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,bool orphaned)445 static void print_prog_json(struct bpf_prog_info *info, int fd, bool orphaned)
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_bool_field(json_wtr, "orphaned", orphaned);
465 jsonw_uint_field(json_wtr, "bytes_xlated", info->xlated_prog_len);
466
467 if (info->jited_prog_len) {
468 jsonw_bool_field(json_wtr, "jited", true);
469 jsonw_uint_field(json_wtr, "bytes_jited", info->jited_prog_len);
470 } else {
471 jsonw_bool_field(json_wtr, "jited", false);
472 }
473
474 memlock = get_fdinfo(fd, "memlock");
475 if (memlock)
476 jsonw_int_field(json_wtr, "bytes_memlock", atoll(memlock));
477 free(memlock);
478
479 if (info->nr_map_ids)
480 show_prog_maps(fd, info->nr_map_ids);
481
482 if (info->btf_id)
483 jsonw_int_field(json_wtr, "btf_id", info->btf_id);
484
485 if (!hashmap__empty(prog_table)) {
486 struct hashmap_entry *entry;
487
488 jsonw_name(json_wtr, "pinned");
489 jsonw_start_array(json_wtr);
490 hashmap__for_each_key_entry(prog_table, entry, info->id)
491 jsonw_string(json_wtr, entry->pvalue);
492 jsonw_end_array(json_wtr);
493 }
494
495 emit_obj_refs_json(refs_table, info->id, json_wtr);
496
497 show_prog_metadata(fd, info->nr_map_ids);
498
499 jsonw_end_object(json_wtr);
500 }
501
print_prog_header_plain(struct bpf_prog_info * info,int fd)502 static void print_prog_header_plain(struct bpf_prog_info *info, int fd)
503 {
504 const char *prog_type_str;
505 char prog_name[MAX_PROG_FULL_NAME];
506
507 printf("%u: ", info->id);
508 prog_type_str = libbpf_bpf_prog_type_str(info->type);
509 if (prog_type_str)
510 printf("%s ", prog_type_str);
511 else
512 printf("type %u ", info->type);
513
514 if (*info->name) {
515 get_prog_full_name(info, fd, prog_name, sizeof(prog_name));
516 printf("name %s ", prog_name);
517 }
518
519 printf("tag ");
520 fprint_hex(stdout, info->tag, BPF_TAG_SIZE, "");
521 print_dev_plain(info->ifindex, info->netns_dev, info->netns_ino);
522 printf("%s", info->gpl_compatible ? " gpl" : "");
523 if (info->run_time_ns)
524 printf(" run_time_ns %llu run_cnt %llu",
525 info->run_time_ns, info->run_cnt);
526 if (info->recursion_misses)
527 printf(" recursion_misses %llu", info->recursion_misses);
528 printf("\n");
529 }
530
print_prog_plain(struct bpf_prog_info * info,int fd,bool orphaned)531 static void print_prog_plain(struct bpf_prog_info *info, int fd, bool orphaned)
532 {
533 char *memlock;
534
535 print_prog_header_plain(info, fd);
536
537 if (info->load_time) {
538 char buf[32];
539
540 print_boot_time(info->load_time, buf, sizeof(buf));
541
542 /* Piggy back on load_time, since 0 uid is a valid one */
543 printf("\tloaded_at %s uid %u\n", buf, info->created_by_uid);
544 }
545
546 printf("\txlated %uB", info->xlated_prog_len);
547
548 if (info->jited_prog_len)
549 printf(" jited %uB", info->jited_prog_len);
550 else
551 printf(" not jited");
552
553 memlock = get_fdinfo(fd, "memlock");
554 if (memlock)
555 printf(" memlock %sB", memlock);
556 free(memlock);
557
558 if (orphaned)
559 printf(" orphaned");
560
561 if (info->nr_map_ids)
562 show_prog_maps(fd, info->nr_map_ids);
563
564 if (!hashmap__empty(prog_table)) {
565 struct hashmap_entry *entry;
566
567 hashmap__for_each_key_entry(prog_table, entry, info->id)
568 printf("\n\tpinned %s", (char *)entry->pvalue);
569 }
570
571 if (info->btf_id)
572 printf("\n\tbtf_id %u", info->btf_id);
573
574 emit_obj_refs_plain(refs_table, info->id, "\n\tpids ");
575
576 printf("\n");
577
578 show_prog_metadata(fd, info->nr_map_ids);
579 }
580
show_prog(int fd)581 static int show_prog(int fd)
582 {
583 struct bpf_prog_info info = {};
584 __u32 len = sizeof(info);
585 int err;
586
587 err = bpf_prog_get_info_by_fd(fd, &info, &len);
588 if (err && err != -ENODEV) {
589 p_err("can't get prog info: %s", strerror(errno));
590 return -1;
591 }
592
593 if (json_output)
594 print_prog_json(&info, fd, err == -ENODEV);
595 else
596 print_prog_plain(&info, fd, err == -ENODEV);
597
598 return 0;
599 }
600
do_show_subset(int argc,char ** argv)601 static int do_show_subset(int argc, char **argv)
602 {
603 int *fds = NULL;
604 int nb_fds, i;
605 int err = -1;
606
607 fds = malloc(sizeof(int));
608 if (!fds) {
609 p_err("mem alloc failed");
610 return -1;
611 }
612 nb_fds = prog_parse_fds(&argc, &argv, &fds);
613 if (nb_fds < 1)
614 goto exit_free;
615
616 if (json_output && nb_fds > 1)
617 jsonw_start_array(json_wtr); /* root array */
618 for (i = 0; i < nb_fds; i++) {
619 err = show_prog(fds[i]);
620 if (err) {
621 for (; i < nb_fds; i++)
622 close(fds[i]);
623 break;
624 }
625 close(fds[i]);
626 }
627 if (json_output && nb_fds > 1)
628 jsonw_end_array(json_wtr); /* root array */
629
630 exit_free:
631 free(fds);
632 return err;
633 }
634
do_show(int argc,char ** argv)635 static int do_show(int argc, char **argv)
636 {
637 __u32 id = 0;
638 int err;
639 int fd;
640
641 if (show_pinned) {
642 prog_table = hashmap__new(hash_fn_for_key_as_id,
643 equal_fn_for_key_as_id, NULL);
644 if (IS_ERR(prog_table)) {
645 p_err("failed to create hashmap for pinned paths");
646 return -1;
647 }
648 build_pinned_obj_table(prog_table, BPF_OBJ_PROG);
649 }
650 build_obj_refs_table(&refs_table, BPF_OBJ_PROG);
651
652 if (argc == 2)
653 return do_show_subset(argc, argv);
654
655 if (argc)
656 return BAD_ARG();
657
658 if (json_output)
659 jsonw_start_array(json_wtr);
660 while (true) {
661 err = bpf_prog_get_next_id(id, &id);
662 if (err) {
663 if (errno == ENOENT) {
664 err = 0;
665 break;
666 }
667 p_err("can't get next program: %s%s", strerror(errno),
668 errno == EINVAL ? " -- kernel too old?" : "");
669 err = -1;
670 break;
671 }
672
673 fd = bpf_prog_get_fd_by_id(id);
674 if (fd < 0) {
675 if (errno == ENOENT)
676 continue;
677 p_err("can't get prog by id (%u): %s",
678 id, strerror(errno));
679 err = -1;
680 break;
681 }
682
683 err = show_prog(fd);
684 close(fd);
685 if (err)
686 break;
687 }
688
689 if (json_output)
690 jsonw_end_array(json_wtr);
691
692 delete_obj_refs_table(refs_table);
693
694 if (show_pinned)
695 delete_pinned_obj_table(prog_table);
696
697 return err;
698 }
699
700 static int
prog_dump(struct bpf_prog_info * info,enum dump_mode mode,char * filepath,bool opcodes,bool visual,bool linum)701 prog_dump(struct bpf_prog_info *info, enum dump_mode mode,
702 char *filepath, bool opcodes, bool visual, bool linum)
703 {
704 struct bpf_prog_linfo *prog_linfo = NULL;
705 const char *disasm_opt = NULL;
706 struct dump_data dd = {};
707 void *func_info = NULL;
708 struct btf *btf = NULL;
709 char func_sig[1024];
710 unsigned char *buf;
711 __u32 member_len;
712 int fd, err = -1;
713 ssize_t n;
714
715 if (mode == DUMP_JITED) {
716 if (info->jited_prog_len == 0 || !info->jited_prog_insns) {
717 p_info("no instructions returned");
718 return -1;
719 }
720 buf = u64_to_ptr(info->jited_prog_insns);
721 member_len = info->jited_prog_len;
722 } else { /* DUMP_XLATED */
723 if (info->xlated_prog_len == 0 || !info->xlated_prog_insns) {
724 p_err("error retrieving insn dump: kernel.kptr_restrict set?");
725 return -1;
726 }
727 buf = u64_to_ptr(info->xlated_prog_insns);
728 member_len = info->xlated_prog_len;
729 }
730
731 if (info->btf_id) {
732 btf = btf__load_from_kernel_by_id(info->btf_id);
733 if (!btf) {
734 p_err("failed to get btf");
735 return -1;
736 }
737 }
738
739 func_info = u64_to_ptr(info->func_info);
740
741 if (info->nr_line_info) {
742 prog_linfo = bpf_prog_linfo__new(info);
743 if (!prog_linfo)
744 p_info("error in processing bpf_line_info. continue without it.");
745 }
746
747 if (filepath) {
748 fd = open(filepath, O_WRONLY | O_CREAT | O_TRUNC, 0600);
749 if (fd < 0) {
750 p_err("can't open file %s: %s", filepath,
751 strerror(errno));
752 goto exit_free;
753 }
754
755 n = write(fd, buf, member_len);
756 close(fd);
757 if (n != (ssize_t)member_len) {
758 p_err("error writing output file: %s",
759 n < 0 ? strerror(errno) : "short write");
760 goto exit_free;
761 }
762
763 if (json_output)
764 jsonw_null(json_wtr);
765 } else if (mode == DUMP_JITED) {
766 const char *name = NULL;
767
768 if (info->ifindex) {
769 name = ifindex_to_arch(info->ifindex, info->netns_dev,
770 info->netns_ino, &disasm_opt);
771 if (!name)
772 goto exit_free;
773 }
774
775 if (info->nr_jited_func_lens && info->jited_func_lens) {
776 struct kernel_sym *sym = NULL;
777 struct bpf_func_info *record;
778 char sym_name[SYM_MAX_NAME];
779 unsigned char *img = buf;
780 __u64 *ksyms = NULL;
781 __u32 *lens;
782 __u32 i;
783 if (info->nr_jited_ksyms) {
784 kernel_syms_load(&dd);
785 ksyms = u64_to_ptr(info->jited_ksyms);
786 }
787
788 if (json_output)
789 jsonw_start_array(json_wtr);
790
791 lens = u64_to_ptr(info->jited_func_lens);
792 for (i = 0; i < info->nr_jited_func_lens; i++) {
793 if (ksyms) {
794 sym = kernel_syms_search(&dd, ksyms[i]);
795 if (sym)
796 sprintf(sym_name, "%s", sym->name);
797 else
798 sprintf(sym_name, "0x%016llx", ksyms[i]);
799 } else {
800 strcpy(sym_name, "unknown");
801 }
802
803 if (func_info) {
804 record = func_info + i * info->func_info_rec_size;
805 btf_dumper_type_only(btf, record->type_id,
806 func_sig,
807 sizeof(func_sig));
808 }
809
810 if (json_output) {
811 jsonw_start_object(json_wtr);
812 if (func_info && func_sig[0] != '\0') {
813 jsonw_name(json_wtr, "proto");
814 jsonw_string(json_wtr, func_sig);
815 }
816 jsonw_name(json_wtr, "name");
817 jsonw_string(json_wtr, sym_name);
818 jsonw_name(json_wtr, "insns");
819 } else {
820 if (func_info && func_sig[0] != '\0')
821 printf("%s:\n", func_sig);
822 printf("%s:\n", sym_name);
823 }
824
825 if (ksyms) {
826 if (disasm_print_insn(img, lens[i], opcodes,
827 name, disasm_opt, btf,
828 prog_linfo, ksyms[i], i,
829 linum))
830 goto exit_free;
831 } else {
832 if (disasm_print_insn(img, lens[i], opcodes,
833 name, disasm_opt, btf,
834 NULL, 0, 0, false))
835 goto exit_free;
836 }
837
838 img += lens[i];
839
840 if (json_output)
841 jsonw_end_object(json_wtr);
842 else
843 printf("\n");
844 }
845
846 if (json_output)
847 jsonw_end_array(json_wtr);
848 } else {
849 if (disasm_print_insn(buf, member_len, opcodes, name,
850 disasm_opt, btf, NULL, 0, 0,
851 false))
852 goto exit_free;
853 }
854 } else {
855 kernel_syms_load(&dd);
856 dd.nr_jited_ksyms = info->nr_jited_ksyms;
857 dd.jited_ksyms = u64_to_ptr(info->jited_ksyms);
858 dd.btf = btf;
859 dd.func_info = func_info;
860 dd.finfo_rec_size = info->func_info_rec_size;
861 dd.prog_linfo = prog_linfo;
862
863 if (json_output)
864 dump_xlated_json(&dd, buf, member_len, opcodes, linum);
865 else if (visual)
866 dump_xlated_cfg(&dd, buf, member_len, opcodes, linum);
867 else
868 dump_xlated_plain(&dd, buf, member_len, opcodes, linum);
869 kernel_syms_destroy(&dd);
870 }
871
872 err = 0;
873
874 exit_free:
875 btf__free(btf);
876 bpf_prog_linfo__free(prog_linfo);
877 return err;
878 }
879
do_dump(int argc,char ** argv)880 static int do_dump(int argc, char **argv)
881 {
882 struct bpf_prog_info info;
883 __u32 info_len = sizeof(info);
884 size_t info_data_sz = 0;
885 void *info_data = NULL;
886 char *filepath = NULL;
887 bool opcodes = false;
888 bool visual = false;
889 enum dump_mode mode;
890 bool linum = false;
891 int nb_fds, i = 0;
892 int *fds = NULL;
893 int err = -1;
894
895 if (is_prefix(*argv, "jited")) {
896 if (disasm_init())
897 return -1;
898 mode = DUMP_JITED;
899 } else if (is_prefix(*argv, "xlated")) {
900 mode = DUMP_XLATED;
901 } else {
902 p_err("expected 'xlated' or 'jited', got: %s", *argv);
903 return -1;
904 }
905 NEXT_ARG();
906
907 if (argc < 2)
908 usage();
909
910 fds = malloc(sizeof(int));
911 if (!fds) {
912 p_err("mem alloc failed");
913 return -1;
914 }
915 nb_fds = prog_parse_fds(&argc, &argv, &fds);
916 if (nb_fds < 1)
917 goto exit_free;
918
919 while (argc) {
920 if (is_prefix(*argv, "file")) {
921 NEXT_ARG();
922 if (!argc) {
923 p_err("expected file path");
924 goto exit_close;
925 }
926 if (nb_fds > 1) {
927 p_err("several programs matched");
928 goto exit_close;
929 }
930
931 filepath = *argv;
932 NEXT_ARG();
933 } else if (is_prefix(*argv, "opcodes")) {
934 opcodes = true;
935 NEXT_ARG();
936 } else if (is_prefix(*argv, "visual")) {
937 if (nb_fds > 1) {
938 p_err("several programs matched");
939 goto exit_close;
940 }
941
942 visual = true;
943 NEXT_ARG();
944 } else if (is_prefix(*argv, "linum")) {
945 linum = true;
946 NEXT_ARG();
947 } else {
948 usage();
949 goto exit_close;
950 }
951 }
952
953 if (filepath && (opcodes || visual || linum)) {
954 p_err("'file' is not compatible with 'opcodes', 'visual', or 'linum'");
955 goto exit_close;
956 }
957 if (json_output && visual) {
958 p_err("'visual' is not compatible with JSON output");
959 goto exit_close;
960 }
961
962 if (json_output && nb_fds > 1)
963 jsonw_start_array(json_wtr); /* root array */
964 for (i = 0; i < nb_fds; i++) {
965 memset(&info, 0, sizeof(info));
966
967 err = bpf_prog_get_info_by_fd(fds[i], &info, &info_len);
968 if (err) {
969 p_err("can't get prog info: %s", strerror(errno));
970 break;
971 }
972
973 err = prep_prog_info(&info, mode, &info_data, &info_data_sz);
974 if (err) {
975 p_err("can't grow prog info_data");
976 break;
977 }
978
979 err = bpf_prog_get_info_by_fd(fds[i], &info, &info_len);
980 if (err) {
981 p_err("can't get prog info: %s", strerror(errno));
982 break;
983 }
984
985 if (json_output && nb_fds > 1) {
986 jsonw_start_object(json_wtr); /* prog object */
987 print_prog_header_json(&info, fds[i]);
988 jsonw_name(json_wtr, "insns");
989 } else if (nb_fds > 1) {
990 print_prog_header_plain(&info, fds[i]);
991 }
992
993 err = prog_dump(&info, mode, filepath, opcodes, visual, linum);
994
995 if (json_output && nb_fds > 1)
996 jsonw_end_object(json_wtr); /* prog object */
997 else if (i != nb_fds - 1 && nb_fds > 1)
998 printf("\n");
999
1000 if (err)
1001 break;
1002 close(fds[i]);
1003 }
1004 if (json_output && nb_fds > 1)
1005 jsonw_end_array(json_wtr); /* root array */
1006
1007 exit_close:
1008 for (; i < nb_fds; i++)
1009 close(fds[i]);
1010 exit_free:
1011 free(info_data);
1012 free(fds);
1013 return err;
1014 }
1015
do_pin(int argc,char ** argv)1016 static int do_pin(int argc, char **argv)
1017 {
1018 int err;
1019
1020 err = do_pin_any(argc, argv, prog_parse_fd);
1021 if (!err && json_output)
1022 jsonw_null(json_wtr);
1023 return err;
1024 }
1025
1026 struct map_replace {
1027 int idx;
1028 int fd;
1029 char *name;
1030 };
1031
map_replace_compar(const void * p1,const void * p2)1032 static int map_replace_compar(const void *p1, const void *p2)
1033 {
1034 const struct map_replace *a = p1, *b = p2;
1035
1036 return a->idx - b->idx;
1037 }
1038
parse_attach_detach_args(int argc,char ** argv,int * progfd,enum bpf_attach_type * attach_type,int * mapfd)1039 static int parse_attach_detach_args(int argc, char **argv, int *progfd,
1040 enum bpf_attach_type *attach_type,
1041 int *mapfd)
1042 {
1043 if (!REQ_ARGS(3))
1044 return -EINVAL;
1045
1046 *progfd = prog_parse_fd(&argc, &argv);
1047 if (*progfd < 0)
1048 return *progfd;
1049
1050 *attach_type = parse_attach_type(*argv);
1051 if (*attach_type == __MAX_BPF_ATTACH_TYPE) {
1052 p_err("invalid attach/detach type");
1053 return -EINVAL;
1054 }
1055
1056 if (*attach_type == BPF_FLOW_DISSECTOR) {
1057 *mapfd = 0;
1058 return 0;
1059 }
1060
1061 NEXT_ARG();
1062 if (!REQ_ARGS(2))
1063 return -EINVAL;
1064
1065 *mapfd = map_parse_fd(&argc, &argv, 0);
1066 if (*mapfd < 0)
1067 return *mapfd;
1068
1069 return 0;
1070 }
1071
do_attach(int argc,char ** argv)1072 static int do_attach(int argc, char **argv)
1073 {
1074 enum bpf_attach_type attach_type;
1075 int err, progfd;
1076 int mapfd;
1077
1078 err = parse_attach_detach_args(argc, argv,
1079 &progfd, &attach_type, &mapfd);
1080 if (err)
1081 return err;
1082
1083 err = bpf_prog_attach(progfd, mapfd, attach_type, 0);
1084 if (err) {
1085 p_err("failed prog attach to map");
1086 return -EINVAL;
1087 }
1088
1089 if (json_output)
1090 jsonw_null(json_wtr);
1091 return 0;
1092 }
1093
do_detach(int argc,char ** argv)1094 static int do_detach(int argc, char **argv)
1095 {
1096 enum bpf_attach_type attach_type;
1097 int err, progfd;
1098 int mapfd;
1099
1100 err = parse_attach_detach_args(argc, argv,
1101 &progfd, &attach_type, &mapfd);
1102 if (err)
1103 return err;
1104
1105 err = bpf_prog_detach2(progfd, mapfd, attach_type);
1106 if (err) {
1107 p_err("failed prog detach from map");
1108 return -EINVAL;
1109 }
1110
1111 if (json_output)
1112 jsonw_null(json_wtr);
1113 return 0;
1114 }
1115
1116 enum prog_tracelog_mode {
1117 TRACE_STDOUT,
1118 TRACE_STDERR,
1119 };
1120
1121 static int
prog_tracelog_stream(int prog_fd,enum prog_tracelog_mode mode)1122 prog_tracelog_stream(int prog_fd, enum prog_tracelog_mode mode)
1123 {
1124 FILE *file = mode == TRACE_STDOUT ? stdout : stderr;
1125 int stream_id = mode == TRACE_STDOUT ? 1 : 2;
1126 char buf[512];
1127 int ret;
1128
1129 ret = 0;
1130 do {
1131 ret = bpf_prog_stream_read(prog_fd, stream_id, buf, sizeof(buf), NULL);
1132 if (ret > 0)
1133 fwrite(buf, sizeof(buf[0]), ret, file);
1134 } while (ret > 0);
1135
1136 fflush(file);
1137 return ret ? -1 : 0;
1138 }
1139
do_tracelog_any(int argc,char ** argv)1140 static int do_tracelog_any(int argc, char **argv)
1141 {
1142 enum prog_tracelog_mode mode;
1143 int fd;
1144
1145 if (argc == 0)
1146 return do_tracelog(argc, argv);
1147 if (!is_prefix(*argv, "stdout") && !is_prefix(*argv, "stderr"))
1148 usage();
1149 mode = is_prefix(*argv, "stdout") ? TRACE_STDOUT : TRACE_STDERR;
1150 NEXT_ARG();
1151
1152 if (!REQ_ARGS(2))
1153 return -1;
1154
1155 fd = prog_parse_fd(&argc, &argv);
1156 if (fd < 0)
1157 return -1;
1158
1159 return prog_tracelog_stream(fd, mode);
1160 }
1161
check_single_stdin(char * file_data_in,char * file_ctx_in)1162 static int check_single_stdin(char *file_data_in, char *file_ctx_in)
1163 {
1164 if (file_data_in && file_ctx_in &&
1165 !strcmp(file_data_in, "-") && !strcmp(file_ctx_in, "-")) {
1166 p_err("cannot use standard input for both data_in and ctx_in");
1167 return -1;
1168 }
1169
1170 return 0;
1171 }
1172
get_run_data(const char * fname,void ** data_ptr,unsigned int * size)1173 static int get_run_data(const char *fname, void **data_ptr, unsigned int *size)
1174 {
1175 size_t block_size = 256;
1176 size_t buf_size = block_size;
1177 size_t nb_read = 0;
1178 void *tmp;
1179 FILE *f;
1180
1181 if (!fname) {
1182 *data_ptr = NULL;
1183 *size = 0;
1184 return 0;
1185 }
1186
1187 if (!strcmp(fname, "-"))
1188 f = stdin;
1189 else
1190 f = fopen(fname, "r");
1191 if (!f) {
1192 p_err("failed to open %s: %s", fname, strerror(errno));
1193 return -1;
1194 }
1195
1196 *data_ptr = malloc(block_size);
1197 if (!*data_ptr) {
1198 p_err("failed to allocate memory for data_in/ctx_in: %s",
1199 strerror(errno));
1200 goto err_fclose;
1201 }
1202
1203 while ((nb_read += fread(*data_ptr + nb_read, 1, block_size, f))) {
1204 if (feof(f))
1205 break;
1206 if (ferror(f)) {
1207 p_err("failed to read data_in/ctx_in from %s: %s",
1208 fname, strerror(errno));
1209 goto err_free;
1210 }
1211 if (nb_read > buf_size - block_size) {
1212 if (buf_size == UINT32_MAX) {
1213 p_err("data_in/ctx_in is too long (max: %u)",
1214 UINT32_MAX);
1215 goto err_free;
1216 }
1217 /* No space for fread()-ing next chunk; realloc() */
1218 buf_size *= 2;
1219 tmp = realloc(*data_ptr, buf_size);
1220 if (!tmp) {
1221 p_err("failed to reallocate data_in/ctx_in: %s",
1222 strerror(errno));
1223 goto err_free;
1224 }
1225 *data_ptr = tmp;
1226 }
1227 }
1228 if (f != stdin)
1229 fclose(f);
1230
1231 *size = nb_read;
1232 return 0;
1233
1234 err_free:
1235 free(*data_ptr);
1236 *data_ptr = NULL;
1237 err_fclose:
1238 if (f != stdin)
1239 fclose(f);
1240 return -1;
1241 }
1242
hex_print(void * data,unsigned int size,FILE * f)1243 static void hex_print(void *data, unsigned int size, FILE *f)
1244 {
1245 size_t i, j;
1246 char c;
1247
1248 for (i = 0; i < size; i += 16) {
1249 /* Row offset */
1250 fprintf(f, "%07zx\t", i);
1251
1252 /* Hexadecimal values */
1253 for (j = i; j < i + 16 && j < size; j++)
1254 fprintf(f, "%02x%s", *(uint8_t *)(data + j),
1255 j % 2 ? " " : "");
1256 for (; j < i + 16; j++)
1257 fprintf(f, " %s", j % 2 ? " " : "");
1258
1259 /* ASCII values (if relevant), '.' otherwise */
1260 fprintf(f, "| ");
1261 for (j = i; j < i + 16 && j < size; j++) {
1262 c = *(char *)(data + j);
1263 if (c < ' ' || c > '~')
1264 c = '.';
1265 fprintf(f, "%c%s", c, j == i + 7 ? " " : "");
1266 }
1267
1268 fprintf(f, "\n");
1269 }
1270 }
1271
1272 static int
print_run_output(void * data,unsigned int size,const char * fname,const char * json_key)1273 print_run_output(void *data, unsigned int size, const char *fname,
1274 const char *json_key)
1275 {
1276 size_t nb_written;
1277 FILE *f;
1278
1279 if (!fname)
1280 return 0;
1281
1282 if (!strcmp(fname, "-")) {
1283 f = stdout;
1284 if (json_output) {
1285 jsonw_name(json_wtr, json_key);
1286 print_data_json(data, size);
1287 } else {
1288 hex_print(data, size, f);
1289 }
1290 return 0;
1291 }
1292
1293 f = fopen(fname, "w");
1294 if (!f) {
1295 p_err("failed to open %s: %s", fname, strerror(errno));
1296 return -1;
1297 }
1298
1299 nb_written = fwrite(data, 1, size, f);
1300 fclose(f);
1301 if (nb_written != size) {
1302 p_err("failed to write output data/ctx: %s", strerror(errno));
1303 return -1;
1304 }
1305
1306 return 0;
1307 }
1308
alloc_run_data(void ** data_ptr,unsigned int size_out)1309 static int alloc_run_data(void **data_ptr, unsigned int size_out)
1310 {
1311 *data_ptr = calloc(size_out, 1);
1312 if (!*data_ptr) {
1313 p_err("failed to allocate memory for output data/ctx: %s",
1314 strerror(errno));
1315 return -1;
1316 }
1317
1318 return 0;
1319 }
1320
do_run(int argc,char ** argv)1321 static int do_run(int argc, char **argv)
1322 {
1323 char *data_fname_in = NULL, *data_fname_out = NULL;
1324 char *ctx_fname_in = NULL, *ctx_fname_out = NULL;
1325 const unsigned int default_size = SZ_32K;
1326 void *data_in = NULL, *data_out = NULL;
1327 void *ctx_in = NULL, *ctx_out = NULL;
1328 unsigned int repeat = 1;
1329 int fd, err;
1330 LIBBPF_OPTS(bpf_test_run_opts, test_attr);
1331
1332 if (!REQ_ARGS(4))
1333 return -1;
1334
1335 fd = prog_parse_fd(&argc, &argv);
1336 if (fd < 0)
1337 return -1;
1338
1339 while (argc) {
1340 if (detect_common_prefix(*argv, "data_in", "data_out",
1341 "data_size_out", NULL))
1342 return -1;
1343 if (detect_common_prefix(*argv, "ctx_in", "ctx_out",
1344 "ctx_size_out", NULL))
1345 return -1;
1346
1347 if (is_prefix(*argv, "data_in")) {
1348 NEXT_ARG();
1349 if (!REQ_ARGS(1))
1350 return -1;
1351
1352 data_fname_in = GET_ARG();
1353 if (check_single_stdin(data_fname_in, ctx_fname_in))
1354 return -1;
1355 } else if (is_prefix(*argv, "data_out")) {
1356 NEXT_ARG();
1357 if (!REQ_ARGS(1))
1358 return -1;
1359
1360 data_fname_out = GET_ARG();
1361 } else if (is_prefix(*argv, "data_size_out")) {
1362 char *endptr;
1363
1364 NEXT_ARG();
1365 if (!REQ_ARGS(1))
1366 return -1;
1367
1368 test_attr.data_size_out = strtoul(*argv, &endptr, 0);
1369 if (*endptr) {
1370 p_err("can't parse %s as output data size",
1371 *argv);
1372 return -1;
1373 }
1374 NEXT_ARG();
1375 } else if (is_prefix(*argv, "ctx_in")) {
1376 NEXT_ARG();
1377 if (!REQ_ARGS(1))
1378 return -1;
1379
1380 ctx_fname_in = GET_ARG();
1381 if (check_single_stdin(data_fname_in, ctx_fname_in))
1382 return -1;
1383 } else if (is_prefix(*argv, "ctx_out")) {
1384 NEXT_ARG();
1385 if (!REQ_ARGS(1))
1386 return -1;
1387
1388 ctx_fname_out = GET_ARG();
1389 } else if (is_prefix(*argv, "ctx_size_out")) {
1390 char *endptr;
1391
1392 NEXT_ARG();
1393 if (!REQ_ARGS(1))
1394 return -1;
1395
1396 test_attr.ctx_size_out = strtoul(*argv, &endptr, 0);
1397 if (*endptr) {
1398 p_err("can't parse %s as output context size",
1399 *argv);
1400 return -1;
1401 }
1402 NEXT_ARG();
1403 } else if (is_prefix(*argv, "repeat")) {
1404 char *endptr;
1405
1406 NEXT_ARG();
1407 if (!REQ_ARGS(1))
1408 return -1;
1409
1410 repeat = strtoul(*argv, &endptr, 0);
1411 if (*endptr) {
1412 p_err("can't parse %s as repeat number",
1413 *argv);
1414 return -1;
1415 }
1416 NEXT_ARG();
1417 } else {
1418 p_err("expected no more arguments, 'data_in', 'data_out', 'data_size_out', 'ctx_in', 'ctx_out', 'ctx_size_out' or 'repeat', got: '%s'?",
1419 *argv);
1420 return -1;
1421 }
1422 }
1423
1424 err = get_run_data(data_fname_in, &data_in, &test_attr.data_size_in);
1425 if (err)
1426 return -1;
1427
1428 if (data_in) {
1429 if (!test_attr.data_size_out)
1430 test_attr.data_size_out = default_size;
1431 err = alloc_run_data(&data_out, test_attr.data_size_out);
1432 if (err)
1433 goto free_data_in;
1434 }
1435
1436 err = get_run_data(ctx_fname_in, &ctx_in, &test_attr.ctx_size_in);
1437 if (err)
1438 goto free_data_out;
1439
1440 if (ctx_in) {
1441 if (!test_attr.ctx_size_out)
1442 test_attr.ctx_size_out = default_size;
1443 err = alloc_run_data(&ctx_out, test_attr.ctx_size_out);
1444 if (err)
1445 goto free_ctx_in;
1446 }
1447
1448 test_attr.repeat = repeat;
1449 test_attr.data_in = data_in;
1450 test_attr.data_out = data_out;
1451 test_attr.ctx_in = ctx_in;
1452 test_attr.ctx_out = ctx_out;
1453
1454 err = bpf_prog_test_run_opts(fd, &test_attr);
1455 if (err) {
1456 p_err("failed to run program: %s", strerror(errno));
1457 goto free_ctx_out;
1458 }
1459
1460 err = 0;
1461
1462 if (json_output)
1463 jsonw_start_object(json_wtr); /* root */
1464
1465 /* Do not exit on errors occurring when printing output data/context,
1466 * we still want to print return value and duration for program run.
1467 */
1468 if (test_attr.data_size_out)
1469 err += print_run_output(test_attr.data_out,
1470 test_attr.data_size_out,
1471 data_fname_out, "data_out");
1472 if (test_attr.ctx_size_out)
1473 err += print_run_output(test_attr.ctx_out,
1474 test_attr.ctx_size_out,
1475 ctx_fname_out, "ctx_out");
1476
1477 if (json_output) {
1478 jsonw_uint_field(json_wtr, "retval", test_attr.retval);
1479 jsonw_uint_field(json_wtr, "duration", test_attr.duration);
1480 jsonw_end_object(json_wtr); /* root */
1481 } else {
1482 fprintf(stdout, "Return value: %u, duration%s: %uns\n",
1483 test_attr.retval,
1484 repeat > 1 ? " (average)" : "", test_attr.duration);
1485 }
1486
1487 free_ctx_out:
1488 free(ctx_out);
1489 free_ctx_in:
1490 free(ctx_in);
1491 free_data_out:
1492 free(data_out);
1493 free_data_in:
1494 free(data_in);
1495
1496 return err;
1497 }
1498
1499 static int
get_prog_type_by_name(const char * name,enum bpf_prog_type * prog_type,enum bpf_attach_type * expected_attach_type)1500 get_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
1501 enum bpf_attach_type *expected_attach_type)
1502 {
1503 libbpf_print_fn_t print_backup;
1504 int ret;
1505
1506 ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type);
1507 if (!ret)
1508 return ret;
1509
1510 /* libbpf_prog_type_by_name() failed, let's re-run with debug level */
1511 print_backup = libbpf_set_print(print_all_levels);
1512 ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type);
1513 libbpf_set_print(print_backup);
1514
1515 return ret;
1516 }
1517
1518 static int
auto_attach_program(struct bpf_program * prog,const char * path)1519 auto_attach_program(struct bpf_program *prog, const char *path)
1520 {
1521 struct bpf_link *link;
1522 int err;
1523
1524 link = bpf_program__attach(prog);
1525 if (!link) {
1526 p_info("Program %s does not support autoattach, falling back to pinning",
1527 bpf_program__name(prog));
1528 return bpf_obj_pin(bpf_program__fd(prog), path);
1529 }
1530
1531 err = bpf_link__pin(link, path);
1532 bpf_link__destroy(link);
1533 return err;
1534 }
1535
1536 static int
auto_attach_programs(struct bpf_object * obj,const char * path)1537 auto_attach_programs(struct bpf_object *obj, const char *path)
1538 {
1539 struct bpf_program *prog;
1540 char buf[PATH_MAX];
1541 int err;
1542
1543 bpf_object__for_each_program(prog, obj) {
1544 err = pathname_concat(buf, sizeof(buf), path, bpf_program__name(prog));
1545 if (err)
1546 goto err_unpin_programs;
1547
1548 err = auto_attach_program(prog, buf);
1549 if (err)
1550 goto err_unpin_programs;
1551 }
1552
1553 return 0;
1554
1555 err_unpin_programs:
1556 while ((prog = bpf_object__prev_program(obj, prog))) {
1557 if (pathname_concat(buf, sizeof(buf), path, bpf_program__name(prog)))
1558 continue;
1559
1560 bpf_program__unpin(prog, buf);
1561 }
1562
1563 return err;
1564 }
1565
load_with_options(int argc,char ** argv,bool first_prog_only)1566 static int load_with_options(int argc, char **argv, bool first_prog_only)
1567 {
1568 enum bpf_prog_type common_prog_type = BPF_PROG_TYPE_UNSPEC;
1569 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts,
1570 .relaxed_maps = relaxed_maps,
1571 );
1572 enum bpf_attach_type expected_attach_type;
1573 struct map_replace *map_replace = NULL;
1574 struct bpf_program *prog = NULL, *pos;
1575 unsigned int old_map_fds = 0;
1576 const char *pinmaps = NULL;
1577 __u32 xdpmeta_ifindex = 0;
1578 __u32 offload_ifindex = 0;
1579 bool auto_attach = false;
1580 struct bpf_object *obj;
1581 struct bpf_map *map;
1582 const char *pinfile;
1583 unsigned int i, j;
1584 const char *file;
1585 int idx, err;
1586
1587
1588 if (!REQ_ARGS(2))
1589 return -1;
1590 file = GET_ARG();
1591 pinfile = GET_ARG();
1592
1593 while (argc) {
1594 if (is_prefix(*argv, "type")) {
1595 NEXT_ARG();
1596
1597 if (common_prog_type != BPF_PROG_TYPE_UNSPEC) {
1598 p_err("program type already specified");
1599 goto err_free_reuse_maps;
1600 }
1601 if (!REQ_ARGS(1))
1602 goto err_free_reuse_maps;
1603
1604 err = libbpf_prog_type_by_name(*argv, &common_prog_type,
1605 &expected_attach_type);
1606 if (err < 0) {
1607 /* Put a '/' at the end of type to appease libbpf */
1608 char *type = malloc(strlen(*argv) + 2);
1609
1610 if (!type) {
1611 p_err("mem alloc failed");
1612 goto err_free_reuse_maps;
1613 }
1614 *type = 0;
1615 strcat(type, *argv);
1616 strcat(type, "/");
1617
1618 err = get_prog_type_by_name(type, &common_prog_type,
1619 &expected_attach_type);
1620 free(type);
1621 if (err < 0)
1622 goto err_free_reuse_maps;
1623 }
1624
1625 NEXT_ARG();
1626 } else if (is_prefix(*argv, "map")) {
1627 void *new_map_replace;
1628 char *endptr, *name;
1629 int fd;
1630
1631 NEXT_ARG();
1632
1633 if (!REQ_ARGS(4))
1634 goto err_free_reuse_maps;
1635
1636 if (is_prefix(*argv, "idx")) {
1637 NEXT_ARG();
1638
1639 idx = strtoul(*argv, &endptr, 0);
1640 if (*endptr) {
1641 p_err("can't parse %s as IDX", *argv);
1642 goto err_free_reuse_maps;
1643 }
1644 name = NULL;
1645 } else if (is_prefix(*argv, "name")) {
1646 NEXT_ARG();
1647
1648 name = *argv;
1649 idx = -1;
1650 } else {
1651 p_err("expected 'idx' or 'name', got: '%s'?",
1652 *argv);
1653 goto err_free_reuse_maps;
1654 }
1655 NEXT_ARG();
1656
1657 fd = map_parse_fd(&argc, &argv, 0);
1658 if (fd < 0)
1659 goto err_free_reuse_maps;
1660
1661 new_map_replace = libbpf_reallocarray(map_replace,
1662 old_map_fds + 1,
1663 sizeof(*map_replace));
1664 if (!new_map_replace) {
1665 p_err("mem alloc failed");
1666 goto err_free_reuse_maps;
1667 }
1668 map_replace = new_map_replace;
1669
1670 map_replace[old_map_fds].idx = idx;
1671 map_replace[old_map_fds].name = name;
1672 map_replace[old_map_fds].fd = fd;
1673 old_map_fds++;
1674 } else if (is_prefix(*argv, "dev")) {
1675 p_info("Warning: 'bpftool prog load [...] dev <ifname>' syntax is deprecated.\n"
1676 "Going further, please use 'offload_dev <ifname>' to offload program to device.\n"
1677 "For applications using XDP hints only, use 'xdpmeta_dev <ifname>'.");
1678 goto offload_dev;
1679 } else if (is_prefix(*argv, "offload_dev")) {
1680 offload_dev:
1681 NEXT_ARG();
1682
1683 if (offload_ifindex) {
1684 p_err("offload_dev already specified");
1685 goto err_free_reuse_maps;
1686 } else if (xdpmeta_ifindex) {
1687 p_err("xdpmeta_dev and offload_dev are mutually exclusive");
1688 goto err_free_reuse_maps;
1689 }
1690 if (!REQ_ARGS(1))
1691 goto err_free_reuse_maps;
1692
1693 offload_ifindex = if_nametoindex(*argv);
1694 if (!offload_ifindex) {
1695 p_err("unrecognized netdevice '%s': %s",
1696 *argv, strerror(errno));
1697 goto err_free_reuse_maps;
1698 }
1699 NEXT_ARG();
1700 } else if (is_prefix(*argv, "xdpmeta_dev")) {
1701 NEXT_ARG();
1702
1703 if (xdpmeta_ifindex) {
1704 p_err("xdpmeta_dev already specified");
1705 goto err_free_reuse_maps;
1706 } else if (offload_ifindex) {
1707 p_err("xdpmeta_dev and offload_dev are mutually exclusive");
1708 goto err_free_reuse_maps;
1709 }
1710 if (!REQ_ARGS(1))
1711 goto err_free_reuse_maps;
1712
1713 xdpmeta_ifindex = if_nametoindex(*argv);
1714 if (!xdpmeta_ifindex) {
1715 p_err("unrecognized netdevice '%s': %s",
1716 *argv, strerror(errno));
1717 goto err_free_reuse_maps;
1718 }
1719 NEXT_ARG();
1720 } else if (is_prefix(*argv, "pinmaps")) {
1721 NEXT_ARG();
1722
1723 if (!REQ_ARGS(1))
1724 goto err_free_reuse_maps;
1725
1726 pinmaps = GET_ARG();
1727 } else if (is_prefix(*argv, "autoattach")) {
1728 auto_attach = true;
1729 NEXT_ARG();
1730 } else if (is_prefix(*argv, "kernel_btf")) {
1731 NEXT_ARG();
1732
1733 if (!REQ_ARGS(1))
1734 goto err_free_reuse_maps;
1735
1736 open_opts.btf_custom_path = GET_ARG();
1737 } else {
1738 p_err("expected no more arguments, "
1739 "'type', 'map', 'offload_dev', 'xdpmeta_dev', 'pinmaps', "
1740 "'autoattach', or 'kernel_btf', got: '%s'?",
1741 *argv);
1742 goto err_free_reuse_maps;
1743 }
1744 }
1745
1746 set_max_rlimit();
1747
1748 if (verifier_logs)
1749 /* log_level1 + log_level2 + stats, but not stable UAPI */
1750 open_opts.kernel_log_level = 1 + 2 + 4;
1751
1752 obj = bpf_object__open_file(file, &open_opts);
1753 if (!obj) {
1754 p_err("failed to open object file");
1755 goto err_free_reuse_maps;
1756 }
1757
1758 bpf_object__for_each_program(pos, obj) {
1759 enum bpf_prog_type prog_type = common_prog_type;
1760
1761 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
1762 const char *sec_name = bpf_program__section_name(pos);
1763
1764 err = get_prog_type_by_name(sec_name, &prog_type,
1765 &expected_attach_type);
1766 if (err < 0)
1767 goto err_close_obj;
1768 }
1769
1770 if (prog_type == BPF_PROG_TYPE_XDP && xdpmeta_ifindex) {
1771 bpf_program__set_flags(pos, BPF_F_XDP_DEV_BOUND_ONLY);
1772 bpf_program__set_ifindex(pos, xdpmeta_ifindex);
1773 } else {
1774 bpf_program__set_ifindex(pos, offload_ifindex);
1775 }
1776 if (bpf_program__type(pos) != prog_type)
1777 bpf_program__set_type(pos, prog_type);
1778 bpf_program__set_expected_attach_type(pos, expected_attach_type);
1779 }
1780
1781 qsort(map_replace, old_map_fds, sizeof(*map_replace),
1782 map_replace_compar);
1783
1784 /* After the sort maps by name will be first on the list, because they
1785 * have idx == -1. Resolve them.
1786 */
1787 j = 0;
1788 while (j < old_map_fds && map_replace[j].name) {
1789 i = 0;
1790 bpf_object__for_each_map(map, obj) {
1791 if (!strcmp(bpf_map__name(map), map_replace[j].name)) {
1792 map_replace[j].idx = i;
1793 break;
1794 }
1795 i++;
1796 }
1797 if (map_replace[j].idx == -1) {
1798 p_err("unable to find map '%s'", map_replace[j].name);
1799 goto err_close_obj;
1800 }
1801 j++;
1802 }
1803 /* Resort if any names were resolved */
1804 if (j)
1805 qsort(map_replace, old_map_fds, sizeof(*map_replace),
1806 map_replace_compar);
1807
1808 /* Set ifindex and name reuse */
1809 j = 0;
1810 idx = 0;
1811 bpf_object__for_each_map(map, obj) {
1812 if (bpf_map__type(map) != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1813 bpf_map__set_ifindex(map, offload_ifindex);
1814
1815 if (j < old_map_fds && idx == map_replace[j].idx) {
1816 err = bpf_map__reuse_fd(map, map_replace[j++].fd);
1817 if (err) {
1818 p_err("unable to set up map reuse: %d", err);
1819 goto err_close_obj;
1820 }
1821
1822 /* Next reuse wants to apply to the same map */
1823 if (j < old_map_fds && map_replace[j].idx == idx) {
1824 p_err("replacement for map idx %d specified more than once",
1825 idx);
1826 goto err_close_obj;
1827 }
1828 }
1829
1830 idx++;
1831 }
1832 if (j < old_map_fds) {
1833 p_err("map idx '%d' not used", map_replace[j].idx);
1834 goto err_close_obj;
1835 }
1836
1837 err = bpf_object__load(obj);
1838 if (err) {
1839 p_err("failed to load object file");
1840 goto err_close_obj;
1841 }
1842
1843 if (first_prog_only)
1844 err = mount_bpffs_for_file(pinfile);
1845 else
1846 err = create_and_mount_bpffs_dir(pinfile);
1847 if (err)
1848 goto err_close_obj;
1849
1850 if (first_prog_only) {
1851 prog = bpf_object__next_program(obj, NULL);
1852 if (!prog) {
1853 p_err("object file doesn't contain any bpf program");
1854 goto err_close_obj;
1855 }
1856
1857 if (auto_attach)
1858 err = auto_attach_program(prog, pinfile);
1859 else
1860 err = bpf_obj_pin(bpf_program__fd(prog), pinfile);
1861 if (err) {
1862 p_err("failed to pin program %s",
1863 bpf_program__section_name(prog));
1864 goto err_close_obj;
1865 }
1866 } else {
1867 if (auto_attach)
1868 err = auto_attach_programs(obj, pinfile);
1869 else
1870 err = bpf_object__pin_programs(obj, pinfile);
1871 if (err) {
1872 p_err("failed to pin all programs");
1873 goto err_close_obj;
1874 }
1875 }
1876
1877 if (pinmaps) {
1878 err = create_and_mount_bpffs_dir(pinmaps);
1879 if (err)
1880 goto err_unpin;
1881
1882 err = bpf_object__pin_maps(obj, pinmaps);
1883 if (err) {
1884 p_err("failed to pin all maps");
1885 goto err_unpin;
1886 }
1887 }
1888
1889 if (json_output)
1890 jsonw_null(json_wtr);
1891
1892 bpf_object__close(obj);
1893 for (i = 0; i < old_map_fds; i++)
1894 close(map_replace[i].fd);
1895 free(map_replace);
1896
1897 return 0;
1898
1899 err_unpin:
1900 if (first_prog_only)
1901 unlink(pinfile);
1902 else
1903 bpf_object__unpin_programs(obj, pinfile);
1904 err_close_obj:
1905 bpf_object__close(obj);
1906 err_free_reuse_maps:
1907 for (i = 0; i < old_map_fds; i++)
1908 close(map_replace[i].fd);
1909 free(map_replace);
1910 return -1;
1911 }
1912
count_open_fds(void)1913 static int count_open_fds(void)
1914 {
1915 DIR *dp = opendir("/proc/self/fd");
1916 struct dirent *de;
1917 int cnt = -3;
1918
1919 if (!dp)
1920 return -1;
1921
1922 while ((de = readdir(dp)))
1923 cnt++;
1924
1925 closedir(dp);
1926 return cnt;
1927 }
1928
try_loader(struct gen_loader_opts * gen)1929 static int try_loader(struct gen_loader_opts *gen)
1930 {
1931 struct bpf_load_and_run_opts opts = {};
1932 struct bpf_loader_ctx *ctx;
1933 int ctx_sz = sizeof(*ctx) + 64 * max(sizeof(struct bpf_map_desc),
1934 sizeof(struct bpf_prog_desc));
1935 int log_buf_sz = (1u << 24) - 1;
1936 int err, fds_before, fd_delta;
1937 char *log_buf = NULL;
1938
1939 ctx = alloca(ctx_sz);
1940 memset(ctx, 0, ctx_sz);
1941 ctx->sz = ctx_sz;
1942 if (verifier_logs) {
1943 ctx->log_level = 1 + 2 + 4;
1944 ctx->log_size = log_buf_sz;
1945 log_buf = malloc(log_buf_sz);
1946 if (!log_buf)
1947 return -ENOMEM;
1948 ctx->log_buf = (long) log_buf;
1949 }
1950 opts.ctx = ctx;
1951 opts.data = gen->data;
1952 opts.data_sz = gen->data_sz;
1953 opts.insns = gen->insns;
1954 opts.insns_sz = gen->insns_sz;
1955 fds_before = count_open_fds();
1956 err = bpf_load_and_run(&opts);
1957 fd_delta = count_open_fds() - fds_before;
1958 if (err < 0 || verifier_logs) {
1959 fprintf(stderr, "err %d\n%s\n%s", err, opts.errstr, log_buf);
1960 if (fd_delta && err < 0)
1961 fprintf(stderr, "loader prog leaked %d FDs\n",
1962 fd_delta);
1963 }
1964 free(log_buf);
1965 return err;
1966 }
1967
do_loader(int argc,char ** argv)1968 static int do_loader(int argc, char **argv)
1969 {
1970 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts);
1971 DECLARE_LIBBPF_OPTS(gen_loader_opts, gen);
1972 struct bpf_object *obj;
1973 const char *file;
1974 int err = 0;
1975
1976 if (!REQ_ARGS(1))
1977 return -1;
1978 file = GET_ARG();
1979
1980 if (verifier_logs)
1981 /* log_level1 + log_level2 + stats, but not stable UAPI */
1982 open_opts.kernel_log_level = 1 + 2 + 4;
1983
1984 obj = bpf_object__open_file(file, &open_opts);
1985 if (!obj) {
1986 err = -1;
1987 p_err("failed to open object file");
1988 goto err_close_obj;
1989 }
1990
1991 err = bpf_object__gen_loader(obj, &gen);
1992 if (err)
1993 goto err_close_obj;
1994
1995 err = bpf_object__load(obj);
1996 if (err) {
1997 p_err("failed to load object file");
1998 goto err_close_obj;
1999 }
2000
2001 if (verifier_logs) {
2002 struct dump_data dd = {};
2003
2004 kernel_syms_load(&dd);
2005 dump_xlated_plain(&dd, (void *)gen.insns, gen.insns_sz, false, false);
2006 kernel_syms_destroy(&dd);
2007 }
2008 err = try_loader(&gen);
2009 err_close_obj:
2010 bpf_object__close(obj);
2011 return err;
2012 }
2013
do_load(int argc,char ** argv)2014 static int do_load(int argc, char **argv)
2015 {
2016 if (use_loader)
2017 return do_loader(argc, argv);
2018 return load_with_options(argc, argv, true);
2019 }
2020
do_loadall(int argc,char ** argv)2021 static int do_loadall(int argc, char **argv)
2022 {
2023 return load_with_options(argc, argv, false);
2024 }
2025
2026 #ifdef BPFTOOL_WITHOUT_SKELETONS
2027
do_profile(int argc,char ** argv)2028 static int do_profile(int argc, char **argv)
2029 {
2030 p_err("bpftool prog profile command is not supported. Please build bpftool with clang >= 10.0.0");
2031 return 0;
2032 }
2033
2034 #else /* BPFTOOL_WITHOUT_SKELETONS */
2035
2036 #include "profiler.skel.h"
2037
2038 struct profile_metric {
2039 const char *name;
2040 struct bpf_perf_event_value val;
2041 struct perf_event_attr attr;
2042 bool selected;
2043
2044 /* calculate ratios like instructions per cycle */
2045 const int ratio_metric; /* 0 for N/A, 1 for index 0 (cycles) */
2046 const char *ratio_desc;
2047 const float ratio_mul;
2048 } metrics[] = {
2049 {
2050 .name = "cycles",
2051 .attr = {
2052 .type = PERF_TYPE_HARDWARE,
2053 .config = PERF_COUNT_HW_CPU_CYCLES,
2054 .exclude_user = 1,
2055 },
2056 },
2057 {
2058 .name = "instructions",
2059 .attr = {
2060 .type = PERF_TYPE_HARDWARE,
2061 .config = PERF_COUNT_HW_INSTRUCTIONS,
2062 .exclude_user = 1,
2063 },
2064 .ratio_metric = 1,
2065 .ratio_desc = "insns per cycle",
2066 .ratio_mul = 1.0,
2067 },
2068 {
2069 .name = "l1d_loads",
2070 .attr = {
2071 .type = PERF_TYPE_HW_CACHE,
2072 .config =
2073 PERF_COUNT_HW_CACHE_L1D |
2074 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2075 (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16),
2076 .exclude_user = 1,
2077 },
2078 },
2079 {
2080 .name = "llc_misses",
2081 .attr = {
2082 .type = PERF_TYPE_HW_CACHE,
2083 .config =
2084 PERF_COUNT_HW_CACHE_LL |
2085 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2086 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2087 .exclude_user = 1
2088 },
2089 .ratio_metric = 2,
2090 .ratio_desc = "LLC misses per million insns",
2091 .ratio_mul = 1e6,
2092 },
2093 {
2094 .name = "itlb_misses",
2095 .attr = {
2096 .type = PERF_TYPE_HW_CACHE,
2097 .config =
2098 PERF_COUNT_HW_CACHE_ITLB |
2099 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2100 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2101 .exclude_user = 1
2102 },
2103 .ratio_metric = 2,
2104 .ratio_desc = "itlb misses per million insns",
2105 .ratio_mul = 1e6,
2106 },
2107 {
2108 .name = "dtlb_misses",
2109 .attr = {
2110 .type = PERF_TYPE_HW_CACHE,
2111 .config =
2112 PERF_COUNT_HW_CACHE_DTLB |
2113 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2114 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2115 .exclude_user = 1
2116 },
2117 .ratio_metric = 2,
2118 .ratio_desc = "dtlb misses per million insns",
2119 .ratio_mul = 1e6,
2120 },
2121 };
2122
2123 static __u64 profile_total_count;
2124
2125 #define MAX_NUM_PROFILE_METRICS 4
2126
profile_parse_metrics(int argc,char ** argv)2127 static int profile_parse_metrics(int argc, char **argv)
2128 {
2129 unsigned int metric_cnt;
2130 int selected_cnt = 0;
2131 unsigned int i;
2132
2133 metric_cnt = ARRAY_SIZE(metrics);
2134
2135 while (argc > 0) {
2136 for (i = 0; i < metric_cnt; i++) {
2137 if (is_prefix(argv[0], metrics[i].name)) {
2138 if (!metrics[i].selected)
2139 selected_cnt++;
2140 metrics[i].selected = true;
2141 break;
2142 }
2143 }
2144 if (i == metric_cnt) {
2145 p_err("unknown metric %s", argv[0]);
2146 return -1;
2147 }
2148 NEXT_ARG();
2149 }
2150 if (selected_cnt > MAX_NUM_PROFILE_METRICS) {
2151 p_err("too many (%d) metrics, please specify no more than %d metrics at a time",
2152 selected_cnt, MAX_NUM_PROFILE_METRICS);
2153 return -1;
2154 }
2155 return selected_cnt;
2156 }
2157
profile_read_values(struct profiler_bpf * obj)2158 static void profile_read_values(struct profiler_bpf *obj)
2159 {
2160 __u32 m, cpu, num_cpu = obj->rodata->num_cpu;
2161 int reading_map_fd, count_map_fd;
2162 __u64 counts[num_cpu];
2163 __u32 key = 0;
2164 int err;
2165
2166 reading_map_fd = bpf_map__fd(obj->maps.accum_readings);
2167 count_map_fd = bpf_map__fd(obj->maps.counts);
2168 if (reading_map_fd < 0 || count_map_fd < 0) {
2169 p_err("failed to get fd for map");
2170 return;
2171 }
2172
2173 err = bpf_map_lookup_elem(count_map_fd, &key, counts);
2174 if (err) {
2175 p_err("failed to read count_map: %s", strerror(errno));
2176 return;
2177 }
2178
2179 profile_total_count = 0;
2180 for (cpu = 0; cpu < num_cpu; cpu++)
2181 profile_total_count += counts[cpu];
2182
2183 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2184 struct bpf_perf_event_value values[num_cpu];
2185
2186 if (!metrics[m].selected)
2187 continue;
2188
2189 err = bpf_map_lookup_elem(reading_map_fd, &key, values);
2190 if (err) {
2191 p_err("failed to read reading_map: %s",
2192 strerror(errno));
2193 return;
2194 }
2195 for (cpu = 0; cpu < num_cpu; cpu++) {
2196 metrics[m].val.counter += values[cpu].counter;
2197 metrics[m].val.enabled += values[cpu].enabled;
2198 metrics[m].val.running += values[cpu].running;
2199 }
2200 key++;
2201 }
2202 }
2203
profile_print_readings_json(void)2204 static void profile_print_readings_json(void)
2205 {
2206 __u32 m;
2207
2208 jsonw_start_array(json_wtr);
2209 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2210 if (!metrics[m].selected)
2211 continue;
2212 jsonw_start_object(json_wtr);
2213 jsonw_string_field(json_wtr, "metric", metrics[m].name);
2214 jsonw_lluint_field(json_wtr, "run_cnt", profile_total_count);
2215 jsonw_lluint_field(json_wtr, "value", metrics[m].val.counter);
2216 jsonw_lluint_field(json_wtr, "enabled", metrics[m].val.enabled);
2217 jsonw_lluint_field(json_wtr, "running", metrics[m].val.running);
2218
2219 jsonw_end_object(json_wtr);
2220 }
2221 jsonw_end_array(json_wtr);
2222 }
2223
profile_print_readings_plain(void)2224 static void profile_print_readings_plain(void)
2225 {
2226 __u32 m;
2227
2228 printf("\n%18llu %-20s\n", profile_total_count, "run_cnt");
2229 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2230 struct bpf_perf_event_value *val = &metrics[m].val;
2231 int r;
2232
2233 if (!metrics[m].selected)
2234 continue;
2235 printf("%18llu %-20s", val->counter, metrics[m].name);
2236
2237 r = metrics[m].ratio_metric - 1;
2238 if (r >= 0 && metrics[r].selected &&
2239 metrics[r].val.counter > 0) {
2240 printf("# %8.2f %-30s",
2241 val->counter * metrics[m].ratio_mul /
2242 metrics[r].val.counter,
2243 metrics[m].ratio_desc);
2244 } else {
2245 printf("%-41s", "");
2246 }
2247
2248 if (val->enabled > val->running)
2249 printf("(%4.2f%%)",
2250 val->running * 100.0 / val->enabled);
2251 printf("\n");
2252 }
2253 }
2254
profile_print_readings(void)2255 static void profile_print_readings(void)
2256 {
2257 if (json_output)
2258 profile_print_readings_json();
2259 else
2260 profile_print_readings_plain();
2261 }
2262
profile_target_name(int tgt_fd)2263 static char *profile_target_name(int tgt_fd)
2264 {
2265 struct bpf_func_info func_info;
2266 struct bpf_prog_info info = {};
2267 __u32 info_len = sizeof(info);
2268 const struct btf_type *t;
2269 __u32 func_info_rec_size;
2270 struct btf *btf = NULL;
2271 char *name = NULL;
2272 int err;
2273
2274 err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len);
2275 if (err) {
2276 p_err("failed to get info for prog FD %d", tgt_fd);
2277 goto out;
2278 }
2279
2280 if (info.btf_id == 0) {
2281 p_err("prog FD %d doesn't have valid btf", tgt_fd);
2282 goto out;
2283 }
2284
2285 func_info_rec_size = info.func_info_rec_size;
2286 if (info.nr_func_info == 0) {
2287 p_err("found 0 func_info for prog FD %d", tgt_fd);
2288 goto out;
2289 }
2290
2291 memset(&info, 0, sizeof(info));
2292 info.nr_func_info = 1;
2293 info.func_info_rec_size = func_info_rec_size;
2294 info.func_info = ptr_to_u64(&func_info);
2295
2296 err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len);
2297 if (err) {
2298 p_err("failed to get func_info for prog FD %d", tgt_fd);
2299 goto out;
2300 }
2301
2302 btf = btf__load_from_kernel_by_id(info.btf_id);
2303 if (!btf) {
2304 p_err("failed to load btf for prog FD %d", tgt_fd);
2305 goto out;
2306 }
2307
2308 t = btf__type_by_id(btf, func_info.type_id);
2309 if (!t) {
2310 p_err("btf %u doesn't have type %u",
2311 info.btf_id, func_info.type_id);
2312 goto out;
2313 }
2314 name = strdup(btf__name_by_offset(btf, t->name_off));
2315 out:
2316 btf__free(btf);
2317 return name;
2318 }
2319
2320 static struct profiler_bpf *profile_obj;
2321 static int profile_tgt_fd = -1;
2322 static char *profile_tgt_name;
2323 static int *profile_perf_events;
2324 static int profile_perf_event_cnt;
2325
profile_close_perf_events(struct profiler_bpf * obj)2326 static void profile_close_perf_events(struct profiler_bpf *obj)
2327 {
2328 int i;
2329
2330 for (i = profile_perf_event_cnt - 1; i >= 0; i--)
2331 close(profile_perf_events[i]);
2332
2333 free(profile_perf_events);
2334 profile_perf_event_cnt = 0;
2335 }
2336
profile_open_perf_event(int mid,int cpu,int map_fd)2337 static int profile_open_perf_event(int mid, int cpu, int map_fd)
2338 {
2339 int pmu_fd;
2340
2341 pmu_fd = syscall(__NR_perf_event_open, &metrics[mid].attr,
2342 -1 /*pid*/, cpu, -1 /*group_fd*/, 0);
2343 if (pmu_fd < 0) {
2344 if (errno == ENODEV) {
2345 p_info("cpu %d may be offline, skip %s profiling.",
2346 cpu, metrics[mid].name);
2347 profile_perf_event_cnt++;
2348 return 0;
2349 }
2350 return -1;
2351 }
2352
2353 if (bpf_map_update_elem(map_fd,
2354 &profile_perf_event_cnt,
2355 &pmu_fd, BPF_ANY) ||
2356 ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0)) {
2357 close(pmu_fd);
2358 return -1;
2359 }
2360
2361 profile_perf_events[profile_perf_event_cnt++] = pmu_fd;
2362 return 0;
2363 }
2364
profile_open_perf_events(struct profiler_bpf * obj)2365 static int profile_open_perf_events(struct profiler_bpf *obj)
2366 {
2367 unsigned int cpu, m;
2368 int map_fd;
2369
2370 profile_perf_events = calloc(
2371 obj->rodata->num_cpu * obj->rodata->num_metric, sizeof(int));
2372 if (!profile_perf_events) {
2373 p_err("failed to allocate memory for perf_event array: %s",
2374 strerror(errno));
2375 return -1;
2376 }
2377 map_fd = bpf_map__fd(obj->maps.events);
2378 if (map_fd < 0) {
2379 p_err("failed to get fd for events map");
2380 return -1;
2381 }
2382
2383 for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2384 if (!metrics[m].selected)
2385 continue;
2386 for (cpu = 0; cpu < obj->rodata->num_cpu; cpu++) {
2387 if (profile_open_perf_event(m, cpu, map_fd)) {
2388 p_err("failed to create event %s on cpu %u",
2389 metrics[m].name, cpu);
2390 return -1;
2391 }
2392 }
2393 }
2394 return 0;
2395 }
2396
profile_print_and_cleanup(void)2397 static void profile_print_and_cleanup(void)
2398 {
2399 profile_close_perf_events(profile_obj);
2400 profile_read_values(profile_obj);
2401 profile_print_readings();
2402 profiler_bpf__destroy(profile_obj);
2403
2404 close(profile_tgt_fd);
2405 free(profile_tgt_name);
2406 }
2407
int_exit(int signo)2408 static void int_exit(int signo)
2409 {
2410 profile_print_and_cleanup();
2411 exit(0);
2412 }
2413
do_profile(int argc,char ** argv)2414 static int do_profile(int argc, char **argv)
2415 {
2416 int num_metric, num_cpu, err = -1;
2417 struct bpf_program *prog;
2418 unsigned long duration;
2419 char *endptr;
2420
2421 /* we at least need two args for the prog and one metric */
2422 if (!REQ_ARGS(3))
2423 return -EINVAL;
2424
2425 /* parse target fd */
2426 profile_tgt_fd = prog_parse_fd(&argc, &argv);
2427 if (profile_tgt_fd < 0) {
2428 p_err("failed to parse fd");
2429 return -1;
2430 }
2431
2432 /* parse profiling optional duration */
2433 if (argc > 2 && is_prefix(argv[0], "duration")) {
2434 NEXT_ARG();
2435 duration = strtoul(*argv, &endptr, 0);
2436 if (*endptr)
2437 usage();
2438 NEXT_ARG();
2439 } else {
2440 duration = UINT_MAX;
2441 }
2442
2443 num_metric = profile_parse_metrics(argc, argv);
2444 if (num_metric <= 0)
2445 goto out;
2446
2447 num_cpu = libbpf_num_possible_cpus();
2448 if (num_cpu <= 0) {
2449 p_err("failed to identify number of CPUs");
2450 goto out;
2451 }
2452
2453 profile_obj = profiler_bpf__open();
2454 if (!profile_obj) {
2455 p_err("failed to open and/or load BPF object");
2456 goto out;
2457 }
2458
2459 profile_obj->rodata->num_cpu = num_cpu;
2460 profile_obj->rodata->num_metric = num_metric;
2461
2462 /* adjust map sizes */
2463 bpf_map__set_max_entries(profile_obj->maps.events, num_metric * num_cpu);
2464 bpf_map__set_max_entries(profile_obj->maps.fentry_readings, num_metric);
2465 bpf_map__set_max_entries(profile_obj->maps.accum_readings, num_metric);
2466 bpf_map__set_max_entries(profile_obj->maps.counts, 1);
2467
2468 /* change target name */
2469 profile_tgt_name = profile_target_name(profile_tgt_fd);
2470 if (!profile_tgt_name)
2471 goto out;
2472
2473 bpf_object__for_each_program(prog, profile_obj->obj) {
2474 err = bpf_program__set_attach_target(prog, profile_tgt_fd,
2475 profile_tgt_name);
2476 if (err) {
2477 p_err("failed to set attach target\n");
2478 goto out;
2479 }
2480 }
2481
2482 set_max_rlimit();
2483 err = profiler_bpf__load(profile_obj);
2484 if (err) {
2485 p_err("failed to load profile_obj");
2486 goto out;
2487 }
2488
2489 err = profile_open_perf_events(profile_obj);
2490 if (err)
2491 goto out;
2492
2493 err = profiler_bpf__attach(profile_obj);
2494 if (err) {
2495 p_err("failed to attach profile_obj");
2496 goto out;
2497 }
2498 signal(SIGINT, int_exit);
2499
2500 sleep(duration);
2501 profile_print_and_cleanup();
2502 return 0;
2503
2504 out:
2505 profile_close_perf_events(profile_obj);
2506 if (profile_obj)
2507 profiler_bpf__destroy(profile_obj);
2508 close(profile_tgt_fd);
2509 free(profile_tgt_name);
2510 return err;
2511 }
2512
2513 #endif /* BPFTOOL_WITHOUT_SKELETONS */
2514
do_help(int argc,char ** argv)2515 static int do_help(int argc, char **argv)
2516 {
2517 if (json_output) {
2518 jsonw_null(json_wtr);
2519 return 0;
2520 }
2521
2522 fprintf(stderr,
2523 "Usage: %1$s %2$s { show | list } [PROG]\n"
2524 " %1$s %2$s dump xlated PROG [{ file FILE | [opcodes] [linum] [visual] }]\n"
2525 " %1$s %2$s dump jited PROG [{ file FILE | [opcodes] [linum] }]\n"
2526 " %1$s %2$s pin PROG FILE\n"
2527 " %1$s %2$s { load | loadall } OBJ PATH \\\n"
2528 " [type TYPE] [{ offload_dev | xdpmeta_dev } NAME] \\\n"
2529 " [map { idx IDX | name NAME } MAP]\\\n"
2530 " [pinmaps MAP_DIR]\n"
2531 " [autoattach]\n"
2532 " [kernel_btf BTF_FILE]\n"
2533 " %1$s %2$s attach PROG ATTACH_TYPE [MAP]\n"
2534 " %1$s %2$s detach PROG ATTACH_TYPE [MAP]\n"
2535 " %1$s %2$s run PROG \\\n"
2536 " data_in FILE \\\n"
2537 " [data_out FILE [data_size_out L]] \\\n"
2538 " [ctx_in FILE [ctx_out FILE [ctx_size_out M]]] \\\n"
2539 " [repeat N]\n"
2540 " %1$s %2$s profile PROG [duration DURATION] METRICs\n"
2541 " %1$s %2$s tracelog\n"
2542 " %1$s %2$s tracelog { stdout | stderr } PROG\n"
2543 " %1$s %2$s help\n"
2544 "\n"
2545 " " HELP_SPEC_MAP "\n"
2546 " " HELP_SPEC_PROGRAM "\n"
2547 " TYPE := { socket | kprobe | kretprobe | classifier | action |\n"
2548 " tracepoint | raw_tracepoint | xdp | perf_event | cgroup/skb |\n"
2549 " cgroup/sock | cgroup/dev | lwt_in | lwt_out | lwt_xmit |\n"
2550 " lwt_seg6local | sockops | sk_skb | sk_msg | lirc_mode2 |\n"
2551 " sk_reuseport | flow_dissector | cgroup/sysctl |\n"
2552 " cgroup/bind4 | cgroup/bind6 | cgroup/post_bind4 |\n"
2553 " cgroup/post_bind6 | cgroup/connect4 | cgroup/connect6 |\n"
2554 " cgroup/connect_unix | cgroup/getpeername4 | cgroup/getpeername6 |\n"
2555 " cgroup/getpeername_unix | cgroup/getsockname4 | cgroup/getsockname6 |\n"
2556 " cgroup/getsockname_unix | cgroup/sendmsg4 | cgroup/sendmsg6 |\n"
2557 " cgroup/sendmsg_unix | cgroup/recvmsg4 | cgroup/recvmsg6 | cgroup/recvmsg_unix |\n"
2558 " cgroup/getsockopt | cgroup/setsockopt | cgroup/sock_release |\n"
2559 " struct_ops | fentry | fexit | freplace | sk_lookup }\n"
2560 " ATTACH_TYPE := { sk_msg_verdict | sk_skb_verdict | sk_skb_stream_verdict |\n"
2561 " sk_skb_stream_parser | flow_dissector }\n"
2562 " METRIC := { cycles | instructions | l1d_loads | llc_misses | itlb_misses | dtlb_misses }\n"
2563 " " HELP_SPEC_OPTIONS " |\n"
2564 " {-f|--bpffs} | {-m|--mapcompat} | {-n|--nomount} |\n"
2565 " {-L|--use-loader} }\n"
2566 "",
2567 bin_name, argv[-2]);
2568
2569 return 0;
2570 }
2571
2572 static const struct cmd cmds[] = {
2573 { "show", do_show },
2574 { "list", do_show },
2575 { "help", do_help },
2576 { "dump", do_dump },
2577 { "pin", do_pin },
2578 { "load", do_load },
2579 { "loadall", do_loadall },
2580 { "attach", do_attach },
2581 { "detach", do_detach },
2582 { "tracelog", do_tracelog_any },
2583 { "run", do_run },
2584 { "profile", do_profile },
2585 { 0 }
2586 };
2587
do_prog(int argc,char ** argv)2588 int do_prog(int argc, char **argv)
2589 {
2590 return cmd_select(cmds, argc, argv, do_help);
2591 }
2592