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 <byteswap.h>
   5#include <endian.h>
   6#include <stdio.h>
   7#include <stdlib.h>
   8#include <string.h>
   9#include <fcntl.h>
  10#include <unistd.h>
  11#include <errno.h>
  12#include <sys/utsname.h>
  13#include <sys/param.h>
  14#include <sys/stat.h>
  15#include <linux/kernel.h>
  16#include <linux/err.h>
  17#include <linux/btf.h>
  18#include <gelf.h>
  19#include "btf.h"
  20#include "bpf.h"
  21#include "libbpf.h"
  22#include "libbpf_internal.h"
  23#include "hashmap.h"
  24#include "strset.h"
  25
  26#define BTF_MAX_NR_TYPES 0x7fffffffU
  27#define BTF_MAX_STR_OFFSET 0x7fffffffU
  28
  29static struct btf_type btf_void;
  30
  31struct btf {
  32        /* raw BTF data in native endianness */
  33        void *raw_data;
  34        /* raw BTF data in non-native endianness */
  35        void *raw_data_swapped;
  36        __u32 raw_size;
  37        /* whether target endianness differs from the native one */
  38        bool swapped_endian;
  39
  40        /*
  41         * When BTF is loaded from an ELF or raw memory it is stored
  42         * in a contiguous memory block. The hdr, type_data, and, strs_data
  43         * point inside that memory region to their respective parts of BTF
  44         * representation:
  45         *
  46         * +--------------------------------+
  47         * |  Header  |  Types  |  Strings  |
  48         * +--------------------------------+
  49         * ^          ^         ^
  50         * |          |         |
  51         * hdr        |         |
  52         * types_data-+         |
  53         * strs_data------------+
  54         *
  55         * If BTF data is later modified, e.g., due to types added or
  56         * removed, BTF deduplication performed, etc, this contiguous
  57         * representation is broken up into three independently allocated
  58         * memory regions to be able to modify them independently.
  59         * raw_data is nulled out at that point, but can be later allocated
  60         * and cached again if user calls btf__get_raw_data(), at which point
  61         * raw_data will contain a contiguous copy of header, types, and
  62         * strings:
  63         *
  64         * +----------+  +---------+  +-----------+
  65         * |  Header  |  |  Types  |  |  Strings  |
  66         * +----------+  +---------+  +-----------+
  67         * ^             ^            ^
  68         * |             |            |
  69         * hdr           |            |
  70         * types_data----+            |
  71         * strset__data(strs_set)-----+
  72         *
  73         *               +----------+---------+-----------+
  74         *               |  Header  |  Types  |  Strings  |
  75         * raw_data----->+----------+---------+-----------+
  76         */
  77        struct btf_header *hdr;
  78
  79        void *types_data;
  80        size_t types_data_cap; /* used size stored in hdr->type_len */
  81
  82        /* type ID to `struct btf_type *` lookup index
  83         * type_offs[0] corresponds to the first non-VOID type:
  84         *   - for base BTF it's type [1];
  85         *   - for split BTF it's the first non-base BTF type.
  86         */
  87        __u32 *type_offs;
  88        size_t type_offs_cap;
  89        /* number of types in this BTF instance:
  90         *   - doesn't include special [0] void type;
  91         *   - for split BTF counts number of types added on top of base BTF.
  92         */
  93        __u32 nr_types;
  94        /* if not NULL, points to the base BTF on top of which the current
  95         * split BTF is based
  96         */
  97        struct btf *base_btf;
  98        /* BTF type ID of the first type in this BTF instance:
  99         *   - for base BTF it's equal to 1;
 100         *   - for split BTF it's equal to biggest type ID of base BTF plus 1.
 101         */
 102        int start_id;
 103        /* logical string offset of this BTF instance:
 104         *   - for base BTF it's equal to 0;
 105         *   - for split BTF it's equal to total size of base BTF's string section size.
 106         */
 107        int start_str_off;
 108
 109        /* only one of strs_data or strs_set can be non-NULL, depending on
 110         * whether BTF is in a modifiable state (strs_set is used) or not
 111         * (strs_data points inside raw_data)
 112         */
 113        void *strs_data;
 114        /* a set of unique strings */
 115        struct strset *strs_set;
 116        /* whether strings are already deduplicated */
 117        bool strs_deduped;
 118
 119        /* BTF object FD, if loaded into kernel */
 120        int fd;
 121
 122        /* Pointer size (in bytes) for a target architecture of this BTF */
 123        int ptr_sz;
 124};
 125
 126static inline __u64 ptr_to_u64(const void *ptr)
 127{
 128        return (__u64) (unsigned long) ptr;
 129}
 130
 131/* Ensure given dynamically allocated memory region pointed to by *data* with
 132 * capacity of *cap_cnt* elements each taking *elem_sz* bytes has enough
 133 * memory to accomodate *add_cnt* new elements, assuming *cur_cnt* elements
 134 * are already used. At most *max_cnt* elements can be ever allocated.
 135 * If necessary, memory is reallocated and all existing data is copied over,
 136 * new pointer to the memory region is stored at *data, new memory region
 137 * capacity (in number of elements) is stored in *cap.
 138 * On success, memory pointer to the beginning of unused memory is returned.
 139 * On error, NULL is returned.
 140 */
 141void *libbpf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz,
 142                     size_t cur_cnt, size_t max_cnt, size_t add_cnt)
 143{
 144        size_t new_cnt;
 145        void *new_data;
 146
 147        if (cur_cnt + add_cnt <= *cap_cnt)
 148                return *data + cur_cnt * elem_sz;
 149
 150        /* requested more than the set limit */
 151        if (cur_cnt + add_cnt > max_cnt)
 152                return NULL;
 153
 154        new_cnt = *cap_cnt;
 155        new_cnt += new_cnt / 4;           /* expand by 25% */
 156        if (new_cnt < 16)                 /* but at least 16 elements */
 157                new_cnt = 16;
 158        if (new_cnt > max_cnt)            /* but not exceeding a set limit */
 159                new_cnt = max_cnt;
 160        if (new_cnt < cur_cnt + add_cnt)  /* also ensure we have enough memory */
 161                new_cnt = cur_cnt + add_cnt;
 162
 163        new_data = libbpf_reallocarray(*data, new_cnt, elem_sz);
 164        if (!new_data)
 165                return NULL;
 166
 167        /* zero out newly allocated portion of memory */
 168        memset(new_data + (*cap_cnt) * elem_sz, 0, (new_cnt - *cap_cnt) * elem_sz);
 169
 170        *data = new_data;
 171        *cap_cnt = new_cnt;
 172        return new_data + cur_cnt * elem_sz;
 173}
 174
 175/* Ensure given dynamically allocated memory region has enough allocated space
 176 * to accommodate *need_cnt* elements of size *elem_sz* bytes each
 177 */
 178int libbpf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt)
 179{
 180        void *p;
 181
 182        if (need_cnt <= *cap_cnt)
 183                return 0;
 184
 185        p = libbpf_add_mem(data, cap_cnt, elem_sz, *cap_cnt, SIZE_MAX, need_cnt - *cap_cnt);
 186        if (!p)
 187                return -ENOMEM;
 188
 189        return 0;
 190}
 191
 192static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
 193{
 194        __u32 *p;
 195
 196        p = libbpf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
 197                           btf->nr_types, BTF_MAX_NR_TYPES, 1);
 198        if (!p)
 199                return -ENOMEM;
 200
 201        *p = type_off;
 202        return 0;
 203}
 204
 205static void btf_bswap_hdr(struct btf_header *h)
 206{
 207        h->magic = bswap_16(h->magic);
 208        h->hdr_len = bswap_32(h->hdr_len);
 209        h->type_off = bswap_32(h->type_off);
 210        h->type_len = bswap_32(h->type_len);
 211        h->str_off = bswap_32(h->str_off);
 212        h->str_len = bswap_32(h->str_len);
 213}
 214
 215static int btf_parse_hdr(struct btf *btf)
 216{
 217        struct btf_header *hdr = btf->hdr;
 218        __u32 meta_left;
 219
 220        if (btf->raw_size < sizeof(struct btf_header)) {
 221                pr_debug("BTF header not found\n");
 222                return -EINVAL;
 223        }
 224
 225        if (hdr->magic == bswap_16(BTF_MAGIC)) {
 226                btf->swapped_endian = true;
 227                if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
 228                        pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
 229                                bswap_32(hdr->hdr_len));
 230                        return -ENOTSUP;
 231                }
 232                btf_bswap_hdr(hdr);
 233        } else if (hdr->magic != BTF_MAGIC) {
 234                pr_debug("Invalid BTF magic:%x\n", hdr->magic);
 235                return -EINVAL;
 236        }
 237
 238        meta_left = btf->raw_size - sizeof(*hdr);
 239        if (meta_left < hdr->str_off + hdr->str_len) {
 240                pr_debug("Invalid BTF total size:%u\n", btf->raw_size);
 241                return -EINVAL;
 242        }
 243
 244        if (hdr->type_off + hdr->type_len > hdr->str_off) {
 245                pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
 246                         hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
 247                return -EINVAL;
 248        }
 249
 250        if (hdr->type_off % 4) {
 251                pr_debug("BTF type section is not aligned to 4 bytes\n");
 252                return -EINVAL;
 253        }
 254
 255        return 0;
 256}
 257
 258static int btf_parse_str_sec(struct btf *btf)
 259{
 260        const struct btf_header *hdr = btf->hdr;
 261        const char *start = btf->strs_data;
 262        const char *end = start + btf->hdr->str_len;
 263
 264        if (btf->base_btf && hdr->str_len == 0)
 265                return 0;
 266        if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET || end[-1]) {
 267                pr_debug("Invalid BTF string section\n");
 268                return -EINVAL;
 269        }
 270        if (!btf->base_btf && start[0]) {
 271                pr_debug("Invalid BTF string section\n");
 272                return -EINVAL;
 273        }
 274        return 0;
 275}
 276
 277static int btf_type_size(const struct btf_type *t)
 278{
 279        const int base_size = sizeof(struct btf_type);
 280        __u16 vlen = btf_vlen(t);
 281
 282        switch (btf_kind(t)) {
 283        case BTF_KIND_FWD:
 284        case BTF_KIND_CONST:
 285        case BTF_KIND_VOLATILE:
 286        case BTF_KIND_RESTRICT:
 287        case BTF_KIND_PTR:
 288        case BTF_KIND_TYPEDEF:
 289        case BTF_KIND_FUNC:
 290        case BTF_KIND_FLOAT:
 291                return base_size;
 292        case BTF_KIND_INT:
 293                return base_size + sizeof(__u32);
 294        case BTF_KIND_ENUM:
 295                return base_size + vlen * sizeof(struct btf_enum);
 296        case BTF_KIND_ARRAY:
 297                return base_size + sizeof(struct btf_array);
 298        case BTF_KIND_STRUCT:
 299        case BTF_KIND_UNION:
 300                return base_size + vlen * sizeof(struct btf_member);
 301        case BTF_KIND_FUNC_PROTO:
 302                return base_size + vlen * sizeof(struct btf_param);
 303        case BTF_KIND_VAR:
 304                return base_size + sizeof(struct btf_var);
 305        case BTF_KIND_DATASEC:
 306                return base_size + vlen * sizeof(struct btf_var_secinfo);
 307        default:
 308                pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
 309                return -EINVAL;
 310        }
 311}
 312
 313static void btf_bswap_type_base(struct btf_type *t)
 314{
 315        t->name_off = bswap_32(t->name_off);
 316        t->info = bswap_32(t->info);
 317        t->type = bswap_32(t->type);
 318}
 319
 320static int btf_bswap_type_rest(struct btf_type *t)
 321{
 322        struct btf_var_secinfo *v;
 323        struct btf_member *m;
 324        struct btf_array *a;
 325        struct btf_param *p;
 326        struct btf_enum *e;
 327        __u16 vlen = btf_vlen(t);
 328        int i;
 329
 330        switch (btf_kind(t)) {
 331        case BTF_KIND_FWD:
 332        case BTF_KIND_CONST:
 333        case BTF_KIND_VOLATILE:
 334        case BTF_KIND_RESTRICT:
 335        case BTF_KIND_PTR:
 336        case BTF_KIND_TYPEDEF:
 337        case BTF_KIND_FUNC:
 338        case BTF_KIND_FLOAT:
 339                return 0;
 340        case BTF_KIND_INT:
 341                *(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
 342                return 0;
 343        case BTF_KIND_ENUM:
 344                for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
 345                        e->name_off = bswap_32(e->name_off);
 346                        e->val = bswap_32(e->val);
 347                }
 348                return 0;
 349        case BTF_KIND_ARRAY:
 350                a = btf_array(t);
 351                a->type = bswap_32(a->type);
 352                a->index_type = bswap_32(a->index_type);
 353                a->nelems = bswap_32(a->nelems);
 354                return 0;
 355        case BTF_KIND_STRUCT:
 356        case BTF_KIND_UNION:
 357                for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
 358                        m->name_off = bswap_32(m->name_off);
 359                        m->type = bswap_32(m->type);
 360                        m->offset = bswap_32(m->offset);
 361                }
 362                return 0;
 363        case BTF_KIND_FUNC_PROTO:
 364                for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
 365                        p->name_off = bswap_32(p->name_off);
 366                        p->type = bswap_32(p->type);
 367                }
 368                return 0;
 369        case BTF_KIND_VAR:
 370                btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
 371                return 0;
 372        case BTF_KIND_DATASEC:
 373                for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
 374                        v->type = bswap_32(v->type);
 375                        v->offset = bswap_32(v->offset);
 376                        v->size = bswap_32(v->size);
 377                }
 378                return 0;
 379        default:
 380                pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
 381                return -EINVAL;
 382        }
 383}
 384
 385static int btf_parse_type_sec(struct btf *btf)
 386{
 387        struct btf_header *hdr = btf->hdr;
 388        void *next_type = btf->types_data;
 389        void *end_type = next_type + hdr->type_len;
 390        int err, type_size;
 391
 392        while (next_type + sizeof(struct btf_type) <= end_type) {
 393                if (btf->swapped_endian)
 394                        btf_bswap_type_base(next_type);
 395
 396                type_size = btf_type_size(next_type);
 397                if (type_size < 0)
 398                        return type_size;
 399                if (next_type + type_size > end_type) {
 400                        pr_warn("BTF type [%d] is malformed\n", btf->start_id + btf->nr_types);
 401                        return -EINVAL;
 402                }
 403
 404                if (btf->swapped_endian && btf_bswap_type_rest(next_type))
 405                        return -EINVAL;
 406
 407                err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
 408                if (err)
 409                        return err;
 410
 411                next_type += type_size;
 412                btf->nr_types++;
 413        }
 414
 415        if (next_type != end_type) {
 416                pr_warn("BTF types data is malformed\n");
 417                return -EINVAL;
 418        }
 419
 420        return 0;
 421}
 422
 423__u32 btf__get_nr_types(const struct btf *btf)
 424{
 425        return btf->start_id + btf->nr_types - 1;
 426}
 427
 428const struct btf *btf__base_btf(const struct btf *btf)
 429{
 430        return btf->base_btf;
 431}
 432
 433/* internal helper returning non-const pointer to a type */
 434struct btf_type *btf_type_by_id(struct btf *btf, __u32 type_id)
 435{
 436        if (type_id == 0)
 437                return &btf_void;
 438        if (type_id < btf->start_id)
 439                return btf_type_by_id(btf->base_btf, type_id);
 440        return btf->types_data + btf->type_offs[type_id - btf->start_id];
 441}
 442
 443const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
 444{
 445        if (type_id >= btf->start_id + btf->nr_types)
 446                return NULL;
 447        return btf_type_by_id((struct btf *)btf, type_id);
 448}
 449
 450static int determine_ptr_size(const struct btf *btf)
 451{
 452        const struct btf_type *t;
 453        const char *name;
 454        int i, n;
 455
 456        if (btf->base_btf && btf->base_btf->ptr_sz > 0)
 457                return btf->base_btf->ptr_sz;
 458
 459        n = btf__get_nr_types(btf);
 460        for (i = 1; i <= n; i++) {
 461                t = btf__type_by_id(btf, i);
 462                if (!btf_is_int(t))
 463                        continue;
 464
 465                name = btf__name_by_offset(btf, t->name_off);
 466                if (!name)
 467                        continue;
 468
 469                if (strcmp(name, "long int") == 0 ||
 470                    strcmp(name, "long unsigned int") == 0) {
 471                        if (t->size != 4 && t->size != 8)
 472                                continue;
 473                        return t->size;
 474                }
 475        }
 476
 477        return -1;
 478}
 479
 480static size_t btf_ptr_sz(const struct btf *btf)
 481{
 482        if (!btf->ptr_sz)
 483                ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
 484        return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
 485}
 486
 487/* Return pointer size this BTF instance assumes. The size is heuristically
 488 * determined by looking for 'long' or 'unsigned long' integer type and
 489 * recording its size in bytes. If BTF type information doesn't have any such
 490 * type, this function returns 0. In the latter case, native architecture's
 491 * pointer size is assumed, so will be either 4 or 8, depending on
 492 * architecture that libbpf was compiled for. It's possible to override
 493 * guessed value by using btf__set_pointer_size() API.
 494 */
 495size_t btf__pointer_size(const struct btf *btf)
 496{
 497        if (!btf->ptr_sz)
 498                ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
 499
 500        if (btf->ptr_sz < 0)
 501                /* not enough BTF type info to guess */
 502                return 0;
 503
 504        return btf->ptr_sz;
 505}
 506
 507/* Override or set pointer size in bytes. Only values of 4 and 8 are
 508 * supported.
 509 */
 510int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
 511{
 512        if (ptr_sz != 4 && ptr_sz != 8)
 513                return -EINVAL;
 514        btf->ptr_sz = ptr_sz;
 515        return 0;
 516}
 517
 518static bool is_host_big_endian(void)
 519{
 520#if __BYTE_ORDER == __LITTLE_ENDIAN
 521        return false;
 522#elif __BYTE_ORDER == __BIG_ENDIAN
 523        return true;
 524#else
 525# error "Unrecognized __BYTE_ORDER__"
 526#endif
 527}
 528
 529enum btf_endianness btf__endianness(const struct btf *btf)
 530{
 531        if (is_host_big_endian())
 532                return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
 533        else
 534                return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
 535}
 536
 537int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
 538{
 539        if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
 540                return -EINVAL;
 541
 542        btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
 543        if (!btf->swapped_endian) {
 544                free(btf->raw_data_swapped);
 545                btf->raw_data_swapped = NULL;
 546        }
 547        return 0;
 548}
 549
 550static bool btf_type_is_void(const struct btf_type *t)
 551{
 552        return t == &btf_void || btf_is_fwd(t);
 553}
 554
 555static bool btf_type_is_void_or_null(const struct btf_type *t)
 556{
 557        return !t || btf_type_is_void(t);
 558}
 559
 560#define MAX_RESOLVE_DEPTH 32
 561
 562__s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
 563{
 564        const struct btf_array *array;
 565        const struct btf_type *t;
 566        __u32 nelems = 1;
 567        __s64 size = -1;
 568        int i;
 569
 570        t = btf__type_by_id(btf, type_id);
 571        for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t);
 572             i++) {
 573                switch (btf_kind(t)) {
 574                case BTF_KIND_INT:
 575                case BTF_KIND_STRUCT:
 576                case BTF_KIND_UNION:
 577                case BTF_KIND_ENUM:
 578                case BTF_KIND_DATASEC:
 579                case BTF_KIND_FLOAT:
 580                        size = t->size;
 581                        goto done;
 582                case BTF_KIND_PTR:
 583                        size = btf_ptr_sz(btf);
 584                        goto done;
 585                case BTF_KIND_TYPEDEF:
 586                case BTF_KIND_VOLATILE:
 587                case BTF_KIND_CONST:
 588                case BTF_KIND_RESTRICT:
 589                case BTF_KIND_VAR:
 590                        type_id = t->type;
 591                        break;
 592                case BTF_KIND_ARRAY:
 593                        array = btf_array(t);
 594                        if (nelems && array->nelems > UINT32_MAX / nelems)
 595                                return -E2BIG;
 596                        nelems *= array->nelems;
 597                        type_id = array->type;
 598                        break;
 599                default:
 600                        return -EINVAL;
 601                }
 602
 603                t = btf__type_by_id(btf, type_id);
 604        }
 605
 606done:
 607        if (size < 0)
 608                return -EINVAL;
 609        if (nelems && size > UINT32_MAX / nelems)
 610                return -E2BIG;
 611
 612        return nelems * size;
 613}
 614
 615int btf__align_of(const struct btf *btf, __u32 id)
 616{
 617        const struct btf_type *t = btf__type_by_id(btf, id);
 618        __u16 kind = btf_kind(t);
 619
 620        switch (kind) {
 621        case BTF_KIND_INT:
 622        case BTF_KIND_ENUM:
 623        case BTF_KIND_FLOAT:
 624                return min(btf_ptr_sz(btf), (size_t)t->size);
 625        case BTF_KIND_PTR:
 626                return btf_ptr_sz(btf);
 627        case BTF_KIND_TYPEDEF:
 628        case BTF_KIND_VOLATILE:
 629        case BTF_KIND_CONST:
 630        case BTF_KIND_RESTRICT:
 631                return btf__align_of(btf, t->type);
 632        case BTF_KIND_ARRAY:
 633                return btf__align_of(btf, btf_array(t)->type);
 634        case BTF_KIND_STRUCT:
 635        case BTF_KIND_UNION: {
 636                const struct btf_member *m = btf_members(t);
 637                __u16 vlen = btf_vlen(t);
 638                int i, max_align = 1, align;
 639
 640                for (i = 0; i < vlen; i++, m++) {
 641                        align = btf__align_of(btf, m->type);
 642                        if (align <= 0)
 643                                return align;
 644                        max_align = max(max_align, align);
 645                }
 646
 647                return max_align;
 648        }
 649        default:
 650                pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
 651                return 0;
 652        }
 653}
 654
 655int btf__resolve_type(const struct btf *btf, __u32 type_id)
 656{
 657        const struct btf_type *t;
 658        int depth = 0;
 659
 660        t = btf__type_by_id(btf, type_id);
 661        while (depth < MAX_RESOLVE_DEPTH &&
 662               !btf_type_is_void_or_null(t) &&
 663               (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
 664                type_id = t->type;
 665                t = btf__type_by_id(btf, type_id);
 666                depth++;
 667        }
 668
 669        if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
 670                return -EINVAL;
 671
 672        return type_id;
 673}
 674
 675__s32 btf__find_by_name(const struct btf *btf, const char *type_name)
 676{
 677        __u32 i, nr_types = btf__get_nr_types(btf);
 678
 679        if (!strcmp(type_name, "void"))
 680                return 0;
 681
 682        for (i = 1; i <= nr_types; i++) {
 683                const struct btf_type *t = btf__type_by_id(btf, i);
 684                const char *name = btf__name_by_offset(btf, t->name_off);
 685
 686                if (name && !strcmp(type_name, name))
 687                        return i;
 688        }
 689
 690        return -ENOENT;
 691}
 692
 693__s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
 694                             __u32 kind)
 695{
 696        __u32 i, nr_types = btf__get_nr_types(btf);
 697
 698        if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
 699                return 0;
 700
 701        for (i = 1; i <= nr_types; i++) {
 702                const struct btf_type *t = btf__type_by_id(btf, i);
 703                const char *name;
 704
 705                if (btf_kind(t) != kind)
 706                        continue;
 707                name = btf__name_by_offset(btf, t->name_off);
 708                if (name && !strcmp(type_name, name))
 709                        return i;
 710        }
 711
 712        return -ENOENT;
 713}
 714
 715static bool btf_is_modifiable(const struct btf *btf)
 716{
 717        return (void *)btf->hdr != btf->raw_data;
 718}
 719
 720void btf__free(struct btf *btf)
 721{
 722        if (IS_ERR_OR_NULL(btf))
 723                return;
 724
 725        if (btf->fd >= 0)
 726                close(btf->fd);
 727
 728        if (btf_is_modifiable(btf)) {
 729                /* if BTF was modified after loading, it will have a split
 730                 * in-memory representation for header, types, and strings
 731                 * sections, so we need to free all of them individually. It
 732                 * might still have a cached contiguous raw data present,
 733                 * which will be unconditionally freed below.
 734                 */
 735                free(btf->hdr);
 736                free(btf->types_data);
 737                strset__free(btf->strs_set);
 738        }
 739        free(btf->raw_data);
 740        free(btf->raw_data_swapped);
 741        free(btf->type_offs);
 742        free(btf);
 743}
 744
 745static struct btf *btf_new_empty(struct btf *base_btf)
 746{
 747        struct btf *btf;
 748
 749        btf = calloc(1, sizeof(*btf));
 750        if (!btf)
 751                return ERR_PTR(-ENOMEM);
 752
 753        btf->nr_types = 0;
 754        btf->start_id = 1;
 755        btf->start_str_off = 0;
 756        btf->fd = -1;
 757        btf->ptr_sz = sizeof(void *);
 758        btf->swapped_endian = false;
 759
 760        if (base_btf) {
 761                btf->base_btf = base_btf;
 762                btf->start_id = btf__get_nr_types(base_btf) + 1;
 763                btf->start_str_off = base_btf->hdr->str_len;
 764        }
 765
 766        /* +1 for empty string at offset 0 */
 767        btf->raw_size = sizeof(struct btf_header) + (base_btf ? 0 : 1);
 768        btf->raw_data = calloc(1, btf->raw_size);
 769        if (!btf->raw_data) {
 770                free(btf);
 771                return ERR_PTR(-ENOMEM);
 772        }
 773
 774        btf->hdr = btf->raw_data;
 775        btf->hdr->hdr_len = sizeof(struct btf_header);
 776        btf->hdr->magic = BTF_MAGIC;
 777        btf->hdr->version = BTF_VERSION;
 778
 779        btf->types_data = btf->raw_data + btf->hdr->hdr_len;
 780        btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
 781        btf->hdr->str_len = base_btf ? 0 : 1; /* empty string at offset 0 */
 782
 783        return btf;
 784}
 785
 786struct btf *btf__new_empty(void)
 787{
 788        return btf_new_empty(NULL);
 789}
 790
 791struct btf *btf__new_empty_split(struct btf *base_btf)
 792{
 793        return btf_new_empty(base_btf);
 794}
 795
 796static struct btf *btf_new(const void *data, __u32 size, struct btf *base_btf)
 797{
 798        struct btf *btf;
 799        int err;
 800
 801        btf = calloc(1, sizeof(struct btf));
 802        if (!btf)
 803                return ERR_PTR(-ENOMEM);
 804
 805        btf->nr_types = 0;
 806        btf->start_id = 1;
 807        btf->start_str_off = 0;
 808
 809        if (base_btf) {
 810                btf->base_btf = base_btf;
 811                btf->start_id = btf__get_nr_types(base_btf) + 1;
 812                btf->start_str_off = base_btf->hdr->str_len;
 813        }
 814
 815        btf->raw_data = malloc(size);
 816        if (!btf->raw_data) {
 817                err = -ENOMEM;
 818                goto done;
 819        }
 820        memcpy(btf->raw_data, data, size);
 821        btf->raw_size = size;
 822
 823        btf->hdr = btf->raw_data;
 824        err = btf_parse_hdr(btf);
 825        if (err)
 826                goto done;
 827
 828        btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
 829        btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
 830
 831        err = btf_parse_str_sec(btf);
 832        err = err ?: btf_parse_type_sec(btf);
 833        if (err)
 834                goto done;
 835
 836        btf->fd = -1;
 837
 838done:
 839        if (err) {
 840                btf__free(btf);
 841                return ERR_PTR(err);
 842        }
 843
 844        return btf;
 845}
 846
 847struct btf *btf__new(const void *data, __u32 size)
 848{
 849        return btf_new(data, size, NULL);
 850}
 851
 852static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,
 853                                 struct btf_ext **btf_ext)
 854{
 855        Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
 856        int err = 0, fd = -1, idx = 0;
 857        struct btf *btf = NULL;
 858        Elf_Scn *scn = NULL;
 859        Elf *elf = NULL;
 860        GElf_Ehdr ehdr;
 861        size_t shstrndx;
 862
 863        if (elf_version(EV_CURRENT) == EV_NONE) {
 864                pr_warn("failed to init libelf for %s\n", path);
 865                return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
 866        }
 867
 868        fd = open(path, O_RDONLY);
 869        if (fd < 0) {
 870                err = -errno;
 871                pr_warn("failed to open %s: %s\n", path, strerror(errno));
 872                return ERR_PTR(err);
 873        }
 874
 875        err = -LIBBPF_ERRNO__FORMAT;
 876
 877        elf = elf_begin(fd, ELF_C_READ, NULL);
 878        if (!elf) {
 879                pr_warn("failed to open %s as ELF file\n", path);
 880                goto done;
 881        }
 882        if (!gelf_getehdr(elf, &ehdr)) {
 883                pr_warn("failed to get EHDR from %s\n", path);
 884                goto done;
 885        }
 886
 887        if (elf_getshdrstrndx(elf, &shstrndx)) {
 888                pr_warn("failed to get section names section index for %s\n",
 889                        path);
 890                goto done;
 891        }
 892
 893        if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
 894                pr_warn("failed to get e_shstrndx from %s\n", path);
 895                goto done;
 896        }
 897
 898        while ((scn = elf_nextscn(elf, scn)) != NULL) {
 899                GElf_Shdr sh;
 900                char *name;
 901
 902                idx++;
 903                if (gelf_getshdr(scn, &sh) != &sh) {
 904                        pr_warn("failed to get section(%d) header from %s\n",
 905                                idx, path);
 906                        goto done;
 907                }
 908                name = elf_strptr(elf, shstrndx, sh.sh_name);
 909                if (!name) {
 910                        pr_warn("failed to get section(%d) name from %s\n",
 911                                idx, path);
 912                        goto done;
 913                }
 914                if (strcmp(name, BTF_ELF_SEC) == 0) {
 915                        btf_data = elf_getdata(scn, 0);
 916                        if (!btf_data) {
 917                                pr_warn("failed to get section(%d, %s) data from %s\n",
 918                                        idx, name, path);
 919                                goto done;
 920                        }
 921                        continue;
 922                } else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
 923                        btf_ext_data = elf_getdata(scn, 0);
 924                        if (!btf_ext_data) {
 925                                pr_warn("failed to get section(%d, %s) data from %s\n",
 926                                        idx, name, path);
 927                                goto done;
 928                        }
 929                        continue;
 930                }
 931        }
 932
 933        err = 0;
 934
 935        if (!btf_data) {
 936                err = -ENOENT;
 937                goto done;
 938        }
 939        btf = btf_new(btf_data->d_buf, btf_data->d_size, base_btf);
 940        if (IS_ERR(btf))
 941                goto done;
 942
 943        switch (gelf_getclass(elf)) {
 944        case ELFCLASS32:
 945                btf__set_pointer_size(btf, 4);
 946                break;
 947        case ELFCLASS64:
 948                btf__set_pointer_size(btf, 8);
 949                break;
 950        default:
 951                pr_warn("failed to get ELF class (bitness) for %s\n", path);
 952                break;
 953        }
 954
 955        if (btf_ext && btf_ext_data) {
 956                *btf_ext = btf_ext__new(btf_ext_data->d_buf,
 957                                        btf_ext_data->d_size);
 958                if (IS_ERR(*btf_ext))
 959                        goto done;
 960        } else if (btf_ext) {
 961                *btf_ext = NULL;
 962        }
 963done:
 964        if (elf)
 965                elf_end(elf);
 966        close(fd);
 967
 968        if (err)
 969                return ERR_PTR(err);
 970        /*
 971         * btf is always parsed before btf_ext, so no need to clean up
 972         * btf_ext, if btf loading failed
 973         */
 974        if (IS_ERR(btf))
 975                return btf;
 976        if (btf_ext && IS_ERR(*btf_ext)) {
 977                btf__free(btf);
 978                err = PTR_ERR(*btf_ext);
 979                return ERR_PTR(err);
 980        }
 981        return btf;
 982}
 983
 984struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
 985{
 986        return btf_parse_elf(path, NULL, btf_ext);
 987}
 988
 989struct btf *btf__parse_elf_split(const char *path, struct btf *base_btf)
 990{
 991        return btf_parse_elf(path, base_btf, NULL);
 992}
 993
 994static struct btf *btf_parse_raw(const char *path, struct btf *base_btf)
 995{
 996        struct btf *btf = NULL;
 997        void *data = NULL;
 998        FILE *f = NULL;
 999        __u16 magic;
1000        int err = 0;
1001        long sz;
1002
1003        f = fopen(path, "rb");
1004        if (!f) {
1005                err = -errno;
1006                goto err_out;
1007        }
1008
1009        /* check BTF magic */
1010        if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
1011                err = -EIO;
1012                goto err_out;
1013        }
1014        if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
1015                /* definitely not a raw BTF */
1016                err = -EPROTO;
1017                goto err_out;
1018        }
1019
1020        /* get file size */
1021        if (fseek(f, 0, SEEK_END)) {
1022                err = -errno;
1023                goto err_out;
1024        }
1025        sz = ftell(f);
1026        if (sz < 0) {
1027                err = -errno;
1028                goto err_out;
1029        }
1030        /* rewind to the start */
1031        if (fseek(f, 0, SEEK_SET)) {
1032                err = -errno;
1033                goto err_out;
1034        }
1035
1036        /* pre-alloc memory and read all of BTF data */
1037        data = malloc(sz);
1038        if (!data) {
1039                err = -ENOMEM;
1040                goto err_out;
1041        }
1042        if (fread(data, 1, sz, f) < sz) {
1043                err = -EIO;
1044                goto err_out;
1045        }
1046
1047        /* finally parse BTF data */
1048        btf = btf_new(data, sz, base_btf);
1049
1050err_out:
1051        free(data);
1052        if (f)
1053                fclose(f);
1054        return err ? ERR_PTR(err) : btf;
1055}
1056
1057struct btf *btf__parse_raw(const char *path)
1058{
1059        return btf_parse_raw(path, NULL);
1060}
1061
1062struct btf *btf__parse_raw_split(const char *path, struct btf *base_btf)
1063{
1064        return btf_parse_raw(path, base_btf);
1065}
1066
1067static struct btf *btf_parse(const char *path, struct btf *base_btf, struct btf_ext **btf_ext)
1068{
1069        struct btf *btf;
1070
1071        if (btf_ext)
1072                *btf_ext = NULL;
1073
1074        btf = btf_parse_raw(path, base_btf);
1075        if (!IS_ERR(btf) || PTR_ERR(btf) != -EPROTO)
1076                return btf;
1077
1078        return btf_parse_elf(path, base_btf, btf_ext);
1079}
1080
1081struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
1082{
1083        return btf_parse(path, NULL, btf_ext);
1084}
1085
1086struct btf *btf__parse_split(const char *path, struct btf *base_btf)
1087{
1088        return btf_parse(path, base_btf, NULL);
1089}
1090
1091static int compare_vsi_off(const void *_a, const void *_b)
1092{
1093        const struct btf_var_secinfo *a = _a;
1094        const struct btf_var_secinfo *b = _b;
1095
1096        return a->offset - b->offset;
1097}
1098
1099static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
1100                             struct btf_type *t)
1101{
1102        __u32 size = 0, off = 0, i, vars = btf_vlen(t);
1103        const char *name = btf__name_by_offset(btf, t->name_off);
1104        const struct btf_type *t_var;
1105        struct btf_var_secinfo *vsi;
1106        const struct btf_var *var;
1107        int ret;
1108
1109        if (!name) {
1110                pr_debug("No name found in string section for DATASEC kind.\n");
1111                return -ENOENT;
1112        }
1113
1114        /* .extern datasec size and var offsets were set correctly during
1115         * extern collection step, so just skip straight to sorting variables
1116         */
1117        if (t->size)
1118                goto sort_vars;
1119
1120        ret = bpf_object__section_size(obj, name, &size);
1121        if (ret || !size || (t->size && t->size != size)) {
1122                pr_debug("Invalid size for section %s: %u bytes\n", name, size);
1123                return -ENOENT;
1124        }
1125
1126        t->size = size;
1127
1128        for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
1129                t_var = btf__type_by_id(btf, vsi->type);
1130                var = btf_var(t_var);
1131
1132                if (!btf_is_var(t_var)) {
1133                        pr_debug("Non-VAR type seen in section %s\n", name);
1134                        return -EINVAL;
1135                }
1136
1137                if (var->linkage == BTF_VAR_STATIC)
1138                        continue;
1139
1140                name = btf__name_by_offset(btf, t_var->name_off);
1141                if (!name) {
1142                        pr_debug("No name found in string section for VAR kind\n");
1143                        return -ENOENT;
1144                }
1145
1146                ret = bpf_object__variable_offset(obj, name, &off);
1147                if (ret) {
1148                        pr_debug("No offset found in symbol table for VAR %s\n",
1149                                 name);
1150                        return -ENOENT;
1151                }
1152
1153                vsi->offset = off;
1154        }
1155
1156sort_vars:
1157        qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
1158        return 0;
1159}
1160
1161int btf__finalize_data(struct bpf_object *obj, struct btf *btf)
1162{
1163        int err = 0;
1164        __u32 i;
1165
1166        for (i = 1; i <= btf->nr_types; i++) {
1167                struct btf_type *t = btf_type_by_id(btf, i);
1168
1169                /* Loader needs to fix up some of the things compiler
1170                 * couldn't get its hands on while emitting BTF. This
1171                 * is section size and global variable offset. We use
1172                 * the info from the ELF itself for this purpose.
1173                 */
1174                if (btf_is_datasec(t)) {
1175                        err = btf_fixup_datasec(obj, btf, t);
1176                        if (err)
1177                                break;
1178                }
1179        }
1180
1181        return err;
1182}
1183
1184static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1185
1186int btf__load(struct btf *btf)
1187{
1188        __u32 log_buf_size = 0, raw_size;
1189        char *log_buf = NULL;
1190        void *raw_data;
1191        int err = 0;
1192
1193        if (btf->fd >= 0)
1194                return -EEXIST;
1195
1196retry_load:
1197        if (log_buf_size) {
1198                log_buf = malloc(log_buf_size);
1199                if (!log_buf)
1200                        return -ENOMEM;
1201
1202                *log_buf = 0;
1203        }
1204
1205        raw_data = btf_get_raw_data(btf, &raw_size, false);
1206        if (!raw_data) {
1207                err = -ENOMEM;
1208                goto done;
1209        }
1210        /* cache native raw data representation */
1211        btf->raw_size = raw_size;
1212        btf->raw_data = raw_data;
1213
1214        btf->fd = bpf_load_btf(raw_data, raw_size, log_buf, log_buf_size, false);
1215        if (btf->fd < 0) {
1216                if (!log_buf || errno == ENOSPC) {
1217                        log_buf_size = max((__u32)BPF_LOG_BUF_SIZE,
1218                                           log_buf_size << 1);
1219                        free(log_buf);
1220                        goto retry_load;
1221                }
1222
1223                err = -errno;
1224                pr_warn("Error loading BTF: %s(%d)\n", strerror(errno), errno);
1225                if (*log_buf)
1226                        pr_warn("%s\n", log_buf);
1227                goto done;
1228        }
1229
1230done:
1231        free(log_buf);
1232        return err;
1233}
1234
1235int btf__fd(const struct btf *btf)
1236{
1237        return btf->fd;
1238}
1239
1240void btf__set_fd(struct btf *btf, int fd)
1241{
1242        btf->fd = fd;
1243}
1244
1245static const void *btf_strs_data(const struct btf *btf)
1246{
1247        return btf->strs_data ? btf->strs_data : strset__data(btf->strs_set);
1248}
1249
1250static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1251{
1252        struct btf_header *hdr = btf->hdr;
1253        struct btf_type *t;
1254        void *data, *p;
1255        __u32 data_sz;
1256        int i;
1257
1258        data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1259        if (data) {
1260                *size = btf->raw_size;
1261                return data;
1262        }
1263
1264        data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1265        data = calloc(1, data_sz);
1266        if (!data)
1267                return NULL;
1268        p = data;
1269
1270        memcpy(p, hdr, hdr->hdr_len);
1271        if (swap_endian)
1272                btf_bswap_hdr(p);
1273        p += hdr->hdr_len;
1274
1275        memcpy(p, btf->types_data, hdr->type_len);
1276        if (swap_endian) {
1277                for (i = 0; i < btf->nr_types; i++) {
1278                        t = p + btf->type_offs[i];
1279                        /* btf_bswap_type_rest() relies on native t->info, so
1280                         * we swap base type info after we swapped all the
1281                         * additional information
1282                         */
1283                        if (btf_bswap_type_rest(t))
1284                                goto err_out;
1285                        btf_bswap_type_base(t);
1286                }
1287        }
1288        p += hdr->type_len;
1289
1290        memcpy(p, btf_strs_data(btf), hdr->str_len);
1291        p += hdr->str_len;
1292
1293        *size = data_sz;
1294        return data;
1295err_out:
1296        free(data);
1297        return NULL;
1298}
1299
1300const void *btf__get_raw_data(const struct btf *btf_ro, __u32 *size)
1301{
1302        struct btf *btf = (struct btf *)btf_ro;
1303        __u32 data_sz;
1304        void *data;
1305
1306        data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1307        if (!data)
1308                return NULL;
1309
1310        btf->raw_size = data_sz;
1311        if (btf->swapped_endian)
1312                btf->raw_data_swapped = data;
1313        else
1314                btf->raw_data = data;
1315        *size = data_sz;
1316        return data;
1317}
1318
1319const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
1320{
1321        if (offset < btf->start_str_off)
1322                return btf__str_by_offset(btf->base_btf, offset);
1323        else if (offset - btf->start_str_off < btf->hdr->str_len)
1324                return btf_strs_data(btf) + (offset - btf->start_str_off);
1325        else
1326                return NULL;
1327}
1328
1329const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1330{
1331        return btf__str_by_offset(btf, offset);
1332}
1333
1334struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf)
1335{
1336        struct bpf_btf_info btf_info;
1337        __u32 len = sizeof(btf_info);
1338        __u32 last_size;
1339        struct btf *btf;
1340        void *ptr;
1341        int err;
1342
1343        /* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
1344         * let's start with a sane default - 4KiB here - and resize it only if
1345         * bpf_obj_get_info_by_fd() needs a bigger buffer.
1346         */
1347        last_size = 4096;
1348        ptr = malloc(last_size);
1349        if (!ptr)
1350                return ERR_PTR(-ENOMEM);
1351
1352        memset(&btf_info, 0, sizeof(btf_info));
1353        btf_info.btf = ptr_to_u64(ptr);
1354        btf_info.btf_size = last_size;
1355        err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1356
1357        if (!err && btf_info.btf_size > last_size) {
1358                void *temp_ptr;
1359
1360                last_size = btf_info.btf_size;
1361                temp_ptr = realloc(ptr, last_size);
1362                if (!temp_ptr) {
1363                        btf = ERR_PTR(-ENOMEM);
1364                        goto exit_free;
1365                }
1366                ptr = temp_ptr;
1367
1368                len = sizeof(btf_info);
1369                memset(&btf_info, 0, sizeof(btf_info));
1370                btf_info.btf = ptr_to_u64(ptr);
1371                btf_info.btf_size = last_size;
1372
1373                err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1374        }
1375
1376        if (err || btf_info.btf_size > last_size) {
1377                btf = err ? ERR_PTR(-errno) : ERR_PTR(-E2BIG);
1378                goto exit_free;
1379        }
1380
1381        btf = btf_new(ptr, btf_info.btf_size, base_btf);
1382
1383exit_free:
1384        free(ptr);
1385        return btf;
1386}
1387
1388int btf__get_from_id(__u32 id, struct btf **btf)
1389{
1390        struct btf *res;
1391        int btf_fd;
1392
1393        *btf = NULL;
1394        btf_fd = bpf_btf_get_fd_by_id(id);
1395        if (btf_fd < 0)
1396                return -errno;
1397
1398        res = btf_get_from_fd(btf_fd, NULL);
1399        close(btf_fd);
1400        if (IS_ERR(res))
1401                return PTR_ERR(res);
1402
1403        *btf = res;
1404        return 0;
1405}
1406
1407int btf__get_map_kv_tids(const struct btf *btf, const char *map_name,
1408                         __u32 expected_key_size, __u32 expected_value_size,
1409                         __u32 *key_type_id, __u32 *value_type_id)
1410{
1411        const struct btf_type *container_type;
1412        const struct btf_member *key, *value;
1413        const size_t max_name = 256;
1414        char container_name[max_name];
1415        __s64 key_size, value_size;
1416        __s32 container_id;
1417
1418        if (snprintf(container_name, max_name, "____btf_map_%s", map_name) ==
1419            max_name) {
1420                pr_warn("map:%s length of '____btf_map_%s' is too long\n",
1421                        map_name, map_name);
1422                return -EINVAL;
1423        }
1424
1425        container_id = btf__find_by_name(btf, container_name);
1426        if (container_id < 0) {
1427                pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
1428                         map_name, container_name);
1429                return container_id;
1430        }
1431
1432        container_type = btf__type_by_id(btf, container_id);
1433        if (!container_type) {
1434                pr_warn("map:%s cannot find BTF type for container_id:%u\n",
1435                        map_name, container_id);
1436                return -EINVAL;
1437        }
1438
1439        if (!btf_is_struct(container_type) || btf_vlen(container_type) < 2) {
1440                pr_warn("map:%s container_name:%s is an invalid container struct\n",
1441                        map_name, container_name);
1442                return -EINVAL;
1443        }
1444
1445        key = btf_members(container_type);
1446        value = key + 1;
1447
1448        key_size = btf__resolve_size(btf, key->type);
1449        if (key_size < 0) {
1450                pr_warn("map:%s invalid BTF key_type_size\n", map_name);
1451                return key_size;
1452        }
1453
1454        if (expected_key_size != key_size) {
1455                pr_warn("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
1456                        map_name, (__u32)key_size, expected_key_size);
1457                return -EINVAL;
1458        }
1459
1460        value_size = btf__resolve_size(btf, value->type);
1461        if (value_size < 0) {
1462                pr_warn("map:%s invalid BTF value_type_size\n", map_name);
1463                return value_size;
1464        }
1465
1466        if (expected_value_size != value_size) {
1467                pr_warn("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1468                        map_name, (__u32)value_size, expected_value_size);
1469                return -EINVAL;
1470        }
1471
1472        *key_type_id = key->type;
1473        *value_type_id = value->type;
1474
1475        return 0;
1476}
1477
1478static void btf_invalidate_raw_data(struct btf *btf)
1479{
1480        if (btf->raw_data) {
1481                free(btf->raw_data);
1482                btf->raw_data = NULL;
1483        }
1484        if (btf->raw_data_swapped) {
1485                free(btf->raw_data_swapped);
1486                btf->raw_data_swapped = NULL;
1487        }
1488}
1489
1490/* Ensure BTF is ready to be modified (by splitting into a three memory
1491 * regions for header, types, and strings). Also invalidate cached
1492 * raw_data, if any.
1493 */
1494static int btf_ensure_modifiable(struct btf *btf)
1495{
1496        void *hdr, *types;
1497        struct strset *set = NULL;
1498        int err = -ENOMEM;
1499
1500        if (btf_is_modifiable(btf)) {
1501                /* any BTF modification invalidates raw_data */
1502                btf_invalidate_raw_data(btf);
1503                return 0;
1504        }
1505
1506        /* split raw data into three memory regions */
1507        hdr = malloc(btf->hdr->hdr_len);
1508        types = malloc(btf->hdr->type_len);
1509        if (!hdr || !types)
1510                goto err_out;
1511
1512        memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1513        memcpy(types, btf->types_data, btf->hdr->type_len);
1514
1515        /* build lookup index for all strings */
1516        set = strset__new(BTF_MAX_STR_OFFSET, btf->strs_data, btf->hdr->str_len);
1517        if (IS_ERR(set)) {
1518                err = PTR_ERR(set);
1519                goto err_out;
1520        }
1521
1522        /* only when everything was successful, update internal state */
1523        btf->hdr = hdr;
1524        btf->types_data = types;
1525        btf->types_data_cap = btf->hdr->type_len;
1526        btf->strs_data = NULL;
1527        btf->strs_set = set;
1528        /* if BTF was created from scratch, all strings are guaranteed to be
1529         * unique and deduplicated
1530         */
1531        if (btf->hdr->str_len == 0)
1532                btf->strs_deduped = true;
1533        if (!btf->base_btf && btf->hdr->str_len == 1)
1534                btf->strs_deduped = true;
1535
1536        /* invalidate raw_data representation */
1537        btf_invalidate_raw_data(btf);
1538
1539        return 0;
1540
1541err_out:
1542        strset__free(set);
1543        free(hdr);
1544        free(types);
1545        return err;
1546}
1547
1548/* Find an offset in BTF string section that corresponds to a given string *s*.
1549 * Returns:
1550 *   - >0 offset into string section, if string is found;
1551 *   - -ENOENT, if string is not in the string section;
1552 *   - <0, on any other error.
1553 */
1554int btf__find_str(struct btf *btf, const char *s)
1555{
1556        int off;
1557
1558        if (btf->base_btf) {
1559                off = btf__find_str(btf->base_btf, s);
1560                if (off != -ENOENT)
1561                        return off;
1562        }
1563
1564        /* BTF needs to be in a modifiable state to build string lookup index */
1565        if (btf_ensure_modifiable(btf))
1566                return -ENOMEM;
1567
1568        off = strset__find_str(btf->strs_set, s);
1569        if (off < 0)
1570                return off;
1571
1572        return btf->start_str_off + off;
1573}
1574
1575/* Add a string s to the BTF string section.
1576 * Returns:
1577 *   - > 0 offset into string section, on success;
1578 *   - < 0, on error.
1579 */
1580int btf__add_str(struct btf *btf, const char *s)
1581{
1582        int off;
1583
1584        if (btf->base_btf) {
1585                off = btf__find_str(btf->base_btf, s);
1586                if (off != -ENOENT)
1587                        return off;
1588        }
1589
1590        if (btf_ensure_modifiable(btf))
1591                return -ENOMEM;
1592
1593        off = strset__add_str(btf->strs_set, s);
1594        if (off < 0)
1595                return off;
1596
1597        btf->hdr->str_len = strset__data_size(btf->strs_set);
1598
1599        return btf->start_str_off + off;
1600}
1601
1602static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1603{
1604        return libbpf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1605                              btf->hdr->type_len, UINT_MAX, add_sz);
1606}
1607
1608static void btf_type_inc_vlen(struct btf_type *t)
1609{
1610        t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1611}
1612
1613static int btf_commit_type(struct btf *btf, int data_sz)
1614{
1615        int err;
1616
1617        err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1618        if (err)
1619                return err;
1620
1621        btf->hdr->type_len += data_sz;
1622        btf->hdr->str_off += data_sz;
1623        btf->nr_types++;
1624        return btf->start_id + btf->nr_types - 1;
1625}
1626
1627struct btf_pipe {
1628        const struct btf *src;
1629        struct btf *dst;
1630};
1631
1632static int btf_rewrite_str(__u32 *str_off, void *ctx)
1633{
1634        struct btf_pipe *p = ctx;
1635        int off;
1636
1637        if (!*str_off) /* nothing to do for empty strings */
1638                return 0;
1639
1640        off = btf__add_str(p->dst, btf__str_by_offset(p->src, *str_off));
1641        if (off < 0)
1642                return off;
1643
1644        *str_off = off;
1645        return 0;
1646}
1647
1648int btf__add_type(struct btf *btf, const struct btf *src_btf, const struct btf_type *src_type)
1649{
1650        struct btf_pipe p = { .src = src_btf, .dst = btf };
1651        struct btf_type *t;
1652        int sz, err;
1653
1654        sz = btf_type_size(src_type);
1655        if (sz < 0)
1656                return sz;
1657
1658        /* deconstruct BTF, if necessary, and invalidate raw_data */
1659        if (btf_ensure_modifiable(btf))
1660                return -ENOMEM;
1661
1662        t = btf_add_type_mem(btf, sz);
1663        if (!t)
1664                return -ENOMEM;
1665
1666        memcpy(t, src_type, sz);
1667
1668        err = btf_type_visit_str_offs(t, btf_rewrite_str, &p);
1669        if (err)
1670                return err;
1671
1672        return btf_commit_type(btf, sz);
1673}
1674
1675/*
1676 * Append new BTF_KIND_INT type with:
1677 *   - *name* - non-empty, non-NULL type name;
1678 *   - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1679 *   - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1680 * Returns:
1681 *   - >0, type ID of newly added BTF type;
1682 *   - <0, on error.
1683 */
1684int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
1685{
1686        struct btf_type *t;
1687        int sz, name_off;
1688
1689        /* non-empty name */
1690        if (!name || !name[0])
1691                return -EINVAL;
1692        /* byte_sz must be power of 2 */
1693        if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
1694                return -EINVAL;
1695        if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
1696                return -EINVAL;
1697
1698        /* deconstruct BTF, if necessary, and invalidate raw_data */
1699        if (btf_ensure_modifiable(btf))
1700                return -ENOMEM;
1701
1702        sz = sizeof(struct btf_type) + sizeof(int);
1703        t = btf_add_type_mem(btf, sz);
1704        if (!t)
1705                return -ENOMEM;
1706
1707        /* if something goes wrong later, we might end up with an extra string,
1708         * but that shouldn't be a problem, because BTF can't be constructed
1709         * completely anyway and will most probably be just discarded
1710         */
1711        name_off = btf__add_str(btf, name);
1712        if (name_off < 0)
1713                return name_off;
1714
1715        t->name_off = name_off;
1716        t->info = btf_type_info(BTF_KIND_INT, 0, 0);
1717        t->size = byte_sz;
1718        /* set INT info, we don't allow setting legacy bit offset/size */
1719        *(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
1720
1721        return btf_commit_type(btf, sz);
1722}
1723
1724/*
1725 * Append new BTF_KIND_FLOAT type with:
1726 *   - *name* - non-empty, non-NULL type name;
1727 *   - *sz* - size of the type, in bytes;
1728 * Returns:
1729 *   - >0, type ID of newly added BTF type;
1730 *   - <0, on error.
1731 */
1732int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
1733{
1734        struct btf_type *t;
1735        int sz, name_off;
1736
1737        /* non-empty name */
1738        if (!name || !name[0])
1739                return -EINVAL;
1740
1741        /* byte_sz must be one of the explicitly allowed values */
1742        if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
1743            byte_sz != 16)
1744                return -EINVAL;
1745
1746        if (btf_ensure_modifiable(btf))
1747                return -ENOMEM;
1748
1749        sz = sizeof(struct btf_type);
1750        t = btf_add_type_mem(btf, sz);
1751        if (!t)
1752                return -ENOMEM;
1753
1754        name_off = btf__add_str(btf, name);
1755        if (name_off < 0)
1756                return name_off;
1757
1758        t->name_off = name_off;
1759        t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
1760        t->size = byte_sz;
1761
1762        return btf_commit_type(btf, sz);
1763}
1764
1765/* it's completely legal to append BTF types with type IDs pointing forward to
1766 * types that haven't been appended yet, so we only make sure that id looks
1767 * sane, we can't guarantee that ID will always be valid
1768 */
1769static int validate_type_id(int id)
1770{
1771        if (id < 0 || id > BTF_MAX_NR_TYPES)
1772                return -EINVAL;
1773        return 0;
1774}
1775
1776/* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
1777static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
1778{
1779        struct btf_type *t;
1780        int sz, name_off = 0;
1781
1782        if (validate_type_id(ref_type_id))
1783                return -EINVAL;
1784
1785        if (btf_ensure_modifiable(btf))
1786                return -ENOMEM;
1787
1788        sz = sizeof(struct btf_type);
1789        t = btf_add_type_mem(btf, sz);
1790        if (!t)
1791                return -ENOMEM;
1792
1793        if (name && name[0]) {
1794                name_off = btf__add_str(btf, name);
1795                if (name_off < 0)
1796                        return name_off;
1797        }
1798
1799        t->name_off = name_off;
1800        t->info = btf_type_info(kind, 0, 0);
1801        t->type = ref_type_id;
1802
1803        return btf_commit_type(btf, sz);
1804}
1805
1806/*
1807 * Append new BTF_KIND_PTR type with:
1808 *   - *ref_type_id* - referenced type ID, it might not exist yet;
1809 * Returns:
1810 *   - >0, type ID of newly added BTF type;
1811 *   - <0, on error.
1812 */
1813int btf__add_ptr(struct btf *btf, int ref_type_id)
1814{
1815        return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
1816}
1817
1818/*
1819 * Append new BTF_KIND_ARRAY type with:
1820 *   - *index_type_id* - type ID of the type describing array index;
1821 *   - *elem_type_id* - type ID of the type describing array element;
1822 *   - *nr_elems* - the size of the array;
1823 * Returns:
1824 *   - >0, type ID of newly added BTF type;
1825 *   - <0, on error.
1826 */
1827int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
1828{
1829        struct btf_type *t;
1830        struct btf_array *a;
1831        int sz;
1832
1833        if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
1834                return -EINVAL;
1835
1836        if (btf_ensure_modifiable(btf))
1837                return -ENOMEM;
1838
1839        sz = sizeof(struct btf_type) + sizeof(struct btf_array);
1840        t = btf_add_type_mem(btf, sz);
1841        if (!t)
1842                return -ENOMEM;
1843
1844        t->name_off = 0;
1845        t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
1846        t->size = 0;
1847
1848        a = btf_array(t);
1849        a->type = elem_type_id;
1850        a->index_type = index_type_id;
1851        a->nelems = nr_elems;
1852
1853        return btf_commit_type(btf, sz);
1854}
1855
1856/* generic STRUCT/UNION append function */
1857static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
1858{
1859        struct btf_type *t;
1860        int sz, name_off = 0;
1861
1862        if (btf_ensure_modifiable(btf))
1863                return -ENOMEM;
1864
1865        sz = sizeof(struct btf_type);
1866        t = btf_add_type_mem(btf, sz);
1867        if (!t)
1868                return -ENOMEM;
1869
1870        if (name && name[0]) {
1871                name_off = btf__add_str(btf, name);
1872                if (name_off < 0)
1873                        return name_off;
1874        }
1875
1876        /* start out with vlen=0 and no kflag; this will be adjusted when
1877         * adding each member
1878         */
1879        t->name_off = name_off;
1880        t->info = btf_type_info(kind, 0, 0);
1881        t->size = bytes_sz;
1882
1883        return btf_commit_type(btf, sz);
1884}
1885
1886/*
1887 * Append new BTF_KIND_STRUCT type with:
1888 *   - *name* - name of the struct, can be NULL or empty for anonymous structs;
1889 *   - *byte_sz* - size of the struct, in bytes;
1890 *
1891 * Struct initially has no fields in it. Fields can be added by
1892 * btf__add_field() right after btf__add_struct() succeeds.
1893 *
1894 * Returns:
1895 *   - >0, type ID of newly added BTF type;
1896 *   - <0, on error.
1897 */
1898int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
1899{
1900        return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
1901}
1902
1903/*
1904 * Append new BTF_KIND_UNION type with:
1905 *   - *name* - name of the union, can be NULL or empty for anonymous union;
1906 *   - *byte_sz* - size of the union, in bytes;
1907 *
1908 * Union initially has no fields in it. Fields can be added by
1909 * btf__add_field() right after btf__add_union() succeeds. All fields
1910 * should have *bit_offset* of 0.
1911 *
1912 * Returns:
1913 *   - >0, type ID of newly added BTF type;
1914 *   - <0, on error.
1915 */
1916int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
1917{
1918        return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
1919}
1920
1921static struct btf_type *btf_last_type(struct btf *btf)
1922{
1923        return btf_type_by_id(btf, btf__get_nr_types(btf));
1924}
1925
1926/*
1927 * Append new field for the current STRUCT/UNION type with:
1928 *   - *name* - name of the field, can be NULL or empty for anonymous field;
1929 *   - *type_id* - type ID for the type describing field type;
1930 *   - *bit_offset* - bit offset of the start of the field within struct/union;
1931 *   - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
1932 * Returns:
1933 *   -  0, on success;
1934 *   - <0, on error.
1935 */
1936int btf__add_field(struct btf *btf, const char *name, int type_id,
1937                   __u32 bit_offset, __u32 bit_size)
1938{
1939        struct btf_type *t;
1940        struct btf_member *m;
1941        bool is_bitfield;
1942        int sz, name_off = 0;
1943
1944        /* last type should be union/struct */
1945        if (btf->nr_types == 0)
1946                return -EINVAL;
1947        t = btf_last_type(btf);
1948        if (!btf_is_composite(t))
1949                return -EINVAL;
1950
1951        if (validate_type_id(type_id))
1952                return -EINVAL;
1953        /* best-effort bit field offset/size enforcement */
1954        is_bitfield = bit_size || (bit_offset % 8 != 0);
1955        if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
1956                return -EINVAL;
1957
1958        /* only offset 0 is allowed for unions */
1959        if (btf_is_union(t) && bit_offset)
1960                return -EINVAL;
1961
1962        /* decompose and invalidate raw data */
1963        if (btf_ensure_modifiable(btf))
1964                return -ENOMEM;
1965
1966        sz = sizeof(struct btf_member);
1967        m = btf_add_type_mem(btf, sz);
1968        if (!m)
1969                return -ENOMEM;
1970
1971        if (name && name[0]) {
1972                name_off = btf__add_str(btf, name);
1973                if (name_off < 0)
1974                        return name_off;
1975        }
1976
1977        m->name_off = name_off;
1978        m->type = type_id;
1979        m->offset = bit_offset | (bit_size << 24);
1980
1981        /* btf_add_type_mem can invalidate t pointer */
1982        t = btf_last_type(btf);
1983        /* update parent type's vlen and kflag */
1984        t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
1985
1986        btf->hdr->type_len += sz;
1987        btf->hdr->str_off += sz;
1988        return 0;
1989}
1990
1991/*
1992 * Append new BTF_KIND_ENUM type with:
1993 *   - *name* - name of the enum, can be NULL or empty for anonymous enums;
1994 *   - *byte_sz* - size of the enum, in bytes.
1995 *
1996 * Enum initially has no enum values in it (and corresponds to enum forward
1997 * declaration). Enumerator values can be added by btf__add_enum_value()
1998 * immediately after btf__add_enum() succeeds.
1999 *
2000 * Returns:
2001 *   - >0, type ID of newly added BTF type;
2002 *   - <0, on error.
2003 */
2004int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2005{
2006        struct btf_type *t;
2007        int sz, name_off = 0;
2008
2009        /* byte_sz must be power of 2 */
2010        if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2011                return -EINVAL;
2012
2013        if (btf_ensure_modifiable(btf))
2014                return -ENOMEM;
2015
2016        sz = sizeof(struct btf_type);
2017        t = btf_add_type_mem(btf, sz);
2018        if (!t)
2019                return -ENOMEM;
2020
2021        if (name && name[0]) {
2022                name_off = btf__add_str(btf, name);
2023                if (name_off < 0)
2024                        return name_off;
2025        }
2026
2027        /* start out with vlen=0; it will be adjusted when adding enum values */
2028        t->name_off = name_off;
2029        t->info = btf_type_info(BTF_KIND_ENUM, 0, 0);
2030        t->size = byte_sz;
2031
2032        return btf_commit_type(btf, sz);
2033}
2034
2035/*
2036 * Append new enum value for the current ENUM type with:
2037 *   - *name* - name of the enumerator value, can't be NULL or empty;
2038 *   - *value* - integer value corresponding to enum value *name*;
2039 * Returns:
2040 *   -  0, on success;
2041 *   - <0, on error.
2042 */
2043int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2044{
2045        struct btf_type *t;
2046        struct btf_enum *v;
2047        int sz, name_off;
2048
2049        /* last type should be BTF_KIND_ENUM */
2050        if (btf->nr_types == 0)
2051                return -EINVAL;
2052        t = btf_last_type(btf);
2053        if (!btf_is_enum(t))
2054                return -EINVAL;
2055
2056        /* non-empty name */
2057        if (!name || !name[0])
2058                return -EINVAL;
2059        if (value < INT_MIN || value > UINT_MAX)
2060                return -E2BIG;
2061
2062        /* decompose and invalidate raw data */
2063        if (btf_ensure_modifiable(btf))
2064                return -ENOMEM;
2065
2066        sz = sizeof(struct btf_enum);
2067        v = btf_add_type_mem(btf, sz);
2068        if (!v)
2069                return -ENOMEM;
2070
2071        name_off = btf__add_str(btf, name);
2072        if (name_off < 0)
2073                return name_off;
2074
2075        v->name_off = name_off;
2076        v->val = value;
2077
2078        /* update parent type's vlen */
2079        t = btf_last_type(btf);
2080        btf_type_inc_vlen(t);
2081
2082        btf->hdr->type_len += sz;
2083        btf->hdr->str_off += sz;
2084        return 0;
2085}
2086
2087/*
2088 * Append new BTF_KIND_FWD type with:
2089 *   - *name*, non-empty/non-NULL name;
2090 *   - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2091 *     BTF_FWD_UNION, or BTF_FWD_ENUM;
2092 * Returns:
2093 *   - >0, type ID of newly added BTF type;
2094 *   - <0, on error.
2095 */
2096int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2097{
2098        if (!name || !name[0])
2099                return -EINVAL;
2100
2101        switch (fwd_kind) {
2102        case BTF_FWD_STRUCT:
2103        case BTF_FWD_UNION: {
2104                struct btf_type *t;
2105                int id;
2106
2107                id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
2108                if (id <= 0)
2109                        return id;
2110                t = btf_type_by_id(btf, id);
2111                t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2112                return id;
2113        }
2114        case BTF_FWD_ENUM:
2115                /* enum forward in BTF currently is just an enum with no enum
2116                 * values; we also assume a standard 4-byte size for it
2117                 */
2118                return btf__add_enum(btf, name, sizeof(int));
2119        default:
2120                return -EINVAL;
2121        }
2122}
2123
2124/*
2125 * Append new BTF_KING_TYPEDEF type with:
2126 *   - *name*, non-empty/non-NULL name;
2127 *   - *ref_type_id* - referenced type ID, it might not exist yet;
2128 * Returns:
2129 *   - >0, type ID of newly added BTF type;
2130 *   - <0, on error.
2131 */
2132int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2133{
2134        if (!name || !name[0])
2135                return -EINVAL;
2136
2137        return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2138}
2139
2140/*
2141 * Append new BTF_KIND_VOLATILE type with:
2142 *   - *ref_type_id* - referenced type ID, it might not exist yet;
2143 * Returns:
2144 *   - >0, type ID of newly added BTF type;
2145 *   - <0, on error.
2146 */
2147int btf__add_volatile(struct btf *btf, int ref_type_id)
2148{
2149        return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2150}
2151
2152/*
2153 * Append new BTF_KIND_CONST type with:
2154 *   - *ref_type_id* - referenced type ID, it might not exist yet;
2155 * Returns:
2156 *   - >0, type ID of newly added BTF type;
2157 *   - <0, on error.
2158 */
2159int btf__add_const(struct btf *btf, int ref_type_id)
2160{
2161        return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2162}
2163
2164/*
2165 * Append new BTF_KIND_RESTRICT type with:
2166 *   - *ref_type_id* - referenced type ID, it might not exist yet;
2167 * Returns:
2168 *   - >0, type ID of newly added BTF type;
2169 *   - <0, on error.
2170 */
2171int btf__add_restrict(struct btf *btf, int ref_type_id)
2172{
2173        return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2174}
2175
2176/*
2177 * Append new BTF_KIND_FUNC type with:
2178 *   - *name*, non-empty/non-NULL name;
2179 *   - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2180 * Returns:
2181 *   - >0, type ID of newly added BTF type;
2182 *   - <0, on error.
2183 */
2184int btf__add_func(struct btf *btf, const char *name,
2185                  enum btf_func_linkage linkage, int proto_type_id)
2186{
2187        int id;
2188
2189        if (!name || !name[0])
2190                return -EINVAL;
2191        if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2192            linkage != BTF_FUNC_EXTERN)
2193                return -EINVAL;
2194
2195        id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2196        if (id > 0) {
2197                struct btf_type *t = btf_type_by_id(btf, id);
2198
2199                t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2200        }
2201        return id;
2202}
2203
2204/*
2205 * Append new BTF_KIND_FUNC_PROTO with:
2206 *   - *ret_type_id* - type ID for return result of a function.
2207 *
2208 * Function prototype initially has no arguments, but they can be added by
2209 * btf__add_func_param() one by one, immediately after
2210 * btf__add_func_proto() succeeded.
2211 *
2212 * Returns:
2213 *   - >0, type ID of newly added BTF type;
2214 *   - <0, on error.
2215 */
2216int btf__add_func_proto(struct btf *btf, int ret_type_id)
2217{
2218        struct btf_type *t;
2219        int sz;
2220
2221        if (validate_type_id(ret_type_id))
2222                return -EINVAL;
2223
2224        if (btf_ensure_modifiable(btf))
2225                return -ENOMEM;
2226
2227        sz = sizeof(struct btf_type);
2228        t = btf_add_type_mem(btf, sz);
2229        if (!t)
2230                return -ENOMEM;
2231
2232        /* start out with vlen=0; this will be adjusted when adding enum
2233         * values, if necessary
2234         */
2235        t->name_off = 0;
2236        t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2237        t->type = ret_type_id;
2238
2239        return btf_commit_type(btf, sz);
2240}
2241
2242/*
2243 * Append new function parameter for current FUNC_PROTO type with:
2244 *   - *name* - parameter name, can be NULL or empty;
2245 *   - *type_id* - type ID describing the type of the parameter.
2246 * Returns:
2247 *   -  0, on success;
2248 *   - <0, on error.
2249 */
2250int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2251{
2252        struct btf_type *t;
2253        struct btf_param *p;
2254        int sz, name_off = 0;
2255
2256        if (validate_type_id(type_id))
2257                return -EINVAL;
2258
2259        /* last type should be BTF_KIND_FUNC_PROTO */
2260        if (btf->nr_types == 0)
2261                return -EINVAL;
2262        t = btf_last_type(btf);
2263        if (!btf_is_func_proto(t))
2264                return -EINVAL;
2265
2266        /* decompose and invalidate raw data */
2267        if (btf_ensure_modifiable(btf))
2268                return -ENOMEM;
2269
2270        sz = sizeof(struct btf_param);
2271        p = btf_add_type_mem(btf, sz);
2272        if (!p)
2273                return -ENOMEM;
2274
2275        if (name && name[0]) {
2276                name_off = btf__add_str(btf, name);
2277                if (name_off < 0)
2278                        return name_off;
2279        }
2280
2281        p->name_off = name_off;
2282        p->type = type_id;
2283
2284        /* update parent type's vlen */
2285        t = btf_last_type(btf);
2286        btf_type_inc_vlen(t);
2287
2288        btf->hdr->type_len += sz;
2289        btf->hdr->str_off += sz;
2290        return 0;
2291}
2292
2293/*
2294 * Append new BTF_KIND_VAR type with:
2295 *   - *name* - non-empty/non-NULL name;
2296 *   - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2297 *     BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2298 *   - *type_id* - type ID of the type describing the type of the variable.
2299 * Returns:
2300 *   - >0, type ID of newly added BTF type;
2301 *   - <0, on error.
2302 */
2303int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2304{
2305        struct btf_type *t;
2306        struct btf_var *v;
2307        int sz, name_off;
2308
2309        /* non-empty name */
2310        if (!name || !name[0])
2311                return -EINVAL;
2312        if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2313            linkage != BTF_VAR_GLOBAL_EXTERN)
2314                return -EINVAL;
2315        if (validate_type_id(type_id))
2316                return -EINVAL;
2317
2318        /* deconstruct BTF, if necessary, and invalidate raw_data */
2319        if (btf_ensure_modifiable(btf))
2320                return -ENOMEM;
2321
2322        sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2323        t = btf_add_type_mem(btf, sz);
2324        if (!t)
2325                return -ENOMEM;
2326
2327        name_off = btf__add_str(btf, name);
2328        if (name_off < 0)
2329                return name_off;
2330
2331        t->name_off = name_off;
2332        t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2333        t->type = type_id;
2334
2335        v = btf_var(t);
2336        v->linkage = linkage;
2337
2338        return btf_commit_type(btf, sz);
2339}
2340
2341/*
2342 * Append new BTF_KIND_DATASEC type with:
2343 *   - *name* - non-empty/non-NULL name;
2344 *   - *byte_sz* - data section size, in bytes.
2345 *
2346 * Data section is initially empty. Variables info can be added with
2347 * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2348 *
2349 * Returns:
2350 *   - >0, type ID of newly added BTF type;
2351 *   - <0, on error.
2352 */
2353int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2354{
2355        struct btf_type *t;
2356        int sz, name_off;
2357
2358        /* non-empty name */
2359        if (!name || !name[0])
2360                return -EINVAL;
2361
2362        if (btf_ensure_modifiable(btf))
2363                return -ENOMEM;
2364
2365        sz = sizeof(struct btf_type);
2366        t = btf_add_type_mem(btf, sz);
2367        if (!t)
2368                return -ENOMEM;
2369
2370        name_off = btf__add_str(btf, name);
2371        if (name_off < 0)
2372                return name_off;
2373
2374        /* start with vlen=0, which will be update as var_secinfos are added */
2375        t->name_off = name_off;
2376        t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2377        t->size = byte_sz;
2378
2379        return btf_commit_type(btf, sz);
2380}
2381
2382/*
2383 * Append new data section variable information entry for current DATASEC type:
2384 *   - *var_type_id* - type ID, describing type of the variable;
2385 *   - *offset* - variable offset within data section, in bytes;
2386 *   - *byte_sz* - variable size, in bytes.
2387 *
2388 * Returns:
2389 *   -  0, on success;
2390 *   - <0, on error.
2391 */
2392int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2393{
2394        struct btf_type *t;
2395        struct btf_var_secinfo *v;
2396        int sz;
2397
2398        /* last type should be BTF_KIND_DATASEC */
2399        if (btf->nr_types == 0)
2400                return -EINVAL;
2401        t = btf_last_type(btf);
2402        if (!btf_is_datasec(t))
2403                return -EINVAL;
2404
2405        if (validate_type_id(var_type_id))
2406                return -EINVAL;
2407
2408        /* decompose and invalidate raw data */
2409        if (btf_ensure_modifiable(btf))
2410                return -ENOMEM;
2411
2412        sz = sizeof(struct btf_var_secinfo);
2413        v = btf_add_type_mem(btf, sz);
2414        if (!v)
2415                return -ENOMEM;
2416
2417        v->type = var_type_id;
2418        v->offset = offset;
2419        v->size = byte_sz;
2420
2421        /* update parent type's vlen */
2422        t = btf_last_type(btf);
2423        btf_type_inc_vlen(t);
2424
2425        btf->hdr->type_len += sz;
2426        btf->hdr->str_off += sz;
2427        return 0;
2428}
2429
2430struct btf_ext_sec_setup_param {
2431        __u32 off;
2432        __u32 len;
2433        __u32 min_rec_size;
2434        struct btf_ext_info *ext_info;
2435        const char *desc;
2436};
2437
2438static int btf_ext_setup_info(struct btf_ext *btf_ext,
2439                              struct btf_ext_sec_setup_param *ext_sec)
2440{
2441        const struct btf_ext_info_sec *sinfo;
2442        struct btf_ext_info *ext_info;
2443        __u32 info_left, record_size;
2444        /* The start of the info sec (including the __u32 record_size). */
2445        void *info;
2446
2447        if (ext_sec->len == 0)
2448                return 0;
2449
2450        if (ext_sec->off & 0x03) {
2451                pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2452                     ext_sec->desc);
2453                return -EINVAL;
2454        }
2455
2456        info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2457        info_left = ext_sec->len;
2458
2459        if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2460                pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2461                         ext_sec->desc, ext_sec->off, ext_sec->len);
2462                return -EINVAL;
2463        }
2464
2465        /* At least a record size */
2466        if (info_left < sizeof(__u32)) {
2467                pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2468                return -EINVAL;
2469        }
2470
2471        /* The record size needs to meet the minimum standard */
2472        record_size = *(__u32 *)info;
2473        if (record_size < ext_sec->min_rec_size ||
2474            record_size & 0x03) {
2475                pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2476                         ext_sec->desc, record_size);
2477                return -EINVAL;
2478        }
2479
2480        sinfo = info + sizeof(__u32);
2481        info_left -= sizeof(__u32);
2482
2483        /* If no records, return failure now so .BTF.ext won't be used. */
2484        if (!info_left) {
2485                pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
2486                return -EINVAL;
2487        }
2488
2489        while (info_left) {
2490                unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2491                __u64 total_record_size;
2492                __u32 num_records;
2493
2494                if (info_left < sec_hdrlen) {
2495                        pr_debug("%s section header is not found in .BTF.ext\n",
2496                             ext_sec->desc);
2497                        return -EINVAL;
2498                }
2499
2500                num_records = sinfo->num_info;
2501                if (num_records == 0) {
2502                        pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2503                             ext_sec->desc);
2504                        return -EINVAL;
2505                }
2506
2507                total_record_size = sec_hdrlen +
2508                                    (__u64)num_records * record_size;
2509                if (info_left < total_record_size) {
2510                        pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2511                             ext_sec->desc);
2512                        return -EINVAL;
2513                }
2514
2515                info_left -= total_record_size;
2516                sinfo = (void *)sinfo + total_record_size;
2517        }
2518
2519        ext_info = ext_sec->ext_info;
2520        ext_info->len = ext_sec->len - sizeof(__u32);
2521        ext_info->rec_size = record_size;
2522        ext_info->info = info + sizeof(__u32);
2523
2524        return 0;
2525}
2526
2527static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
2528{
2529        struct btf_ext_sec_setup_param param = {
2530                .off = btf_ext->hdr->func_info_off,
2531                .len = btf_ext->hdr->func_info_len,
2532                .min_rec_size = sizeof(struct bpf_func_info_min),
2533                .ext_info = &btf_ext->func_info,
2534                .desc = "func_info"
2535        };
2536
2537        return btf_ext_setup_info(btf_ext, &param);
2538}
2539
2540static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
2541{
2542        struct btf_ext_sec_setup_param param = {
2543                .off = btf_ext->hdr->line_info_off,
2544                .len = btf_ext->hdr->line_info_len,
2545                .min_rec_size = sizeof(struct bpf_line_info_min),
2546                .ext_info = &btf_ext->line_info,
2547                .desc = "line_info",
2548        };
2549
2550        return btf_ext_setup_info(btf_ext, &param);
2551}
2552
2553static int btf_ext_setup_core_relos(struct btf_ext *btf_ext)
2554{
2555        struct btf_ext_sec_setup_param param = {
2556                .off = btf_ext->hdr->core_relo_off,
2557                .len = btf_ext->hdr->core_relo_len,
2558                .min_rec_size = sizeof(struct bpf_core_relo),
2559                .ext_info = &btf_ext->core_relo_info,
2560                .desc = "core_relo",
2561        };
2562
2563        return btf_ext_setup_info(btf_ext, &param);
2564}
2565
2566static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
2567{
2568        const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
2569
2570        if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
2571            data_size < hdr->hdr_len) {
2572                pr_debug("BTF.ext header not found");
2573                return -EINVAL;
2574        }
2575
2576        if (hdr->magic == bswap_16(BTF_MAGIC)) {
2577                pr_warn("BTF.ext in non-native endianness is not supported\n");
2578                return -ENOTSUP;
2579        } else if (hdr->magic != BTF_MAGIC) {
2580                pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
2581                return -EINVAL;
2582        }
2583
2584        if (hdr->version != BTF_VERSION) {
2585                pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
2586                return -ENOTSUP;
2587        }
2588
2589        if (hdr->flags) {
2590                pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
2591                return -ENOTSUP;
2592        }
2593
2594        if (data_size == hdr->hdr_len) {
2595                pr_debug("BTF.ext has no data\n");
2596                return -EINVAL;
2597        }
2598
2599        return 0;
2600}
2601
2602void btf_ext__free(struct btf_ext *btf_ext)
2603{
2604        if (IS_ERR_OR_NULL(btf_ext))
2605                return;
2606        free(btf_ext->data);
2607        free(btf_ext);
2608}
2609
2610struct btf_ext *btf_ext__new(__u8 *data, __u32 size)
2611{
2612        struct btf_ext *btf_ext;
2613        int err;
2614
2615        err = btf_ext_parse_hdr(data, size);
2616        if (err)
2617                return ERR_PTR(err);
2618
2619        btf_ext = calloc(1, sizeof(struct btf_ext));
2620        if (!btf_ext)
2621                return ERR_PTR(-ENOMEM);
2622
2623        btf_ext->data_size = size;
2624        btf_ext->data = malloc(size);
2625        if (!btf_ext->data) {
2626                err = -ENOMEM;
2627                goto done;
2628        }
2629        memcpy(btf_ext->data, data, size);
2630
2631        if (btf_ext->hdr->hdr_len <
2632            offsetofend(struct btf_ext_header, line_info_len))
2633                goto done;
2634        err = btf_ext_setup_func_info(btf_ext);
2635        if (err)
2636                goto done;
2637
2638        err = btf_ext_setup_line_info(btf_ext);
2639        if (err)
2640                goto done;
2641
2642        if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
2643                goto done;
2644        err = btf_ext_setup_core_relos(btf_ext);
2645        if (err)
2646                goto done;
2647
2648done:
2649        if (err) {
2650                btf_ext__free(btf_ext);
2651                return ERR_PTR(err);
2652        }
2653
2654        return btf_ext;
2655}
2656
2657const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
2658{
2659        *size = btf_ext->data_size;
2660        return btf_ext->data;
2661}
2662
2663static int btf_ext_reloc_info(const struct btf *btf,
2664                              const struct btf_ext_info *ext_info,
2665                              const char *sec_name, __u32 insns_cnt,
2666                              void **info, __u32 *cnt)
2667{
2668        __u32 sec_hdrlen = sizeof(struct btf_ext_info_sec);
2669        __u32 i, record_size, existing_len, records_len;
2670        struct btf_ext_info_sec *sinfo;
2671        const char *info_sec_name;
2672        __u64 remain_len;
2673        void *data;
2674
2675        record_size = ext_info->rec_size;
2676        sinfo = ext_info->info;
2677        remain_len = ext_info->len;
2678        while (remain_len > 0) {
2679                records_len = sinfo->num_info * record_size;
2680                info_sec_name = btf__name_by_offset(btf, sinfo->sec_name_off);
2681                if (strcmp(info_sec_name, sec_name)) {
2682                        remain_len -= sec_hdrlen + records_len;
2683                        sinfo = (void *)sinfo + sec_hdrlen + records_len;
2684                        continue;
2685                }
2686
2687                existing_len = (*cnt) * record_size;
2688                data = realloc(*info, existing_len + records_len);
2689                if (!data)
2690                        return -ENOMEM;
2691
2692                memcpy(data + existing_len, sinfo->data, records_len);
2693                /* adjust insn_off only, the rest data will be passed
2694                 * to the kernel.
2695                 */
2696                for (i = 0; i < sinfo->num_info; i++) {
2697                        __u32 *insn_off;
2698
2699                        insn_off = data + existing_len + (i * record_size);
2700                        *insn_off = *insn_off / sizeof(struct bpf_insn) +
2701                                insns_cnt;
2702                }
2703                *info = data;
2704                *cnt += sinfo->num_info;
2705                return 0;
2706        }
2707
2708        return -ENOENT;
2709}
2710
2711int btf_ext__reloc_func_info(const struct btf *btf,
2712                             const struct btf_ext *btf_ext,
2713                             const char *sec_name, __u32 insns_cnt,
2714                             void **func_info, __u32 *cnt)
2715{
2716        return btf_ext_reloc_info(btf, &btf_ext->func_info, sec_name,
2717                                  insns_cnt, func_info, cnt);
2718}
2719
2720int btf_ext__reloc_line_info(const struct btf *btf,
2721                             const struct btf_ext *btf_ext,
2722                             const char *sec_name, __u32 insns_cnt,
2723                             void **line_info, __u32 *cnt)
2724{
2725        return btf_ext_reloc_info(btf, &btf_ext->line_info, sec_name,
2726                                  insns_cnt, line_info, cnt);
2727}
2728
2729__u32 btf_ext__func_info_rec_size(const struct btf_ext *btf_ext)
2730{
2731        return btf_ext->func_info.rec_size;
2732}
2733
2734__u32 btf_ext__line_info_rec_size(const struct btf_ext *btf_ext)
2735{
2736        return btf_ext->line_info.rec_size;
2737}
2738
2739struct btf_dedup;
2740
2741static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
2742                                       const struct btf_dedup_opts *opts);
2743static void btf_dedup_free(struct btf_dedup *d);
2744static int btf_dedup_prep(struct btf_dedup *d);
2745static int btf_dedup_strings(struct btf_dedup *d);
2746static int btf_dedup_prim_types(struct btf_dedup *d);
2747static int btf_dedup_struct_types(struct btf_dedup *d);
2748static int btf_dedup_ref_types(struct btf_dedup *d);
2749static int btf_dedup_compact_types(struct btf_dedup *d);
2750static int btf_dedup_remap_types(struct btf_dedup *d);
2751
2752/*
2753 * Deduplicate BTF types and strings.
2754 *
2755 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
2756 * section with all BTF type descriptors and string data. It overwrites that
2757 * memory in-place with deduplicated types and strings without any loss of
2758 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
2759 * is provided, all the strings referenced from .BTF.ext section are honored
2760 * and updated to point to the right offsets after deduplication.
2761 *
2762 * If function returns with error, type/string data might be garbled and should
2763 * be discarded.
2764 *
2765 * More verbose and detailed description of both problem btf_dedup is solving,
2766 * as well as solution could be found at:
2767 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
2768 *
2769 * Problem description and justification
2770 * =====================================
2771 *
2772 * BTF type information is typically emitted either as a result of conversion
2773 * from DWARF to BTF or directly by compiler. In both cases, each compilation
2774 * unit contains information about a subset of all the types that are used
2775 * in an application. These subsets are frequently overlapping and contain a lot
2776 * of duplicated information when later concatenated together into a single
2777 * binary. This algorithm ensures that each unique type is represented by single
2778 * BTF type descriptor, greatly reducing resulting size of BTF data.
2779 *
2780 * Compilation unit isolation and subsequent duplication of data is not the only
2781 * problem. The same type hierarchy (e.g., struct and all the type that struct
2782 * references) in different compilation units can be represented in BTF to
2783 * various degrees of completeness (or, rather, incompleteness) due to
2784 * struct/union forward declarations.
2785 *
2786 * Let's take a look at an example, that we'll use to better understand the
2787 * problem (and solution). Suppose we have two compilation units, each using
2788 * same `struct S`, but each of them having incomplete type information about
2789 * struct's fields:
2790 *
2791 * // CU #1:
2792 * struct S;
2793 * struct A {
2794 *      int a;
2795 *      struct A* self;
2796 *      struct S* parent;
2797 * };
2798 * struct B;
2799 * struct S {
2800 *      struct A* a_ptr;
2801 *      struct B* b_ptr;
2802 * };
2803 *
2804 * // CU #2:
2805 * struct S;
2806 * struct A;
2807 * struct B {
2808 *      int b;
2809 *      struct B* self;
2810 *      struct S* parent;
2811 * };
2812 * struct S {
2813 *      struct A* a_ptr;
2814 *      struct B* b_ptr;
2815 * };
2816 *
2817 * In case of CU #1, BTF data will know only that `struct B` exist (but no
2818 * more), but will know the complete type information about `struct A`. While
2819 * for CU #2, it will know full type information about `struct B`, but will
2820 * only know about forward declaration of `struct A` (in BTF terms, it will
2821 * have `BTF_KIND_FWD` type descriptor with name `B`).
2822 *
2823 * This compilation unit isolation means that it's possible that there is no
2824 * single CU with complete type information describing structs `S`, `A`, and
2825 * `B`. Also, we might get tons of duplicated and redundant type information.
2826 *
2827 * Additional complication we need to keep in mind comes from the fact that
2828 * types, in general, can form graphs containing cycles, not just DAGs.
2829 *
2830 * While algorithm does deduplication, it also merges and resolves type
2831 * information (unless disabled throught `struct btf_opts`), whenever possible.
2832 * E.g., in the example above with two compilation units having partial type
2833 * information for structs `A` and `B`, the output of algorithm will emit
2834 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
2835 * (as well as type information for `int` and pointers), as if they were defined
2836 * in a single compilation unit as:
2837 *
2838 * struct A {
2839 *      int a;
2840 *      struct A* self;
2841 *      struct S* parent;
2842 * };
2843 * struct B {
2844 *      int b;
2845 *      struct B* self;
2846 *      struct S* parent;
2847 * };
2848 * struct S {
2849 *      struct A* a_ptr;
2850 *      struct B* b_ptr;
2851 * };
2852 *
2853 * Algorithm summary
2854 * =================
2855 *
2856 * Algorithm completes its work in 6 separate passes:
2857 *
2858 * 1. Strings deduplication.
2859 * 2. Primitive types deduplication (int, enum, fwd).
2860 * 3. Struct/union types deduplication.
2861 * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
2862 *    protos, and const/volatile/restrict modifiers).
2863 * 5. Types compaction.
2864 * 6. Types remapping.
2865 *
2866 * Algorithm determines canonical type descriptor, which is a single
2867 * representative type for each truly unique type. This canonical type is the
2868 * one that will go into final deduplicated BTF type information. For
2869 * struct/unions, it is also the type that algorithm will merge additional type
2870 * information into (while resolving FWDs), as it discovers it from data in
2871 * other CUs. Each input BTF type eventually gets either mapped to itself, if
2872 * that type is canonical, or to some other type, if that type is equivalent
2873 * and was chosen as canonical representative. This mapping is stored in
2874 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
2875 * FWD type got resolved to.
2876 *
2877 * To facilitate fast discovery of canonical types, we also maintain canonical
2878 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
2879 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
2880 * that match that signature. With sufficiently good choice of type signature
2881 * hashing function, we can limit number of canonical types for each unique type
2882 * signature to a very small number, allowing to find canonical type for any
2883 * duplicated type very quickly.
2884 *
2885 * Struct/union deduplication is the most critical part and algorithm for
2886 * deduplicating structs/unions is described in greater details in comments for
2887 * `btf_dedup_is_equiv` function.
2888 */
2889int btf__dedup(struct btf *btf, struct btf_ext *btf_ext,
2890               const struct btf_dedup_opts *opts)
2891{
2892        struct btf_dedup *d = btf_dedup_new(btf, btf_ext, opts);
2893        int err;
2894
2895        if (IS_ERR(d)) {
2896                pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
2897                return -EINVAL;
2898        }
2899
2900        if (btf_ensure_modifiable(btf))
2901                return -ENOMEM;
2902
2903        err = btf_dedup_prep(d);
2904        if (err) {
2905                pr_debug("btf_dedup_prep failed:%d\n", err);
2906                goto done;
2907        }
2908        err = btf_dedup_strings(d);
2909        if (err < 0) {
2910                pr_debug("btf_dedup_strings failed:%d\n", err);
2911                goto done;
2912        }
2913        err = btf_dedup_prim_types(d);
2914        if (err < 0) {
2915                pr_debug("btf_dedup_prim_types failed:%d\n", err);
2916                goto done;
2917        }
2918        err = btf_dedup_struct_types(d);
2919        if (err < 0) {
2920                pr_debug("btf_dedup_struct_types failed:%d\n", err);
2921                goto done;
2922        }
2923        err = btf_dedup_ref_types(d);
2924        if (err < 0) {
2925                pr_debug("btf_dedup_ref_types failed:%d\n", err);
2926                goto done;
2927        }
2928        err = btf_dedup_compact_types(d);
2929        if (err < 0) {
2930                pr_debug("btf_dedup_compact_types failed:%d\n", err);
2931                goto done;
2932        }
2933        err = btf_dedup_remap_types(d);
2934        if (err < 0) {
2935                pr_debug("btf_dedup_remap_types failed:%d\n", err);
2936                goto done;
2937        }
2938
2939done:
2940        btf_dedup_free(d);
2941        return err;
2942}
2943
2944#define BTF_UNPROCESSED_ID ((__u32)-1)
2945#define BTF_IN_PROGRESS_ID ((__u32)-2)
2946
2947struct btf_dedup {
2948        /* .BTF section to be deduped in-place */
2949        struct btf *btf;
2950        /*
2951         * Optional .BTF.ext section. When provided, any strings referenced
2952         * from it will be taken into account when deduping strings
2953         */
2954        struct btf_ext *btf_ext;
2955        /*
2956         * This is a map from any type's signature hash to a list of possible
2957         * canonical representative type candidates. Hash collisions are
2958         * ignored, so even types of various kinds can share same list of
2959         * candidates, which is fine because we rely on subsequent
2960         * btf_xxx_equal() checks to authoritatively verify type equality.
2961         */
2962        struct hashmap *dedup_table;
2963        /* Canonical types map */
2964        __u32 *map;
2965        /* Hypothetical mapping, used during type graph equivalence checks */
2966        __u32 *hypot_map;
2967        __u32 *hypot_list;
2968        size_t hypot_cnt;
2969        size_t hypot_cap;
2970        /* Whether hypothetical mapping, if successful, would need to adjust
2971         * already canonicalized types (due to a new forward declaration to
2972         * concrete type resolution). In such case, during split BTF dedup
2973         * candidate type would still be considered as different, because base
2974         * BTF is considered to be immutable.
2975         */
2976        bool hypot_adjust_canon;
2977        /* Various option modifying behavior of algorithm */
2978        struct btf_dedup_opts opts;
2979        /* temporary strings deduplication state */
2980        struct strset *strs_set;
2981};
2982
2983static long hash_combine(long h, long value)
2984{
2985        return h * 31 + value;
2986}
2987
2988#define for_each_dedup_cand(d, node, hash) \
2989        hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
2990
2991static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
2992{
2993        return hashmap__append(d->dedup_table,
2994                               (void *)hash, (void *)(long)type_id);
2995}
2996
2997static int btf_dedup_hypot_map_add(struct btf_dedup *d,
2998                                   __u32 from_id, __u32 to_id)
2999{
3000        if (d->hypot_cnt == d->hypot_cap) {
3001                __u32 *new_list;
3002
3003                d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3004                new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3005                if (!new_list)
3006                        return -ENOMEM;
3007                d->hypot_list = new_list;
3008        }
3009        d->hypot_list[d->hypot_cnt++] = from_id;
3010        d->hypot_map[from_id] = to_id;
3011        return 0;
3012}
3013
3014static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3015{
3016        int i;
3017
3018        for (i = 0; i < d->hypot_cnt; i++)
3019                d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3020        d->hypot_cnt = 0;
3021        d->hypot_adjust_canon = false;
3022}
3023
3024static void btf_dedup_free(struct btf_dedup *d)
3025{
3026        hashmap__free(d->dedup_table);
3027        d->dedup_table = NULL;
3028
3029        free(d->map);
3030        d->map = NULL;
3031
3032        free(d->hypot_map);
3033        d->hypot_map = NULL;
3034
3035        free(d->hypot_list);
3036        d->hypot_list = NULL;
3037
3038        free(d);
3039}
3040
3041static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
3042{
3043        return (size_t)key;
3044}
3045
3046static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
3047{
3048        return 0;
3049}
3050
3051static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
3052{
3053        return k1 == k2;
3054}
3055
3056static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
3057                                       const struct btf_dedup_opts *opts)
3058{
3059        struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3060        hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3061        int i, err = 0, type_cnt;
3062
3063        if (!d)
3064                return ERR_PTR(-ENOMEM);
3065
3066        d->opts.dont_resolve_fwds = opts && opts->dont_resolve_fwds;
3067        /* dedup_table_size is now used only to force collisions in tests */
3068        if (opts && opts->dedup_table_size == 1)
3069                hash_fn = btf_dedup_collision_hash_fn;
3070
3071        d->btf = btf;
3072        d->btf_ext = btf_ext;
3073
3074        d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3075        if (IS_ERR(d->dedup_table)) {
3076                err = PTR_ERR(d->dedup_table);
3077                d->dedup_table = NULL;
3078                goto done;
3079        }
3080
3081        type_cnt = btf__get_nr_types(btf) + 1;
3082        d->map = malloc(sizeof(__u32) * type_cnt);
3083        if (!d->map) {
3084                err = -ENOMEM;
3085                goto done;
3086        }
3087        /* special BTF "void" type is made canonical immediately */
3088        d->map[0] = 0;
3089        for (i = 1; i < type_cnt; i++) {
3090                struct btf_type *t = btf_type_by_id(d->btf, i);
3091
3092                /* VAR and DATASEC are never deduped and are self-canonical */
3093                if (btf_is_var(t) || btf_is_datasec(t))
3094                        d->map[i] = i;
3095                else
3096                        d->map[i] = BTF_UNPROCESSED_ID;
3097        }
3098
3099        d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3100        if (!d->hypot_map) {
3101                err = -ENOMEM;
3102                goto done;
3103        }
3104        for (i = 0; i < type_cnt; i++)
3105                d->hypot_map[i] = BTF_UNPROCESSED_ID;
3106
3107done:
3108        if (err) {
3109                btf_dedup_free(d);
3110                return ERR_PTR(err);
3111        }
3112
3113        return d;
3114}
3115
3116/*
3117 * Iterate over all possible places in .BTF and .BTF.ext that can reference
3118 * string and pass pointer to it to a provided callback `fn`.
3119 */
3120static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3121{
3122        int i, r;
3123
3124        for (i = 0; i < d->btf->nr_types; i++) {
3125                struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3126
3127                r = btf_type_visit_str_offs(t, fn, ctx);
3128                if (r)
3129                        return r;
3130        }
3131
3132        if (!d->btf_ext)
3133                return 0;
3134
3135        r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3136        if (r)
3137                return r;
3138
3139        return 0;
3140}
3141
3142static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3143{
3144        struct btf_dedup *d = ctx;
3145        __u32 str_off = *str_off_ptr;
3146        const char *s;
3147        int off, err;
3148
3149        /* don't touch empty string or string in main BTF */
3150        if (str_off == 0 || str_off < d->btf->start_str_off)
3151                return 0;
3152
3153        s = btf__str_by_offset(d->btf, str_off);
3154        if (d->btf->base_btf) {
3155                err = btf__find_str(d->btf->base_btf, s);
3156                if (err >= 0) {
3157                        *str_off_ptr = err;
3158                        return 0;
3159                }
3160                if (err != -ENOENT)
3161                        return err;
3162        }
3163
3164        off = strset__add_str(d->strs_set, s);
3165        if (off < 0)
3166                return off;
3167
3168        *str_off_ptr = d->btf->start_str_off + off;
3169        return 0;
3170}
3171
3172/*
3173 * Dedup string and filter out those that are not referenced from either .BTF
3174 * or .BTF.ext (if provided) sections.
3175 *
3176 * This is done by building index of all strings in BTF's string section,
3177 * then iterating over all entities that can reference strings (e.g., type
3178 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3179 * strings as used. After that all used strings are deduped and compacted into
3180 * sequential blob of memory and new offsets are calculated. Then all the string
3181 * references are iterated again and rewritten using new offsets.
3182 */
3183static int btf_dedup_strings(struct btf_dedup *d)
3184{
3185        int err;
3186
3187        if (d->btf->strs_deduped)
3188                return 0;
3189
3190        d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3191        if (IS_ERR(d->strs_set)) {
3192                err = PTR_ERR(d->strs_set);
3193                goto err_out;
3194        }
3195
3196        if (!d->btf->base_btf) {
3197                /* insert empty string; we won't be looking it up during strings
3198                 * dedup, but it's good to have it for generic BTF string lookups
3199                 */
3200                err = strset__add_str(d->strs_set, "");
3201                if (err < 0)
3202                        goto err_out;
3203        }
3204
3205        /* remap string offsets */
3206        err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3207        if (err)
3208                goto err_out;
3209
3210        /* replace BTF string data and hash with deduped ones */
3211        strset__free(d->btf->strs_set);
3212        d->btf->hdr->str_len = strset__data_size(d->strs_set);
3213        d->btf->strs_set = d->strs_set;
3214        d->strs_set = NULL;
3215        d->btf->strs_deduped = true;
3216        return 0;
3217
3218err_out:
3219        strset__free(d->strs_set);
3220        d->strs_set = NULL;
3221
3222        return err;
3223}
3224
3225static long btf_hash_common(struct btf_type *t)
3226{
3227        long h;
3228
3229        h = hash_combine(0, t->name_off);
3230        h = hash_combine(h, t->info);
3231        h = hash_combine(h, t->size);
3232        return h;
3233}
3234
3235static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3236{
3237        return t1->name_off == t2->name_off &&
3238               t1->info == t2->info &&
3239               t1->size == t2->size;
3240}
3241
3242/* Calculate type signature hash of INT. */
3243static long btf_hash_int(struct btf_type *t)
3244{
3245        __u32 info = *(__u32 *)(t + 1);
3246        long h;
3247
3248        h = btf_hash_common(t);
3249        h = hash_combine(h, info);
3250        return h;
3251}
3252
3253/* Check structural equality of two INTs. */
3254static bool btf_equal_int(struct btf_type *t1, struct btf_type *t2)
3255{
3256        __u32 info1, info2;
3257
3258        if (!btf_equal_common(t1, t2))
3259                return false;
3260        info1 = *(__u32 *)(t1 + 1);
3261        info2 = *(__u32 *)(t2 + 1);
3262        return info1 == info2;
3263}
3264
3265/* Calculate type signature hash of ENUM. */
3266static long btf_hash_enum(struct btf_type *t)
3267{
3268        long h;
3269
3270        /* don't hash vlen and enum members to support enum fwd resolving */
3271        h = hash_combine(0, t->name_off);
3272        h = hash_combine(h, t->info & ~0xffff);
3273        h = hash_combine(h, t->size);
3274        return h;
3275}
3276
3277/* Check structural equality of two ENUMs. */
3278static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3279{
3280        const struct btf_enum *m1, *m2;
3281        __u16 vlen;
3282        int i;
3283
3284        if (!btf_equal_common(t1, t2))
3285                return false;
3286
3287        vlen = btf_vlen(t1);
3288        m1 = btf_enum(t1);
3289        m2 = btf_enum(t2);
3290        for (i = 0; i < vlen; i++) {
3291                if (m1->name_off != m2->name_off || m1->val != m2->val)
3292                        return false;
3293                m1++;
3294                m2++;
3295        }
3296        return true;
3297}
3298
3299static inline bool btf_is_enum_fwd(struct btf_type *t)
3300{
3301        return btf_is_enum(t) && btf_vlen(t) == 0;
3302}
3303
3304static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3305{
3306        if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3307                return btf_equal_enum(t1, t2);
3308        /* ignore vlen when comparing */
3309        return t1->name_off == t2->name_off &&
3310               (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
3311               t1->size == t2->size;
3312}
3313
3314/*
3315 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3316 * as referenced type IDs equivalence is established separately during type
3317 * graph equivalence check algorithm.
3318 */
3319static long btf_hash_struct(struct btf_type *t)
3320{
3321        const struct btf_member *member = btf_members(t);
3322        __u32 vlen = btf_vlen(t);
3323        long h = btf_hash_common(t);
3324        int i;
3325
3326        for (i = 0; i < vlen; i++) {
3327                h = hash_combine(h, member->name_off);
3328                h = hash_combine(h, member->offset);
3329                /* no hashing of referenced type ID, it can be unresolved yet */
3330                member++;
3331        }
3332        return h;
3333}
3334
3335/*
3336 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3337 * IDs. This check is performed during type graph equivalence check and
3338 * referenced types equivalence is checked separately.
3339 */
3340static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3341{
3342        const struct btf_member *m1, *m2;
3343        __u16 vlen;
3344        int i;
3345
3346        if (!btf_equal_common(t1, t2))
3347                return false;
3348
3349        vlen = btf_vlen(t1);
3350        m1 = btf_members(t1);
3351        m2 = btf_members(t2);
3352        for (i = 0; i < vlen; i++) {
3353                if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3354                        return false;
3355                m1++;
3356                m2++;
3357        }
3358        return true;
3359}
3360
3361/*
3362 * Calculate type signature hash of ARRAY, including referenced type IDs,
3363 * under assumption that they were already resolved to canonical type IDs and
3364 * are not going to change.
3365 */
3366static long btf_hash_array(struct btf_type *t)
3367{
3368        const struct btf_array *info = btf_array(t);
3369        long h = btf_hash_common(t);
3370
3371        h = hash_combine(h, info->type);
3372        h = hash_combine(h, info->index_type);
3373        h = hash_combine(h, info->nelems);
3374        return h;
3375}
3376
3377/*
3378 * Check exact equality of two ARRAYs, taking into account referenced
3379 * type IDs, under assumption that they were already resolved to canonical
3380 * type IDs and are not going to change.
3381 * This function is called during reference types deduplication to compare
3382 * ARRAY to potential canonical representative.
3383 */
3384static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3385{
3386        const struct btf_array *info1, *info2;
3387
3388        if (!btf_equal_common(t1, t2))
3389                return false;
3390
3391        info1 = btf_array(t1);
3392        info2 = btf_array(t2);
3393        return info1->type == info2->type &&
3394               info1->index_type == info2->index_type &&
3395               info1->nelems == info2->nelems;
3396}
3397
3398/*
3399 * Check structural compatibility of two ARRAYs, ignoring referenced type
3400 * IDs. This check is performed during type graph equivalence check and
3401 * referenced types equivalence is checked separately.
3402 */
3403static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
3404{
3405        if (!btf_equal_common(t1, t2))
3406                return false;
3407
3408        return btf_array(t1)->nelems == btf_array(t2)->nelems;
3409}
3410
3411/*
3412 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
3413 * under assumption that they were already resolved to canonical type IDs and
3414 * are not going to change.
3415 */
3416static long btf_hash_fnproto(struct btf_type *t)
3417{
3418        const struct btf_param *member = btf_params(t);
3419        __u16 vlen = btf_vlen(t);
3420        long h = btf_hash_common(t);
3421        int i;
3422
3423        for (i = 0; i < vlen; i++) {
3424                h = hash_combine(h, member->name_off);
3425                h = hash_combine(h, member->type);
3426                member++;
3427        }
3428        return h;
3429}
3430
3431/*
3432 * Check exact equality of two FUNC_PROTOs, taking into account referenced
3433 * type IDs, under assumption that they were already resolved to canonical
3434 * type IDs and are not going to change.
3435 * This function is called during reference types deduplication to compare
3436 * FUNC_PROTO to potential canonical representative.
3437 */
3438static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
3439{
3440        const struct btf_param *m1, *m2;
3441        __u16 vlen;
3442        int i;
3443
3444        if (!btf_equal_common(t1, t2))
3445                return false;
3446
3447        vlen = btf_vlen(t1);
3448        m1 = btf_params(t1);
3449        m2 = btf_params(t2);
3450        for (i = 0; i < vlen; i++) {
3451                if (m1->name_off != m2->name_off || m1->type != m2->type)
3452                        return false;
3453                m1++;
3454                m2++;
3455        }
3456        return true;
3457}
3458
3459/*
3460 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3461 * IDs. This check is performed during type graph equivalence check and
3462 * referenced types equivalence is checked separately.
3463 */
3464static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
3465{
3466        const struct btf_param *m1, *m2;
3467        __u16 vlen;
3468        int i;
3469
3470        /* skip return type ID */
3471        if (t1->name_off != t2->name_off || t1->info != t2->info)
3472                return false;
3473
3474        vlen = btf_vlen(t1);
3475        m1 = btf_params(t1);
3476        m2 = btf_params(t2);
3477        for (i = 0; i < vlen; i++) {
3478                if (m1->name_off != m2->name_off)
3479                        return false;
3480                m1++;
3481                m2++;
3482        }
3483        return true;
3484}
3485
3486/* Prepare split BTF for deduplication by calculating hashes of base BTF's
3487 * types and initializing the rest of the state (canonical type mapping) for
3488 * the fixed base BTF part.
3489 */
3490static int btf_dedup_prep(struct btf_dedup *d)
3491{
3492        struct btf_type *t;
3493        int type_id;
3494        long h;
3495
3496        if (!d->btf->base_btf)
3497                return 0;
3498
3499        for (type_id = 1; type_id < d->btf->start_id; type_id++) {
3500                t = btf_type_by_id(d->btf, type_id);
3501
3502                /* all base BTF types are self-canonical by definition */
3503                d->map[type_id] = type_id;
3504
3505                switch (btf_kind(t)) {
3506                case BTF_KIND_VAR:
3507                case BTF_KIND_DATASEC:
3508                        /* VAR and DATASEC are never hash/deduplicated */
3509                        continue;
3510                case BTF_KIND_CONST:
3511                case BTF_KIND_VOLATILE:
3512                case BTF_KIND_RESTRICT:
3513                case BTF_KIND_PTR:
3514                case BTF_KIND_FWD:
3515                case BTF_KIND_TYPEDEF:
3516                case BTF_KIND_FUNC:
3517                case BTF_KIND_FLOAT:
3518                        h = btf_hash_common(t);
3519                        break;
3520                case BTF_KIND_INT:
3521                        h = btf_hash_int(t);
3522                        break;
3523                case BTF_KIND_ENUM:
3524                        h = btf_hash_enum(t);
3525                        break;
3526                case BTF_KIND_STRUCT:
3527                case BTF_KIND_UNION:
3528                        h = btf_hash_struct(t);
3529                        break;
3530                case BTF_KIND_ARRAY:
3531                        h = btf_hash_array(t);
3532                        break;
3533                case BTF_KIND_FUNC_PROTO:
3534                        h = btf_hash_fnproto(t);
3535                        break;
3536                default:
3537                        pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
3538                        return -EINVAL;
3539                }
3540                if (btf_dedup_table_add(d, h, type_id))
3541                        return -ENOMEM;
3542        }
3543
3544        return 0;
3545}
3546
3547/*
3548 * Deduplicate primitive types, that can't reference other types, by calculating
3549 * their type signature hash and comparing them with any possible canonical
3550 * candidate. If no canonical candidate matches, type itself is marked as
3551 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
3552 */
3553static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
3554{
3555        struct btf_type *t = btf_type_by_id(d->btf, type_id);
3556        struct hashmap_entry *hash_entry;
3557        struct btf_type *cand;
3558        /* if we don't find equivalent type, then we are canonical */
3559        __u32 new_id = type_id;
3560        __u32 cand_id;
3561        long h;
3562
3563        switch (btf_kind(t)) {
3564        case BTF_KIND_CONST:
3565        case BTF_KIND_VOLATILE:
3566        case BTF_KIND_RESTRICT:
3567        case BTF_KIND_PTR:
3568        case BTF_KIND_TYPEDEF:
3569        case BTF_KIND_ARRAY:
3570        case BTF_KIND_STRUCT:
3571        case BTF_KIND_UNION:
3572        case BTF_KIND_FUNC:
3573        case BTF_KIND_FUNC_PROTO:
3574        case BTF_KIND_VAR:
3575        case BTF_KIND_DATASEC:
3576                return 0;
3577
3578        case BTF_KIND_INT:
3579                h = btf_hash_int(t);
3580                for_each_dedup_cand(d, hash_entry, h) {
3581                        cand_id = (__u32)(long)hash_entry->value;
3582                        cand = btf_type_by_id(d->btf, cand_id);
3583                        if (btf_equal_int(t, cand)) {
3584                                new_id = cand_id;
3585                                break;
3586                        }
3587                }
3588                break;
3589
3590        case BTF_KIND_ENUM:
3591                h = btf_hash_enum(t);
3592                for_each_dedup_cand(d, hash_entry, h) {
3593                        cand_id = (__u32)(long)hash_entry->value;
3594                        cand = btf_type_by_id(d->btf, cand_id);
3595                        if (btf_equal_enum(t, cand)) {
3596                                new_id = cand_id;
3597                                break;
3598                        }
3599                        if (d->opts.dont_resolve_fwds)
3600                                continue;
3601                        if (btf_compat_enum(t, cand)) {
3602                                if (btf_is_enum_fwd(t)) {
3603                                        /* resolve fwd to full enum */
3604                                        new_id = cand_id;
3605                                        break;
3606                                }
3607                                /* resolve canonical enum fwd to full enum */
3608                                d->map[cand_id] = type_id;
3609                        }
3610                }
3611                break;
3612
3613        case BTF_KIND_FWD:
3614        case BTF_KIND_FLOAT:
3615                h = btf_hash_common(t);
3616                for_each_dedup_cand(d, hash_entry, h) {
3617                        cand_id = (__u32)(long)hash_entry->value;
3618                        cand = btf_type_by_id(d->btf, cand_id);
3619                        if (btf_equal_common(t, cand)) {
3620                                new_id = cand_id;
3621                                break;
3622                        }
3623                }
3624                break;
3625
3626        default:
3627                return -EINVAL;
3628        }
3629
3630        d->map[type_id] = new_id;
3631        if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
3632                return -ENOMEM;
3633
3634        return 0;
3635}
3636
3637static int btf_dedup_prim_types(struct btf_dedup *d)
3638{
3639        int i, err;
3640
3641        for (i = 0; i < d->btf->nr_types; i++) {
3642                err = btf_dedup_prim_type(d, d->btf->start_id + i);
3643                if (err)
3644                        return err;
3645        }
3646        return 0;
3647}
3648
3649/*
3650 * Check whether type is already mapped into canonical one (could be to itself).
3651 */
3652static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
3653{
3654        return d->map[type_id] <= BTF_MAX_NR_TYPES;
3655}
3656
3657/*
3658 * Resolve type ID into its canonical type ID, if any; otherwise return original
3659 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
3660 * STRUCT/UNION link and resolve it into canonical type ID as well.
3661 */
3662static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
3663{
3664        while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3665                type_id = d->map[type_id];
3666        return type_id;
3667}
3668
3669/*
3670 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
3671 * type ID.
3672 */
3673static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
3674{
3675        __u32 orig_type_id = type_id;
3676
3677        if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3678                return type_id;
3679
3680        while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3681                type_id = d->map[type_id];
3682
3683        if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3684                return type_id;
3685
3686        return orig_type_id;
3687}
3688
3689
3690static inline __u16 btf_fwd_kind(struct btf_type *t)
3691{
3692        return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
3693}
3694
3695/* Check if given two types are identical ARRAY definitions */
3696static int btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
3697{
3698        struct btf_type *t1, *t2;
3699
3700        t1 = btf_type_by_id(d->btf, id1);
3701        t2 = btf_type_by_id(d->btf, id2);
3702        if (!btf_is_array(t1) || !btf_is_array(t2))
3703                return 0;
3704
3705        return btf_equal_array(t1, t2);
3706}
3707
3708/*
3709 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
3710 * call it "candidate graph" in this description for brevity) to a type graph
3711 * formed by (potential) canonical struct/union ("canonical graph" for brevity
3712 * here, though keep in mind that not all types in canonical graph are
3713 * necessarily canonical representatives themselves, some of them might be
3714 * duplicates or its uniqueness might not have been established yet).
3715 * Returns:
3716 *  - >0, if type graphs are equivalent;
3717 *  -  0, if not equivalent;
3718 *  - <0, on error.
3719 *
3720 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
3721 * equivalence of BTF types at each step. If at any point BTF types in candidate
3722 * and canonical graphs are not compatible structurally, whole graphs are
3723 * incompatible. If types are structurally equivalent (i.e., all information
3724 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
3725 * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
3726 * If a type references other types, then those referenced types are checked
3727 * for equivalence recursively.
3728 *
3729 * During DFS traversal, if we find that for current `canon_id` type we
3730 * already have some mapping in hypothetical map, we check for two possible
3731 * situations:
3732 *   - `canon_id` is mapped to exactly the same type as `cand_id`. This will
3733 *     happen when type graphs have cycles. In this case we assume those two
3734 *     types are equivalent.
3735 *   - `canon_id` is mapped to different type. This is contradiction in our
3736 *     hypothetical mapping, because same graph in canonical graph corresponds
3737 *     to two different types in candidate graph, which for equivalent type
3738 *     graphs shouldn't happen. This condition terminates equivalence check
3739 *     with negative result.
3740 *
3741 * If type graphs traversal exhausts types to check and find no contradiction,
3742 * then type graphs are equivalent.
3743 *
3744 * When checking types for equivalence, there is one special case: FWD types.
3745 * If FWD type resolution is allowed and one of the types (either from canonical
3746 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
3747 * flag) and their names match, hypothetical mapping is updated to point from
3748 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
3749 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
3750 *
3751 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
3752 * if there are two exactly named (or anonymous) structs/unions that are
3753 * compatible structurally, one of which has FWD field, while other is concrete
3754 * STRUCT/UNION, but according to C sources they are different structs/unions
3755 * that are referencing different types with the same name. This is extremely
3756 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
3757 * this logic is causing problems.
3758 *
3759 * Doing FWD resolution means that both candidate and/or canonical graphs can
3760 * consists of portions of the graph that come from multiple compilation units.
3761 * This is due to the fact that types within single compilation unit are always
3762 * deduplicated and FWDs are already resolved, if referenced struct/union
3763 * definiton is available. So, if we had unresolved FWD and found corresponding
3764 * STRUCT/UNION, they will be from different compilation units. This
3765 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
3766 * type graph will likely have at least two different BTF types that describe
3767 * same type (e.g., most probably there will be two different BTF types for the
3768 * same 'int' primitive type) and could even have "overlapping" parts of type
3769 * graph that describe same subset of types.
3770 *
3771 * This in turn means that our assumption that each type in canonical graph
3772 * must correspond to exactly one type in candidate graph might not hold
3773 * anymore and will make it harder to detect contradictions using hypothetical
3774 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
3775 * resolution only in canonical graph. FWDs in candidate graphs are never
3776 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
3777 * that can occur:
3778 *   - Both types in canonical and candidate graphs are FWDs. If they are
3779 *     structurally equivalent, then they can either be both resolved to the
3780 *     same STRUCT/UNION or not resolved at all. In both cases they are
3781 *     equivalent and there is no need to resolve FWD on candidate side.
3782 *   - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
3783 *     so nothing to resolve as well, algorithm will check equivalence anyway.
3784 *   - Type in canonical graph is FWD, while type in candidate is concrete
3785 *     STRUCT/UNION. In this case candidate graph comes from single compilation
3786 *     unit, so there is exactly one BTF type for each unique C type. After
3787 *     resolving FWD into STRUCT/UNION, there might be more than one BTF type
3788 *     in canonical graph mapping to single BTF type in candidate graph, but
3789 *     because hypothetical mapping maps from canonical to candidate types, it's
3790 *     alright, and we still maintain the property of having single `canon_id`
3791 *     mapping to single `cand_id` (there could be two different `canon_id`
3792 *     mapped to the same `cand_id`, but it's not contradictory).
3793 *   - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
3794 *     graph is FWD. In this case we are just going to check compatibility of
3795 *     STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
3796 *     assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
3797 *     a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
3798 *     turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
3799 *     canonical graph.
3800 */
3801static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
3802                              __u32 canon_id)
3803{
3804        struct btf_type *cand_type;
3805        struct btf_type *canon_type;
3806        __u32 hypot_type_id;
3807        __u16 cand_kind;
3808        __u16 canon_kind;
3809        int i, eq;
3810
3811        /* if both resolve to the same canonical, they must be equivalent */
3812        if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
3813                return 1;
3814
3815        canon_id = resolve_fwd_id(d, canon_id);
3816
3817        hypot_type_id = d->hypot_map[canon_id];
3818        if (hypot_type_id <= BTF_MAX_NR_TYPES) {
3819                /* In some cases compiler will generate different DWARF types
3820                 * for *identical* array type definitions and use them for
3821                 * different fields within the *same* struct. This breaks type
3822                 * equivalence check, which makes an assumption that candidate
3823                 * types sub-graph has a consistent and deduped-by-compiler
3824                 * types within a single CU. So work around that by explicitly
3825                 * allowing identical array types here.
3826                 */
3827                return hypot_type_id == cand_id ||
3828                       btf_dedup_identical_arrays(d, hypot_type_id, cand_id);
3829        }
3830
3831        if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
3832                return -ENOMEM;
3833
3834        cand_type = btf_type_by_id(d->btf, cand_id);
3835        canon_type = btf_type_by_id(d->btf, canon_id);
3836        cand_kind = btf_kind(cand_type);
3837        canon_kind = btf_kind(canon_type);
3838
3839        if (cand_type->name_off != canon_type->name_off)
3840                return 0;
3841
3842        /* FWD <--> STRUCT/UNION equivalence check, if enabled */
3843        if (!d->opts.dont_resolve_fwds
3844            && (cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
3845            && cand_kind != canon_kind) {
3846                __u16 real_kind;
3847                __u16 fwd_kind;
3848
3849                if (cand_kind == BTF_KIND_FWD) {
3850                        real_kind = canon_kind;
3851                        fwd_kind = btf_fwd_kind(cand_type);
3852                } else {
3853                        real_kind = cand_kind;
3854                        fwd_kind = btf_fwd_kind(canon_type);
3855                        /* we'd need to resolve base FWD to STRUCT/UNION */
3856                        if (fwd_kind == real_kind && canon_id < d->btf->start_id)
3857                                d->hypot_adjust_canon = true;
3858                }
3859                return fwd_kind == real_kind;
3860        }
3861
3862        if (cand_kind != canon_kind)
3863                return 0;
3864
3865        switch (cand_kind) {
3866        case BTF_KIND_INT:
3867                return btf_equal_int(cand_type, canon_type);
3868
3869        case BTF_KIND_ENUM:
3870                if (d->opts.dont_resolve_fwds)
3871                        return btf_equal_enum(cand_type, canon_type);
3872                else
3873                        return btf_compat_enum(cand_type, canon_type);
3874
3875        case BTF_KIND_FWD:
3876        case BTF_KIND_FLOAT:
3877                return btf_equal_common(cand_type, canon_type);
3878
3879        case BTF_KIND_CONST:
3880        case BTF_KIND_VOLATILE:
3881        case BTF_KIND_RESTRICT:
3882        case BTF_KIND_PTR:
3883        case BTF_KIND_TYPEDEF:
3884        case BTF_KIND_FUNC:
3885                if (cand_type->info != canon_type->info)
3886                        return 0;
3887                return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
3888
3889        case BTF_KIND_ARRAY: {
3890                const struct btf_array *cand_arr, *canon_arr;
3891
3892                if (!btf_compat_array(cand_type, canon_type))
3893                        return 0;
3894                cand_arr = btf_array(cand_type);
3895                canon_arr = btf_array(canon_type);
3896                eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
3897                if (eq <= 0)
3898                        return eq;
3899                return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
3900        }
3901
3902        case BTF_KIND_STRUCT:
3903        case BTF_KIND_UNION: {
3904                const struct btf_member *cand_m, *canon_m;
3905                __u16 vlen;
3906
3907                if (!btf_shallow_equal_struct(cand_type, canon_type))
3908                        return 0;
3909                vlen = btf_vlen(cand_type);
3910                cand_m = btf_members(cand_type);
3911                canon_m = btf_members(canon_type);
3912                for (i = 0; i < vlen; i++) {
3913                        eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
3914                        if (eq <= 0)
3915                                return eq;
3916                        cand_m++;
3917                        canon_m++;
3918                }
3919
3920                return 1;
3921        }
3922
3923        case BTF_KIND_FUNC_PROTO: {
3924                const struct btf_param *cand_p, *canon_p;
3925                __u16 vlen;
3926
3927                if (!btf_compat_fnproto(cand_type, canon_type))
3928                        return 0;
3929                eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
3930                if (eq <= 0)
3931                        return eq;
3932                vlen = btf_vlen(cand_type);
3933                cand_p = btf_params(cand_type);
3934                canon_p = btf_params(canon_type);
3935                for (i = 0; i < vlen; i++) {
3936                        eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
3937                        if (eq <= 0)
3938                                return eq;
3939                        cand_p++;
3940                        canon_p++;
3941                }
3942                return 1;
3943        }
3944
3945        default:
3946                return -EINVAL;
3947        }
3948        return 0;
3949}
3950
3951/*
3952 * Use hypothetical mapping, produced by successful type graph equivalence
3953 * check, to augment existing struct/union canonical mapping, where possible.
3954 *
3955 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
3956 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
3957 * it doesn't matter if FWD type was part of canonical graph or candidate one,
3958 * we are recording the mapping anyway. As opposed to carefulness required
3959 * for struct/union correspondence mapping (described below), for FWD resolution
3960 * it's not important, as by the time that FWD type (reference type) will be
3961 * deduplicated all structs/unions will be deduped already anyway.
3962 *
3963 * Recording STRUCT/UNION mapping is purely a performance optimization and is
3964 * not required for correctness. It needs to be done carefully to ensure that
3965 * struct/union from candidate's type graph is not mapped into corresponding
3966 * struct/union from canonical type graph that itself hasn't been resolved into
3967 * canonical representative. The only guarantee we have is that canonical
3968 * struct/union was determined as canonical and that won't change. But any
3969 * types referenced through that struct/union fields could have been not yet
3970 * resolved, so in case like that it's too early to establish any kind of
3971 * correspondence between structs/unions.
3972 *
3973 * No canonical correspondence is derived for primitive types (they are already
3974 * deduplicated completely already anyway) or reference types (they rely on
3975 * stability of struct/union canonical relationship for equivalence checks).
3976 */
3977static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
3978{
3979        __u32 canon_type_id, targ_type_id;
3980        __u16 t_kind, c_kind;
3981        __u32 t_id, c_id;
3982        int i;
3983
3984        for (i = 0; i < d->hypot_cnt; i++) {
3985                canon_type_id = d->hypot_list[i];
3986                targ_type_id = d->hypot_map[canon_type_id];
3987                t_id = resolve_type_id(d, targ_type_id);
3988                c_id = resolve_type_id(d, canon_type_id);
3989                t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
3990                c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
3991                /*
3992                 * Resolve FWD into STRUCT/UNION.
3993                 * It's ok to resolve FWD into STRUCT/UNION that's not yet
3994                 * mapped to canonical representative (as opposed to
3995                 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
3996                 * eventually that struct is going to be mapped and all resolved
3997                 * FWDs will automatically resolve to correct canonical
3998                 * representative. This will happen before ref type deduping,
3999                 * which critically depends on stability of these mapping. This
4000                 * stability is not a requirement for STRUCT/UNION equivalence
4001                 * checks, though.
4002                 */
4003
4004                /* if it's the split BTF case, we still need to point base FWD
4005                 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4006                 * will be resolved against base FWD. If we don't point base
4007                 * canonical FWD to the resolved STRUCT/UNION, then all the
4008                 * FWDs in split BTF won't be correctly resolved to a proper
4009                 * STRUCT/UNION.
4010                 */
4011                if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4012                        d->map[c_id] = t_id;
4013
4014                /* if graph equivalence determined that we'd need to adjust
4015                 * base canonical types, then we need to only point base FWDs
4016                 * to STRUCTs/UNIONs and do no more modifications. For all
4017                 * other purposes the type graphs were not equivalent.
4018                 */
4019                if (d->hypot_adjust_canon)
4020                        continue;
4021                
4022                if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4023                        d->map[t_id] = c_id;
4024
4025                if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4026                    c_kind != BTF_KIND_FWD &&
4027                    is_type_mapped(d, c_id) &&
4028                    !is_type_mapped(d, t_id)) {
4029                        /*
4030                         * as a perf optimization, we can map struct/union
4031                         * that's part of type graph we just verified for
4032                         * equivalence. We can do that for struct/union that has
4033                         * canonical representative only, though.
4034                         */
4035                        d->map[t_id] = c_id;
4036                }
4037        }
4038}
4039
4040/*
4041 * Deduplicate struct/union types.
4042 *
4043 * For each struct/union type its type signature hash is calculated, taking
4044 * into account type's name, size, number, order and names of fields, but
4045 * ignoring type ID's referenced from fields, because they might not be deduped
4046 * completely until after reference types deduplication phase. This type hash
4047 * is used to iterate over all potential canonical types, sharing same hash.
4048 * For each canonical candidate we check whether type graphs that they form
4049 * (through referenced types in fields and so on) are equivalent using algorithm
4050 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4051 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4052 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4053 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4054 * potentially map other structs/unions to their canonical representatives,
4055 * if such relationship hasn't yet been established. This speeds up algorithm
4056 * by eliminating some of the duplicate work.
4057 *
4058 * If no matching canonical representative was found, struct/union is marked
4059 * as canonical for itself and is added into btf_dedup->dedup_table hash map
4060 * for further look ups.
4061 */
4062static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4063{
4064        struct btf_type *cand_type, *t;
4065        struct hashmap_entry *hash_entry;
4066        /* if we don't find equivalent type, then we are canonical */
4067        __u32 new_id = type_id;
4068        __u16 kind;
4069        long h;
4070
4071        /* already deduped or is in process of deduping (loop detected) */
4072        if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4073                return 0;
4074
4075        t = btf_type_by_id(d->btf, type_id);
4076        kind = btf_kind(t);
4077
4078        if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4079                return 0;
4080
4081        h = btf_hash_struct(t);
4082        for_each_dedup_cand(d, hash_entry, h) {
4083                __u32 cand_id = (__u32)(long)hash_entry->value;
4084                int eq;
4085
4086                /*
4087                 * Even though btf_dedup_is_equiv() checks for
4088                 * btf_shallow_equal_struct() internally when checking two
4089                 * structs (unions) for equivalence, we need to guard here
4090                 * from picking matching FWD type as a dedup candidate.
4091                 * This can happen due to hash collision. In such case just
4092                 * relying on btf_dedup_is_equiv() would lead to potentially
4093                 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4094                 * FWD and compatible STRUCT/UNION are considered equivalent.
4095                 */
4096                cand_type = btf_type_by_id(d->btf, cand_id);
4097                if (!btf_shallow_equal_struct(t, cand_type))
4098                        continue;
4099
4100                btf_dedup_clear_hypot_map(d);
4101                eq = btf_dedup_is_equiv(d, type_id, cand_id);
4102                if (eq < 0)
4103                        return eq;
4104                if (!eq)
4105                        continue;
4106                btf_dedup_merge_hypot_map(d);
4107                if (d->hypot_adjust_canon) /* not really equivalent */
4108                        continue;
4109                new_id = cand_id;
4110                break;
4111        }
4112
4113        d->map[type_id] = new_id;
4114        if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4115                return -ENOMEM;
4116
4117        return 0;
4118}
4119
4120static int btf_dedup_struct_types(struct btf_dedup *d)
4121{
4122        int i, err;
4123
4124        for (i = 0; i < d->btf->nr_types; i++) {
4125                err = btf_dedup_struct_type(d, d->btf->start_id + i);
4126                if (err)
4127                        return err;
4128        }
4129        return 0;
4130}
4131
4132/*
4133 * Deduplicate reference type.
4134 *
4135 * Once all primitive and struct/union types got deduplicated, we can easily
4136 * deduplicate all other (reference) BTF types. This is done in two steps:
4137 *
4138 * 1. Resolve all referenced type IDs into their canonical type IDs. This
4139 * resolution can be done either immediately for primitive or struct/union types
4140 * (because they were deduped in previous two phases) or recursively for
4141 * reference types. Recursion will always terminate at either primitive or
4142 * struct/union type, at which point we can "unwind" chain of reference types
4143 * one by one. There is no danger of encountering cycles because in C type
4144 * system the only way to form type cycle is through struct/union, so any chain
4145 * of reference types, even those taking part in a type cycle, will inevitably
4146 * reach struct/union at some point.
4147 *
4148 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4149 * becomes "stable", in the sense that no further deduplication will cause
4150 * any changes to it. With that, it's now possible to calculate type's signature
4151 * hash (this time taking into account referenced type IDs) and loop over all
4152 * potential canonical representatives. If no match was found, current type
4153 * will become canonical representative of itself and will be added into
4154 * btf_dedup->dedup_table as another possible canonical representative.
4155 */
4156static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4157{
4158        struct hashmap_entry *hash_entry;
4159        __u32 new_id = type_id, cand_id;
4160        struct btf_type *t, *cand;
4161        /* if we don't find equivalent type, then we are representative type */
4162        int ref_type_id;
4163        long h;
4164
4165        if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4166                return -ELOOP;
4167        if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4168                return resolve_type_id(d, type_id);
4169
4170        t = btf_type_by_id(d->btf, type_id);
4171        d->map[type_id] = BTF_IN_PROGRESS_ID;
4172
4173        switch (btf_kind(t)) {
4174        case BTF_KIND_CONST:
4175        case BTF_KIND_VOLATILE:
4176        case BTF_KIND_RESTRICT:
4177        case BTF_KIND_PTR:
4178        case BTF_KIND_TYPEDEF:
4179        case BTF_KIND_FUNC:
4180                ref_type_id = btf_dedup_ref_type(d, t->type);
4181                if (ref_type_id < 0)
4182                        return ref_type_id;
4183                t->type = ref_type_id;
4184
4185                h = btf_hash_common(t);
4186                for_each_dedup_cand(d, hash_entry, h) {
4187                        cand_id = (__u32)(long)hash_entry->value;
4188                        cand = btf_type_by_id(d->btf, cand_id);
4189                        if (btf_equal_common(t, cand)) {
4190                                new_id = cand_id;
4191                                break;
4192                        }
4193                }
4194                break;
4195
4196        case BTF_KIND_ARRAY: {
4197                struct btf_array *info = btf_array(t);
4198
4199                ref_type_id = btf_dedup_ref_type(d, info->type);
4200                if (ref_type_id < 0)
4201                        return ref_type_id;
4202                info->type = ref_type_id;
4203
4204                ref_type_id = btf_dedup_ref_type(d, info->index_type);
4205                if (ref_type_id < 0)
4206                        return ref_type_id;
4207                info->index_type = ref_type_id;
4208
4209                h = btf_hash_array(t);
4210                for_each_dedup_cand(d, hash_entry, h) {
4211                        cand_id = (__u32)(long)hash_entry->value;
4212                        cand = btf_type_by_id(d->btf, cand_id);
4213                        if (btf_equal_array(t, cand)) {
4214                                new_id = cand_id;
4215                                break;
4216                        }
4217                }
4218                break;
4219        }
4220
4221        case BTF_KIND_FUNC_PROTO: {
4222                struct btf_param *param;
4223                __u16 vlen;
4224                int i;
4225
4226                ref_type_id = btf_dedup_ref_type(d, t->type);
4227                if (ref_type_id < 0)
4228                        return ref_type_id;
4229                t->type = ref_type_id;
4230
4231                vlen = btf_vlen(t);
4232                param = btf_params(t);
4233                for (i = 0; i < vlen; i++) {
4234                        ref_type_id = btf_dedup_ref_type(d, param->type);
4235                        if (ref_type_id < 0)
4236                                return ref_type_id;
4237                        param->type = ref_type_id;
4238                        param++;
4239                }
4240
4241                h = btf_hash_fnproto(t);
4242                for_each_dedup_cand(d, hash_entry, h) {
4243                        cand_id = (__u32)(long)hash_entry->value;
4244                        cand = btf_type_by_id(d->btf, cand_id);
4245                        if (btf_equal_fnproto(t, cand)) {
4246                                new_id = cand_id;
4247                                break;
4248                        }
4249                }
4250                break;
4251        }
4252
4253        default:
4254                return -EINVAL;
4255        }
4256
4257        d->map[type_id] = new_id;
4258        if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4259                return -ENOMEM;
4260
4261        return new_id;
4262}
4263
4264static int btf_dedup_ref_types(struct btf_dedup *d)
4265{
4266        int i, err;
4267
4268        for (i = 0; i < d->btf->nr_types; i++) {
4269                err = btf_dedup_ref_type(d, d->btf->start_id + i);
4270                if (err < 0)
4271                        return err;
4272        }
4273        /* we won't need d->dedup_table anymore */
4274        hashmap__free(d->dedup_table);
4275        d->dedup_table = NULL;
4276        return 0;
4277}
4278
4279/*
4280 * Compact types.
4281 *
4282 * After we established for each type its corresponding canonical representative
4283 * type, we now can eliminate types that are not canonical and leave only
4284 * canonical ones layed out sequentially in memory by copying them over
4285 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
4286 * a map from original type ID to a new compacted type ID, which will be used
4287 * during next phase to "fix up" type IDs, referenced from struct/union and
4288 * reference types.
4289 */
4290static int btf_dedup_compact_types(struct btf_dedup *d)
4291{
4292        __u32 *new_offs;
4293        __u32 next_type_id = d->btf->start_id;
4294        const struct btf_type *t;
4295        void *p;
4296        int i, id, len;
4297
4298        /* we are going to reuse hypot_map to store compaction remapping */
4299        d->hypot_map[0] = 0;
4300        /* base BTF types are not renumbered */
4301        for (id = 1; id < d->btf->start_id; id++)
4302                d->hypot_map[id] = id;
4303        for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
4304                d->hypot_map[id] = BTF_UNPROCESSED_ID;
4305
4306        p = d->btf->types_data;
4307
4308        for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
4309                if (d->map[id] != id)
4310                        continue;
4311
4312                t = btf__type_by_id(d->btf, id);
4313                len = btf_type_size(t);
4314                if (len < 0)
4315                        return len;
4316
4317                memmove(p, t, len);
4318                d->hypot_map[id] = next_type_id;
4319                d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
4320                p += len;
4321                next_type_id++;
4322        }
4323
4324        /* shrink struct btf's internal types index and update btf_header */
4325        d->btf->nr_types = next_type_id - d->btf->start_id;
4326        d->btf->type_offs_cap = d->btf->nr_types;
4327        d->btf->hdr->type_len = p - d->btf->types_data;
4328        new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
4329                                       sizeof(*new_offs));
4330        if (d->btf->type_offs_cap && !new_offs)
4331                return -ENOMEM;
4332        d->btf->type_offs = new_offs;
4333        d->btf->hdr->str_off = d->btf->hdr->type_len;
4334        d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
4335        return 0;
4336}
4337
4338/*
4339 * Figure out final (deduplicated and compacted) type ID for provided original
4340 * `type_id` by first resolving it into corresponding canonical type ID and
4341 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
4342 * which is populated during compaction phase.
4343 */
4344static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
4345{
4346        struct btf_dedup *d = ctx;
4347        __u32 resolved_type_id, new_type_id;
4348
4349        resolved_type_id = resolve_type_id(d, *type_id);
4350        new_type_id = d->hypot_map[resolved_type_id];
4351        if (new_type_id > BTF_MAX_NR_TYPES)
4352                return -EINVAL;
4353
4354        *type_id = new_type_id;
4355        return 0;
4356}
4357
4358/*
4359 * Remap referenced type IDs into deduped type IDs.
4360 *
4361 * After BTF types are deduplicated and compacted, their final type IDs may
4362 * differ from original ones. The map from original to a corresponding
4363 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
4364 * compaction phase. During remapping phase we are rewriting all type IDs
4365 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
4366 * their final deduped type IDs.
4367 */
4368static int btf_dedup_remap_types(struct btf_dedup *d)
4369{
4370        int i, r;
4371
4372        for (i = 0; i < d->btf->nr_types; i++) {
4373                struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
4374
4375                r = btf_type_visit_type_ids(t, btf_dedup_remap_type_id, d);
4376                if (r)
4377                        return r;
4378        }
4379
4380        if (!d->btf_ext)
4381                return 0;
4382
4383        r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
4384        if (r)
4385                return r;
4386
4387        return 0;
4388}
4389
4390/*
4391 * Probe few well-known locations for vmlinux kernel image and try to load BTF
4392 * data out of it to use for target BTF.
4393 */
4394struct btf *libbpf_find_kernel_btf(void)
4395{
4396        struct {
4397                const char *path_fmt;
4398                bool raw_btf;
4399        } locations[] = {
4400                /* try canonical vmlinux BTF through sysfs first */
4401                { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
4402                /* fall back to trying to find vmlinux ELF on disk otherwise */
4403                { "/boot/vmlinux-%1$s" },
4404                { "/lib/modules/%1$s/vmlinux-%1$s" },
4405                { "/lib/modules/%1$s/build/vmlinux" },
4406                { "/usr/lib/modules/%1$s/kernel/vmlinux" },
4407                { "/usr/lib/debug/boot/vmlinux-%1$s" },
4408                { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
4409                { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
4410        };
4411        char path[PATH_MAX + 1];
4412        struct utsname buf;
4413        struct btf *btf;
4414        int i;
4415
4416        uname(&buf);
4417
4418        for (i = 0; i < ARRAY_SIZE(locations); i++) {
4419                snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
4420
4421                if (access(path, R_OK))
4422                        continue;
4423
4424                if (locations[i].raw_btf)
4425                        btf = btf__parse_raw(path);
4426                else
4427                        btf = btf__parse_elf(path, NULL);
4428
4429                pr_debug("loading kernel BTF '%s': %ld\n",
4430                         path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
4431                if (IS_ERR(btf))
4432                        continue;
4433
4434                return btf;
4435        }
4436
4437        pr_warn("failed to find valid kernel BTF\n");
4438        return ERR_PTR(-ESRCH);
4439}
4440
4441int btf_type_visit_type_ids(struct btf_type *t, type_id_visit_fn visit, void *ctx)
4442{
4443        int i, n, err;
4444
4445        switch (btf_kind(t)) {
4446        case BTF_KIND_INT:
4447        case BTF_KIND_FLOAT:
4448        case BTF_KIND_ENUM:
4449                return 0;
4450
4451        case BTF_KIND_FWD:
4452        case BTF_KIND_CONST:
4453        case BTF_KIND_VOLATILE:
4454        case BTF_KIND_RESTRICT:
4455        case BTF_KIND_PTR:
4456        case BTF_KIND_TYPEDEF:
4457        case BTF_KIND_FUNC:
4458        case BTF_KIND_VAR:
4459                return visit(&t->type, ctx);
4460
4461        case BTF_KIND_ARRAY: {
4462                struct btf_array *a = btf_array(t);
4463
4464                err = visit(&a->type, ctx);
4465                err = err ?: visit(&a->index_type, ctx);
4466                return err;
4467        }
4468
4469        case BTF_KIND_STRUCT:
4470        case BTF_KIND_UNION: {
4471                struct btf_member *m = btf_members(t);
4472
4473                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4474                        err = visit(&m->type, ctx);
4475                        if (err)
4476                                return err;
4477                }
4478                return 0;
4479        }
4480
4481        case BTF_KIND_FUNC_PROTO: {
4482                struct btf_param *m = btf_params(t);
4483
4484                err = visit(&t->type, ctx);
4485                if (err)
4486                        return err;
4487                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4488                        err = visit(&m->type, ctx);
4489                        if (err)
4490                                return err;
4491                }
4492                return 0;
4493        }
4494
4495        case BTF_KIND_DATASEC: {
4496                struct btf_var_secinfo *m = btf_var_secinfos(t);
4497
4498                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4499                        err = visit(&m->type, ctx);
4500                        if (err)
4501                                return err;
4502                }
4503                return 0;
4504        }
4505
4506        default:
4507                return -EINVAL;
4508        }
4509}
4510
4511int btf_type_visit_str_offs(struct btf_type *t, str_off_visit_fn visit, void *ctx)
4512{
4513        int i, n, err;
4514
4515        err = visit(&t->name_off, ctx);
4516        if (err)
4517                return err;
4518
4519        switch (btf_kind(t)) {
4520        case BTF_KIND_STRUCT:
4521        case BTF_KIND_UNION: {
4522                struct btf_member *m = btf_members(t);
4523
4524                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4525                        err = visit(&m->name_off, ctx);
4526                        if (err)
4527                                return err;
4528                }
4529                break;
4530        }
4531        case BTF_KIND_ENUM: {
4532                struct btf_enum *m = btf_enum(t);
4533
4534                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4535                        err = visit(&m->name_off, ctx);
4536                        if (err)
4537                                return err;
4538                }
4539                break;
4540        }
4541        case BTF_KIND_FUNC_PROTO: {
4542                struct btf_param *m = btf_params(t);
4543
4544                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4545                        err = visit(&m->name_off, ctx);
4546                        if (err)
4547                                return err;
4548                }
4549                break;
4550        }
4551        default:
4552                break;
4553        }
4554
4555        return 0;
4556}
4557
4558int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
4559{
4560        const struct btf_ext_info *seg;
4561        struct btf_ext_info_sec *sec;
4562        int i, err;
4563
4564        seg = &btf_ext->func_info;
4565        for_each_btf_ext_sec(seg, sec) {
4566                struct bpf_func_info_min *rec;
4567
4568                for_each_btf_ext_rec(seg, sec, i, rec) {
4569                        err = visit(&rec->type_id, ctx);
4570                        if (err < 0)
4571                                return err;
4572                }
4573        }
4574
4575        seg = &btf_ext->core_relo_info;
4576        for_each_btf_ext_sec(seg, sec) {
4577                struct bpf_core_relo *rec;
4578
4579                for_each_btf_ext_rec(seg, sec, i, rec) {
4580                        err = visit(&rec->type_id, ctx);
4581                        if (err < 0)
4582                                return err;
4583                }
4584        }
4585
4586        return 0;
4587}
4588
4589int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
4590{
4591        const struct btf_ext_info *seg;
4592        struct btf_ext_info_sec *sec;
4593        int i, err;
4594
4595        seg = &btf_ext->func_info;
4596        for_each_btf_ext_sec(seg, sec) {
4597                err = visit(&sec->sec_name_off, ctx);
4598                if (err)
4599                        return err;
4600        }
4601
4602        seg = &btf_ext->line_info;
4603        for_each_btf_ext_sec(seg, sec) {
4604                struct bpf_line_info_min *rec;
4605
4606                err = visit(&sec->sec_name_off, ctx);
4607                if (err)
4608                        return err;
4609
4610                for_each_btf_ext_rec(seg, sec, i, rec) {
4611                        err = visit(&rec->file_name_off, ctx);
4612                        if (err)
4613                                return err;
4614                        err = visit(&rec->line_off, ctx);
4615                        if (err)
4616                                return err;
4617                }
4618        }
4619
4620        seg = &btf_ext->core_relo_info;
4621        for_each_btf_ext_sec(seg, sec) {
4622                struct bpf_core_relo *rec;
4623
4624                err = visit(&sec->sec_name_off, ctx);
4625                if (err)
4626                        return err;
4627
4628                for_each_btf_ext_rec(seg, sec, i, rec) {
4629                        err = visit(&rec->access_str_off, ctx);
4630                        if (err)
4631                                return err;
4632                }
4633        }
4634
4635        return 0;
4636}
4637