linux/samples/bpf/bpf_load.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2#include <stdio.h>
   3#include <sys/types.h>
   4#include <sys/stat.h>
   5#include <fcntl.h>
   6#include <libelf.h>
   7#include <gelf.h>
   8#include <errno.h>
   9#include <unistd.h>
  10#include <string.h>
  11#include <stdbool.h>
  12#include <stdlib.h>
  13#include <linux/bpf.h>
  14#include <linux/filter.h>
  15#include <linux/perf_event.h>
  16#include <linux/netlink.h>
  17#include <linux/rtnetlink.h>
  18#include <linux/types.h>
  19#include <sys/types.h>
  20#include <sys/socket.h>
  21#include <sys/syscall.h>
  22#include <sys/ioctl.h>
  23#include <sys/mman.h>
  24#include <poll.h>
  25#include <ctype.h>
  26#include <assert.h>
  27#include "libbpf.h"
  28#include "bpf_load.h"
  29#include "perf-sys.h"
  30
  31#define DEBUGFS "/sys/kernel/debug/tracing/"
  32
  33static char license[128];
  34static int kern_version;
  35static bool processed_sec[128];
  36char bpf_log_buf[BPF_LOG_BUF_SIZE];
  37int map_fd[MAX_MAPS];
  38int prog_fd[MAX_PROGS];
  39int event_fd[MAX_PROGS];
  40int prog_cnt;
  41int prog_array_fd = -1;
  42
  43struct bpf_map_data map_data[MAX_MAPS];
  44int map_data_count = 0;
  45
  46static int populate_prog_array(const char *event, int prog_fd)
  47{
  48        int ind = atoi(event), err;
  49
  50        err = bpf_map_update_elem(prog_array_fd, &ind, &prog_fd, BPF_ANY);
  51        if (err < 0) {
  52                printf("failed to store prog_fd in prog_array\n");
  53                return -1;
  54        }
  55        return 0;
  56}
  57
  58static int load_and_attach(const char *event, struct bpf_insn *prog, int size)
  59{
  60        bool is_socket = strncmp(event, "socket", 6) == 0;
  61        bool is_kprobe = strncmp(event, "kprobe/", 7) == 0;
  62        bool is_kretprobe = strncmp(event, "kretprobe/", 10) == 0;
  63        bool is_tracepoint = strncmp(event, "tracepoint/", 11) == 0;
  64        bool is_xdp = strncmp(event, "xdp", 3) == 0;
  65        bool is_perf_event = strncmp(event, "perf_event", 10) == 0;
  66        bool is_cgroup_skb = strncmp(event, "cgroup/skb", 10) == 0;
  67        bool is_cgroup_sk = strncmp(event, "cgroup/sock", 11) == 0;
  68        bool is_sockops = strncmp(event, "sockops", 7) == 0;
  69        bool is_sk_skb = strncmp(event, "sk_skb", 6) == 0;
  70        size_t insns_cnt = size / sizeof(struct bpf_insn);
  71        enum bpf_prog_type prog_type;
  72        char buf[256];
  73        int fd, efd, err, id;
  74        struct perf_event_attr attr = {};
  75
  76        attr.type = PERF_TYPE_TRACEPOINT;
  77        attr.sample_type = PERF_SAMPLE_RAW;
  78        attr.sample_period = 1;
  79        attr.wakeup_events = 1;
  80
  81        if (is_socket) {
  82                prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
  83        } else if (is_kprobe || is_kretprobe) {
  84                prog_type = BPF_PROG_TYPE_KPROBE;
  85        } else if (is_tracepoint) {
  86                prog_type = BPF_PROG_TYPE_TRACEPOINT;
  87        } else if (is_xdp) {
  88                prog_type = BPF_PROG_TYPE_XDP;
  89        } else if (is_perf_event) {
  90                prog_type = BPF_PROG_TYPE_PERF_EVENT;
  91        } else if (is_cgroup_skb) {
  92                prog_type = BPF_PROG_TYPE_CGROUP_SKB;
  93        } else if (is_cgroup_sk) {
  94                prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
  95        } else if (is_sockops) {
  96                prog_type = BPF_PROG_TYPE_SOCK_OPS;
  97        } else if (is_sk_skb) {
  98                prog_type = BPF_PROG_TYPE_SK_SKB;
  99        } else {
 100                printf("Unknown event '%s'\n", event);
 101                return -1;
 102        }
 103
 104        fd = bpf_load_program(prog_type, prog, insns_cnt, license, kern_version,
 105                              bpf_log_buf, BPF_LOG_BUF_SIZE);
 106        if (fd < 0) {
 107                printf("bpf_load_program() err=%d\n%s", errno, bpf_log_buf);
 108                return -1;
 109        }
 110
 111        prog_fd[prog_cnt++] = fd;
 112
 113        if (is_xdp || is_perf_event || is_cgroup_skb || is_cgroup_sk)
 114                return 0;
 115
 116        if (is_socket || is_sockops || is_sk_skb) {
 117                if (is_socket)
 118                        event += 6;
 119                else
 120                        event += 7;
 121                if (*event != '/')
 122                        return 0;
 123                event++;
 124                if (!isdigit(*event)) {
 125                        printf("invalid prog number\n");
 126                        return -1;
 127                }
 128                return populate_prog_array(event, fd);
 129        }
 130
 131        if (is_kprobe || is_kretprobe) {
 132                if (is_kprobe)
 133                        event += 7;
 134                else
 135                        event += 10;
 136
 137                if (*event == 0) {
 138                        printf("event name cannot be empty\n");
 139                        return -1;
 140                }
 141
 142                if (isdigit(*event))
 143                        return populate_prog_array(event, fd);
 144
 145                snprintf(buf, sizeof(buf),
 146                         "echo '%c:%s %s' >> /sys/kernel/debug/tracing/kprobe_events",
 147                         is_kprobe ? 'p' : 'r', event, event);
 148                err = system(buf);
 149                if (err < 0) {
 150                        printf("failed to create kprobe '%s' error '%s'\n",
 151                               event, strerror(errno));
 152                        return -1;
 153                }
 154
 155                strcpy(buf, DEBUGFS);
 156                strcat(buf, "events/kprobes/");
 157                strcat(buf, event);
 158                strcat(buf, "/id");
 159        } else if (is_tracepoint) {
 160                event += 11;
 161
 162                if (*event == 0) {
 163                        printf("event name cannot be empty\n");
 164                        return -1;
 165                }
 166                strcpy(buf, DEBUGFS);
 167                strcat(buf, "events/");
 168                strcat(buf, event);
 169                strcat(buf, "/id");
 170        }
 171
 172        efd = open(buf, O_RDONLY, 0);
 173        if (efd < 0) {
 174                printf("failed to open event %s\n", event);
 175                return -1;
 176        }
 177
 178        err = read(efd, buf, sizeof(buf));
 179        if (err < 0 || err >= sizeof(buf)) {
 180                printf("read from '%s' failed '%s'\n", event, strerror(errno));
 181                return -1;
 182        }
 183
 184        close(efd);
 185
 186        buf[err] = 0;
 187        id = atoi(buf);
 188        attr.config = id;
 189
 190        efd = sys_perf_event_open(&attr, -1/*pid*/, 0/*cpu*/, -1/*group_fd*/, 0);
 191        if (efd < 0) {
 192                printf("event %d fd %d err %s\n", id, efd, strerror(errno));
 193                return -1;
 194        }
 195        event_fd[prog_cnt - 1] = efd;
 196        ioctl(efd, PERF_EVENT_IOC_ENABLE, 0);
 197        ioctl(efd, PERF_EVENT_IOC_SET_BPF, fd);
 198
 199        return 0;
 200}
 201
 202static int load_maps(struct bpf_map_data *maps, int nr_maps,
 203                     fixup_map_cb fixup_map)
 204{
 205        int i, numa_node;
 206
 207        for (i = 0; i < nr_maps; i++) {
 208                if (fixup_map) {
 209                        fixup_map(&maps[i], i);
 210                        /* Allow userspace to assign map FD prior to creation */
 211                        if (maps[i].fd != -1) {
 212                                map_fd[i] = maps[i].fd;
 213                                continue;
 214                        }
 215                }
 216
 217                numa_node = maps[i].def.map_flags & BPF_F_NUMA_NODE ?
 218                        maps[i].def.numa_node : -1;
 219
 220                if (maps[i].def.type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
 221                    maps[i].def.type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 222                        int inner_map_fd = map_fd[maps[i].def.inner_map_idx];
 223
 224                        map_fd[i] = bpf_create_map_in_map_node(maps[i].def.type,
 225                                                        maps[i].def.key_size,
 226                                                        inner_map_fd,
 227                                                        maps[i].def.max_entries,
 228                                                        maps[i].def.map_flags,
 229                                                        numa_node);
 230                } else {
 231                        map_fd[i] = bpf_create_map_node(maps[i].def.type,
 232                                                        maps[i].def.key_size,
 233                                                        maps[i].def.value_size,
 234                                                        maps[i].def.max_entries,
 235                                                        maps[i].def.map_flags,
 236                                                        numa_node);
 237                }
 238                if (map_fd[i] < 0) {
 239                        printf("failed to create a map: %d %s\n",
 240                               errno, strerror(errno));
 241                        return 1;
 242                }
 243                maps[i].fd = map_fd[i];
 244
 245                if (maps[i].def.type == BPF_MAP_TYPE_PROG_ARRAY)
 246                        prog_array_fd = map_fd[i];
 247        }
 248        return 0;
 249}
 250
 251static int get_sec(Elf *elf, int i, GElf_Ehdr *ehdr, char **shname,
 252                   GElf_Shdr *shdr, Elf_Data **data)
 253{
 254        Elf_Scn *scn;
 255
 256        scn = elf_getscn(elf, i);
 257        if (!scn)
 258                return 1;
 259
 260        if (gelf_getshdr(scn, shdr) != shdr)
 261                return 2;
 262
 263        *shname = elf_strptr(elf, ehdr->e_shstrndx, shdr->sh_name);
 264        if (!*shname || !shdr->sh_size)
 265                return 3;
 266
 267        *data = elf_getdata(scn, 0);
 268        if (!*data || elf_getdata(scn, *data) != NULL)
 269                return 4;
 270
 271        return 0;
 272}
 273
 274static int parse_relo_and_apply(Elf_Data *data, Elf_Data *symbols,
 275                                GElf_Shdr *shdr, struct bpf_insn *insn,
 276                                struct bpf_map_data *maps, int nr_maps)
 277{
 278        int i, nrels;
 279
 280        nrels = shdr->sh_size / shdr->sh_entsize;
 281
 282        for (i = 0; i < nrels; i++) {
 283                GElf_Sym sym;
 284                GElf_Rel rel;
 285                unsigned int insn_idx;
 286                bool match = false;
 287                int j, map_idx;
 288
 289                gelf_getrel(data, i, &rel);
 290
 291                insn_idx = rel.r_offset / sizeof(struct bpf_insn);
 292
 293                gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym);
 294
 295                if (insn[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
 296                        printf("invalid relo for insn[%d].code 0x%x\n",
 297                               insn_idx, insn[insn_idx].code);
 298                        return 1;
 299                }
 300                insn[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
 301
 302                /* Match FD relocation against recorded map_data[] offset */
 303                for (map_idx = 0; map_idx < nr_maps; map_idx++) {
 304                        if (maps[map_idx].elf_offset == sym.st_value) {
 305                                match = true;
 306                                break;
 307                        }
 308                }
 309                if (match) {
 310                        insn[insn_idx].imm = maps[map_idx].fd;
 311                } else {
 312                        printf("invalid relo for insn[%d] no map_data match\n",
 313                               insn_idx);
 314                        return 1;
 315                }
 316        }
 317
 318        return 0;
 319}
 320
 321static int cmp_symbols(const void *l, const void *r)
 322{
 323        const GElf_Sym *lsym = (const GElf_Sym *)l;
 324        const GElf_Sym *rsym = (const GElf_Sym *)r;
 325
 326        if (lsym->st_value < rsym->st_value)
 327                return -1;
 328        else if (lsym->st_value > rsym->st_value)
 329                return 1;
 330        else
 331                return 0;
 332}
 333
 334static int load_elf_maps_section(struct bpf_map_data *maps, int maps_shndx,
 335                                 Elf *elf, Elf_Data *symbols, int strtabidx)
 336{
 337        int map_sz_elf, map_sz_copy;
 338        bool validate_zero = false;
 339        Elf_Data *data_maps;
 340        int i, nr_maps;
 341        GElf_Sym *sym;
 342        Elf_Scn *scn;
 343        int copy_sz;
 344
 345        if (maps_shndx < 0)
 346                return -EINVAL;
 347        if (!symbols)
 348                return -EINVAL;
 349
 350        /* Get data for maps section via elf index */
 351        scn = elf_getscn(elf, maps_shndx);
 352        if (scn)
 353                data_maps = elf_getdata(scn, NULL);
 354        if (!scn || !data_maps) {
 355                printf("Failed to get Elf_Data from maps section %d\n",
 356                       maps_shndx);
 357                return -EINVAL;
 358        }
 359
 360        /* For each map get corrosponding symbol table entry */
 361        sym = calloc(MAX_MAPS+1, sizeof(GElf_Sym));
 362        for (i = 0, nr_maps = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 363                assert(nr_maps < MAX_MAPS+1);
 364                if (!gelf_getsym(symbols, i, &sym[nr_maps]))
 365                        continue;
 366                if (sym[nr_maps].st_shndx != maps_shndx)
 367                        continue;
 368                /* Only increment iif maps section */
 369                nr_maps++;
 370        }
 371
 372        /* Align to map_fd[] order, via sort on offset in sym.st_value */
 373        qsort(sym, nr_maps, sizeof(GElf_Sym), cmp_symbols);
 374
 375        /* Keeping compatible with ELF maps section changes
 376         * ------------------------------------------------
 377         * The program size of struct bpf_map_def is known by loader
 378         * code, but struct stored in ELF file can be different.
 379         *
 380         * Unfortunately sym[i].st_size is zero.  To calculate the
 381         * struct size stored in the ELF file, assume all struct have
 382         * the same size, and simply divide with number of map
 383         * symbols.
 384         */
 385        map_sz_elf = data_maps->d_size / nr_maps;
 386        map_sz_copy = sizeof(struct bpf_map_def);
 387        if (map_sz_elf < map_sz_copy) {
 388                /*
 389                 * Backward compat, loading older ELF file with
 390                 * smaller struct, keeping remaining bytes zero.
 391                 */
 392                map_sz_copy = map_sz_elf;
 393        } else if (map_sz_elf > map_sz_copy) {
 394                /*
 395                 * Forward compat, loading newer ELF file with larger
 396                 * struct with unknown features. Assume zero means
 397                 * feature not used.  Thus, validate rest of struct
 398                 * data is zero.
 399                 */
 400                validate_zero = true;
 401        }
 402
 403        /* Memcpy relevant part of ELF maps data to loader maps */
 404        for (i = 0; i < nr_maps; i++) {
 405                unsigned char *addr, *end;
 406                struct bpf_map_def *def;
 407                const char *map_name;
 408                size_t offset;
 409
 410                map_name = elf_strptr(elf, strtabidx, sym[i].st_name);
 411                maps[i].name = strdup(map_name);
 412                if (!maps[i].name) {
 413                        printf("strdup(%s): %s(%d)\n", map_name,
 414                               strerror(errno), errno);
 415                        free(sym);
 416                        return -errno;
 417                }
 418
 419                /* Symbol value is offset into ELF maps section data area */
 420                offset = sym[i].st_value;
 421                def = (struct bpf_map_def *)(data_maps->d_buf + offset);
 422                maps[i].elf_offset = offset;
 423                memset(&maps[i].def, 0, sizeof(struct bpf_map_def));
 424                memcpy(&maps[i].def, def, map_sz_copy);
 425
 426                /* Verify no newer features were requested */
 427                if (validate_zero) {
 428                        addr = (unsigned char*) def + map_sz_copy;
 429                        end  = (unsigned char*) def + map_sz_elf;
 430                        for (; addr < end; addr++) {
 431                                if (*addr != 0) {
 432                                        free(sym);
 433                                        return -EFBIG;
 434                                }
 435                        }
 436                }
 437        }
 438
 439        free(sym);
 440        return nr_maps;
 441}
 442
 443static int do_load_bpf_file(const char *path, fixup_map_cb fixup_map)
 444{
 445        int fd, i, ret, maps_shndx = -1, strtabidx = -1;
 446        Elf *elf;
 447        GElf_Ehdr ehdr;
 448        GElf_Shdr shdr, shdr_prog;
 449        Elf_Data *data, *data_prog, *data_maps = NULL, *symbols = NULL;
 450        char *shname, *shname_prog;
 451        int nr_maps = 0;
 452
 453        /* reset global variables */
 454        kern_version = 0;
 455        memset(license, 0, sizeof(license));
 456        memset(processed_sec, 0, sizeof(processed_sec));
 457
 458        if (elf_version(EV_CURRENT) == EV_NONE)
 459                return 1;
 460
 461        fd = open(path, O_RDONLY, 0);
 462        if (fd < 0)
 463                return 1;
 464
 465        elf = elf_begin(fd, ELF_C_READ, NULL);
 466
 467        if (!elf)
 468                return 1;
 469
 470        if (gelf_getehdr(elf, &ehdr) != &ehdr)
 471                return 1;
 472
 473        /* clear all kprobes */
 474        i = system("echo \"\" > /sys/kernel/debug/tracing/kprobe_events");
 475
 476        /* scan over all elf sections to get license and map info */
 477        for (i = 1; i < ehdr.e_shnum; i++) {
 478
 479                if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
 480                        continue;
 481
 482                if (0) /* helpful for llvm debugging */
 483                        printf("section %d:%s data %p size %zd link %d flags %d\n",
 484                               i, shname, data->d_buf, data->d_size,
 485                               shdr.sh_link, (int) shdr.sh_flags);
 486
 487                if (strcmp(shname, "license") == 0) {
 488                        processed_sec[i] = true;
 489                        memcpy(license, data->d_buf, data->d_size);
 490                } else if (strcmp(shname, "version") == 0) {
 491                        processed_sec[i] = true;
 492                        if (data->d_size != sizeof(int)) {
 493                                printf("invalid size of version section %zd\n",
 494                                       data->d_size);
 495                                return 1;
 496                        }
 497                        memcpy(&kern_version, data->d_buf, sizeof(int));
 498                } else if (strcmp(shname, "maps") == 0) {
 499                        int j;
 500
 501                        maps_shndx = i;
 502                        data_maps = data;
 503                        for (j = 0; j < MAX_MAPS; j++)
 504                                map_data[j].fd = -1;
 505                } else if (shdr.sh_type == SHT_SYMTAB) {
 506                        strtabidx = shdr.sh_link;
 507                        symbols = data;
 508                }
 509        }
 510
 511        ret = 1;
 512
 513        if (!symbols) {
 514                printf("missing SHT_SYMTAB section\n");
 515                goto done;
 516        }
 517
 518        if (data_maps) {
 519                nr_maps = load_elf_maps_section(map_data, maps_shndx,
 520                                                elf, symbols, strtabidx);
 521                if (nr_maps < 0) {
 522                        printf("Error: Failed loading ELF maps (errno:%d):%s\n",
 523                               nr_maps, strerror(-nr_maps));
 524                        ret = 1;
 525                        goto done;
 526                }
 527                if (load_maps(map_data, nr_maps, fixup_map))
 528                        goto done;
 529                map_data_count = nr_maps;
 530
 531                processed_sec[maps_shndx] = true;
 532        }
 533
 534        /* process all relo sections, and rewrite bpf insns for maps */
 535        for (i = 1; i < ehdr.e_shnum; i++) {
 536                if (processed_sec[i])
 537                        continue;
 538
 539                if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
 540                        continue;
 541
 542                if (shdr.sh_type == SHT_REL) {
 543                        struct bpf_insn *insns;
 544
 545                        /* locate prog sec that need map fixup (relocations) */
 546                        if (get_sec(elf, shdr.sh_info, &ehdr, &shname_prog,
 547                                    &shdr_prog, &data_prog))
 548                                continue;
 549
 550                        if (shdr_prog.sh_type != SHT_PROGBITS ||
 551                            !(shdr_prog.sh_flags & SHF_EXECINSTR))
 552                                continue;
 553
 554                        insns = (struct bpf_insn *) data_prog->d_buf;
 555                        processed_sec[i] = true; /* relo section */
 556
 557                        if (parse_relo_and_apply(data, symbols, &shdr, insns,
 558                                                 map_data, nr_maps))
 559                                continue;
 560                }
 561        }
 562
 563        /* load programs */
 564        for (i = 1; i < ehdr.e_shnum; i++) {
 565
 566                if (processed_sec[i])
 567                        continue;
 568
 569                if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
 570                        continue;
 571
 572                if (memcmp(shname, "kprobe/", 7) == 0 ||
 573                    memcmp(shname, "kretprobe/", 10) == 0 ||
 574                    memcmp(shname, "tracepoint/", 11) == 0 ||
 575                    memcmp(shname, "xdp", 3) == 0 ||
 576                    memcmp(shname, "perf_event", 10) == 0 ||
 577                    memcmp(shname, "socket", 6) == 0 ||
 578                    memcmp(shname, "cgroup/", 7) == 0 ||
 579                    memcmp(shname, "sockops", 7) == 0 ||
 580                    memcmp(shname, "sk_skb", 6) == 0) {
 581                        ret = load_and_attach(shname, data->d_buf,
 582                                              data->d_size);
 583                        if (ret != 0)
 584                                goto done;
 585                }
 586        }
 587
 588        ret = 0;
 589done:
 590        close(fd);
 591        return ret;
 592}
 593
 594int load_bpf_file(char *path)
 595{
 596        return do_load_bpf_file(path, NULL);
 597}
 598
 599int load_bpf_file_fixup_map(const char *path, fixup_map_cb fixup_map)
 600{
 601        return do_load_bpf_file(path, fixup_map);
 602}
 603
 604void read_trace_pipe(void)
 605{
 606        int trace_fd;
 607
 608        trace_fd = open(DEBUGFS "trace_pipe", O_RDONLY, 0);
 609        if (trace_fd < 0)
 610                return;
 611
 612        while (1) {
 613                static char buf[4096];
 614                ssize_t sz;
 615
 616                sz = read(trace_fd, buf, sizeof(buf));
 617                if (sz > 0) {
 618                        buf[sz] = 0;
 619                        puts(buf);
 620                }
 621        }
 622}
 623
 624#define MAX_SYMS 300000
 625static struct ksym syms[MAX_SYMS];
 626static int sym_cnt;
 627
 628static int ksym_cmp(const void *p1, const void *p2)
 629{
 630        return ((struct ksym *)p1)->addr - ((struct ksym *)p2)->addr;
 631}
 632
 633int load_kallsyms(void)
 634{
 635        FILE *f = fopen("/proc/kallsyms", "r");
 636        char func[256], buf[256];
 637        char symbol;
 638        void *addr;
 639        int i = 0;
 640
 641        if (!f)
 642                return -ENOENT;
 643
 644        while (!feof(f)) {
 645                if (!fgets(buf, sizeof(buf), f))
 646                        break;
 647                if (sscanf(buf, "%p %c %s", &addr, &symbol, func) != 3)
 648                        break;
 649                if (!addr)
 650                        continue;
 651                syms[i].addr = (long) addr;
 652                syms[i].name = strdup(func);
 653                i++;
 654        }
 655        sym_cnt = i;
 656        qsort(syms, sym_cnt, sizeof(struct ksym), ksym_cmp);
 657        return 0;
 658}
 659
 660struct ksym *ksym_search(long key)
 661{
 662        int start = 0, end = sym_cnt;
 663        int result;
 664
 665        while (start < end) {
 666                size_t mid = start + (end - start) / 2;
 667
 668                result = key - syms[mid].addr;
 669                if (result < 0)
 670                        end = mid;
 671                else if (result > 0)
 672                        start = mid + 1;
 673                else
 674                        return &syms[mid];
 675        }
 676
 677        if (start >= 1 && syms[start - 1].addr < key &&
 678            key < syms[start].addr)
 679                /* valid ksym */
 680                return &syms[start - 1];
 681
 682        /* out of range. return _stext */
 683        return &syms[0];
 684}
 685
 686int set_link_xdp_fd(int ifindex, int fd, __u32 flags)
 687{
 688        struct sockaddr_nl sa;
 689        int sock, seq = 0, len, ret = -1;
 690        char buf[4096];
 691        struct nlattr *nla, *nla_xdp;
 692        struct {
 693                struct nlmsghdr  nh;
 694                struct ifinfomsg ifinfo;
 695                char             attrbuf[64];
 696        } req;
 697        struct nlmsghdr *nh;
 698        struct nlmsgerr *err;
 699
 700        memset(&sa, 0, sizeof(sa));
 701        sa.nl_family = AF_NETLINK;
 702
 703        sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
 704        if (sock < 0) {
 705                printf("open netlink socket: %s\n", strerror(errno));
 706                return -1;
 707        }
 708
 709        if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
 710                printf("bind to netlink: %s\n", strerror(errno));
 711                goto cleanup;
 712        }
 713
 714        memset(&req, 0, sizeof(req));
 715        req.nh.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg));
 716        req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
 717        req.nh.nlmsg_type = RTM_SETLINK;
 718        req.nh.nlmsg_pid = 0;
 719        req.nh.nlmsg_seq = ++seq;
 720        req.ifinfo.ifi_family = AF_UNSPEC;
 721        req.ifinfo.ifi_index = ifindex;
 722
 723        /* started nested attribute for XDP */
 724        nla = (struct nlattr *)(((char *)&req)
 725                                + NLMSG_ALIGN(req.nh.nlmsg_len));
 726        nla->nla_type = NLA_F_NESTED | 43/*IFLA_XDP*/;
 727        nla->nla_len = NLA_HDRLEN;
 728
 729        /* add XDP fd */
 730        nla_xdp = (struct nlattr *)((char *)nla + nla->nla_len);
 731        nla_xdp->nla_type = 1/*IFLA_XDP_FD*/;
 732        nla_xdp->nla_len = NLA_HDRLEN + sizeof(int);
 733        memcpy((char *)nla_xdp + NLA_HDRLEN, &fd, sizeof(fd));
 734        nla->nla_len += nla_xdp->nla_len;
 735
 736        /* if user passed in any flags, add those too */
 737        if (flags) {
 738                nla_xdp = (struct nlattr *)((char *)nla + nla->nla_len);
 739                nla_xdp->nla_type = 3/*IFLA_XDP_FLAGS*/;
 740                nla_xdp->nla_len = NLA_HDRLEN + sizeof(flags);
 741                memcpy((char *)nla_xdp + NLA_HDRLEN, &flags, sizeof(flags));
 742                nla->nla_len += nla_xdp->nla_len;
 743        }
 744
 745        req.nh.nlmsg_len += NLA_ALIGN(nla->nla_len);
 746
 747        if (send(sock, &req, req.nh.nlmsg_len, 0) < 0) {
 748                printf("send to netlink: %s\n", strerror(errno));
 749                goto cleanup;
 750        }
 751
 752        len = recv(sock, buf, sizeof(buf), 0);
 753        if (len < 0) {
 754                printf("recv from netlink: %s\n", strerror(errno));
 755                goto cleanup;
 756        }
 757
 758        for (nh = (struct nlmsghdr *)buf; NLMSG_OK(nh, len);
 759             nh = NLMSG_NEXT(nh, len)) {
 760                if (nh->nlmsg_pid != getpid()) {
 761                        printf("Wrong pid %d, expected %d\n",
 762                               nh->nlmsg_pid, getpid());
 763                        goto cleanup;
 764                }
 765                if (nh->nlmsg_seq != seq) {
 766                        printf("Wrong seq %d, expected %d\n",
 767                               nh->nlmsg_seq, seq);
 768                        goto cleanup;
 769                }
 770                switch (nh->nlmsg_type) {
 771                case NLMSG_ERROR:
 772                        err = (struct nlmsgerr *)NLMSG_DATA(nh);
 773                        if (!err->error)
 774                                continue;
 775                        printf("nlmsg error %s\n", strerror(-err->error));
 776                        goto cleanup;
 777                case NLMSG_DONE:
 778                        break;
 779                }
 780        }
 781
 782        ret = 0;
 783
 784cleanup:
 785        close(sock);
 786        return ret;
 787}
 788