Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.15.
   1// SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
   2/* Copyright (c) 2018 Facebook */
   3
   4#include <endian.h>
   5#include <stdio.h>
   6#include <stdlib.h>
   7#include <string.h>
   8#include <fcntl.h>
   9#include <unistd.h>
  10#include <errno.h>
  11#include <sys/utsname.h>
  12#include <sys/param.h>
  13#include <sys/stat.h>
  14#include <linux/kernel.h>
  15#include <linux/err.h>
  16#include <linux/btf.h>
  17#include <gelf.h>
  18#include "btf.h"
  19#include "bpf.h"
  20#include "libbpf.h"
  21#include "libbpf_internal.h"
  22#include "hashmap.h"
  23
  24/* make sure libbpf doesn't use kernel-only integer typedefs */
  25#pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64
  26
  27#define BTF_MAX_NR_TYPES 0x7fffffffU
  28#define BTF_MAX_STR_OFFSET 0x7fffffffU
  29
  30static struct btf_type btf_void;
  31
  32struct btf {
  33	union {
  34		struct btf_header *hdr;
  35		void *data;
  36	};
  37	struct btf_type **types;
  38	const char *strings;
  39	void *nohdr_data;
  40	__u32 nr_types;
  41	__u32 types_size;
  42	__u32 data_size;
  43	int fd;
  44	int ptr_sz;
  45};
  46
  47static inline __u64 ptr_to_u64(const void *ptr)
  48{
  49	return (__u64) (unsigned long) ptr;
  50}
  51
  52static int btf_add_type(struct btf *btf, struct btf_type *t)
  53{
  54	if (btf->types_size - btf->nr_types < 2) {
  55		struct btf_type **new_types;
  56		__u32 expand_by, new_size;
  57
  58		if (btf->types_size == BTF_MAX_NR_TYPES)
  59			return -E2BIG;
  60
  61		expand_by = max(btf->types_size >> 2, 16U);
  62		new_size = min(BTF_MAX_NR_TYPES, btf->types_size + expand_by);
  63
  64		new_types = realloc(btf->types, sizeof(*new_types) * new_size);
  65		if (!new_types)
  66			return -ENOMEM;
  67
  68		if (btf->nr_types == 0)
  69			new_types[0] = &btf_void;
  70
  71		btf->types = new_types;
  72		btf->types_size = new_size;
  73	}
  74
  75	btf->types[++(btf->nr_types)] = t;
  76
  77	return 0;
  78}
  79
  80static int btf_parse_hdr(struct btf *btf)
  81{
  82	const struct btf_header *hdr = btf->hdr;
  83	__u32 meta_left;
  84
  85	if (btf->data_size < sizeof(struct btf_header)) {
  86		pr_debug("BTF header not found\n");
  87		return -EINVAL;
  88	}
  89
  90	if (hdr->magic != BTF_MAGIC) {
  91		pr_debug("Invalid BTF magic:%x\n", hdr->magic);
  92		return -EINVAL;
  93	}
  94
  95	if (hdr->version != BTF_VERSION) {
  96		pr_debug("Unsupported BTF version:%u\n", hdr->version);
  97		return -ENOTSUP;
  98	}
  99
 100	if (hdr->flags) {
 101		pr_debug("Unsupported BTF flags:%x\n", hdr->flags);
 102		return -ENOTSUP;
 103	}
 104
 105	meta_left = btf->data_size - sizeof(*hdr);
 106	if (!meta_left) {
 107		pr_debug("BTF has no data\n");
 108		return -EINVAL;
 109	}
 110
 111	if (meta_left < hdr->type_off) {
 112		pr_debug("Invalid BTF type section offset:%u\n", hdr->type_off);
 113		return -EINVAL;
 114	}
 115
 116	if (meta_left < hdr->str_off) {
 117		pr_debug("Invalid BTF string section offset:%u\n", hdr->str_off);
 118		return -EINVAL;
 119	}
 120
 121	if (hdr->type_off >= hdr->str_off) {
 122		pr_debug("BTF type section offset >= string section offset. No type?\n");
 123		return -EINVAL;
 124	}
 125
 126	if (hdr->type_off & 0x02) {
 127		pr_debug("BTF type section is not aligned to 4 bytes\n");
 128		return -EINVAL;
 129	}
 130
 131	btf->nohdr_data = btf->hdr + 1;
 132
 133	return 0;
 134}
 135
 136static int btf_parse_str_sec(struct btf *btf)
 137{
 138	const struct btf_header *hdr = btf->hdr;
 139	const char *start = btf->nohdr_data + hdr->str_off;
 140	const char *end = start + btf->hdr->str_len;
 141
 142	if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET ||
 143	    start[0] || end[-1]) {
 144		pr_debug("Invalid BTF string section\n");
 145		return -EINVAL;
 146	}
 147
 148	btf->strings = start;
 149
 150	return 0;
 151}
 152
 153static int btf_type_size(struct btf_type *t)
 154{
 155	int base_size = sizeof(struct btf_type);
 156	__u16 vlen = btf_vlen(t);
 157
 158	switch (btf_kind(t)) {
 159	case BTF_KIND_FWD:
 160	case BTF_KIND_CONST:
 161	case BTF_KIND_VOLATILE:
 162	case BTF_KIND_RESTRICT:
 163	case BTF_KIND_PTR:
 164	case BTF_KIND_TYPEDEF:
 165	case BTF_KIND_FUNC:
 166		return base_size;
 167	case BTF_KIND_INT:
 168		return base_size + sizeof(__u32);
 169	case BTF_KIND_ENUM:
 170		return base_size + vlen * sizeof(struct btf_enum);
 171	case BTF_KIND_ARRAY:
 172		return base_size + sizeof(struct btf_array);
 173	case BTF_KIND_STRUCT:
 174	case BTF_KIND_UNION:
 175		return base_size + vlen * sizeof(struct btf_member);
 176	case BTF_KIND_FUNC_PROTO:
 177		return base_size + vlen * sizeof(struct btf_param);
 178	case BTF_KIND_VAR:
 179		return base_size + sizeof(struct btf_var);
 180	case BTF_KIND_DATASEC:
 181		return base_size + vlen * sizeof(struct btf_var_secinfo);
 182	default:
 183		pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
 184		return -EINVAL;
 185	}
 186}
 187
 188static int btf_parse_type_sec(struct btf *btf)
 189{
 190	struct btf_header *hdr = btf->hdr;
 191	void *nohdr_data = btf->nohdr_data;
 192	void *next_type = nohdr_data + hdr->type_off;
 193	void *end_type = nohdr_data + hdr->str_off;
 194
 195	while (next_type < end_type) {
 196		struct btf_type *t = next_type;
 197		int type_size;
 198		int err;
 199
 200		type_size = btf_type_size(t);
 201		if (type_size < 0)
 202			return type_size;
 203		next_type += type_size;
 204		err = btf_add_type(btf, t);
 205		if (err)
 206			return err;
 207	}
 208
 209	return 0;
 210}
 211
 212__u32 btf__get_nr_types(const struct btf *btf)
 213{
 214	return btf->nr_types;
 215}
 216
 217const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
 218{
 219	if (type_id > btf->nr_types)
 220		return NULL;
 221
 222	return btf->types[type_id];
 223}
 224
 225static int determine_ptr_size(const struct btf *btf)
 226{
 227	const struct btf_type *t;
 228	const char *name;
 229	int i;
 230
 231	for (i = 1; i <= btf->nr_types; i++) {
 232		t = btf__type_by_id(btf, i);
 233		if (!btf_is_int(t))
 234			continue;
 235
 236		name = btf__name_by_offset(btf, t->name_off);
 237		if (!name)
 238			continue;
 239
 240		if (strcmp(name, "long int") == 0 ||
 241		    strcmp(name, "long unsigned int") == 0) {
 242			if (t->size != 4 && t->size != 8)
 243				continue;
 244			return t->size;
 245		}
 246	}
 247
 248	return -1;
 249}
 250
 251static size_t btf_ptr_sz(const struct btf *btf)
 252{
 253	if (!btf->ptr_sz)
 254		((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
 255	return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
 256}
 257
 258/* Return pointer size this BTF instance assumes. The size is heuristically
 259 * determined by looking for 'long' or 'unsigned long' integer type and
 260 * recording its size in bytes. If BTF type information doesn't have any such
 261 * type, this function returns 0. In the latter case, native architecture's
 262 * pointer size is assumed, so will be either 4 or 8, depending on
 263 * architecture that libbpf was compiled for. It's possible to override
 264 * guessed value by using btf__set_pointer_size() API.
 265 */
 266size_t btf__pointer_size(const struct btf *btf)
 267{
 268	if (!btf->ptr_sz)
 269		((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
 270
 271	if (btf->ptr_sz < 0)
 272		/* not enough BTF type info to guess */
 273		return 0;
 274
 275	return btf->ptr_sz;
 276}
 277
 278/* Override or set pointer size in bytes. Only values of 4 and 8 are
 279 * supported.
 280 */
 281int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
 282{
 283	if (ptr_sz != 4 && ptr_sz != 8)
 284		return -EINVAL;
 285	btf->ptr_sz = ptr_sz;
 286	return 0;
 287}
 288
 289static bool btf_type_is_void(const struct btf_type *t)
 290{
 291	return t == &btf_void || btf_is_fwd(t);
 292}
 293
 294static bool btf_type_is_void_or_null(const struct btf_type *t)
 295{
 296	return !t || btf_type_is_void(t);
 297}
 298
 299#define MAX_RESOLVE_DEPTH 32
 300
 301__s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
 302{
 303	const struct btf_array *array;
 304	const struct btf_type *t;
 305	__u32 nelems = 1;
 306	__s64 size = -1;
 307	int i;
 308
 309	t = btf__type_by_id(btf, type_id);
 310	for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t);
 311	     i++) {
 312		switch (btf_kind(t)) {
 313		case BTF_KIND_INT:
 314		case BTF_KIND_STRUCT:
 315		case BTF_KIND_UNION:
 316		case BTF_KIND_ENUM:
 317		case BTF_KIND_DATASEC:
 318			size = t->size;
 319			goto done;
 320		case BTF_KIND_PTR:
 321			size = btf_ptr_sz(btf);
 322			goto done;
 323		case BTF_KIND_TYPEDEF:
 324		case BTF_KIND_VOLATILE:
 325		case BTF_KIND_CONST:
 326		case BTF_KIND_RESTRICT:
 327		case BTF_KIND_VAR:
 328			type_id = t->type;
 329			break;
 330		case BTF_KIND_ARRAY:
 331			array = btf_array(t);
 332			if (nelems && array->nelems > UINT32_MAX / nelems)
 333				return -E2BIG;
 334			nelems *= array->nelems;
 335			type_id = array->type;
 336			break;
 337		default:
 338			return -EINVAL;
 339		}
 340
 341		t = btf__type_by_id(btf, type_id);
 342	}
 343
 344done:
 345	if (size < 0)
 346		return -EINVAL;
 347	if (nelems && size > UINT32_MAX / nelems)
 348		return -E2BIG;
 349
 350	return nelems * size;
 351}
 352
 353int btf__align_of(const struct btf *btf, __u32 id)
 354{
 355	const struct btf_type *t = btf__type_by_id(btf, id);
 356	__u16 kind = btf_kind(t);
 357
 358	switch (kind) {
 359	case BTF_KIND_INT:
 360	case BTF_KIND_ENUM:
 361		return min(btf_ptr_sz(btf), (size_t)t->size);
 362	case BTF_KIND_PTR:
 363		return btf_ptr_sz(btf);
 364	case BTF_KIND_TYPEDEF:
 365	case BTF_KIND_VOLATILE:
 366	case BTF_KIND_CONST:
 367	case BTF_KIND_RESTRICT:
 368		return btf__align_of(btf, t->type);
 369	case BTF_KIND_ARRAY:
 370		return btf__align_of(btf, btf_array(t)->type);
 371	case BTF_KIND_STRUCT:
 372	case BTF_KIND_UNION: {
 373		const struct btf_member *m = btf_members(t);
 374		__u16 vlen = btf_vlen(t);
 375		int i, max_align = 1, align;
 376
 377		for (i = 0; i < vlen; i++, m++) {
 378			align = btf__align_of(btf, m->type);
 379			if (align <= 0)
 380				return align;
 381			max_align = max(max_align, align);
 382		}
 383
 384		return max_align;
 385	}
 386	default:
 387		pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
 388		return 0;
 389	}
 390}
 391
 392int btf__resolve_type(const struct btf *btf, __u32 type_id)
 393{
 394	const struct btf_type *t;
 395	int depth = 0;
 396
 397	t = btf__type_by_id(btf, type_id);
 398	while (depth < MAX_RESOLVE_DEPTH &&
 399	       !btf_type_is_void_or_null(t) &&
 400	       (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
 401		type_id = t->type;
 402		t = btf__type_by_id(btf, type_id);
 403		depth++;
 404	}
 405
 406	if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
 407		return -EINVAL;
 408
 409	return type_id;
 410}
 411
 412__s32 btf__find_by_name(const struct btf *btf, const char *type_name)
 413{
 414	__u32 i;
 415
 416	if (!strcmp(type_name, "void"))
 417		return 0;
 418
 419	for (i = 1; i <= btf->nr_types; i++) {
 420		const struct btf_type *t = btf->types[i];
 421		const char *name = btf__name_by_offset(btf, t->name_off);
 422
 423		if (name && !strcmp(type_name, name))
 424			return i;
 425	}
 426
 427	return -ENOENT;
 428}
 429
 430__s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
 431			     __u32 kind)
 432{
 433	__u32 i;
 434
 435	if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
 436		return 0;
 437
 438	for (i = 1; i <= btf->nr_types; i++) {
 439		const struct btf_type *t = btf->types[i];
 440		const char *name;
 441
 442		if (btf_kind(t) != kind)
 443			continue;
 444		name = btf__name_by_offset(btf, t->name_off);
 445		if (name && !strcmp(type_name, name))
 446			return i;
 447	}
 448
 449	return -ENOENT;
 450}
 451
 452void btf__free(struct btf *btf)
 453{
 454	if (IS_ERR_OR_NULL(btf))
 455		return;
 456
 457	if (btf->fd >= 0)
 458		close(btf->fd);
 459
 460	free(btf->data);
 461	free(btf->types);
 462	free(btf);
 463}
 464
 465struct btf *btf__new(const void *data, __u32 size)
 466{
 467	struct btf *btf;
 468	int err;
 469
 470	btf = calloc(1, sizeof(struct btf));
 471	if (!btf)
 472		return ERR_PTR(-ENOMEM);
 473
 474	btf->fd = -1;
 475
 476	btf->data = malloc(size);
 477	if (!btf->data) {
 478		err = -ENOMEM;
 479		goto done;
 480	}
 481
 482	memcpy(btf->data, data, size);
 483	btf->data_size = size;
 484
 485	err = btf_parse_hdr(btf);
 486	if (err)
 487		goto done;
 488
 489	err = btf_parse_str_sec(btf);
 490	if (err)
 491		goto done;
 492
 493	err = btf_parse_type_sec(btf);
 494
 495done:
 496	if (err) {
 497		btf__free(btf);
 498		return ERR_PTR(err);
 499	}
 500
 501	return btf;
 502}
 503
 504static bool btf_check_endianness(const GElf_Ehdr *ehdr)
 505{
 506#if __BYTE_ORDER == __LITTLE_ENDIAN
 507	return ehdr->e_ident[EI_DATA] == ELFDATA2LSB;
 508#elif __BYTE_ORDER == __BIG_ENDIAN
 509	return ehdr->e_ident[EI_DATA] == ELFDATA2MSB;
 510#else
 511# error "Unrecognized __BYTE_ORDER__"
 512#endif
 513}
 514
 515struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
 516{
 517	Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
 518	int err = 0, fd = -1, idx = 0;
 519	struct btf *btf = NULL;
 520	Elf_Scn *scn = NULL;
 521	Elf *elf = NULL;
 522	GElf_Ehdr ehdr;
 523
 524	if (elf_version(EV_CURRENT) == EV_NONE) {
 525		pr_warn("failed to init libelf for %s\n", path);
 526		return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
 527	}
 528
 529	fd = open(path, O_RDONLY);
 530	if (fd < 0) {
 531		err = -errno;
 532		pr_warn("failed to open %s: %s\n", path, strerror(errno));
 533		return ERR_PTR(err);
 534	}
 535
 536	err = -LIBBPF_ERRNO__FORMAT;
 537
 538	elf = elf_begin(fd, ELF_C_READ, NULL);
 539	if (!elf) {
 540		pr_warn("failed to open %s as ELF file\n", path);
 541		goto done;
 542	}
 543	if (!gelf_getehdr(elf, &ehdr)) {
 544		pr_warn("failed to get EHDR from %s\n", path);
 545		goto done;
 546	}
 547	if (!btf_check_endianness(&ehdr)) {
 548		pr_warn("non-native ELF endianness is not supported\n");
 549		goto done;
 550	}
 551	if (!elf_rawdata(elf_getscn(elf, ehdr.e_shstrndx), NULL)) {
 552		pr_warn("failed to get e_shstrndx from %s\n", path);
 553		goto done;
 554	}
 555
 556	while ((scn = elf_nextscn(elf, scn)) != NULL) {
 557		GElf_Shdr sh;
 558		char *name;
 559
 560		idx++;
 561		if (gelf_getshdr(scn, &sh) != &sh) {
 562			pr_warn("failed to get section(%d) header from %s\n",
 563				idx, path);
 564			goto done;
 565		}
 566		name = elf_strptr(elf, ehdr.e_shstrndx, sh.sh_name);
 567		if (!name) {
 568			pr_warn("failed to get section(%d) name from %s\n",
 569				idx, path);
 570			goto done;
 571		}
 572		if (strcmp(name, BTF_ELF_SEC) == 0) {
 573			btf_data = elf_getdata(scn, 0);
 574			if (!btf_data) {
 575				pr_warn("failed to get section(%d, %s) data from %s\n",
 576					idx, name, path);
 577				goto done;
 578			}
 579			continue;
 580		} else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
 581			btf_ext_data = elf_getdata(scn, 0);
 582			if (!btf_ext_data) {
 583				pr_warn("failed to get section(%d, %s) data from %s\n",
 584					idx, name, path);
 585				goto done;
 586			}
 587			continue;
 588		}
 589	}
 590
 591	err = 0;
 592
 593	if (!btf_data) {
 594		err = -ENOENT;
 595		goto done;
 596	}
 597	btf = btf__new(btf_data->d_buf, btf_data->d_size);
 598	if (IS_ERR(btf))
 599		goto done;
 600
 601	switch (gelf_getclass(elf)) {
 602	case ELFCLASS32:
 603		btf__set_pointer_size(btf, 4);
 604		break;
 605	case ELFCLASS64:
 606		btf__set_pointer_size(btf, 8);
 607		break;
 608	default:
 609		pr_warn("failed to get ELF class (bitness) for %s\n", path);
 610		break;
 611	}
 612
 613	if (btf_ext && btf_ext_data) {
 614		*btf_ext = btf_ext__new(btf_ext_data->d_buf,
 615					btf_ext_data->d_size);
 616		if (IS_ERR(*btf_ext))
 617			goto done;
 618	} else if (btf_ext) {
 619		*btf_ext = NULL;
 620	}
 621done:
 622	if (elf)
 623		elf_end(elf);
 624	close(fd);
 625
 626	if (err)
 627		return ERR_PTR(err);
 628	/*
 629	 * btf is always parsed before btf_ext, so no need to clean up
 630	 * btf_ext, if btf loading failed
 631	 */
 632	if (IS_ERR(btf))
 633		return btf;
 634	if (btf_ext && IS_ERR(*btf_ext)) {
 635		btf__free(btf);
 636		err = PTR_ERR(*btf_ext);
 637		return ERR_PTR(err);
 638	}
 639	return btf;
 640}
 641
 642struct btf *btf__parse_raw(const char *path)
 643{
 644	struct btf *btf = NULL;
 645	void *data = NULL;
 646	FILE *f = NULL;
 647	__u16 magic;
 648	int err = 0;
 649	long sz;
 650
 651	f = fopen(path, "rb");
 652	if (!f) {
 653		err = -errno;
 654		goto err_out;
 655	}
 656
 657	/* check BTF magic */
 658	if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
 659		err = -EIO;
 660		goto err_out;
 661	}
 662	if (magic == __bswap_16(BTF_MAGIC)) {
 663		/* non-native endian raw BTF */
 664		pr_warn("non-native BTF endianness is not supported\n");
 665		err = -LIBBPF_ERRNO__ENDIAN;
 666		goto err_out;
 667	}
 668	if (magic != BTF_MAGIC) {
 669		/* definitely not a raw BTF */
 670		err = -EPROTO;
 671		goto err_out;
 672	}
 673
 674	/* get file size */
 675	if (fseek(f, 0, SEEK_END)) {
 676		err = -errno;
 677		goto err_out;
 678	}
 679	sz = ftell(f);
 680	if (sz < 0) {
 681		err = -errno;
 682		goto err_out;
 683	}
 684	/* rewind to the start */
 685	if (fseek(f, 0, SEEK_SET)) {
 686		err = -errno;
 687		goto err_out;
 688	}
 689
 690	/* pre-alloc memory and read all of BTF data */
 691	data = malloc(sz);
 692	if (!data) {
 693		err = -ENOMEM;
 694		goto err_out;
 695	}
 696	if (fread(data, 1, sz, f) < sz) {
 697		err = -EIO;
 698		goto err_out;
 699	}
 700
 701	/* finally parse BTF data */
 702	btf = btf__new(data, sz);
 703
 704err_out:
 705	free(data);
 706	if (f)
 707		fclose(f);
 708	return err ? ERR_PTR(err) : btf;
 709}
 710
 711struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
 712{
 713	struct btf *btf;
 714
 715	if (btf_ext)
 716		*btf_ext = NULL;
 717
 718	btf = btf__parse_raw(path);
 719	if (!IS_ERR(btf) || PTR_ERR(btf) != -EPROTO)
 720		return btf;
 721
 722	return btf__parse_elf(path, btf_ext);
 723}
 724
 725static int compare_vsi_off(const void *_a, const void *_b)
 726{
 727	const struct btf_var_secinfo *a = _a;
 728	const struct btf_var_secinfo *b = _b;
 729
 730	return a->offset - b->offset;
 731}
 732
 733static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
 734			     struct btf_type *t)
 735{
 736	__u32 size = 0, off = 0, i, vars = btf_vlen(t);
 737	const char *name = btf__name_by_offset(btf, t->name_off);
 738	const struct btf_type *t_var;
 739	struct btf_var_secinfo *vsi;
 740	const struct btf_var *var;
 741	int ret;
 742
 743	if (!name) {
 744		pr_debug("No name found in string section for DATASEC kind.\n");
 745		return -ENOENT;
 746	}
 747
 748	/* .extern datasec size and var offsets were set correctly during
 749	 * extern collection step, so just skip straight to sorting variables
 750	 */
 751	if (t->size)
 752		goto sort_vars;
 753
 754	ret = bpf_object__section_size(obj, name, &size);
 755	if (ret || !size || (t->size && t->size != size)) {
 756		pr_debug("Invalid size for section %s: %u bytes\n", name, size);
 757		return -ENOENT;
 758	}
 759
 760	t->size = size;
 761
 762	for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
 763		t_var = btf__type_by_id(btf, vsi->type);
 764		var = btf_var(t_var);
 765
 766		if (!btf_is_var(t_var)) {
 767			pr_debug("Non-VAR type seen in section %s\n", name);
 768			return -EINVAL;
 769		}
 770
 771		if (var->linkage == BTF_VAR_STATIC)
 772			continue;
 773
 774		name = btf__name_by_offset(btf, t_var->name_off);
 775		if (!name) {
 776			pr_debug("No name found in string section for VAR kind\n");
 777			return -ENOENT;
 778		}
 779
 780		ret = bpf_object__variable_offset(obj, name, &off);
 781		if (ret) {
 782			pr_debug("No offset found in symbol table for VAR %s\n",
 783				 name);
 784			return -ENOENT;
 785		}
 786
 787		vsi->offset = off;
 788	}
 789
 790sort_vars:
 791	qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
 792	return 0;
 793}
 794
 795int btf__finalize_data(struct bpf_object *obj, struct btf *btf)
 796{
 797	int err = 0;
 798	__u32 i;
 799
 800	for (i = 1; i <= btf->nr_types; i++) {
 801		struct btf_type *t = btf->types[i];
 802
 803		/* Loader needs to fix up some of the things compiler
 804		 * couldn't get its hands on while emitting BTF. This
 805		 * is section size and global variable offset. We use
 806		 * the info from the ELF itself for this purpose.
 807		 */
 808		if (btf_is_datasec(t)) {
 809			err = btf_fixup_datasec(obj, btf, t);
 810			if (err)
 811				break;
 812		}
 813	}
 814
 815	return err;
 816}
 817
 818int btf__load(struct btf *btf)
 819{
 820	__u32 log_buf_size = 0;
 821	char *log_buf = NULL;
 822	int err = 0;
 823
 824	if (btf->fd >= 0)
 825		return -EEXIST;
 826
 827retry_load:
 828	if (log_buf_size) {
 829		log_buf = malloc(log_buf_size);
 830		if (!log_buf)
 831			return -ENOMEM;
 832
 833		*log_buf = 0;
 834	}
 835
 836	btf->fd = bpf_load_btf(btf->data, btf->data_size,
 837			       log_buf, log_buf_size, false);
 838	if (btf->fd < 0) {
 839		if (!log_buf || errno == ENOSPC) {
 840			log_buf_size = max((__u32)BPF_LOG_BUF_SIZE,
 841					   log_buf_size << 1);
 842			free(log_buf);
 843			goto retry_load;
 844		}
 845
 846		err = -errno;
 847		pr_warn("Error loading BTF: %s(%d)\n", strerror(errno), errno);
 848		if (*log_buf)
 849			pr_warn("%s\n", log_buf);
 850		goto done;
 851	}
 852
 853done:
 854	free(log_buf);
 855	return err;
 856}
 857
 858int btf__fd(const struct btf *btf)
 859{
 860	return btf->fd;
 861}
 862
 863void btf__set_fd(struct btf *btf, int fd)
 864{
 865	btf->fd = fd;
 866}
 867
 868const void *btf__get_raw_data(const struct btf *btf, __u32 *size)
 869{
 870	*size = btf->data_size;
 871	return btf->data;
 872}
 873
 874const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
 875{
 876	if (offset < btf->hdr->str_len)
 877		return &btf->strings[offset];
 878	else
 879		return NULL;
 880}
 881
 882int btf__get_from_id(__u32 id, struct btf **btf)
 883{
 884	struct bpf_btf_info btf_info = { 0 };
 885	__u32 len = sizeof(btf_info);
 886	__u32 last_size;
 887	int btf_fd;
 888	void *ptr;
 889	int err;
 890
 891	err = 0;
 892	*btf = NULL;
 893	btf_fd = bpf_btf_get_fd_by_id(id);
 894	if (btf_fd < 0)
 895		return 0;
 896
 897	/* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
 898	 * let's start with a sane default - 4KiB here - and resize it only if
 899	 * bpf_obj_get_info_by_fd() needs a bigger buffer.
 900	 */
 901	btf_info.btf_size = 4096;
 902	last_size = btf_info.btf_size;
 903	ptr = malloc(last_size);
 904	if (!ptr) {
 905		err = -ENOMEM;
 906		goto exit_free;
 907	}
 908
 909	memset(ptr, 0, last_size);
 910	btf_info.btf = ptr_to_u64(ptr);
 911	err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
 912
 913	if (!err && btf_info.btf_size > last_size) {
 914		void *temp_ptr;
 915
 916		last_size = btf_info.btf_size;
 917		temp_ptr = realloc(ptr, last_size);
 918		if (!temp_ptr) {
 919			err = -ENOMEM;
 920			goto exit_free;
 921		}
 922		ptr = temp_ptr;
 923		memset(ptr, 0, last_size);
 924		btf_info.btf = ptr_to_u64(ptr);
 925		err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
 926	}
 927
 928	if (err || btf_info.btf_size > last_size) {
 929		err = errno;
 930		goto exit_free;
 931	}
 932
 933	*btf = btf__new((__u8 *)(long)btf_info.btf, btf_info.btf_size);
 934	if (IS_ERR(*btf)) {
 935		err = PTR_ERR(*btf);
 936		*btf = NULL;
 937	}
 938
 939exit_free:
 940	close(btf_fd);
 941	free(ptr);
 942
 943	return err;
 944}
 945
 946int btf__get_map_kv_tids(const struct btf *btf, const char *map_name,
 947			 __u32 expected_key_size, __u32 expected_value_size,
 948			 __u32 *key_type_id, __u32 *value_type_id)
 949{
 950	const struct btf_type *container_type;
 951	const struct btf_member *key, *value;
 952	const size_t max_name = 256;
 953	char container_name[max_name];
 954	__s64 key_size, value_size;
 955	__s32 container_id;
 956
 957	if (snprintf(container_name, max_name, "____btf_map_%s", map_name) ==
 958	    max_name) {
 959		pr_warn("map:%s length of '____btf_map_%s' is too long\n",
 960			map_name, map_name);
 961		return -EINVAL;
 962	}
 963
 964	container_id = btf__find_by_name(btf, container_name);
 965	if (container_id < 0) {
 966		pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
 967			 map_name, container_name);
 968		return container_id;
 969	}
 970
 971	container_type = btf__type_by_id(btf, container_id);
 972	if (!container_type) {
 973		pr_warn("map:%s cannot find BTF type for container_id:%u\n",
 974			map_name, container_id);
 975		return -EINVAL;
 976	}
 977
 978	if (!btf_is_struct(container_type) || btf_vlen(container_type) < 2) {
 979		pr_warn("map:%s container_name:%s is an invalid container struct\n",
 980			map_name, container_name);
 981		return -EINVAL;
 982	}
 983
 984	key = btf_members(container_type);
 985	value = key + 1;
 986
 987	key_size = btf__resolve_size(btf, key->type);
 988	if (key_size < 0) {
 989		pr_warn("map:%s invalid BTF key_type_size\n", map_name);
 990		return key_size;
 991	}
 992
 993	if (expected_key_size != key_size) {
 994		pr_warn("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
 995			map_name, (__u32)key_size, expected_key_size);
 996		return -EINVAL;
 997	}
 998
 999	value_size = btf__resolve_size(btf, value->type);
1000	if (value_size < 0) {
1001		pr_warn("map:%s invalid BTF value_type_size\n", map_name);
1002		return value_size;
1003	}
1004
1005	if (expected_value_size != value_size) {
1006		pr_warn("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1007			map_name, (__u32)value_size, expected_value_size);
1008		return -EINVAL;
1009	}
1010
1011	*key_type_id = key->type;
1012	*value_type_id = value->type;
1013
1014	return 0;
1015}
1016
1017struct btf_ext_sec_setup_param {
1018	__u32 off;
1019	__u32 len;
1020	__u32 min_rec_size;
1021	struct btf_ext_info *ext_info;
1022	const char *desc;
1023};
1024
1025static int btf_ext_setup_info(struct btf_ext *btf_ext,
1026			      struct btf_ext_sec_setup_param *ext_sec)
1027{
1028	const struct btf_ext_info_sec *sinfo;
1029	struct btf_ext_info *ext_info;
1030	__u32 info_left, record_size;
1031	/* The start of the info sec (including the __u32 record_size). */
1032	void *info;
1033
1034	if (ext_sec->len == 0)
1035		return 0;
1036
1037	if (ext_sec->off & 0x03) {
1038		pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
1039		     ext_sec->desc);
1040		return -EINVAL;
1041	}
1042
1043	info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
1044	info_left = ext_sec->len;
1045
1046	if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
1047		pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
1048			 ext_sec->desc, ext_sec->off, ext_sec->len);
1049		return -EINVAL;
1050	}
1051
1052	/* At least a record size */
1053	if (info_left < sizeof(__u32)) {
1054		pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
1055		return -EINVAL;
1056	}
1057
1058	/* The record size needs to meet the minimum standard */
1059	record_size = *(__u32 *)info;
1060	if (record_size < ext_sec->min_rec_size ||
1061	    record_size & 0x03) {
1062		pr_debug("%s section in .BTF.ext has invalid record size %u\n",
1063			 ext_sec->desc, record_size);
1064		return -EINVAL;
1065	}
1066
1067	sinfo = info + sizeof(__u32);
1068	info_left -= sizeof(__u32);
1069
1070	/* If no records, return failure now so .BTF.ext won't be used. */
1071	if (!info_left) {
1072		pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
1073		return -EINVAL;
1074	}
1075
1076	while (info_left) {
1077		unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
1078		__u64 total_record_size;
1079		__u32 num_records;
1080
1081		if (info_left < sec_hdrlen) {
1082			pr_debug("%s section header is not found in .BTF.ext\n",
1083			     ext_sec->desc);
1084			return -EINVAL;
1085		}
1086
1087		num_records = sinfo->num_info;
1088		if (num_records == 0) {
1089			pr_debug("%s section has incorrect num_records in .BTF.ext\n",
1090			     ext_sec->desc);
1091			return -EINVAL;
1092		}
1093
1094		total_record_size = sec_hdrlen +
1095				    (__u64)num_records * record_size;
1096		if (info_left < total_record_size) {
1097			pr_debug("%s section has incorrect num_records in .BTF.ext\n",
1098			     ext_sec->desc);
1099			return -EINVAL;
1100		}
1101
1102		info_left -= total_record_size;
1103		sinfo = (void *)sinfo + total_record_size;
1104	}
1105
1106	ext_info = ext_sec->ext_info;
1107	ext_info->len = ext_sec->len - sizeof(__u32);
1108	ext_info->rec_size = record_size;
1109	ext_info->info = info + sizeof(__u32);
1110
1111	return 0;
1112}
1113
1114static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
1115{
1116	struct btf_ext_sec_setup_param param = {
1117		.off = btf_ext->hdr->func_info_off,
1118		.len = btf_ext->hdr->func_info_len,
1119		.min_rec_size = sizeof(struct bpf_func_info_min),
1120		.ext_info = &btf_ext->func_info,
1121		.desc = "func_info"
1122	};
1123
1124	return btf_ext_setup_info(btf_ext, &param);
1125}
1126
1127static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
1128{
1129	struct btf_ext_sec_setup_param param = {
1130		.off = btf_ext->hdr->line_info_off,
1131		.len = btf_ext->hdr->line_info_len,
1132		.min_rec_size = sizeof(struct bpf_line_info_min),
1133		.ext_info = &btf_ext->line_info,
1134		.desc = "line_info",
1135	};
1136
1137	return btf_ext_setup_info(btf_ext, &param);
1138}
1139
1140static int btf_ext_setup_field_reloc(struct btf_ext *btf_ext)
1141{
1142	struct btf_ext_sec_setup_param param = {
1143		.off = btf_ext->hdr->field_reloc_off,
1144		.len = btf_ext->hdr->field_reloc_len,
1145		.min_rec_size = sizeof(struct bpf_field_reloc),
1146		.ext_info = &btf_ext->field_reloc_info,
1147		.desc = "field_reloc",
1148	};
1149
1150	return btf_ext_setup_info(btf_ext, &param);
1151}
1152
1153static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
1154{
1155	const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
1156
1157	if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
1158	    data_size < hdr->hdr_len) {
1159		pr_debug("BTF.ext header not found");
1160		return -EINVAL;
1161	}
1162
1163	if (hdr->magic != BTF_MAGIC) {
1164		pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
1165		return -EINVAL;
1166	}
1167
1168	if (hdr->version != BTF_VERSION) {
1169		pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
1170		return -ENOTSUP;
1171	}
1172
1173	if (hdr->flags) {
1174		pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
1175		return -ENOTSUP;
1176	}
1177
1178	if (data_size == hdr->hdr_len) {
1179		pr_debug("BTF.ext has no data\n");
1180		return -EINVAL;
1181	}
1182
1183	return 0;
1184}
1185
1186void btf_ext__free(struct btf_ext *btf_ext)
1187{
1188	if (IS_ERR_OR_NULL(btf_ext))
1189		return;
1190	free(btf_ext->data);
1191	free(btf_ext);
1192}
1193
1194struct btf_ext *btf_ext__new(__u8 *data, __u32 size)
1195{
1196	struct btf_ext *btf_ext;
1197	int err;
1198
1199	err = btf_ext_parse_hdr(data, size);
1200	if (err)
1201		return ERR_PTR(err);
1202
1203	btf_ext = calloc(1, sizeof(struct btf_ext));
1204	if (!btf_ext)
1205		return ERR_PTR(-ENOMEM);
1206
1207	btf_ext->data_size = size;
1208	btf_ext->data = malloc(size);
1209	if (!btf_ext->data) {
1210		err = -ENOMEM;
1211		goto done;
1212	}
1213	memcpy(btf_ext->data, data, size);
1214
1215	if (btf_ext->hdr->hdr_len <
1216	    offsetofend(struct btf_ext_header, line_info_len))
1217		goto done;
1218	err = btf_ext_setup_func_info(btf_ext);
1219	if (err)
1220		goto done;
1221
1222	err = btf_ext_setup_line_info(btf_ext);
1223	if (err)
1224		goto done;
1225
1226	if (btf_ext->hdr->hdr_len <
1227	    offsetofend(struct btf_ext_header, field_reloc_len))
1228		goto done;
1229	err = btf_ext_setup_field_reloc(btf_ext);
1230	if (err)
1231		goto done;
1232
1233done:
1234	if (err) {
1235		btf_ext__free(btf_ext);
1236		return ERR_PTR(err);
1237	}
1238
1239	return btf_ext;
1240}
1241
1242const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
1243{
1244	*size = btf_ext->data_size;
1245	return btf_ext->data;
1246}
1247
1248static int btf_ext_reloc_info(const struct btf *btf,
1249			      const struct btf_ext_info *ext_info,
1250			      const char *sec_name, __u32 insns_cnt,
1251			      void **info, __u32 *cnt)
1252{
1253	__u32 sec_hdrlen = sizeof(struct btf_ext_info_sec);
1254	__u32 i, record_size, existing_len, records_len;
1255	struct btf_ext_info_sec *sinfo;
1256	const char *info_sec_name;
1257	__u64 remain_len;
1258	void *data;
1259
1260	record_size = ext_info->rec_size;
1261	sinfo = ext_info->info;
1262	remain_len = ext_info->len;
1263	while (remain_len > 0) {
1264		records_len = sinfo->num_info * record_size;
1265		info_sec_name = btf__name_by_offset(btf, sinfo->sec_name_off);
1266		if (strcmp(info_sec_name, sec_name)) {
1267			remain_len -= sec_hdrlen + records_len;
1268			sinfo = (void *)sinfo + sec_hdrlen + records_len;
1269			continue;
1270		}
1271
1272		existing_len = (*cnt) * record_size;
1273		data = realloc(*info, existing_len + records_len);
1274		if (!data)
1275			return -ENOMEM;
1276
1277		memcpy(data + existing_len, sinfo->data, records_len);
1278		/* adjust insn_off only, the rest data will be passed
1279		 * to the kernel.
1280		 */
1281		for (i = 0; i < sinfo->num_info; i++) {
1282			__u32 *insn_off;
1283
1284			insn_off = data + existing_len + (i * record_size);
1285			*insn_off = *insn_off / sizeof(struct bpf_insn) +
1286				insns_cnt;
1287		}
1288		*info = data;
1289		*cnt += sinfo->num_info;
1290		return 0;
1291	}
1292
1293	return -ENOENT;
1294}
1295
1296int btf_ext__reloc_func_info(const struct btf *btf,
1297			     const struct btf_ext *btf_ext,
1298			     const char *sec_name, __u32 insns_cnt,
1299			     void **func_info, __u32 *cnt)
1300{
1301	return btf_ext_reloc_info(btf, &btf_ext->func_info, sec_name,
1302				  insns_cnt, func_info, cnt);
1303}
1304
1305int btf_ext__reloc_line_info(const struct btf *btf,
1306			     const struct btf_ext *btf_ext,
1307			     const char *sec_name, __u32 insns_cnt,
1308			     void **line_info, __u32 *cnt)
1309{
1310	return btf_ext_reloc_info(btf, &btf_ext->line_info, sec_name,
1311				  insns_cnt, line_info, cnt);
1312}
1313
1314__u32 btf_ext__func_info_rec_size(const struct btf_ext *btf_ext)
1315{
1316	return btf_ext->func_info.rec_size;
1317}
1318
1319__u32 btf_ext__line_info_rec_size(const struct btf_ext *btf_ext)
1320{
1321	return btf_ext->line_info.rec_size;
1322}
1323
1324struct btf_dedup;
1325
1326static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
1327				       const struct btf_dedup_opts *opts);
1328static void btf_dedup_free(struct btf_dedup *d);
1329static int btf_dedup_strings(struct btf_dedup *d);
1330static int btf_dedup_prim_types(struct btf_dedup *d);
1331static int btf_dedup_struct_types(struct btf_dedup *d);
1332static int btf_dedup_ref_types(struct btf_dedup *d);
1333static int btf_dedup_compact_types(struct btf_dedup *d);
1334static int btf_dedup_remap_types(struct btf_dedup *d);
1335
1336/*
1337 * Deduplicate BTF types and strings.
1338 *
1339 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
1340 * section with all BTF type descriptors and string data. It overwrites that
1341 * memory in-place with deduplicated types and strings without any loss of
1342 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
1343 * is provided, all the strings referenced from .BTF.ext section are honored
1344 * and updated to point to the right offsets after deduplication.
1345 *
1346 * If function returns with error, type/string data might be garbled and should
1347 * be discarded.
1348 *
1349 * More verbose and detailed description of both problem btf_dedup is solving,
1350 * as well as solution could be found at:
1351 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
1352 *
1353 * Problem description and justification
1354 * =====================================
1355 *
1356 * BTF type information is typically emitted either as a result of conversion
1357 * from DWARF to BTF or directly by compiler. In both cases, each compilation
1358 * unit contains information about a subset of all the types that are used
1359 * in an application. These subsets are frequently overlapping and contain a lot
1360 * of duplicated information when later concatenated together into a single
1361 * binary. This algorithm ensures that each unique type is represented by single
1362 * BTF type descriptor, greatly reducing resulting size of BTF data.
1363 *
1364 * Compilation unit isolation and subsequent duplication of data is not the only
1365 * problem. The same type hierarchy (e.g., struct and all the type that struct
1366 * references) in different compilation units can be represented in BTF to
1367 * various degrees of completeness (or, rather, incompleteness) due to
1368 * struct/union forward declarations.
1369 *
1370 * Let's take a look at an example, that we'll use to better understand the
1371 * problem (and solution). Suppose we have two compilation units, each using
1372 * same `struct S`, but each of them having incomplete type information about
1373 * struct's fields:
1374 *
1375 * // CU #1:
1376 * struct S;
1377 * struct A {
1378 *	int a;
1379 *	struct A* self;
1380 *	struct S* parent;
1381 * };
1382 * struct B;
1383 * struct S {
1384 *	struct A* a_ptr;
1385 *	struct B* b_ptr;
1386 * };
1387 *
1388 * // CU #2:
1389 * struct S;
1390 * struct A;
1391 * struct B {
1392 *	int b;
1393 *	struct B* self;
1394 *	struct S* parent;
1395 * };
1396 * struct S {
1397 *	struct A* a_ptr;
1398 *	struct B* b_ptr;
1399 * };
1400 *
1401 * In case of CU #1, BTF data will know only that `struct B` exist (but no
1402 * more), but will know the complete type information about `struct A`. While
1403 * for CU #2, it will know full type information about `struct B`, but will
1404 * only know about forward declaration of `struct A` (in BTF terms, it will
1405 * have `BTF_KIND_FWD` type descriptor with name `B`).
1406 *
1407 * This compilation unit isolation means that it's possible that there is no
1408 * single CU with complete type information describing structs `S`, `A`, and
1409 * `B`. Also, we might get tons of duplicated and redundant type information.
1410 *
1411 * Additional complication we need to keep in mind comes from the fact that
1412 * types, in general, can form graphs containing cycles, not just DAGs.
1413 *
1414 * While algorithm does deduplication, it also merges and resolves type
1415 * information (unless disabled throught `struct btf_opts`), whenever possible.
1416 * E.g., in the example above with two compilation units having partial type
1417 * information for structs `A` and `B`, the output of algorithm will emit
1418 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
1419 * (as well as type information for `int` and pointers), as if they were defined
1420 * in a single compilation unit as:
1421 *
1422 * struct A {
1423 *	int a;
1424 *	struct A* self;
1425 *	struct S* parent;
1426 * };
1427 * struct B {
1428 *	int b;
1429 *	struct B* self;
1430 *	struct S* parent;
1431 * };
1432 * struct S {
1433 *	struct A* a_ptr;
1434 *	struct B* b_ptr;
1435 * };
1436 *
1437 * Algorithm summary
1438 * =================
1439 *
1440 * Algorithm completes its work in 6 separate passes:
1441 *
1442 * 1. Strings deduplication.
1443 * 2. Primitive types deduplication (int, enum, fwd).
1444 * 3. Struct/union types deduplication.
1445 * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
1446 *    protos, and const/volatile/restrict modifiers).
1447 * 5. Types compaction.
1448 * 6. Types remapping.
1449 *
1450 * Algorithm determines canonical type descriptor, which is a single
1451 * representative type for each truly unique type. This canonical type is the
1452 * one that will go into final deduplicated BTF type information. For
1453 * struct/unions, it is also the type that algorithm will merge additional type
1454 * information into (while resolving FWDs), as it discovers it from data in
1455 * other CUs. Each input BTF type eventually gets either mapped to itself, if
1456 * that type is canonical, or to some other type, if that type is equivalent
1457 * and was chosen as canonical representative. This mapping is stored in
1458 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
1459 * FWD type got resolved to.
1460 *
1461 * To facilitate fast discovery of canonical types, we also maintain canonical
1462 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
1463 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
1464 * that match that signature. With sufficiently good choice of type signature
1465 * hashing function, we can limit number of canonical types for each unique type
1466 * signature to a very small number, allowing to find canonical type for any
1467 * duplicated type very quickly.
1468 *
1469 * Struct/union deduplication is the most critical part and algorithm for
1470 * deduplicating structs/unions is described in greater details in comments for
1471 * `btf_dedup_is_equiv` function.
1472 */
1473int btf__dedup(struct btf *btf, struct btf_ext *btf_ext,
1474	       const struct btf_dedup_opts *opts)
1475{
1476	struct btf_dedup *d = btf_dedup_new(btf, btf_ext, opts);
1477	int err;
1478
1479	if (IS_ERR(d)) {
1480		pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
1481		return -EINVAL;
1482	}
1483
1484	err = btf_dedup_strings(d);
1485	if (err < 0) {
1486		pr_debug("btf_dedup_strings failed:%d\n", err);
1487		goto done;
1488	}
1489	err = btf_dedup_prim_types(d);
1490	if (err < 0) {
1491		pr_debug("btf_dedup_prim_types failed:%d\n", err);
1492		goto done;
1493	}
1494	err = btf_dedup_struct_types(d);
1495	if (err < 0) {
1496		pr_debug("btf_dedup_struct_types failed:%d\n", err);
1497		goto done;
1498	}
1499	err = btf_dedup_ref_types(d);
1500	if (err < 0) {
1501		pr_debug("btf_dedup_ref_types failed:%d\n", err);
1502		goto done;
1503	}
1504	err = btf_dedup_compact_types(d);
1505	if (err < 0) {
1506		pr_debug("btf_dedup_compact_types failed:%d\n", err);
1507		goto done;
1508	}
1509	err = btf_dedup_remap_types(d);
1510	if (err < 0) {
1511		pr_debug("btf_dedup_remap_types failed:%d\n", err);
1512		goto done;
1513	}
1514
1515done:
1516	btf_dedup_free(d);
1517	return err;
1518}
1519
1520#define BTF_UNPROCESSED_ID ((__u32)-1)
1521#define BTF_IN_PROGRESS_ID ((__u32)-2)
1522
1523struct btf_dedup {
1524	/* .BTF section to be deduped in-place */
1525	struct btf *btf;
1526	/*
1527	 * Optional .BTF.ext section. When provided, any strings referenced
1528	 * from it will be taken into account when deduping strings
1529	 */
1530	struct btf_ext *btf_ext;
1531	/*
1532	 * This is a map from any type's signature hash to a list of possible
1533	 * canonical representative type candidates. Hash collisions are
1534	 * ignored, so even types of various kinds can share same list of
1535	 * candidates, which is fine because we rely on subsequent
1536	 * btf_xxx_equal() checks to authoritatively verify type equality.
1537	 */
1538	struct hashmap *dedup_table;
1539	/* Canonical types map */
1540	__u32 *map;
1541	/* Hypothetical mapping, used during type graph equivalence checks */
1542	__u32 *hypot_map;
1543	__u32 *hypot_list;
1544	size_t hypot_cnt;
1545	size_t hypot_cap;
1546	/* Various option modifying behavior of algorithm */
1547	struct btf_dedup_opts opts;
1548};
1549
1550struct btf_str_ptr {
1551	const char *str;
1552	__u32 new_off;
1553	bool used;
1554};
1555
1556struct btf_str_ptrs {
1557	struct btf_str_ptr *ptrs;
1558	const char *data;
1559	__u32 cnt;
1560	__u32 cap;
1561};
1562
1563static long hash_combine(long h, long value)
1564{
1565	return h * 31 + value;
1566}
1567
1568#define for_each_dedup_cand(d, node, hash) \
1569	hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
1570
1571static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
1572{
1573	return hashmap__append(d->dedup_table,
1574			       (void *)hash, (void *)(long)type_id);
1575}
1576
1577static int btf_dedup_hypot_map_add(struct btf_dedup *d,
1578				   __u32 from_id, __u32 to_id)
1579{
1580	if (d->hypot_cnt == d->hypot_cap) {
1581		__u32 *new_list;
1582
1583		d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
1584		new_list = realloc(d->hypot_list, sizeof(__u32) * d->hypot_cap);
1585		if (!new_list)
1586			return -ENOMEM;
1587		d->hypot_list = new_list;
1588	}
1589	d->hypot_list[d->hypot_cnt++] = from_id;
1590	d->hypot_map[from_id] = to_id;
1591	return 0;
1592}
1593
1594static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
1595{
1596	int i;
1597
1598	for (i = 0; i < d->hypot_cnt; i++)
1599		d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
1600	d->hypot_cnt = 0;
1601}
1602
1603static void btf_dedup_free(struct btf_dedup *d)
1604{
1605	hashmap__free(d->dedup_table);
1606	d->dedup_table = NULL;
1607
1608	free(d->map);
1609	d->map = NULL;
1610
1611	free(d->hypot_map);
1612	d->hypot_map = NULL;
1613
1614	free(d->hypot_list);
1615	d->hypot_list = NULL;
1616
1617	free(d);
1618}
1619
1620static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
1621{
1622	return (size_t)key;
1623}
1624
1625static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
1626{
1627	return 0;
1628}
1629
1630static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
1631{
1632	return k1 == k2;
1633}
1634
1635static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
1636				       const struct btf_dedup_opts *opts)
1637{
1638	struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
1639	hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
1640	int i, err = 0;
1641
1642	if (!d)
1643		return ERR_PTR(-ENOMEM);
1644
1645	d->opts.dont_resolve_fwds = opts && opts->dont_resolve_fwds;
1646	/* dedup_table_size is now used only to force collisions in tests */
1647	if (opts && opts->dedup_table_size == 1)
1648		hash_fn = btf_dedup_collision_hash_fn;
1649
1650	d->btf = btf;
1651	d->btf_ext = btf_ext;
1652
1653	d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
1654	if (IS_ERR(d->dedup_table)) {
1655		err = PTR_ERR(d->dedup_table);
1656		d->dedup_table = NULL;
1657		goto done;
1658	}
1659
1660	d->map = malloc(sizeof(__u32) * (1 + btf->nr_types));
1661	if (!d->map) {
1662		err = -ENOMEM;
1663		goto done;
1664	}
1665	/* special BTF "void" type is made canonical immediately */
1666	d->map[0] = 0;
1667	for (i = 1; i <= btf->nr_types; i++) {
1668		struct btf_type *t = d->btf->types[i];
1669
1670		/* VAR and DATASEC are never deduped and are self-canonical */
1671		if (btf_is_var(t) || btf_is_datasec(t))
1672			d->map[i] = i;
1673		else
1674			d->map[i] = BTF_UNPROCESSED_ID;
1675	}
1676
1677	d->hypot_map = malloc(sizeof(__u32) * (1 + btf->nr_types));
1678	if (!d->hypot_map) {
1679		err = -ENOMEM;
1680		goto done;
1681	}
1682	for (i = 0; i <= btf->nr_types; i++)
1683		d->hypot_map[i] = BTF_UNPROCESSED_ID;
1684
1685done:
1686	if (err) {
1687		btf_dedup_free(d);
1688		return ERR_PTR(err);
1689	}
1690
1691	return d;
1692}
1693
1694typedef int (*str_off_fn_t)(__u32 *str_off_ptr, void *ctx);
1695
1696/*
1697 * Iterate over all possible places in .BTF and .BTF.ext that can reference
1698 * string and pass pointer to it to a provided callback `fn`.
1699 */
1700static int btf_for_each_str_off(struct btf_dedup *d, str_off_fn_t fn, void *ctx)
1701{
1702	void *line_data_cur, *line_data_end;
1703	int i, j, r, rec_size;
1704	struct btf_type *t;
1705
1706	for (i = 1; i <= d->btf->nr_types; i++) {
1707		t = d->btf->types[i];
1708		r = fn(&t->name_off, ctx);
1709		if (r)
1710			return r;
1711
1712		switch (btf_kind(t)) {
1713		case BTF_KIND_STRUCT:
1714		case BTF_KIND_UNION: {
1715			struct btf_member *m = btf_members(t);
1716			__u16 vlen = btf_vlen(t);
1717
1718			for (j = 0; j < vlen; j++) {
1719				r = fn(&m->name_off, ctx);
1720				if (r)
1721					return r;
1722				m++;
1723			}
1724			break;
1725		}
1726		case BTF_KIND_ENUM: {
1727			struct btf_enum *m = btf_enum(t);
1728			__u16 vlen = btf_vlen(t);
1729
1730			for (j = 0; j < vlen; j++) {
1731				r = fn(&m->name_off, ctx);
1732				if (r)
1733					return r;
1734				m++;
1735			}
1736			break;
1737		}
1738		case BTF_KIND_FUNC_PROTO: {
1739			struct btf_param *m = btf_params(t);
1740			__u16 vlen = btf_vlen(t);
1741
1742			for (j = 0; j < vlen; j++) {
1743				r = fn(&m->name_off, ctx);
1744				if (r)
1745					return r;
1746				m++;
1747			}
1748			break;
1749		}
1750		default:
1751			break;
1752		}
1753	}
1754
1755	if (!d->btf_ext)
1756		return 0;
1757
1758	line_data_cur = d->btf_ext->line_info.info;
1759	line_data_end = d->btf_ext->line_info.info + d->btf_ext->line_info.len;
1760	rec_size = d->btf_ext->line_info.rec_size;
1761
1762	while (line_data_cur < line_data_end) {
1763		struct btf_ext_info_sec *sec = line_data_cur;
1764		struct bpf_line_info_min *line_info;
1765		__u32 num_info = sec->num_info;
1766
1767		r = fn(&sec->sec_name_off, ctx);
1768		if (r)
1769			return r;
1770
1771		line_data_cur += sizeof(struct btf_ext_info_sec);
1772		for (i = 0; i < num_info; i++) {
1773			line_info = line_data_cur;
1774			r = fn(&line_info->file_name_off, ctx);
1775			if (r)
1776				return r;
1777			r = fn(&line_info->line_off, ctx);
1778			if (r)
1779				return r;
1780			line_data_cur += rec_size;
1781		}
1782	}
1783
1784	return 0;
1785}
1786
1787static int str_sort_by_content(const void *a1, const void *a2)
1788{
1789	const struct btf_str_ptr *p1 = a1;
1790	const struct btf_str_ptr *p2 = a2;
1791
1792	return strcmp(p1->str, p2->str);
1793}
1794
1795static int str_sort_by_offset(const void *a1, const void *a2)
1796{
1797	const struct btf_str_ptr *p1 = a1;
1798	const struct btf_str_ptr *p2 = a2;
1799
1800	if (p1->str != p2->str)
1801		return p1->str < p2->str ? -1 : 1;
1802	return 0;
1803}
1804
1805static int btf_dedup_str_ptr_cmp(const void *str_ptr, const void *pelem)
1806{
1807	const struct btf_str_ptr *p = pelem;
1808
1809	if (str_ptr != p->str)
1810		return (const char *)str_ptr < p->str ? -1 : 1;
1811	return 0;
1812}
1813
1814static int btf_str_mark_as_used(__u32 *str_off_ptr, void *ctx)
1815{
1816	struct btf_str_ptrs *strs;
1817	struct btf_str_ptr *s;
1818
1819	if (*str_off_ptr == 0)
1820		return 0;
1821
1822	strs = ctx;
1823	s = bsearch(strs->data + *str_off_ptr, strs->ptrs, strs->cnt,
1824		    sizeof(struct btf_str_ptr), btf_dedup_str_ptr_cmp);
1825	if (!s)
1826		return -EINVAL;
1827	s->used = true;
1828	return 0;
1829}
1830
1831static int btf_str_remap_offset(__u32 *str_off_ptr, void *ctx)
1832{
1833	struct btf_str_ptrs *strs;
1834	struct btf_str_ptr *s;
1835
1836	if (*str_off_ptr == 0)
1837		return 0;
1838
1839	strs = ctx;
1840	s = bsearch(strs->data + *str_off_ptr, strs->ptrs, strs->cnt,
1841		    sizeof(struct btf_str_ptr), btf_dedup_str_ptr_cmp);
1842	if (!s)
1843		return -EINVAL;
1844	*str_off_ptr = s->new_off;
1845	return 0;
1846}
1847
1848/*
1849 * Dedup string and filter out those that are not referenced from either .BTF
1850 * or .BTF.ext (if provided) sections.
1851 *
1852 * This is done by building index of all strings in BTF's string section,
1853 * then iterating over all entities that can reference strings (e.g., type
1854 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
1855 * strings as used. After that all used strings are deduped and compacted into
1856 * sequential blob of memory and new offsets are calculated. Then all the string
1857 * references are iterated again and rewritten using new offsets.
1858 */
1859static int btf_dedup_strings(struct btf_dedup *d)
1860{
1861	const struct btf_header *hdr = d->btf->hdr;
1862	char *start = (char *)d->btf->nohdr_data + hdr->str_off;
1863	char *end = start + d->btf->hdr->str_len;
1864	char *p = start, *tmp_strs = NULL;
1865	struct btf_str_ptrs strs = {
1866		.cnt = 0,
1867		.cap = 0,
1868		.ptrs = NULL,
1869		.data = start,
1870	};
1871	int i, j, err = 0, grp_idx;
1872	bool grp_used;
1873
1874	/* build index of all strings */
1875	while (p < end) {
1876		if (strs.cnt + 1 > strs.cap) {
1877			struct btf_str_ptr *new_ptrs;
1878
1879			strs.cap += max(strs.cnt / 2, 16U);
1880			new_ptrs = realloc(strs.ptrs,
1881					   sizeof(strs.ptrs[0]) * strs.cap);
1882			if (!new_ptrs) {
1883				err = -ENOMEM;
1884				goto done;
1885			}
1886			strs.ptrs = new_ptrs;
1887		}
1888
1889		strs.ptrs[strs.cnt].str = p;
1890		strs.ptrs[strs.cnt].used = false;
1891
1892		p += strlen(p) + 1;
1893		strs.cnt++;
1894	}
1895
1896	/* temporary storage for deduplicated strings */
1897	tmp_strs = malloc(d->btf->hdr->str_len);
1898	if (!tmp_strs) {
1899		err = -ENOMEM;
1900		goto done;
1901	}
1902
1903	/* mark all used strings */
1904	strs.ptrs[0].used = true;
1905	err = btf_for_each_str_off(d, btf_str_mark_as_used, &strs);
1906	if (err)
1907		goto done;
1908
1909	/* sort strings by context, so that we can identify duplicates */
1910	qsort(strs.ptrs, strs.cnt, sizeof(strs.ptrs[0]), str_sort_by_content);
1911
1912	/*
1913	 * iterate groups of equal strings and if any instance in a group was
1914	 * referenced, emit single instance and remember new offset
1915	 */
1916	p = tmp_strs;
1917	grp_idx = 0;
1918	grp_used = strs.ptrs[0].used;
1919	/* iterate past end to avoid code duplication after loop */
1920	for (i = 1; i <= strs.cnt; i++) {
1921		/*
1922		 * when i == strs.cnt, we want to skip string comparison and go
1923		 * straight to handling last group of strings (otherwise we'd
1924		 * need to handle last group after the loop w/ duplicated code)
1925		 */
1926		if (i < strs.cnt &&
1927		    !strcmp(strs.ptrs[i].str, strs.ptrs[grp_idx].str)) {
1928			grp_used = grp_used || strs.ptrs[i].used;
1929			continue;
1930		}
1931
1932		/*
1933		 * this check would have been required after the loop to handle
1934		 * last group of strings, but due to <= condition in a loop
1935		 * we avoid that duplication
1936		 */
1937		if (grp_used) {
1938			int new_off = p - tmp_strs;
1939			__u32 len = strlen(strs.ptrs[grp_idx].str);
1940
1941			memmove(p, strs.ptrs[grp_idx].str, len + 1);
1942			for (j = grp_idx; j < i; j++)
1943				strs.ptrs[j].new_off = new_off;
1944			p += len + 1;
1945		}
1946
1947		if (i < strs.cnt) {
1948			grp_idx = i;
1949			grp_used = strs.ptrs[i].used;
1950		}
1951	}
1952
1953	/* replace original strings with deduped ones */
1954	d->btf->hdr->str_len = p - tmp_strs;
1955	memmove(start, tmp_strs, d->btf->hdr->str_len);
1956	end = start + d->btf->hdr->str_len;
1957
1958	/* restore original order for further binary search lookups */
1959	qsort(strs.ptrs, strs.cnt, sizeof(strs.ptrs[0]), str_sort_by_offset);
1960
1961	/* remap string offsets */
1962	err = btf_for_each_str_off(d, btf_str_remap_offset, &strs);
1963	if (err)
1964		goto done;
1965
1966	d->btf->hdr->str_len = end - start;
1967
1968done:
1969	free(tmp_strs);
1970	free(strs.ptrs);
1971	return err;
1972}
1973
1974static long btf_hash_common(struct btf_type *t)
1975{
1976	long h;
1977
1978	h = hash_combine(0, t->name_off);
1979	h = hash_combine(h, t->info);
1980	h = hash_combine(h, t->size);
1981	return h;
1982}
1983
1984static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
1985{
1986	return t1->name_off == t2->name_off &&
1987	       t1->info == t2->info &&
1988	       t1->size == t2->size;
1989}
1990
1991/* Calculate type signature hash of INT. */
1992static long btf_hash_int(struct btf_type *t)
1993{
1994	__u32 info = *(__u32 *)(t + 1);
1995	long h;
1996
1997	h = btf_hash_common(t);
1998	h = hash_combine(h, info);
1999	return h;
2000}
2001
2002/* Check structural equality of two INTs. */
2003static bool btf_equal_int(struct btf_type *t1, struct btf_type *t2)
2004{
2005	__u32 info1, info2;
2006
2007	if (!btf_equal_common(t1, t2))
2008		return false;
2009	info1 = *(__u32 *)(t1 + 1);
2010	info2 = *(__u32 *)(t2 + 1);
2011	return info1 == info2;
2012}
2013
2014/* Calculate type signature hash of ENUM. */
2015static long btf_hash_enum(struct btf_type *t)
2016{
2017	long h;
2018
2019	/* don't hash vlen and enum members to support enum fwd resolving */
2020	h = hash_combine(0, t->name_off);
2021	h = hash_combine(h, t->info & ~0xffff);
2022	h = hash_combine(h, t->size);
2023	return h;
2024}
2025
2026/* Check structural equality of two ENUMs. */
2027static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
2028{
2029	const struct btf_enum *m1, *m2;
2030	__u16 vlen;
2031	int i;
2032
2033	if (!btf_equal_common(t1, t2))
2034		return false;
2035
2036	vlen = btf_vlen(t1);
2037	m1 = btf_enum(t1);
2038	m2 = btf_enum(t2);
2039	for (i = 0; i < vlen; i++) {
2040		if (m1->name_off != m2->name_off || m1->val != m2->val)
2041			return false;
2042		m1++;
2043		m2++;
2044	}
2045	return true;
2046}
2047
2048static inline bool btf_is_enum_fwd(struct btf_type *t)
2049{
2050	return btf_is_enum(t) && btf_vlen(t) == 0;
2051}
2052
2053static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
2054{
2055	if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
2056		return btf_equal_enum(t1, t2);
2057	/* ignore vlen when comparing */
2058	return t1->name_off == t2->name_off &&
2059	       (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
2060	       t1->size == t2->size;
2061}
2062
2063/*
2064 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
2065 * as referenced type IDs equivalence is established separately during type
2066 * graph equivalence check algorithm.
2067 */
2068static long btf_hash_struct(struct btf_type *t)
2069{
2070	const struct btf_member *member = btf_members(t);
2071	__u32 vlen = btf_vlen(t);
2072	long h = btf_hash_common(t);
2073	int i;
2074
2075	for (i = 0; i < vlen; i++) {
2076		h = hash_combine(h, member->name_off);
2077		h = hash_combine(h, member->offset);
2078		/* no hashing of referenced type ID, it can be unresolved yet */
2079		member++;
2080	}
2081	return h;
2082}
2083
2084/*
2085 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
2086 * IDs. This check is performed during type graph equivalence check and
2087 * referenced types equivalence is checked separately.
2088 */
2089static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
2090{
2091	const struct btf_member *m1, *m2;
2092	__u16 vlen;
2093	int i;
2094
2095	if (!btf_equal_common(t1, t2))
2096		return false;
2097
2098	vlen = btf_vlen(t1);
2099	m1 = btf_members(t1);
2100	m2 = btf_members(t2);
2101	for (i = 0; i < vlen; i++) {
2102		if (m1->name_off != m2->name_off || m1->offset != m2->offset)
2103			return false;
2104		m1++;
2105		m2++;
2106	}
2107	return true;
2108}
2109
2110/*
2111 * Calculate type signature hash of ARRAY, including referenced type IDs,
2112 * under assumption that they were already resolved to canonical type IDs and
2113 * are not going to change.
2114 */
2115static long btf_hash_array(struct btf_type *t)
2116{
2117	const struct btf_array *info = btf_array(t);
2118	long h = btf_hash_common(t);
2119
2120	h = hash_combine(h, info->type);
2121	h = hash_combine(h, info->index_type);
2122	h = hash_combine(h, info->nelems);
2123	return h;
2124}
2125
2126/*
2127 * Check exact equality of two ARRAYs, taking into account referenced
2128 * type IDs, under assumption that they were already resolved to canonical
2129 * type IDs and are not going to change.
2130 * This function is called during reference types deduplication to compare
2131 * ARRAY to potential canonical representative.
2132 */
2133static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
2134{
2135	const struct btf_array *info1, *info2;
2136
2137	if (!btf_equal_common(t1, t2))
2138		return false;
2139
2140	info1 = btf_array(t1);
2141	info2 = btf_array(t2);
2142	return info1->type == info2->type &&
2143	       info1->index_type == info2->index_type &&
2144	       info1->nelems == info2->nelems;
2145}
2146
2147/*
2148 * Check structural compatibility of two ARRAYs, ignoring referenced type
2149 * IDs. This check is performed during type graph equivalence check and
2150 * referenced types equivalence is checked separately.
2151 */
2152static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
2153{
2154	if (!btf_equal_common(t1, t2))
2155		return false;
2156
2157	return btf_array(t1)->nelems == btf_array(t2)->nelems;
2158}
2159
2160/*
2161 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
2162 * under assumption that they were already resolved to canonical type IDs and
2163 * are not going to change.
2164 */
2165static long btf_hash_fnproto(struct btf_type *t)
2166{
2167	const struct btf_param *member = btf_params(t);
2168	__u16 vlen = btf_vlen(t);
2169	long h = btf_hash_common(t);
2170	int i;
2171
2172	for (i = 0; i < vlen; i++) {
2173		h = hash_combine(h, member->name_off);
2174		h = hash_combine(h, member->type);
2175		member++;
2176	}
2177	return h;
2178}
2179
2180/*
2181 * Check exact equality of two FUNC_PROTOs, taking into account referenced
2182 * type IDs, under assumption that they were already resolved to canonical
2183 * type IDs and are not going to change.
2184 * This function is called during reference types deduplication to compare
2185 * FUNC_PROTO to potential canonical representative.
2186 */
2187static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
2188{
2189	const struct btf_param *m1, *m2;
2190	__u16 vlen;
2191	int i;
2192
2193	if (!btf_equal_common(t1, t2))
2194		return false;
2195
2196	vlen = btf_vlen(t1);
2197	m1 = btf_params(t1);
2198	m2 = btf_params(t2);
2199	for (i = 0; i < vlen; i++) {
2200		if (m1->name_off != m2->name_off || m1->type != m2->type)
2201			return false;
2202		m1++;
2203		m2++;
2204	}
2205	return true;
2206}
2207
2208/*
2209 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
2210 * IDs. This check is performed during type graph equivalence check and
2211 * referenced types equivalence is checked separately.
2212 */
2213static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
2214{
2215	const struct btf_param *m1, *m2;
2216	__u16 vlen;
2217	int i;
2218
2219	/* skip return type ID */
2220	if (t1->name_off != t2->name_off || t1->info != t2->info)
2221		return false;
2222
2223	vlen = btf_vlen(t1);
2224	m1 = btf_params(t1);
2225	m2 = btf_params(t2);
2226	for (i = 0; i < vlen; i++) {
2227		if (m1->name_off != m2->name_off)
2228			return false;
2229		m1++;
2230		m2++;
2231	}
2232	return true;
2233}
2234
2235/*
2236 * Deduplicate primitive types, that can't reference other types, by calculating
2237 * their type signature hash and comparing them with any possible canonical
2238 * candidate. If no canonical candidate matches, type itself is marked as
2239 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
2240 */
2241static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
2242{
2243	struct btf_type *t = d->btf->types[type_id];
2244	struct hashmap_entry *hash_entry;
2245	struct btf_type *cand;
2246	/* if we don't find equivalent type, then we are canonical */
2247	__u32 new_id = type_id;
2248	__u32 cand_id;
2249	long h;
2250
2251	switch (btf_kind(t)) {
2252	case BTF_KIND_CONST:
2253	case BTF_KIND_VOLATILE:
2254	case BTF_KIND_RESTRICT:
2255	case BTF_KIND_PTR:
2256	case BTF_KIND_TYPEDEF:
2257	case BTF_KIND_ARRAY:
2258	case BTF_KIND_STRUCT:
2259	case BTF_KIND_UNION:
2260	case BTF_KIND_FUNC:
2261	case BTF_KIND_FUNC_PROTO:
2262	case BTF_KIND_VAR:
2263	case BTF_KIND_DATASEC:
2264		return 0;
2265
2266	case BTF_KIND_INT:
2267		h = btf_hash_int(t);
2268		for_each_dedup_cand(d, hash_entry, h) {
2269			cand_id = (__u32)(long)hash_entry->value;
2270			cand = d->btf->types[cand_id];
2271			if (btf_equal_int(t, cand)) {
2272				new_id = cand_id;
2273				break;
2274			}
2275		}
2276		break;
2277
2278	case BTF_KIND_ENUM:
2279		h = btf_hash_enum(t);
2280		for_each_dedup_cand(d, hash_entry, h) {
2281			cand_id = (__u32)(long)hash_entry->value;
2282			cand = d->btf->types[cand_id];
2283			if (btf_equal_enum(t, cand)) {
2284				new_id = cand_id;
2285				break;
2286			}
2287			if (d->opts.dont_resolve_fwds)
2288				continue;
2289			if (btf_compat_enum(t, cand)) {
2290				if (btf_is_enum_fwd(t)) {
2291					/* resolve fwd to full enum */
2292					new_id = cand_id;
2293					break;
2294				}
2295				/* resolve canonical enum fwd to full enum */
2296				d->map[cand_id] = type_id;
2297			}
2298		}
2299		break;
2300
2301	case BTF_KIND_FWD:
2302		h = btf_hash_common(t);
2303		for_each_dedup_cand(d, hash_entry, h) {
2304			cand_id = (__u32)(long)hash_entry->value;
2305			cand = d->btf->types[cand_id];
2306			if (btf_equal_common(t, cand)) {
2307				new_id = cand_id;
2308				break;
2309			}
2310		}
2311		break;
2312
2313	default:
2314		return -EINVAL;
2315	}
2316
2317	d->map[type_id] = new_id;
2318	if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
2319		return -ENOMEM;
2320
2321	return 0;
2322}
2323
2324static int btf_dedup_prim_types(struct btf_dedup *d)
2325{
2326	int i, err;
2327
2328	for (i = 1; i <= d->btf->nr_types; i++) {
2329		err = btf_dedup_prim_type(d, i);
2330		if (err)
2331			return err;
2332	}
2333	return 0;
2334}
2335
2336/*
2337 * Check whether type is already mapped into canonical one (could be to itself).
2338 */
2339static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
2340{
2341	return d->map[type_id] <= BTF_MAX_NR_TYPES;
2342}
2343
2344/*
2345 * Resolve type ID into its canonical type ID, if any; otherwise return original
2346 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
2347 * STRUCT/UNION link and resolve it into canonical type ID as well.
2348 */
2349static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
2350{
2351	while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
2352		type_id = d->map[type_id];
2353	return type_id;
2354}
2355
2356/*
2357 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
2358 * type ID.
2359 */
2360static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
2361{
2362	__u32 orig_type_id = type_id;
2363
2364	if (!btf_is_fwd(d->btf->types[type_id]))
2365		return type_id;
2366
2367	while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
2368		type_id = d->map[type_id];
2369
2370	if (!btf_is_fwd(d->btf->types[type_id]))
2371		return type_id;
2372
2373	return orig_type_id;
2374}
2375
2376
2377static inline __u16 btf_fwd_kind(struct btf_type *t)
2378{
2379	return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
2380}
2381
2382/*
2383 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
2384 * call it "candidate graph" in this description for brevity) to a type graph
2385 * formed by (potential) canonical struct/union ("canonical graph" for brevity
2386 * here, though keep in mind that not all types in canonical graph are
2387 * necessarily canonical representatives themselves, some of them might be
2388 * duplicates or its uniqueness might not have been established yet).
2389 * Returns:
2390 *  - >0, if type graphs are equivalent;
2391 *  -  0, if not equivalent;
2392 *  - <0, on error.
2393 *
2394 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
2395 * equivalence of BTF types at each step. If at any point BTF types in candidate
2396 * and canonical graphs are not compatible structurally, whole graphs are
2397 * incompatible. If types are structurally equivalent (i.e., all information
2398 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
2399 * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
2400 * If a type references other types, then those referenced types are checked
2401 * for equivalence recursively.
2402 *
2403 * During DFS traversal, if we find that for current `canon_id` type we
2404 * already have some mapping in hypothetical map, we check for two possible
2405 * situations:
2406 *   - `canon_id` is mapped to exactly the same type as `cand_id`. This will
2407 *     happen when type graphs have cycles. In this case we assume those two
2408 *     types are equivalent.
2409 *   - `canon_id` is mapped to different type. This is contradiction in our
2410 *     hypothetical mapping, because same graph in canonical graph corresponds
2411 *     to two different types in candidate graph, which for equivalent type
2412 *     graphs shouldn't happen. This condition terminates equivalence check
2413 *     with negative result.
2414 *
2415 * If type graphs traversal exhausts types to check and find no contradiction,
2416 * then type graphs are equivalent.
2417 *
2418 * When checking types for equivalence, there is one special case: FWD types.
2419 * If FWD type resolution is allowed and one of the types (either from canonical
2420 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
2421 * flag) and their names match, hypothetical mapping is updated to point from
2422 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
2423 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
2424 *
2425 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
2426 * if there are two exactly named (or anonymous) structs/unions that are
2427 * compatible structurally, one of which has FWD field, while other is concrete
2428 * STRUCT/UNION, but according to C sources they are different structs/unions
2429 * that are referencing different types with the same name. This is extremely
2430 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
2431 * this logic is causing problems.
2432 *
2433 * Doing FWD resolution means that both candidate and/or canonical graphs can
2434 * consists of portions of the graph that come from multiple compilation units.
2435 * This is due to the fact that types within single compilation unit are always
2436 * deduplicated and FWDs are already resolved, if referenced struct/union
2437 * definiton is available. So, if we had unresolved FWD and found corresponding
2438 * STRUCT/UNION, they will be from different compilation units. This
2439 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
2440 * type graph will likely have at least two different BTF types that describe
2441 * same type (e.g., most probably there will be two different BTF types for the
2442 * same 'int' primitive type) and could even have "overlapping" parts of type
2443 * graph that describe same subset of types.
2444 *
2445 * This in turn means that our assumption that each type in canonical graph
2446 * must correspond to exactly one type in candidate graph might not hold
2447 * anymore and will make it harder to detect contradictions using hypothetical
2448 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
2449 * resolution only in canonical graph. FWDs in candidate graphs are never
2450 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
2451 * that can occur:
2452 *   - Both types in canonical and candidate graphs are FWDs. If they are
2453 *     structurally equivalent, then they can either be both resolved to the
2454 *     same STRUCT/UNION or not resolved at all. In both cases they are
2455 *     equivalent and there is no need to resolve FWD on candidate side.
2456 *   - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
2457 *     so nothing to resolve as well, algorithm will check equivalence anyway.
2458 *   - Type in canonical graph is FWD, while type in candidate is concrete
2459 *     STRUCT/UNION. In this case candidate graph comes from single compilation
2460 *     unit, so there is exactly one BTF type for each unique C type. After
2461 *     resolving FWD into STRUCT/UNION, there might be more than one BTF type
2462 *     in canonical graph mapping to single BTF type in candidate graph, but
2463 *     because hypothetical mapping maps from canonical to candidate types, it's
2464 *     alright, and we still maintain the property of having single `canon_id`
2465 *     mapping to single `cand_id` (there could be two different `canon_id`
2466 *     mapped to the same `cand_id`, but it's not contradictory).
2467 *   - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
2468 *     graph is FWD. In this case we are just going to check compatibility of
2469 *     STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
2470 *     assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
2471 *     a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
2472 *     turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
2473 *     canonical graph.
2474 */
2475static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
2476			      __u32 canon_id)
2477{
2478	struct btf_type *cand_type;
2479	struct btf_type *canon_type;
2480	__u32 hypot_type_id;
2481	__u16 cand_kind;
2482	__u16 canon_kind;
2483	int i, eq;
2484
2485	/* if both resolve to the same canonical, they must be equivalent */
2486	if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
2487		return 1;
2488
2489	canon_id = resolve_fwd_id(d, canon_id);
2490
2491	hypot_type_id = d->hypot_map[canon_id];
2492	if (hypot_type_id <= BTF_MAX_NR_TYPES)
2493		return hypot_type_id == cand_id;
2494
2495	if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
2496		return -ENOMEM;
2497
2498	cand_type = d->btf->types[cand_id];
2499	canon_type = d->btf->types[canon_id];
2500	cand_kind = btf_kind(cand_type);
2501	canon_kind = btf_kind(canon_type);
2502
2503	if (cand_type->name_off != canon_type->name_off)
2504		return 0;
2505
2506	/* FWD <--> STRUCT/UNION equivalence check, if enabled */
2507	if (!d->opts.dont_resolve_fwds
2508	    && (cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
2509	    && cand_kind != canon_kind) {
2510		__u16 real_kind;
2511		__u16 fwd_kind;
2512
2513		if (cand_kind == BTF_KIND_FWD) {
2514			real_kind = canon_kind;
2515			fwd_kind = btf_fwd_kind(cand_type);
2516		} else {
2517			real_kind = cand_kind;
2518			fwd_kind = btf_fwd_kind(canon_type);
2519		}
2520		return fwd_kind == real_kind;
2521	}
2522
2523	if (cand_kind != canon_kind)
2524		return 0;
2525
2526	switch (cand_kind) {
2527	case BTF_KIND_INT:
2528		return btf_equal_int(cand_type, canon_type);
2529
2530	case BTF_KIND_ENUM:
2531		if (d->opts.dont_resolve_fwds)
2532			return btf_equal_enum(cand_type, canon_type);
2533		else
2534			return btf_compat_enum(cand_type, canon_type);
2535
2536	case BTF_KIND_FWD:
2537		return btf_equal_common(cand_type, canon_type);
2538
2539	case BTF_KIND_CONST:
2540	case BTF_KIND_VOLATILE:
2541	case BTF_KIND_RESTRICT:
2542	case BTF_KIND_PTR:
2543	case BTF_KIND_TYPEDEF:
2544	case BTF_KIND_FUNC:
2545		if (cand_type->info != canon_type->info)
2546			return 0;
2547		return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
2548
2549	case BTF_KIND_ARRAY: {
2550		const struct btf_array *cand_arr, *canon_arr;
2551
2552		if (!btf_compat_array(cand_type, canon_type))
2553			return 0;
2554		cand_arr = btf_array(cand_type);
2555		canon_arr = btf_array(canon_type);
2556		eq = btf_dedup_is_equiv(d,
2557			cand_arr->index_type, canon_arr->index_type);
2558		if (eq <= 0)
2559			return eq;
2560		return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
2561	}
2562
2563	case BTF_KIND_STRUCT:
2564	case BTF_KIND_UNION: {
2565		const struct btf_member *cand_m, *canon_m;
2566		__u16 vlen;
2567
2568		if (!btf_shallow_equal_struct(cand_type, canon_type))
2569			return 0;
2570		vlen = btf_vlen(cand_type);
2571		cand_m = btf_members(cand_type);
2572		canon_m = btf_members(canon_type);
2573		for (i = 0; i < vlen; i++) {
2574			eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
2575			if (eq <= 0)
2576				return eq;
2577			cand_m++;
2578			canon_m++;
2579		}
2580
2581		return 1;
2582	}
2583
2584	case BTF_KIND_FUNC_PROTO: {
2585		const struct btf_param *cand_p, *canon_p;
2586		__u16 vlen;
2587
2588		if (!btf_compat_fnproto(cand_type, canon_type))
2589			return 0;
2590		eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
2591		if (eq <= 0)
2592			return eq;
2593		vlen = btf_vlen(cand_type);
2594		cand_p = btf_params(cand_type);
2595		canon_p = btf_params(canon_type);
2596		for (i = 0; i < vlen; i++) {
2597			eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
2598			if (eq <= 0)
2599				return eq;
2600			cand_p++;
2601			canon_p++;
2602		}
2603		return 1;
2604	}
2605
2606	default:
2607		return -EINVAL;
2608	}
2609	return 0;
2610}
2611
2612/*
2613 * Use hypothetical mapping, produced by successful type graph equivalence
2614 * check, to augment existing struct/union canonical mapping, where possible.
2615 *
2616 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
2617 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
2618 * it doesn't matter if FWD type was part of canonical graph or candidate one,
2619 * we are recording the mapping anyway. As opposed to carefulness required
2620 * for struct/union correspondence mapping (described below), for FWD resolution
2621 * it's not important, as by the time that FWD type (reference type) will be
2622 * deduplicated all structs/unions will be deduped already anyway.
2623 *
2624 * Recording STRUCT/UNION mapping is purely a performance optimization and is
2625 * not required for correctness. It needs to be done carefully to ensure that
2626 * struct/union from candidate's type graph is not mapped into corresponding
2627 * struct/union from canonical type graph that itself hasn't been resolved into
2628 * canonical representative. The only guarantee we have is that canonical
2629 * struct/union was determined as canonical and that won't change. But any
2630 * types referenced through that struct/union fields could have been not yet
2631 * resolved, so in case like that it's too early to establish any kind of
2632 * correspondence between structs/unions.
2633 *
2634 * No canonical correspondence is derived for primitive types (they are already
2635 * deduplicated completely already anyway) or reference types (they rely on
2636 * stability of struct/union canonical relationship for equivalence checks).
2637 */
2638static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
2639{
2640	__u32 cand_type_id, targ_type_id;
2641	__u16 t_kind, c_kind;
2642	__u32 t_id, c_id;
2643	int i;
2644
2645	for (i = 0; i < d->hypot_cnt; i++) {
2646		cand_type_id = d->hypot_list[i];
2647		targ_type_id = d->hypot_map[cand_type_id];
2648		t_id = resolve_type_id(d, targ_type_id);
2649		c_id = resolve_type_id(d, cand_type_id);
2650		t_kind = btf_kind(d->btf->types[t_id]);
2651		c_kind = btf_kind(d->btf->types[c_id]);
2652		/*
2653		 * Resolve FWD into STRUCT/UNION.
2654		 * It's ok to resolve FWD into STRUCT/UNION that's not yet
2655		 * mapped to canonical representative (as opposed to
2656		 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
2657		 * eventually that struct is going to be mapped and all resolved
2658		 * FWDs will automatically resolve to correct canonical
2659		 * representative. This will happen before ref type deduping,
2660		 * which critically depends on stability of these mapping. This
2661		 * stability is not a requirement for STRUCT/UNION equivalence
2662		 * checks, though.
2663		 */
2664		if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
2665			d->map[c_id] = t_id;
2666		else if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
2667			d->map[t_id] = c_id;
2668
2669		if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
2670		    c_kind != BTF_KIND_FWD &&
2671		    is_type_mapped(d, c_id) &&
2672		    !is_type_mapped(d, t_id)) {
2673			/*
2674			 * as a perf optimization, we can map struct/union
2675			 * that's part of type graph we just verified for
2676			 * equivalence. We can do that for struct/union that has
2677			 * canonical representative only, though.
2678			 */
2679			d->map[t_id] = c_id;
2680		}
2681	}
2682}
2683
2684/*
2685 * Deduplicate struct/union types.
2686 *
2687 * For each struct/union type its type signature hash is calculated, taking
2688 * into account type's name, size, number, order and names of fields, but
2689 * ignoring type ID's referenced from fields, because they might not be deduped
2690 * completely until after reference types deduplication phase. This type hash
2691 * is used to iterate over all potential canonical types, sharing same hash.
2692 * For each canonical candidate we check whether type graphs that they form
2693 * (through referenced types in fields and so on) are equivalent using algorithm
2694 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
2695 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
2696 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
2697 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
2698 * potentially map other structs/unions to their canonical representatives,
2699 * if such relationship hasn't yet been established. This speeds up algorithm
2700 * by eliminating some of the duplicate work.
2701 *
2702 * If no matching canonical representative was found, struct/union is marked
2703 * as canonical for itself and is added into btf_dedup->dedup_table hash map
2704 * for further look ups.
2705 */
2706static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
2707{
2708	struct btf_type *cand_type, *t;
2709	struct hashmap_entry *hash_entry;
2710	/* if we don't find equivalent type, then we are canonical */
2711	__u32 new_id = type_id;
2712	__u16 kind;
2713	long h;
2714
2715	/* already deduped or is in process of deduping (loop detected) */
2716	if (d->map[type_id] <= BTF_MAX_NR_TYPES)
2717		return 0;
2718
2719	t = d->btf->types[type_id];
2720	kind = btf_kind(t);
2721
2722	if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
2723		return 0;
2724
2725	h = btf_hash_struct(t);
2726	for_each_dedup_cand(d, hash_entry, h) {
2727		__u32 cand_id = (__u32)(long)hash_entry->value;
2728		int eq;
2729
2730		/*
2731		 * Even though btf_dedup_is_equiv() checks for
2732		 * btf_shallow_equal_struct() internally when checking two
2733		 * structs (unions) for equivalence, we need to guard here
2734		 * from picking matching FWD type as a dedup candidate.
2735		 * This can happen due to hash collision. In such case just
2736		 * relying on btf_dedup_is_equiv() would lead to potentially
2737		 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
2738		 * FWD and compatible STRUCT/UNION are considered equivalent.
2739		 */
2740		cand_type = d->btf->types[cand_id];
2741		if (!btf_shallow_equal_struct(t, cand_type))
2742			continue;
2743
2744		btf_dedup_clear_hypot_map(d);
2745		eq = btf_dedup_is_equiv(d, type_id, cand_id);
2746		if (eq < 0)
2747			return eq;
2748		if (!eq)
2749			continue;
2750		new_id = cand_id;
2751		btf_dedup_merge_hypot_map(d);
2752		break;
2753	}
2754
2755	d->map[type_id] = new_id;
2756	if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
2757		return -ENOMEM;
2758
2759	return 0;
2760}
2761
2762static int btf_dedup_struct_types(struct btf_dedup *d)
2763{
2764	int i, err;
2765
2766	for (i = 1; i <= d->btf->nr_types; i++) {
2767		err = btf_dedup_struct_type(d, i);
2768		if (err)
2769			return err;
2770	}
2771	return 0;
2772}
2773
2774/*
2775 * Deduplicate reference type.
2776 *
2777 * Once all primitive and struct/union types got deduplicated, we can easily
2778 * deduplicate all other (reference) BTF types. This is done in two steps:
2779 *
2780 * 1. Resolve all referenced type IDs into their canonical type IDs. This
2781 * resolution can be done either immediately for primitive or struct/union types
2782 * (because they were deduped in previous two phases) or recursively for
2783 * reference types. Recursion will always terminate at either primitive or
2784 * struct/union type, at which point we can "unwind" chain of reference types
2785 * one by one. There is no danger of encountering cycles because in C type
2786 * system the only way to form type cycle is through struct/union, so any chain
2787 * of reference types, even those taking part in a type cycle, will inevitably
2788 * reach struct/union at some point.
2789 *
2790 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
2791 * becomes "stable", in the sense that no further deduplication will cause
2792 * any changes to it. With that, it's now possible to calculate type's signature
2793 * hash (this time taking into account referenced type IDs) and loop over all
2794 * potential canonical representatives. If no match was found, current type
2795 * will become canonical representative of itself and will be added into
2796 * btf_dedup->dedup_table as another possible canonical representative.
2797 */
2798static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
2799{
2800	struct hashmap_entry *hash_entry;
2801	__u32 new_id = type_id, cand_id;
2802	struct btf_type *t, *cand;
2803	/* if we don't find equivalent type, then we are representative type */
2804	int ref_type_id;
2805	long h;
2806
2807	if (d->map[type_id] == BTF_IN_PROGRESS_ID)
2808		return -ELOOP;
2809	if (d->map[type_id] <= BTF_MAX_NR_TYPES)
2810		return resolve_type_id(d, type_id);
2811
2812	t = d->btf->types[type_id];
2813	d->map[type_id] = BTF_IN_PROGRESS_ID;
2814
2815	switch (btf_kind(t)) {
2816	case BTF_KIND_CONST:
2817	case BTF_KIND_VOLATILE:
2818	case BTF_KIND_RESTRICT:
2819	case BTF_KIND_PTR:
2820	case BTF_KIND_TYPEDEF:
2821	case BTF_KIND_FUNC:
2822		ref_type_id = btf_dedup_ref_type(d, t->type);
2823		if (ref_type_id < 0)
2824			return ref_type_id;
2825		t->type = ref_type_id;
2826
2827		h = btf_hash_common(t);
2828		for_each_dedup_cand(d, hash_entry, h) {
2829			cand_id = (__u32)(long)hash_entry->value;
2830			cand = d->btf->types[cand_id];
2831			if (btf_equal_common(t, cand)) {
2832				new_id = cand_id;
2833				break;
2834			}
2835		}
2836		break;
2837
2838	case BTF_KIND_ARRAY: {
2839		struct btf_array *info = btf_array(t);
2840
2841		ref_type_id = btf_dedup_ref_type(d, info->type);
2842		if (ref_type_id < 0)
2843			return ref_type_id;
2844		info->type = ref_type_id;
2845
2846		ref_type_id = btf_dedup_ref_type(d, info->index_type);
2847		if (ref_type_id < 0)
2848			return ref_type_id;
2849		info->index_type = ref_type_id;
2850
2851		h = btf_hash_array(t);
2852		for_each_dedup_cand(d, hash_entry, h) {
2853			cand_id = (__u32)(long)hash_entry->value;
2854			cand = d->btf->types[cand_id];
2855			if (btf_equal_array(t, cand)) {
2856				new_id = cand_id;
2857				break;
2858			}
2859		}
2860		break;
2861	}
2862
2863	case BTF_KIND_FUNC_PROTO: {
2864		struct btf_param *param;
2865		__u16 vlen;
2866		int i;
2867
2868		ref_type_id = btf_dedup_ref_type(d, t->type);
2869		if (ref_type_id < 0)
2870			return ref_type_id;
2871		t->type = ref_type_id;
2872
2873		vlen = btf_vlen(t);
2874		param = btf_params(t);
2875		for (i = 0; i < vlen; i++) {
2876			ref_type_id = btf_dedup_ref_type(d, param->type);
2877			if (ref_type_id < 0)
2878				return ref_type_id;
2879			param->type = ref_type_id;
2880			param++;
2881		}
2882
2883		h = btf_hash_fnproto(t);
2884		for_each_dedup_cand(d, hash_entry, h) {
2885			cand_id = (__u32)(long)hash_entry->value;
2886			cand = d->btf->types[cand_id];
2887			if (btf_equal_fnproto(t, cand)) {
2888				new_id = cand_id;
2889				break;
2890			}
2891		}
2892		break;
2893	}
2894
2895	default:
2896		return -EINVAL;
2897	}
2898
2899	d->map[type_id] = new_id;
2900	if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
2901		return -ENOMEM;
2902
2903	return new_id;
2904}
2905
2906static int btf_dedup_ref_types(struct btf_dedup *d)
2907{
2908	int i, err;
2909
2910	for (i = 1; i <= d->btf->nr_types; i++) {
2911		err = btf_dedup_ref_type(d, i);
2912		if (err < 0)
2913			return err;
2914	}
2915	/* we won't need d->dedup_table anymore */
2916	hashmap__free(d->dedup_table);
2917	d->dedup_table = NULL;
2918	return 0;
2919}
2920
2921/*
2922 * Compact types.
2923 *
2924 * After we established for each type its corresponding canonical representative
2925 * type, we now can eliminate types that are not canonical and leave only
2926 * canonical ones layed out sequentially in memory by copying them over
2927 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
2928 * a map from original type ID to a new compacted type ID, which will be used
2929 * during next phase to "fix up" type IDs, referenced from struct/union and
2930 * reference types.
2931 */
2932static int btf_dedup_compact_types(struct btf_dedup *d)
2933{
2934	struct btf_type **new_types;
2935	__u32 next_type_id = 1;
2936	char *types_start, *p;
2937	int i, len;
2938
2939	/* we are going to reuse hypot_map to store compaction remapping */
2940	d->hypot_map[0] = 0;
2941	for (i = 1; i <= d->btf->nr_types; i++)
2942		d->hypot_map[i] = BTF_UNPROCESSED_ID;
2943
2944	types_start = d->btf->nohdr_data + d->btf->hdr->type_off;
2945	p = types_start;
2946
2947	for (i = 1; i <= d->btf->nr_types; i++) {
2948		if (d->map[i] != i)
2949			continue;
2950
2951		len = btf_type_size(d->btf->types[i]);
2952		if (len < 0)
2953			return len;
2954
2955		memmove(p, d->btf->types[i], len);
2956		d->hypot_map[i] = next_type_id;
2957		d->btf->types[next_type_id] = (struct btf_type *)p;
2958		p += len;
2959		next_type_id++;
2960	}
2961
2962	/* shrink struct btf's internal types index and update btf_header */
2963	d->btf->nr_types = next_type_id - 1;
2964	d->btf->types_size = d->btf->nr_types;
2965	d->btf->hdr->type_len = p - types_start;
2966	new_types = realloc(d->btf->types,
2967			    (1 + d->btf->nr_types) * sizeof(struct btf_type *));
2968	if (!new_types)
2969		return -ENOMEM;
2970	d->btf->types = new_types;
2971
2972	/* make sure string section follows type information without gaps */
2973	d->btf->hdr->str_off = p - (char *)d->btf->nohdr_data;
2974	memmove(p, d->btf->strings, d->btf->hdr->str_len);
2975	d->btf->strings = p;
2976	p += d->btf->hdr->str_len;
2977
2978	d->btf->data_size = p - (char *)d->btf->data;
2979	return 0;
2980}
2981
2982/*
2983 * Figure out final (deduplicated and compacted) type ID for provided original
2984 * `type_id` by first resolving it into corresponding canonical type ID and
2985 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
2986 * which is populated during compaction phase.
2987 */
2988static int btf_dedup_remap_type_id(struct btf_dedup *d, __u32 type_id)
2989{
2990	__u32 resolved_type_id, new_type_id;
2991
2992	resolved_type_id = resolve_type_id(d, type_id);
2993	new_type_id = d->hypot_map[resolved_type_id];
2994	if (new_type_id > BTF_MAX_NR_TYPES)
2995		return -EINVAL;
2996	return new_type_id;
2997}
2998
2999/*
3000 * Remap referenced type IDs into deduped type IDs.
3001 *
3002 * After BTF types are deduplicated and compacted, their final type IDs may
3003 * differ from original ones. The map from original to a corresponding
3004 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
3005 * compaction phase. During remapping phase we are rewriting all type IDs
3006 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
3007 * their final deduped type IDs.
3008 */
3009static int btf_dedup_remap_type(struct btf_dedup *d, __u32 type_id)
3010{
3011	struct btf_type *t = d->btf->types[type_id];
3012	int i, r;
3013
3014	switch (btf_kind(t)) {
3015	case BTF_KIND_INT:
3016	case BTF_KIND_ENUM:
3017		break;
3018
3019	case BTF_KIND_FWD:
3020	case BTF_KIND_CONST:
3021	case BTF_KIND_VOLATILE:
3022	case BTF_KIND_RESTRICT:
3023	case BTF_KIND_PTR:
3024	case BTF_KIND_TYPEDEF:
3025	case BTF_KIND_FUNC:
3026	case BTF_KIND_VAR:
3027		r = btf_dedup_remap_type_id(d, t->type);
3028		if (r < 0)
3029			return r;
3030		t->type = r;
3031		break;
3032
3033	case BTF_KIND_ARRAY: {
3034		struct btf_array *arr_info = btf_array(t);
3035
3036		r = btf_dedup_remap_type_id(d, arr_info->type);
3037		if (r < 0)
3038			return r;
3039		arr_info->type = r;
3040		r = btf_dedup_remap_type_id(d, arr_info->index_type);
3041		if (r < 0)
3042			return r;
3043		arr_info->index_type = r;
3044		break;
3045	}
3046
3047	case BTF_KIND_STRUCT:
3048	case BTF_KIND_UNION: {
3049		struct btf_member *member = btf_members(t);
3050		__u16 vlen = btf_vlen(t);
3051
3052		for (i = 0; i < vlen; i++) {
3053			r = btf_dedup_remap_type_id(d, member->type);
3054			if (r < 0)
3055				return r;
3056			member->type = r;
3057			member++;
3058		}
3059		break;
3060	}
3061
3062	case BTF_KIND_FUNC_PROTO: {
3063		struct btf_param *param = btf_params(t);
3064		__u16 vlen = btf_vlen(t);
3065
3066		r = btf_dedup_remap_type_id(d, t->type);
3067		if (r < 0)
3068			return r;
3069		t->type = r;
3070
3071		for (i = 0; i < vlen; i++) {
3072			r = btf_dedup_remap_type_id(d, param->type);
3073			if (r < 0)
3074				return r;
3075			param->type = r;
3076			param++;
3077		}
3078		break;
3079	}
3080
3081	case BTF_KIND_DATASEC: {
3082		struct btf_var_secinfo *var = btf_var_secinfos(t);
3083		__u16 vlen = btf_vlen(t);
3084
3085		for (i = 0; i < vlen; i++) {
3086			r = btf_dedup_remap_type_id(d, var->type);
3087			if (r < 0)
3088				return r;
3089			var->type = r;
3090			var++;
3091		}
3092		break;
3093	}
3094
3095	default:
3096		return -EINVAL;
3097	}
3098
3099	return 0;
3100}
3101
3102static int btf_dedup_remap_types(struct btf_dedup *d)
3103{
3104	int i, r;
3105
3106	for (i = 1; i <= d->btf->nr_types; i++) {
3107		r = btf_dedup_remap_type(d, i);
3108		if (r < 0)
3109			return r;
3110	}
3111	return 0;
3112}
3113
3114/*
3115 * Probe few well-known locations for vmlinux kernel image and try to load BTF
3116 * data out of it to use for target BTF.
3117 */
3118struct btf *libbpf_find_kernel_btf(void)
3119{
3120	struct {
3121		const char *path_fmt;
3122		bool raw_btf;
3123	} locations[] = {
3124		/* try canonical vmlinux BTF through sysfs first */
3125		{ "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
3126		/* fall back to trying to find vmlinux ELF on disk otherwise */
3127		{ "/boot/vmlinux-%1$s" },
3128		{ "/lib/modules/%1$s/vmlinux-%1$s" },
3129		{ "/lib/modules/%1$s/build/vmlinux" },
3130		{ "/usr/lib/modules/%1$s/kernel/vmlinux" },
3131		{ "/usr/lib/debug/boot/vmlinux-%1$s" },
3132		{ "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
3133		{ "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
3134	};
3135	char path[PATH_MAX + 1];
3136	struct utsname buf;
3137	struct btf *btf;
3138	int i;
3139
3140	uname(&buf);
3141
3142	for (i = 0; i < ARRAY_SIZE(locations); i++) {
3143		snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
3144
3145		if (access(path, R_OK))
3146			continue;
3147
3148		if (locations[i].raw_btf)
3149			btf = btf__parse_raw(path);
3150		else
3151			btf = btf__parse_elf(path, NULL);
3152
3153		pr_debug("loading kernel BTF '%s': %ld\n",
3154			 path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
3155		if (IS_ERR(btf))
3156			continue;
3157
3158		return btf;
3159	}
3160
3161	pr_warn("failed to find valid kernel BTF\n");
3162	return ERR_PTR(-ESRCH);
3163}