samples/bpf/bpf_load.c

   1 #include <stdio.h>
   2 #include <sys/types.h>
   3 #include <sys/stat.h>
   4 #include <fcntl.h>
   5 #include <libelf.h>
   6 #include <gelf.h>
   7 #include <errno.h>
   8 #include <unistd.h>
   9 #include <string.h>
  10 #include <stdbool.h>
  11 #include <stdlib.h>
  12 #include <linux/bpf.h>
  13 #include <linux/filter.h>
  14 #include <linux/perf_event.h>
  15 #include <linux/netlink.h>
  16 #include <linux/rtnetlink.h>
  17 #include <linux/types.h>
  18 #include <sys/types.h>
  19 #include <sys/socket.h>
  20 #include <sys/syscall.h>
  21 #include <sys/ioctl.h>
  22 #include <sys/mman.h>
  23 #include <poll.h>
  24 #include <ctype.h>
  25 #include <assert.h>
  26 #include "libbpf.h"
  27 #include "bpf_load.h"
  28 #include "perf-sys.h"
  29
  30 #define DEBUGFS "/sys/kernel/debug/tracing/"
  31
  32 static char license[128];
  33 static int kern_version;
  34 static bool processed_sec[128];
  35 char bpf_log_buf[BPF_LOG_BUF_SIZE];
  36 int map_fd[MAX_MAPS];
  37 int prog_fd[MAX_PROGS];
  38 int event_fd[MAX_PROGS];
  39 int prog_cnt;
  40 int prog_array_fd = -1;
  41
  42 struct bpf_map_data map_data[MAX_MAPS];
  43 int map_data_count = 0;
  44
  45 static int populate_prog_array(const char *event, int prog_fd)
  46 {
  47         int ind = atoi(event), err;
  48
  49         err = bpf_map_update_elem(prog_array_fd, &ind, &prog_fd, BPF_ANY);
  50         if (err < 0) {
  51                 printf("failed to store prog_fd in prog_array\n");
  52                 return -1;
  53         }
  54         return 0;
  55 }
  56
  57 static int load_and_attach(const char *event, struct bpf_insn *prog, int size)
  58 {
  59         bool is_socket = strncmp(event, "socket", 6) == 0;
  60         bool is_kprobe = strncmp(event, "kprobe/", 7) == 0;
  61         bool is_kretprobe = strncmp(event, "kretprobe/", 10) == 0;
  62         bool is_tracepoint = strncmp(event, "tracepoint/", 11) == 0;
  63         bool is_xdp = strncmp(event, "xdp", 3) == 0;
  64         bool is_perf_event = strncmp(event, "perf_event", 10) == 0;
  65         bool is_cgroup_skb = strncmp(event, "cgroup/skb", 10) == 0;
  66         bool is_cgroup_sk = strncmp(event, "cgroup/sock", 11) == 0;
  67         size_t insns_cnt = size / sizeof(struct bpf_insn);
  68         enum bpf_prog_type prog_type;
  69         char buf[256];
  70         int fd, efd, err, id;
  71         struct perf_event_attr attr = {};
  72
  73         attr.type = PERF_TYPE_TRACEPOINT;
  74         attr.sample_type = PERF_SAMPLE_RAW;
  75         attr.sample_period = 1;
  76         attr.wakeup_events = 1;
  77
  78         if (is_socket) {
  79                 prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
  80         } else if (is_kprobe || is_kretprobe) {
  81                 prog_type = BPF_PROG_TYPE_KPROBE;
  82         } else if (is_tracepoint) {
  83                 prog_type = BPF_PROG_TYPE_TRACEPOINT;
  84         } else if (is_xdp) {
  85                 prog_type = BPF_PROG_TYPE_XDP;
  86         } else if (is_perf_event) {
  87                 prog_type = BPF_PROG_TYPE_PERF_EVENT;
  88         } else if (is_cgroup_skb) {
  89                 prog_type = BPF_PROG_TYPE_CGROUP_SKB;
  90         } else if (is_cgroup_sk) {
  91                 prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
  92         } else {
  93                 printf("Unknown event '%s'\n", event);
  94                 return -1;
  95         }
  96
  97         fd = bpf_load_program(prog_type, prog, insns_cnt, license, kern_version,
  98                               bpf_log_buf, BPF_LOG_BUF_SIZE);
  99         if (fd < 0) {
 100                 printf("bpf_load_program() err=%d\n%s", errno, bpf_log_buf);
 101                 return -1;
 102         }
 103
 104         prog_fd[prog_cnt++] = fd;
 105
 106         if (is_xdp || is_perf_event || is_cgroup_skb || is_cgroup_sk)
 107                 return 0;
 108
 109         if (is_socket) {
 110                 event += 6;
 111                 if (*event != '/')
 112                         return 0;
 113                 event++;
 114                 if (!isdigit(*event)) {
 115                         printf("invalid prog number\n");
 116                         return -1;
 117                 }
 118                 return populate_prog_array(event, fd);
 119         }
 120
 121         if (is_kprobe || is_kretprobe) {
 122                 if (is_kprobe)
 123                         event += 7;
 124                 else
 125                         event += 10;
 126
 127                 if (*event == 0) {
 128                         printf("event name cannot be empty\n");
 129                         return -1;
 130                 }
 131
 132                 if (isdigit(*event))
 133                         return populate_prog_array(event, fd);
 134
 135                 snprintf(buf, sizeof(buf),
 136                          "echo '%c:%s %s' >> /sys/kernel/debug/tracing/kprobe_events",
 137                          is_kprobe ? 'p' : 'r', event, event);
 138                 err = system(buf);
 139                 if (err < 0) {
 140                         printf("failed to create kprobe '%s' error '%s'\n",
 141                                event, strerror(errno));
 142                         return -1;
 143                 }
 144
 145                 strcpy(buf, DEBUGFS);
 146                 strcat(buf, "events/kprobes/");
 147                 strcat(buf, event);
 148                 strcat(buf, "/id");
 149         } else if (is_tracepoint) {
 150                 event += 11;
 151
 152                 if (*event == 0) {
 153                         printf("event name cannot be empty\n");
 154                         return -1;
 155                 }
 156                 strcpy(buf, DEBUGFS);
 157                 strcat(buf, "events/");
 158                 strcat(buf, event);
 159                 strcat(buf, "/id");
 160         }
 161
 162         efd = open(buf, O_RDONLY, 0);
 163         if (efd < 0) {
 164                 printf("failed to open event %s\n", event);
 165                 return -1;
 166         }
 167
 168         err = read(efd, buf, sizeof(buf));
 169         if (err < 0 || err >= sizeof(buf)) {
 170                 printf("read from '%s' failed '%s'\n", event, strerror(errno));
 171                 return -1;
 172         }
 173
 174         close(efd);
 175
 176         buf[err] = 0;
 177         id = atoi(buf);
 178         attr.config = id;
 179
 180         efd = sys_perf_event_open(&attr, -1/*pid*/, 0/*cpu*/, -1/*group_fd*/, 0);
 181         if (efd < 0) {
 182                 printf("event %d fd %d err %s\n", id, efd, strerror(errno));
 183                 return -1;
 184         }
 185         event_fd[prog_cnt - 1] = efd;
 186         ioctl(efd, PERF_EVENT_IOC_ENABLE, 0);
 187         ioctl(efd, PERF_EVENT_IOC_SET_BPF, fd);
 188
 189         return 0;
 190 }
 191
 192 static int load_maps(struct bpf_map_data *maps, int nr_maps,
 193                      fixup_map_cb fixup_map)
 194 {
 195         int i;
 196
 197         for (i = 0; i < nr_maps; i++) {
 198                 if (fixup_map) {
 199                         fixup_map(&maps[i], i);
 200                         /* Allow userspace to assign map FD prior to creation */
 201                         if (maps[i].fd != -1) {
 202                                 map_fd[i] = maps[i].fd;
 203                                 continue;
 204                         }
 205                 }
 206
 207                 if (maps[i].def.type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
 208                     maps[i].def.type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 209                         int inner_map_fd = map_fd[maps[i].def.inner_map_idx];
 210
 211                         map_fd[i] = bpf_create_map_in_map(maps[i].def.type,
 212                                                         maps[i].def.key_size,
 213                                                         inner_map_fd,
 214                                                         maps[i].def.max_entries,
 215                                                         maps[i].def.map_flags);
 216                 } else {
 217                         map_fd[i] = bpf_create_map(maps[i].def.type,
 218                                                    maps[i].def.key_size,
 219                                                    maps[i].def.value_size,
 220                                                    maps[i].def.max_entries,
 221                                                    maps[i].def.map_flags);
 222                 }
 223                 if (map_fd[i] < 0) {
 224                         printf("failed to create a map: %d %s\n",
 225                                errno, strerror(errno));
 226                         return 1;
 227                 }
 228                 maps[i].fd = map_fd[i];
 229
 230                 if (maps[i].def.type == BPF_MAP_TYPE_PROG_ARRAY)
 231                         prog_array_fd = map_fd[i];
 232         }
 233         return 0;
 234 }
 235
 236 static int get_sec(Elf *elf, int i, GElf_Ehdr *ehdr, char **shname,
 237                    GElf_Shdr *shdr, Elf_Data **data)
 238 {
 239         Elf_Scn *scn;
 240
 241         scn = elf_getscn(elf, i);
 242         if (!scn)
 243                 return 1;
 244
 245         if (gelf_getshdr(scn, shdr) != shdr)
 246                 return 2;
 247
 248         *shname = elf_strptr(elf, ehdr->e_shstrndx, shdr->sh_name);
 249         if (!*shname || !shdr->sh_size)
 250                 return 3;
 251
 252         *data = elf_getdata(scn, 0);
 253         if (!*data || elf_getdata(scn, *data) != NULL)
 254                 return 4;
 255
 256         return 0;
 257 }
 258
 259 static int parse_relo_and_apply(Elf_Data *data, Elf_Data *symbols,
 260                                 GElf_Shdr *shdr, struct bpf_insn *insn,
 261                                 struct bpf_map_data *maps, int nr_maps)
 262 {
 263         int i, nrels;
 264
 265         nrels = shdr->sh_size / shdr->sh_entsize;
 266
 267         for (i = 0; i < nrels; i++) {
 268                 GElf_Sym sym;
 269                 GElf_Rel rel;
 270                 unsigned int insn_idx;
 271                 bool match = false;
 272                 int j, map_idx;
 273
 274                 gelf_getrel(data, i, &rel);
 275
 276                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
 277
 278                 gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym);
 279
 280                 if (insn[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
 281                         printf("invalid relo for insn[%d].code 0x%x\n",
 282                                insn_idx, insn[insn_idx].code);
 283                         return 1;
 284                 }
 285                 insn[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
 286
 287                 /* Match FD relocation against recorded map_data[] offset */
 288                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
 289                         if (maps[map_idx].elf_offset == sym.st_value) {
 290                                 match = true;
 291                                 break;
 292                         }
 293                 }
 294                 if (match) {
 295                         insn[insn_idx].imm = maps[map_idx].fd;
 296                 } else {
 297                         printf("invalid relo for insn[%d] no map_data match\n",
 298                                insn_idx);
 299                         return 1;
 300                 }
 301         }
 302
 303         return 0;
 304 }
 305
 306 static int cmp_symbols(const void *l, const void *r)
 307 {
 308         const GElf_Sym *lsym = (const GElf_Sym *)l;
 309         const GElf_Sym *rsym = (const GElf_Sym *)r;
 310
 311         if (lsym->st_value < rsym->st_value)
 312                 return -1;
 313         else if (lsym->st_value > rsym->st_value)
 314                 return 1;
 315         else
 316                 return 0;
 317 }
 318
 319 static int load_elf_maps_section(struct bpf_map_data *maps, int maps_shndx,
 320                                  Elf *elf, Elf_Data *symbols, int strtabidx)
 321 {
 322         int map_sz_elf, map_sz_copy;
 323         bool validate_zero = false;
 324         Elf_Data *data_maps;
 325         int i, nr_maps;
 326         GElf_Sym *sym;
 327         Elf_Scn *scn;
 328         int copy_sz;
 329
 330         if (maps_shndx < 0)
 331                 return -EINVAL;
 332         if (!symbols)
 333                 return -EINVAL;
 334
 335         /* Get data for maps section via elf index */
 336         scn = elf_getscn(elf, maps_shndx);
 337         if (scn)
 338                 data_maps = elf_getdata(scn, NULL);
 339         if (!scn || !data_maps) {
 340                 printf("Failed to get Elf_Data from maps section %d\n",
 341                        maps_shndx);
 342                 return -EINVAL;
 343         }
 344
 345         /* For each map get corrosponding symbol table entry */
 346         sym = calloc(MAX_MAPS+1, sizeof(GElf_Sym));
 347         for (i = 0, nr_maps = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 348                 assert(nr_maps < MAX_MAPS+1);
 349                 if (!gelf_getsym(symbols, i, &sym[nr_maps]))
 350                         continue;
 351                 if (sym[nr_maps].st_shndx != maps_shndx)
 352                         continue;
 353                 /* Only increment iif maps section */
 354                 nr_maps++;
 355         }
 356
 357         /* Align to map_fd[] order, via sort on offset in sym.st_value */
 358         qsort(sym, nr_maps, sizeof(GElf_Sym), cmp_symbols);
 359
 360         /* Keeping compatible with ELF maps section changes
 361          * ------------------------------------------------
 362          * The program size of struct bpf_map_def is known by loader
 363          * code, but struct stored in ELF file can be different.
 364          *
 365          * Unfortunately sym[i].st_size is zero.  To calculate the
 366          * struct size stored in the ELF file, assume all struct have
 367          * the same size, and simply divide with number of map
 368          * symbols.
 369          */
 370         map_sz_elf = data_maps->d_size / nr_maps;
 371         map_sz_copy = sizeof(struct bpf_map_def);
 372         if (map_sz_elf < map_sz_copy) {
 373                 /*
 374                  * Backward compat, loading older ELF file with
 375                  * smaller struct, keeping remaining bytes zero.
 376                  */
 377                 map_sz_copy = map_sz_elf;
 378         } else if (map_sz_elf > map_sz_copy) {
 379                 /*
 380                  * Forward compat, loading newer ELF file with larger
 381                  * struct with unknown features. Assume zero means
 382                  * feature not used.  Thus, validate rest of struct
 383                  * data is zero.
 384                  */
 385                 validate_zero = true;
 386         }
 387
 388         /* Memcpy relevant part of ELF maps data to loader maps */
 389         for (i = 0; i < nr_maps; i++) {
 390                 unsigned char *addr, *end;
 391                 struct bpf_map_def *def;
 392                 const char *map_name;
 393                 size_t offset;
 394
 395                 map_name = elf_strptr(elf, strtabidx, sym[i].st_name);
 396                 maps[i].name = strdup(map_name);
 397                 if (!maps[i].name) {
 398                         printf("strdup(%s): %s(%d)\n", map_name,
 399                                strerror(errno), errno);
 400                         free(sym);
 401                         return -errno;
 402                 }
 403
 404                 /* Symbol value is offset into ELF maps section data area */
 405                 offset = sym[i].st_value;
 406                 def = (struct bpf_map_def *)(data_maps->d_buf + offset);
 407                 maps[i].elf_offset = offset;
 408                 memset(&maps[i].def, 0, sizeof(struct bpf_map_def));
 409                 memcpy(&maps[i].def, def, map_sz_copy);
 410
 411                 /* Verify no newer features were requested */
 412                 if (validate_zero) {
 413                         addr = (unsigned char*) def + map_sz_copy;
 414                         end  = (unsigned char*) def + map_sz_elf;
 415                         for (; addr < end; addr++) {
 416                                 if (*addr != 0) {
 417                                         free(sym);
 418                                         return -EFBIG;
 419                                 }
 420                         }
 421                 }
 422         }
 423
 424         free(sym);
 425         return nr_maps;
 426 }
 427
 428 static int do_load_bpf_file(const char *path, fixup_map_cb fixup_map)
 429 {
 430         int fd, i, ret, maps_shndx = -1, strtabidx = -1;
 431         Elf *elf;
 432         GElf_Ehdr ehdr;
 433         GElf_Shdr shdr, shdr_prog;
 434         Elf_Data *data, *data_prog, *data_maps = NULL, *symbols = NULL;
 435         char *shname, *shname_prog;
 436         int nr_maps = 0;
 437
 438         /* reset global variables */
 439         kern_version = 0;
 440         memset(license, 0, sizeof(license));
 441         memset(processed_sec, 0, sizeof(processed_sec));
 442
 443         if (elf_version(EV_CURRENT) == EV_NONE)
 444                 return 1;
 445
 446         fd = open(path, O_RDONLY, 0);
 447         if (fd < 0)
 448                 return 1;
 449
 450         elf = elf_begin(fd, ELF_C_READ, NULL);
 451
 452         if (!elf)
 453                 return 1;
 454
 455         if (gelf_getehdr(elf, &ehdr) != &ehdr)
 456                 return 1;
 457
 458         /* clear all kprobes */
 459         i = system("echo \"\" > /sys/kernel/debug/tracing/kprobe_events");
 460
 461         /* scan over all elf sections to get license and map info */
 462         for (i = 1; i < ehdr.e_shnum; i++) {
 463
 464                 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
 465                         continue;
 466
 467                 if (0) /* helpful for llvm debugging */
 468                         printf("section %d:%s data %p size %zd link %d flags %d\n",
 469                                i, shname, data->d_buf, data->d_size,
 470                                shdr.sh_link, (int) shdr.sh_flags);
 471
 472                 if (strcmp(shname, "license") == 0) {
 473                         processed_sec[i] = true;
 474                         memcpy(license, data->d_buf, data->d_size);
 475                 } else if (strcmp(shname, "version") == 0) {
 476                         processed_sec[i] = true;
 477                         if (data->d_size != sizeof(int)) {
 478                                 printf("invalid size of version section %zd\n",
 479                                        data->d_size);
 480                                 return 1;
 481                         }
 482                         memcpy(&kern_version, data->d_buf, sizeof(int));
 483                 } else if (strcmp(shname, "maps") == 0) {
 484                         int j;
 485
 486                         maps_shndx = i;
 487                         data_maps = data;
 488                         for (j = 0; j < MAX_MAPS; j++)
 489                                 map_data[j].fd = -1;
 490                 } else if (shdr.sh_type == SHT_SYMTAB) {
 491                         strtabidx = shdr.sh_link;
 492                         symbols = data;
 493                 }
 494         }
 495
 496         ret = 1;
 497
 498         if (!symbols) {
 499                 printf("missing SHT_SYMTAB section\n");
 500                 goto done;
 501         }
 502
 503         if (data_maps) {
 504                 nr_maps = load_elf_maps_section(map_data, maps_shndx,
 505                                                 elf, symbols, strtabidx);
 506                 if (nr_maps < 0) {
 507                         printf("Error: Failed loading ELF maps (errno:%d):%s\n",
 508                                nr_maps, strerror(-nr_maps));
 509                         ret = 1;
 510                         goto done;
 511                 }
 512                 if (load_maps(map_data, nr_maps, fixup_map))
 513                         goto done;
 514                 map_data_count = nr_maps;
 515
 516                 processed_sec[maps_shndx] = true;
 517         }
 518
 519         /* load programs that need map fixup (relocations) */
 520         for (i = 1; i < ehdr.e_shnum; i++) {
 521                 if (processed_sec[i])
 522                         continue;
 523
 524                 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
 525                         continue;
 526                 if (shdr.sh_type == SHT_REL) {
 527                         struct bpf_insn *insns;
 528
 529                         if (get_sec(elf, shdr.sh_info, &ehdr, &shname_prog,
 530                                     &shdr_prog, &data_prog))
 531                                 continue;
 532
 533                         if (shdr_prog.sh_type != SHT_PROGBITS ||
 534                             !(shdr_prog.sh_flags & SHF_EXECINSTR))
 535                                 continue;
 536
 537                         insns = (struct bpf_insn *) data_prog->d_buf;
 538
 539                         processed_sec[shdr.sh_info] = true;
 540                         processed_sec[i] = true;
 541
 542                         if (parse_relo_and_apply(data, symbols, &shdr, insns,
 543                                                  map_data, nr_maps))
 544                                 continue;
 545
 546                         if (memcmp(shname_prog, "kprobe/", 7) == 0 ||
 547                             memcmp(shname_prog, "kretprobe/", 10) == 0 ||
 548                             memcmp(shname_prog, "tracepoint/", 11) == 0 ||
 549                             memcmp(shname_prog, "xdp", 3) == 0 ||
 550                             memcmp(shname_prog, "perf_event", 10) == 0 ||
 551                             memcmp(shname_prog, "socket", 6) == 0 ||
 552                             memcmp(shname_prog, "cgroup/", 7) == 0)
 553                                 load_and_attach(shname_prog, insns, data_prog->d_size);
 554                 }
 555         }
 556
 557         /* load programs that don't use maps */
 558         for (i = 1; i < ehdr.e_shnum; i++) {
 559
 560                 if (processed_sec[i])
 561                         continue;
 562
 563                 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
 564                         continue;
 565
 566                 if (memcmp(shname, "kprobe/", 7) == 0 ||
 567                     memcmp(shname, "kretprobe/", 10) == 0 ||
 568                     memcmp(shname, "tracepoint/", 11) == 0 ||
 569                     memcmp(shname, "xdp", 3) == 0 ||
 570                     memcmp(shname, "perf_event", 10) == 0 ||
 571                     memcmp(shname, "socket", 6) == 0 ||
 572                     memcmp(shname, "cgroup/", 7) == 0)
 573                         load_and_attach(shname, data->d_buf, data->d_size);
 574         }
 575
 576         ret = 0;
 577 done:
 578         close(fd);
 579         return ret;
 580 }
 581
 582 int load_bpf_file(char *path)
 583 {
 584         return do_load_bpf_file(path, NULL);
 585 }
 586
 587 int load_bpf_file_fixup_map(const char *path, fixup_map_cb fixup_map)
 588 {
 589         return do_load_bpf_file(path, fixup_map);
 590 }
 591
 592 void read_trace_pipe(void)
 593 {
 594         int trace_fd;
 595
 596         trace_fd = open(DEBUGFS "trace_pipe", O_RDONLY, 0);
 597         if (trace_fd < 0)
 598                 return;
 599
 600         while (1) {
 601                 static char buf[4096];
 602                 ssize_t sz;
 603
 604                 sz = read(trace_fd, buf, sizeof(buf));
 605                 if (sz > 0) {
 606                         buf[sz] = 0;
 607                         puts(buf);
 608                 }
 609         }
 610 }
 611
 612 #define MAX_SYMS 300000
 613 static struct ksym syms[MAX_SYMS];
 614 static int sym_cnt;
 615
 616 static int ksym_cmp(const void *p1, const void *p2)
 617 {
 618         return ((struct ksym *)p1)->addr - ((struct ksym *)p2)->addr;
 619 }
 620
 621 int load_kallsyms(void)
 622 {
 623         FILE *f = fopen("/proc/kallsyms", "r");
 624         char func[256], buf[256];
 625         char symbol;
 626         void *addr;
 627         int i = 0;
 628
 629         if (!f)
 630                 return -ENOENT;
 631
 632         while (!feof(f)) {
 633                 if (!fgets(buf, sizeof(buf), f))
 634                         break;
 635                 if (sscanf(buf, "%p %c %s", &addr, &symbol, func) != 3)
 636                         break;
 637                 if (!addr)
 638                         continue;
 639                 syms[i].addr = (long) addr;
 640                 syms[i].name = strdup(func);
 641                 i++;
 642         }
 643         sym_cnt = i;
 644         qsort(syms, sym_cnt, sizeof(struct ksym), ksym_cmp);
 645         return 0;
 646 }
 647
 648 struct ksym *ksym_search(long key)
 649 {
 650         int start = 0, end = sym_cnt;
 651         int result;
 652
 653         while (start < end) {
 654                 size_t mid = start + (end - start) / 2;
 655
 656                 result = key - syms[mid].addr;
 657                 if (result < 0)
 658                         end = mid;
 659                 else if (result > 0)
 660                         start = mid + 1;
 661                 else
 662                         return &syms[mid];
 663         }
 664
 665         if (start >= 1 && syms[start - 1].addr < key &&
 666             key < syms[start].addr)
 667                 /* valid ksym */
 668                 return &syms[start - 1];
 669
 670         /* out of range. return _stext */
 671         return &syms[0];
 672 }
 673
 674 int set_link_xdp_fd(int ifindex, int fd, __u32 flags)
 675 {
 676         struct sockaddr_nl sa;
 677         int sock, seq = 0, len, ret = -1;
 678         char buf[4096];
 679         struct nlattr *nla, *nla_xdp;
 680         struct {
 681                 struct nlmsghdr  nh;
 682                 struct ifinfomsg ifinfo;
 683                 char             attrbuf[64];
 684         } req;
 685         struct nlmsghdr *nh;
 686         struct nlmsgerr *err;
 687
 688         memset(&sa, 0, sizeof(sa));
 689         sa.nl_family = AF_NETLINK;
 690
 691         sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
 692         if (sock < 0) {
 693                 printf("open netlink socket: %s\n", strerror(errno));
 694                 return -1;
 695         }
 696
 697         if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
 698                 printf("bind to netlink: %s\n", strerror(errno));
 699                 goto cleanup;
 700         }
 701
 702         memset(&req, 0, sizeof(req));
 703         req.nh.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg));
 704         req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
 705         req.nh.nlmsg_type = RTM_SETLINK;
 706         req.nh.nlmsg_pid = 0;
 707         req.nh.nlmsg_seq = ++seq;
 708         req.ifinfo.ifi_family = AF_UNSPEC;
 709         req.ifinfo.ifi_index = ifindex;
 710
 711         /* started nested attribute for XDP */
 712         nla = (struct nlattr *)(((char *)&req)
 713                                 + NLMSG_ALIGN(req.nh.nlmsg_len));
 714         nla->nla_type = NLA_F_NESTED | 43/*IFLA_XDP*/;
 715         nla->nla_len = NLA_HDRLEN;
 716
 717         /* add XDP fd */
 718         nla_xdp = (struct nlattr *)((char *)nla + nla->nla_len);
 719         nla_xdp->nla_type = 1/*IFLA_XDP_FD*/;
 720         nla_xdp->nla_len = NLA_HDRLEN + sizeof(int);
 721         memcpy((char *)nla_xdp + NLA_HDRLEN, &fd, sizeof(fd));
 722         nla->nla_len += nla_xdp->nla_len;
 723
 724         /* if user passed in any flags, add those too */
 725         if (flags) {
 726                 nla_xdp = (struct nlattr *)((char *)nla + nla->nla_len);
 727                 nla_xdp->nla_type = 3/*IFLA_XDP_FLAGS*/;
 728                 nla_xdp->nla_len = NLA_HDRLEN + sizeof(flags);
 729                 memcpy((char *)nla_xdp + NLA_HDRLEN, &flags, sizeof(flags));
 730                 nla->nla_len += nla_xdp->nla_len;
 731         }
 732
 733         req.nh.nlmsg_len += NLA_ALIGN(nla->nla_len);
 734
 735         if (send(sock, &req, req.nh.nlmsg_len, 0) < 0) {
 736                 printf("send to netlink: %s\n", strerror(errno));
 737                 goto cleanup;
 738         }
 739
 740         len = recv(sock, buf, sizeof(buf), 0);
 741         if (len < 0) {
 742                 printf("recv from netlink: %s\n", strerror(errno));
 743                 goto cleanup;
 744         }
 745
 746         for (nh = (struct nlmsghdr *)buf; NLMSG_OK(nh, len);
 747              nh = NLMSG_NEXT(nh, len)) {
 748                 if (nh->nlmsg_pid != getpid()) {
 749                         printf("Wrong pid %d, expected %d\n",
 750                                nh->nlmsg_pid, getpid());
 751                         goto cleanup;
 752                 }
 753                 if (nh->nlmsg_seq != seq) {
 754                         printf("Wrong seq %d, expected %d\n",
 755                                nh->nlmsg_seq, seq);
 756                         goto cleanup;
 757                 }
 758                 switch (nh->nlmsg_type) {
 759                 case NLMSG_ERROR:
 760                         err = (struct nlmsgerr *)NLMSG_DATA(nh);
 761                         if (!err->error)
 762                                 continue;
 763                         printf("nlmsg error %s\n", strerror(-err->error));
 764                         goto cleanup;
 765                 case NLMSG_DONE:
 766                         break;
 767                 }
 768         }
 769
 770         ret = 0;
 771
 772 cleanup:
 773         close(sock);
 774         return ret;
 775 }