linux/tools/lib/bpf/btf.c
<<
>>
Prefs
   1// SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
   2/* Copyright (c) 2018 Facebook */
   3
   4#include <endian.h>
   5#include <stdio.h>
   6#include <stdlib.h>
   7#include <string.h>
   8#include <fcntl.h>
   9#include <unistd.h>
  10#include <errno.h>
  11#include <linux/err.h>
  12#include <linux/btf.h>
  13#include <gelf.h>
  14#include "btf.h"
  15#include "bpf.h"
  16#include "libbpf.h"
  17#include "libbpf_internal.h"
  18#include "hashmap.h"
  19
  20#define BTF_MAX_NR_TYPES 0x7fffffff
  21#define BTF_MAX_STR_OFFSET 0x7fffffff
  22
  23static struct btf_type btf_void;
  24
  25struct btf {
  26        union {
  27                struct btf_header *hdr;
  28                void *data;
  29        };
  30        struct btf_type **types;
  31        const char *strings;
  32        void *nohdr_data;
  33        __u32 nr_types;
  34        __u32 types_size;
  35        __u32 data_size;
  36        int fd;
  37};
  38
  39static inline __u64 ptr_to_u64(const void *ptr)
  40{
  41        return (__u64) (unsigned long) ptr;
  42}
  43
  44static int btf_add_type(struct btf *btf, struct btf_type *t)
  45{
  46        if (btf->types_size - btf->nr_types < 2) {
  47                struct btf_type **new_types;
  48                __u32 expand_by, new_size;
  49
  50                if (btf->types_size == BTF_MAX_NR_TYPES)
  51                        return -E2BIG;
  52
  53                expand_by = max(btf->types_size >> 2, 16);
  54                new_size = min(BTF_MAX_NR_TYPES, btf->types_size + expand_by);
  55
  56                new_types = realloc(btf->types, sizeof(*new_types) * new_size);
  57                if (!new_types)
  58                        return -ENOMEM;
  59
  60                if (btf->nr_types == 0)
  61                        new_types[0] = &btf_void;
  62
  63                btf->types = new_types;
  64                btf->types_size = new_size;
  65        }
  66
  67        btf->types[++(btf->nr_types)] = t;
  68
  69        return 0;
  70}
  71
  72static int btf_parse_hdr(struct btf *btf)
  73{
  74        const struct btf_header *hdr = btf->hdr;
  75        __u32 meta_left;
  76
  77        if (btf->data_size < sizeof(struct btf_header)) {
  78                pr_debug("BTF header not found\n");
  79                return -EINVAL;
  80        }
  81
  82        if (hdr->magic != BTF_MAGIC) {
  83                pr_debug("Invalid BTF magic:%x\n", hdr->magic);
  84                return -EINVAL;
  85        }
  86
  87        if (hdr->version != BTF_VERSION) {
  88                pr_debug("Unsupported BTF version:%u\n", hdr->version);
  89                return -ENOTSUP;
  90        }
  91
  92        if (hdr->flags) {
  93                pr_debug("Unsupported BTF flags:%x\n", hdr->flags);
  94                return -ENOTSUP;
  95        }
  96
  97        meta_left = btf->data_size - sizeof(*hdr);
  98        if (!meta_left) {
  99                pr_debug("BTF has no data\n");
 100                return -EINVAL;
 101        }
 102
 103        if (meta_left < hdr->type_off) {
 104                pr_debug("Invalid BTF type section offset:%u\n", hdr->type_off);
 105                return -EINVAL;
 106        }
 107
 108        if (meta_left < hdr->str_off) {
 109                pr_debug("Invalid BTF string section offset:%u\n", hdr->str_off);
 110                return -EINVAL;
 111        }
 112
 113        if (hdr->type_off >= hdr->str_off) {
 114                pr_debug("BTF type section offset >= string section offset. No type?\n");
 115                return -EINVAL;
 116        }
 117
 118        if (hdr->type_off & 0x02) {
 119                pr_debug("BTF type section is not aligned to 4 bytes\n");
 120                return -EINVAL;
 121        }
 122
 123        btf->nohdr_data = btf->hdr + 1;
 124
 125        return 0;
 126}
 127
 128static int btf_parse_str_sec(struct btf *btf)
 129{
 130        const struct btf_header *hdr = btf->hdr;
 131        const char *start = btf->nohdr_data + hdr->str_off;
 132        const char *end = start + btf->hdr->str_len;
 133
 134        if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET ||
 135            start[0] || end[-1]) {
 136                pr_debug("Invalid BTF string section\n");
 137                return -EINVAL;
 138        }
 139
 140        btf->strings = start;
 141
 142        return 0;
 143}
 144
 145static int btf_type_size(struct btf_type *t)
 146{
 147        int base_size = sizeof(struct btf_type);
 148        __u16 vlen = btf_vlen(t);
 149
 150        switch (btf_kind(t)) {
 151        case BTF_KIND_FWD:
 152        case BTF_KIND_CONST:
 153        case BTF_KIND_VOLATILE:
 154        case BTF_KIND_RESTRICT:
 155        case BTF_KIND_PTR:
 156        case BTF_KIND_TYPEDEF:
 157        case BTF_KIND_FUNC:
 158                return base_size;
 159        case BTF_KIND_INT:
 160                return base_size + sizeof(__u32);
 161        case BTF_KIND_ENUM:
 162                return base_size + vlen * sizeof(struct btf_enum);
 163        case BTF_KIND_ARRAY:
 164                return base_size + sizeof(struct btf_array);
 165        case BTF_KIND_STRUCT:
 166        case BTF_KIND_UNION:
 167                return base_size + vlen * sizeof(struct btf_member);
 168        case BTF_KIND_FUNC_PROTO:
 169                return base_size + vlen * sizeof(struct btf_param);
 170        case BTF_KIND_VAR:
 171                return base_size + sizeof(struct btf_var);
 172        case BTF_KIND_DATASEC:
 173                return base_size + vlen * sizeof(struct btf_var_secinfo);
 174        default:
 175                pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
 176                return -EINVAL;
 177        }
 178}
 179
 180static int btf_parse_type_sec(struct btf *btf)
 181{
 182        struct btf_header *hdr = btf->hdr;
 183        void *nohdr_data = btf->nohdr_data;
 184        void *next_type = nohdr_data + hdr->type_off;
 185        void *end_type = nohdr_data + hdr->str_off;
 186
 187        while (next_type < end_type) {
 188                struct btf_type *t = next_type;
 189                int type_size;
 190                int err;
 191
 192                type_size = btf_type_size(t);
 193                if (type_size < 0)
 194                        return type_size;
 195                next_type += type_size;
 196                err = btf_add_type(btf, t);
 197                if (err)
 198                        return err;
 199        }
 200
 201        return 0;
 202}
 203
 204__u32 btf__get_nr_types(const struct btf *btf)
 205{
 206        return btf->nr_types;
 207}
 208
 209const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
 210{
 211        if (type_id > btf->nr_types)
 212                return NULL;
 213
 214        return btf->types[type_id];
 215}
 216
 217static bool btf_type_is_void(const struct btf_type *t)
 218{
 219        return t == &btf_void || btf_is_fwd(t);
 220}
 221
 222static bool btf_type_is_void_or_null(const struct btf_type *t)
 223{
 224        return !t || btf_type_is_void(t);
 225}
 226
 227#define MAX_RESOLVE_DEPTH 32
 228
 229__s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
 230{
 231        const struct btf_array *array;
 232        const struct btf_type *t;
 233        __u32 nelems = 1;
 234        __s64 size = -1;
 235        int i;
 236
 237        t = btf__type_by_id(btf, type_id);
 238        for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t);
 239             i++) {
 240                switch (btf_kind(t)) {
 241                case BTF_KIND_INT:
 242                case BTF_KIND_STRUCT:
 243                case BTF_KIND_UNION:
 244                case BTF_KIND_ENUM:
 245                case BTF_KIND_DATASEC:
 246                        size = t->size;
 247                        goto done;
 248                case BTF_KIND_PTR:
 249                        size = sizeof(void *);
 250                        goto done;
 251                case BTF_KIND_TYPEDEF:
 252                case BTF_KIND_VOLATILE:
 253                case BTF_KIND_CONST:
 254                case BTF_KIND_RESTRICT:
 255                case BTF_KIND_VAR:
 256                        type_id = t->type;
 257                        break;
 258                case BTF_KIND_ARRAY:
 259                        array = btf_array(t);
 260                        if (nelems && array->nelems > UINT32_MAX / nelems)
 261                                return -E2BIG;
 262                        nelems *= array->nelems;
 263                        type_id = array->type;
 264                        break;
 265                default:
 266                        return -EINVAL;
 267                }
 268
 269                t = btf__type_by_id(btf, type_id);
 270        }
 271
 272        if (size < 0)
 273                return -EINVAL;
 274
 275done:
 276        if (nelems && size > UINT32_MAX / nelems)
 277                return -E2BIG;
 278
 279        return nelems * size;
 280}
 281
 282int btf__resolve_type(const struct btf *btf, __u32 type_id)
 283{
 284        const struct btf_type *t;
 285        int depth = 0;
 286
 287        t = btf__type_by_id(btf, type_id);
 288        while (depth < MAX_RESOLVE_DEPTH &&
 289               !btf_type_is_void_or_null(t) &&
 290               (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
 291                type_id = t->type;
 292                t = btf__type_by_id(btf, type_id);
 293                depth++;
 294        }
 295
 296        if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
 297                return -EINVAL;
 298
 299        return type_id;
 300}
 301
 302__s32 btf__find_by_name(const struct btf *btf, const char *type_name)
 303{
 304        __u32 i;
 305
 306        if (!strcmp(type_name, "void"))
 307                return 0;
 308
 309        for (i = 1; i <= btf->nr_types; i++) {
 310                const struct btf_type *t = btf->types[i];
 311                const char *name = btf__name_by_offset(btf, t->name_off);
 312
 313                if (name && !strcmp(type_name, name))
 314                        return i;
 315        }
 316
 317        return -ENOENT;
 318}
 319
 320void btf__free(struct btf *btf)
 321{
 322        if (!btf)
 323                return;
 324
 325        if (btf->fd != -1)
 326                close(btf->fd);
 327
 328        free(btf->data);
 329        free(btf->types);
 330        free(btf);
 331}
 332
 333struct btf *btf__new(__u8 *data, __u32 size)
 334{
 335        struct btf *btf;
 336        int err;
 337
 338        btf = calloc(1, sizeof(struct btf));
 339        if (!btf)
 340                return ERR_PTR(-ENOMEM);
 341
 342        btf->fd = -1;
 343
 344        btf->data = malloc(size);
 345        if (!btf->data) {
 346                err = -ENOMEM;
 347                goto done;
 348        }
 349
 350        memcpy(btf->data, data, size);
 351        btf->data_size = size;
 352
 353        err = btf_parse_hdr(btf);
 354        if (err)
 355                goto done;
 356
 357        err = btf_parse_str_sec(btf);
 358        if (err)
 359                goto done;
 360
 361        err = btf_parse_type_sec(btf);
 362
 363done:
 364        if (err) {
 365                btf__free(btf);
 366                return ERR_PTR(err);
 367        }
 368
 369        return btf;
 370}
 371
 372static bool btf_check_endianness(const GElf_Ehdr *ehdr)
 373{
 374#if __BYTE_ORDER == __LITTLE_ENDIAN
 375        return ehdr->e_ident[EI_DATA] == ELFDATA2LSB;
 376#elif __BYTE_ORDER == __BIG_ENDIAN
 377        return ehdr->e_ident[EI_DATA] == ELFDATA2MSB;
 378#else
 379# error "Unrecognized __BYTE_ORDER__"
 380#endif
 381}
 382
 383struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
 384{
 385        Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
 386        int err = 0, fd = -1, idx = 0;
 387        struct btf *btf = NULL;
 388        Elf_Scn *scn = NULL;
 389        Elf *elf = NULL;
 390        GElf_Ehdr ehdr;
 391
 392        if (elf_version(EV_CURRENT) == EV_NONE) {
 393                pr_warning("failed to init libelf for %s\n", path);
 394                return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
 395        }
 396
 397        fd = open(path, O_RDONLY);
 398        if (fd < 0) {
 399                err = -errno;
 400                pr_warning("failed to open %s: %s\n", path, strerror(errno));
 401                return ERR_PTR(err);
 402        }
 403
 404        err = -LIBBPF_ERRNO__FORMAT;
 405
 406        elf = elf_begin(fd, ELF_C_READ, NULL);
 407        if (!elf) {
 408                pr_warning("failed to open %s as ELF file\n", path);
 409                goto done;
 410        }
 411        if (!gelf_getehdr(elf, &ehdr)) {
 412                pr_warning("failed to get EHDR from %s\n", path);
 413                goto done;
 414        }
 415        if (!btf_check_endianness(&ehdr)) {
 416                pr_warning("non-native ELF endianness is not supported\n");
 417                goto done;
 418        }
 419        if (!elf_rawdata(elf_getscn(elf, ehdr.e_shstrndx), NULL)) {
 420                pr_warning("failed to get e_shstrndx from %s\n", path);
 421                goto done;
 422        }
 423
 424        while ((scn = elf_nextscn(elf, scn)) != NULL) {
 425                GElf_Shdr sh;
 426                char *name;
 427
 428                idx++;
 429                if (gelf_getshdr(scn, &sh) != &sh) {
 430                        pr_warning("failed to get section(%d) header from %s\n",
 431                                   idx, path);
 432                        goto done;
 433                }
 434                name = elf_strptr(elf, ehdr.e_shstrndx, sh.sh_name);
 435                if (!name) {
 436                        pr_warning("failed to get section(%d) name from %s\n",
 437                                   idx, path);
 438                        goto done;
 439                }
 440                if (strcmp(name, BTF_ELF_SEC) == 0) {
 441                        btf_data = elf_getdata(scn, 0);
 442                        if (!btf_data) {
 443                                pr_warning("failed to get section(%d, %s) data from %s\n",
 444                                           idx, name, path);
 445                                goto done;
 446                        }
 447                        continue;
 448                } else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
 449                        btf_ext_data = elf_getdata(scn, 0);
 450                        if (!btf_ext_data) {
 451                                pr_warning("failed to get section(%d, %s) data from %s\n",
 452                                           idx, name, path);
 453                                goto done;
 454                        }
 455                        continue;
 456                }
 457        }
 458
 459        err = 0;
 460
 461        if (!btf_data) {
 462                err = -ENOENT;
 463                goto done;
 464        }
 465        btf = btf__new(btf_data->d_buf, btf_data->d_size);
 466        if (IS_ERR(btf))
 467                goto done;
 468
 469        if (btf_ext && btf_ext_data) {
 470                *btf_ext = btf_ext__new(btf_ext_data->d_buf,
 471                                        btf_ext_data->d_size);
 472                if (IS_ERR(*btf_ext))
 473                        goto done;
 474        } else if (btf_ext) {
 475                *btf_ext = NULL;
 476        }
 477done:
 478        if (elf)
 479                elf_end(elf);
 480        close(fd);
 481
 482        if (err)
 483                return ERR_PTR(err);
 484        /*
 485         * btf is always parsed before btf_ext, so no need to clean up
 486         * btf_ext, if btf loading failed
 487         */
 488        if (IS_ERR(btf))
 489                return btf;
 490        if (btf_ext && IS_ERR(*btf_ext)) {
 491                btf__free(btf);
 492                err = PTR_ERR(*btf_ext);
 493                return ERR_PTR(err);
 494        }
 495        return btf;
 496}
 497
 498static int compare_vsi_off(const void *_a, const void *_b)
 499{
 500        const struct btf_var_secinfo *a = _a;
 501        const struct btf_var_secinfo *b = _b;
 502
 503        return a->offset - b->offset;
 504}
 505
 506static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
 507                             struct btf_type *t)
 508{
 509        __u32 size = 0, off = 0, i, vars = btf_vlen(t);
 510        const char *name = btf__name_by_offset(btf, t->name_off);
 511        const struct btf_type *t_var;
 512        struct btf_var_secinfo *vsi;
 513        const struct btf_var *var;
 514        int ret;
 515
 516        if (!name) {
 517                pr_debug("No name found in string section for DATASEC kind.\n");
 518                return -ENOENT;
 519        }
 520
 521        ret = bpf_object__section_size(obj, name, &size);
 522        if (ret || !size || (t->size && t->size != size)) {
 523                pr_debug("Invalid size for section %s: %u bytes\n", name, size);
 524                return -ENOENT;
 525        }
 526
 527        t->size = size;
 528
 529        for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
 530                t_var = btf__type_by_id(btf, vsi->type);
 531                var = btf_var(t_var);
 532
 533                if (!btf_is_var(t_var)) {
 534                        pr_debug("Non-VAR type seen in section %s\n", name);
 535                        return -EINVAL;
 536                }
 537
 538                if (var->linkage == BTF_VAR_STATIC)
 539                        continue;
 540
 541                name = btf__name_by_offset(btf, t_var->name_off);
 542                if (!name) {
 543                        pr_debug("No name found in string section for VAR kind\n");
 544                        return -ENOENT;
 545                }
 546
 547                ret = bpf_object__variable_offset(obj, name, &off);
 548                if (ret) {
 549                        pr_debug("No offset found in symbol table for VAR %s\n",
 550                                 name);
 551                        return -ENOENT;
 552                }
 553
 554                vsi->offset = off;
 555        }
 556
 557        qsort(t + 1, vars, sizeof(*vsi), compare_vsi_off);
 558        return 0;
 559}
 560
 561int btf__finalize_data(struct bpf_object *obj, struct btf *btf)
 562{
 563        int err = 0;
 564        __u32 i;
 565
 566        for (i = 1; i <= btf->nr_types; i++) {
 567                struct btf_type *t = btf->types[i];
 568
 569                /* Loader needs to fix up some of the things compiler
 570                 * couldn't get its hands on while emitting BTF. This
 571                 * is section size and global variable offset. We use
 572                 * the info from the ELF itself for this purpose.
 573                 */
 574                if (btf_is_datasec(t)) {
 575                        err = btf_fixup_datasec(obj, btf, t);
 576                        if (err)
 577                                break;
 578                }
 579        }
 580
 581        return err;
 582}
 583
 584int btf__load(struct btf *btf)
 585{
 586        __u32 log_buf_size = BPF_LOG_BUF_SIZE;
 587        char *log_buf = NULL;
 588        int err = 0;
 589
 590        if (btf->fd >= 0)
 591                return -EEXIST;
 592
 593        log_buf = malloc(log_buf_size);
 594        if (!log_buf)
 595                return -ENOMEM;
 596
 597        *log_buf = 0;
 598
 599        btf->fd = bpf_load_btf(btf->data, btf->data_size,
 600                               log_buf, log_buf_size, false);
 601        if (btf->fd < 0) {
 602                err = -errno;
 603                pr_warning("Error loading BTF: %s(%d)\n", strerror(errno), errno);
 604                if (*log_buf)
 605                        pr_warning("%s\n", log_buf);
 606                goto done;
 607        }
 608
 609done:
 610        free(log_buf);
 611        return err;
 612}
 613
 614int btf__fd(const struct btf *btf)
 615{
 616        return btf->fd;
 617}
 618
 619const void *btf__get_raw_data(const struct btf *btf, __u32 *size)
 620{
 621        *size = btf->data_size;
 622        return btf->data;
 623}
 624
 625const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
 626{
 627        if (offset < btf->hdr->str_len)
 628                return &btf->strings[offset];
 629        else
 630                return NULL;
 631}
 632
 633int btf__get_from_id(__u32 id, struct btf **btf)
 634{
 635        struct bpf_btf_info btf_info = { 0 };
 636        __u32 len = sizeof(btf_info);
 637        __u32 last_size;
 638        int btf_fd;
 639        void *ptr;
 640        int err;
 641
 642        err = 0;
 643        *btf = NULL;
 644        btf_fd = bpf_btf_get_fd_by_id(id);
 645        if (btf_fd < 0)
 646                return 0;
 647
 648        /* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
 649         * let's start with a sane default - 4KiB here - and resize it only if
 650         * bpf_obj_get_info_by_fd() needs a bigger buffer.
 651         */
 652        btf_info.btf_size = 4096;
 653        last_size = btf_info.btf_size;
 654        ptr = malloc(last_size);
 655        if (!ptr) {
 656                err = -ENOMEM;
 657                goto exit_free;
 658        }
 659
 660        memset(ptr, 0, last_size);
 661        btf_info.btf = ptr_to_u64(ptr);
 662        err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
 663
 664        if (!err && btf_info.btf_size > last_size) {
 665                void *temp_ptr;
 666
 667                last_size = btf_info.btf_size;
 668                temp_ptr = realloc(ptr, last_size);
 669                if (!temp_ptr) {
 670                        err = -ENOMEM;
 671                        goto exit_free;
 672                }
 673                ptr = temp_ptr;
 674                memset(ptr, 0, last_size);
 675                btf_info.btf = ptr_to_u64(ptr);
 676                err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
 677        }
 678
 679        if (err || btf_info.btf_size > last_size) {
 680                err = errno;
 681                goto exit_free;
 682        }
 683
 684        *btf = btf__new((__u8 *)(long)btf_info.btf, btf_info.btf_size);
 685        if (IS_ERR(*btf)) {
 686                err = PTR_ERR(*btf);
 687                *btf = NULL;
 688        }
 689
 690exit_free:
 691        close(btf_fd);
 692        free(ptr);
 693
 694        return err;
 695}
 696
 697int btf__get_map_kv_tids(const struct btf *btf, const char *map_name,
 698                         __u32 expected_key_size, __u32 expected_value_size,
 699                         __u32 *key_type_id, __u32 *value_type_id)
 700{
 701        const struct btf_type *container_type;
 702        const struct btf_member *key, *value;
 703        const size_t max_name = 256;
 704        char container_name[max_name];
 705        __s64 key_size, value_size;
 706        __s32 container_id;
 707
 708        if (snprintf(container_name, max_name, "____btf_map_%s", map_name) ==
 709            max_name) {
 710                pr_warning("map:%s length of '____btf_map_%s' is too long\n",
 711                           map_name, map_name);
 712                return -EINVAL;
 713        }
 714
 715        container_id = btf__find_by_name(btf, container_name);
 716        if (container_id < 0) {
 717                pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
 718                         map_name, container_name);
 719                return container_id;
 720        }
 721
 722        container_type = btf__type_by_id(btf, container_id);
 723        if (!container_type) {
 724                pr_warning("map:%s cannot find BTF type for container_id:%u\n",
 725                           map_name, container_id);
 726                return -EINVAL;
 727        }
 728
 729        if (!btf_is_struct(container_type) || btf_vlen(container_type) < 2) {
 730                pr_warning("map:%s container_name:%s is an invalid container struct\n",
 731                           map_name, container_name);
 732                return -EINVAL;
 733        }
 734
 735        key = btf_members(container_type);
 736        value = key + 1;
 737
 738        key_size = btf__resolve_size(btf, key->type);
 739        if (key_size < 0) {
 740                pr_warning("map:%s invalid BTF key_type_size\n", map_name);
 741                return key_size;
 742        }
 743
 744        if (expected_key_size != key_size) {
 745                pr_warning("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
 746                           map_name, (__u32)key_size, expected_key_size);
 747                return -EINVAL;
 748        }
 749
 750        value_size = btf__resolve_size(btf, value->type);
 751        if (value_size < 0) {
 752                pr_warning("map:%s invalid BTF value_type_size\n", map_name);
 753                return value_size;
 754        }
 755
 756        if (expected_value_size != value_size) {
 757                pr_warning("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
 758                           map_name, (__u32)value_size, expected_value_size);
 759                return -EINVAL;
 760        }
 761
 762        *key_type_id = key->type;
 763        *value_type_id = value->type;
 764
 765        return 0;
 766}
 767
 768struct btf_ext_sec_setup_param {
 769        __u32 off;
 770        __u32 len;
 771        __u32 min_rec_size;
 772        struct btf_ext_info *ext_info;
 773        const char *desc;
 774};
 775
 776static int btf_ext_setup_info(struct btf_ext *btf_ext,
 777                              struct btf_ext_sec_setup_param *ext_sec)
 778{
 779        const struct btf_ext_info_sec *sinfo;
 780        struct btf_ext_info *ext_info;
 781        __u32 info_left, record_size;
 782        /* The start of the info sec (including the __u32 record_size). */
 783        void *info;
 784
 785        if (ext_sec->len == 0)
 786                return 0;
 787
 788        if (ext_sec->off & 0x03) {
 789                pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
 790                     ext_sec->desc);
 791                return -EINVAL;
 792        }
 793
 794        info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
 795        info_left = ext_sec->len;
 796
 797        if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
 798                pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
 799                         ext_sec->desc, ext_sec->off, ext_sec->len);
 800                return -EINVAL;
 801        }
 802
 803        /* At least a record size */
 804        if (info_left < sizeof(__u32)) {
 805                pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
 806                return -EINVAL;
 807        }
 808
 809        /* The record size needs to meet the minimum standard */
 810        record_size = *(__u32 *)info;
 811        if (record_size < ext_sec->min_rec_size ||
 812            record_size & 0x03) {
 813                pr_debug("%s section in .BTF.ext has invalid record size %u\n",
 814                         ext_sec->desc, record_size);
 815                return -EINVAL;
 816        }
 817
 818        sinfo = info + sizeof(__u32);
 819        info_left -= sizeof(__u32);
 820
 821        /* If no records, return failure now so .BTF.ext won't be used. */
 822        if (!info_left) {
 823                pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
 824                return -EINVAL;
 825        }
 826
 827        while (info_left) {
 828                unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
 829                __u64 total_record_size;
 830                __u32 num_records;
 831
 832                if (info_left < sec_hdrlen) {
 833                        pr_debug("%s section header is not found in .BTF.ext\n",
 834                             ext_sec->desc);
 835                        return -EINVAL;
 836                }
 837
 838                num_records = sinfo->num_info;
 839                if (num_records == 0) {
 840                        pr_debug("%s section has incorrect num_records in .BTF.ext\n",
 841                             ext_sec->desc);
 842                        return -EINVAL;
 843                }
 844
 845                total_record_size = sec_hdrlen +
 846                                    (__u64)num_records * record_size;
 847                if (info_left < total_record_size) {
 848                        pr_debug("%s section has incorrect num_records in .BTF.ext\n",
 849                             ext_sec->desc);
 850                        return -EINVAL;
 851                }
 852
 853                info_left -= total_record_size;
 854                sinfo = (void *)sinfo + total_record_size;
 855        }
 856
 857        ext_info = ext_sec->ext_info;
 858        ext_info->len = ext_sec->len - sizeof(__u32);
 859        ext_info->rec_size = record_size;
 860        ext_info->info = info + sizeof(__u32);
 861
 862        return 0;
 863}
 864
 865static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
 866{
 867        struct btf_ext_sec_setup_param param = {
 868                .off = btf_ext->hdr->func_info_off,
 869                .len = btf_ext->hdr->func_info_len,
 870                .min_rec_size = sizeof(struct bpf_func_info_min),
 871                .ext_info = &btf_ext->func_info,
 872                .desc = "func_info"
 873        };
 874
 875        return btf_ext_setup_info(btf_ext, &param);
 876}
 877
 878static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
 879{
 880        struct btf_ext_sec_setup_param param = {
 881                .off = btf_ext->hdr->line_info_off,
 882                .len = btf_ext->hdr->line_info_len,
 883                .min_rec_size = sizeof(struct bpf_line_info_min),
 884                .ext_info = &btf_ext->line_info,
 885                .desc = "line_info",
 886        };
 887
 888        return btf_ext_setup_info(btf_ext, &param);
 889}
 890
 891static int btf_ext_setup_offset_reloc(struct btf_ext *btf_ext)
 892{
 893        struct btf_ext_sec_setup_param param = {
 894                .off = btf_ext->hdr->offset_reloc_off,
 895                .len = btf_ext->hdr->offset_reloc_len,
 896                .min_rec_size = sizeof(struct bpf_offset_reloc),
 897                .ext_info = &btf_ext->offset_reloc_info,
 898                .desc = "offset_reloc",
 899        };
 900
 901        return btf_ext_setup_info(btf_ext, &param);
 902}
 903
 904static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
 905{
 906        const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
 907
 908        if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
 909            data_size < hdr->hdr_len) {
 910                pr_debug("BTF.ext header not found");
 911                return -EINVAL;
 912        }
 913
 914        if (hdr->magic != BTF_MAGIC) {
 915                pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
 916                return -EINVAL;
 917        }
 918
 919        if (hdr->version != BTF_VERSION) {
 920                pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
 921                return -ENOTSUP;
 922        }
 923
 924        if (hdr->flags) {
 925                pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
 926                return -ENOTSUP;
 927        }
 928
 929        if (data_size == hdr->hdr_len) {
 930                pr_debug("BTF.ext has no data\n");
 931                return -EINVAL;
 932        }
 933
 934        return 0;
 935}
 936
 937void btf_ext__free(struct btf_ext *btf_ext)
 938{
 939        if (!btf_ext)
 940                return;
 941        free(btf_ext->data);
 942        free(btf_ext);
 943}
 944
 945struct btf_ext *btf_ext__new(__u8 *data, __u32 size)
 946{
 947        struct btf_ext *btf_ext;
 948        int err;
 949
 950        err = btf_ext_parse_hdr(data, size);
 951        if (err)
 952                return ERR_PTR(err);
 953
 954        btf_ext = calloc(1, sizeof(struct btf_ext));
 955        if (!btf_ext)
 956                return ERR_PTR(-ENOMEM);
 957
 958        btf_ext->data_size = size;
 959        btf_ext->data = malloc(size);
 960        if (!btf_ext->data) {
 961                err = -ENOMEM;
 962                goto done;
 963        }
 964        memcpy(btf_ext->data, data, size);
 965
 966        if (btf_ext->hdr->hdr_len <
 967            offsetofend(struct btf_ext_header, line_info_len))
 968                goto done;
 969        err = btf_ext_setup_func_info(btf_ext);
 970        if (err)
 971                goto done;
 972
 973        err = btf_ext_setup_line_info(btf_ext);
 974        if (err)
 975                goto done;
 976
 977        if (btf_ext->hdr->hdr_len <
 978            offsetofend(struct btf_ext_header, offset_reloc_len))
 979                goto done;
 980        err = btf_ext_setup_offset_reloc(btf_ext);
 981        if (err)
 982                goto done;
 983
 984done:
 985        if (err) {
 986                btf_ext__free(btf_ext);
 987                return ERR_PTR(err);
 988        }
 989
 990        return btf_ext;
 991}
 992
 993const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
 994{
 995        *size = btf_ext->data_size;
 996        return btf_ext->data;
 997}
 998
 999static int btf_ext_reloc_info(const struct btf *btf,
1000                              const struct btf_ext_info *ext_info,
1001                              const char *sec_name, __u32 insns_cnt,
1002                              void **info, __u32 *cnt)
1003{
1004        __u32 sec_hdrlen = sizeof(struct btf_ext_info_sec);
1005        __u32 i, record_size, existing_len, records_len;
1006        struct btf_ext_info_sec *sinfo;
1007        const char *info_sec_name;
1008        __u64 remain_len;
1009        void *data;
1010
1011        record_size = ext_info->rec_size;
1012        sinfo = ext_info->info;
1013        remain_len = ext_info->len;
1014        while (remain_len > 0) {
1015                records_len = sinfo->num_info * record_size;
1016                info_sec_name = btf__name_by_offset(btf, sinfo->sec_name_off);
1017                if (strcmp(info_sec_name, sec_name)) {
1018                        remain_len -= sec_hdrlen + records_len;
1019                        sinfo = (void *)sinfo + sec_hdrlen + records_len;
1020                        continue;
1021                }
1022
1023                existing_len = (*cnt) * record_size;
1024                data = realloc(*info, existing_len + records_len);
1025                if (!data)
1026                        return -ENOMEM;
1027
1028                memcpy(data + existing_len, sinfo->data, records_len);
1029                /* adjust insn_off only, the rest data will be passed
1030                 * to the kernel.
1031                 */
1032                for (i = 0; i < sinfo->num_info; i++) {
1033                        __u32 *insn_off;
1034
1035                        insn_off = data + existing_len + (i * record_size);
1036                        *insn_off = *insn_off / sizeof(struct bpf_insn) +
1037                                insns_cnt;
1038                }
1039                *info = data;
1040                *cnt += sinfo->num_info;
1041                return 0;
1042        }
1043
1044        return -ENOENT;
1045}
1046
1047int btf_ext__reloc_func_info(const struct btf *btf,
1048                             const struct btf_ext *btf_ext,
1049                             const char *sec_name, __u32 insns_cnt,
1050                             void **func_info, __u32 *cnt)
1051{
1052        return btf_ext_reloc_info(btf, &btf_ext->func_info, sec_name,
1053                                  insns_cnt, func_info, cnt);
1054}
1055
1056int btf_ext__reloc_line_info(const struct btf *btf,
1057                             const struct btf_ext *btf_ext,
1058                             const char *sec_name, __u32 insns_cnt,
1059                             void **line_info, __u32 *cnt)
1060{
1061        return btf_ext_reloc_info(btf, &btf_ext->line_info, sec_name,
1062                                  insns_cnt, line_info, cnt);
1063}
1064
1065__u32 btf_ext__func_info_rec_size(const struct btf_ext *btf_ext)
1066{
1067        return btf_ext->func_info.rec_size;
1068}
1069
1070__u32 btf_ext__line_info_rec_size(const struct btf_ext *btf_ext)
1071{
1072        return btf_ext->line_info.rec_size;
1073}
1074
1075struct btf_dedup;
1076
1077static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
1078                                       const struct btf_dedup_opts *opts);
1079static void btf_dedup_free(struct btf_dedup *d);
1080static int btf_dedup_strings(struct btf_dedup *d);
1081static int btf_dedup_prim_types(struct btf_dedup *d);
1082static int btf_dedup_struct_types(struct btf_dedup *d);
1083static int btf_dedup_ref_types(struct btf_dedup *d);
1084static int btf_dedup_compact_types(struct btf_dedup *d);
1085static int btf_dedup_remap_types(struct btf_dedup *d);
1086
1087/*
1088 * Deduplicate BTF types and strings.
1089 *
1090 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
1091 * section with all BTF type descriptors and string data. It overwrites that
1092 * memory in-place with deduplicated types and strings without any loss of
1093 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
1094 * is provided, all the strings referenced from .BTF.ext section are honored
1095 * and updated to point to the right offsets after deduplication.
1096 *
1097 * If function returns with error, type/string data might be garbled and should
1098 * be discarded.
1099 *
1100 * More verbose and detailed description of both problem btf_dedup is solving,
1101 * as well as solution could be found at:
1102 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
1103 *
1104 * Problem description and justification
1105 * =====================================
1106 *
1107 * BTF type information is typically emitted either as a result of conversion
1108 * from DWARF to BTF or directly by compiler. In both cases, each compilation
1109 * unit contains information about a subset of all the types that are used
1110 * in an application. These subsets are frequently overlapping and contain a lot
1111 * of duplicated information when later concatenated together into a single
1112 * binary. This algorithm ensures that each unique type is represented by single
1113 * BTF type descriptor, greatly reducing resulting size of BTF data.
1114 *
1115 * Compilation unit isolation and subsequent duplication of data is not the only
1116 * problem. The same type hierarchy (e.g., struct and all the type that struct
1117 * references) in different compilation units can be represented in BTF to
1118 * various degrees of completeness (or, rather, incompleteness) due to
1119 * struct/union forward declarations.
1120 *
1121 * Let's take a look at an example, that we'll use to better understand the
1122 * problem (and solution). Suppose we have two compilation units, each using
1123 * same `struct S`, but each of them having incomplete type information about
1124 * struct's fields:
1125 *
1126 * // CU #1:
1127 * struct S;
1128 * struct A {
1129 *      int a;
1130 *      struct A* self;
1131 *      struct S* parent;
1132 * };
1133 * struct B;
1134 * struct S {
1135 *      struct A* a_ptr;
1136 *      struct B* b_ptr;
1137 * };
1138 *
1139 * // CU #2:
1140 * struct S;
1141 * struct A;
1142 * struct B {
1143 *      int b;
1144 *      struct B* self;
1145 *      struct S* parent;
1146 * };
1147 * struct S {
1148 *      struct A* a_ptr;
1149 *      struct B* b_ptr;
1150 * };
1151 *
1152 * In case of CU #1, BTF data will know only that `struct B` exist (but no
1153 * more), but will know the complete type information about `struct A`. While
1154 * for CU #2, it will know full type information about `struct B`, but will
1155 * only know about forward declaration of `struct A` (in BTF terms, it will
1156 * have `BTF_KIND_FWD` type descriptor with name `B`).
1157 *
1158 * This compilation unit isolation means that it's possible that there is no
1159 * single CU with complete type information describing structs `S`, `A`, and
1160 * `B`. Also, we might get tons of duplicated and redundant type information.
1161 *
1162 * Additional complication we need to keep in mind comes from the fact that
1163 * types, in general, can form graphs containing cycles, not just DAGs.
1164 *
1165 * While algorithm does deduplication, it also merges and resolves type
1166 * information (unless disabled throught `struct btf_opts`), whenever possible.
1167 * E.g., in the example above with two compilation units having partial type
1168 * information for structs `A` and `B`, the output of algorithm will emit
1169 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
1170 * (as well as type information for `int` and pointers), as if they were defined
1171 * in a single compilation unit as:
1172 *
1173 * struct A {
1174 *      int a;
1175 *      struct A* self;
1176 *      struct S* parent;
1177 * };
1178 * struct B {
1179 *      int b;
1180 *      struct B* self;
1181 *      struct S* parent;
1182 * };
1183 * struct S {
1184 *      struct A* a_ptr;
1185 *      struct B* b_ptr;
1186 * };
1187 *
1188 * Algorithm summary
1189 * =================
1190 *
1191 * Algorithm completes its work in 6 separate passes:
1192 *
1193 * 1. Strings deduplication.
1194 * 2. Primitive types deduplication (int, enum, fwd).
1195 * 3. Struct/union types deduplication.
1196 * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
1197 *    protos, and const/volatile/restrict modifiers).
1198 * 5. Types compaction.
1199 * 6. Types remapping.
1200 *
1201 * Algorithm determines canonical type descriptor, which is a single
1202 * representative type for each truly unique type. This canonical type is the
1203 * one that will go into final deduplicated BTF type information. For
1204 * struct/unions, it is also the type that algorithm will merge additional type
1205 * information into (while resolving FWDs), as it discovers it from data in
1206 * other CUs. Each input BTF type eventually gets either mapped to itself, if
1207 * that type is canonical, or to some other type, if that type is equivalent
1208 * and was chosen as canonical representative. This mapping is stored in
1209 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
1210 * FWD type got resolved to.
1211 *
1212 * To facilitate fast discovery of canonical types, we also maintain canonical
1213 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
1214 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
1215 * that match that signature. With sufficiently good choice of type signature
1216 * hashing function, we can limit number of canonical types for each unique type
1217 * signature to a very small number, allowing to find canonical type for any
1218 * duplicated type very quickly.
1219 *
1220 * Struct/union deduplication is the most critical part and algorithm for
1221 * deduplicating structs/unions is described in greater details in comments for
1222 * `btf_dedup_is_equiv` function.
1223 */
1224int btf__dedup(struct btf *btf, struct btf_ext *btf_ext,
1225               const struct btf_dedup_opts *opts)
1226{
1227        struct btf_dedup *d = btf_dedup_new(btf, btf_ext, opts);
1228        int err;
1229
1230        if (IS_ERR(d)) {
1231                pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
1232                return -EINVAL;
1233        }
1234
1235        err = btf_dedup_strings(d);
1236        if (err < 0) {
1237                pr_debug("btf_dedup_strings failed:%d\n", err);
1238                goto done;
1239        }
1240        err = btf_dedup_prim_types(d);
1241        if (err < 0) {
1242                pr_debug("btf_dedup_prim_types failed:%d\n", err);
1243                goto done;
1244        }
1245        err = btf_dedup_struct_types(d);
1246        if (err < 0) {
1247                pr_debug("btf_dedup_struct_types failed:%d\n", err);
1248                goto done;
1249        }
1250        err = btf_dedup_ref_types(d);
1251        if (err < 0) {
1252                pr_debug("btf_dedup_ref_types failed:%d\n", err);
1253                goto done;
1254        }
1255        err = btf_dedup_compact_types(d);
1256        if (err < 0) {
1257                pr_debug("btf_dedup_compact_types failed:%d\n", err);
1258                goto done;
1259        }
1260        err = btf_dedup_remap_types(d);
1261        if (err < 0) {
1262                pr_debug("btf_dedup_remap_types failed:%d\n", err);
1263                goto done;
1264        }
1265
1266done:
1267        btf_dedup_free(d);
1268        return err;
1269}
1270
1271#define BTF_UNPROCESSED_ID ((__u32)-1)
1272#define BTF_IN_PROGRESS_ID ((__u32)-2)
1273
1274struct btf_dedup {
1275        /* .BTF section to be deduped in-place */
1276        struct btf *btf;
1277        /*
1278         * Optional .BTF.ext section. When provided, any strings referenced
1279         * from it will be taken into account when deduping strings
1280         */
1281        struct btf_ext *btf_ext;
1282        /*
1283         * This is a map from any type's signature hash to a list of possible
1284         * canonical representative type candidates. Hash collisions are
1285         * ignored, so even types of various kinds can share same list of
1286         * candidates, which is fine because we rely on subsequent
1287         * btf_xxx_equal() checks to authoritatively verify type equality.
1288         */
1289        struct hashmap *dedup_table;
1290        /* Canonical types map */
1291        __u32 *map;
1292        /* Hypothetical mapping, used during type graph equivalence checks */
1293        __u32 *hypot_map;
1294        __u32 *hypot_list;
1295        size_t hypot_cnt;
1296        size_t hypot_cap;
1297        /* Various option modifying behavior of algorithm */
1298        struct btf_dedup_opts opts;
1299};
1300
1301struct btf_str_ptr {
1302        const char *str;
1303        __u32 new_off;
1304        bool used;
1305};
1306
1307struct btf_str_ptrs {
1308        struct btf_str_ptr *ptrs;
1309        const char *data;
1310        __u32 cnt;
1311        __u32 cap;
1312};
1313
1314static long hash_combine(long h, long value)
1315{
1316        return h * 31 + value;
1317}
1318
1319#define for_each_dedup_cand(d, node, hash) \
1320        hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
1321
1322static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
1323{
1324        return hashmap__append(d->dedup_table,
1325                               (void *)hash, (void *)(long)type_id);
1326}
1327
1328static int btf_dedup_hypot_map_add(struct btf_dedup *d,
1329                                   __u32 from_id, __u32 to_id)
1330{
1331        if (d->hypot_cnt == d->hypot_cap) {
1332                __u32 *new_list;
1333
1334                d->hypot_cap += max(16, d->hypot_cap / 2);
1335                new_list = realloc(d->hypot_list, sizeof(__u32) * d->hypot_cap);
1336                if (!new_list)
1337                        return -ENOMEM;
1338                d->hypot_list = new_list;
1339        }
1340        d->hypot_list[d->hypot_cnt++] = from_id;
1341        d->hypot_map[from_id] = to_id;
1342        return 0;
1343}
1344
1345static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
1346{
1347        int i;
1348
1349        for (i = 0; i < d->hypot_cnt; i++)
1350                d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
1351        d->hypot_cnt = 0;
1352}
1353
1354static void btf_dedup_free(struct btf_dedup *d)
1355{
1356        hashmap__free(d->dedup_table);
1357        d->dedup_table = NULL;
1358
1359        free(d->map);
1360        d->map = NULL;
1361
1362        free(d->hypot_map);
1363        d->hypot_map = NULL;
1364
1365        free(d->hypot_list);
1366        d->hypot_list = NULL;
1367
1368        free(d);
1369}
1370
1371static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
1372{
1373        return (size_t)key;
1374}
1375
1376static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
1377{
1378        return 0;
1379}
1380
1381static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
1382{
1383        return k1 == k2;
1384}
1385
1386static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
1387                                       const struct btf_dedup_opts *opts)
1388{
1389        struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
1390        hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
1391        int i, err = 0;
1392
1393        if (!d)
1394                return ERR_PTR(-ENOMEM);
1395
1396        d->opts.dont_resolve_fwds = opts && opts->dont_resolve_fwds;
1397        /* dedup_table_size is now used only to force collisions in tests */
1398        if (opts && opts->dedup_table_size == 1)
1399                hash_fn = btf_dedup_collision_hash_fn;
1400
1401        d->btf = btf;
1402        d->btf_ext = btf_ext;
1403
1404        d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
1405        if (IS_ERR(d->dedup_table)) {
1406                err = PTR_ERR(d->dedup_table);
1407                d->dedup_table = NULL;
1408                goto done;
1409        }
1410
1411        d->map = malloc(sizeof(__u32) * (1 + btf->nr_types));
1412        if (!d->map) {
1413                err = -ENOMEM;
1414                goto done;
1415        }
1416        /* special BTF "void" type is made canonical immediately */
1417        d->map[0] = 0;
1418        for (i = 1; i <= btf->nr_types; i++) {
1419                struct btf_type *t = d->btf->types[i];
1420
1421                /* VAR and DATASEC are never deduped and are self-canonical */
1422                if (btf_is_var(t) || btf_is_datasec(t))
1423                        d->map[i] = i;
1424                else
1425                        d->map[i] = BTF_UNPROCESSED_ID;
1426        }
1427
1428        d->hypot_map = malloc(sizeof(__u32) * (1 + btf->nr_types));
1429        if (!d->hypot_map) {
1430                err = -ENOMEM;
1431                goto done;
1432        }
1433        for (i = 0; i <= btf->nr_types; i++)
1434                d->hypot_map[i] = BTF_UNPROCESSED_ID;
1435
1436done:
1437        if (err) {
1438                btf_dedup_free(d);
1439                return ERR_PTR(err);
1440        }
1441
1442        return d;
1443}
1444
1445typedef int (*str_off_fn_t)(__u32 *str_off_ptr, void *ctx);
1446
1447/*
1448 * Iterate over all possible places in .BTF and .BTF.ext that can reference
1449 * string and pass pointer to it to a provided callback `fn`.
1450 */
1451static int btf_for_each_str_off(struct btf_dedup *d, str_off_fn_t fn, void *ctx)
1452{
1453        void *line_data_cur, *line_data_end;
1454        int i, j, r, rec_size;
1455        struct btf_type *t;
1456
1457        for (i = 1; i <= d->btf->nr_types; i++) {
1458                t = d->btf->types[i];
1459                r = fn(&t->name_off, ctx);
1460                if (r)
1461                        return r;
1462
1463                switch (btf_kind(t)) {
1464                case BTF_KIND_STRUCT:
1465                case BTF_KIND_UNION: {
1466                        struct btf_member *m = btf_members(t);
1467                        __u16 vlen = btf_vlen(t);
1468
1469                        for (j = 0; j < vlen; j++) {
1470                                r = fn(&m->name_off, ctx);
1471                                if (r)
1472                                        return r;
1473                                m++;
1474                        }
1475                        break;
1476                }
1477                case BTF_KIND_ENUM: {
1478                        struct btf_enum *m = btf_enum(t);
1479                        __u16 vlen = btf_vlen(t);
1480
1481                        for (j = 0; j < vlen; j++) {
1482                                r = fn(&m->name_off, ctx);
1483                                if (r)
1484                                        return r;
1485                                m++;
1486                        }
1487                        break;
1488                }
1489                case BTF_KIND_FUNC_PROTO: {
1490                        struct btf_param *m = btf_params(t);
1491                        __u16 vlen = btf_vlen(t);
1492
1493                        for (j = 0; j < vlen; j++) {
1494                                r = fn(&m->name_off, ctx);
1495                                if (r)
1496                                        return r;
1497                                m++;
1498                        }
1499                        break;
1500                }
1501                default:
1502                        break;
1503                }
1504        }
1505
1506        if (!d->btf_ext)
1507                return 0;
1508
1509        line_data_cur = d->btf_ext->line_info.info;
1510        line_data_end = d->btf_ext->line_info.info + d->btf_ext->line_info.len;
1511        rec_size = d->btf_ext->line_info.rec_size;
1512
1513        while (line_data_cur < line_data_end) {
1514                struct btf_ext_info_sec *sec = line_data_cur;
1515                struct bpf_line_info_min *line_info;
1516                __u32 num_info = sec->num_info;
1517
1518                r = fn(&sec->sec_name_off, ctx);
1519                if (r)
1520                        return r;
1521
1522                line_data_cur += sizeof(struct btf_ext_info_sec);
1523                for (i = 0; i < num_info; i++) {
1524                        line_info = line_data_cur;
1525                        r = fn(&line_info->file_name_off, ctx);
1526                        if (r)
1527                                return r;
1528                        r = fn(&line_info->line_off, ctx);
1529                        if (r)
1530                                return r;
1531                        line_data_cur += rec_size;
1532                }
1533        }
1534
1535        return 0;
1536}
1537
1538static int str_sort_by_content(const void *a1, const void *a2)
1539{
1540        const struct btf_str_ptr *p1 = a1;
1541        const struct btf_str_ptr *p2 = a2;
1542
1543        return strcmp(p1->str, p2->str);
1544}
1545
1546static int str_sort_by_offset(const void *a1, const void *a2)
1547{
1548        const struct btf_str_ptr *p1 = a1;
1549        const struct btf_str_ptr *p2 = a2;
1550
1551        if (p1->str != p2->str)
1552                return p1->str < p2->str ? -1 : 1;
1553        return 0;
1554}
1555
1556static int btf_dedup_str_ptr_cmp(const void *str_ptr, const void *pelem)
1557{
1558        const struct btf_str_ptr *p = pelem;
1559
1560        if (str_ptr != p->str)
1561                return (const char *)str_ptr < p->str ? -1 : 1;
1562        return 0;
1563}
1564
1565static int btf_str_mark_as_used(__u32 *str_off_ptr, void *ctx)
1566{
1567        struct btf_str_ptrs *strs;
1568        struct btf_str_ptr *s;
1569
1570        if (*str_off_ptr == 0)
1571                return 0;
1572
1573        strs = ctx;
1574        s = bsearch(strs->data + *str_off_ptr, strs->ptrs, strs->cnt,
1575                    sizeof(struct btf_str_ptr), btf_dedup_str_ptr_cmp);
1576        if (!s)
1577                return -EINVAL;
1578        s->used = true;
1579        return 0;
1580}
1581
1582static int btf_str_remap_offset(__u32 *str_off_ptr, void *ctx)
1583{
1584        struct btf_str_ptrs *strs;
1585        struct btf_str_ptr *s;
1586
1587        if (*str_off_ptr == 0)
1588                return 0;
1589
1590        strs = ctx;
1591        s = bsearch(strs->data + *str_off_ptr, strs->ptrs, strs->cnt,
1592                    sizeof(struct btf_str_ptr), btf_dedup_str_ptr_cmp);
1593        if (!s)
1594                return -EINVAL;
1595        *str_off_ptr = s->new_off;
1596        return 0;
1597}
1598
1599/*
1600 * Dedup string and filter out those that are not referenced from either .BTF
1601 * or .BTF.ext (if provided) sections.
1602 *
1603 * This is done by building index of all strings in BTF's string section,
1604 * then iterating over all entities that can reference strings (e.g., type
1605 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
1606 * strings as used. After that all used strings are deduped and compacted into
1607 * sequential blob of memory and new offsets are calculated. Then all the string
1608 * references are iterated again and rewritten using new offsets.
1609 */
1610static int btf_dedup_strings(struct btf_dedup *d)
1611{
1612        const struct btf_header *hdr = d->btf->hdr;
1613        char *start = (char *)d->btf->nohdr_data + hdr->str_off;
1614        char *end = start + d->btf->hdr->str_len;
1615        char *p = start, *tmp_strs = NULL;
1616        struct btf_str_ptrs strs = {
1617                .cnt = 0,
1618                .cap = 0,
1619                .ptrs = NULL,
1620                .data = start,
1621        };
1622        int i, j, err = 0, grp_idx;
1623        bool grp_used;
1624
1625        /* build index of all strings */
1626        while (p < end) {
1627                if (strs.cnt + 1 > strs.cap) {
1628                        struct btf_str_ptr *new_ptrs;
1629
1630                        strs.cap += max(strs.cnt / 2, 16);
1631                        new_ptrs = realloc(strs.ptrs,
1632                                           sizeof(strs.ptrs[0]) * strs.cap);
1633                        if (!new_ptrs) {
1634                                err = -ENOMEM;
1635                                goto done;
1636                        }
1637                        strs.ptrs = new_ptrs;
1638                }
1639
1640                strs.ptrs[strs.cnt].str = p;
1641                strs.ptrs[strs.cnt].used = false;
1642
1643                p += strlen(p) + 1;
1644                strs.cnt++;
1645        }
1646
1647        /* temporary storage for deduplicated strings */
1648        tmp_strs = malloc(d->btf->hdr->str_len);
1649        if (!tmp_strs) {
1650                err = -ENOMEM;
1651                goto done;
1652        }
1653
1654        /* mark all used strings */
1655        strs.ptrs[0].used = true;
1656        err = btf_for_each_str_off(d, btf_str_mark_as_used, &strs);
1657        if (err)
1658                goto done;
1659
1660        /* sort strings by context, so that we can identify duplicates */
1661        qsort(strs.ptrs, strs.cnt, sizeof(strs.ptrs[0]), str_sort_by_content);
1662
1663        /*
1664         * iterate groups of equal strings and if any instance in a group was
1665         * referenced, emit single instance and remember new offset
1666         */
1667        p = tmp_strs;
1668        grp_idx = 0;
1669        grp_used = strs.ptrs[0].used;
1670        /* iterate past end to avoid code duplication after loop */
1671        for (i = 1; i <= strs.cnt; i++) {
1672                /*
1673                 * when i == strs.cnt, we want to skip string comparison and go
1674                 * straight to handling last group of strings (otherwise we'd
1675                 * need to handle last group after the loop w/ duplicated code)
1676                 */
1677                if (i < strs.cnt &&
1678                    !strcmp(strs.ptrs[i].str, strs.ptrs[grp_idx].str)) {
1679                        grp_used = grp_used || strs.ptrs[i].used;
1680                        continue;
1681                }
1682
1683                /*
1684                 * this check would have been required after the loop to handle
1685                 * last group of strings, but due to <= condition in a loop
1686                 * we avoid that duplication
1687                 */
1688                if (grp_used) {
1689                        int new_off = p - tmp_strs;
1690                        __u32 len = strlen(strs.ptrs[grp_idx].str);
1691
1692                        memmove(p, strs.ptrs[grp_idx].str, len + 1);
1693                        for (j = grp_idx; j < i; j++)
1694                                strs.ptrs[j].new_off = new_off;
1695                        p += len + 1;
1696                }
1697
1698                if (i < strs.cnt) {
1699                        grp_idx = i;
1700                        grp_used = strs.ptrs[i].used;
1701                }
1702        }
1703
1704        /* replace original strings with deduped ones */
1705        d->btf->hdr->str_len = p - tmp_strs;
1706        memmove(start, tmp_strs, d->btf->hdr->str_len);
1707        end = start + d->btf->hdr->str_len;
1708
1709        /* restore original order for further binary search lookups */
1710        qsort(strs.ptrs, strs.cnt, sizeof(strs.ptrs[0]), str_sort_by_offset);
1711
1712        /* remap string offsets */
1713        err = btf_for_each_str_off(d, btf_str_remap_offset, &strs);
1714        if (err)
1715                goto done;
1716
1717        d->btf->hdr->str_len = end - start;
1718
1719done:
1720        free(tmp_strs);
1721        free(strs.ptrs);
1722        return err;
1723}
1724
1725static long btf_hash_common(struct btf_type *t)
1726{
1727        long h;
1728
1729        h = hash_combine(0, t->name_off);
1730        h = hash_combine(h, t->info);
1731        h = hash_combine(h, t->size);
1732        return h;
1733}
1734
1735static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
1736{
1737        return t1->name_off == t2->name_off &&
1738               t1->info == t2->info &&
1739               t1->size == t2->size;
1740}
1741
1742/* Calculate type signature hash of INT. */
1743static long btf_hash_int(struct btf_type *t)
1744{
1745        __u32 info = *(__u32 *)(t + 1);
1746        long h;
1747
1748        h = btf_hash_common(t);
1749        h = hash_combine(h, info);
1750        return h;
1751}
1752
1753/* Check structural equality of two INTs. */
1754static bool btf_equal_int(struct btf_type *t1, struct btf_type *t2)
1755{
1756        __u32 info1, info2;
1757
1758        if (!btf_equal_common(t1, t2))
1759                return false;
1760        info1 = *(__u32 *)(t1 + 1);
1761        info2 = *(__u32 *)(t2 + 1);
1762        return info1 == info2;
1763}
1764
1765/* Calculate type signature hash of ENUM. */
1766static long btf_hash_enum(struct btf_type *t)
1767{
1768        long h;
1769
1770        /* don't hash vlen and enum members to support enum fwd resolving */
1771        h = hash_combine(0, t->name_off);
1772        h = hash_combine(h, t->info & ~0xffff);
1773        h = hash_combine(h, t->size);
1774        return h;
1775}
1776
1777/* Check structural equality of two ENUMs. */
1778static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
1779{
1780        const struct btf_enum *m1, *m2;
1781        __u16 vlen;
1782        int i;
1783
1784        if (!btf_equal_common(t1, t2))
1785                return false;
1786
1787        vlen = btf_vlen(t1);
1788        m1 = btf_enum(t1);
1789        m2 = btf_enum(t2);
1790        for (i = 0; i < vlen; i++) {
1791                if (m1->name_off != m2->name_off || m1->val != m2->val)
1792                        return false;
1793                m1++;
1794                m2++;
1795        }
1796        return true;
1797}
1798
1799static inline bool btf_is_enum_fwd(struct btf_type *t)
1800{
1801        return btf_is_enum(t) && btf_vlen(t) == 0;
1802}
1803
1804static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
1805{
1806        if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
1807                return btf_equal_enum(t1, t2);
1808        /* ignore vlen when comparing */
1809        return t1->name_off == t2->name_off &&
1810               (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
1811               t1->size == t2->size;
1812}
1813
1814/*
1815 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
1816 * as referenced type IDs equivalence is established separately during type
1817 * graph equivalence check algorithm.
1818 */
1819static long btf_hash_struct(struct btf_type *t)
1820{
1821        const struct btf_member *member = btf_members(t);
1822        __u32 vlen = btf_vlen(t);
1823        long h = btf_hash_common(t);
1824        int i;
1825
1826        for (i = 0; i < vlen; i++) {
1827                h = hash_combine(h, member->name_off);
1828                h = hash_combine(h, member->offset);
1829                /* no hashing of referenced type ID, it can be unresolved yet */
1830                member++;
1831        }
1832        return h;
1833}
1834
1835/*
1836 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
1837 * IDs. This check is performed during type graph equivalence check and
1838 * referenced types equivalence is checked separately.
1839 */
1840static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
1841{
1842        const struct btf_member *m1, *m2;
1843        __u16 vlen;
1844        int i;
1845
1846        if (!btf_equal_common(t1, t2))
1847                return false;
1848
1849        vlen = btf_vlen(t1);
1850        m1 = btf_members(t1);
1851        m2 = btf_members(t2);
1852        for (i = 0; i < vlen; i++) {
1853                if (m1->name_off != m2->name_off || m1->offset != m2->offset)
1854                        return false;
1855                m1++;
1856                m2++;
1857        }
1858        return true;
1859}
1860
1861/*
1862 * Calculate type signature hash of ARRAY, including referenced type IDs,
1863 * under assumption that they were already resolved to canonical type IDs and
1864 * are not going to change.
1865 */
1866static long btf_hash_array(struct btf_type *t)
1867{
1868        const struct btf_array *info = btf_array(t);
1869        long h = btf_hash_common(t);
1870
1871        h = hash_combine(h, info->type);
1872        h = hash_combine(h, info->index_type);
1873        h = hash_combine(h, info->nelems);
1874        return h;
1875}
1876
1877/*
1878 * Check exact equality of two ARRAYs, taking into account referenced
1879 * type IDs, under assumption that they were already resolved to canonical
1880 * type IDs and are not going to change.
1881 * This function is called during reference types deduplication to compare
1882 * ARRAY to potential canonical representative.
1883 */
1884static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
1885{
1886        const struct btf_array *info1, *info2;
1887
1888        if (!btf_equal_common(t1, t2))
1889                return false;
1890
1891        info1 = btf_array(t1);
1892        info2 = btf_array(t2);
1893        return info1->type == info2->type &&
1894               info1->index_type == info2->index_type &&
1895               info1->nelems == info2->nelems;
1896}
1897
1898/*
1899 * Check structural compatibility of two ARRAYs, ignoring referenced type
1900 * IDs. This check is performed during type graph equivalence check and
1901 * referenced types equivalence is checked separately.
1902 */
1903static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
1904{
1905        if (!btf_equal_common(t1, t2))
1906                return false;
1907
1908        return btf_array(t1)->nelems == btf_array(t2)->nelems;
1909}
1910
1911/*
1912 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
1913 * under assumption that they were already resolved to canonical type IDs and
1914 * are not going to change.
1915 */
1916static long btf_hash_fnproto(struct btf_type *t)
1917{
1918        const struct btf_param *member = btf_params(t);
1919        __u16 vlen = btf_vlen(t);
1920        long h = btf_hash_common(t);
1921        int i;
1922
1923        for (i = 0; i < vlen; i++) {
1924                h = hash_combine(h, member->name_off);
1925                h = hash_combine(h, member->type);
1926                member++;
1927        }
1928        return h;
1929}
1930
1931/*
1932 * Check exact equality of two FUNC_PROTOs, taking into account referenced
1933 * type IDs, under assumption that they were already resolved to canonical
1934 * type IDs and are not going to change.
1935 * This function is called during reference types deduplication to compare
1936 * FUNC_PROTO to potential canonical representative.
1937 */
1938static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
1939{
1940        const struct btf_param *m1, *m2;
1941        __u16 vlen;
1942        int i;
1943
1944        if (!btf_equal_common(t1, t2))
1945                return false;
1946
1947        vlen = btf_vlen(t1);
1948        m1 = btf_params(t1);
1949        m2 = btf_params(t2);
1950        for (i = 0; i < vlen; i++) {
1951                if (m1->name_off != m2->name_off || m1->type != m2->type)
1952                        return false;
1953                m1++;
1954                m2++;
1955        }
1956        return true;
1957}
1958
1959/*
1960 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
1961 * IDs. This check is performed during type graph equivalence check and
1962 * referenced types equivalence is checked separately.
1963 */
1964static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
1965{
1966        const struct btf_param *m1, *m2;
1967        __u16 vlen;
1968        int i;
1969
1970        /* skip return type ID */
1971        if (t1->name_off != t2->name_off || t1->info != t2->info)
1972                return false;
1973
1974        vlen = btf_vlen(t1);
1975        m1 = btf_params(t1);
1976        m2 = btf_params(t2);
1977        for (i = 0; i < vlen; i++) {
1978                if (m1->name_off != m2->name_off)
1979                        return false;
1980                m1++;
1981                m2++;
1982        }
1983        return true;
1984}
1985
1986/*
1987 * Deduplicate primitive types, that can't reference other types, by calculating
1988 * their type signature hash and comparing them with any possible canonical
1989 * candidate. If no canonical candidate matches, type itself is marked as
1990 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
1991 */
1992static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
1993{
1994        struct btf_type *t = d->btf->types[type_id];
1995        struct hashmap_entry *hash_entry;
1996        struct btf_type *cand;
1997        /* if we don't find equivalent type, then we are canonical */
1998        __u32 new_id = type_id;
1999        __u32 cand_id;
2000        long h;
2001
2002        switch (btf_kind(t)) {
2003        case BTF_KIND_CONST:
2004        case BTF_KIND_VOLATILE:
2005        case BTF_KIND_RESTRICT:
2006        case BTF_KIND_PTR:
2007        case BTF_KIND_TYPEDEF:
2008        case BTF_KIND_ARRAY:
2009        case BTF_KIND_STRUCT:
2010        case BTF_KIND_UNION:
2011        case BTF_KIND_FUNC:
2012        case BTF_KIND_FUNC_PROTO:
2013        case BTF_KIND_VAR:
2014        case BTF_KIND_DATASEC:
2015                return 0;
2016
2017        case BTF_KIND_INT:
2018                h = btf_hash_int(t);
2019                for_each_dedup_cand(d, hash_entry, h) {
2020                        cand_id = (__u32)(long)hash_entry->value;
2021                        cand = d->btf->types[cand_id];
2022                        if (btf_equal_int(t, cand)) {
2023                                new_id = cand_id;
2024                                break;
2025                        }
2026                }
2027                break;
2028
2029        case BTF_KIND_ENUM:
2030                h = btf_hash_enum(t);
2031                for_each_dedup_cand(d, hash_entry, h) {
2032                        cand_id = (__u32)(long)hash_entry->value;
2033                        cand = d->btf->types[cand_id];
2034                        if (btf_equal_enum(t, cand)) {
2035                                new_id = cand_id;
2036                                break;
2037                        }
2038                        if (d->opts.dont_resolve_fwds)
2039                                continue;
2040                        if (btf_compat_enum(t, cand)) {
2041                                if (btf_is_enum_fwd(t)) {
2042                                        /* resolve fwd to full enum */
2043                                        new_id = cand_id;
2044                                        break;
2045                                }
2046                                /* resolve canonical enum fwd to full enum */
2047                                d->map[cand_id] = type_id;
2048                        }
2049                }
2050                break;
2051
2052        case BTF_KIND_FWD:
2053                h = btf_hash_common(t);
2054                for_each_dedup_cand(d, hash_entry, h) {
2055                        cand_id = (__u32)(long)hash_entry->value;
2056                        cand = d->btf->types[cand_id];
2057                        if (btf_equal_common(t, cand)) {
2058                                new_id = cand_id;
2059                                break;
2060                        }
2061                }
2062                break;
2063
2064        default:
2065                return -EINVAL;
2066        }
2067
2068        d->map[type_id] = new_id;
2069        if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
2070                return -ENOMEM;
2071
2072        return 0;
2073}
2074
2075static int btf_dedup_prim_types(struct btf_dedup *d)
2076{
2077        int i, err;
2078
2079        for (i = 1; i <= d->btf->nr_types; i++) {
2080                err = btf_dedup_prim_type(d, i);
2081                if (err)
2082                        return err;
2083        }
2084        return 0;
2085}
2086
2087/*
2088 * Check whether type is already mapped into canonical one (could be to itself).
2089 */
2090static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
2091{
2092        return d->map[type_id] <= BTF_MAX_NR_TYPES;
2093}
2094
2095/*
2096 * Resolve type ID into its canonical type ID, if any; otherwise return original
2097 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
2098 * STRUCT/UNION link and resolve it into canonical type ID as well.
2099 */
2100static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
2101{
2102        while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
2103                type_id = d->map[type_id];
2104        return type_id;
2105}
2106
2107/*
2108 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
2109 * type ID.
2110 */
2111static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
2112{
2113        __u32 orig_type_id = type_id;
2114
2115        if (!btf_is_fwd(d->btf->types[type_id]))
2116                return type_id;
2117
2118        while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
2119                type_id = d->map[type_id];
2120
2121        if (!btf_is_fwd(d->btf->types[type_id]))
2122                return type_id;
2123
2124        return orig_type_id;
2125}
2126
2127
2128static inline __u16 btf_fwd_kind(struct btf_type *t)
2129{
2130        return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
2131}
2132
2133/*
2134 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
2135 * call it "candidate graph" in this description for brevity) to a type graph
2136 * formed by (potential) canonical struct/union ("canonical graph" for brevity
2137 * here, though keep in mind that not all types in canonical graph are
2138 * necessarily canonical representatives themselves, some of them might be
2139 * duplicates or its uniqueness might not have been established yet).
2140 * Returns:
2141 *  - >0, if type graphs are equivalent;
2142 *  -  0, if not equivalent;
2143 *  - <0, on error.
2144 *
2145 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
2146 * equivalence of BTF types at each step. If at any point BTF types in candidate
2147 * and canonical graphs are not compatible structurally, whole graphs are
2148 * incompatible. If types are structurally equivalent (i.e., all information
2149 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
2150 * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
2151 * If a type references other types, then those referenced types are checked
2152 * for equivalence recursively.
2153 *
2154 * During DFS traversal, if we find that for current `canon_id` type we
2155 * already have some mapping in hypothetical map, we check for two possible
2156 * situations:
2157 *   - `canon_id` is mapped to exactly the same type as `cand_id`. This will
2158 *     happen when type graphs have cycles. In this case we assume those two
2159 *     types are equivalent.
2160 *   - `canon_id` is mapped to different type. This is contradiction in our
2161 *     hypothetical mapping, because same graph in canonical graph corresponds
2162 *     to two different types in candidate graph, which for equivalent type
2163 *     graphs shouldn't happen. This condition terminates equivalence check
2164 *     with negative result.
2165 *
2166 * If type graphs traversal exhausts types to check and find no contradiction,
2167 * then type graphs are equivalent.
2168 *
2169 * When checking types for equivalence, there is one special case: FWD types.
2170 * If FWD type resolution is allowed and one of the types (either from canonical
2171 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
2172 * flag) and their names match, hypothetical mapping is updated to point from
2173 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
2174 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
2175 *
2176 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
2177 * if there are two exactly named (or anonymous) structs/unions that are
2178 * compatible structurally, one of which has FWD field, while other is concrete
2179 * STRUCT/UNION, but according to C sources they are different structs/unions
2180 * that are referencing different types with the same name. This is extremely
2181 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
2182 * this logic is causing problems.
2183 *
2184 * Doing FWD resolution means that both candidate and/or canonical graphs can
2185 * consists of portions of the graph that come from multiple compilation units.
2186 * This is due to the fact that types within single compilation unit are always
2187 * deduplicated and FWDs are already resolved, if referenced struct/union
2188 * definiton is available. So, if we had unresolved FWD and found corresponding
2189 * STRUCT/UNION, they will be from different compilation units. This
2190 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
2191 * type graph will likely have at least two different BTF types that describe
2192 * same type (e.g., most probably there will be two different BTF types for the
2193 * same 'int' primitive type) and could even have "overlapping" parts of type
2194 * graph that describe same subset of types.
2195 *
2196 * This in turn means that our assumption that each type in canonical graph
2197 * must correspond to exactly one type in candidate graph might not hold
2198 * anymore and will make it harder to detect contradictions using hypothetical
2199 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
2200 * resolution only in canonical graph. FWDs in candidate graphs are never
2201 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
2202 * that can occur:
2203 *   - Both types in canonical and candidate graphs are FWDs. If they are
2204 *     structurally equivalent, then they can either be both resolved to the
2205 *     same STRUCT/UNION or not resolved at all. In both cases they are
2206 *     equivalent and there is no need to resolve FWD on candidate side.
2207 *   - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
2208 *     so nothing to resolve as well, algorithm will check equivalence anyway.
2209 *   - Type in canonical graph is FWD, while type in candidate is concrete
2210 *     STRUCT/UNION. In this case candidate graph comes from single compilation
2211 *     unit, so there is exactly one BTF type for each unique C type. After
2212 *     resolving FWD into STRUCT/UNION, there might be more than one BTF type
2213 *     in canonical graph mapping to single BTF type in candidate graph, but
2214 *     because hypothetical mapping maps from canonical to candidate types, it's
2215 *     alright, and we still maintain the property of having single `canon_id`
2216 *     mapping to single `cand_id` (there could be two different `canon_id`
2217 *     mapped to the same `cand_id`, but it's not contradictory).
2218 *   - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
2219 *     graph is FWD. In this case we are just going to check compatibility of
2220 *     STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
2221 *     assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
2222 *     a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
2223 *     turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
2224 *     canonical graph.
2225 */
2226static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
2227                              __u32 canon_id)
2228{
2229        struct btf_type *cand_type;
2230        struct btf_type *canon_type;
2231        __u32 hypot_type_id;
2232        __u16 cand_kind;
2233        __u16 canon_kind;
2234        int i, eq;
2235
2236        /* if both resolve to the same canonical, they must be equivalent */
2237        if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
2238                return 1;
2239
2240        canon_id = resolve_fwd_id(d, canon_id);
2241
2242        hypot_type_id = d->hypot_map[canon_id];
2243        if (hypot_type_id <= BTF_MAX_NR_TYPES)
2244                return hypot_type_id == cand_id;
2245
2246        if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
2247                return -ENOMEM;
2248
2249        cand_type = d->btf->types[cand_id];
2250        canon_type = d->btf->types[canon_id];
2251        cand_kind = btf_kind(cand_type);
2252        canon_kind = btf_kind(canon_type);
2253
2254        if (cand_type->name_off != canon_type->name_off)
2255                return 0;
2256
2257        /* FWD <--> STRUCT/UNION equivalence check, if enabled */
2258        if (!d->opts.dont_resolve_fwds
2259            && (cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
2260            && cand_kind != canon_kind) {
2261                __u16 real_kind;
2262                __u16 fwd_kind;
2263
2264                if (cand_kind == BTF_KIND_FWD) {
2265                        real_kind = canon_kind;
2266                        fwd_kind = btf_fwd_kind(cand_type);
2267                } else {
2268                        real_kind = cand_kind;
2269                        fwd_kind = btf_fwd_kind(canon_type);
2270                }
2271                return fwd_kind == real_kind;
2272        }
2273
2274        if (cand_kind != canon_kind)
2275                return 0;
2276
2277        switch (cand_kind) {
2278        case BTF_KIND_INT:
2279                return btf_equal_int(cand_type, canon_type);
2280
2281        case BTF_KIND_ENUM:
2282                if (d->opts.dont_resolve_fwds)
2283                        return btf_equal_enum(cand_type, canon_type);
2284                else
2285                        return btf_compat_enum(cand_type, canon_type);
2286
2287        case BTF_KIND_FWD:
2288                return btf_equal_common(cand_type, canon_type);
2289
2290        case BTF_KIND_CONST:
2291        case BTF_KIND_VOLATILE:
2292        case BTF_KIND_RESTRICT:
2293        case BTF_KIND_PTR:
2294        case BTF_KIND_TYPEDEF:
2295        case BTF_KIND_FUNC:
2296                if (cand_type->info != canon_type->info)
2297                        return 0;
2298                return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
2299
2300        case BTF_KIND_ARRAY: {
2301                const struct btf_array *cand_arr, *canon_arr;
2302
2303                if (!btf_compat_array(cand_type, canon_type))
2304                        return 0;
2305                cand_arr = btf_array(cand_type);
2306                canon_arr = btf_array(canon_type);
2307                eq = btf_dedup_is_equiv(d,
2308                        cand_arr->index_type, canon_arr->index_type);
2309                if (eq <= 0)
2310                        return eq;
2311                return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
2312        }
2313
2314        case BTF_KIND_STRUCT:
2315        case BTF_KIND_UNION: {
2316                const struct btf_member *cand_m, *canon_m;
2317                __u16 vlen;
2318
2319                if (!btf_shallow_equal_struct(cand_type, canon_type))
2320                        return 0;
2321                vlen = btf_vlen(cand_type);
2322                cand_m = btf_members(cand_type);
2323                canon_m = btf_members(canon_type);
2324                for (i = 0; i < vlen; i++) {
2325                        eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
2326                        if (eq <= 0)
2327                                return eq;
2328                        cand_m++;
2329                        canon_m++;
2330                }
2331
2332                return 1;
2333        }
2334
2335        case BTF_KIND_FUNC_PROTO: {
2336                const struct btf_param *cand_p, *canon_p;
2337                __u16 vlen;
2338
2339                if (!btf_compat_fnproto(cand_type, canon_type))
2340                        return 0;
2341                eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
2342                if (eq <= 0)
2343                        return eq;
2344                vlen = btf_vlen(cand_type);
2345                cand_p = btf_params(cand_type);
2346                canon_p = btf_params(canon_type);
2347                for (i = 0; i < vlen; i++) {
2348                        eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
2349                        if (eq <= 0)
2350                                return eq;
2351                        cand_p++;
2352                        canon_p++;
2353                }
2354                return 1;
2355        }
2356
2357        default:
2358                return -EINVAL;
2359        }
2360        return 0;
2361}
2362
2363/*
2364 * Use hypothetical mapping, produced by successful type graph equivalence
2365 * check, to augment existing struct/union canonical mapping, where possible.
2366 *
2367 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
2368 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
2369 * it doesn't matter if FWD type was part of canonical graph or candidate one,
2370 * we are recording the mapping anyway. As opposed to carefulness required
2371 * for struct/union correspondence mapping (described below), for FWD resolution
2372 * it's not important, as by the time that FWD type (reference type) will be
2373 * deduplicated all structs/unions will be deduped already anyway.
2374 *
2375 * Recording STRUCT/UNION mapping is purely a performance optimization and is
2376 * not required for correctness. It needs to be done carefully to ensure that
2377 * struct/union from candidate's type graph is not mapped into corresponding
2378 * struct/union from canonical type graph that itself hasn't been resolved into
2379 * canonical representative. The only guarantee we have is that canonical
2380 * struct/union was determined as canonical and that won't change. But any
2381 * types referenced through that struct/union fields could have been not yet
2382 * resolved, so in case like that it's too early to establish any kind of
2383 * correspondence between structs/unions.
2384 *
2385 * No canonical correspondence is derived for primitive types (they are already
2386 * deduplicated completely already anyway) or reference types (they rely on
2387 * stability of struct/union canonical relationship for equivalence checks).
2388 */
2389static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
2390{
2391        __u32 cand_type_id, targ_type_id;
2392        __u16 t_kind, c_kind;
2393        __u32 t_id, c_id;
2394        int i;
2395
2396        for (i = 0; i < d->hypot_cnt; i++) {
2397                cand_type_id = d->hypot_list[i];
2398                targ_type_id = d->hypot_map[cand_type_id];
2399                t_id = resolve_type_id(d, targ_type_id);
2400                c_id = resolve_type_id(d, cand_type_id);
2401                t_kind = btf_kind(d->btf->types[t_id]);
2402                c_kind = btf_kind(d->btf->types[c_id]);
2403                /*
2404                 * Resolve FWD into STRUCT/UNION.
2405                 * It's ok to resolve FWD into STRUCT/UNION that's not yet
2406                 * mapped to canonical representative (as opposed to
2407                 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
2408                 * eventually that struct is going to be mapped and all resolved
2409                 * FWDs will automatically resolve to correct canonical
2410                 * representative. This will happen before ref type deduping,
2411                 * which critically depends on stability of these mapping. This
2412                 * stability is not a requirement for STRUCT/UNION equivalence
2413                 * checks, though.
2414                 */
2415                if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
2416                        d->map[c_id] = t_id;
2417                else if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
2418                        d->map[t_id] = c_id;
2419
2420                if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
2421                    c_kind != BTF_KIND_FWD &&
2422                    is_type_mapped(d, c_id) &&
2423                    !is_type_mapped(d, t_id)) {
2424                        /*
2425                         * as a perf optimization, we can map struct/union
2426                         * that's part of type graph we just verified for
2427                         * equivalence. We can do that for struct/union that has
2428                         * canonical representative only, though.
2429                         */
2430                        d->map[t_id] = c_id;
2431                }
2432        }
2433}
2434
2435/*
2436 * Deduplicate struct/union types.
2437 *
2438 * For each struct/union type its type signature hash is calculated, taking
2439 * into account type's name, size, number, order and names of fields, but
2440 * ignoring type ID's referenced from fields, because they might not be deduped
2441 * completely until after reference types deduplication phase. This type hash
2442 * is used to iterate over all potential canonical types, sharing same hash.
2443 * For each canonical candidate we check whether type graphs that they form
2444 * (through referenced types in fields and so on) are equivalent using algorithm
2445 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
2446 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
2447 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
2448 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
2449 * potentially map other structs/unions to their canonical representatives,
2450 * if such relationship hasn't yet been established. This speeds up algorithm
2451 * by eliminating some of the duplicate work.
2452 *
2453 * If no matching canonical representative was found, struct/union is marked
2454 * as canonical for itself and is added into btf_dedup->dedup_table hash map
2455 * for further look ups.
2456 */
2457static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
2458{
2459        struct btf_type *cand_type, *t;
2460        struct hashmap_entry *hash_entry;
2461        /* if we don't find equivalent type, then we are canonical */
2462        __u32 new_id = type_id;
2463        __u16 kind;
2464        long h;
2465
2466        /* already deduped or is in process of deduping (loop detected) */
2467        if (d->map[type_id] <= BTF_MAX_NR_TYPES)
2468                return 0;
2469
2470        t = d->btf->types[type_id];
2471        kind = btf_kind(t);
2472
2473        if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
2474                return 0;
2475
2476        h = btf_hash_struct(t);
2477        for_each_dedup_cand(d, hash_entry, h) {
2478                __u32 cand_id = (__u32)(long)hash_entry->value;
2479                int eq;
2480
2481                /*
2482                 * Even though btf_dedup_is_equiv() checks for
2483                 * btf_shallow_equal_struct() internally when checking two
2484                 * structs (unions) for equivalence, we need to guard here
2485                 * from picking matching FWD type as a dedup candidate.
2486                 * This can happen due to hash collision. In such case just
2487                 * relying on btf_dedup_is_equiv() would lead to potentially
2488                 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
2489                 * FWD and compatible STRUCT/UNION are considered equivalent.
2490                 */
2491                cand_type = d->btf->types[cand_id];
2492                if (!btf_shallow_equal_struct(t, cand_type))
2493                        continue;
2494
2495                btf_dedup_clear_hypot_map(d);
2496                eq = btf_dedup_is_equiv(d, type_id, cand_id);
2497                if (eq < 0)
2498                        return eq;
2499                if (!eq)
2500                        continue;
2501                new_id = cand_id;
2502                btf_dedup_merge_hypot_map(d);
2503                break;
2504        }
2505
2506        d->map[type_id] = new_id;
2507        if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
2508                return -ENOMEM;
2509
2510        return 0;
2511}
2512
2513static int btf_dedup_struct_types(struct btf_dedup *d)
2514{
2515        int i, err;
2516
2517        for (i = 1; i <= d->btf->nr_types; i++) {
2518                err = btf_dedup_struct_type(d, i);
2519                if (err)
2520                        return err;
2521        }
2522        return 0;
2523}
2524
2525/*
2526 * Deduplicate reference type.
2527 *
2528 * Once all primitive and struct/union types got deduplicated, we can easily
2529 * deduplicate all other (reference) BTF types. This is done in two steps:
2530 *
2531 * 1. Resolve all referenced type IDs into their canonical type IDs. This
2532 * resolution can be done either immediately for primitive or struct/union types
2533 * (because they were deduped in previous two phases) or recursively for
2534 * reference types. Recursion will always terminate at either primitive or
2535 * struct/union type, at which point we can "unwind" chain of reference types
2536 * one by one. There is no danger of encountering cycles because in C type
2537 * system the only way to form type cycle is through struct/union, so any chain
2538 * of reference types, even those taking part in a type cycle, will inevitably
2539 * reach struct/union at some point.
2540 *
2541 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
2542 * becomes "stable", in the sense that no further deduplication will cause
2543 * any changes to it. With that, it's now possible to calculate type's signature
2544 * hash (this time taking into account referenced type IDs) and loop over all
2545 * potential canonical representatives. If no match was found, current type
2546 * will become canonical representative of itself and will be added into
2547 * btf_dedup->dedup_table as another possible canonical representative.
2548 */
2549static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
2550{
2551        struct hashmap_entry *hash_entry;
2552        __u32 new_id = type_id, cand_id;
2553        struct btf_type *t, *cand;
2554        /* if we don't find equivalent type, then we are representative type */
2555        int ref_type_id;
2556        long h;
2557
2558        if (d->map[type_id] == BTF_IN_PROGRESS_ID)
2559                return -ELOOP;
2560        if (d->map[type_id] <= BTF_MAX_NR_TYPES)
2561                return resolve_type_id(d, type_id);
2562
2563        t = d->btf->types[type_id];
2564        d->map[type_id] = BTF_IN_PROGRESS_ID;
2565
2566        switch (btf_kind(t)) {
2567        case BTF_KIND_CONST:
2568        case BTF_KIND_VOLATILE:
2569        case BTF_KIND_RESTRICT:
2570        case BTF_KIND_PTR:
2571        case BTF_KIND_TYPEDEF:
2572        case BTF_KIND_FUNC:
2573                ref_type_id = btf_dedup_ref_type(d, t->type);
2574                if (ref_type_id < 0)
2575                        return ref_type_id;
2576                t->type = ref_type_id;
2577
2578                h = btf_hash_common(t);
2579                for_each_dedup_cand(d, hash_entry, h) {
2580                        cand_id = (__u32)(long)hash_entry->value;
2581                        cand = d->btf->types[cand_id];
2582                        if (btf_equal_common(t, cand)) {
2583                                new_id = cand_id;
2584                                break;
2585                        }
2586                }
2587                break;
2588
2589        case BTF_KIND_ARRAY: {
2590                struct btf_array *info = btf_array(t);
2591
2592                ref_type_id = btf_dedup_ref_type(d, info->type);
2593                if (ref_type_id < 0)
2594                        return ref_type_id;
2595                info->type = ref_type_id;
2596
2597                ref_type_id = btf_dedup_ref_type(d, info->index_type);
2598                if (ref_type_id < 0)
2599                        return ref_type_id;
2600                info->index_type = ref_type_id;
2601
2602                h = btf_hash_array(t);
2603                for_each_dedup_cand(d, hash_entry, h) {
2604                        cand_id = (__u32)(long)hash_entry->value;
2605                        cand = d->btf->types[cand_id];
2606                        if (btf_equal_array(t, cand)) {
2607                                new_id = cand_id;
2608                                break;
2609                        }
2610                }
2611                break;
2612        }
2613
2614        case BTF_KIND_FUNC_PROTO: {
2615                struct btf_param *param;
2616                __u16 vlen;
2617                int i;
2618
2619                ref_type_id = btf_dedup_ref_type(d, t->type);
2620                if (ref_type_id < 0)
2621                        return ref_type_id;
2622                t->type = ref_type_id;
2623
2624                vlen = btf_vlen(t);
2625                param = btf_params(t);
2626                for (i = 0; i < vlen; i++) {
2627                        ref_type_id = btf_dedup_ref_type(d, param->type);
2628                        if (ref_type_id < 0)
2629                                return ref_type_id;
2630                        param->type = ref_type_id;
2631                        param++;
2632                }
2633
2634                h = btf_hash_fnproto(t);
2635                for_each_dedup_cand(d, hash_entry, h) {
2636                        cand_id = (__u32)(long)hash_entry->value;
2637                        cand = d->btf->types[cand_id];
2638                        if (btf_equal_fnproto(t, cand)) {
2639                                new_id = cand_id;
2640                                break;
2641                        }
2642                }
2643                break;
2644        }
2645
2646        default:
2647                return -EINVAL;
2648        }
2649
2650        d->map[type_id] = new_id;
2651        if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
2652                return -ENOMEM;
2653
2654        return new_id;
2655}
2656
2657static int btf_dedup_ref_types(struct btf_dedup *d)
2658{
2659        int i, err;
2660
2661        for (i = 1; i <= d->btf->nr_types; i++) {
2662                err = btf_dedup_ref_type(d, i);
2663                if (err < 0)
2664                        return err;
2665        }
2666        /* we won't need d->dedup_table anymore */
2667        hashmap__free(d->dedup_table);
2668        d->dedup_table = NULL;
2669        return 0;
2670}
2671
2672/*
2673 * Compact types.
2674 *
2675 * After we established for each type its corresponding canonical representative
2676 * type, we now can eliminate types that are not canonical and leave only
2677 * canonical ones layed out sequentially in memory by copying them over
2678 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
2679 * a map from original type ID to a new compacted type ID, which will be used
2680 * during next phase to "fix up" type IDs, referenced from struct/union and
2681 * reference types.
2682 */
2683static int btf_dedup_compact_types(struct btf_dedup *d)
2684{
2685        struct btf_type **new_types;
2686        __u32 next_type_id = 1;
2687        char *types_start, *p;
2688        int i, len;
2689
2690        /* we are going to reuse hypot_map to store compaction remapping */
2691        d->hypot_map[0] = 0;
2692        for (i = 1; i <= d->btf->nr_types; i++)
2693                d->hypot_map[i] = BTF_UNPROCESSED_ID;
2694
2695        types_start = d->btf->nohdr_data + d->btf->hdr->type_off;
2696        p = types_start;
2697
2698        for (i = 1; i <= d->btf->nr_types; i++) {
2699                if (d->map[i] != i)
2700                        continue;
2701
2702                len = btf_type_size(d->btf->types[i]);
2703                if (len < 0)
2704                        return len;
2705
2706                memmove(p, d->btf->types[i], len);
2707                d->hypot_map[i] = next_type_id;
2708                d->btf->types[next_type_id] = (struct btf_type *)p;
2709                p += len;
2710                next_type_id++;
2711        }
2712
2713        /* shrink struct btf's internal types index and update btf_header */
2714        d->btf->nr_types = next_type_id - 1;
2715        d->btf->types_size = d->btf->nr_types;
2716        d->btf->hdr->type_len = p - types_start;
2717        new_types = realloc(d->btf->types,
2718                            (1 + d->btf->nr_types) * sizeof(struct btf_type *));
2719        if (!new_types)
2720                return -ENOMEM;
2721        d->btf->types = new_types;
2722
2723        /* make sure string section follows type information without gaps */
2724        d->btf->hdr->str_off = p - (char *)d->btf->nohdr_data;
2725        memmove(p, d->btf->strings, d->btf->hdr->str_len);
2726        d->btf->strings = p;
2727        p += d->btf->hdr->str_len;
2728
2729        d->btf->data_size = p - (char *)d->btf->data;
2730        return 0;
2731}
2732
2733/*
2734 * Figure out final (deduplicated and compacted) type ID for provided original
2735 * `type_id` by first resolving it into corresponding canonical type ID and
2736 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
2737 * which is populated during compaction phase.
2738 */
2739static int btf_dedup_remap_type_id(struct btf_dedup *d, __u32 type_id)
2740{
2741        __u32 resolved_type_id, new_type_id;
2742
2743        resolved_type_id = resolve_type_id(d, type_id);
2744        new_type_id = d->hypot_map[resolved_type_id];
2745        if (new_type_id > BTF_MAX_NR_TYPES)
2746                return -EINVAL;
2747        return new_type_id;
2748}
2749
2750/*
2751 * Remap referenced type IDs into deduped type IDs.
2752 *
2753 * After BTF types are deduplicated and compacted, their final type IDs may
2754 * differ from original ones. The map from original to a corresponding
2755 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
2756 * compaction phase. During remapping phase we are rewriting all type IDs
2757 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
2758 * their final deduped type IDs.
2759 */
2760static int btf_dedup_remap_type(struct btf_dedup *d, __u32 type_id)
2761{
2762        struct btf_type *t = d->btf->types[type_id];
2763        int i, r;
2764
2765        switch (btf_kind(t)) {
2766        case BTF_KIND_INT:
2767        case BTF_KIND_ENUM:
2768                break;
2769
2770        case BTF_KIND_FWD:
2771        case BTF_KIND_CONST:
2772        case BTF_KIND_VOLATILE:
2773        case BTF_KIND_RESTRICT:
2774        case BTF_KIND_PTR:
2775        case BTF_KIND_TYPEDEF:
2776        case BTF_KIND_FUNC:
2777        case BTF_KIND_VAR:
2778                r = btf_dedup_remap_type_id(d, t->type);
2779                if (r < 0)
2780                        return r;
2781                t->type = r;
2782                break;
2783
2784        case BTF_KIND_ARRAY: {
2785                struct btf_array *arr_info = btf_array(t);
2786
2787                r = btf_dedup_remap_type_id(d, arr_info->type);
2788                if (r < 0)
2789                        return r;
2790                arr_info->type = r;
2791                r = btf_dedup_remap_type_id(d, arr_info->index_type);
2792                if (r < 0)
2793                        return r;
2794                arr_info->index_type = r;
2795                break;
2796        }
2797
2798        case BTF_KIND_STRUCT:
2799        case BTF_KIND_UNION: {
2800                struct btf_member *member = btf_members(t);
2801                __u16 vlen = btf_vlen(t);
2802
2803                for (i = 0; i < vlen; i++) {
2804                        r = btf_dedup_remap_type_id(d, member->type);
2805                        if (r < 0)
2806                                return r;
2807                        member->type = r;
2808                        member++;
2809                }
2810                break;
2811        }
2812
2813        case BTF_KIND_FUNC_PROTO: {
2814                struct btf_param *param = btf_params(t);
2815                __u16 vlen = btf_vlen(t);
2816
2817                r = btf_dedup_remap_type_id(d, t->type);
2818                if (r < 0)
2819                        return r;
2820                t->type = r;
2821
2822                for (i = 0; i < vlen; i++) {
2823                        r = btf_dedup_remap_type_id(d, param->type);
2824                        if (r < 0)
2825                                return r;
2826                        param->type = r;
2827                        param++;
2828                }
2829                break;
2830        }
2831
2832        case BTF_KIND_DATASEC: {
2833                struct btf_var_secinfo *var = btf_var_secinfos(t);
2834                __u16 vlen = btf_vlen(t);
2835
2836                for (i = 0; i < vlen; i++) {
2837                        r = btf_dedup_remap_type_id(d, var->type);
2838                        if (r < 0)
2839                                return r;
2840                        var->type = r;
2841                        var++;
2842                }
2843                break;
2844        }
2845
2846        default:
2847                return -EINVAL;
2848        }
2849
2850        return 0;
2851}
2852
2853static int btf_dedup_remap_types(struct btf_dedup *d)
2854{
2855        int i, r;
2856
2857        for (i = 1; i <= d->btf->nr_types; i++) {
2858                r = btf_dedup_remap_type(d, i);
2859                if (r < 0)
2860                        return r;
2861        }
2862        return 0;
2863}
2864