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, bool orphaned)
 446 {
 447         char *memlock;
 448
 449         jsonw_start_object(json_wtr);
 450         print_prog_header_json(info, fd);
 451         print_dev_json(info->ifindex, info->netns_dev, info->netns_ino);
 452
 453         if (info->load_time) {
 454                 char buf[32];
 455
 456                 print_boot_time(info->load_time, buf, sizeof(buf));
 457
 458                 /* Piggy back on load_time, since 0 uid is a valid one */
 459                 jsonw_name(json_wtr, "loaded_at");
 460                 jsonw_printf(json_wtr, "%s", buf);
 461                 jsonw_uint_field(json_wtr, "uid", info->created_by_uid);
 462         }
 463
 464         jsonw_bool_field(json_wtr, "orphaned", orphaned);
 465         jsonw_uint_field(json_wtr, "bytes_xlated", info->xlated_prog_len);
 466
 467         if (info->jited_prog_len) {
 468                 jsonw_bool_field(json_wtr, "jited", true);
 469                 jsonw_uint_field(json_wtr, "bytes_jited", info->jited_prog_len);
 470         } else {
 471                 jsonw_bool_field(json_wtr, "jited", false);
 472         }
 473
 474         memlock = get_fdinfo(fd, "memlock");
 475         if (memlock)
 476                 jsonw_int_field(json_wtr, "bytes_memlock", atoll(memlock));
 477         free(memlock);
 478
 479         if (info->nr_map_ids)
 480                 show_prog_maps(fd, info->nr_map_ids);
 481
 482         if (info->btf_id)
 483                 jsonw_int_field(json_wtr, "btf_id", info->btf_id);
 484
 485         if (!hashmap__empty(prog_table)) {
 486                 struct hashmap_entry *entry;
 487
 488                 jsonw_name(json_wtr, "pinned");
 489                 jsonw_start_array(json_wtr);
 490                 hashmap__for_each_key_entry(prog_table, entry, info->id)
 491                         jsonw_string(json_wtr, entry->pvalue);
 492                 jsonw_end_array(json_wtr);
 493         }
 494
 495         emit_obj_refs_json(refs_table, info->id, json_wtr);
 496
 497         show_prog_metadata(fd, info->nr_map_ids);
 498
 499         jsonw_end_object(json_wtr);
 500 }
 501
 502 static void print_prog_header_plain(struct bpf_prog_info *info, int fd)
 503 {
 504         const char *prog_type_str;
 505         char prog_name[MAX_PROG_FULL_NAME];
 506
 507         printf("%u: ", info->id);
 508         prog_type_str = libbpf_bpf_prog_type_str(info->type);
 509         if (prog_type_str)
 510                 printf("%s  ", prog_type_str);
 511         else
 512                 printf("type %u  ", info->type);
 513
 514         if (*info->name) {
 515                 get_prog_full_name(info, fd, prog_name, sizeof(prog_name));
 516                 printf("name %s  ", prog_name);
 517         }
 518
 519         printf("tag ");
 520         fprint_hex(stdout, info->tag, BPF_TAG_SIZE, "");
 521         print_dev_plain(info->ifindex, info->netns_dev, info->netns_ino);
 522         printf("%s", info->gpl_compatible ? "  gpl" : "");
 523         if (info->run_time_ns)
 524                 printf(" run_time_ns %lld run_cnt %lld",
 525                        info->run_time_ns, info->run_cnt);
 526         if (info->recursion_misses)
 527                 printf(" recursion_misses %lld", info->recursion_misses);
 528         printf("\n");
 529 }
 530
 531 static void print_prog_plain(struct bpf_prog_info *info, int fd, bool orphaned)
 532 {
 533         char *memlock;
 534
 535         print_prog_header_plain(info, fd);
 536
 537         if (info->load_time) {
 538                 char buf[32];
 539
 540                 print_boot_time(info->load_time, buf, sizeof(buf));
 541
 542                 /* Piggy back on load_time, since 0 uid is a valid one */
 543                 printf("\tloaded_at %s  uid %u\n", buf, info->created_by_uid);
 544         }
 545
 546         printf("\txlated %uB", info->xlated_prog_len);
 547
 548         if (info->jited_prog_len)
 549                 printf("  jited %uB", info->jited_prog_len);
 550         else
 551                 printf("  not jited");
 552
 553         memlock = get_fdinfo(fd, "memlock");
 554         if (memlock)
 555                 printf("  memlock %sB", memlock);
 556         free(memlock);
 557
 558         if (orphaned)
 559                 printf("  orphaned");
 560
 561         if (info->nr_map_ids)
 562                 show_prog_maps(fd, info->nr_map_ids);
 563
 564         if (!hashmap__empty(prog_table)) {
 565                 struct hashmap_entry *entry;
 566
 567                 hashmap__for_each_key_entry(prog_table, entry, info->id)
 568                         printf("\n\tpinned %s", (char *)entry->pvalue);
 569         }
 570
 571         if (info->btf_id)
 572                 printf("\n\tbtf_id %d", info->btf_id);
 573
 574         emit_obj_refs_plain(refs_table, info->id, "\n\tpids ");
 575
 576         printf("\n");
 577
 578         show_prog_metadata(fd, info->nr_map_ids);
 579 }
 580
 581 static int show_prog(int fd)
 582 {
 583         struct bpf_prog_info info = {};
 584         __u32 len = sizeof(info);
 585         int err;
 586
 587         err = bpf_prog_get_info_by_fd(fd, &info, &len);
 588         if (err && err != -ENODEV) {
 589                 p_err("can't get prog info: %s", strerror(errno));
 590                 return -1;
 591         }
 592
 593         if (json_output)
 594                 print_prog_json(&info, fd, err == -ENODEV);
 595         else
 596                 print_prog_plain(&info, fd, err == -ENODEV);
 597
 598         return 0;
 599 }
 600
 601 static int do_show_subset(int argc, char **argv)
 602 {
 603         int *fds = NULL;
 604         int nb_fds, i;
 605         int err = -1;
 606
 607         fds = malloc(sizeof(int));
 608         if (!fds) {
 609                 p_err("mem alloc failed");
 610                 return -1;
 611         }
 612         nb_fds = prog_parse_fds(&argc, &argv, &fds);
 613         if (nb_fds < 1)
 614                 goto exit_free;
 615
 616         if (json_output && nb_fds > 1)
 617                 jsonw_start_array(json_wtr);    /* root array */
 618         for (i = 0; i < nb_fds; i++) {
 619                 err = show_prog(fds[i]);
 620                 if (err) {
 621                         for (; i < nb_fds; i++)
 622                                 close(fds[i]);
 623                         break;
 624                 }
 625                 close(fds[i]);
 626         }
 627         if (json_output && nb_fds > 1)
 628                 jsonw_end_array(json_wtr);      /* root array */
 629
 630 exit_free:
 631         free(fds);
 632         return err;
 633 }
 634
 635 static int do_show(int argc, char **argv)
 636 {
 637         __u32 id = 0;
 638         int err;
 639         int fd;
 640
 641         if (show_pinned) {
 642                 prog_table = hashmap__new(hash_fn_for_key_as_id,
 643                                           equal_fn_for_key_as_id, NULL);
 644                 if (IS_ERR(prog_table)) {
 645                         p_err("failed to create hashmap for pinned paths");
 646                         return -1;
 647                 }
 648                 build_pinned_obj_table(prog_table, BPF_OBJ_PROG);
 649         }
 650         build_obj_refs_table(&refs_table, BPF_OBJ_PROG);
 651
 652         if (argc == 2)
 653                 return do_show_subset(argc, argv);
 654
 655         if (argc)
 656                 return BAD_ARG();
 657
 658         if (json_output)
 659                 jsonw_start_array(json_wtr);
 660         while (true) {
 661                 err = bpf_prog_get_next_id(id, &id);
 662                 if (err) {
 663                         if (errno == ENOENT) {
 664                                 err = 0;
 665                                 break;
 666                         }
 667                         p_err("can't get next program: %s%s", strerror(errno),
 668                               errno == EINVAL ? " -- kernel too old?" : "");
 669                         err = -1;
 670                         break;
 671                 }
 672
 673                 fd = bpf_prog_get_fd_by_id(id);
 674                 if (fd < 0) {
 675                         if (errno == ENOENT)
 676                                 continue;
 677                         p_err("can't get prog by id (%u): %s",
 678                               id, strerror(errno));
 679                         err = -1;
 680                         break;
 681                 }
 682
 683                 err = show_prog(fd);
 684                 close(fd);
 685                 if (err)
 686                         break;
 687         }
 688
 689         if (json_output)
 690                 jsonw_end_array(json_wtr);
 691
 692         delete_obj_refs_table(refs_table);
 693
 694         if (show_pinned)
 695                 delete_pinned_obj_table(prog_table);
 696
 697         return err;
 698 }
 699
 700 static int
 701 prog_dump(struct bpf_prog_info *info, enum dump_mode mode,
 702           char *filepath, bool opcodes, bool visual, bool linum)
 703 {
 704         struct bpf_prog_linfo *prog_linfo = NULL;
 705         const char *disasm_opt = NULL;
 706         struct dump_data dd = {};
 707         void *func_info = NULL;
 708         struct btf *btf = NULL;
 709         char func_sig[1024];
 710         unsigned char *buf;
 711         __u32 member_len;
 712         int fd, err = -1;
 713         ssize_t n;
 714
 715         if (mode == DUMP_JITED) {
 716                 if (info->jited_prog_len == 0 || !info->jited_prog_insns) {
 717                         p_info("no instructions returned");
 718                         return -1;
 719                 }
 720                 buf = u64_to_ptr(info->jited_prog_insns);
 721                 member_len = info->jited_prog_len;
 722         } else {        /* DUMP_XLATED */
 723                 if (info->xlated_prog_len == 0 || !info->xlated_prog_insns) {
 724                         p_err("error retrieving insn dump: kernel.kptr_restrict set?");
 725                         return -1;
 726                 }
 727                 buf = u64_to_ptr(info->xlated_prog_insns);
 728                 member_len = info->xlated_prog_len;
 729         }
 730
 731         if (info->btf_id) {
 732                 btf = btf__load_from_kernel_by_id(info->btf_id);
 733                 if (!btf) {
 734                         p_err("failed to get btf");
 735                         return -1;
 736                 }
 737         }
 738
 739         func_info = u64_to_ptr(info->func_info);
 740
 741         if (info->nr_line_info) {
 742                 prog_linfo = bpf_prog_linfo__new(info);
 743                 if (!prog_linfo)
 744                         p_info("error in processing bpf_line_info.  continue without it.");
 745         }
 746
 747         if (filepath) {
 748                 fd = open(filepath, O_WRONLY | O_CREAT | O_TRUNC, 0600);
 749                 if (fd < 0) {
 750                         p_err("can't open file %s: %s", filepath,
 751                               strerror(errno));
 752                         goto exit_free;
 753                 }
 754
 755                 n = write(fd, buf, member_len);
 756                 close(fd);
 757                 if (n != (ssize_t)member_len) {
 758                         p_err("error writing output file: %s",
 759                               n < 0 ? strerror(errno) : "short write");
 760                         goto exit_free;
 761                 }
 762
 763                 if (json_output)
 764                         jsonw_null(json_wtr);
 765         } else if (mode == DUMP_JITED) {
 766                 const char *name = NULL;
 767
 768                 if (info->ifindex) {
 769                         name = ifindex_to_arch(info->ifindex, info->netns_dev,
 770                                                info->netns_ino, &disasm_opt);
 771                         if (!name)
 772                                 goto exit_free;
 773                 }
 774
 775                 if (info->nr_jited_func_lens && info->jited_func_lens) {
 776                         struct kernel_sym *sym = NULL;
 777                         struct bpf_func_info *record;
 778                         char sym_name[SYM_MAX_NAME];
 779                         unsigned char *img = buf;
 780                         __u64 *ksyms = NULL;
 781                         __u32 *lens;
 782                         __u32 i;
 783                         if (info->nr_jited_ksyms) {
 784                                 kernel_syms_load(&dd);
 785                                 ksyms = u64_to_ptr(info->jited_ksyms);
 786                         }
 787
 788                         if (json_output)
 789                                 jsonw_start_array(json_wtr);
 790
 791                         lens = u64_to_ptr(info->jited_func_lens);
 792                         for (i = 0; i < info->nr_jited_func_lens; i++) {
 793                                 if (ksyms) {
 794                                         sym = kernel_syms_search(&dd, ksyms[i]);
 795                                         if (sym)
 796                                                 sprintf(sym_name, "%s", sym->name);
 797                                         else
 798                                                 sprintf(sym_name, "0x%016llx", ksyms[i]);
 799                                 } else {
 800                                         strcpy(sym_name, "unknown");
 801                                 }
 802
 803                                 if (func_info) {
 804                                         record = func_info + i * info->func_info_rec_size;
 805                                         btf_dumper_type_only(btf, record->type_id,
 806                                                              func_sig,
 807                                                              sizeof(func_sig));
 808                                 }
 809
 810                                 if (json_output) {
 811                                         jsonw_start_object(json_wtr);
 812                                         if (func_info && func_sig[0] != '\0') {
 813                                                 jsonw_name(json_wtr, "proto");
 814                                                 jsonw_string(json_wtr, func_sig);
 815                                         }
 816                                         jsonw_name(json_wtr, "name");
 817                                         jsonw_string(json_wtr, sym_name);
 818                                         jsonw_name(json_wtr, "insns");
 819                                 } else {
 820                                         if (func_info && func_sig[0] != '\0')
 821                                                 printf("%s:\n", func_sig);
 822                                         printf("%s:\n", sym_name);
 823                                 }
 824
 825                                 if (disasm_print_insn(img, lens[i], opcodes,
 826                                                       name, disasm_opt, btf,
 827                                                       prog_linfo, ksyms[i], i,
 828                                                       linum))
 829                                         goto exit_free;
 830
 831                                 img += lens[i];
 832
 833                                 if (json_output)
 834                                         jsonw_end_object(json_wtr);
 835                                 else
 836                                         printf("\n");
 837                         }
 838
 839                         if (json_output)
 840                                 jsonw_end_array(json_wtr);
 841                 } else {
 842                         if (disasm_print_insn(buf, member_len, opcodes, name,
 843                                               disasm_opt, btf, NULL, 0, 0,
 844                                               false))
 845                                 goto exit_free;
 846                 }
 847         } else {
 848                 kernel_syms_load(&dd);
 849                 dd.nr_jited_ksyms = info->nr_jited_ksyms;
 850                 dd.jited_ksyms = u64_to_ptr(info->jited_ksyms);
 851                 dd.btf = btf;
 852                 dd.func_info = func_info;
 853                 dd.finfo_rec_size = info->func_info_rec_size;
 854                 dd.prog_linfo = prog_linfo;
 855
 856                 if (json_output)
 857                         dump_xlated_json(&dd, buf, member_len, opcodes, linum);
 858                 else if (visual)
 859                         dump_xlated_cfg(&dd, buf, member_len, opcodes, linum);
 860                 else
 861                         dump_xlated_plain(&dd, buf, member_len, opcodes, linum);
 862                 kernel_syms_destroy(&dd);
 863         }
 864
 865         err = 0;
 866
 867 exit_free:
 868         btf__free(btf);
 869         bpf_prog_linfo__free(prog_linfo);
 870         return err;
 871 }
 872
 873 static int do_dump(int argc, char **argv)
 874 {
 875         struct bpf_prog_info info;
 876         __u32 info_len = sizeof(info);
 877         size_t info_data_sz = 0;
 878         void *info_data = NULL;
 879         char *filepath = NULL;
 880         bool opcodes = false;
 881         bool visual = false;
 882         enum dump_mode mode;
 883         bool linum = false;
 884         int nb_fds, i = 0;
 885         int *fds = NULL;
 886         int err = -1;
 887
 888         if (is_prefix(*argv, "jited")) {
 889                 if (disasm_init())
 890                         return -1;
 891                 mode = DUMP_JITED;
 892         } else if (is_prefix(*argv, "xlated")) {
 893                 mode = DUMP_XLATED;
 894         } else {
 895                 p_err("expected 'xlated' or 'jited', got: %s", *argv);
 896                 return -1;
 897         }
 898         NEXT_ARG();
 899
 900         if (argc < 2)
 901                 usage();
 902
 903         fds = malloc(sizeof(int));
 904         if (!fds) {
 905                 p_err("mem alloc failed");
 906                 return -1;
 907         }
 908         nb_fds = prog_parse_fds(&argc, &argv, &fds);
 909         if (nb_fds < 1)
 910                 goto exit_free;
 911
 912         while (argc) {
 913                 if (is_prefix(*argv, "file")) {
 914                         NEXT_ARG();
 915                         if (!argc) {
 916                                 p_err("expected file path");
 917                                 goto exit_close;
 918                         }
 919                         if (nb_fds > 1) {
 920                                 p_err("several programs matched");
 921                                 goto exit_close;
 922                         }
 923
 924                         filepath = *argv;
 925                         NEXT_ARG();
 926                 } else if (is_prefix(*argv, "opcodes")) {
 927                         opcodes = true;
 928                         NEXT_ARG();
 929                 } else if (is_prefix(*argv, "visual")) {
 930                         if (nb_fds > 1) {
 931                                 p_err("several programs matched");
 932                                 goto exit_close;
 933                         }
 934
 935                         visual = true;
 936                         NEXT_ARG();
 937                 } else if (is_prefix(*argv, "linum")) {
 938                         linum = true;
 939                         NEXT_ARG();
 940                 } else {
 941                         usage();
 942                         goto exit_close;
 943                 }
 944         }
 945
 946         if (filepath && (opcodes || visual || linum)) {
 947                 p_err("'file' is not compatible with 'opcodes', 'visual', or 'linum'");
 948                 goto exit_close;
 949         }
 950         if (json_output && visual) {
 951                 p_err("'visual' is not compatible with JSON output");
 952                 goto exit_close;
 953         }
 954
 955         if (json_output && nb_fds > 1)
 956                 jsonw_start_array(json_wtr);    /* root array */
 957         for (i = 0; i < nb_fds; i++) {
 958                 memset(&info, 0, sizeof(info));
 959
 960                 err = bpf_prog_get_info_by_fd(fds[i], &info, &info_len);
 961                 if (err) {
 962                         p_err("can't get prog info: %s", strerror(errno));
 963                         break;
 964                 }
 965
 966                 err = prep_prog_info(&info, mode, &info_data, &info_data_sz);
 967                 if (err) {
 968                         p_err("can't grow prog info_data");
 969                         break;
 970                 }
 971
 972                 err = bpf_prog_get_info_by_fd(fds[i], &info, &info_len);
 973                 if (err) {
 974                         p_err("can't get prog info: %s", strerror(errno));
 975                         break;
 976                 }
 977
 978                 if (json_output && nb_fds > 1) {
 979                         jsonw_start_object(json_wtr);   /* prog object */
 980                         print_prog_header_json(&info, fds[i]);
 981                         jsonw_name(json_wtr, "insns");
 982                 } else if (nb_fds > 1) {
 983                         print_prog_header_plain(&info, fds[i]);
 984                 }
 985
 986                 err = prog_dump(&info, mode, filepath, opcodes, visual, linum);
 987
 988                 if (json_output && nb_fds > 1)
 989                         jsonw_end_object(json_wtr);     /* prog object */
 990                 else if (i != nb_fds - 1 && nb_fds > 1)
 991                         printf("\n");
 992
 993                 if (err)
 994                         break;
 995                 close(fds[i]);
 996         }
 997         if (json_output && nb_fds > 1)
 998                 jsonw_end_array(json_wtr);      /* root array */
 999
1000 exit_close:
1001         for (; i < nb_fds; i++)
1002                 close(fds[i]);
1003 exit_free:
1004         free(info_data);
1005         free(fds);
1006         return err;
1007 }
1008
1009 static int do_pin(int argc, char **argv)
1010 {
1011         int err;
1012
1013         err = do_pin_any(argc, argv, prog_parse_fd);
1014         if (!err && json_output)
1015                 jsonw_null(json_wtr);
1016         return err;
1017 }
1018
1019 struct map_replace {
1020         int idx;
1021         int fd;
1022         char *name;
1023 };
1024
1025 static int map_replace_compar(const void *p1, const void *p2)
1026 {
1027         const struct map_replace *a = p1, *b = p2;
1028
1029         return a->idx - b->idx;
1030 }
1031
1032 static int parse_attach_detach_args(int argc, char **argv, int *progfd,
1033                                     enum bpf_attach_type *attach_type,
1034                                     int *mapfd)
1035 {
1036         if (!REQ_ARGS(3))
1037                 return -EINVAL;
1038
1039         *progfd = prog_parse_fd(&argc, &argv);
1040         if (*progfd < 0)
1041                 return *progfd;
1042
1043         *attach_type = parse_attach_type(*argv);
1044         if (*attach_type == __MAX_BPF_ATTACH_TYPE) {
1045                 p_err("invalid attach/detach type");
1046                 return -EINVAL;
1047         }
1048
1049         if (*attach_type == BPF_FLOW_DISSECTOR) {
1050                 *mapfd = 0;
1051                 return 0;
1052         }
1053
1054         NEXT_ARG();
1055         if (!REQ_ARGS(2))
1056                 return -EINVAL;
1057
1058         *mapfd = map_parse_fd(&argc, &argv);
1059         if (*mapfd < 0)
1060                 return *mapfd;
1061
1062         return 0;
1063 }
1064
1065 static int do_attach(int argc, char **argv)
1066 {
1067         enum bpf_attach_type attach_type;
1068         int err, progfd;
1069         int mapfd;
1070
1071         err = parse_attach_detach_args(argc, argv,
1072                                        &progfd, &attach_type, &mapfd);
1073         if (err)
1074                 return err;
1075
1076         err = bpf_prog_attach(progfd, mapfd, attach_type, 0);
1077         if (err) {
1078                 p_err("failed prog attach to map");
1079                 return -EINVAL;
1080         }
1081
1082         if (json_output)
1083                 jsonw_null(json_wtr);
1084         return 0;
1085 }
1086
1087 static int do_detach(int argc, char **argv)
1088 {
1089         enum bpf_attach_type attach_type;
1090         int err, progfd;
1091         int mapfd;
1092
1093         err = parse_attach_detach_args(argc, argv,
1094                                        &progfd, &attach_type, &mapfd);
1095         if (err)
1096                 return err;
1097
1098         err = bpf_prog_detach2(progfd, mapfd, attach_type);
1099         if (err) {
1100                 p_err("failed prog detach from map");
1101                 return -EINVAL;
1102         }
1103
1104         if (json_output)
1105                 jsonw_null(json_wtr);
1106         return 0;
1107 }
1108
1109 static int check_single_stdin(char *file_data_in, char *file_ctx_in)
1110 {
1111         if (file_data_in && file_ctx_in &&
1112             !strcmp(file_data_in, "-") && !strcmp(file_ctx_in, "-")) {
1113                 p_err("cannot use standard input for both data_in and ctx_in");
1114                 return -1;
1115         }
1116
1117         return 0;
1118 }
1119
1120 static int get_run_data(const char *fname, void **data_ptr, unsigned int *size)
1121 {
1122         size_t block_size = 256;
1123         size_t buf_size = block_size;
1124         size_t nb_read = 0;
1125         void *tmp;
1126         FILE *f;
1127
1128         if (!fname) {
1129                 *data_ptr = NULL;
1130                 *size = 0;
1131                 return 0;
1132         }
1133
1134         if (!strcmp(fname, "-"))
1135                 f = stdin;
1136         else
1137                 f = fopen(fname, "r");
1138         if (!f) {
1139                 p_err("failed to open %s: %s", fname, strerror(errno));
1140                 return -1;
1141         }
1142
1143         *data_ptr = malloc(block_size);
1144         if (!*data_ptr) {
1145                 p_err("failed to allocate memory for data_in/ctx_in: %s",
1146                       strerror(errno));
1147                 goto err_fclose;
1148         }
1149
1150         while ((nb_read += fread(*data_ptr + nb_read, 1, block_size, f))) {
1151                 if (feof(f))
1152                         break;
1153                 if (ferror(f)) {
1154                         p_err("failed to read data_in/ctx_in from %s: %s",
1155                               fname, strerror(errno));
1156                         goto err_free;
1157                 }
1158                 if (nb_read > buf_size - block_size) {
1159                         if (buf_size == UINT32_MAX) {
1160                                 p_err("data_in/ctx_in is too long (max: %d)",
1161                                       UINT32_MAX);
1162                                 goto err_free;
1163                         }
1164                         /* No space for fread()-ing next chunk; realloc() */
1165                         buf_size *= 2;
1166                         tmp = realloc(*data_ptr, buf_size);
1167                         if (!tmp) {
1168                                 p_err("failed to reallocate data_in/ctx_in: %s",
1169                                       strerror(errno));
1170                                 goto err_free;
1171                         }
1172                         *data_ptr = tmp;
1173                 }
1174         }
1175         if (f != stdin)
1176                 fclose(f);
1177
1178         *size = nb_read;
1179         return 0;
1180
1181 err_free:
1182         free(*data_ptr);
1183         *data_ptr = NULL;
1184 err_fclose:
1185         if (f != stdin)
1186                 fclose(f);
1187         return -1;
1188 }
1189
1190 static void hex_print(void *data, unsigned int size, FILE *f)
1191 {
1192         size_t i, j;
1193         char c;
1194
1195         for (i = 0; i < size; i += 16) {
1196                 /* Row offset */
1197                 fprintf(f, "%07zx\t", i);
1198
1199                 /* Hexadecimal values */
1200                 for (j = i; j < i + 16 && j < size; j++)
1201                         fprintf(f, "%02x%s", *(uint8_t *)(data + j),
1202                                 j % 2 ? " " : "");
1203                 for (; j < i + 16; j++)
1204                         fprintf(f, "  %s", j % 2 ? " " : "");
1205
1206                 /* ASCII values (if relevant), '.' otherwise */
1207                 fprintf(f, "| ");
1208                 for (j = i; j < i + 16 && j < size; j++) {
1209                         c = *(char *)(data + j);
1210                         if (c < ' ' || c > '~')
1211                                 c = '.';
1212                         fprintf(f, "%c%s", c, j == i + 7 ? " " : "");
1213                 }
1214
1215                 fprintf(f, "\n");
1216         }
1217 }
1218
1219 static int
1220 print_run_output(void *data, unsigned int size, const char *fname,
1221                  const char *json_key)
1222 {
1223         size_t nb_written;
1224         FILE *f;
1225
1226         if (!fname)
1227                 return 0;
1228
1229         if (!strcmp(fname, "-")) {
1230                 f = stdout;
1231                 if (json_output) {
1232                         jsonw_name(json_wtr, json_key);
1233                         print_data_json(data, size);
1234                 } else {
1235                         hex_print(data, size, f);
1236                 }
1237                 return 0;
1238         }
1239
1240         f = fopen(fname, "w");
1241         if (!f) {
1242                 p_err("failed to open %s: %s", fname, strerror(errno));
1243                 return -1;
1244         }
1245
1246         nb_written = fwrite(data, 1, size, f);
1247         fclose(f);
1248         if (nb_written != size) {
1249                 p_err("failed to write output data/ctx: %s", strerror(errno));
1250                 return -1;
1251         }
1252
1253         return 0;
1254 }
1255
1256 static int alloc_run_data(void **data_ptr, unsigned int size_out)
1257 {
1258         *data_ptr = calloc(size_out, 1);
1259         if (!*data_ptr) {
1260                 p_err("failed to allocate memory for output data/ctx: %s",
1261                       strerror(errno));
1262                 return -1;
1263         }
1264
1265         return 0;
1266 }
1267
1268 static int do_run(int argc, char **argv)
1269 {
1270         char *data_fname_in = NULL, *data_fname_out = NULL;
1271         char *ctx_fname_in = NULL, *ctx_fname_out = NULL;
1272         const unsigned int default_size = SZ_32K;
1273         void *data_in = NULL, *data_out = NULL;
1274         void *ctx_in = NULL, *ctx_out = NULL;
1275         unsigned int repeat = 1;
1276         int fd, err;
1277         LIBBPF_OPTS(bpf_test_run_opts, test_attr);
1278
1279         if (!REQ_ARGS(4))
1280                 return -1;
1281
1282         fd = prog_parse_fd(&argc, &argv);
1283         if (fd < 0)
1284                 return -1;
1285
1286         while (argc) {
1287                 if (detect_common_prefix(*argv, "data_in", "data_out",
1288                                          "data_size_out", NULL))
1289                         return -1;
1290                 if (detect_common_prefix(*argv, "ctx_in", "ctx_out",
1291                                          "ctx_size_out", NULL))
1292                         return -1;
1293
1294                 if (is_prefix(*argv, "data_in")) {
1295                         NEXT_ARG();
1296                         if (!REQ_ARGS(1))
1297                                 return -1;
1298
1299                         data_fname_in = GET_ARG();
1300                         if (check_single_stdin(data_fname_in, ctx_fname_in))
1301                                 return -1;
1302                 } else if (is_prefix(*argv, "data_out")) {
1303                         NEXT_ARG();
1304                         if (!REQ_ARGS(1))
1305                                 return -1;
1306
1307                         data_fname_out = GET_ARG();
1308                 } else if (is_prefix(*argv, "data_size_out")) {
1309                         char *endptr;
1310
1311                         NEXT_ARG();
1312                         if (!REQ_ARGS(1))
1313                                 return -1;
1314
1315                         test_attr.data_size_out = strtoul(*argv, &endptr, 0);
1316                         if (*endptr) {
1317                                 p_err("can't parse %s as output data size",
1318                                       *argv);
1319                                 return -1;
1320                         }
1321                         NEXT_ARG();
1322                 } else if (is_prefix(*argv, "ctx_in")) {
1323                         NEXT_ARG();
1324                         if (!REQ_ARGS(1))
1325                                 return -1;
1326
1327                         ctx_fname_in = GET_ARG();
1328                         if (check_single_stdin(data_fname_in, ctx_fname_in))
1329                                 return -1;
1330                 } else if (is_prefix(*argv, "ctx_out")) {
1331                         NEXT_ARG();
1332                         if (!REQ_ARGS(1))
1333                                 return -1;
1334
1335                         ctx_fname_out = GET_ARG();
1336                 } else if (is_prefix(*argv, "ctx_size_out")) {
1337                         char *endptr;
1338
1339                         NEXT_ARG();
1340                         if (!REQ_ARGS(1))
1341                                 return -1;
1342
1343                         test_attr.ctx_size_out = strtoul(*argv, &endptr, 0);
1344                         if (*endptr) {
1345                                 p_err("can't parse %s as output context size",
1346                                       *argv);
1347                                 return -1;
1348                         }
1349                         NEXT_ARG();
1350                 } else if (is_prefix(*argv, "repeat")) {
1351                         char *endptr;
1352
1353                         NEXT_ARG();
1354                         if (!REQ_ARGS(1))
1355                                 return -1;
1356
1357                         repeat = strtoul(*argv, &endptr, 0);
1358                         if (*endptr) {
1359                                 p_err("can't parse %s as repeat number",
1360                                       *argv);
1361                                 return -1;
1362                         }
1363                         NEXT_ARG();
1364                 } else {
1365                         p_err("expected no more arguments, 'data_in', 'data_out', 'data_size_out', 'ctx_in', 'ctx_out', 'ctx_size_out' or 'repeat', got: '%s'?",
1366                               *argv);
1367                         return -1;
1368                 }
1369         }
1370
1371         err = get_run_data(data_fname_in, &data_in, &test_attr.data_size_in);
1372         if (err)
1373                 return -1;
1374
1375         if (data_in) {
1376                 if (!test_attr.data_size_out)
1377                         test_attr.data_size_out = default_size;
1378                 err = alloc_run_data(&data_out, test_attr.data_size_out);
1379                 if (err)
1380                         goto free_data_in;
1381         }
1382
1383         err = get_run_data(ctx_fname_in, &ctx_in, &test_attr.ctx_size_in);
1384         if (err)
1385                 goto free_data_out;
1386
1387         if (ctx_in) {
1388                 if (!test_attr.ctx_size_out)
1389                         test_attr.ctx_size_out = default_size;
1390                 err = alloc_run_data(&ctx_out, test_attr.ctx_size_out);
1391                 if (err)
1392                         goto free_ctx_in;
1393         }
1394
1395         test_attr.repeat        = repeat;
1396         test_attr.data_in       = data_in;
1397         test_attr.data_out      = data_out;
1398         test_attr.ctx_in        = ctx_in;
1399         test_attr.ctx_out       = ctx_out;
1400
1401         err = bpf_prog_test_run_opts(fd, &test_attr);
1402         if (err) {
1403                 p_err("failed to run program: %s", strerror(errno));
1404                 goto free_ctx_out;
1405         }
1406
1407         err = 0;
1408
1409         if (json_output)
1410                 jsonw_start_object(json_wtr);   /* root */
1411
1412         /* Do not exit on errors occurring when printing output data/context,
1413          * we still want to print return value and duration for program run.
1414          */
1415         if (test_attr.data_size_out)
1416                 err += print_run_output(test_attr.data_out,
1417                                         test_attr.data_size_out,
1418                                         data_fname_out, "data_out");
1419         if (test_attr.ctx_size_out)
1420                 err += print_run_output(test_attr.ctx_out,
1421                                         test_attr.ctx_size_out,
1422                                         ctx_fname_out, "ctx_out");
1423
1424         if (json_output) {
1425                 jsonw_uint_field(json_wtr, "retval", test_attr.retval);
1426                 jsonw_uint_field(json_wtr, "duration", test_attr.duration);
1427                 jsonw_end_object(json_wtr);     /* root */
1428         } else {
1429                 fprintf(stdout, "Return value: %u, duration%s: %uns\n",
1430                         test_attr.retval,
1431                         repeat > 1 ? " (average)" : "", test_attr.duration);
1432         }
1433
1434 free_ctx_out:
1435         free(ctx_out);
1436 free_ctx_in:
1437         free(ctx_in);
1438 free_data_out:
1439         free(data_out);
1440 free_data_in:
1441         free(data_in);
1442
1443         return err;
1444 }
1445
1446 static int
1447 get_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
1448                       enum bpf_attach_type *expected_attach_type)
1449 {
1450         libbpf_print_fn_t print_backup;
1451         int ret;
1452
1453         ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type);
1454         if (!ret)
1455                 return ret;
1456
1457         /* libbpf_prog_type_by_name() failed, let's re-run with debug level */
1458         print_backup = libbpf_set_print(print_all_levels);
1459         ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type);
1460         libbpf_set_print(print_backup);
1461
1462         return ret;
1463 }
1464
1465 static int
1466 auto_attach_program(struct bpf_program *prog, const char *path)
1467 {
1468         struct bpf_link *link;
1469         int err;
1470
1471         link = bpf_program__attach(prog);
1472         if (!link) {
1473                 p_info("Program %s does not support autoattach, falling back to pinning",
1474                        bpf_program__name(prog));
1475                 return bpf_obj_pin(bpf_program__fd(prog), path);
1476         }
1477
1478         err = bpf_link__pin(link, path);
1479         bpf_link__destroy(link);
1480         return err;
1481 }
1482
1483 static int
1484 auto_attach_programs(struct bpf_object *obj, const char *path)
1485 {
1486         struct bpf_program *prog;
1487         char buf[PATH_MAX];
1488         int err;
1489
1490         bpf_object__for_each_program(prog, obj) {
1491                 err = pathname_concat(buf, sizeof(buf), path, bpf_program__name(prog));
1492                 if (err)
1493                         goto err_unpin_programs;
1494
1495                 err = auto_attach_program(prog, buf);
1496                 if (err)
1497                         goto err_unpin_programs;
1498         }
1499
1500         return 0;
1501
1502 err_unpin_programs:
1503         while ((prog = bpf_object__prev_program(obj, prog))) {
1504                 if (pathname_concat(buf, sizeof(buf), path, bpf_program__name(prog)))
1505                         continue;
1506
1507                 bpf_program__unpin(prog, buf);
1508         }
1509
1510         return err;
1511 }
1512
1513 static int load_with_options(int argc, char **argv, bool first_prog_only)
1514 {
1515         enum bpf_prog_type common_prog_type = BPF_PROG_TYPE_UNSPEC;
1516         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts,
1517                 .relaxed_maps = relaxed_maps,
1518         );
1519         enum bpf_attach_type expected_attach_type;
1520         struct map_replace *map_replace = NULL;
1521         struct bpf_program *prog = NULL, *pos;
1522         unsigned int old_map_fds = 0;
1523         const char *pinmaps = NULL;
1524         __u32 xdpmeta_ifindex = 0;
1525         __u32 offload_ifindex = 0;
1526         bool auto_attach = false;
1527         struct bpf_object *obj;
1528         struct bpf_map *map;
1529         const char *pinfile;
1530         unsigned int i, j;
1531         const char *file;
1532         int idx, err;
1533
1534
1535         if (!REQ_ARGS(2))
1536                 return -1;
1537         file = GET_ARG();
1538         pinfile = GET_ARG();
1539
1540         while (argc) {
1541                 if (is_prefix(*argv, "type")) {
1542                         NEXT_ARG();
1543
1544                         if (common_prog_type != BPF_PROG_TYPE_UNSPEC) {
1545                                 p_err("program type already specified");
1546                                 goto err_free_reuse_maps;
1547                         }
1548                         if (!REQ_ARGS(1))
1549                                 goto err_free_reuse_maps;
1550
1551                         err = libbpf_prog_type_by_name(*argv, &common_prog_type,
1552                                                        &expected_attach_type);
1553                         if (err < 0) {
1554                                 /* Put a '/' at the end of type to appease libbpf */
1555                                 char *type = malloc(strlen(*argv) + 2);
1556
1557                                 if (!type) {
1558                                         p_err("mem alloc failed");
1559                                         goto err_free_reuse_maps;
1560                                 }
1561                                 *type = 0;
1562                                 strcat(type, *argv);
1563                                 strcat(type, "/");
1564
1565                                 err = get_prog_type_by_name(type, &common_prog_type,
1566                                                             &expected_attach_type);
1567                                 free(type);
1568                                 if (err < 0)
1569                                         goto err_free_reuse_maps;
1570                         }
1571
1572                         NEXT_ARG();
1573                 } else if (is_prefix(*argv, "map")) {
1574                         void *new_map_replace;
1575                         char *endptr, *name;
1576                         int fd;
1577
1578                         NEXT_ARG();
1579
1580                         if (!REQ_ARGS(4))
1581                                 goto err_free_reuse_maps;
1582
1583                         if (is_prefix(*argv, "idx")) {
1584                                 NEXT_ARG();
1585
1586                                 idx = strtoul(*argv, &endptr, 0);
1587                                 if (*endptr) {
1588                                         p_err("can't parse %s as IDX", *argv);
1589                                         goto err_free_reuse_maps;
1590                                 }
1591                                 name = NULL;
1592                         } else if (is_prefix(*argv, "name")) {
1593                                 NEXT_ARG();
1594
1595                                 name = *argv;
1596                                 idx = -1;
1597                         } else {
1598                                 p_err("expected 'idx' or 'name', got: '%s'?",
1599                                       *argv);
1600                                 goto err_free_reuse_maps;
1601                         }
1602                         NEXT_ARG();
1603
1604                         fd = map_parse_fd(&argc, &argv);
1605                         if (fd < 0)
1606                                 goto err_free_reuse_maps;
1607
1608                         new_map_replace = libbpf_reallocarray(map_replace,
1609                                                               old_map_fds + 1,
1610                                                               sizeof(*map_replace));
1611                         if (!new_map_replace) {
1612                                 p_err("mem alloc failed");
1613                                 goto err_free_reuse_maps;
1614                         }
1615                         map_replace = new_map_replace;
1616
1617                         map_replace[old_map_fds].idx = idx;
1618                         map_replace[old_map_fds].name = name;
1619                         map_replace[old_map_fds].fd = fd;
1620                         old_map_fds++;
1621                 } else if (is_prefix(*argv, "dev")) {
1622                         p_info("Warning: 'bpftool prog load [...] dev <ifname>' syntax is deprecated.\n"
1623                                "Going further, please use 'offload_dev <ifname>' to offload program to device.\n"
1624                                "For applications using XDP hints only, use 'xdpmeta_dev <ifname>'.");
1625                         goto offload_dev;
1626                 } else if (is_prefix(*argv, "offload_dev")) {
1627 offload_dev:
1628                         NEXT_ARG();
1629
1630                         if (offload_ifindex) {
1631                                 p_err("offload_dev already specified");
1632                                 goto err_free_reuse_maps;
1633                         } else if (xdpmeta_ifindex) {
1634                                 p_err("xdpmeta_dev and offload_dev are mutually exclusive");
1635                                 goto err_free_reuse_maps;
1636                         }
1637                         if (!REQ_ARGS(1))
1638                                 goto err_free_reuse_maps;
1639
1640                         offload_ifindex = if_nametoindex(*argv);
1641                         if (!offload_ifindex) {
1642                                 p_err("unrecognized netdevice '%s': %s",
1643                                       *argv, strerror(errno));
1644                                 goto err_free_reuse_maps;
1645                         }
1646                         NEXT_ARG();
1647                 } else if (is_prefix(*argv, "xdpmeta_dev")) {
1648                         NEXT_ARG();
1649
1650                         if (xdpmeta_ifindex) {
1651                                 p_err("xdpmeta_dev already specified");
1652                                 goto err_free_reuse_maps;
1653                         } else if (offload_ifindex) {
1654                                 p_err("xdpmeta_dev and offload_dev are mutually exclusive");
1655                                 goto err_free_reuse_maps;
1656                         }
1657                         if (!REQ_ARGS(1))
1658                                 goto err_free_reuse_maps;
1659
1660                         xdpmeta_ifindex = if_nametoindex(*argv);
1661                         if (!xdpmeta_ifindex) {
1662                                 p_err("unrecognized netdevice '%s': %s",
1663                                       *argv, strerror(errno));
1664                                 goto err_free_reuse_maps;
1665                         }
1666                         NEXT_ARG();
1667                 } else if (is_prefix(*argv, "pinmaps")) {
1668                         NEXT_ARG();
1669
1670                         if (!REQ_ARGS(1))
1671                                 goto err_free_reuse_maps;
1672
1673                         pinmaps = GET_ARG();
1674                 } else if (is_prefix(*argv, "autoattach")) {
1675                         auto_attach = true;
1676                         NEXT_ARG();
1677                 } else {
1678                         p_err("expected no more arguments, 'type', 'map' or 'dev', got: '%s'?",
1679                               *argv);
1680                         goto err_free_reuse_maps;
1681                 }
1682         }
1683
1684         set_max_rlimit();
1685
1686         if (verifier_logs)
1687                 /* log_level1 + log_level2 + stats, but not stable UAPI */
1688                 open_opts.kernel_log_level = 1 + 2 + 4;
1689
1690         obj = bpf_object__open_file(file, &open_opts);
1691         if (!obj) {
1692                 p_err("failed to open object file");
1693                 goto err_free_reuse_maps;
1694         }
1695
1696         bpf_object__for_each_program(pos, obj) {
1697                 enum bpf_prog_type prog_type = common_prog_type;
1698
1699                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
1700                         const char *sec_name = bpf_program__section_name(pos);
1701
1702                         err = get_prog_type_by_name(sec_name, &prog_type,
1703                                                     &expected_attach_type);
1704                         if (err < 0)
1705                                 goto err_close_obj;
1706                 }
1707
1708                 if (prog_type == BPF_PROG_TYPE_XDP && xdpmeta_ifindex) {
1709                         bpf_program__set_flags(pos, BPF_F_XDP_DEV_BOUND_ONLY);
1710                         bpf_program__set_ifindex(pos, xdpmeta_ifindex);
1711                 } else {
1712                         bpf_program__set_ifindex(pos, offload_ifindex);
1713                 }
1714                 if (bpf_program__type(pos) != prog_type)
1715                         bpf_program__set_type(pos, prog_type);
1716                 bpf_program__set_expected_attach_type(pos, expected_attach_type);
1717         }
1718
1719         qsort(map_replace, old_map_fds, sizeof(*map_replace),
1720               map_replace_compar);
1721
1722         /* After the sort maps by name will be first on the list, because they
1723          * have idx == -1.  Resolve them.
1724          */
1725         j = 0;
1726         while (j < old_map_fds && map_replace[j].name) {
1727                 i = 0;
1728                 bpf_object__for_each_map(map, obj) {
1729                         if (!strcmp(bpf_map__name(map), map_replace[j].name)) {
1730                                 map_replace[j].idx = i;
1731                                 break;
1732                         }
1733                         i++;
1734                 }
1735                 if (map_replace[j].idx == -1) {
1736                         p_err("unable to find map '%s'", map_replace[j].name);
1737                         goto err_close_obj;
1738                 }
1739                 j++;
1740         }
1741         /* Resort if any names were resolved */
1742         if (j)
1743                 qsort(map_replace, old_map_fds, sizeof(*map_replace),
1744                       map_replace_compar);
1745
1746         /* Set ifindex and name reuse */
1747         j = 0;
1748         idx = 0;
1749         bpf_object__for_each_map(map, obj) {
1750                 if (bpf_map__type(map) != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1751                         bpf_map__set_ifindex(map, offload_ifindex);
1752
1753                 if (j < old_map_fds && idx == map_replace[j].idx) {
1754                         err = bpf_map__reuse_fd(map, map_replace[j++].fd);
1755                         if (err) {
1756                                 p_err("unable to set up map reuse: %d", err);
1757                                 goto err_close_obj;
1758                         }
1759
1760                         /* Next reuse wants to apply to the same map */
1761                         if (j < old_map_fds && map_replace[j].idx == idx) {
1762                                 p_err("replacement for map idx %d specified more than once",
1763                                       idx);
1764                                 goto err_close_obj;
1765                         }
1766                 }
1767
1768                 idx++;
1769         }
1770         if (j < old_map_fds) {
1771                 p_err("map idx '%d' not used", map_replace[j].idx);
1772                 goto err_close_obj;
1773         }
1774
1775         err = bpf_object__load(obj);
1776         if (err) {
1777                 p_err("failed to load object file");
1778                 goto err_close_obj;
1779         }
1780
1781         err = mount_bpffs_for_pin(pinfile, !first_prog_only);
1782         if (err)
1783                 goto err_close_obj;
1784
1785         if (first_prog_only) {
1786                 prog = bpf_object__next_program(obj, NULL);
1787                 if (!prog) {
1788                         p_err("object file doesn't contain any bpf program");
1789                         goto err_close_obj;
1790                 }
1791
1792                 if (auto_attach)
1793                         err = auto_attach_program(prog, pinfile);
1794                 else
1795                         err = bpf_obj_pin(bpf_program__fd(prog), pinfile);
1796                 if (err) {
1797                         p_err("failed to pin program %s",
1798                               bpf_program__section_name(prog));
1799                         goto err_close_obj;
1800                 }
1801         } else {
1802                 if (auto_attach)
1803                         err = auto_attach_programs(obj, pinfile);
1804                 else
1805                         err = bpf_object__pin_programs(obj, pinfile);
1806                 if (err) {
1807                         p_err("failed to pin all programs");
1808                         goto err_close_obj;
1809                 }
1810         }
1811
1812         if (pinmaps) {
1813                 err = bpf_object__pin_maps(obj, pinmaps);
1814                 if (err) {
1815                         p_err("failed to pin all maps");
1816                         goto err_unpin;
1817                 }
1818         }
1819
1820         if (json_output)
1821                 jsonw_null(json_wtr);
1822
1823         bpf_object__close(obj);
1824         for (i = 0; i < old_map_fds; i++)
1825                 close(map_replace[i].fd);
1826         free(map_replace);
1827
1828         return 0;
1829
1830 err_unpin:
1831         if (first_prog_only)
1832                 unlink(pinfile);
1833         else
1834                 bpf_object__unpin_programs(obj, pinfile);
1835 err_close_obj:
1836         bpf_object__close(obj);
1837 err_free_reuse_maps:
1838         for (i = 0; i < old_map_fds; i++)
1839                 close(map_replace[i].fd);
1840         free(map_replace);
1841         return -1;
1842 }
1843
1844 static int count_open_fds(void)
1845 {
1846         DIR *dp = opendir("/proc/self/fd");
1847         struct dirent *de;
1848         int cnt = -3;
1849
1850         if (!dp)
1851                 return -1;
1852
1853         while ((de = readdir(dp)))
1854                 cnt++;
1855
1856         closedir(dp);
1857         return cnt;
1858 }
1859
1860 static int try_loader(struct gen_loader_opts *gen)
1861 {
1862         struct bpf_load_and_run_opts opts = {};
1863         struct bpf_loader_ctx *ctx;
1864         int ctx_sz = sizeof(*ctx) + 64 * max(sizeof(struct bpf_map_desc),
1865                                              sizeof(struct bpf_prog_desc));
1866         int log_buf_sz = (1u << 24) - 1;
1867         int err, fds_before, fd_delta;
1868         char *log_buf = NULL;
1869
1870         ctx = alloca(ctx_sz);
1871         memset(ctx, 0, ctx_sz);
1872         ctx->sz = ctx_sz;
1873         if (verifier_logs) {
1874                 ctx->log_level = 1 + 2 + 4;
1875                 ctx->log_size = log_buf_sz;
1876                 log_buf = malloc(log_buf_sz);
1877                 if (!log_buf)
1878                         return -ENOMEM;
1879                 ctx->log_buf = (long) log_buf;
1880         }
1881         opts.ctx = ctx;
1882         opts.data = gen->data;
1883         opts.data_sz = gen->data_sz;
1884         opts.insns = gen->insns;
1885         opts.insns_sz = gen->insns_sz;
1886         fds_before = count_open_fds();
1887         err = bpf_load_and_run(&opts);
1888         fd_delta = count_open_fds() - fds_before;
1889         if (err < 0 || verifier_logs) {
1890                 fprintf(stderr, "err %d\n%s\n%s", err, opts.errstr, log_buf);
1891                 if (fd_delta && err < 0)
1892                         fprintf(stderr, "loader prog leaked %d FDs\n",
1893                                 fd_delta);
1894         }
1895         free(log_buf);
1896         return err;
1897 }
1898
1899 static int do_loader(int argc, char **argv)
1900 {
1901         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts);
1902         DECLARE_LIBBPF_OPTS(gen_loader_opts, gen);
1903         struct bpf_object *obj;
1904         const char *file;
1905         int err = 0;
1906
1907         if (!REQ_ARGS(1))
1908                 return -1;
1909         file = GET_ARG();
1910
1911         if (verifier_logs)
1912                 /* log_level1 + log_level2 + stats, but not stable UAPI */
1913                 open_opts.kernel_log_level = 1 + 2 + 4;
1914
1915         obj = bpf_object__open_file(file, &open_opts);
1916         if (!obj) {
1917                 p_err("failed to open object file");
1918                 goto err_close_obj;
1919         }
1920
1921         err = bpf_object__gen_loader(obj, &gen);
1922         if (err)
1923                 goto err_close_obj;
1924
1925         err = bpf_object__load(obj);
1926         if (err) {
1927                 p_err("failed to load object file");
1928                 goto err_close_obj;
1929         }
1930
1931         if (verifier_logs) {
1932                 struct dump_data dd = {};
1933
1934                 kernel_syms_load(&dd);
1935                 dump_xlated_plain(&dd, (void *)gen.insns, gen.insns_sz, false, false);
1936                 kernel_syms_destroy(&dd);
1937         }
1938         err = try_loader(&gen);
1939 err_close_obj:
1940         bpf_object__close(obj);
1941         return err;
1942 }
1943
1944 static int do_load(int argc, char **argv)
1945 {
1946         if (use_loader)
1947                 return do_loader(argc, argv);
1948         return load_with_options(argc, argv, true);
1949 }
1950
1951 static int do_loadall(int argc, char **argv)
1952 {
1953         return load_with_options(argc, argv, false);
1954 }
1955
1956 #ifdef BPFTOOL_WITHOUT_SKELETONS
1957
1958 static int do_profile(int argc, char **argv)
1959 {
1960         p_err("bpftool prog profile command is not supported. Please build bpftool with clang >= 10.0.0");
1961         return 0;
1962 }
1963
1964 #else /* BPFTOOL_WITHOUT_SKELETONS */
1965
1966 #include "profiler.skel.h"
1967
1968 struct profile_metric {
1969         const char *name;
1970         struct bpf_perf_event_value val;
1971         struct perf_event_attr attr;
1972         bool selected;
1973
1974         /* calculate ratios like instructions per cycle */
1975         const int ratio_metric; /* 0 for N/A, 1 for index 0 (cycles) */
1976         const char *ratio_desc;
1977         const float ratio_mul;
1978 } metrics[] = {
1979         {
1980                 .name = "cycles",
1981                 .attr = {
1982                         .type = PERF_TYPE_HARDWARE,
1983                         .config = PERF_COUNT_HW_CPU_CYCLES,
1984                         .exclude_user = 1,
1985                 },
1986         },
1987         {
1988                 .name = "instructions",
1989                 .attr = {
1990                         .type = PERF_TYPE_HARDWARE,
1991                         .config = PERF_COUNT_HW_INSTRUCTIONS,
1992                         .exclude_user = 1,
1993                 },
1994                 .ratio_metric = 1,
1995                 .ratio_desc = "insns per cycle",
1996                 .ratio_mul = 1.0,
1997         },
1998         {
1999                 .name = "l1d_loads",
2000                 .attr = {
2001                         .type = PERF_TYPE_HW_CACHE,
2002                         .config =
2003                                 PERF_COUNT_HW_CACHE_L1D |
2004                                 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2005                                 (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16),
2006                         .exclude_user = 1,
2007                 },
2008         },
2009         {
2010                 .name = "llc_misses",
2011                 .attr = {
2012                         .type = PERF_TYPE_HW_CACHE,
2013                         .config =
2014                                 PERF_COUNT_HW_CACHE_LL |
2015                                 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2016                                 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2017                         .exclude_user = 1
2018                 },
2019                 .ratio_metric = 2,
2020                 .ratio_desc = "LLC misses per million insns",
2021                 .ratio_mul = 1e6,
2022         },
2023         {
2024                 .name = "itlb_misses",
2025                 .attr = {
2026                         .type = PERF_TYPE_HW_CACHE,
2027                         .config =
2028                                 PERF_COUNT_HW_CACHE_ITLB |
2029                                 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2030                                 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2031                         .exclude_user = 1
2032                 },
2033                 .ratio_metric = 2,
2034                 .ratio_desc = "itlb misses per million insns",
2035                 .ratio_mul = 1e6,
2036         },
2037         {
2038                 .name = "dtlb_misses",
2039                 .attr = {
2040                         .type = PERF_TYPE_HW_CACHE,
2041                         .config =
2042                                 PERF_COUNT_HW_CACHE_DTLB |
2043                                 (PERF_COUNT_HW_CACHE_OP_READ << 8) |
2044                                 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
2045                         .exclude_user = 1
2046                 },
2047                 .ratio_metric = 2,
2048                 .ratio_desc = "dtlb misses per million insns",
2049                 .ratio_mul = 1e6,
2050         },
2051 };
2052
2053 static __u64 profile_total_count;
2054
2055 #define MAX_NUM_PROFILE_METRICS 4
2056
2057 static int profile_parse_metrics(int argc, char **argv)
2058 {
2059         unsigned int metric_cnt;
2060         int selected_cnt = 0;
2061         unsigned int i;
2062
2063         metric_cnt = ARRAY_SIZE(metrics);
2064
2065         while (argc > 0) {
2066                 for (i = 0; i < metric_cnt; i++) {
2067                         if (is_prefix(argv[0], metrics[i].name)) {
2068                                 if (!metrics[i].selected)
2069                                         selected_cnt++;
2070                                 metrics[i].selected = true;
2071                                 break;
2072                         }
2073                 }
2074                 if (i == metric_cnt) {
2075                         p_err("unknown metric %s", argv[0]);
2076                         return -1;
2077                 }
2078                 NEXT_ARG();
2079         }
2080         if (selected_cnt > MAX_NUM_PROFILE_METRICS) {
2081                 p_err("too many (%d) metrics, please specify no more than %d metrics at at time",
2082                       selected_cnt, MAX_NUM_PROFILE_METRICS);
2083                 return -1;
2084         }
2085         return selected_cnt;
2086 }
2087
2088 static void profile_read_values(struct profiler_bpf *obj)
2089 {
2090         __u32 m, cpu, num_cpu = obj->rodata->num_cpu;
2091         int reading_map_fd, count_map_fd;
2092         __u64 counts[num_cpu];
2093         __u32 key = 0;
2094         int err;
2095
2096         reading_map_fd = bpf_map__fd(obj->maps.accum_readings);
2097         count_map_fd = bpf_map__fd(obj->maps.counts);
2098         if (reading_map_fd < 0 || count_map_fd < 0) {
2099                 p_err("failed to get fd for map");
2100                 return;
2101         }
2102
2103         err = bpf_map_lookup_elem(count_map_fd, &key, counts);
2104         if (err) {
2105                 p_err("failed to read count_map: %s", strerror(errno));
2106                 return;
2107         }
2108
2109         profile_total_count = 0;
2110         for (cpu = 0; cpu < num_cpu; cpu++)
2111                 profile_total_count += counts[cpu];
2112
2113         for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2114                 struct bpf_perf_event_value values[num_cpu];
2115
2116                 if (!metrics[m].selected)
2117                         continue;
2118
2119                 err = bpf_map_lookup_elem(reading_map_fd, &key, values);
2120                 if (err) {
2121                         p_err("failed to read reading_map: %s",
2122                               strerror(errno));
2123                         return;
2124                 }
2125                 for (cpu = 0; cpu < num_cpu; cpu++) {
2126                         metrics[m].val.counter += values[cpu].counter;
2127                         metrics[m].val.enabled += values[cpu].enabled;
2128                         metrics[m].val.running += values[cpu].running;
2129                 }
2130                 key++;
2131         }
2132 }
2133
2134 static void profile_print_readings_json(void)
2135 {
2136         __u32 m;
2137
2138         jsonw_start_array(json_wtr);
2139         for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2140                 if (!metrics[m].selected)
2141                         continue;
2142                 jsonw_start_object(json_wtr);
2143                 jsonw_string_field(json_wtr, "metric", metrics[m].name);
2144                 jsonw_lluint_field(json_wtr, "run_cnt", profile_total_count);
2145                 jsonw_lluint_field(json_wtr, "value", metrics[m].val.counter);
2146                 jsonw_lluint_field(json_wtr, "enabled", metrics[m].val.enabled);
2147                 jsonw_lluint_field(json_wtr, "running", metrics[m].val.running);
2148
2149                 jsonw_end_object(json_wtr);
2150         }
2151         jsonw_end_array(json_wtr);
2152 }
2153
2154 static void profile_print_readings_plain(void)
2155 {
2156         __u32 m;
2157
2158         printf("\n%18llu %-20s\n", profile_total_count, "run_cnt");
2159         for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2160                 struct bpf_perf_event_value *val = &metrics[m].val;
2161                 int r;
2162
2163                 if (!metrics[m].selected)
2164                         continue;
2165                 printf("%18llu %-20s", val->counter, metrics[m].name);
2166
2167                 r = metrics[m].ratio_metric - 1;
2168                 if (r >= 0 && metrics[r].selected &&
2169                     metrics[r].val.counter > 0) {
2170                         printf("# %8.2f %-30s",
2171                                val->counter * metrics[m].ratio_mul /
2172                                metrics[r].val.counter,
2173                                metrics[m].ratio_desc);
2174                 } else {
2175                         printf("%-41s", "");
2176                 }
2177
2178                 if (val->enabled > val->running)
2179                         printf("(%4.2f%%)",
2180                                val->running * 100.0 / val->enabled);
2181                 printf("\n");
2182         }
2183 }
2184
2185 static void profile_print_readings(void)
2186 {
2187         if (json_output)
2188                 profile_print_readings_json();
2189         else
2190                 profile_print_readings_plain();
2191 }
2192
2193 static char *profile_target_name(int tgt_fd)
2194 {
2195         struct bpf_func_info func_info;
2196         struct bpf_prog_info info = {};
2197         __u32 info_len = sizeof(info);
2198         const struct btf_type *t;
2199         __u32 func_info_rec_size;
2200         struct btf *btf = NULL;
2201         char *name = NULL;
2202         int err;
2203
2204         err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len);
2205         if (err) {
2206                 p_err("failed to get info for prog FD %d", tgt_fd);
2207                 goto out;
2208         }
2209
2210         if (info.btf_id == 0) {
2211                 p_err("prog FD %d doesn't have valid btf", tgt_fd);
2212                 goto out;
2213         }
2214
2215         func_info_rec_size = info.func_info_rec_size;
2216         if (info.nr_func_info == 0) {
2217                 p_err("found 0 func_info for prog FD %d", tgt_fd);
2218                 goto out;
2219         }
2220
2221         memset(&info, 0, sizeof(info));
2222         info.nr_func_info = 1;
2223         info.func_info_rec_size = func_info_rec_size;
2224         info.func_info = ptr_to_u64(&func_info);
2225
2226         err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len);
2227         if (err) {
2228                 p_err("failed to get func_info for prog FD %d", tgt_fd);
2229                 goto out;
2230         }
2231
2232         btf = btf__load_from_kernel_by_id(info.btf_id);
2233         if (!btf) {
2234                 p_err("failed to load btf for prog FD %d", tgt_fd);
2235                 goto out;
2236         }
2237
2238         t = btf__type_by_id(btf, func_info.type_id);
2239         if (!t) {
2240                 p_err("btf %d doesn't have type %d",
2241                       info.btf_id, func_info.type_id);
2242                 goto out;
2243         }
2244         name = strdup(btf__name_by_offset(btf, t->name_off));
2245 out:
2246         btf__free(btf);
2247         return name;
2248 }
2249
2250 static struct profiler_bpf *profile_obj;
2251 static int profile_tgt_fd = -1;
2252 static char *profile_tgt_name;
2253 static int *profile_perf_events;
2254 static int profile_perf_event_cnt;
2255
2256 static void profile_close_perf_events(struct profiler_bpf *obj)
2257 {
2258         int i;
2259
2260         for (i = profile_perf_event_cnt - 1; i >= 0; i--)
2261                 close(profile_perf_events[i]);
2262
2263         free(profile_perf_events);
2264         profile_perf_event_cnt = 0;
2265 }
2266
2267 static int profile_open_perf_event(int mid, int cpu, int map_fd)
2268 {
2269         int pmu_fd;
2270
2271         pmu_fd = syscall(__NR_perf_event_open, &metrics[mid].attr,
2272                          -1 /*pid*/, cpu, -1 /*group_fd*/, 0);
2273         if (pmu_fd < 0) {
2274                 if (errno == ENODEV) {
2275                         p_info("cpu %d may be offline, skip %s profiling.",
2276                                 cpu, metrics[mid].name);
2277                         profile_perf_event_cnt++;
2278                         return 0;
2279                 }
2280                 return -1;
2281         }
2282
2283         if (bpf_map_update_elem(map_fd,
2284                                 &profile_perf_event_cnt,
2285                                 &pmu_fd, BPF_ANY) ||
2286             ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0)) {
2287                 close(pmu_fd);
2288                 return -1;
2289         }
2290
2291         profile_perf_events[profile_perf_event_cnt++] = pmu_fd;
2292         return 0;
2293 }
2294
2295 static int profile_open_perf_events(struct profiler_bpf *obj)
2296 {
2297         unsigned int cpu, m;
2298         int map_fd;
2299
2300         profile_perf_events = calloc(
2301                 obj->rodata->num_cpu * obj->rodata->num_metric, sizeof(int));
2302         if (!profile_perf_events) {
2303                 p_err("failed to allocate memory for perf_event array: %s",
2304                       strerror(errno));
2305                 return -1;
2306         }
2307         map_fd = bpf_map__fd(obj->maps.events);
2308         if (map_fd < 0) {
2309                 p_err("failed to get fd for events map");
2310                 return -1;
2311         }
2312
2313         for (m = 0; m < ARRAY_SIZE(metrics); m++) {
2314                 if (!metrics[m].selected)
2315                         continue;
2316                 for (cpu = 0; cpu < obj->rodata->num_cpu; cpu++) {
2317                         if (profile_open_perf_event(m, cpu, map_fd)) {
2318                                 p_err("failed to create event %s on cpu %d",
2319                                       metrics[m].name, cpu);
2320                                 return -1;
2321                         }
2322                 }
2323         }
2324         return 0;
2325 }
2326
2327 static void profile_print_and_cleanup(void)
2328 {
2329         profile_close_perf_events(profile_obj);
2330         profile_read_values(profile_obj);
2331         profile_print_readings();
2332         profiler_bpf__destroy(profile_obj);
2333
2334         close(profile_tgt_fd);
2335         free(profile_tgt_name);
2336 }
2337
2338 static void int_exit(int signo)
2339 {
2340         profile_print_and_cleanup();
2341         exit(0);
2342 }
2343
2344 static int do_profile(int argc, char **argv)
2345 {
2346         int num_metric, num_cpu, err = -1;
2347         struct bpf_program *prog;
2348         unsigned long duration;
2349         char *endptr;
2350
2351         /* we at least need two args for the prog and one metric */
2352         if (!REQ_ARGS(3))
2353                 return -EINVAL;
2354
2355         /* parse target fd */
2356         profile_tgt_fd = prog_parse_fd(&argc, &argv);
2357         if (profile_tgt_fd < 0) {
2358                 p_err("failed to parse fd");
2359                 return -1;
2360         }
2361
2362         /* parse profiling optional duration */
2363         if (argc > 2 && is_prefix(argv[0], "duration")) {
2364                 NEXT_ARG();
2365                 duration = strtoul(*argv, &endptr, 0);
2366                 if (*endptr)
2367                         usage();
2368                 NEXT_ARG();
2369         } else {
2370                 duration = UINT_MAX;
2371         }
2372
2373         num_metric = profile_parse_metrics(argc, argv);
2374         if (num_metric <= 0)
2375                 goto out;
2376
2377         num_cpu = libbpf_num_possible_cpus();
2378         if (num_cpu <= 0) {
2379                 p_err("failed to identify number of CPUs");
2380                 goto out;
2381         }
2382
2383         profile_obj = profiler_bpf__open();
2384         if (!profile_obj) {
2385                 p_err("failed to open and/or load BPF object");
2386                 goto out;
2387         }
2388
2389         profile_obj->rodata->num_cpu = num_cpu;
2390         profile_obj->rodata->num_metric = num_metric;
2391
2392         /* adjust map sizes */
2393         bpf_map__set_max_entries(profile_obj->maps.events, num_metric * num_cpu);
2394         bpf_map__set_max_entries(profile_obj->maps.fentry_readings, num_metric);
2395         bpf_map__set_max_entries(profile_obj->maps.accum_readings, num_metric);
2396         bpf_map__set_max_entries(profile_obj->maps.counts, 1);
2397
2398         /* change target name */
2399         profile_tgt_name = profile_target_name(profile_tgt_fd);
2400         if (!profile_tgt_name)
2401                 goto out;
2402
2403         bpf_object__for_each_program(prog, profile_obj->obj) {
2404                 err = bpf_program__set_attach_target(prog, profile_tgt_fd,
2405                                                      profile_tgt_name);
2406                 if (err) {
2407                         p_err("failed to set attach target\n");
2408                         goto out;
2409                 }
2410         }
2411
2412         set_max_rlimit();
2413         err = profiler_bpf__load(profile_obj);
2414         if (err) {
2415                 p_err("failed to load profile_obj");
2416                 goto out;
2417         }
2418
2419         err = profile_open_perf_events(profile_obj);
2420         if (err)
2421                 goto out;
2422
2423         err = profiler_bpf__attach(profile_obj);
2424         if (err) {
2425                 p_err("failed to attach profile_obj");
2426                 goto out;
2427         }
2428         signal(SIGINT, int_exit);
2429
2430         sleep(duration);
2431         profile_print_and_cleanup();
2432         return 0;
2433
2434 out:
2435         profile_close_perf_events(profile_obj);
2436         if (profile_obj)
2437                 profiler_bpf__destroy(profile_obj);
2438         close(profile_tgt_fd);
2439         free(profile_tgt_name);
2440         return err;
2441 }
2442
2443 #endif /* BPFTOOL_WITHOUT_SKELETONS */
2444
2445 static int do_help(int argc, char **argv)
2446 {
2447         if (json_output) {
2448                 jsonw_null(json_wtr);
2449                 return 0;
2450         }
2451
2452         fprintf(stderr,
2453                 "Usage: %1$s %2$s { show | list } [PROG]\n"
2454                 "       %1$s %2$s dump xlated PROG [{ file FILE | [opcodes] [linum] [visual] }]\n"
2455                 "       %1$s %2$s dump jited  PROG [{ file FILE | [opcodes] [linum] }]\n"
2456                 "       %1$s %2$s pin   PROG FILE\n"
2457                 "       %1$s %2$s { load | loadall } OBJ  PATH \\\n"
2458                 "                         [type TYPE] [{ offload_dev | xdpmeta_dev } NAME] \\\n"
2459                 "                         [map { idx IDX | name NAME } MAP]\\\n"
2460                 "                         [pinmaps MAP_DIR]\n"
2461                 "                         [autoattach]\n"
2462                 "       %1$s %2$s attach PROG ATTACH_TYPE [MAP]\n"
2463                 "       %1$s %2$s detach PROG ATTACH_TYPE [MAP]\n"
2464                 "       %1$s %2$s run PROG \\\n"
2465                 "                         data_in FILE \\\n"
2466                 "                         [data_out FILE [data_size_out L]] \\\n"
2467                 "                         [ctx_in FILE [ctx_out FILE [ctx_size_out M]]] \\\n"
2468                 "                         [repeat N]\n"
2469                 "       %1$s %2$s profile PROG [duration DURATION] METRICs\n"
2470                 "       %1$s %2$s tracelog\n"
2471                 "       %1$s %2$s help\n"
2472                 "\n"
2473                 "       " HELP_SPEC_MAP "\n"
2474                 "       " HELP_SPEC_PROGRAM "\n"
2475                 "       TYPE := { socket | kprobe | kretprobe | classifier | action |\n"
2476                 "                 tracepoint | raw_tracepoint | xdp | perf_event | cgroup/skb |\n"
2477                 "                 cgroup/sock | cgroup/dev | lwt_in | lwt_out | lwt_xmit |\n"
2478                 "                 lwt_seg6local | sockops | sk_skb | sk_msg | lirc_mode2 |\n"
2479                 "                 sk_reuseport | flow_dissector | cgroup/sysctl |\n"
2480                 "                 cgroup/bind4 | cgroup/bind6 | cgroup/post_bind4 |\n"
2481                 "                 cgroup/post_bind6 | cgroup/connect4 | cgroup/connect6 |\n"
2482                 "                 cgroup/connect_unix | cgroup/getpeername4 | cgroup/getpeername6 |\n"
2483                 "                 cgroup/getpeername_unix | cgroup/getsockname4 | cgroup/getsockname6 |\n"
2484                 "                 cgroup/getsockname_unix | cgroup/sendmsg4 | cgroup/sendmsg6 |\n"
2485                 "                 cgroup/sendmsg°unix | cgroup/recvmsg4 | cgroup/recvmsg6 | cgroup/recvmsg_unix |\n"
2486                 "                 cgroup/getsockopt | cgroup/setsockopt | cgroup/sock_release |\n"
2487                 "                 struct_ops | fentry | fexit | freplace | sk_lookup }\n"
2488                 "       ATTACH_TYPE := { sk_msg_verdict | sk_skb_verdict | sk_skb_stream_verdict |\n"
2489                 "                        sk_skb_stream_parser | flow_dissector }\n"
2490                 "       METRIC := { cycles | instructions | l1d_loads | llc_misses | itlb_misses | dtlb_misses }\n"
2491                 "       " HELP_SPEC_OPTIONS " |\n"
2492                 "                    {-f|--bpffs} | {-m|--mapcompat} | {-n|--nomount} |\n"
2493                 "                    {-L|--use-loader} }\n"
2494                 "",
2495                 bin_name, argv[-2]);
2496
2497         return 0;
2498 }
2499
2500 static const struct cmd cmds[] = {
2501         { "show",       do_show },
2502         { "list",       do_show },
2503         { "help",       do_help },
2504         { "dump",       do_dump },
2505         { "pin",        do_pin },
2506         { "load",       do_load },
2507         { "loadall",    do_loadall },
2508         { "attach",     do_attach },
2509         { "detach",     do_detach },
2510         { "tracelog",   do_tracelog },
2511         { "run",        do_run },
2512         { "profile",    do_profile },
2513         { 0 }
2514 };
2515
2516 int do_prog(int argc, char **argv)
2517 {
2518         return cmd_select(cmds, argc, argv, do_help);
2519 }