tools/bpf/bpftool/prog.c

   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
  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
  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
 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
 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
 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
 299 static bool has_metadata_prefix(const char *s)
 300 {
 301         return strncmp(s, BPF_METADATA_PREFIX, BPF_METADATA_PREFIX_LEN) == 0;
 302 }
 303
 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
 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
 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
 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
 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
 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
 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
 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
 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
 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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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         bool auto_attach = false;
1521         struct bpf_object *obj;
1522         struct bpf_map *map;
1523         const char *pinfile;
1524         unsigned int i, j;
1525         __u32 ifindex = 0;
1526         const char *file;
1527         int idx, err;
1528
1529
1530         if (!REQ_ARGS(2))
1531                 return -1;
1532         file = GET_ARG();
1533         pinfile = GET_ARG();
1534
1535         while (argc) {
1536                 if (is_prefix(*argv, "type")) {
1537                         NEXT_ARG();
1538
1539                         if (common_prog_type != BPF_PROG_TYPE_UNSPEC) {
1540                                 p_err("program type already specified");
1541                                 goto err_free_reuse_maps;
1542                         }
1543                         if (!REQ_ARGS(1))
1544                                 goto err_free_reuse_maps;
1545
1546                         err = libbpf_prog_type_by_name(*argv, &common_prog_type,
1547                                                        &expected_attach_type);
1548                         if (err < 0) {
1549                                 /* Put a '/' at the end of type to appease libbpf */
1550                                 char *type = malloc(strlen(*argv) + 2);
1551
1552                                 if (!type) {
1553                                         p_err("mem alloc failed");
1554                                         goto err_free_reuse_maps;
1555                                 }
1556                                 *type = 0;
1557                                 strcat(type, *argv);
1558                                 strcat(type, "/");
1559
1560                                 err = get_prog_type_by_name(type, &common_prog_type,
1561                                                             &expected_attach_type);
1562                                 free(type);
1563                                 if (err < 0)
1564                                         goto err_free_reuse_maps;
1565                         }
1566
1567                         NEXT_ARG();
1568                 } else if (is_prefix(*argv, "map")) {
1569                         void *new_map_replace;
1570                         char *endptr, *name;
1571                         int fd;
1572
1573                         NEXT_ARG();
1574
1575                         if (!REQ_ARGS(4))
1576                                 goto err_free_reuse_maps;
1577
1578                         if (is_prefix(*argv, "idx")) {
1579                                 NEXT_ARG();
1580
1581                                 idx = strtoul(*argv, &endptr, 0);
1582                                 if (*endptr) {
1583                                         p_err("can't parse %s as IDX", *argv);
1584                                         goto err_free_reuse_maps;
1585                                 }
1586                                 name = NULL;
1587                         } else if (is_prefix(*argv, "name")) {
1588                                 NEXT_ARG();
1589
1590                                 name = *argv;
1591                                 idx = -1;
1592                         } else {
1593                                 p_err("expected 'idx' or 'name', got: '%s'?",
1594                                       *argv);
1595                                 goto err_free_reuse_maps;
1596                         }
1597                         NEXT_ARG();
1598
1599                         fd = map_parse_fd(&argc, &argv);
1600                         if (fd < 0)
1601                                 goto err_free_reuse_maps;
1602
1603                         new_map_replace = libbpf_reallocarray(map_replace,
1604                                                               old_map_fds + 1,
1605                                                               sizeof(*map_replace));
1606                         if (!new_map_replace) {
1607                                 p_err("mem alloc failed");
1608                                 goto err_free_reuse_maps;
1609                         }
1610                         map_replace = new_map_replace;
1611
1612                         map_replace[old_map_fds].idx = idx;
1613                         map_replace[old_map_fds].name = name;
1614                         map_replace[old_map_fds].fd = fd;
1615                         old_map_fds++;
1616                 } else if (is_prefix(*argv, "dev")) {
1617                         NEXT_ARG();
1618
1619                         if (ifindex) {
1620                                 p_err("offload device already specified");
1621                                 goto err_free_reuse_maps;
1622                         }
1623                         if (!REQ_ARGS(1))
1624                                 goto err_free_reuse_maps;
1625
1626                         ifindex = if_nametoindex(*argv);
1627                         if (!ifindex) {
1628                                 p_err("unrecognized netdevice '%s': %s",
1629                                       *argv, strerror(errno));
1630                                 goto err_free_reuse_maps;
1631                         }
1632                         NEXT_ARG();
1633                 } else if (is_prefix(*argv, "pinmaps")) {
1634                         NEXT_ARG();
1635
1636                         if (!REQ_ARGS(1))
1637                                 goto err_free_reuse_maps;
1638
1639                         pinmaps = GET_ARG();
1640                 } else if (is_prefix(*argv, "autoattach")) {
1641                         auto_attach = true;
1642                         NEXT_ARG();
1643                 } else {
1644                         p_err("expected no more arguments, 'type', 'map' or 'dev', got: '%s'?",
1645                               *argv);
1646                         goto err_free_reuse_maps;
1647                 }
1648         }
1649
1650         set_max_rlimit();
1651
1652         if (verifier_logs)
1653                 /* log_level1 + log_level2 + stats, but not stable UAPI */
1654                 open_opts.kernel_log_level = 1 + 2 + 4;
1655
1656         obj = bpf_object__open_file(file, &open_opts);
1657         if (!obj) {
1658                 p_err("failed to open object file");
1659                 goto err_free_reuse_maps;
1660         }
1661
1662         bpf_object__for_each_program(pos, obj) {
1663                 enum bpf_prog_type prog_type = common_prog_type;
1664
1665                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
1666                         const char *sec_name = bpf_program__section_name(pos);
1667
1668                         err = get_prog_type_by_name(sec_name, &prog_type,
1669                                                     &expected_attach_type);
1670                         if (err < 0)
1671                                 goto err_close_obj;
1672                 }
1673
1674                 bpf_program__set_ifindex(pos, ifindex);
1675                 if (bpf_program__type(pos) != prog_type)
1676                         bpf_program__set_type(pos, prog_type);
1677                 bpf_program__set_expected_attach_type(pos, expected_attach_type);
1678         }
1679
1680         qsort(map_replace, old_map_fds, sizeof(*map_replace),
1681               map_replace_compar);
1682
1683         /* After the sort maps by name will be first on the list, because they
1684          * have idx == -1.  Resolve them.
1685          */
1686         j = 0;
1687         while (j < old_map_fds && map_replace[j].name) {
1688                 i = 0;
1689                 bpf_object__for_each_map(map, obj) {
1690                         if (!strcmp(bpf_map__name(map), map_replace[j].name)) {
1691                                 map_replace[j].idx = i;
1692                                 break;
1693                         }
1694                         i++;
1695                 }
1696                 if (map_replace[j].idx == -1) {
1697                         p_err("unable to find map '%s'", map_replace[j].name);
1698                         goto err_close_obj;
1699                 }
1700                 j++;
1701         }
1702         /* Resort if any names were resolved */
1703         if (j)
1704                 qsort(map_replace, old_map_fds, sizeof(*map_replace),
1705                       map_replace_compar);
1706
1707         /* Set ifindex and name reuse */
1708         j = 0;
1709         idx = 0;
1710         bpf_object__for_each_map(map, obj) {
1711                 if (bpf_map__type(map) != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1712                         bpf_map__set_ifindex(map, ifindex);
1713
1714                 if (j < old_map_fds && idx == map_replace[j].idx) {
1715                         err = bpf_map__reuse_fd(map, map_replace[j++].fd);
1716                         if (err) {
1717                                 p_err("unable to set up map reuse: %d", err);
1718                                 goto err_close_obj;
1719                         }
1720
1721                         /* Next reuse wants to apply to the same map */
1722                         if (j < old_map_fds && map_replace[j].idx == idx) {
1723                                 p_err("replacement for map idx %d specified more than once",
1724                                       idx);
1725                                 goto err_close_obj;
1726                         }
1727                 }
1728
1729                 idx++;
1730         }
1731         if (j < old_map_fds) {
1732                 p_err("map idx '%d' not used", map_replace[j].idx);
1733                 goto err_close_obj;
1734         }
1735
1736         err = bpf_object__load(obj);
1737         if (err) {
1738                 p_err("failed to load object file");
1739                 goto err_close_obj;
1740         }
1741
1742         err = mount_bpffs_for_pin(pinfile);
1743         if (err)
1744                 goto err_close_obj;
1745
1746         if (first_prog_only) {
1747                 prog = bpf_object__next_program(obj, NULL);
1748                 if (!prog) {
1749                         p_err("object file doesn't contain any bpf program");
1750                         goto err_close_obj;
1751                 }
1752
1753                 if (auto_attach)
1754                         err = auto_attach_program(prog, pinfile);
1755                 else
1756                         err = bpf_obj_pin(bpf_program__fd(prog), pinfile);
1757                 if (err) {
1758                         p_err("failed to pin program %s",
1759                               bpf_program__section_name(prog));
1760                         goto err_close_obj;
1761                 }
1762         } else {
1763                 if (auto_attach)
1764                         err = auto_attach_programs(obj, pinfile);
1765                 else
1766                         err = bpf_object__pin_programs(obj, pinfile);
1767                 if (err) {
1768                         p_err("failed to pin all programs");
1769                         goto err_close_obj;
1770                 }
1771         }
1772
1773         if (pinmaps) {
1774                 err = bpf_object__pin_maps(obj, pinmaps);
1775                 if (err) {
1776                         p_err("failed to pin all maps");
1777                         goto err_unpin;
1778                 }
1779         }
1780
1781         if (json_output)
1782                 jsonw_null(json_wtr);
1783
1784         bpf_object__close(obj);
1785         for (i = 0; i < old_map_fds; i++)
1786                 close(map_replace[i].fd);
1787         free(map_replace);
1788
1789         return 0;
1790
1791 err_unpin:
1792         if (first_prog_only)
1793                 unlink(pinfile);
1794         else
1795                 bpf_object__unpin_programs(obj, pinfile);
1796 err_close_obj:
1797         bpf_object__close(obj);
1798 err_free_reuse_maps:
1799         for (i = 0; i < old_map_fds; i++)
1800                 close(map_replace[i].fd);
1801         free(map_replace);
1802         return -1;
1803 }
1804
1805 static int count_open_fds(void)
1806 {
1807         DIR *dp = opendir("/proc/self/fd");
1808         struct dirent *de;
1809         int cnt = -3;
1810
1811         if (!dp)
1812                 return -1;
1813
1814         while ((de = readdir(dp)))
1815                 cnt++;
1816
1817         closedir(dp);
1818         return cnt;
1819 }
1820
1821 static int try_loader(struct gen_loader_opts *gen)
1822 {
1823         struct bpf_load_and_run_opts opts = {};
1824         struct bpf_loader_ctx *ctx;
1825         int ctx_sz = sizeof(*ctx) + 64 * max(sizeof(struct bpf_map_desc),
1826                                              sizeof(struct bpf_prog_desc));
1827         int log_buf_sz = (1u << 24) - 1;
1828         int err, fds_before, fd_delta;
1829         char *log_buf = NULL;
1830
1831         ctx = alloca(ctx_sz);
1832         memset(ctx, 0, ctx_sz);
1833         ctx->sz = ctx_sz;
1834         if (verifier_logs) {
1835                 ctx->log_level = 1 + 2 + 4;
1836                 ctx->log_size = log_buf_sz;
1837                 log_buf = malloc(log_buf_sz);
1838                 if (!log_buf)
1839                         return -ENOMEM;
1840                 ctx->log_buf = (long) log_buf;
1841         }
1842         opts.ctx = ctx;
1843         opts.data = gen->data;
1844         opts.data_sz = gen->data_sz;
1845         opts.insns = gen->insns;
1846         opts.insns_sz = gen->insns_sz;
1847         fds_before = count_open_fds();
1848         err = bpf_load_and_run(&opts);
1849         fd_delta = count_open_fds() - fds_before;
1850         if (err < 0 || verifier_logs) {
1851                 fprintf(stderr, "err %d\n%s\n%s", err, opts.errstr, log_buf);
1852                 if (fd_delta && err < 0)
1853                         fprintf(stderr, "loader prog leaked %d FDs\n",
1854                                 fd_delta);
1855         }
1856         free(log_buf);
1857         return err;
1858 }
1859
1860 static int do_loader(int argc, char **argv)
1861 {
1862         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts);
1863         DECLARE_LIBBPF_OPTS(gen_loader_opts, gen);
1864         struct bpf_object *obj;
1865         const char *file;
1866         int err = 0;
1867
1868         if (!REQ_ARGS(1))
1869                 return -1;
1870         file = GET_ARG();
1871
1872         if (verifier_logs)
1873                 /* log_level1 + log_level2 + stats, but not stable UAPI */
1874                 open_opts.kernel_log_level = 1 + 2 + 4;
1875
1876         obj = bpf_object__open_file(file, &open_opts);
1877         if (!obj) {
1878                 p_err("failed to open object file");
1879                 goto err_close_obj;
1880         }
1881
1882         err = bpf_object__gen_loader(obj, &gen);
1883         if (err)
1884                 goto err_close_obj;
1885
1886         err = bpf_object__load(obj);
1887         if (err) {
1888                 p_err("failed to load object file");
1889                 goto err_close_obj;
1890         }
1891
1892         if (verifier_logs) {
1893                 struct dump_data dd = {};
1894
1895                 kernel_syms_load(&dd);
1896                 dump_xlated_plain(&dd, (void *)gen.insns, gen.insns_sz, false, false);
1897                 kernel_syms_destroy(&dd);
1898         }
1899         err = try_loader(&gen);
1900 err_close_obj:
1901         bpf_object__close(obj);
1902         return err;
1903 }
1904
1905 static int do_load(int argc, char **argv)
1906 {
1907         if (use_loader)
1908                 return do_loader(argc, argv);
1909         return load_with_options(argc, argv, true);
1910 }
1911
1912 static int do_loadall(int argc, char **argv)
1913 {
1914         return load_with_options(argc, argv, false);
1915 }
1916
1917 #ifdef BPFTOOL_WITHOUT_SKELETONS
1918
1919 static int do_profile(int argc, char **argv)
1920 {
1921         p_err("bpftool prog profile command is not supported. Please build bpftool with clang >= 10.0.0");
1922         return 0;
1923 }
1924
1925 #else /* BPFTOOL_WITHOUT_SKELETONS */
1926
1927 #include "profiler.skel.h"
1928
1929 struct profile_metric {
1930         const char *name;
1931         struct bpf_perf_event_value val;
1932         struct perf_event_attr attr;
1933         bool selected;
1934
1935         /* calculate ratios like instructions per cycle */
1936         const int ratio_metric; /* 0 for N/A, 1 for index 0 (cycles) */
1937         const char *ratio_desc;
1938         const float ratio_mul;
1939 } metrics[] = {
1940         {
1941                 .name = "cycles",
1942                 .attr = {
1943                         .type = PERF_TYPE_HARDWARE,
1944                         .config = PERF_COUNT_HW_CPU_CYCLES,
1945                         .exclude_user = 1,
1946                 },
1947         },
1948         {
1949                 .name = "instructions",
1950                 .attr = {
1951                         .type = PERF_TYPE_HARDWARE,
1952                         .config = PERF_COUNT_HW_INSTRUCTIONS,
1953                         .exclude_user = 1,
1954                 },
1955                 .ratio_metric = 1,
1956                 .ratio_desc = "insns per cycle",
1957                 .ratio_mul = 1.0,
1958         },
1959         {
1960                 .name = "l1d_loads",
1961                 .attr = {
1962                         .type = PERF_TYPE_HW_CACHE,
1963                         .config =
1964                                 PERF_COUNT_HW_CACHE_L1D |
1965                                 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
1966                                 (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16),
1967                         .exclude_user = 1,
1968                 },
1969         },
1970         {
1971                 .name = "llc_misses",
1972                 .attr = {
1973                         .type = PERF_TYPE_HW_CACHE,
1974                         .config =
1975                                 PERF_COUNT_HW_CACHE_LL |
1976                                 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
1977                                 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
1978                         .exclude_user = 1
1979                 },
1980                 .ratio_metric = 2,
1981                 .ratio_desc = "LLC misses per million insns",
1982                 .ratio_mul = 1e6,
1983         },
1984         {
1985                 .name = "itlb_misses",
1986                 .attr = {
1987                         .type = PERF_TYPE_HW_CACHE,
1988                         .config =
1989                                 PERF_COUNT_HW_CACHE_ITLB |
1990                                 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
1991                                 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
1992                         .exclude_user = 1
1993                 },
1994                 .ratio_metric = 2,
1995                 .ratio_desc = "itlb misses per million insns",
1996                 .ratio_mul = 1e6,
1997         },
1998         {
1999                 .name = "dtlb_misses",
2000                 .attr = {
2001                         .type = PERF_TYPE_HW_CACHE,
2002                         .config =
2003                                 PERF_COUNT_HW_CACHE_DTLB |
2004                                 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2005                                 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2006                         .exclude_user = 1
2007                 },
2008                 .ratio_metric = 2,
2009                 .ratio_desc = "dtlb misses per million insns",
2010                 .ratio_mul = 1e6,
2011         },
2012 };
2013
2014 static __u64 profile_total_count;
2015
2016 #define MAX_NUM_PROFILE_METRICS 4
2017
2018 static int profile_parse_metrics(int argc, char **argv)
2019 {
2020         unsigned int metric_cnt;
2021         int selected_cnt = 0;
2022         unsigned int i;
2023
2024         metric_cnt = ARRAY_SIZE(metrics);
2025
2026         while (argc > 0) {
2027                 for (i = 0; i < metric_cnt; i++) {
2028                         if (is_prefix(argv[0], metrics[i].name)) {
2029                                 if (!metrics[i].selected)
2030                                         selected_cnt++;
2031                                 metrics[i].selected = true;
2032                                 break;
2033                         }
2034                 }
2035                 if (i == metric_cnt) {
2036                         p_err("unknown metric %s", argv[0]);
2037                         return -1;
2038                 }
2039                 NEXT_ARG();
2040         }
2041         if (selected_cnt > MAX_NUM_PROFILE_METRICS) {
2042                 p_err("too many (%d) metrics, please specify no more than %d metrics at at time",
2043                       selected_cnt, MAX_NUM_PROFILE_METRICS);
2044                 return -1;
2045         }
2046         return selected_cnt;
2047 }
2048
2049 static void profile_read_values(struct profiler_bpf *obj)
2050 {
2051         __u32 m, cpu, num_cpu = obj->rodata->num_cpu;
2052         int reading_map_fd, count_map_fd;
2053         __u64 counts[num_cpu];
2054         __u32 key = 0;
2055         int err;
2056
2057         reading_map_fd = bpf_map__fd(obj->maps.accum_readings);
2058         count_map_fd = bpf_map__fd(obj->maps.counts);
2059         if (reading_map_fd < 0 || count_map_fd < 0) {
2060                 p_err("failed to get fd for map");
2061                 return;
2062         }
2063
2064         err = bpf_map_lookup_elem(count_map_fd, &key, counts);
2065         if (err) {
2066                 p_err("failed to read count_map: %s", strerror(errno));
2067                 return;
2068         }
2069
2070         profile_total_count = 0;
2071         for (cpu = 0; cpu < num_cpu; cpu++)
2072                 profile_total_count += counts[cpu];
2073
2074         for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2075                 struct bpf_perf_event_value values[num_cpu];
2076
2077                 if (!metrics[m].selected)
2078                         continue;
2079
2080                 err = bpf_map_lookup_elem(reading_map_fd, &key, values);
2081                 if (err) {
2082                         p_err("failed to read reading_map: %s",
2083                               strerror(errno));
2084                         return;
2085                 }
2086                 for (cpu = 0; cpu < num_cpu; cpu++) {
2087                         metrics[m].val.counter += values[cpu].counter;
2088                         metrics[m].val.enabled += values[cpu].enabled;
2089                         metrics[m].val.running += values[cpu].running;
2090                 }
2091                 key++;
2092         }
2093 }
2094
2095 static void profile_print_readings_json(void)
2096 {
2097         __u32 m;
2098
2099         jsonw_start_array(json_wtr);
2100         for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2101                 if (!metrics[m].selected)
2102                         continue;
2103                 jsonw_start_object(json_wtr);
2104                 jsonw_string_field(json_wtr, "metric", metrics[m].name);
2105                 jsonw_lluint_field(json_wtr, "run_cnt", profile_total_count);
2106                 jsonw_lluint_field(json_wtr, "value", metrics[m].val.counter);
2107                 jsonw_lluint_field(json_wtr, "enabled", metrics[m].val.enabled);
2108                 jsonw_lluint_field(json_wtr, "running", metrics[m].val.running);
2109
2110                 jsonw_end_object(json_wtr);
2111         }
2112         jsonw_end_array(json_wtr);
2113 }
2114
2115 static void profile_print_readings_plain(void)
2116 {
2117         __u32 m;
2118
2119         printf("\n%18llu %-20s\n", profile_total_count, "run_cnt");
2120         for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2121                 struct bpf_perf_event_value *val = &metrics[m].val;
2122                 int r;
2123
2124                 if (!metrics[m].selected)
2125                         continue;
2126                 printf("%18llu %-20s", val->counter, metrics[m].name);
2127
2128                 r = metrics[m].ratio_metric - 1;
2129                 if (r >= 0 && metrics[r].selected &&
2130                     metrics[r].val.counter > 0) {
2131                         printf("# %8.2f %-30s",
2132                                val->counter * metrics[m].ratio_mul /
2133                                metrics[r].val.counter,
2134                                metrics[m].ratio_desc);
2135                 } else {
2136                         printf("%-41s", "");
2137                 }
2138
2139                 if (val->enabled > val->running)
2140                         printf("(%4.2f%%)",
2141                                val->running * 100.0 / val->enabled);
2142                 printf("\n");
2143         }
2144 }
2145
2146 static void profile_print_readings(void)
2147 {
2148         if (json_output)
2149                 profile_print_readings_json();
2150         else
2151                 profile_print_readings_plain();
2152 }
2153
2154 static char *profile_target_name(int tgt_fd)
2155 {
2156         struct bpf_func_info func_info;
2157         struct bpf_prog_info info = {};
2158         __u32 info_len = sizeof(info);
2159         const struct btf_type *t;
2160         __u32 func_info_rec_size;
2161         struct btf *btf = NULL;
2162         char *name = NULL;
2163         int err;
2164
2165         err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len);
2166         if (err) {
2167                 p_err("failed to get info for prog FD %d", tgt_fd);
2168                 goto out;
2169         }
2170
2171         if (info.btf_id == 0) {
2172                 p_err("prog FD %d doesn't have valid btf", tgt_fd);
2173                 goto out;
2174         }
2175
2176         func_info_rec_size = info.func_info_rec_size;
2177         if (info.nr_func_info == 0) {
2178                 p_err("found 0 func_info for prog FD %d", tgt_fd);
2179                 goto out;
2180         }
2181
2182         memset(&info, 0, sizeof(info));
2183         info.nr_func_info = 1;
2184         info.func_info_rec_size = func_info_rec_size;
2185         info.func_info = ptr_to_u64(&func_info);
2186
2187         err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len);
2188         if (err) {
2189                 p_err("failed to get func_info for prog FD %d", tgt_fd);
2190                 goto out;
2191         }
2192
2193         btf = btf__load_from_kernel_by_id(info.btf_id);
2194         if (!btf) {
2195                 p_err("failed to load btf for prog FD %d", tgt_fd);
2196                 goto out;
2197         }
2198
2199         t = btf__type_by_id(btf, func_info.type_id);
2200         if (!t) {
2201                 p_err("btf %d doesn't have type %d",
2202                       info.btf_id, func_info.type_id);
2203                 goto out;
2204         }
2205         name = strdup(btf__name_by_offset(btf, t->name_off));
2206 out:
2207         btf__free(btf);
2208         return name;
2209 }
2210
2211 static struct profiler_bpf *profile_obj;
2212 static int profile_tgt_fd = -1;
2213 static char *profile_tgt_name;
2214 static int *profile_perf_events;
2215 static int profile_perf_event_cnt;
2216
2217 static void profile_close_perf_events(struct profiler_bpf *obj)
2218 {
2219         int i;
2220
2221         for (i = profile_perf_event_cnt - 1; i >= 0; i--)
2222                 close(profile_perf_events[i]);
2223
2224         free(profile_perf_events);
2225         profile_perf_event_cnt = 0;
2226 }
2227
2228 static int profile_open_perf_event(int mid, int cpu, int map_fd)
2229 {
2230         int pmu_fd;
2231
2232         pmu_fd = syscall(__NR_perf_event_open, &metrics[mid].attr,
2233                          -1 /*pid*/, cpu, -1 /*group_fd*/, 0);
2234         if (pmu_fd < 0) {
2235                 if (errno == ENODEV) {
2236                         p_info("cpu %d may be offline, skip %s profiling.",
2237                                 cpu, metrics[mid].name);
2238                         profile_perf_event_cnt++;
2239                         return 0;
2240                 }
2241                 return -1;
2242         }
2243
2244         if (bpf_map_update_elem(map_fd,
2245                                 &profile_perf_event_cnt,
2246                                 &pmu_fd, BPF_ANY) ||
2247             ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0)) {
2248                 close(pmu_fd);
2249                 return -1;
2250         }
2251
2252         profile_perf_events[profile_perf_event_cnt++] = pmu_fd;
2253         return 0;
2254 }
2255
2256 static int profile_open_perf_events(struct profiler_bpf *obj)
2257 {
2258         unsigned int cpu, m;
2259         int map_fd;
2260
2261         profile_perf_events = calloc(
2262                 sizeof(int), obj->rodata->num_cpu * obj->rodata->num_metric);
2263         if (!profile_perf_events) {
2264                 p_err("failed to allocate memory for perf_event array: %s",
2265                       strerror(errno));
2266                 return -1;
2267         }
2268         map_fd = bpf_map__fd(obj->maps.events);
2269         if (map_fd < 0) {
2270                 p_err("failed to get fd for events map");
2271                 return -1;
2272         }
2273
2274         for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2275                 if (!metrics[m].selected)
2276                         continue;
2277                 for (cpu = 0; cpu < obj->rodata->num_cpu; cpu++) {
2278                         if (profile_open_perf_event(m, cpu, map_fd)) {
2279                                 p_err("failed to create event %s on cpu %d",
2280                                       metrics[m].name, cpu);
2281                                 return -1;
2282                         }
2283                 }
2284         }
2285         return 0;
2286 }
2287
2288 static void profile_print_and_cleanup(void)
2289 {
2290         profile_close_perf_events(profile_obj);
2291         profile_read_values(profile_obj);
2292         profile_print_readings();
2293         profiler_bpf__destroy(profile_obj);
2294
2295         close(profile_tgt_fd);
2296         free(profile_tgt_name);
2297 }
2298
2299 static void int_exit(int signo)
2300 {
2301         profile_print_and_cleanup();
2302         exit(0);
2303 }
2304
2305 static int do_profile(int argc, char **argv)
2306 {
2307         int num_metric, num_cpu, err = -1;
2308         struct bpf_program *prog;
2309         unsigned long duration;
2310         char *endptr;
2311
2312         /* we at least need two args for the prog and one metric */
2313         if (!REQ_ARGS(3))
2314                 return -EINVAL;
2315
2316         /* parse target fd */
2317         profile_tgt_fd = prog_parse_fd(&argc, &argv);
2318         if (profile_tgt_fd < 0) {
2319                 p_err("failed to parse fd");
2320                 return -1;
2321         }
2322
2323         /* parse profiling optional duration */
2324         if (argc > 2 && is_prefix(argv[0], "duration")) {
2325                 NEXT_ARG();
2326                 duration = strtoul(*argv, &endptr, 0);
2327                 if (*endptr)
2328                         usage();
2329                 NEXT_ARG();
2330         } else {
2331                 duration = UINT_MAX;
2332         }
2333
2334         num_metric = profile_parse_metrics(argc, argv);
2335         if (num_metric <= 0)
2336                 goto out;
2337
2338         num_cpu = libbpf_num_possible_cpus();
2339         if (num_cpu <= 0) {
2340                 p_err("failed to identify number of CPUs");
2341                 goto out;
2342         }
2343
2344         profile_obj = profiler_bpf__open();
2345         if (!profile_obj) {
2346                 p_err("failed to open and/or load BPF object");
2347                 goto out;
2348         }
2349
2350         profile_obj->rodata->num_cpu = num_cpu;
2351         profile_obj->rodata->num_metric = num_metric;
2352
2353         /* adjust map sizes */
2354         bpf_map__set_max_entries(profile_obj->maps.events, num_metric * num_cpu);
2355         bpf_map__set_max_entries(profile_obj->maps.fentry_readings, num_metric);
2356         bpf_map__set_max_entries(profile_obj->maps.accum_readings, num_metric);
2357         bpf_map__set_max_entries(profile_obj->maps.counts, 1);
2358
2359         /* change target name */
2360         profile_tgt_name = profile_target_name(profile_tgt_fd);
2361         if (!profile_tgt_name)
2362                 goto out;
2363
2364         bpf_object__for_each_program(prog, profile_obj->obj) {
2365                 err = bpf_program__set_attach_target(prog, profile_tgt_fd,
2366                                                      profile_tgt_name);
2367                 if (err) {
2368                         p_err("failed to set attach target\n");
2369                         goto out;
2370                 }
2371         }
2372
2373         set_max_rlimit();
2374         err = profiler_bpf__load(profile_obj);
2375         if (err) {
2376                 p_err("failed to load profile_obj");
2377                 goto out;
2378         }
2379
2380         err = profile_open_perf_events(profile_obj);
2381         if (err)
2382                 goto out;
2383
2384         err = profiler_bpf__attach(profile_obj);
2385         if (err) {
2386                 p_err("failed to attach profile_obj");
2387                 goto out;
2388         }
2389         signal(SIGINT, int_exit);
2390
2391         sleep(duration);
2392         profile_print_and_cleanup();
2393         return 0;
2394
2395 out:
2396         profile_close_perf_events(profile_obj);
2397         if (profile_obj)
2398                 profiler_bpf__destroy(profile_obj);
2399         close(profile_tgt_fd);
2400         free(profile_tgt_name);
2401         return err;
2402 }
2403
2404 #endif /* BPFTOOL_WITHOUT_SKELETONS */
2405
2406 static int do_help(int argc, char **argv)
2407 {
2408         if (json_output) {
2409                 jsonw_null(json_wtr);
2410                 return 0;
2411         }
2412
2413         fprintf(stderr,
2414                 "Usage: %1$s %2$s { show | list } [PROG]\n"
2415                 "       %1$s %2$s dump xlated PROG [{ file FILE | [opcodes] [linum] [visual] }]\n"
2416                 "       %1$s %2$s dump jited  PROG [{ file FILE | [opcodes] [linum] }]\n"
2417                 "       %1$s %2$s pin   PROG FILE\n"
2418                 "       %1$s %2$s { load | loadall } OBJ  PATH \\\n"
2419                 "                         [type TYPE] [dev NAME] \\\n"
2420                 "                         [map { idx IDX | name NAME } MAP]\\\n"
2421                 "                         [pinmaps MAP_DIR]\n"
2422                 "                         [autoattach]\n"
2423                 "       %1$s %2$s attach PROG ATTACH_TYPE [MAP]\n"
2424                 "       %1$s %2$s detach PROG ATTACH_TYPE [MAP]\n"
2425                 "       %1$s %2$s run PROG \\\n"
2426                 "                         data_in FILE \\\n"
2427                 "                         [data_out FILE [data_size_out L]] \\\n"
2428                 "                         [ctx_in FILE [ctx_out FILE [ctx_size_out M]]] \\\n"
2429                 "                         [repeat N]\n"
2430                 "       %1$s %2$s profile PROG [duration DURATION] METRICs\n"
2431                 "       %1$s %2$s tracelog\n"
2432                 "       %1$s %2$s help\n"
2433                 "\n"
2434                 "       " HELP_SPEC_MAP "\n"
2435                 "       " HELP_SPEC_PROGRAM "\n"
2436                 "       TYPE := { socket | kprobe | kretprobe | classifier | action |\n"
2437                 "                 tracepoint | raw_tracepoint | xdp | perf_event | cgroup/skb |\n"
2438                 "                 cgroup/sock | cgroup/dev | lwt_in | lwt_out | lwt_xmit |\n"
2439                 "                 lwt_seg6local | sockops | sk_skb | sk_msg | lirc_mode2 |\n"
2440                 "                 sk_reuseport | flow_dissector | cgroup/sysctl |\n"
2441                 "                 cgroup/bind4 | cgroup/bind6 | cgroup/post_bind4 |\n"
2442                 "                 cgroup/post_bind6 | cgroup/connect4 | cgroup/connect6 |\n"
2443                 "                 cgroup/getpeername4 | cgroup/getpeername6 |\n"
2444                 "                 cgroup/getsockname4 | cgroup/getsockname6 | cgroup/sendmsg4 |\n"
2445                 "                 cgroup/sendmsg6 | cgroup/recvmsg4 | cgroup/recvmsg6 |\n"
2446                 "                 cgroup/getsockopt | cgroup/setsockopt | cgroup/sock_release |\n"
2447                 "                 struct_ops | fentry | fexit | freplace | sk_lookup }\n"
2448                 "       ATTACH_TYPE := { sk_msg_verdict | sk_skb_verdict | sk_skb_stream_verdict |\n"
2449                 "                        sk_skb_stream_parser | flow_dissector }\n"
2450                 "       METRIC := { cycles | instructions | l1d_loads | llc_misses | itlb_misses | dtlb_misses }\n"
2451                 "       " HELP_SPEC_OPTIONS " |\n"
2452                 "                    {-f|--bpffs} | {-m|--mapcompat} | {-n|--nomount} |\n"
2453                 "                    {-L|--use-loader} }\n"
2454                 "",
2455                 bin_name, argv[-2]);
2456
2457         return 0;
2458 }
2459
2460 static const struct cmd cmds[] = {
2461         { "show",       do_show },
2462         { "list",       do_show },
2463         { "help",       do_help },
2464         { "dump",       do_dump },
2465         { "pin",        do_pin },
2466         { "load",       do_load },
2467         { "loadall",    do_loadall },
2468         { "attach",     do_attach },
2469         { "detach",     do_detach },
2470         { "tracelog",   do_tracelog },
2471         { "run",        do_run },
2472         { "profile",    do_profile },
2473         { 0 }
2474 };
2475
2476 int do_prog(int argc, char **argv)
2477 {
2478         return cmd_select(cmds, argc, argv, do_help);
2479 }