_tgt_name);
2439 return err;
2440 }
2441
2442 #endif /* BPFTOOL_WITHOUT_SKELETONS */
2443
do_help(int argc,char ** argv)2444 static int do_help(int argc, char **argv)
2445 {
2446 if (json_output) {
2447 jsonw_null(json_wtr);
2448 return 0;
2449 }
2450
2451 fprintf(stderr,
2452 "Usage: %1$s %2$s { show | list } [PROG]\n"
2453 " %1$s %2$s dump xlated PROG [{ file FILE | [opcodes] [linum] [visual] }]\n"
2454 " %1$s %2$s dump jited PROG [{ file FILE | [opcodes] [linum] }]\n"
2455 " %1$s %2$s pin PROG FILE\n"
2456 " %1$s %2$s { load | loadall } OBJ PATH \\\n"
2457 " [type TYPE] [{ offload_dev | xdpmeta_dev } NAME] \\\n"
2458 " [map { idx IDX | name NAME } MAP]\\\n"
2459 " [pinmaps MAP_DIR]\n"
2460 " [autoattach]\n"
2461 " %1$s %2$s attach PROG ATTACH_TYPE [MAP]\n"
2462 " %1$s %2$s detach PROG ATTACH_TYPE [MAP]\n"
2463 " %1$s %2$s run PROG \\\n"
2464 " data_in FILE \\\n"
2465 " [data_out FILE [data_size_out L]] \\\n"
2466 " [ctx_in FILE [ctx_out FILE [ctx_size_out M]]] \\\n"
2467 " [repeat N]\n"
2468 " %1$s %2$s profile PROG [duration DURATION] METRICs\n"
2469 " %1$s %2$s tracelog\n"
2470 " %1$s %2$s help\n"
2471 "\n"
2472 " " HELP_SPEC_MAP "\n"
2473 " " HELP_SPEC_PROGRAM "\n"
2474 " TYPE := { socket | kprobe | kretprobe | classifier | action |\n"
2475 " tracepoint | raw_tracepoint | xdp | perf_event | cgroup/skb |\n"
2476 " cgroup/sock | cgroup/dev | lwt_in | lwt_out | lwt_xmit |\n"
2477 " lwt_seg6local | sockops | sk_skb | sk_msg | lirc_mode2 |\n"
2478 " sk_reuseport | flow_dissector | cgroup/sysctl |\n"
2479 " cgroup/bind4 | cgroup/bind6 | cgroup/post_bind4 |\n"
2480 " cgroup/post_bind6 | cgroup/connect4 | cgroup/connect6 |\n"
2481 " cgroup/getpeername4 | cgroup/getpeername6 |\n"
2482 " cgroup/getsockname4 | cgroup/getsockname6 | cgroup/sendmsg4 |\n"
2483 " cgroup/sendmsg6 | cgroup/recvmsg4 | cgroup/recvmsg6 |\n"
2484 " cgroup/getsockopt | cgroup/setsockopt | cgroup/sock_release |\n"
2485 " struct_ops | fentry | fexit | freplace | sk_lookup }\n"
2486 " ATTACH_TYPE := { sk_msg_verdict | sk_skb_verdict | sk_skb_stream_verdict |\n"
2487 " sk_skb_stream_parser | flow_dissector }\n"
2488 " METRIC := { cycles | instructions | l1d_loads | llc_misses | itlb_misses | dtlb_misses }\n"
2489 " " HELP_SPEC_OPTIONS " |\n"
2490 " {-f|--bpffs} | {-m|--mapcompat} | {-n|--nomount} |\n"
2491 " {-L|--use-loader} }\n"
2492 "",
2493 bin_name, argv[-2]);
2494
2495 return 0;
2496 }
2497
2498 static const struct cmd cmds[] = {
2499 { "show", do_show },
2500 { "list", do_show },
2501 { "help", do_help },
2502 { "dump", do_dump },
2503 { "pin", do_pin },
2504 { "load", do_load },
2505 { "loadall", do_loadall },
2506 { "attach", do_attach },
2507 { "detach", do_detach },
2508 { "tracelog", do_tracelog },
2509 { "run", do_run },
2510 { "profile", do_profile },
2511 { 0 }
2512 };
2513
do_prog(int argc,char ** argv)2514 int do_prog(int argc, char **argv)
2515 {
2516 return cmd_select(cmds, argc, argv, do_help);
2517 }
2518