projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bpf: Attachment verification for BPF_MODIFY_RETURN
[linux-2.6-block.git] / tools / lib / bpf / libbpf.c
CommitLineData
1bc38b8f 1// SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
6061a3d6 2
1b76c13e
WN		 3/*
4 * Common eBPF ELF object loading operations.
5 *
6 * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
7 * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
8 * Copyright (C) 2015 Huawei Inc.
f367540c 9 * Copyright (C) 2017 Nicira, Inc.
d859900c 10 * Copyright (C) 2019 Isovalent, Inc.
1b76c13e
WN		 11 */
12
b4269954 13#ifndef _GNU_SOURCE
531b014e 14#define _GNU_SOURCE
b4269954 15#endif
1b76c13e 16#include <stdlib.h>
b3f59d66
WN		 17#include <stdio.h>
18#include <stdarg.h>
f367540c 19#include <libgen.h>
34090915 20#include <inttypes.h>
8ab9da57 21#include <limits.h>
b3f59d66 22#include <string.h>
1b76c13e 23#include <unistd.h>
cdb2f920 24#include <endian.h>
1a5e3fb1
WN		 25#include <fcntl.h>
26#include <errno.h>
113e6b7e 27#include <ctype.h>
1b76c13e 28#include <asm/unistd.h>
e28ff1a8 29#include <linux/err.h>
cb1e5e96 30#include <linux/kernel.h>
1b76c13e 31#include <linux/bpf.h>
38d5d3b3 32#include <linux/btf.h>
47eff617 33#include <linux/filter.h>
9a208eff 34#include <linux/list.h>
f367540c 35#include <linux/limits.h>
438363c0 36#include <linux/perf_event.h>
a64af0ef 37#include <linux/ring_buffer.h>
5e61f270 38#include <linux/version.h>
fb84b822 39#include <sys/epoll.h>
63f2f5ee 40#include <sys/ioctl.h>
fb84b822 41#include <sys/mman.h>
f367540c
JS		 42#include <sys/stat.h>
43#include <sys/types.h>
44#include <sys/vfs.h>
ddc7c304 45#include <sys/utsname.h>
dc3a2d25 46#include <sys/resource.h>
531b014e 47#include <tools/libc_compat.h>
1a5e3fb1
WN		 48#include <libelf.h>
49#include <gelf.h>
166750bc 50#include <zlib.h>
1b76c13e
WN		 51
52#include "libbpf.h"
52d3352e 53#include "bpf.h"
8a138aed 54#include "btf.h"
6d41907c 55#include "str_error.h"
d7c4b398 56#include "libbpf_internal.h"
ddc7c304 57#include "hashmap.h"
b3f59d66 58
1d1a3bcf
AN		 59/* make sure libbpf doesn't use kernel-only integer typedefs */
60#pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64
61
9b16137a
WN		 62#ifndef EM_BPF
63#define EM_BPF 247
64#endif
65
f367540c
JS		 66#ifndef BPF_FS_MAGIC
67#define BPF_FS_MAGIC 0xcafe4a11
68#endif
69
ff466b58
AI		 70/* vsprintf() in __base_pr() uses nonliteral format string. It may break
71 * compilation if user enables corresponding warning. Disable it explicitly.
72 */
73#pragma GCC diagnostic ignored "-Wformat-nonliteral"
74
b3f59d66
WN		 75#define __printf(a, b) __attribute__((format(printf, a, b)))
76
590a0088
MKL		 77static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
78static struct bpf_program *bpf_object__find_prog_by_idx(struct bpf_object *obj,
79 int idx);
80static const struct btf_type *
81skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id);
82
a8a1f7d0
SF		 83static int __base_pr(enum libbpf_print_level level, const char *format,
84 va_list args)
b3f59d66 85{
6f1ae8b6
YS		 86 if (level == LIBBPF_DEBUG)
87 return 0;
88
a8a1f7d0 89 return vfprintf(stderr, format, args);
b3f59d66
WN		 90}
91
a8a1f7d0 92static libbpf_print_fn_t __libbpf_pr = __base_pr;
b3f59d66 93
e87fd8ba 94libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
b3f59d66 95{
e87fd8ba
AN		 96 libbpf_print_fn_t old_print_fn = __libbpf_pr;
97
6f1ae8b6 98 __libbpf_pr = fn;
e87fd8ba 99 return old_print_fn;
b3f59d66 100}
1a5e3fb1 101
8461ef8b
YS		 102__printf(2, 3)
103void libbpf_print(enum libbpf_print_level level, const char *format, ...)
104{
105 va_list args;
106
6f1ae8b6
YS		 107 if (!__libbpf_pr)
108 return;
109
8461ef8b 110 va_start(args, format);
6f1ae8b6 111 __libbpf_pr(level, format, args);
8461ef8b
YS		 112 va_end(args);
113}
114
dc3a2d25
THJ		 115static void pr_perm_msg(int err)
116{
117 struct rlimit limit;
118 char buf[100];
119
120 if (err != -EPERM || geteuid() != 0)
121 return;
122
123 err = getrlimit(RLIMIT_MEMLOCK, &limit);
124 if (err)
125 return;
126
127 if (limit.rlim_cur == RLIM_INFINITY)
128 return;
129
130 if (limit.rlim_cur < 1024)
b5c7d0d0 131 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
dc3a2d25
THJ		 132 else if (limit.rlim_cur < 1024*1024)
133 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
134 else
135 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
136
137 pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
138 buf);
139}
140
6371ca3b
WN		 141#define STRERR_BUFSIZE 128
142
1a5e3fb1
WN		 143/* Copied from tools/perf/util/util.h */
144#ifndef zfree
145# define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
146#endif
147
148#ifndef zclose
149# define zclose(fd) ({ \
150 int ___err = 0; \
151 if ((fd) >= 0) \
152 ___err = close((fd)); \
153 fd = -1; \
154 ___err; })
155#endif
156
157#ifdef HAVE_LIBELF_MMAP_SUPPORT
158# define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
159#else
160# define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
161#endif
162
34be1646
SL		 163static inline __u64 ptr_to_u64(const void *ptr)
164{
165 return (__u64) (unsigned long) ptr;
166}
167
47eff617
SF		 168struct bpf_capabilities {
169 /* v4.14: kernel support for program & map names. */
170 __u32 name:1;
8837fe5d
DB		 171 /* v5.2: kernel support for global data sections. */
172 __u32 global_data:1;
d7c4b398
AN		 173 /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
174 __u32 btf_func:1;
175 /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
176 __u32 btf_datasec:1;
7fe74b43
AN		 177 /* BPF_F_MMAPABLE is supported for arrays */
178 __u32 array_mmap:1;
2d3eb67f
AS		 179 /* BTF_FUNC_GLOBAL is supported */
180 __u32 btf_func_global:1;
47eff617
SF		 181};
182
166750bc
AN		 183enum reloc_type {
184 RELO_LD64,
185 RELO_CALL,
186 RELO_DATA,
187 RELO_EXTERN,
188};
189
190struct reloc_desc {
191 enum reloc_type type;
192 int insn_idx;
193 int map_idx;
194 int sym_off;
195};
196
a5b8bd47
WN		 197/*
198 * bpf_prog should be a better name but it has been used in
199 * linux/filter.h.
200 */
201struct bpf_program {
202 /* Index in elf obj file, for relocation use. */
203 int idx;
88cda1c9 204 char *name;
f0307a7e 205 int prog_ifindex;
a5b8bd47 206 char *section_name;
33a2c75c
SF		 207 /* section_name with / replaced by _; makes recursive pinning
208 * in bpf_object__pin_programs easier
209 */
210 char *pin_name;
a5b8bd47 211 struct bpf_insn *insns;
48cca7e4 212 size_t insns_cnt, main_prog_cnt;
5f44e4c8 213 enum bpf_prog_type type;
34090915 214
166750bc 215 struct reloc_desc *reloc_desc;
34090915 216 int nr_reloc;
da11b417 217 int log_level;
55cffde2 218
b580563e
WN		 219 struct {
220 int nr;
221 int *fds;
222 } instances;
223 bpf_program_prep_t preprocessor;
aa9b1ac3
WN		 224
225 struct bpf_object *obj;
226 void *priv;
227 bpf_program_clear_priv_t clear_priv;
d7be143b
AI		 228
229 enum bpf_attach_type expected_attach_type;
12a8654b 230 __u32 attach_btf_id;
e7bf94db 231 __u32 attach_prog_fd;
2993e051
YS		 232 void *func_info;
233 __u32 func_info_rec_size;
f0187f0b 234 __u32 func_info_cnt;
47eff617
SF		 235
236 struct bpf_capabilities *caps;
3d650141
MKL		 237
238 void *line_info;
239 __u32 line_info_rec_size;
240 __u32 line_info_cnt;
04656198 241 __u32 prog_flags;
a5b8bd47
WN		 242};
243
590a0088
MKL		 244struct bpf_struct_ops {
245 const char *tname;
246 const struct btf_type *type;
247 struct bpf_program **progs;
248 __u32 *kern_func_off;
249 /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
250 void *data;
251 /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
252 * btf_vmlinux's format.
253 * struct bpf_struct_ops_tcp_congestion_ops {
254 * [... some other kernel fields ...]
255 * struct tcp_congestion_ops data;
256 * }
257 * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
258 * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
259 * from "data".
260 */
261 void *kern_vdata;
262 __u32 type_id;
263};
264
ac9d1389
AN		 265#define DATA_SEC ".data"
266#define BSS_SEC ".bss"
267#define RODATA_SEC ".rodata"
81bfdd08 268#define KCONFIG_SEC ".kconfig"
590a0088 269#define STRUCT_OPS_SEC ".struct_ops"
ac9d1389 270
d859900c
DB		 271enum libbpf_map_type {
272 LIBBPF_MAP_UNSPEC,
273 LIBBPF_MAP_DATA,
274 LIBBPF_MAP_BSS,
275 LIBBPF_MAP_RODATA,
81bfdd08 276 LIBBPF_MAP_KCONFIG,
d859900c
DB		 277};
278
279static const char * const libbpf_type_to_btf_name[] = {
ac9d1389
AN		 280 [LIBBPF_MAP_DATA] = DATA_SEC,
281 [LIBBPF_MAP_BSS] = BSS_SEC,
282 [LIBBPF_MAP_RODATA] = RODATA_SEC,
81bfdd08 283 [LIBBPF_MAP_KCONFIG] = KCONFIG_SEC,
d859900c
DB		 284};
285
9d759a9b 286struct bpf_map {
561bbcca 287 char *name;
01af3bf0 288 int fd;
db48814b
AN		 289 int sec_idx;
290 size_t sec_offset;
f0307a7e 291 int map_ifindex;
addb9fc9 292 int inner_map_fd;
9d759a9b 293 struct bpf_map_def def;
5b891af7
MKL		 294 __u32 btf_key_type_id;
295 __u32 btf_value_type_id;
590a0088 296 __u32 btf_vmlinux_value_type_id;
9d759a9b
WN		 297 void *priv;
298 bpf_map_clear_priv_t clear_priv;
d859900c 299 enum libbpf_map_type libbpf_type;
eba9c5f4 300 void *mmaped;
590a0088 301 struct bpf_struct_ops *st_ops;
4580b25f
THJ		 302 char *pin_path;
303 bool pinned;
ec6d5f47 304 bool reused;
d859900c
DB		 305};
306
166750bc
AN		 307enum extern_type {
308 EXT_UNKNOWN,
309 EXT_CHAR,
310 EXT_BOOL,
311 EXT_INT,
312 EXT_TRISTATE,
313 EXT_CHAR_ARR,
314};
315
316struct extern_desc {
317 const char *name;
318 int sym_idx;
319 int btf_id;
320 enum extern_type type;
321 int sz;
322 int align;
323 int data_off;
324 bool is_signed;
325 bool is_weak;
326 bool is_set;
327};
328
9a208eff
WN		 329static LIST_HEAD(bpf_objects_list);
330
1a5e3fb1 331struct bpf_object {
d859900c 332 char name[BPF_OBJ_NAME_LEN];
cb1e5e96 333 char license[64];
438363c0 334 __u32 kern_version;
0b3d1efa 335
a5b8bd47
WN		 336 struct bpf_program *programs;
337 size_t nr_programs;
9d759a9b
WN		 338 struct bpf_map *maps;
339 size_t nr_maps;
bf829271 340 size_t maps_cap;
9d759a9b 341
8601fd42 342 char *kconfig;
166750bc
AN		 343 struct extern_desc *externs;
344 int nr_extern;
81bfdd08 345 int kconfig_map_idx;
166750bc 346
52d3352e 347 bool loaded;
9a94f277 348 bool has_pseudo_calls;
a5b8bd47 349
1a5e3fb1
WN		 350 /*
351 * Information when doing elf related work. Only valid if fd
352 * is valid.
353 */
354 struct {
355 int fd;
5e61f270 356 const void *obj_buf;
6c956392 357 size_t obj_buf_sz;
1a5e3fb1
WN		 358 Elf *elf;
359 GElf_Ehdr ehdr;
bec7d68c 360 Elf_Data *symbols;
d859900c
DB		 361 Elf_Data *data;
362 Elf_Data *rodata;
363 Elf_Data *bss;
590a0088 364 Elf_Data *st_ops_data;
77ba9a5b 365 size_t strtabidx;
b62f06e8
WN		 366 struct {
367 GElf_Shdr shdr;
368 Elf_Data *data;
1f8e2bcb
AN		 369 } *reloc_sects;
370 int nr_reloc_sects;
666810e8 371 int maps_shndx;
abd29c93 372 int btf_maps_shndx;
48cca7e4 373 int text_shndx;
166750bc 374 int symbols_shndx;
d859900c
DB		 375 int data_shndx;
376 int rodata_shndx;
377 int bss_shndx;
590a0088 378 int st_ops_shndx;
1a5e3fb1 379 } efile;
9a208eff
WN		 380 /*
381 * All loaded bpf_object is linked in a list, which is
382 * hidden to caller. bpf_objects__<func> handlers deal with
383 * all objects.
384 */
385 struct list_head list;
10931d24 386
8a138aed 387 struct btf *btf;
a6ed02ca
KS		 388 /* Parse and load BTF vmlinux if any of the programs in the object need
389 * it at load time.
390 */
391 struct btf *btf_vmlinux;
2993e051 392 struct btf_ext *btf_ext;
8a138aed 393
10931d24
WN		 394 void *priv;
395 bpf_object_clear_priv_t clear_priv;
396
47eff617
SF		 397 struct bpf_capabilities caps;
398
1a5e3fb1
WN		 399 char path[];
400};
401#define obj_elf_valid(o) ((o)->efile.elf)
402
29cd77f4 403void bpf_program__unload(struct bpf_program *prog)
55cffde2 404{
b580563e
WN		 405 int i;
406
55cffde2
WN		 407 if (!prog)
408 return;
409
b580563e
WN		 410 /*
411 * If the object is opened but the program was never loaded,
412 * it is possible that prog->instances.nr == -1.
413 */
414 if (prog->instances.nr > 0) {
415 for (i = 0; i < prog->instances.nr; i++)
416 zclose(prog->instances.fds[i]);
417 } else if (prog->instances.nr != -1) {
be18010e
KW		 418 pr_warn("Internal error: instances.nr is %d\n",
419 prog->instances.nr);
b580563e
WN		 420 }
421
422 prog->instances.nr = -1;
423 zfree(&prog->instances.fds);
2993e051 424
2993e051 425 zfree(&prog->func_info);
07a09d1b 426 zfree(&prog->line_info);
55cffde2
WN		 427}
428
a5b8bd47
WN		 429static void bpf_program__exit(struct bpf_program *prog)
430{
431 if (!prog)
432 return;
433
aa9b1ac3
WN		 434 if (prog->clear_priv)
435 prog->clear_priv(prog, prog->priv);
436
437 prog->priv = NULL;
438 prog->clear_priv = NULL;
439
55cffde2 440 bpf_program__unload(prog);
88cda1c9 441 zfree(&prog->name);
a5b8bd47 442 zfree(&prog->section_name);
33a2c75c 443 zfree(&prog->pin_name);
a5b8bd47 444 zfree(&prog->insns);
34090915
WN		 445 zfree(&prog->reloc_desc);
446
447 prog->nr_reloc = 0;
a5b8bd47
WN		 448 prog->insns_cnt = 0;
449 prog->idx = -1;
450}
451
33a2c75c
SF		 452static char *__bpf_program__pin_name(struct bpf_program *prog)
453{
454 char *name, *p;
455
456 name = p = strdup(prog->section_name);
457 while ((p = strchr(p, '/')))
458 *p = '_';
459
460 return name;
461}
462
a5b8bd47 463static int
88cda1c9
MKL		 464bpf_program__init(void *data, size_t size, char *section_name, int idx,
465 struct bpf_program *prog)
a5b8bd47 466{
8ca990ce
AN		 467 const size_t bpf_insn_sz = sizeof(struct bpf_insn);
468
469 if (size == 0 || size % bpf_insn_sz) {
be18010e
KW		 470 pr_warn("corrupted section '%s', size: %zu\n",
471 section_name, size);
a5b8bd47
WN		 472 return -EINVAL;
473 }
474
1ad9cbb8 475 memset(prog, 0, sizeof(*prog));
a5b8bd47 476
88cda1c9 477 prog->section_name = strdup(section_name);
a5b8bd47 478 if (!prog->section_name) {
be18010e
KW		 479 pr_warn("failed to alloc name for prog under section(%d) %s\n",
480 idx, section_name);
a5b8bd47
WN		 481 goto errout;
482 }
483
33a2c75c
SF		 484 prog->pin_name = __bpf_program__pin_name(prog);
485 if (!prog->pin_name) {
be18010e
KW		 486 pr_warn("failed to alloc pin name for prog under section(%d) %s\n",
487 idx, section_name);
33a2c75c
SF		 488 goto errout;
489 }
490
a5b8bd47
WN		 491 prog->insns = malloc(size);
492 if (!prog->insns) {
be18010e
KW		 493 pr_warn("failed to alloc insns for prog under section %s\n",
494 section_name);
a5b8bd47
WN		 495 goto errout;
496 }
8ca990ce
AN		 497 prog->insns_cnt = size / bpf_insn_sz;
498 memcpy(prog->insns, data, size);
a5b8bd47 499 prog->idx = idx;
b580563e
WN		 500 prog->instances.fds = NULL;
501 prog->instances.nr = -1;
47ae7e3d 502 prog->type = BPF_PROG_TYPE_UNSPEC;
a5b8bd47
WN		 503
504 return 0;
505errout:
506 bpf_program__exit(prog);
507 return -ENOMEM;
508}
509
510static int
511bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
88cda1c9 512 char *section_name, int idx)
a5b8bd47
WN		 513{
514 struct bpf_program prog, *progs;
515 int nr_progs, err;
516
88cda1c9 517 err = bpf_program__init(data, size, section_name, idx, &prog);
a5b8bd47
WN		 518 if (err)
519 return err;
520
47eff617 521 prog.caps = &obj->caps;
a5b8bd47
WN		 522 progs = obj->programs;
523 nr_progs = obj->nr_programs;
524
531b014e 525 progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
a5b8bd47
WN		 526 if (!progs) {
527 /*
528 * In this case the original obj->programs
529 * is still valid, so don't need special treat for
530 * bpf_close_object().
531 */
be18010e
KW		 532 pr_warn("failed to alloc a new program under section '%s'\n",
533 section_name);
a5b8bd47
WN		 534 bpf_program__exit(&prog);
535 return -ENOMEM;
536 }
537
538 pr_debug("found program %s\n", prog.section_name);
539 obj->programs = progs;
540 obj->nr_programs = nr_progs + 1;
aa9b1ac3 541 prog.obj = obj;
a5b8bd47
WN		 542 progs[nr_progs] = prog;
543 return 0;
544}
545
88cda1c9
MKL		 546static int
547bpf_object__init_prog_names(struct bpf_object *obj)
548{
549 Elf_Data *symbols = obj->efile.symbols;
550 struct bpf_program *prog;
551 size_t pi, si;
552
553 for (pi = 0; pi < obj->nr_programs; pi++) {
48cca7e4 554 const char *name = NULL;
88cda1c9
MKL		 555
556 prog = &obj->programs[pi];
557
558 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
559 si++) {
560 GElf_Sym sym;
561
562 if (!gelf_getsym(symbols, si, &sym))
563 continue;
564 if (sym.st_shndx != prog->idx)
565 continue;
fe4d44b2
RG		 566 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
567 continue;
88cda1c9
MKL		 568
569 name = elf_strptr(obj->efile.elf,
570 obj->efile.strtabidx,
571 sym.st_name);
572 if (!name) {
be18010e
KW		 573 pr_warn("failed to get sym name string for prog %s\n",
574 prog->section_name);
88cda1c9
MKL		 575 return -LIBBPF_ERRNO__LIBELF;
576 }
577 }
578
9a94f277
JK		 579 if (!name && prog->idx == obj->efile.text_shndx)
580 name = ".text";
581
88cda1c9 582 if (!name) {
be18010e
KW		 583 pr_warn("failed to find sym for prog %s\n",
584 prog->section_name);
88cda1c9
MKL		 585 return -EINVAL;
586 }
9a94f277 587
88cda1c9
MKL		 588 prog->name = strdup(name);
589 if (!prog->name) {
be18010e
KW		 590 pr_warn("failed to allocate memory for prog sym %s\n",
591 name);
88cda1c9
MKL		 592 return -ENOMEM;
593 }
594 }
595
596 return 0;
597}
598
5e61f270
AN		 599static __u32 get_kernel_version(void)
600{
601 __u32 major, minor, patch;
602 struct utsname info;
603
604 uname(&info);
605 if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
606 return 0;
607 return KERNEL_VERSION(major, minor, patch);
608}
609
590a0088
MKL		 610static const struct btf_member *
611find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
612{
613 struct btf_member *m;
614 int i;
615
616 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
617 if (btf_member_bit_offset(t, i) == bit_offset)
618 return m;
619 }
620
621 return NULL;
622}
623
624static const struct btf_member *
625find_member_by_name(const struct btf *btf, const struct btf_type *t,
626 const char *name)
627{
628 struct btf_member *m;
629 int i;
630
631 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
632 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
633 return m;
634 }
635
636 return NULL;
637}
638
639#define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
a6ed02ca
KS		 640static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
641 const char *name, __u32 kind);
590a0088
MKL		 642
643static int
644find_struct_ops_kern_types(const struct btf *btf, const char *tname,
645 const struct btf_type **type, __u32 *type_id,
646 const struct btf_type **vtype, __u32 *vtype_id,
647 const struct btf_member **data_member)
648{
649 const struct btf_type *kern_type, *kern_vtype;
650 const struct btf_member *kern_data_member;
651 __s32 kern_vtype_id, kern_type_id;
590a0088
MKL		 652 __u32 i;
653
654 kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
655 if (kern_type_id < 0) {
656 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
657 tname);
658 return kern_type_id;
659 }
660 kern_type = btf__type_by_id(btf, kern_type_id);
661
662 /* Find the corresponding "map_value" type that will be used
663 * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example,
664 * find "struct bpf_struct_ops_tcp_congestion_ops" from the
665 * btf_vmlinux.
666 */
a6ed02ca
KS		 667 kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
668 tname, BTF_KIND_STRUCT);
590a0088 669 if (kern_vtype_id < 0) {
a6ed02ca
KS		 670 pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
671 STRUCT_OPS_VALUE_PREFIX, tname);
590a0088
MKL		 672 return kern_vtype_id;
673 }
674 kern_vtype = btf__type_by_id(btf, kern_vtype_id);
675
676 /* Find "struct tcp_congestion_ops" from
677 * struct bpf_struct_ops_tcp_congestion_ops {
678 * [ ... ]
679 * struct tcp_congestion_ops data;
680 * }
681 */
682 kern_data_member = btf_members(kern_vtype);
683 for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
684 if (kern_data_member->type == kern_type_id)
685 break;
686 }
687 if (i == btf_vlen(kern_vtype)) {
a6ed02ca
KS		 688 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
689 tname, STRUCT_OPS_VALUE_PREFIX, tname);
590a0088
MKL		 690 return -EINVAL;
691 }
692
693 *type = kern_type;
694 *type_id = kern_type_id;
695 *vtype = kern_vtype;
696 *vtype_id = kern_vtype_id;
697 *data_member = kern_data_member;
698
699 return 0;
700}
701
702static bool bpf_map__is_struct_ops(const struct bpf_map *map)
703{
704 return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
705}
706
707/* Init the map's fields that depend on kern_btf */
708static int bpf_map__init_kern_struct_ops(struct bpf_map *map,
709 const struct btf *btf,
710 const struct btf *kern_btf)
711{
712 const struct btf_member *member, *kern_member, *kern_data_member;
713 const struct btf_type *type, *kern_type, *kern_vtype;
714 __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
715 struct bpf_struct_ops *st_ops;
716 void *data, *kern_data;
717 const char *tname;
718 int err;
719
720 st_ops = map->st_ops;
721 type = st_ops->type;
722 tname = st_ops->tname;
723 err = find_struct_ops_kern_types(kern_btf, tname,
724 &kern_type, &kern_type_id,
725 &kern_vtype, &kern_vtype_id,
726 &kern_data_member);
727 if (err)
728 return err;
729
730 pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
731 map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
732
733 map->def.value_size = kern_vtype->size;
734 map->btf_vmlinux_value_type_id = kern_vtype_id;
735
736 st_ops->kern_vdata = calloc(1, kern_vtype->size);
737 if (!st_ops->kern_vdata)
738 return -ENOMEM;
739
740 data = st_ops->data;
741 kern_data_off = kern_data_member->offset / 8;
742 kern_data = st_ops->kern_vdata + kern_data_off;
743
744 member = btf_members(type);
745 for (i = 0; i < btf_vlen(type); i++, member++) {
746 const struct btf_type *mtype, *kern_mtype;
747 __u32 mtype_id, kern_mtype_id;
748 void *mdata, *kern_mdata;
749 __s64 msize, kern_msize;
750 __u32 moff, kern_moff;
751 __u32 kern_member_idx;
752 const char *mname;
753
754 mname = btf__name_by_offset(btf, member->name_off);
755 kern_member = find_member_by_name(kern_btf, kern_type, mname);
756 if (!kern_member) {
757 pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
758 map->name, mname);
759 return -ENOTSUP;
760 }
761
762 kern_member_idx = kern_member - btf_members(kern_type);
763 if (btf_member_bitfield_size(type, i) ||
764 btf_member_bitfield_size(kern_type, kern_member_idx)) {
765 pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
766 map->name, mname);
767 return -ENOTSUP;
768 }
769
770 moff = member->offset / 8;
771 kern_moff = kern_member->offset / 8;
772
773 mdata = data + moff;
774 kern_mdata = kern_data + kern_moff;
775
776 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
777 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
778 &kern_mtype_id);
779 if (BTF_INFO_KIND(mtype->info) !=
780 BTF_INFO_KIND(kern_mtype->info)) {
781 pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
782 map->name, mname, BTF_INFO_KIND(mtype->info),
783 BTF_INFO_KIND(kern_mtype->info));
784 return -ENOTSUP;
785 }
786
787 if (btf_is_ptr(mtype)) {
788 struct bpf_program *prog;
789
790 mtype = skip_mods_and_typedefs(btf, mtype->type, &mtype_id);
791 kern_mtype = skip_mods_and_typedefs(kern_btf,
792 kern_mtype->type,
793 &kern_mtype_id);
794 if (!btf_is_func_proto(mtype) ||
795 !btf_is_func_proto(kern_mtype)) {
796 pr_warn("struct_ops init_kern %s: non func ptr %s is not supported\n",
797 map->name, mname);
798 return -ENOTSUP;
799 }
800
801 prog = st_ops->progs[i];
802 if (!prog) {
803 pr_debug("struct_ops init_kern %s: func ptr %s is not set\n",
804 map->name, mname);
805 continue;
806 }
807
808 prog->attach_btf_id = kern_type_id;
809 prog->expected_attach_type = kern_member_idx;
810
811 st_ops->kern_func_off[i] = kern_data_off + kern_moff;
812
813 pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
814 map->name, mname, prog->name, moff,
815 kern_moff);
816
817 continue;
818 }
819
820 msize = btf__resolve_size(btf, mtype_id);
821 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
822 if (msize < 0 || kern_msize < 0 || msize != kern_msize) {
823 pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
824 map->name, mname, (ssize_t)msize,
825 (ssize_t)kern_msize);
826 return -ENOTSUP;
827 }
828
829 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
830 map->name, mname, (unsigned int)msize,
831 moff, kern_moff);
832 memcpy(kern_mdata, mdata, msize);
833 }
834
835 return 0;
836}
837
838static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
839{
590a0088
MKL		 840 struct bpf_map *map;
841 size_t i;
842 int err;
843
844 for (i = 0; i < obj->nr_maps; i++) {
845 map = &obj->maps[i];
846
847 if (!bpf_map__is_struct_ops(map))
848 continue;
849
a6ed02ca
KS		 850 err = bpf_map__init_kern_struct_ops(map, obj->btf,
851 obj->btf_vmlinux);
852 if (err)
590a0088 853 return err;
590a0088
MKL		 854 }
855
590a0088
MKL		 856 return 0;
857}
858
859static int bpf_object__init_struct_ops_maps(struct bpf_object *obj)
860{
861 const struct btf_type *type, *datasec;
862 const struct btf_var_secinfo *vsi;
863 struct bpf_struct_ops *st_ops;
864 const char *tname, *var_name;
865 __s32 type_id, datasec_id;
866 const struct btf *btf;
867 struct bpf_map *map;
868 __u32 i;
869
870 if (obj->efile.st_ops_shndx == -1)
871 return 0;
872
873 btf = obj->btf;
874 datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC,
875 BTF_KIND_DATASEC);
876 if (datasec_id < 0) {
877 pr_warn("struct_ops init: DATASEC %s not found\n",
878 STRUCT_OPS_SEC);
879 return -EINVAL;
880 }
881
882 datasec = btf__type_by_id(btf, datasec_id);
883 vsi = btf_var_secinfos(datasec);
884 for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
885 type = btf__type_by_id(obj->btf, vsi->type);
886 var_name = btf__name_by_offset(obj->btf, type->name_off);
887
888 type_id = btf__resolve_type(obj->btf, vsi->type);
889 if (type_id < 0) {
890 pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
891 vsi->type, STRUCT_OPS_SEC);
892 return -EINVAL;
893 }
894
895 type = btf__type_by_id(obj->btf, type_id);
896 tname = btf__name_by_offset(obj->btf, type->name_off);
897 if (!tname[0]) {
898 pr_warn("struct_ops init: anonymous type is not supported\n");
899 return -ENOTSUP;
900 }
901 if (!btf_is_struct(type)) {
902 pr_warn("struct_ops init: %s is not a struct\n", tname);
903 return -EINVAL;
904 }
905
906 map = bpf_object__add_map(obj);
907 if (IS_ERR(map))
908 return PTR_ERR(map);
909
910 map->sec_idx = obj->efile.st_ops_shndx;
911 map->sec_offset = vsi->offset;
912 map->name = strdup(var_name);
913 if (!map->name)
914 return -ENOMEM;
915
916 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
917 map->def.key_size = sizeof(int);
918 map->def.value_size = type->size;
919 map->def.max_entries = 1;
920
921 map->st_ops = calloc(1, sizeof(*map->st_ops));
922 if (!map->st_ops)
923 return -ENOMEM;
924 st_ops = map->st_ops;
925 st_ops->data = malloc(type->size);
926 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
927 st_ops->kern_func_off = malloc(btf_vlen(type) *
928 sizeof(*st_ops->kern_func_off));
929 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
930 return -ENOMEM;
931
932 if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) {
933 pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
934 var_name, STRUCT_OPS_SEC);
935 return -EINVAL;
936 }
937
938 memcpy(st_ops->data,
939 obj->efile.st_ops_data->d_buf + vsi->offset,
940 type->size);
941 st_ops->tname = tname;
942 st_ops->type = type;
943 st_ops->type_id = type_id;
944
945 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
946 tname, type_id, var_name, vsi->offset);
947 }
948
949 return 0;
950}
951
6c956392 952static struct bpf_object *bpf_object__new(const char *path,
5e61f270 953 const void *obj_buf,
2ce8450e
AN		 954 size_t obj_buf_sz,
955 const char *obj_name)
1a5e3fb1
WN		 956{
957 struct bpf_object *obj;
d859900c 958 char *end;
1a5e3fb1
WN		 959
960 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
961 if (!obj) {
be18010e 962 pr_warn("alloc memory failed for %s\n", path);
6371ca3b 963 return ERR_PTR(-ENOMEM);
1a5e3fb1
WN		 964 }
965
966 strcpy(obj->path, path);
2ce8450e
AN		 967 if (obj_name) {
968 strncpy(obj->name, obj_name, sizeof(obj->name) - 1);
969 obj->name[sizeof(obj->name) - 1] = 0;
970 } else {
971 /* Using basename() GNU version which doesn't modify arg. */
972 strncpy(obj->name, basename((void *)path),
973 sizeof(obj->name) - 1);
974 end = strchr(obj->name, '.');
975 if (end)
976 *end = 0;
977 }
6c956392 978
d859900c 979 obj->efile.fd = -1;
6c956392 980 /*
76e1022b 981 * Caller of this function should also call
6c956392
WN		 982 * bpf_object__elf_finish() after data collection to return
983 * obj_buf to user. If not, we should duplicate the buffer to
984 * avoid user freeing them before elf finish.
985 */
986 obj->efile.obj_buf = obj_buf;
987 obj->efile.obj_buf_sz = obj_buf_sz;
666810e8 988 obj->efile.maps_shndx = -1;
abd29c93 989 obj->efile.btf_maps_shndx = -1;
d859900c
DB		 990 obj->efile.data_shndx = -1;
991 obj->efile.rodata_shndx = -1;
992 obj->efile.bss_shndx = -1;
590a0088 993 obj->efile.st_ops_shndx = -1;
81bfdd08 994 obj->kconfig_map_idx = -1;
6c956392 995
5e61f270 996 obj->kern_version = get_kernel_version();
52d3352e 997 obj->loaded = false;
9a208eff
WN		 998
999 INIT_LIST_HEAD(&obj->list);
1000 list_add(&obj->list, &bpf_objects_list);
1a5e3fb1
WN		 1001 return obj;
1002}
1003
1004static void bpf_object__elf_finish(struct bpf_object *obj)
1005{
1006 if (!obj_elf_valid(obj))
1007 return;
1008
1009 if (obj->efile.elf) {
1010 elf_end(obj->efile.elf);
1011 obj->efile.elf = NULL;
1012 }
bec7d68c 1013 obj->efile.symbols = NULL;
d859900c
DB		 1014 obj->efile.data = NULL;
1015 obj->efile.rodata = NULL;
1016 obj->efile.bss = NULL;
590a0088 1017 obj->efile.st_ops_data = NULL;
b62f06e8 1018
1f8e2bcb
AN		 1019 zfree(&obj->efile.reloc_sects);
1020 obj->efile.nr_reloc_sects = 0;
1a5e3fb1 1021 zclose(obj->efile.fd);
6c956392
WN		 1022 obj->efile.obj_buf = NULL;
1023 obj->efile.obj_buf_sz = 0;
1a5e3fb1
WN		 1024}
1025
1026static int bpf_object__elf_init(struct bpf_object *obj)
1027{
1028 int err = 0;
1029 GElf_Ehdr *ep;
1030
1031 if (obj_elf_valid(obj)) {
be18010e 1032 pr_warn("elf init: internal error\n");
6371ca3b 1033 return -LIBBPF_ERRNO__LIBELF;
1a5e3fb1
WN		 1034 }
1035
6c956392
WN		 1036 if (obj->efile.obj_buf_sz > 0) {
1037 /*
1038 * obj_buf should have been validated by
1039 * bpf_object__open_buffer().
1040 */
5e61f270 1041 obj->efile.elf = elf_memory((char *)obj->efile.obj_buf,
6c956392
WN		 1042 obj->efile.obj_buf_sz);
1043 } else {
1044 obj->efile.fd = open(obj->path, O_RDONLY);
1045 if (obj->efile.fd < 0) {
be5c5d4e 1046 char errmsg[STRERR_BUFSIZE], *cp;
1ce6a9fc 1047
be5c5d4e
AN		 1048 err = -errno;
1049 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
be18010e 1050 pr_warn("failed to open %s: %s\n", obj->path, cp);
be5c5d4e 1051 return err;
6c956392
WN		 1052 }
1053
1054 obj->efile.elf = elf_begin(obj->efile.fd,
76e1022b 1055 LIBBPF_ELF_C_READ_MMAP, NULL);
1a5e3fb1
WN		 1056 }
1057
1a5e3fb1 1058 if (!obj->efile.elf) {
be18010e 1059 pr_warn("failed to open %s as ELF file\n", obj->path);
6371ca3b 1060 err = -LIBBPF_ERRNO__LIBELF;
1a5e3fb1
WN		 1061 goto errout;
1062 }
1063
1064 if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
be18010e 1065 pr_warn("failed to get EHDR from %s\n", obj->path);
6371ca3b 1066 err = -LIBBPF_ERRNO__FORMAT;
1a5e3fb1
WN		 1067 goto errout;
1068 }
1069 ep = &obj->efile.ehdr;
1070
9b16137a 1071 /* Old LLVM set e_machine to EM_NONE */
76e1022b
AN		 1072 if (ep->e_type != ET_REL ||
1073 (ep->e_machine && ep->e_machine != EM_BPF)) {
be18010e 1074 pr_warn("%s is not an eBPF object file\n", obj->path);
6371ca3b 1075 err = -LIBBPF_ERRNO__FORMAT;
1a5e3fb1
WN		 1076 goto errout;
1077 }
1078
1079 return 0;
1080errout:
1081 bpf_object__elf_finish(obj);
1082 return err;
1083}
1084
12ef5634 1085static int bpf_object__check_endianness(struct bpf_object *obj)
cc4228d5 1086{
cdb2f920 1087#if __BYTE_ORDER == __LITTLE_ENDIAN
12ef5634
AN		 1088 if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
1089 return 0;
cdb2f920 1090#elif __BYTE_ORDER == __BIG_ENDIAN
12ef5634
AN		 1091 if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
1092 return 0;
1093#else
1094# error "Unrecognized __BYTE_ORDER__"
1095#endif
be18010e 1096 pr_warn("endianness mismatch.\n");
6371ca3b 1097 return -LIBBPF_ERRNO__ENDIAN;
cc4228d5
WN		 1098}
1099
cb1e5e96 1100static int
399dc65e 1101bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
cb1e5e96 1102{
399dc65e 1103 memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
cb1e5e96
WN		 1104 pr_debug("license of %s is %s\n", obj->path, obj->license);
1105 return 0;
1106}
1107
54b8625c
JF		 1108static int
1109bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1110{
1111 __u32 kver;
1112
1113 if (size != sizeof(kver)) {
be18010e 1114 pr_warn("invalid kver section in %s\n", obj->path);
54b8625c
JF		 1115 return -LIBBPF_ERRNO__FORMAT;
1116 }
1117 memcpy(&kver, data, sizeof(kver));
1118 obj->kern_version = kver;
1119 pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1120 return 0;
1121}
1122
addb9fc9
NS		 1123static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1124{
1125 if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1126 type == BPF_MAP_TYPE_HASH_OF_MAPS)
1127 return true;
1128 return false;
1129}
1130
1713d68b
DB		 1131static int bpf_object_search_section_size(const struct bpf_object *obj,
1132 const char *name, size_t *d_size)
1133{
1134 const GElf_Ehdr *ep = &obj->efile.ehdr;
1135 Elf *elf = obj->efile.elf;
1136 Elf_Scn *scn = NULL;
1137 int idx = 0;
1138
1139 while ((scn = elf_nextscn(elf, scn)) != NULL) {
1140 const char *sec_name;
1141 Elf_Data *data;
1142 GElf_Shdr sh;
1143
1144 idx++;
1145 if (gelf_getshdr(scn, &sh) != &sh) {
be18010e
KW		 1146 pr_warn("failed to get section(%d) header from %s\n",
1147 idx, obj->path);
1713d68b
DB		 1148 return -EIO;
1149 }
1150
1151 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1152 if (!sec_name) {
be18010e
KW		 1153 pr_warn("failed to get section(%d) name from %s\n",
1154 idx, obj->path);
1713d68b
DB		 1155 return -EIO;
1156 }
1157
1158 if (strcmp(name, sec_name))
1159 continue;
1160
1161 data = elf_getdata(scn, 0);
1162 if (!data) {
be18010e
KW		 1163 pr_warn("failed to get section(%d) data from %s(%s)\n",
1164 idx, name, obj->path);
1713d68b
DB		 1165 return -EIO;
1166 }
1167
1168 *d_size = data->d_size;
1169 return 0;
1170 }
1171
1172 return -ENOENT;
1173}
1174
1175int bpf_object__section_size(const struct bpf_object *obj, const char *name,
1176 __u32 *size)
1177{
1178 int ret = -ENOENT;
1179 size_t d_size;
1180
1181 *size = 0;
1182 if (!name) {
1183 return -EINVAL;
ac9d1389 1184 } else if (!strcmp(name, DATA_SEC)) {
1713d68b
DB		 1185 if (obj->efile.data)
1186 *size = obj->efile.data->d_size;
ac9d1389 1187 } else if (!strcmp(name, BSS_SEC)) {
1713d68b
DB		 1188 if (obj->efile.bss)
1189 *size = obj->efile.bss->d_size;
ac9d1389 1190 } else if (!strcmp(name, RODATA_SEC)) {
1713d68b
DB		 1191 if (obj->efile.rodata)
1192 *size = obj->efile.rodata->d_size;
590a0088
MKL		 1193 } else if (!strcmp(name, STRUCT_OPS_SEC)) {
1194 if (obj->efile.st_ops_data)
1195 *size = obj->efile.st_ops_data->d_size;
1713d68b
DB		 1196 } else {
1197 ret = bpf_object_search_section_size(obj, name, &d_size);
1198 if (!ret)
1199 *size = d_size;
1200 }
1201
1202 return *size ? 0 : ret;
1203}
1204
1205int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
1206 __u32 *off)
1207{
1208 Elf_Data *symbols = obj->efile.symbols;
1209 const char *sname;
1210 size_t si;
1211
1212 if (!name || !off)
1213 return -EINVAL;
1214
1215 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
1216 GElf_Sym sym;
1217
1218 if (!gelf_getsym(symbols, si, &sym))
1219 continue;
1220 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
1221 GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
1222 continue;
1223
1224 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1225 sym.st_name);
1226 if (!sname) {
be18010e
KW		 1227 pr_warn("failed to get sym name string for var %s\n",
1228 name);
1713d68b
DB		 1229 return -EIO;
1230 }
1231 if (strcmp(name, sname) == 0) {
1232 *off = sym.st_value;
1233 return 0;
1234 }
1235 }
1236
1237 return -ENOENT;
1238}
1239
bf829271 1240static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
d859900c 1241{
bf829271
AN		 1242 struct bpf_map *new_maps;
1243 size_t new_cap;
1244 int i;
1245
1246 if (obj->nr_maps < obj->maps_cap)
1247 return &obj->maps[obj->nr_maps++];
1248
95064979 1249 new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
bf829271
AN		 1250 new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
1251 if (!new_maps) {
be18010e 1252 pr_warn("alloc maps for object failed\n");
bf829271
AN		 1253 return ERR_PTR(-ENOMEM);
1254 }
1255
1256 obj->maps_cap = new_cap;
1257 obj->maps = new_maps;
1258
1259 /* zero out new maps */
1260 memset(obj->maps + obj->nr_maps, 0,
1261 (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
1262 /*
1263 * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
1264 * when failure (zclose won't close negative fd)).
1265 */
1266 for (i = obj->nr_maps; i < obj->maps_cap; i++) {
1267 obj->maps[i].fd = -1;
1268 obj->maps[i].inner_map_fd = -1;
1269 }
1270
1271 return &obj->maps[obj->nr_maps++];
d859900c
DB		 1272}
1273
eba9c5f4
AN		 1274static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1275{
1276 long page_sz = sysconf(_SC_PAGE_SIZE);
1277 size_t map_sz;
1278
c701917e 1279 map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries;
eba9c5f4
AN		 1280 map_sz = roundup(map_sz, page_sz);
1281 return map_sz;
1282}
1283
81bfdd08
AN		 1284static char *internal_map_name(struct bpf_object *obj,
1285 enum libbpf_map_type type)
1286{
113e6b7e 1287 char map_name[BPF_OBJ_NAME_LEN], *p;
81bfdd08
AN		 1288 const char *sfx = libbpf_type_to_btf_name[type];
1289 int sfx_len = max((size_t)7, strlen(sfx));
1290 int pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1,
1291 strlen(obj->name));
1292
1293 snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1294 sfx_len, libbpf_type_to_btf_name[type]);
1295
113e6b7e
THJ		 1296 /* sanitise map name to characters allowed by kernel */
1297 for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1298 if (!isalnum(*p) && *p != '_' && *p != '.')
1299 *p = '_';
1300
81bfdd08
AN		 1301 return strdup(map_name);
1302}
1303
d859900c 1304static int
bf829271 1305bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
eba9c5f4 1306 int sec_idx, void *data, size_t data_sz)
d859900c 1307{
bf829271
AN		 1308 struct bpf_map_def *def;
1309 struct bpf_map *map;
eba9c5f4 1310 int err;
bf829271
AN		 1311
1312 map = bpf_object__add_map(obj);
1313 if (IS_ERR(map))
1314 return PTR_ERR(map);
d859900c
DB		 1315
1316 map->libbpf_type = type;
db48814b
AN		 1317 map->sec_idx = sec_idx;
1318 map->sec_offset = 0;
81bfdd08 1319 map->name = internal_map_name(obj, type);
d859900c 1320 if (!map->name) {
be18010e 1321 pr_warn("failed to alloc map name\n");
d859900c
DB		 1322 return -ENOMEM;
1323 }
1324
bf829271 1325 def = &map->def;
d859900c
DB		 1326 def->type = BPF_MAP_TYPE_ARRAY;
1327 def->key_size = sizeof(int);
eba9c5f4 1328 def->value_size = data_sz;
d859900c 1329 def->max_entries = 1;
81bfdd08 1330 def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
166750bc 1331 ? BPF_F_RDONLY_PROG : 0;
0d13bfce 1332 def->map_flags |= BPF_F_MMAPABLE;
7fe74b43
AN		 1333
1334 pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
81bfdd08 1335 map->name, map->sec_idx, map->sec_offset, def->map_flags);
7fe74b43 1336
eba9c5f4
AN		 1337 map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
1338 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1339 if (map->mmaped == MAP_FAILED) {
1340 err = -errno;
1341 map->mmaped = NULL;
1342 pr_warn("failed to alloc map '%s' content buffer: %d\n",
1343 map->name, err);
1344 zfree(&map->name);
1345 return err;
d859900c
DB		 1346 }
1347
166750bc 1348 if (data)
eba9c5f4
AN		 1349 memcpy(map->mmaped, data, data_sz);
1350
e1d1dc46 1351 pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
d859900c
DB		 1352 return 0;
1353}
1354
bf829271
AN		 1355static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1356{
1357 int err;
1358
bf829271
AN		 1359 /*
1360 * Populate obj->maps with libbpf internal maps.
1361 */
1362 if (obj->efile.data_shndx >= 0) {
1363 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
db48814b 1364 obj->efile.data_shndx,
eba9c5f4
AN		 1365 obj->efile.data->d_buf,
1366 obj->efile.data->d_size);
bf829271
AN		 1367 if (err)
1368 return err;
1369 }
1370 if (obj->efile.rodata_shndx >= 0) {
1371 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
db48814b 1372 obj->efile.rodata_shndx,
eba9c5f4
AN		 1373 obj->efile.rodata->d_buf,
1374 obj->efile.rodata->d_size);
bf829271
AN		 1375 if (err)
1376 return err;
1377 }
1378 if (obj->efile.bss_shndx >= 0) {
1379 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
db48814b 1380 obj->efile.bss_shndx,
eba9c5f4
AN		 1381 NULL,
1382 obj->efile.bss->d_size);
bf829271
AN		 1383 if (err)
1384 return err;
1385 }
1386 return 0;
1387}
1388
166750bc
AN		 1389
1390static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1391 const void *name)
1392{
1393 int i;
1394
1395 for (i = 0; i < obj->nr_extern; i++) {
1396 if (strcmp(obj->externs[i].name, name) == 0)
1397 return &obj->externs[i];
1398 }
1399 return NULL;
1400}
1401
1402static int set_ext_value_tri(struct extern_desc *ext, void *ext_val,
1403 char value)
1404{
1405 switch (ext->type) {
1406 case EXT_BOOL:
1407 if (value == 'm') {
1408 pr_warn("extern %s=%c should be tristate or char\n",
1409 ext->name, value);
1410 return -EINVAL;
1411 }
1412 *(bool *)ext_val = value == 'y' ? true : false;
1413 break;
1414 case EXT_TRISTATE:
1415 if (value == 'y')
1416 *(enum libbpf_tristate *)ext_val = TRI_YES;
1417 else if (value == 'm')
1418 *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1419 else /* value == 'n' */
1420 *(enum libbpf_tristate *)ext_val = TRI_NO;
1421 break;
1422 case EXT_CHAR:
1423 *(char *)ext_val = value;
1424 break;
1425 case EXT_UNKNOWN:
1426 case EXT_INT:
1427 case EXT_CHAR_ARR:
1428 default:
1429 pr_warn("extern %s=%c should be bool, tristate, or char\n",
1430 ext->name, value);
1431 return -EINVAL;
1432 }
1433 ext->is_set = true;
1434 return 0;
1435}
1436
1437static int set_ext_value_str(struct extern_desc *ext, char *ext_val,
1438 const char *value)
1439{
1440 size_t len;
1441
1442 if (ext->type != EXT_CHAR_ARR) {
1443 pr_warn("extern %s=%s should char array\n", ext->name, value);
1444 return -EINVAL;
1445 }
1446
1447 len = strlen(value);
1448 if (value[len - 1] != '"') {
1449 pr_warn("extern '%s': invalid string config '%s'\n",
1450 ext->name, value);
1451 return -EINVAL;
1452 }
1453
1454 /* strip quotes */
1455 len -= 2;
1456 if (len >= ext->sz) {
1457 pr_warn("extern '%s': long string config %s of (%zu bytes) truncated to %d bytes\n",
1458 ext->name, value, len, ext->sz - 1);
1459 len = ext->sz - 1;
1460 }
1461 memcpy(ext_val, value + 1, len);
1462 ext_val[len] = '\0';
1463 ext->is_set = true;
1464 return 0;
1465}
1466
1467static int parse_u64(const char *value, __u64 *res)
1468{
1469 char *value_end;
1470 int err;
1471
1472 errno = 0;
1473 *res = strtoull(value, &value_end, 0);
1474 if (errno) {
1475 err = -errno;
1476 pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1477 return err;
1478 }
1479 if (*value_end) {
1480 pr_warn("failed to parse '%s' as integer completely\n", value);
1481 return -EINVAL;
1482 }
1483 return 0;
1484}
1485
1486static bool is_ext_value_in_range(const struct extern_desc *ext, __u64 v)
1487{
1488 int bit_sz = ext->sz * 8;
1489
1490 if (ext->sz == 8)
1491 return true;
1492
1493 /* Validate that value stored in u64 fits in integer of `ext->sz`
1494 * bytes size without any loss of information. If the target integer
1495 * is signed, we rely on the following limits of integer type of
1496 * Y bits and subsequent transformation:
1497 *
1498 * -2^(Y-1) <= X <= 2^(Y-1) - 1
1499 * 0 <= X + 2^(Y-1) <= 2^Y - 1
1500 * 0 <= X + 2^(Y-1) < 2^Y
1501 *
1502 * For unsigned target integer, check that all the (64 - Y) bits are
1503 * zero.
1504 */
1505 if (ext->is_signed)
1506 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1507 else
1508 return (v >> bit_sz) == 0;
1509}
1510
1511static int set_ext_value_num(struct extern_desc *ext, void *ext_val,
1512 __u64 value)
1513{
1514 if (ext->type != EXT_INT && ext->type != EXT_CHAR) {
1515 pr_warn("extern %s=%llu should be integer\n",
7745ff98 1516 ext->name, (unsigned long long)value);
166750bc
AN		 1517 return -EINVAL;
1518 }
1519 if (!is_ext_value_in_range(ext, value)) {
1520 pr_warn("extern %s=%llu value doesn't fit in %d bytes\n",
7745ff98 1521 ext->name, (unsigned long long)value, ext->sz);
166750bc
AN		 1522 return -ERANGE;
1523 }
1524 switch (ext->sz) {
1525 case 1: *(__u8 *)ext_val = value; break;
1526 case 2: *(__u16 *)ext_val = value; break;
1527 case 4: *(__u32 *)ext_val = value; break;
1528 case 8: *(__u64 *)ext_val = value; break;
1529 default:
1530 return -EINVAL;
1531 }
1532 ext->is_set = true;
1533 return 0;
1534}
1535
8601fd42
AN		 1536static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1537 char *buf, void *data)
166750bc 1538{
166750bc 1539 struct extern_desc *ext;
8601fd42 1540 char *sep, *value;
166750bc
AN		 1541 int len, err = 0;
1542 void *ext_val;
1543 __u64 num;
166750bc 1544
8601fd42
AN		 1545 if (strncmp(buf, "CONFIG_", 7))
1546 return 0;
166750bc 1547
8601fd42
AN		 1548 sep = strchr(buf, '=');
1549 if (!sep) {
1550 pr_warn("failed to parse '%s': no separator\n", buf);
1551 return -EINVAL;
1552 }
1553
1554 /* Trim ending '\n' */
1555 len = strlen(buf);
1556 if (buf[len - 1] == '\n')
1557 buf[len - 1] = '\0';
1558 /* Split on '=' and ensure that a value is present. */
1559 *sep = '\0';
1560 if (!sep[1]) {
1561 *sep = '=';
1562 pr_warn("failed to parse '%s': no value\n", buf);
1563 return -EINVAL;
1564 }
1565
1566 ext = find_extern_by_name(obj, buf);
1567 if (!ext || ext->is_set)
1568 return 0;
1569
1570 ext_val = data + ext->data_off;
1571 value = sep + 1;
1572
1573 switch (*value) {
1574 case 'y': case 'n': case 'm':
1575 err = set_ext_value_tri(ext, ext_val, *value);
1576 break;
1577 case '"':
1578 err = set_ext_value_str(ext, ext_val, value);
1579 break;
1580 default:
1581 /* assume integer */
1582 err = parse_u64(value, &num);
1583 if (err) {
1584 pr_warn("extern %s=%s should be integer\n",
1585 ext->name, value);
1586 return err;
1587 }
1588 err = set_ext_value_num(ext, ext_val, num);
1589 break;
166750bc 1590 }
8601fd42
AN		 1591 if (err)
1592 return err;
1593 pr_debug("extern %s=%s\n", ext->name, value);
1594 return 0;
1595}
1596
1597static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
1598{
1599 char buf[PATH_MAX];
1600 struct utsname uts;
1601 int len, err = 0;
1602 gzFile file;
1603
1604 uname(&uts);
1605 len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
1606 if (len < 0)
1607 return -EINVAL;
1608 else if (len >= PATH_MAX)
1609 return -ENAMETOOLONG;
1610
1611 /* gzopen also accepts uncompressed files. */
1612 file = gzopen(buf, "r");
1613 if (!file)
1614 file = gzopen("/proc/config.gz", "r");
1615
166750bc 1616 if (!file) {
8601fd42 1617 pr_warn("failed to open system Kconfig\n");
166750bc
AN		 1618 return -ENOENT;
1619 }
1620
1621 while (gzgets(file, buf, sizeof(buf))) {
8601fd42
AN		 1622 err = bpf_object__process_kconfig_line(obj, buf, data);
1623 if (err) {
1624 pr_warn("error parsing system Kconfig line '%s': %d\n",
1625 buf, err);
166750bc
AN		 1626 goto out;
1627 }
8601fd42 1628 }
166750bc 1629
8601fd42
AN		 1630out:
1631 gzclose(file);
1632 return err;
1633}
166750bc 1634
8601fd42
AN		 1635static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
1636 const char *config, void *data)
1637{
1638 char buf[PATH_MAX];
1639 int err = 0;
1640 FILE *file;
166750bc 1641
8601fd42
AN		 1642 file = fmemopen((void *)config, strlen(config), "r");
1643 if (!file) {
1644 err = -errno;
1645 pr_warn("failed to open in-memory Kconfig: %d\n", err);
1646 return err;
1647 }
1648
1649 while (fgets(buf, sizeof(buf), file)) {
1650 err = bpf_object__process_kconfig_line(obj, buf, data);
1651 if (err) {
1652 pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
1653 buf, err);
166750bc
AN		 1654 break;
1655 }
166750bc
AN		 1656 }
1657
8601fd42 1658 fclose(file);
166750bc
AN		 1659 return err;
1660}
1661
81bfdd08 1662static int bpf_object__init_kconfig_map(struct bpf_object *obj)
166750bc
AN		 1663{
1664 struct extern_desc *last_ext;
1665 size_t map_sz;
1666 int err;
1667
1668 if (obj->nr_extern == 0)
1669 return 0;
1670
1671 last_ext = &obj->externs[obj->nr_extern - 1];
1672 map_sz = last_ext->data_off + last_ext->sz;
1673
81bfdd08 1674 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
166750bc
AN		 1675 obj->efile.symbols_shndx,
1676 NULL, map_sz);
1677 if (err)
1678 return err;
1679
81bfdd08 1680 obj->kconfig_map_idx = obj->nr_maps - 1;
166750bc
AN		 1681
1682 return 0;
1683}
1684
bf829271 1685static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
561bbcca 1686{
561bbcca 1687 Elf_Data *symbols = obj->efile.symbols;
bf829271 1688 int i, map_def_sz = 0, nr_maps = 0, nr_syms;
d859900c 1689 Elf_Data *data = NULL;
bf829271
AN		 1690 Elf_Scn *scn;
1691
1692 if (obj->efile.maps_shndx < 0)
1693 return 0;
561bbcca 1694
4708bbda
EL		 1695 if (!symbols)
1696 return -EINVAL;
1697
bf829271
AN		 1698 scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
1699 if (scn)
1700 data = elf_getdata(scn, NULL);
1701 if (!scn || !data) {
be18010e
KW		 1702 pr_warn("failed to get Elf_Data from map section %d\n",
1703 obj->efile.maps_shndx);
bf829271 1704 return -EINVAL;
4708bbda 1705 }
561bbcca 1706
4708bbda
EL		 1707 /*
1708 * Count number of maps. Each map has a name.
1709 * Array of maps is not supported: only the first element is
1710 * considered.
1711 *
1712 * TODO: Detect array of map and report error.
1713 */
bf829271
AN		 1714 nr_syms = symbols->d_size / sizeof(GElf_Sym);
1715 for (i = 0; i < nr_syms; i++) {
561bbcca 1716 GElf_Sym sym;
4708bbda
EL		 1717
1718 if (!gelf_getsym(symbols, i, &sym))
1719 continue;
1720 if (sym.st_shndx != obj->efile.maps_shndx)
1721 continue;
1722 nr_maps++;
1723 }
b13c5c14 1724 /* Assume equally sized map definitions */
bf829271
AN		 1725 pr_debug("maps in %s: %d maps in %zd bytes\n",
1726 obj->path, nr_maps, data->d_size);
1727
98e527af 1728 if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) {
8983b731 1729 pr_warn("unable to determine map definition size section %s, %d maps in %zd bytes\n",
be18010e 1730 obj->path, nr_maps, data->d_size);
bf829271 1731 return -EINVAL;
addb9fc9 1732 }
98e527af 1733 map_def_sz = data->d_size / nr_maps;
4708bbda 1734
bf829271
AN		 1735 /* Fill obj->maps using data in "maps" section. */
1736 for (i = 0; i < nr_syms; i++) {
4708bbda 1737 GElf_Sym sym;
561bbcca 1738 const char *map_name;
4708bbda 1739 struct bpf_map_def *def;
bf829271 1740 struct bpf_map *map;
561bbcca
WN		 1741
1742 if (!gelf_getsym(symbols, i, &sym))
1743 continue;
666810e8 1744 if (sym.st_shndx != obj->efile.maps_shndx)
561bbcca
WN		 1745 continue;
1746
bf829271
AN		 1747 map = bpf_object__add_map(obj);
1748 if (IS_ERR(map))
1749 return PTR_ERR(map);
1750
1751 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
561bbcca 1752 sym.st_name);
c51829bb 1753 if (!map_name) {
be18010e
KW		 1754 pr_warn("failed to get map #%d name sym string for obj %s\n",
1755 i, obj->path);
c51829bb
AN		 1756 return -LIBBPF_ERRNO__FORMAT;
1757 }
d859900c 1758
bf829271 1759 map->libbpf_type = LIBBPF_MAP_UNSPEC;
db48814b
AN		 1760 map->sec_idx = sym.st_shndx;
1761 map->sec_offset = sym.st_value;
1762 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
1763 map_name, map->sec_idx, map->sec_offset);
b13c5c14 1764 if (sym.st_value + map_def_sz > data->d_size) {
be18010e
KW		 1765 pr_warn("corrupted maps section in %s: last map \"%s\" too small\n",
1766 obj->path, map_name);
4708bbda 1767 return -EINVAL;
561bbcca 1768 }
4708bbda 1769
bf829271
AN		 1770 map->name = strdup(map_name);
1771 if (!map->name) {
be18010e 1772 pr_warn("failed to alloc map name\n");
973170e6
WN		 1773 return -ENOMEM;
1774 }
bf829271 1775 pr_debug("map %d is \"%s\"\n", i, map->name);
4708bbda 1776 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
b13c5c14
CG		 1777 /*
1778 * If the definition of the map in the object file fits in
1779 * bpf_map_def, copy it. Any extra fields in our version
1780 * of bpf_map_def will default to zero as a result of the
1781 * calloc above.
1782 */
1783 if (map_def_sz <= sizeof(struct bpf_map_def)) {
bf829271 1784 memcpy(&map->def, def, map_def_sz);
b13c5c14
CG		 1785 } else {
1786 /*
1787 * Here the map structure being read is bigger than what
1788 * we expect, truncate if the excess bits are all zero.
1789 * If they are not zero, reject this map as
1790 * incompatible.
1791 */
1792 char *b;
8983b731 1793
b13c5c14
CG		 1794 for (b = ((char *)def) + sizeof(struct bpf_map_def);
1795 b < ((char *)def) + map_def_sz; b++) {
1796 if (*b != 0) {
8983b731 1797 pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n",
be18010e 1798 obj->path, map_name);
c034a177
JF		 1799 if (strict)
1800 return -EINVAL;
b13c5c14
CG		 1801 }
1802 }
bf829271 1803 memcpy(&map->def, def, sizeof(struct bpf_map_def));
b13c5c14 1804 }
561bbcca 1805 }
bf829271
AN		 1806 return 0;
1807}
4708bbda 1808
ddc7c304
AN		 1809static const struct btf_type *
1810skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
abd29c93
AN		 1811{
1812 const struct btf_type *t = btf__type_by_id(btf, id);
8837fe5d 1813
ddc7c304
AN		 1814 if (res_id)
1815 *res_id = id;
1816
1817 while (btf_is_mod(t) || btf_is_typedef(t)) {
1818 if (res_id)
1819 *res_id = t->type;
1820 t = btf__type_by_id(btf, t->type);
abd29c93 1821 }
ddc7c304
AN		 1822
1823 return t;
abd29c93
AN		 1824}
1825
590a0088
MKL		 1826static const struct btf_type *
1827resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
1828{
1829 const struct btf_type *t;
1830
1831 t = skip_mods_and_typedefs(btf, id, NULL);
1832 if (!btf_is_ptr(t))
1833 return NULL;
1834
1835 t = skip_mods_and_typedefs(btf, t->type, res_id);
1836
1837 return btf_is_func_proto(t) ? t : NULL;
1838}
1839
ef99b02b
AN		 1840/*
1841 * Fetch integer attribute of BTF map definition. Such attributes are
1842 * represented using a pointer to an array, in which dimensionality of array
1843 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1844 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1845 * type definition, while using only sizeof(void *) space in ELF data section.
1846 */
1847static bool get_map_field_int(const char *map_name, const struct btf *btf,
abd29c93 1848 const struct btf_type *def,
8983b731
AN		 1849 const struct btf_member *m, __u32 *res)
1850{
ddc7c304 1851 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
abd29c93 1852 const char *name = btf__name_by_offset(btf, m->name_off);
ef99b02b
AN		 1853 const struct btf_array *arr_info;
1854 const struct btf_type *arr_t;
abd29c93 1855
b03bc685 1856 if (!btf_is_ptr(t)) {
be18010e
KW		 1857 pr_warn("map '%s': attr '%s': expected PTR, got %u.\n",
1858 map_name, name, btf_kind(t));
abd29c93
AN		 1859 return false;
1860 }
ef99b02b
AN		 1861
1862 arr_t = btf__type_by_id(btf, t->type);
1863 if (!arr_t) {
be18010e
KW		 1864 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
1865 map_name, name, t->type);
abd29c93
AN		 1866 return false;
1867 }
b03bc685 1868 if (!btf_is_array(arr_t)) {
be18010e
KW		 1869 pr_warn("map '%s': attr '%s': expected ARRAY, got %u.\n",
1870 map_name, name, btf_kind(arr_t));
abd29c93
AN		 1871 return false;
1872 }
b03bc685 1873 arr_info = btf_array(arr_t);
ef99b02b 1874 *res = arr_info->nelems;
abd29c93
AN		 1875 return true;
1876}
1877
57a00f41
THJ		 1878static int build_map_pin_path(struct bpf_map *map, const char *path)
1879{
1880 char buf[PATH_MAX];
1881 int err, len;
1882
1883 if (!path)
1884 path = "/sys/fs/bpf";
1885
1886 len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map));
1887 if (len < 0)
1888 return -EINVAL;
1889 else if (len >= PATH_MAX)
1890 return -ENAMETOOLONG;
1891
1892 err = bpf_map__set_pin_path(map, buf);
1893 if (err)
1894 return err;
1895
1896 return 0;
1897}
1898
abd29c93
AN		 1899static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1900 const struct btf_type *sec,
1901 int var_idx, int sec_idx,
57a00f41
THJ		 1902 const Elf_Data *data, bool strict,
1903 const char *pin_root_path)
abd29c93
AN		 1904{
1905 const struct btf_type *var, *def, *t;
1906 const struct btf_var_secinfo *vi;
1907 const struct btf_var *var_extra;
1908 const struct btf_member *m;
abd29c93
AN		 1909 const char *map_name;
1910 struct bpf_map *map;
1911 int vlen, i;
1912
b03bc685 1913 vi = btf_var_secinfos(sec) + var_idx;
abd29c93 1914 var = btf__type_by_id(obj->btf, vi->type);
b03bc685 1915 var_extra = btf_var(var);
abd29c93 1916 map_name = btf__name_by_offset(obj->btf, var->name_off);
b03bc685 1917 vlen = btf_vlen(var);
abd29c93
AN		 1918
1919 if (map_name == NULL || map_name[0] == '\0') {
be18010e 1920 pr_warn("map #%d: empty name.\n", var_idx);
abd29c93
AN		 1921 return -EINVAL;
1922 }
1923 if ((__u64)vi->offset + vi->size > data->d_size) {
be18010e 1924 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
abd29c93
AN		 1925 return -EINVAL;
1926 }
b03bc685 1927 if (!btf_is_var(var)) {
be18010e
KW		 1928 pr_warn("map '%s': unexpected var kind %u.\n",
1929 map_name, btf_kind(var));
abd29c93
AN		 1930 return -EINVAL;
1931 }
1932 if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1933 var_extra->linkage != BTF_VAR_STATIC) {
be18010e
KW		 1934 pr_warn("map '%s': unsupported var linkage %u.\n",
1935 map_name, var_extra->linkage);
abd29c93
AN		 1936 return -EOPNOTSUPP;
1937 }
1938
ddc7c304 1939 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
b03bc685 1940 if (!btf_is_struct(def)) {
be18010e
KW		 1941 pr_warn("map '%s': unexpected def kind %u.\n",
1942 map_name, btf_kind(var));
abd29c93
AN		 1943 return -EINVAL;
1944 }
1945 if (def->size > vi->size) {
be18010e 1946 pr_warn("map '%s': invalid def size.\n", map_name);
abd29c93
AN		 1947 return -EINVAL;
1948 }
1949
1950 map = bpf_object__add_map(obj);
1951 if (IS_ERR(map))
1952 return PTR_ERR(map);
1953 map->name = strdup(map_name);
1954 if (!map->name) {
be18010e 1955 pr_warn("map '%s': failed to alloc map name.\n", map_name);
abd29c93
AN		 1956 return -ENOMEM;
1957 }
1958 map->libbpf_type = LIBBPF_MAP_UNSPEC;
1959 map->def.type = BPF_MAP_TYPE_UNSPEC;
1960 map->sec_idx = sec_idx;
1961 map->sec_offset = vi->offset;
1962 pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1963 map_name, map->sec_idx, map->sec_offset);
1964
b03bc685
AN		 1965 vlen = btf_vlen(def);
1966 m = btf_members(def);
abd29c93
AN		 1967 for (i = 0; i < vlen; i++, m++) {
1968 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1969
1970 if (!name) {
be18010e 1971 pr_warn("map '%s': invalid field #%d.\n", map_name, i);
abd29c93
AN		 1972 return -EINVAL;
1973 }
1974 if (strcmp(name, "type") == 0) {
1975 if (!get_map_field_int(map_name, obj->btf, def, m,
ef99b02b 1976 &map->def.type))
abd29c93
AN		 1977 return -EINVAL;
1978 pr_debug("map '%s': found type = %u.\n",
1979 map_name, map->def.type);
1980 } else if (strcmp(name, "max_entries") == 0) {
1981 if (!get_map_field_int(map_name, obj->btf, def, m,
ef99b02b 1982 &map->def.max_entries))
abd29c93
AN		 1983 return -EINVAL;
1984 pr_debug("map '%s': found max_entries = %u.\n",
1985 map_name, map->def.max_entries);
1986 } else if (strcmp(name, "map_flags") == 0) {
1987 if (!get_map_field_int(map_name, obj->btf, def, m,
ef99b02b 1988 &map->def.map_flags))
abd29c93
AN		 1989 return -EINVAL;
1990 pr_debug("map '%s': found map_flags = %u.\n",
1991 map_name, map->def.map_flags);
1992 } else if (strcmp(name, "key_size") == 0) {
1993 __u32 sz;
1994
1995 if (!get_map_field_int(map_name, obj->btf, def, m,
ef99b02b 1996 &sz))
abd29c93
AN		 1997 return -EINVAL;
1998 pr_debug("map '%s': found key_size = %u.\n",
1999 map_name, sz);
2000 if (map->def.key_size && map->def.key_size != sz) {
be18010e
KW		 2001 pr_warn("map '%s': conflicting key size %u != %u.\n",
2002 map_name, map->def.key_size, sz);
abd29c93
AN		 2003 return -EINVAL;
2004 }
2005 map->def.key_size = sz;
2006 } else if (strcmp(name, "key") == 0) {
2007 __s64 sz;
2008
2009 t = btf__type_by_id(obj->btf, m->type);
2010 if (!t) {
be18010e
KW		 2011 pr_warn("map '%s': key type [%d] not found.\n",
2012 map_name, m->type);
abd29c93
AN		 2013 return -EINVAL;
2014 }
b03bc685 2015 if (!btf_is_ptr(t)) {
be18010e
KW		 2016 pr_warn("map '%s': key spec is not PTR: %u.\n",
2017 map_name, btf_kind(t));
abd29c93
AN		 2018 return -EINVAL;
2019 }
2020 sz = btf__resolve_size(obj->btf, t->type);
2021 if (sz < 0) {
679152d3
AN		 2022 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2023 map_name, t->type, (ssize_t)sz);
abd29c93
AN		 2024 return sz;
2025 }
679152d3
AN		 2026 pr_debug("map '%s': found key [%u], sz = %zd.\n",
2027 map_name, t->type, (ssize_t)sz);
abd29c93 2028 if (map->def.key_size && map->def.key_size != sz) {
679152d3
AN		 2029 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2030 map_name, map->def.key_size, (ssize_t)sz);
abd29c93
AN		 2031 return -EINVAL;
2032 }
2033 map->def.key_size = sz;
2034 map->btf_key_type_id = t->type;
2035 } else if (strcmp(name, "value_size") == 0) {
2036 __u32 sz;
2037
2038 if (!get_map_field_int(map_name, obj->btf, def, m,
ef99b02b 2039 &sz))
abd29c93
AN		 2040 return -EINVAL;
2041 pr_debug("map '%s': found value_size = %u.\n",
2042 map_name, sz);
2043 if (map->def.value_size && map->def.value_size != sz) {
be18010e
KW		 2044 pr_warn("map '%s': conflicting value size %u != %u.\n",
2045 map_name, map->def.value_size, sz);
abd29c93
AN		 2046 return -EINVAL;
2047 }
2048 map->def.value_size = sz;
2049 } else if (strcmp(name, "value") == 0) {
2050 __s64 sz;
2051
2052 t = btf__type_by_id(obj->btf, m->type);
2053 if (!t) {
be18010e
KW		 2054 pr_warn("map '%s': value type [%d] not found.\n",
2055 map_name, m->type);
abd29c93
AN		 2056 return -EINVAL;
2057 }
b03bc685 2058 if (!btf_is_ptr(t)) {
be18010e
KW		 2059 pr_warn("map '%s': value spec is not PTR: %u.\n",
2060 map_name, btf_kind(t));
abd29c93
AN		 2061 return -EINVAL;
2062 }
2063 sz = btf__resolve_size(obj->btf, t->type);
2064 if (sz < 0) {
679152d3
AN		 2065 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2066 map_name, t->type, (ssize_t)sz);
abd29c93
AN		 2067 return sz;
2068 }
679152d3
AN		 2069 pr_debug("map '%s': found value [%u], sz = %zd.\n",
2070 map_name, t->type, (ssize_t)sz);
abd29c93 2071 if (map->def.value_size && map->def.value_size != sz) {
679152d3
AN		 2072 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2073 map_name, map->def.value_size, (ssize_t)sz);
abd29c93
AN		 2074 return -EINVAL;
2075 }
2076 map->def.value_size = sz;
2077 map->btf_value_type_id = t->type;
57a00f41
THJ		 2078 } else if (strcmp(name, "pinning") == 0) {
2079 __u32 val;
2080 int err;
2081
2082 if (!get_map_field_int(map_name, obj->btf, def, m,
2083 &val))
2084 return -EINVAL;
2085 pr_debug("map '%s': found pinning = %u.\n",
2086 map_name, val);
2087
2088 if (val != LIBBPF_PIN_NONE &&
2089 val != LIBBPF_PIN_BY_NAME) {
2090 pr_warn("map '%s': invalid pinning value %u.\n",
2091 map_name, val);
2092 return -EINVAL;
2093 }
2094 if (val == LIBBPF_PIN_BY_NAME) {
2095 err = build_map_pin_path(map, pin_root_path);
2096 if (err) {
2097 pr_warn("map '%s': couldn't build pin path.\n",
2098 map_name);
2099 return err;
2100 }
2101 }
abd29c93
AN		 2102 } else {
2103 if (strict) {
be18010e
KW		 2104 pr_warn("map '%s': unknown field '%s'.\n",
2105 map_name, name);
abd29c93
AN		 2106 return -ENOTSUP;
2107 }
2108 pr_debug("map '%s': ignoring unknown field '%s'.\n",
2109 map_name, name);
2110 }
2111 }
2112
2113 if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
be18010e 2114 pr_warn("map '%s': map type isn't specified.\n", map_name);
abd29c93
AN		 2115 return -EINVAL;
2116 }
2117
2118 return 0;
2119}
2120
57a00f41
THJ		 2121static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2122 const char *pin_root_path)
abd29c93
AN		 2123{
2124 const struct btf_type *sec = NULL;
2125 int nr_types, i, vlen, err;
2126 const struct btf_type *t;
2127 const char *name;
2128 Elf_Data *data;
2129 Elf_Scn *scn;
2130
2131 if (obj->efile.btf_maps_shndx < 0)
2132 return 0;
2133
2134 scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
2135 if (scn)
2136 data = elf_getdata(scn, NULL);
2137 if (!scn || !data) {
be18010e
KW		 2138 pr_warn("failed to get Elf_Data from map section %d (%s)\n",
2139 obj->efile.maps_shndx, MAPS_ELF_SEC);
abd29c93
AN		 2140 return -EINVAL;
2141 }
2142
2143 nr_types = btf__get_nr_types(obj->btf);
2144 for (i = 1; i <= nr_types; i++) {
2145 t = btf__type_by_id(obj->btf, i);
b03bc685 2146 if (!btf_is_datasec(t))
abd29c93
AN		 2147 continue;
2148 name = btf__name_by_offset(obj->btf, t->name_off);
2149 if (strcmp(name, MAPS_ELF_SEC) == 0) {
2150 sec = t;
2151 break;
2152 }
2153 }
2154
2155 if (!sec) {
be18010e 2156 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
abd29c93
AN		 2157 return -ENOENT;
2158 }
2159
b03bc685 2160 vlen = btf_vlen(sec);
abd29c93
AN		 2161 for (i = 0; i < vlen; i++) {
2162 err = bpf_object__init_user_btf_map(obj, sec, i,
2163 obj->efile.btf_maps_shndx,
8983b731
AN		 2164 data, strict,
2165 pin_root_path);
abd29c93
AN		 2166 if (err)
2167 return err;
2168 }
2169
2170 return 0;
2171}
2172
0d13bfce 2173static int bpf_object__init_maps(struct bpf_object *obj,
01af3bf0 2174 const struct bpf_object_open_opts *opts)
bf829271 2175{
166750bc
AN		 2176 const char *pin_root_path;
2177 bool strict;
bf829271 2178 int err;
8837fe5d 2179
166750bc
AN		 2180 strict = !OPTS_GET(opts, relaxed_maps, false);
2181 pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
bf829271 2182
166750bc
AN		 2183 err = bpf_object__init_user_maps(obj, strict);
2184 err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
2185 err = err ?: bpf_object__init_global_data_maps(obj);
81bfdd08 2186 err = err ?: bpf_object__init_kconfig_map(obj);
590a0088 2187 err = err ?: bpf_object__init_struct_ops_maps(obj);
bf829271
AN		 2188 if (err)
2189 return err;
2190
bf829271 2191 return 0;
561bbcca
WN		 2192}
2193
e3d91b0c
JDB		 2194static bool section_have_execinstr(struct bpf_object *obj, int idx)
2195{
2196 Elf_Scn *scn;
2197 GElf_Shdr sh;
2198
2199 scn = elf_getscn(obj->efile.elf, idx);
2200 if (!scn)
2201 return false;
2202
2203 if (gelf_getshdr(scn, &sh) != &sh)
2204 return false;
2205
2206 if (sh.sh_flags & SHF_EXECINSTR)
2207 return true;
2208
2209 return false;
2210}
2211
d7c4b398
AN		 2212static void bpf_object__sanitize_btf(struct bpf_object *obj)
2213{
2d3eb67f 2214 bool has_func_global = obj->caps.btf_func_global;
d7c4b398
AN		 2215 bool has_datasec = obj->caps.btf_datasec;
2216 bool has_func = obj->caps.btf_func;
2217 struct btf *btf = obj->btf;
2218 struct btf_type *t;
2219 int i, j, vlen;
d7c4b398 2220
2d3eb67f 2221 if (!obj->btf || (has_func && has_datasec && has_func_global))
d7c4b398
AN		 2222 return;
2223
2224 for (i = 1; i <= btf__get_nr_types(btf); i++) {
2225 t = (struct btf_type *)btf__type_by_id(btf, i);
d7c4b398 2226
b03bc685 2227 if (!has_datasec && btf_is_var(t)) {
d7c4b398
AN		 2228 /* replace VAR with INT */
2229 t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1d4126c4
AN		 2230 /*
2231 * using size = 1 is the safest choice, 4 will be too
2232 * big and cause kernel BTF validation failure if
2233 * original variable took less than 4 bytes
2234 */
2235 t->size = 1;
708852dc 2236 *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
b03bc685 2237 } else if (!has_datasec && btf_is_datasec(t)) {
d7c4b398 2238 /* replace DATASEC with STRUCT */
b03bc685
AN		 2239 const struct btf_var_secinfo *v = btf_var_secinfos(t);
2240 struct btf_member *m = btf_members(t);
d7c4b398
AN		 2241 struct btf_type *vt;
2242 char *name;
2243
2244 name = (char *)btf__name_by_offset(btf, t->name_off);
2245 while (*name) {
2246 if (*name == '.')
2247 *name = '_';
2248 name++;
2249 }
2250
b03bc685 2251 vlen = btf_vlen(t);
d7c4b398
AN		 2252 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
2253 for (j = 0; j < vlen; j++, v++, m++) {
2254 /* order of field assignments is important */
2255 m->offset = v->offset * 8;
2256 m->type = v->type;
2257 /* preserve variable name as member name */
2258 vt = (void *)btf__type_by_id(btf, v->type);
2259 m->name_off = vt->name_off;
2260 }
b03bc685 2261 } else if (!has_func && btf_is_func_proto(t)) {
d7c4b398 2262 /* replace FUNC_PROTO with ENUM */
b03bc685 2263 vlen = btf_vlen(t);
d7c4b398
AN		 2264 t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
2265 t->size = sizeof(__u32); /* kernel enforced */
b03bc685 2266 } else if (!has_func && btf_is_func(t)) {
d7c4b398
AN		 2267 /* replace FUNC with TYPEDEF */
2268 t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
2d3eb67f
AS		 2269 } else if (!has_func_global && btf_is_func(t)) {
2270 /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
2271 t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
d7c4b398
AN		 2272 }
2273 }
2274}
2275
2276static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
2277{
2278 if (!obj->btf_ext)
2279 return;
2280
2281 if (!obj->caps.btf_func) {
2282 btf_ext__free(obj->btf_ext);
2283 obj->btf_ext = NULL;
2284 }
2285}
2286
abd29c93
AN		 2287static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
2288{
53276446 2289 return obj->efile.st_ops_shndx >= 0 || obj->nr_extern > 0;
abd29c93
AN		 2290}
2291
063183bf 2292static int bpf_object__init_btf(struct bpf_object *obj,
9c6660d0
AN		 2293 Elf_Data *btf_data,
2294 Elf_Data *btf_ext_data)
2295{
b7d7f3e1 2296 int err = -ENOENT;
9c6660d0
AN		 2297
2298 if (btf_data) {
2299 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
2300 if (IS_ERR(obj->btf)) {
b7d7f3e1
AN		 2301 err = PTR_ERR(obj->btf);
2302 obj->btf = NULL;
be18010e
KW		 2303 pr_warn("Error loading ELF section %s: %d.\n",
2304 BTF_ELF_SEC, err);
9c6660d0
AN		 2305 goto out;
2306 }
b7d7f3e1 2307 err = 0;
9c6660d0
AN		 2308 }
2309 if (btf_ext_data) {
2310 if (!obj->btf) {
2311 pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2312 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2313 goto out;
2314 }
2315 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
2316 btf_ext_data->d_size);
2317 if (IS_ERR(obj->btf_ext)) {
be18010e
KW		 2318 pr_warn("Error loading ELF section %s: %ld. Ignored and continue.\n",
2319 BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
9c6660d0
AN		 2320 obj->btf_ext = NULL;
2321 goto out;
2322 }
9c6660d0
AN		 2323 }
2324out:
b7d7f3e1 2325 if (err && bpf_object__is_btf_mandatory(obj)) {
be18010e 2326 pr_warn("BTF is required, but is missing or corrupted.\n");
b7d7f3e1 2327 return err;
abd29c93 2328 }
9c6660d0
AN		 2329 return 0;
2330}
2331
166750bc
AN		 2332static int bpf_object__finalize_btf(struct bpf_object *obj)
2333{
2334 int err;
2335
2336 if (!obj->btf)
2337 return 0;
2338
2339 err = btf__finalize_data(obj, obj->btf);
2340 if (!err)
2341 return 0;
2342
2343 pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err);
2344 btf__free(obj->btf);
2345 obj->btf = NULL;
2346 btf_ext__free(obj->btf_ext);
2347 obj->btf_ext = NULL;
2348
2349 if (bpf_object__is_btf_mandatory(obj)) {
2350 pr_warn("BTF is required, but is missing or corrupted.\n");
2351 return -ENOENT;
2352 }
2353 return 0;
2354}
2355
a6ed02ca
KS		 2356static inline bool libbpf_prog_needs_vmlinux_btf(struct bpf_program *prog)
2357{
2358 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS)
2359 return true;
2360
2361 /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
2362 * also need vmlinux BTF
2363 */
2364 if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
2365 return true;
2366
2367 return false;
2368}
2369
2370static int bpf_object__load_vmlinux_btf(struct bpf_object *obj)
2371{
2372 struct bpf_program *prog;
2373 int err;
2374
2375 bpf_object__for_each_program(prog, obj) {
2376 if (libbpf_prog_needs_vmlinux_btf(prog)) {
2377 obj->btf_vmlinux = libbpf_find_kernel_btf();
2378 if (IS_ERR(obj->btf_vmlinux)) {
2379 err = PTR_ERR(obj->btf_vmlinux);
2380 pr_warn("Error loading vmlinux BTF: %d\n", err);
2381 obj->btf_vmlinux = NULL;
2382 return err;
2383 }
2384 return 0;
2385 }
2386 }
2387
2388 return 0;
2389}
2390
063183bf
AN		 2391static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
2392{
2393 int err = 0;
2394
2395 if (!obj->btf)
2396 return 0;
2397
2398 bpf_object__sanitize_btf(obj);
2399 bpf_object__sanitize_btf_ext(obj);
2400
2401 err = btf__load(obj->btf);
2402 if (err) {
be18010e
KW		 2403 pr_warn("Error loading %s into kernel: %d.\n",
2404 BTF_ELF_SEC, err);
063183bf
AN		 2405 btf__free(obj->btf);
2406 obj->btf = NULL;
04efe591
AN		 2407 /* btf_ext can't exist without btf, so free it as well */
2408 if (obj->btf_ext) {
2409 btf_ext__free(obj->btf_ext);
2410 obj->btf_ext = NULL;
2411 }
2412
abd29c93
AN		 2413 if (bpf_object__is_btf_mandatory(obj))
2414 return err;
063183bf
AN		 2415 }
2416 return 0;
2417}
2418
0d13bfce 2419static int bpf_object__elf_collect(struct bpf_object *obj)
29603665
WN		 2420{
2421 Elf *elf = obj->efile.elf;
2422 GElf_Ehdr *ep = &obj->efile.ehdr;
f0187f0b 2423 Elf_Data *btf_ext_data = NULL;
1713d68b 2424 Elf_Data *btf_data = NULL;
29603665 2425 Elf_Scn *scn = NULL;
666810e8 2426 int idx = 0, err = 0;
29603665
WN		 2427
2428 /* Elf is corrupted/truncated, avoid calling elf_strptr. */
2429 if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
be18010e 2430 pr_warn("failed to get e_shstrndx from %s\n", obj->path);
6371ca3b 2431 return -LIBBPF_ERRNO__FORMAT;
29603665
WN		 2432 }
2433
2434 while ((scn = elf_nextscn(elf, scn)) != NULL) {
2435 char *name;
2436 GElf_Shdr sh;
2437 Elf_Data *data;
2438
2439 idx++;
2440 if (gelf_getshdr(scn, &sh) != &sh) {
be18010e
KW		 2441 pr_warn("failed to get section(%d) header from %s\n",
2442 idx, obj->path);
01b29d1d 2443 return -LIBBPF_ERRNO__FORMAT;
29603665
WN		 2444 }
2445
2446 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
2447 if (!name) {
be18010e
KW		 2448 pr_warn("failed to get section(%d) name from %s\n",
2449 idx, obj->path);
01b29d1d 2450 return -LIBBPF_ERRNO__FORMAT;
29603665
WN		 2451 }
2452
2453 data = elf_getdata(scn, 0);
2454 if (!data) {
be18010e
KW		 2455 pr_warn("failed to get section(%d) data from %s(%s)\n",
2456 idx, name, obj->path);
01b29d1d 2457 return -LIBBPF_ERRNO__FORMAT;
29603665 2458 }
077c066a
JDB		 2459 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
2460 idx, name, (unsigned long)data->d_size,
29603665
WN		 2461 (int)sh.sh_link, (unsigned long)sh.sh_flags,
2462 (int)sh.sh_type);
cb1e5e96 2463
1713d68b 2464 if (strcmp(name, "license") == 0) {
cb1e5e96
WN		 2465 err = bpf_object__init_license(obj,
2466 data->d_buf,
2467 data->d_size);
01b29d1d
AN		 2468 if (err)
2469 return err;
1713d68b 2470 } else if (strcmp(name, "version") == 0) {
54b8625c
JF		 2471 err = bpf_object__init_kversion(obj,
2472 data->d_buf,
2473 data->d_size);
2474 if (err)
2475 return err;
1713d68b 2476 } else if (strcmp(name, "maps") == 0) {
666810e8 2477 obj->efile.maps_shndx = idx;
abd29c93
AN		 2478 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
2479 obj->efile.btf_maps_shndx = idx;
1713d68b
DB		 2480 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
2481 btf_data = data;
2993e051 2482 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
f0187f0b 2483 btf_ext_data = data;
8a138aed 2484 } else if (sh.sh_type == SHT_SYMTAB) {
bec7d68c 2485 if (obj->efile.symbols) {
be18010e
KW		 2486 pr_warn("bpf: multiple SYMTAB in %s\n",
2487 obj->path);
01b29d1d 2488 return -LIBBPF_ERRNO__FORMAT;
77ba9a5b 2489 }
01b29d1d 2490 obj->efile.symbols = data;
166750bc 2491 obj->efile.symbols_shndx = idx;
01b29d1d 2492 obj->efile.strtabidx = sh.sh_link;
f8c7a4d4
JS		 2493 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
2494 if (sh.sh_flags & SHF_EXECINSTR) {
2495 if (strcmp(name, ".text") == 0)
2496 obj->efile.text_shndx = idx;
2497 err = bpf_object__add_program(obj, data->d_buf,
8983b731
AN		 2498 data->d_size,
2499 name, idx);
f8c7a4d4
JS		 2500 if (err) {
2501 char errmsg[STRERR_BUFSIZE];
8983b731 2502 char *cp;
f8c7a4d4 2503
8983b731
AN		 2504 cp = libbpf_strerror_r(-err, errmsg,
2505 sizeof(errmsg));
be18010e
KW		 2506 pr_warn("failed to alloc program %s (%s): %s",
2507 name, obj->path, cp);
01b29d1d 2508 return err;
f8c7a4d4 2509 }
ac9d1389 2510 } else if (strcmp(name, DATA_SEC) == 0) {
d859900c
DB		 2511 obj->efile.data = data;
2512 obj->efile.data_shndx = idx;
ac9d1389 2513 } else if (strcmp(name, RODATA_SEC) == 0) {
d859900c
DB		 2514 obj->efile.rodata = data;
2515 obj->efile.rodata_shndx = idx;
590a0088
MKL		 2516 } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
2517 obj->efile.st_ops_data = data;
2518 obj->efile.st_ops_shndx = idx;
d859900c
DB		 2519 } else {
2520 pr_debug("skip section(%d) %s\n", idx, name);
a5b8bd47 2521 }
b62f06e8 2522 } else if (sh.sh_type == SHT_REL) {
1f8e2bcb
AN		 2523 int nr_sects = obj->efile.nr_reloc_sects;
2524 void *sects = obj->efile.reloc_sects;
e3d91b0c
JDB		 2525 int sec = sh.sh_info; /* points to other section */
2526
2527 /* Only do relo for section with exec instructions */
590a0088
MKL		 2528 if (!section_have_execinstr(obj, sec) &&
2529 strcmp(name, ".rel" STRUCT_OPS_SEC)) {
e3d91b0c
JDB		 2530 pr_debug("skip relo %s(%d) for section(%d)\n",
2531 name, idx, sec);
2532 continue;
2533 }
b62f06e8 2534
1f8e2bcb
AN		 2535 sects = reallocarray(sects, nr_sects + 1,
2536 sizeof(*obj->efile.reloc_sects));
2537 if (!sects) {
2538 pr_warn("reloc_sects realloc failed\n");
01b29d1d
AN		 2539 return -ENOMEM;
2540 }
b62f06e8 2541
1f8e2bcb
AN		 2542 obj->efile.reloc_sects = sects;
2543 obj->efile.nr_reloc_sects++;
b62f06e8 2544
1f8e2bcb
AN		 2545 obj->efile.reloc_sects[nr_sects].shdr = sh;
2546 obj->efile.reloc_sects[nr_sects].data = data;
ac9d1389
AN		 2547 } else if (sh.sh_type == SHT_NOBITS &&
2548 strcmp(name, BSS_SEC) == 0) {
d859900c
DB		 2549 obj->efile.bss = data;
2550 obj->efile.bss_shndx = idx;
077c066a
JDB		 2551 } else {
2552 pr_debug("skip section(%d) %s\n", idx, name);
bec7d68c 2553 }
29603665 2554 }
561bbcca 2555
d3a3aa0c 2556 if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
be18010e 2557 pr_warn("Corrupted ELF file: index of strtab invalid\n");
f102154d 2558 return -LIBBPF_ERRNO__FORMAT;
77ba9a5b 2559 }
0d13bfce 2560 return bpf_object__init_btf(obj, btf_data, btf_ext_data);
29603665
WN		 2561}
2562
166750bc
AN		 2563static bool sym_is_extern(const GElf_Sym *sym)
2564{
2565 int bind = GELF_ST_BIND(sym->st_info);
2566 /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
2567 return sym->st_shndx == SHN_UNDEF &&
2568 (bind == STB_GLOBAL || bind == STB_WEAK) &&
2569 GELF_ST_TYPE(sym->st_info) == STT_NOTYPE;
2570}
2571
2572static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
2573{
2574 const struct btf_type *t;
2575 const char *var_name;
2576 int i, n;
2577
2578 if (!btf)
2579 return -ESRCH;
2580
2581 n = btf__get_nr_types(btf);
2582 for (i = 1; i <= n; i++) {
2583 t = btf__type_by_id(btf, i);
2584
2585 if (!btf_is_var(t))
2586 continue;
2587
2588 var_name = btf__name_by_offset(btf, t->name_off);
2589 if (strcmp(var_name, ext_name))
2590 continue;
2591
2592 if (btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
2593 return -EINVAL;
2594
2595 return i;
2596 }
2597
2598 return -ENOENT;
2599}
2600
2601static enum extern_type find_extern_type(const struct btf *btf, int id,
2602 bool *is_signed)
2603{
2604 const struct btf_type *t;
2605 const char *name;
2606
2607 t = skip_mods_and_typedefs(btf, id, NULL);
2608 name = btf__name_by_offset(btf, t->name_off);
2609
2610 if (is_signed)
2611 *is_signed = false;
2612 switch (btf_kind(t)) {
2613 case BTF_KIND_INT: {
2614 int enc = btf_int_encoding(t);
2615
2616 if (enc & BTF_INT_BOOL)
2617 return t->size == 1 ? EXT_BOOL : EXT_UNKNOWN;
2618 if (is_signed)
2619 *is_signed = enc & BTF_INT_SIGNED;
2620 if (t->size == 1)
2621 return EXT_CHAR;
2622 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
2623 return EXT_UNKNOWN;
2624 return EXT_INT;
2625 }
2626 case BTF_KIND_ENUM:
2627 if (t->size != 4)
2628 return EXT_UNKNOWN;
2629 if (strcmp(name, "libbpf_tristate"))
2630 return EXT_UNKNOWN;
2631 return EXT_TRISTATE;
2632 case BTF_KIND_ARRAY:
2633 if (btf_array(t)->nelems == 0)
2634 return EXT_UNKNOWN;
2635 if (find_extern_type(btf, btf_array(t)->type, NULL) != EXT_CHAR)
2636 return EXT_UNKNOWN;
2637 return EXT_CHAR_ARR;
2638 default:
2639 return EXT_UNKNOWN;
2640 }
2641}
2642
2643static int cmp_externs(const void *_a, const void *_b)
2644{
2645 const struct extern_desc *a = _a;
2646 const struct extern_desc *b = _b;
2647
2648 /* descending order by alignment requirements */
2649 if (a->align != b->align)
2650 return a->align > b->align ? -1 : 1;
2651 /* ascending order by size, within same alignment class */
2652 if (a->sz != b->sz)
2653 return a->sz < b->sz ? -1 : 1;
2654 /* resolve ties by name */
2655 return strcmp(a->name, b->name);
2656}
2657
2658static int bpf_object__collect_externs(struct bpf_object *obj)
2659{
2660 const struct btf_type *t;
2661 struct extern_desc *ext;
2662 int i, n, off, btf_id;
2663 struct btf_type *sec;
2664 const char *ext_name;
2665 Elf_Scn *scn;
2666 GElf_Shdr sh;
2667
2668 if (!obj->efile.symbols)
2669 return 0;
2670
2671 scn = elf_getscn(obj->efile.elf, obj->efile.symbols_shndx);
2672 if (!scn)
2673 return -LIBBPF_ERRNO__FORMAT;
2674 if (gelf_getshdr(scn, &sh) != &sh)
2675 return -LIBBPF_ERRNO__FORMAT;
2676 n = sh.sh_size / sh.sh_entsize;
2677
2678 pr_debug("looking for externs among %d symbols...\n", n);
2679 for (i = 0; i < n; i++) {
2680 GElf_Sym sym;
2681
2682 if (!gelf_getsym(obj->efile.symbols, i, &sym))
2683 return -LIBBPF_ERRNO__FORMAT;
2684 if (!sym_is_extern(&sym))
2685 continue;
2686 ext_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
2687 sym.st_name);
2688 if (!ext_name || !ext_name[0])
2689 continue;
2690
2691 ext = obj->externs;
2692 ext = reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
2693 if (!ext)
2694 return -ENOMEM;
2695 obj->externs = ext;
2696 ext = &ext[obj->nr_extern];
2697 memset(ext, 0, sizeof(*ext));
2698 obj->nr_extern++;
2699
2700 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
2701 if (ext->btf_id <= 0) {
2702 pr_warn("failed to find BTF for extern '%s': %d\n",
2703 ext_name, ext->btf_id);
2704 return ext->btf_id;
2705 }
2706 t = btf__type_by_id(obj->btf, ext->btf_id);
2707 ext->name = btf__name_by_offset(obj->btf, t->name_off);
2708 ext->sym_idx = i;
2709 ext->is_weak = GELF_ST_BIND(sym.st_info) == STB_WEAK;
2710 ext->sz = btf__resolve_size(obj->btf, t->type);
2711 if (ext->sz <= 0) {
2712 pr_warn("failed to resolve size of extern '%s': %d\n",
2713 ext_name, ext->sz);
2714 return ext->sz;
2715 }
2716 ext->align = btf__align_of(obj->btf, t->type);
2717 if (ext->align <= 0) {
2718 pr_warn("failed to determine alignment of extern '%s': %d\n",
2719 ext_name, ext->align);
2720 return -EINVAL;
2721 }
2722 ext->type = find_extern_type(obj->btf, t->type,
2723 &ext->is_signed);
2724 if (ext->type == EXT_UNKNOWN) {
2725 pr_warn("extern '%s' type is unsupported\n", ext_name);
2726 return -ENOTSUP;
2727 }
2728 }
2729 pr_debug("collected %d externs total\n", obj->nr_extern);
2730
2731 if (!obj->nr_extern)
2732 return 0;
2733
2734 /* sort externs by (alignment, size, name) and calculate their offsets
2735 * within a map */
2736 qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
2737 off = 0;
2738 for (i = 0; i < obj->nr_extern; i++) {
2739 ext = &obj->externs[i];
2740 ext->data_off = roundup(off, ext->align);
2741 off = ext->data_off + ext->sz;
2742 pr_debug("extern #%d: symbol %d, off %u, name %s\n",
2743 i, ext->sym_idx, ext->data_off, ext->name);
2744 }
2745
81bfdd08 2746 btf_id = btf__find_by_name(obj->btf, KCONFIG_SEC);
166750bc 2747 if (btf_id <= 0) {
81bfdd08 2748 pr_warn("no BTF info found for '%s' datasec\n", KCONFIG_SEC);
166750bc
AN		 2749 return -ESRCH;
2750 }
2751
2752 sec = (struct btf_type *)btf__type_by_id(obj->btf, btf_id);
2753 sec->size = off;
2754 n = btf_vlen(sec);
2755 for (i = 0; i < n; i++) {
2756 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
2757
2758 t = btf__type_by_id(obj->btf, vs->type);
2759 ext_name = btf__name_by_offset(obj->btf, t->name_off);
2760 ext = find_extern_by_name(obj, ext_name);
2761 if (!ext) {
2762 pr_warn("failed to find extern definition for BTF var '%s'\n",
2763 ext_name);
2764 return -ESRCH;
2765 }
2766 vs->offset = ext->data_off;
2767 btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
2768 }
2769
2770 return 0;
2771}
2772
34090915
WN		 2773static struct bpf_program *
2774bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
2775{
2776 struct bpf_program *prog;
2777 size_t i;
2778
2779 for (i = 0; i < obj->nr_programs; i++) {
2780 prog = &obj->programs[i];
2781 if (prog->idx == idx)
2782 return prog;
2783 }
2784 return NULL;
2785}
2786
6d4b198b 2787struct bpf_program *
a324aae3
AN		 2788bpf_object__find_program_by_title(const struct bpf_object *obj,
2789 const char *title)
6d4b198b
JK		 2790{
2791 struct bpf_program *pos;
2792
2793 bpf_object__for_each_program(pos, obj) {
2794 if (pos->section_name && !strcmp(pos->section_name, title))
2795 return pos;
2796 }
2797 return NULL;
2798}
2799
01af3bf0
AN		 2800struct bpf_program *
2801bpf_object__find_program_by_name(const struct bpf_object *obj,
2802 const char *name)
2803{
2804 struct bpf_program *prog;
2805
2806 bpf_object__for_each_program(prog, obj) {
2807 if (!strcmp(prog->name, name))
2808 return prog;
2809 }
2810 return NULL;
2811}
2812
d859900c
DB		 2813static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
2814 int shndx)
2815{
2816 return shndx == obj->efile.data_shndx ||
2817 shndx == obj->efile.bss_shndx ||
2818 shndx == obj->efile.rodata_shndx;
2819}
2820
2821static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
2822 int shndx)
2823{
abd29c93
AN		 2824 return shndx == obj->efile.maps_shndx ||
2825 shndx == obj->efile.btf_maps_shndx;
d859900c
DB		 2826}
2827
d859900c
DB		 2828static enum libbpf_map_type
2829bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
2830{
2831 if (shndx == obj->efile.data_shndx)
2832 return LIBBPF_MAP_DATA;
2833 else if (shndx == obj->efile.bss_shndx)
2834 return LIBBPF_MAP_BSS;
2835 else if (shndx == obj->efile.rodata_shndx)
2836 return LIBBPF_MAP_RODATA;
166750bc 2837 else if (shndx == obj->efile.symbols_shndx)
81bfdd08 2838 return LIBBPF_MAP_KCONFIG;
d859900c
DB		 2839 else
2840 return LIBBPF_MAP_UNSPEC;
2841}
2842
1f8e2bcb
AN		 2843static int bpf_program__record_reloc(struct bpf_program *prog,
2844 struct reloc_desc *reloc_desc,
2845 __u32 insn_idx, const char *name,
2846 const GElf_Sym *sym, const GElf_Rel *rel)
2847{
2848 struct bpf_insn *insn = &prog->insns[insn_idx];
2849 size_t map_idx, nr_maps = prog->obj->nr_maps;
2850 struct bpf_object *obj = prog->obj;
2851 __u32 shdr_idx = sym->st_shndx;
2852 enum libbpf_map_type type;
2853 struct bpf_map *map;
2854
2855 /* sub-program call relocation */
2856 if (insn->code == (BPF_JMP | BPF_CALL)) {
2857 if (insn->src_reg != BPF_PSEUDO_CALL) {
2858 pr_warn("incorrect bpf_call opcode\n");
2859 return -LIBBPF_ERRNO__RELOC;
2860 }
2861 /* text_shndx can be 0, if no default "main" program exists */
2862 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
2863 pr_warn("bad call relo against section %u\n", shdr_idx);
2864 return -LIBBPF_ERRNO__RELOC;
2865 }
2866 if (sym->st_value % 8) {
679152d3
AN		 2867 pr_warn("bad call relo offset: %zu\n",
2868 (size_t)sym->st_value);
1f8e2bcb
AN		 2869 return -LIBBPF_ERRNO__RELOC;
2870 }
2871 reloc_desc->type = RELO_CALL;
2872 reloc_desc->insn_idx = insn_idx;
53f8dd43 2873 reloc_desc->sym_off = sym->st_value;
1f8e2bcb
AN		 2874 obj->has_pseudo_calls = true;
2875 return 0;
2876 }
2877
2878 if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) {
8983b731 2879 pr_warn("invalid relo for insns[%d].code 0x%x\n",
1f8e2bcb
AN		 2880 insn_idx, insn->code);
2881 return -LIBBPF_ERRNO__RELOC;
2882 }
166750bc
AN		 2883
2884 if (sym_is_extern(sym)) {
2885 int sym_idx = GELF_R_SYM(rel->r_info);
2886 int i, n = obj->nr_extern;
2887 struct extern_desc *ext;
2888
2889 for (i = 0; i < n; i++) {
2890 ext = &obj->externs[i];
2891 if (ext->sym_idx == sym_idx)
2892 break;
2893 }
2894 if (i >= n) {
2895 pr_warn("extern relo failed to find extern for sym %d\n",
2896 sym_idx);
2897 return -LIBBPF_ERRNO__RELOC;
2898 }
2899 pr_debug("found extern #%d '%s' (sym %d, off %u) for insn %u\n",
2900 i, ext->name, ext->sym_idx, ext->data_off, insn_idx);
2901 reloc_desc->type = RELO_EXTERN;
2902 reloc_desc->insn_idx = insn_idx;
2903 reloc_desc->sym_off = ext->data_off;
2904 return 0;
2905 }
2906
1f8e2bcb 2907 if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
393cdfbe
AN		 2908 pr_warn("invalid relo for \'%s\' in special section 0x%x; forgot to initialize global var?..\n",
2909 name, shdr_idx);
1f8e2bcb
AN		 2910 return -LIBBPF_ERRNO__RELOC;
2911 }
2912
2913 type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
2914
2915 /* generic map reference relocation */
2916 if (type == LIBBPF_MAP_UNSPEC) {
2917 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
2918 pr_warn("bad map relo against section %u\n",
2919 shdr_idx);
2920 return -LIBBPF_ERRNO__RELOC;
2921 }
2922 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
2923 map = &obj->maps[map_idx];
2924 if (map->libbpf_type != type ||
2925 map->sec_idx != sym->st_shndx ||
2926 map->sec_offset != sym->st_value)
2927 continue;
2928 pr_debug("found map %zd (%s, sec %d, off %zu) for insn %u\n",
2929 map_idx, map->name, map->sec_idx,
2930 map->sec_offset, insn_idx);
2931 break;
2932 }
2933 if (map_idx >= nr_maps) {
679152d3
AN		 2934 pr_warn("map relo failed to find map for sec %u, off %zu\n",
2935 shdr_idx, (size_t)sym->st_value);
1f8e2bcb
AN		 2936 return -LIBBPF_ERRNO__RELOC;
2937 }
2938 reloc_desc->type = RELO_LD64;
2939 reloc_desc->insn_idx = insn_idx;
2940 reloc_desc->map_idx = map_idx;
53f8dd43 2941 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
1f8e2bcb
AN		 2942 return 0;
2943 }
2944
2945 /* global data map relocation */
2946 if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
2947 pr_warn("bad data relo against section %u\n", shdr_idx);
2948 return -LIBBPF_ERRNO__RELOC;
1f8e2bcb 2949 }
1f8e2bcb
AN		 2950 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
2951 map = &obj->maps[map_idx];
2952 if (map->libbpf_type != type)
2953 continue;
2954 pr_debug("found data map %zd (%s, sec %d, off %zu) for insn %u\n",
2955 map_idx, map->name, map->sec_idx, map->sec_offset,
2956 insn_idx);
2957 break;
2958 }
2959 if (map_idx >= nr_maps) {
2960 pr_warn("data relo failed to find map for sec %u\n",
2961 shdr_idx);
2962 return -LIBBPF_ERRNO__RELOC;
2963 }
2964
2965 reloc_desc->type = RELO_DATA;
2966 reloc_desc->insn_idx = insn_idx;
2967 reloc_desc->map_idx = map_idx;
53f8dd43 2968 reloc_desc->sym_off = sym->st_value;
1f8e2bcb
AN		 2969 return 0;
2970}
2971
34090915 2972static int
48cca7e4
AS		 2973bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
2974 Elf_Data *data, struct bpf_object *obj)
34090915 2975{
48cca7e4 2976 Elf_Data *symbols = obj->efile.symbols;
1f8e2bcb 2977 int err, i, nrels;
34090915 2978
399dc65e 2979 pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
34090915
WN		 2980 nrels = shdr->sh_size / shdr->sh_entsize;
2981
2982 prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
2983 if (!prog->reloc_desc) {
be18010e 2984 pr_warn("failed to alloc memory in relocation\n");
34090915
WN		 2985 return -ENOMEM;
2986 }
2987 prog->nr_reloc = nrels;
2988
2989 for (i = 0; i < nrels; i++) {
d859900c 2990 const char *name;
1f8e2bcb 2991 __u32 insn_idx;
abd29c93
AN		 2992 GElf_Sym sym;
2993 GElf_Rel rel;
34090915
WN		 2994
2995 if (!gelf_getrel(data, i, &rel)) {
be18010e 2996 pr_warn("relocation: failed to get %d reloc\n", i);
6371ca3b 2997 return -LIBBPF_ERRNO__FORMAT;
34090915 2998 }
399dc65e 2999 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
be18010e
KW		 3000 pr_warn("relocation: symbol %"PRIx64" not found\n",
3001 GELF_R_SYM(rel.r_info));
6371ca3b 3002 return -LIBBPF_ERRNO__FORMAT;
34090915 3003 }
1f8e2bcb
AN		 3004 if (rel.r_offset % sizeof(struct bpf_insn))
3005 return -LIBBPF_ERRNO__FORMAT;
d859900c 3006
1f8e2bcb 3007 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
d859900c
DB		 3008 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
3009 sym.st_name) ? : "<?>";
3010
679152d3
AN		 3011 pr_debug("relo for shdr %u, symb %zu, value %zu, type %d, bind %d, name %d (\'%s\'), insn %u\n",
3012 (__u32)sym.st_shndx, (size_t)GELF_R_SYM(rel.r_info),
3013 (size_t)sym.st_value, GELF_ST_TYPE(sym.st_info),
1f8e2bcb
AN		 3014 GELF_ST_BIND(sym.st_info), sym.st_name, name,
3015 insn_idx);
666810e8 3016
1f8e2bcb
AN		 3017 err = bpf_program__record_reloc(prog, &prog->reloc_desc[i],
3018 insn_idx, name, &sym, &rel);
3019 if (err)
3020 return err;
34090915
WN		 3021 }
3022 return 0;
3023}
3024
abd29c93 3025static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
8a138aed
MKL		 3026{
3027 struct bpf_map_def *def = &map->def;
d859900c 3028 __u32 key_type_id = 0, value_type_id = 0;
96408c43 3029 int ret;
8a138aed 3030
590a0088
MKL		 3031 /* if it's BTF-defined map, we don't need to search for type IDs.
3032 * For struct_ops map, it does not need btf_key_type_id and
3033 * btf_value_type_id.
3034 */
3035 if (map->sec_idx == obj->efile.btf_maps_shndx ||
3036 bpf_map__is_struct_ops(map))
abd29c93
AN		 3037 return 0;
3038
d859900c 3039 if (!bpf_map__is_internal(map)) {
abd29c93 3040 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
d859900c
DB		 3041 def->value_size, &key_type_id,
3042 &value_type_id);
3043 } else {
3044 /*
3045 * LLVM annotates global data differently in BTF, that is,
3046 * only as '.data', '.bss' or '.rodata'.
3047 */
abd29c93 3048 ret = btf__find_by_name(obj->btf,
d859900c
DB		 3049 libbpf_type_to_btf_name[map->libbpf_type]);
3050 }
3051 if (ret < 0)
96408c43 3052 return ret;
8a138aed 3053
96408c43 3054 map->btf_key_type_id = key_type_id;
d859900c
DB		 3055 map->btf_value_type_id = bpf_map__is_internal(map) ?
3056 ret : value_type_id;
8a138aed
MKL		 3057 return 0;
3058}
3059
26736eb9
JK		 3060int bpf_map__reuse_fd(struct bpf_map *map, int fd)
3061{
3062 struct bpf_map_info info = {};
3063 __u32 len = sizeof(info);
3064 int new_fd, err;
3065 char *new_name;
3066
3067 err = bpf_obj_get_info_by_fd(fd, &info, &len);
3068 if (err)
3069 return err;
3070
3071 new_name = strdup(info.name);
3072 if (!new_name)
3073 return -errno;
3074
3075 new_fd = open("/", O_RDONLY | O_CLOEXEC);
d1b4574a
THJ		 3076 if (new_fd < 0) {
3077 err = -errno;
26736eb9 3078 goto err_free_new_name;
d1b4574a 3079 }
26736eb9
JK		 3080
3081 new_fd = dup3(fd, new_fd, O_CLOEXEC);
d1b4574a
THJ		 3082 if (new_fd < 0) {
3083 err = -errno;
26736eb9 3084 goto err_close_new_fd;
d1b4574a 3085 }
26736eb9
JK		 3086
3087 err = zclose(map->fd);
d1b4574a
THJ		 3088 if (err) {
3089 err = -errno;
26736eb9 3090 goto err_close_new_fd;
d1b4574a 3091 }
26736eb9
JK		 3092 free(map->name);
3093
3094 map->fd = new_fd;
3095 map->name = new_name;
3096 map->def.type = info.type;
3097 map->def.key_size = info.key_size;
3098 map->def.value_size = info.value_size;
3099 map->def.max_entries = info.max_entries;
3100 map->def.map_flags = info.map_flags;
3101 map->btf_key_type_id = info.btf_key_type_id;
3102 map->btf_value_type_id = info.btf_value_type_id;
ec6d5f47 3103 map->reused = true;
26736eb9
JK		 3104
3105 return 0;
3106
3107err_close_new_fd:
3108 close(new_fd);
3109err_free_new_name:
3110 free(new_name);
d1b4574a 3111 return err;
26736eb9
JK		 3112}
3113
1a11a4c7
AI		 3114int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
3115{
3116 if (!map || !max_entries)
3117 return -EINVAL;
3118
3119 /* If map already created, its attributes can't be changed. */
3120 if (map->fd >= 0)
3121 return -EBUSY;
3122
3123 map->def.max_entries = max_entries;
3124
3125 return 0;
3126}
3127
47eff617
SF		 3128static int
3129bpf_object__probe_name(struct bpf_object *obj)
3130{
3131 struct bpf_load_program_attr attr;
3132 char *cp, errmsg[STRERR_BUFSIZE];
3133 struct bpf_insn insns[] = {
3134 BPF_MOV64_IMM(BPF_REG_0, 0),
3135 BPF_EXIT_INSN(),
3136 };
3137 int ret;
3138
3139 /* make sure basic loading works */
3140
3141 memset(&attr, 0, sizeof(attr));
3142 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3143 attr.insns = insns;
3144 attr.insns_cnt = ARRAY_SIZE(insns);
3145 attr.license = "GPL";
3146
3147 ret = bpf_load_program_xattr(&attr, NULL, 0);
3148 if (ret < 0) {
3149 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
be18010e
KW		 3150 pr_warn("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
3151 __func__, cp, errno);
47eff617
SF		 3152 return -errno;
3153 }
3154 close(ret);
3155
3156 /* now try the same program, but with the name */
3157
3158 attr.name = "test";
3159 ret = bpf_load_program_xattr(&attr, NULL, 0);
3160 if (ret >= 0) {
3161 obj->caps.name = 1;
3162 close(ret);
3163 }
3164
3165 return 0;
3166}
3167
8837fe5d
DB		 3168static int
3169bpf_object__probe_global_data(struct bpf_object *obj)
3170{
3171 struct bpf_load_program_attr prg_attr;
3172 struct bpf_create_map_attr map_attr;
3173 char *cp, errmsg[STRERR_BUFSIZE];
3174 struct bpf_insn insns[] = {
3175 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
3176 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
3177 BPF_MOV64_IMM(BPF_REG_0, 0),
3178 BPF_EXIT_INSN(),
3179 };
3180 int ret, map;
3181
3182 memset(&map_attr, 0, sizeof(map_attr));
3183 map_attr.map_type = BPF_MAP_TYPE_ARRAY;
3184 map_attr.key_size = sizeof(int);
3185 map_attr.value_size = 32;
3186 map_attr.max_entries = 1;
3187
3188 map = bpf_create_map_xattr(&map_attr);
3189 if (map < 0) {
3190 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
be18010e
KW		 3191 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
3192 __func__, cp, errno);
8837fe5d
DB		 3193 return -errno;
3194 }
3195
3196 insns[0].imm = map;
3197
3198 memset(&prg_attr, 0, sizeof(prg_attr));
3199 prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3200 prg_attr.insns = insns;
3201 prg_attr.insns_cnt = ARRAY_SIZE(insns);
3202 prg_attr.license = "GPL";
3203
3204 ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
3205 if (ret >= 0) {
3206 obj->caps.global_data = 1;
3207 close(ret);
3208 }
3209
3210 close(map);
3211 return 0;
3212}
3213
d7c4b398
AN		 3214static int bpf_object__probe_btf_func(struct bpf_object *obj)
3215{
8983b731 3216 static const char strs[] = "\0int\0x\0a";
d7c4b398
AN		 3217 /* void x(int a) {} */
3218 __u32 types[] = {
3219 /* int */
3220 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
3221 /* FUNC_PROTO */ /* [2] */
3222 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
3223 BTF_PARAM_ENC(7, 1),
3224 /* FUNC x */ /* [3] */
3225 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
3226 };
cfd49210 3227 int btf_fd;
d7c4b398 3228
cfd49210
MR		 3229 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3230 strs, sizeof(strs));
3231 if (btf_fd >= 0) {
d7c4b398 3232 obj->caps.btf_func = 1;
cfd49210
MR		 3233 close(btf_fd);
3234 return 1;
3235 }
3236
d7c4b398
AN		 3237 return 0;
3238}
3239
2d3eb67f
AS		 3240static int bpf_object__probe_btf_func_global(struct bpf_object *obj)
3241{
3242 static const char strs[] = "\0int\0x\0a";
3243 /* static void x(int a) {} */
3244 __u32 types[] = {
3245 /* int */
3246 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
3247 /* FUNC_PROTO */ /* [2] */
3248 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
3249 BTF_PARAM_ENC(7, 1),
3250 /* FUNC x BTF_FUNC_GLOBAL */ /* [3] */
3251 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
3252 };
3253 int btf_fd;
3254
3255 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3256 strs, sizeof(strs));
3257 if (btf_fd >= 0) {
3258 obj->caps.btf_func_global = 1;
3259 close(btf_fd);
3260 return 1;
3261 }
3262
3263 return 0;
3264}
3265
d7c4b398
AN		 3266static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
3267{
8983b731 3268 static const char strs[] = "\0x\0.data";
d7c4b398
AN		 3269 /* static int a; */
3270 __u32 types[] = {
3271 /* int */
3272 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
3273 /* VAR x */ /* [2] */
3274 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
3275 BTF_VAR_STATIC,
3276 /* DATASEC val */ /* [3] */
3277 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
3278 BTF_VAR_SECINFO_ENC(2, 0, 4),
3279 };
cfd49210 3280 int btf_fd;
d7c4b398 3281
cfd49210
MR		 3282 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3283 strs, sizeof(strs));
3284 if (btf_fd >= 0) {
d7c4b398 3285 obj->caps.btf_datasec = 1;
cfd49210
MR		 3286 close(btf_fd);
3287 return 1;
3288 }
3289
d7c4b398
AN		 3290 return 0;
3291}
3292
7fe74b43
AN		 3293static int bpf_object__probe_array_mmap(struct bpf_object *obj)
3294{
3295 struct bpf_create_map_attr attr = {
3296 .map_type = BPF_MAP_TYPE_ARRAY,
3297 .map_flags = BPF_F_MMAPABLE,
3298 .key_size = sizeof(int),
3299 .value_size = sizeof(int),
3300 .max_entries = 1,
3301 };
3302 int fd;
3303
3304 fd = bpf_create_map_xattr(&attr);
3305 if (fd >= 0) {
3306 obj->caps.array_mmap = 1;
3307 close(fd);
3308 return 1;
3309 }
3310
3311 return 0;
3312}
3313
47eff617
SF		 3314static int
3315bpf_object__probe_caps(struct bpf_object *obj)
3316{
8837fe5d
DB		 3317 int (*probe_fn[])(struct bpf_object *obj) = {
3318 bpf_object__probe_name,
3319 bpf_object__probe_global_data,
d7c4b398 3320 bpf_object__probe_btf_func,
2d3eb67f 3321 bpf_object__probe_btf_func_global,
d7c4b398 3322 bpf_object__probe_btf_datasec,
7fe74b43 3323 bpf_object__probe_array_mmap,
8837fe5d
DB		 3324 };
3325 int i, ret;
3326
3327 for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
3328 ret = probe_fn[i](obj);
3329 if (ret < 0)
15ea164e 3330 pr_debug("Probe #%d failed with %d.\n", i, ret);
8837fe5d
DB		 3331 }
3332
3333 return 0;
47eff617
SF		 3334}
3335
57a00f41
THJ		 3336static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
3337{
3338 struct bpf_map_info map_info = {};
3339 char msg[STRERR_BUFSIZE];
3340 __u32 map_info_len;
3341
3342 map_info_len = sizeof(map_info);
3343
3344 if (bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len)) {
3345 pr_warn("failed to get map info for map FD %d: %s\n",
3346 map_fd, libbpf_strerror_r(errno, msg, sizeof(msg)));
3347 return false;
3348 }
3349
3350 return (map_info.type == map->def.type &&
3351 map_info.key_size == map->def.key_size &&
3352 map_info.value_size == map->def.value_size &&
3353 map_info.max_entries == map->def.max_entries &&
3354 map_info.map_flags == map->def.map_flags);
3355}
3356
3357static int
3358bpf_object__reuse_map(struct bpf_map *map)
3359{
3360 char *cp, errmsg[STRERR_BUFSIZE];
3361 int err, pin_fd;
3362
3363 pin_fd = bpf_obj_get(map->pin_path);
3364 if (pin_fd < 0) {
3365 err = -errno;
3366 if (err == -ENOENT) {
3367 pr_debug("found no pinned map to reuse at '%s'\n",
3368 map->pin_path);
3369 return 0;
3370 }
3371
3372 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
3373 pr_warn("couldn't retrieve pinned map '%s': %s\n",
3374 map->pin_path, cp);
3375 return err;
3376 }
3377
3378 if (!map_is_reuse_compat(map, pin_fd)) {
3379 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
3380 map->pin_path);
3381 close(pin_fd);
3382 return -EINVAL;
3383 }
3384
3385 err = bpf_map__reuse_fd(map, pin_fd);
3386 if (err) {
3387 close(pin_fd);
3388 return err;
3389 }
3390 map->pinned = true;
3391 pr_debug("reused pinned map at '%s'\n", map->pin_path);
3392
3393 return 0;
3394}
3395
d859900c
DB		 3396static int
3397bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
3398{
166750bc 3399 enum libbpf_map_type map_type = map->libbpf_type;
d859900c
DB		 3400 char *cp, errmsg[STRERR_BUFSIZE];
3401 int err, zero = 0;
d859900c 3402
166750bc
AN		 3403 /* kernel already zero-initializes .bss map. */
3404 if (map_type == LIBBPF_MAP_BSS)
d859900c
DB		 3405 return 0;
3406
eba9c5f4
AN		 3407 err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
3408 if (err) {
3409 err = -errno;
3410 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3411 pr_warn("Error setting initial map(%s) contents: %s\n",
3412 map->name, cp);
3413 return err;
3414 }
d859900c 3415
81bfdd08
AN		 3416 /* Freeze .rodata and .kconfig map as read-only from syscall side. */
3417 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
d859900c
DB		 3418 err = bpf_map_freeze(map->fd);
3419 if (err) {
eba9c5f4
AN		 3420 err = -errno;
3421 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
be18010e
KW		 3422 pr_warn("Error freezing map(%s) as read-only: %s\n",
3423 map->name, cp);
eba9c5f4 3424 return err;
d859900c
DB		 3425 }
3426 }
eba9c5f4 3427 return 0;
d859900c
DB		 3428}
3429
52d3352e
WN		 3430static int
3431bpf_object__create_maps(struct bpf_object *obj)
3432{
8a138aed 3433 struct bpf_create_map_attr create_attr = {};
d7ff34d5 3434 int nr_cpus = 0;
52d3352e 3435 unsigned int i;
8a138aed 3436 int err;
52d3352e 3437
9d759a9b 3438 for (i = 0; i < obj->nr_maps; i++) {
8a138aed
MKL		 3439 struct bpf_map *map = &obj->maps[i];
3440 struct bpf_map_def *def = &map->def;
1ce6a9fc 3441 char *cp, errmsg[STRERR_BUFSIZE];
8a138aed
MKL		 3442 int *pfd = &map->fd;
3443
57a00f41
THJ		 3444 if (map->pin_path) {
3445 err = bpf_object__reuse_map(map);
3446 if (err) {
3447 pr_warn("error reusing pinned map %s\n",
3448 map->name);
3449 return err;
3450 }
3451 }
3452
26736eb9
JK		 3453 if (map->fd >= 0) {
3454 pr_debug("skip map create (preset) %s: fd=%d\n",
3455 map->name, map->fd);
3456 continue;
3457 }
3458
94cb310c
SF		 3459 if (obj->caps.name)
3460 create_attr.name = map->name;
f0307a7e 3461 create_attr.map_ifindex = map->map_ifindex;
8a138aed
MKL		 3462 create_attr.map_type = def->type;
3463 create_attr.map_flags = def->map_flags;
3464 create_attr.key_size = def->key_size;
3465 create_attr.value_size = def->value_size;
d7ff34d5
AN		 3466 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
3467 !def->max_entries) {
3468 if (!nr_cpus)
3469 nr_cpus = libbpf_num_possible_cpus();
3470 if (nr_cpus < 0) {
be18010e
KW		 3471 pr_warn("failed to determine number of system CPUs: %d\n",
3472 nr_cpus);
d7ff34d5
AN		 3473 err = nr_cpus;
3474 goto err_out;
3475 }
3476 pr_debug("map '%s': setting size to %d\n",
3477 map->name, nr_cpus);
3478 create_attr.max_entries = nr_cpus;
3479 } else {
3480 create_attr.max_entries = def->max_entries;
3481 }
e55d54f4 3482 create_attr.btf_fd = 0;
61746dbe
MKL		 3483 create_attr.btf_key_type_id = 0;
3484 create_attr.btf_value_type_id = 0;
addb9fc9
NS		 3485 if (bpf_map_type__is_map_in_map(def->type) &&
3486 map->inner_map_fd >= 0)
3487 create_attr.inner_map_fd = map->inner_map_fd;
590a0088
MKL		 3488 if (bpf_map__is_struct_ops(map))
3489 create_attr.btf_vmlinux_value_type_id =
3490 map->btf_vmlinux_value_type_id;
8a138aed 3491
abd29c93 3492 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
8a138aed 3493 create_attr.btf_fd = btf__fd(obj->btf);
61746dbe
MKL		 3494 create_attr.btf_key_type_id = map->btf_key_type_id;
3495 create_attr.btf_value_type_id = map->btf_value_type_id;
8a138aed
MKL		 3496 }
3497
3498 *pfd = bpf_create_map_xattr(&create_attr);
e55d54f4
AN		 3499 if (*pfd < 0 && (create_attr.btf_key_type_id ||
3500 create_attr.btf_value_type_id)) {
d7ff34d5
AN		 3501 err = -errno;
3502 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
be18010e
KW		 3503 pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
3504 map->name, cp, err);
e55d54f4 3505 create_attr.btf_fd = 0;
61746dbe
MKL		 3506 create_attr.btf_key_type_id = 0;
3507 create_attr.btf_value_type_id = 0;
3508 map->btf_key_type_id = 0;
3509 map->btf_value_type_id = 0;
8a138aed
MKL		 3510 *pfd = bpf_create_map_xattr(&create_attr);
3511 }
3512
52d3352e
WN		 3513 if (*pfd < 0) {
3514 size_t j;
52d3352e 3515
d7ff34d5 3516 err = -errno;
d859900c 3517err_out:
d7ff34d5 3518 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
be18010e
KW		 3519 pr_warn("failed to create map (name: '%s'): %s(%d)\n",
3520 map->name, cp, err);
dc3a2d25 3521 pr_perm_msg(err);
52d3352e 3522 for (j = 0; j < i; j++)
9d759a9b 3523 zclose(obj->maps[j].fd);
52d3352e
WN		 3524 return err;
3525 }
d859900c
DB		 3526
3527 if (bpf_map__is_internal(map)) {
3528 err = bpf_object__populate_internal_map(obj, map);
3529 if (err < 0) {
3530 zclose(*pfd);
3531 goto err_out;
3532 }
3533 }
3534
57a00f41
THJ		 3535 if (map->pin_path && !map->pinned) {
3536 err = bpf_map__pin(map, NULL);
3537 if (err) {
3538 pr_warn("failed to auto-pin map name '%s' at '%s'\n",
3539 map->name, map->pin_path);
3540 return err;
3541 }
3542 }
3543
76e1022b 3544 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
52d3352e
WN		 3545 }
3546
52d3352e
WN		 3547 return 0;
3548}
3549
f0187f0b
MKL		 3550static int
3551check_btf_ext_reloc_err(struct bpf_program *prog, int err,
3552 void *btf_prog_info, const char *info_name)
3553{
3554 if (err != -ENOENT) {
be18010e
KW		 3555 pr_warn("Error in loading %s for sec %s.\n",
3556 info_name, prog->section_name);
f0187f0b
MKL		 3557 return err;
3558 }
3559
3560 /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
3561
3562 if (btf_prog_info) {
3563 /*
3564 * Some info has already been found but has problem
399dc65e 3565 * in the last btf_ext reloc. Must have to error out.
f0187f0b 3566 */
be18010e
KW		 3567 pr_warn("Error in relocating %s for sec %s.\n",
3568 info_name, prog->section_name);
f0187f0b
MKL		 3569 return err;
3570 }
3571
399dc65e 3572 /* Have problem loading the very first info. Ignore the rest. */
be18010e
KW		 3573 pr_warn("Cannot find %s for main program sec %s. Ignore all %s.\n",
3574 info_name, prog->section_name, info_name);
f0187f0b
MKL		 3575 return 0;
3576}
3577
3578static int
3579bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
3580 const char *section_name, __u32 insn_offset)
3581{
3582 int err;
3583
3584 if (!insn_offset || prog->func_info) {
3585 /*
3586 * !insn_offset => main program
3587 *
3588 * For sub prog, the main program's func_info has to
3589 * be loaded first (i.e. prog->func_info != NULL)
3590 */
3591 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
3592 section_name, insn_offset,
3593 &prog->func_info,
3594 &prog->func_info_cnt);
3595 if (err)
3596 return check_btf_ext_reloc_err(prog, err,
3597 prog->func_info,
3598 "bpf_func_info");
3599
3600 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
3601 }
3602
3d650141
MKL		 3603 if (!insn_offset || prog->line_info) {
3604 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
3605 section_name, insn_offset,
3606 &prog->line_info,
3607 &prog->line_info_cnt);
3608 if (err)
3609 return check_btf_ext_reloc_err(prog, err,
3610 prog->line_info,
3611 "bpf_line_info");
3612
3613 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
3614 }
3615
f0187f0b
MKL		 3616 return 0;
3617}
3618
ddc7c304
AN		 3619#define BPF_CORE_SPEC_MAX_LEN 64
3620
3621/* represents BPF CO-RE field or array element accessor */
3622struct bpf_core_accessor {
3623 __u32 type_id; /* struct/union type or array element type */
3624 __u32 idx; /* field index or array index */
3625 const char *name; /* field name or NULL for array accessor */
3626};
3627
3628struct bpf_core_spec {
3629 const struct btf *btf;
3630 /* high-level spec: named fields and array indices only */
3631 struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
3632 /* high-level spec length */
3633 int len;
3634 /* raw, low-level spec: 1-to-1 with accessor spec string */
3635 int raw_spec[BPF_CORE_SPEC_MAX_LEN];
3636 /* raw spec length */
3637 int raw_len;
ee26dade
AN		 3638 /* field bit offset represented by spec */
3639 __u32 bit_offset;
ddc7c304
AN		 3640};
3641
3642static bool str_is_empty(const char *s)
3643{
3644 return !s || !s[0];
3645}
3646
1b484b30
AN		 3647static bool is_flex_arr(const struct btf *btf,
3648 const struct bpf_core_accessor *acc,
3649 const struct btf_array *arr)
3650{
3651 const struct btf_type *t;
3652
3653 /* not a flexible array, if not inside a struct or has non-zero size */
3654 if (!acc->name || arr->nelems > 0)
3655 return false;
3656
3657 /* has to be the last member of enclosing struct */
3658 t = btf__type_by_id(btf, acc->type_id);
3659 return acc->idx == btf_vlen(t) - 1;
3660}
3661
ddc7c304 3662/*
511bb008 3663 * Turn bpf_field_reloc into a low- and high-level spec representation,
ddc7c304 3664 * validating correctness along the way, as well as calculating resulting
ee26dade
AN		 3665 * field bit offset, specified by accessor string. Low-level spec captures
3666 * every single level of nestedness, including traversing anonymous
ddc7c304
AN		 3667 * struct/union members. High-level one only captures semantically meaningful
3668 * "turning points": named fields and array indicies.
3669 * E.g., for this case:
3670 *
3671 * struct sample {
3672 * int __unimportant;
3673 * struct {
3674 * int __1;
3675 * int __2;
3676 * int a[7];
3677 * };
3678 * };
3679 *
3680 * struct sample *s = ...;
3681 *
3682 * int x = &s->a[3]; // access string = '0:1:2:3'
3683 *
3684 * Low-level spec has 1:1 mapping with each element of access string (it's
3685 * just a parsed access string representation): [0, 1, 2, 3].
3686 *
3687 * High-level spec will capture only 3 points:
3688 * - intial zero-index access by pointer (&s->... is the same as &s[0]...);
3689 * - field 'a' access (corresponds to '2' in low-level spec);
3690 * - array element #3 access (corresponds to '3' in low-level spec).
3691 *
3692 */
3693static int bpf_core_spec_parse(const struct btf *btf,
3694 __u32 type_id,
3695 const char *spec_str,
3696 struct bpf_core_spec *spec)
3697{
3698 int access_idx, parsed_len, i;
1b484b30 3699 struct bpf_core_accessor *acc;
ddc7c304
AN		 3700 const struct btf_type *t;
3701 const char *name;
3702 __u32 id;
3703 __s64 sz;
3704
3705 if (str_is_empty(spec_str) || *spec_str == ':')
3706 return -EINVAL;
3707
3708 memset(spec, 0, sizeof(*spec));
3709 spec->btf = btf;
3710
3711 /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
3712 while (*spec_str) {
3713 if (*spec_str == ':')
3714 ++spec_str;
3715 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
3716 return -EINVAL;
3717 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
3718 return -E2BIG;
3719 spec_str += parsed_len;
3720 spec->raw_spec[spec->raw_len++] = access_idx;
3721 }
3722
3723 if (spec->raw_len == 0)
3724 return -EINVAL;
3725
3726 /* first spec value is always reloc type array index */
3727 t = skip_mods_and_typedefs(btf, type_id, &id);
3728 if (!t)
3729 return -EINVAL;
3730
3731 access_idx = spec->raw_spec[0];
3732 spec->spec[0].type_id = id;
3733 spec->spec[0].idx = access_idx;
3734 spec->len++;
3735
3736 sz = btf__resolve_size(btf, id);
3737 if (sz < 0)
3738 return sz;
ee26dade 3739 spec->bit_offset = access_idx * sz * 8;
ddc7c304
AN		 3740
3741 for (i = 1; i < spec->raw_len; i++) {
3742 t = skip_mods_and_typedefs(btf, id, &id);
3743 if (!t)
3744 return -EINVAL;
3745
3746 access_idx = spec->raw_spec[i];
1b484b30 3747 acc = &spec->spec[spec->len];
ddc7c304
AN		 3748
3749 if (btf_is_composite(t)) {
3750 const struct btf_member *m;
ee26dade 3751 __u32 bit_offset;
ddc7c304
AN		 3752
3753 if (access_idx >= btf_vlen(t))
3754 return -EINVAL;
ddc7c304 3755
ee26dade
AN		 3756 bit_offset = btf_member_bit_offset(t, access_idx);
3757 spec->bit_offset += bit_offset;
ddc7c304
AN		 3758
3759 m = btf_members(t) + access_idx;
3760 if (m->name_off) {
3761 name = btf__name_by_offset(btf, m->name_off);
3762 if (str_is_empty(name))
3763 return -EINVAL;
3764
1b484b30
AN		 3765 acc->type_id = id;
3766 acc->idx = access_idx;
3767 acc->name = name;
ddc7c304
AN		 3768 spec->len++;
3769 }
3770
3771 id = m->type;
3772 } else if (btf_is_array(t)) {
3773 const struct btf_array *a = btf_array(t);
1b484b30 3774 bool flex;
ddc7c304
AN		 3775
3776 t = skip_mods_and_typedefs(btf, a->type, &id);
1b484b30
AN		 3777 if (!t)
3778 return -EINVAL;
3779
3780 flex = is_flex_arr(btf, acc - 1, a);
3781 if (!flex && access_idx >= a->nelems)
ddc7c304
AN		 3782 return -EINVAL;
3783
3784 spec->spec[spec->len].type_id = id;
3785 spec->spec[spec->len].idx = access_idx;
3786 spec->len++;
3787
3788 sz = btf__resolve_size(btf, id);
3789 if (sz < 0)
3790 return sz;
ee26dade 3791 spec->bit_offset += access_idx * sz * 8;
ddc7c304 3792 } else {
be18010e
KW		 3793 pr_warn("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
3794 type_id, spec_str, i, id, btf_kind(t));
ddc7c304
AN		 3795 return -EINVAL;
3796 }
3797 }
3798
3799 return 0;
3800}
3801
3802static bool bpf_core_is_flavor_sep(const char *s)
3803{
3804 /* check X___Y name pattern, where X and Y are not underscores */
3805 return s[0] != '_' && /* X */
3806 s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */
3807 s[4] != '_'; /* Y */
3808}
3809
3810/* Given 'some_struct_name___with_flavor' return the length of a name prefix
3811 * before last triple underscore. Struct name part after last triple
3812 * underscore is ignored by BPF CO-RE relocation during relocation matching.
3813 */
3814static size_t bpf_core_essential_name_len(const char *name)
3815{
3816 size_t n = strlen(name);
3817 int i;
3818
3819 for (i = n - 5; i >= 0; i--) {
3820 if (bpf_core_is_flavor_sep(name + i))
3821 return i + 1;
3822 }
3823 return n;
3824}
3825
3826/* dynamically sized list of type IDs */
3827struct ids_vec {
3828 __u32 *data;
3829 int len;
3830};
3831
3832static void bpf_core_free_cands(struct ids_vec *cand_ids)
3833{
3834 free(cand_ids->data);
3835 free(cand_ids);
3836}
3837
3838static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
3839 __u32 local_type_id,
3840 const struct btf *targ_btf)
3841{
3842 size_t local_essent_len, targ_essent_len;
3843 const char *local_name, *targ_name;
3844 const struct btf_type *t;
3845 struct ids_vec *cand_ids;
3846 __u32 *new_ids;
3847 int i, err, n;
3848
3849 t = btf__type_by_id(local_btf, local_type_id);
3850 if (!t)
3851 return ERR_PTR(-EINVAL);
3852
3853 local_name = btf__name_by_offset(local_btf, t->name_off);
3854 if (str_is_empty(local_name))
3855 return ERR_PTR(-EINVAL);
3856 local_essent_len = bpf_core_essential_name_len(local_name);
3857
3858 cand_ids = calloc(1, sizeof(*cand_ids));
3859 if (!cand_ids)
3860 return ERR_PTR(-ENOMEM);
3861
3862 n = btf__get_nr_types(targ_btf);
3863 for (i = 1; i <= n; i++) {
3864 t = btf__type_by_id(targ_btf, i);
3865 targ_name = btf__name_by_offset(targ_btf, t->name_off);
3866 if (str_is_empty(targ_name))
3867 continue;
3868
3869 targ_essent_len = bpf_core_essential_name_len(targ_name);
3870 if (targ_essent_len != local_essent_len)
3871 continue;
3872
3873 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
3874 pr_debug("[%d] %s: found candidate [%d] %s\n",
3875 local_type_id, local_name, i, targ_name);
35b9211c
AN		 3876 new_ids = reallocarray(cand_ids->data,
3877 cand_ids->len + 1,
3878 sizeof(*cand_ids->data));
ddc7c304
AN		 3879 if (!new_ids) {
3880 err = -ENOMEM;
3881 goto err_out;
3882 }
3883 cand_ids->data = new_ids;
3884 cand_ids->data[cand_ids->len++] = i;
3885 }
3886 }
3887 return cand_ids;
3888err_out:
3889 bpf_core_free_cands(cand_ids);
3890 return ERR_PTR(err);
3891}
3892
3893/* Check two types for compatibility, skipping const/volatile/restrict and
ee26dade 3894 * typedefs, to ensure we are relocating compatible entities:
ddc7c304 3895 * - any two STRUCTs/UNIONs are compatible and can be mixed;
94f060e9 3896 * - any two FWDs are compatible, if their names match (modulo flavor suffix);
ddc7c304 3897 * - any two PTRs are always compatible;
94f060e9
AN		 3898 * - for ENUMs, names should be the same (ignoring flavor suffix) or at
3899 * least one of enums should be anonymous;
ddc7c304 3900 * - for ENUMs, check sizes, names are ignored;
ee26dade 3901 * - for INT, size and signedness are ignored;
ddc7c304
AN		 3902 * - for ARRAY, dimensionality is ignored, element types are checked for
3903 * compatibility recursively;
3904 * - everything else shouldn't be ever a target of relocation.
3905 * These rules are not set in stone and probably will be adjusted as we get
3906 * more experience with using BPF CO-RE relocations.
3907 */
3908static int bpf_core_fields_are_compat(const struct btf *local_btf,
3909 __u32 local_id,
3910 const struct btf *targ_btf,
3911 __u32 targ_id)
3912{
3913 const struct btf_type *local_type, *targ_type;
3914
3915recur:
3916 local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
3917 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
3918 if (!local_type || !targ_type)
3919 return -EINVAL;
3920
3921 if (btf_is_composite(local_type) && btf_is_composite(targ_type))
3922 return 1;
3923 if (btf_kind(local_type) != btf_kind(targ_type))
3924 return 0;
3925
3926 switch (btf_kind(local_type)) {
ddc7c304
AN		 3927 case BTF_KIND_PTR:
3928 return 1;
94f060e9
AN		 3929 case BTF_KIND_FWD:
3930 case BTF_KIND_ENUM: {
3931 const char *local_name, *targ_name;
3932 size_t local_len, targ_len;
3933
3934 local_name = btf__name_by_offset(local_btf,
3935 local_type->name_off);
3936 targ_name = btf__name_by_offset(targ_btf, targ_type->name_off);
3937 local_len = bpf_core_essential_name_len(local_name);
3938 targ_len = bpf_core_essential_name_len(targ_name);
3939 /* one of them is anonymous or both w/ same flavor-less names */
3940 return local_len == 0 || targ_len == 0 ||
3941 (local_len == targ_len &&
3942 strncmp(local_name, targ_name, local_len) == 0);
3943 }
ddc7c304 3944 case BTF_KIND_INT:
ee26dade
AN		 3945 /* just reject deprecated bitfield-like integers; all other
3946 * integers are by default compatible between each other
3947 */
ddc7c304 3948 return btf_int_offset(local_type) == 0 &&
ee26dade 3949 btf_int_offset(targ_type) == 0;
ddc7c304
AN		 3950 case BTF_KIND_ARRAY:
3951 local_id = btf_array(local_type)->type;
3952 targ_id = btf_array(targ_type)->type;
3953 goto recur;
3954 default:
be18010e
KW		 3955 pr_warn("unexpected kind %d relocated, local [%d], target [%d]\n",
3956 btf_kind(local_type), local_id, targ_id);
ddc7c304
AN		 3957 return 0;
3958 }
3959}
3960
3961/*
3962 * Given single high-level named field accessor in local type, find
3963 * corresponding high-level accessor for a target type. Along the way,
3964 * maintain low-level spec for target as well. Also keep updating target
ee26dade 3965 * bit offset.
ddc7c304
AN		 3966 *
3967 * Searching is performed through recursive exhaustive enumeration of all
3968 * fields of a struct/union. If there are any anonymous (embedded)
3969 * structs/unions, they are recursively searched as well. If field with
3970 * desired name is found, check compatibility between local and target types,
3971 * before returning result.
3972 *
3973 * 1 is returned, if field is found.
3974 * 0 is returned if no compatible field is found.
3975 * <0 is returned on error.
3976 */
3977static int bpf_core_match_member(const struct btf *local_btf,
3978 const struct bpf_core_accessor *local_acc,
3979 const struct btf *targ_btf,
3980 __u32 targ_id,
3981 struct bpf_core_spec *spec,
3982 __u32 *next_targ_id)
3983{
3984 const struct btf_type *local_type, *targ_type;
3985 const struct btf_member *local_member, *m;
3986 const char *local_name, *targ_name;
3987 __u32 local_id;
3988 int i, n, found;
3989
3990 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
3991 if (!targ_type)
3992 return -EINVAL;
3993 if (!btf_is_composite(targ_type))
3994 return 0;
3995
3996 local_id = local_acc->type_id;
3997 local_type = btf__type_by_id(local_btf, local_id);
3998 local_member = btf_members(local_type) + local_acc->idx;
3999 local_name = btf__name_by_offset(local_btf, local_member->name_off);
4000
4001 n = btf_vlen(targ_type);
4002 m = btf_members(targ_type);
4003 for (i = 0; i < n; i++, m++) {
ee26dade 4004 __u32 bit_offset;
ddc7c304 4005
ee26dade 4006 bit_offset = btf_member_bit_offset(targ_type, i);
ddc7c304
AN		 4007
4008 /* too deep struct/union/array nesting */
4009 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4010 return -E2BIG;
4011
4012 /* speculate this member will be the good one */
ee26dade 4013 spec->bit_offset += bit_offset;
ddc7c304
AN		 4014 spec->raw_spec[spec->raw_len++] = i;
4015
4016 targ_name = btf__name_by_offset(targ_btf, m->name_off);
4017 if (str_is_empty(targ_name)) {
4018 /* embedded struct/union, we need to go deeper */
4019 found = bpf_core_match_member(local_btf, local_acc,
4020 targ_btf, m->type,
4021 spec, next_targ_id);
4022 if (found) /* either found or error */
4023 return found;
4024 } else if (strcmp(local_name, targ_name) == 0) {
4025 /* matching named field */
4026 struct bpf_core_accessor *targ_acc;
4027
4028 targ_acc = &spec->spec[spec->len++];
4029 targ_acc->type_id = targ_id;
4030 targ_acc->idx = i;
4031 targ_acc->name = targ_name;
4032
4033 *next_targ_id = m->type;
4034 found = bpf_core_fields_are_compat(local_btf,
4035 local_member->type,
4036 targ_btf, m->type);
4037 if (!found)
4038 spec->len--; /* pop accessor */
4039 return found;
4040 }
4041 /* member turned out not to be what we looked for */
ee26dade 4042 spec->bit_offset -= bit_offset;
ddc7c304
AN		 4043 spec->raw_len--;
4044 }
4045
4046 return 0;
4047}
4048
4049/*
4050 * Try to match local spec to a target type and, if successful, produce full
ee26dade 4051 * target spec (high-level, low-level + bit offset).
ddc7c304
AN		 4052 */
4053static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
4054 const struct btf *targ_btf, __u32 targ_id,
4055 struct bpf_core_spec *targ_spec)
4056{
4057 const struct btf_type *targ_type;
4058 const struct bpf_core_accessor *local_acc;
4059 struct bpf_core_accessor *targ_acc;
4060 int i, sz, matched;
4061
4062 memset(targ_spec, 0, sizeof(*targ_spec));
4063 targ_spec->btf = targ_btf;
4064
4065 local_acc = &local_spec->spec[0];
4066 targ_acc = &targ_spec->spec[0];
4067
4068 for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
4069 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
4070 &targ_id);
4071 if (!targ_type)
4072 return -EINVAL;
4073
4074 if (local_acc->name) {
4075 matched = bpf_core_match_member(local_spec->btf,
4076 local_acc,
4077 targ_btf, targ_id,
4078 targ_spec, &targ_id);
4079 if (matched <= 0)
4080 return matched;
4081 } else {
4082 /* for i=0, targ_id is already treated as array element
4083 * type (because it's the original struct), for others
4084 * we should find array element type first
4085 */
4086 if (i > 0) {
4087 const struct btf_array *a;
1b484b30 4088 bool flex;
ddc7c304
AN		 4089
4090 if (!btf_is_array(targ_type))
4091 return 0;
4092
4093 a = btf_array(targ_type);
1b484b30
AN		 4094 flex = is_flex_arr(targ_btf, targ_acc - 1, a);
4095 if (!flex && local_acc->idx >= a->nelems)
ddc7c304
AN		 4096 return 0;
4097 if (!skip_mods_and_typedefs(targ_btf, a->type,
4098 &targ_id))
4099 return -EINVAL;
4100 }
4101
4102 /* too deep struct/union/array nesting */
4103 if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4104 return -E2BIG;
4105
4106 targ_acc->type_id = targ_id;
4107 targ_acc->idx = local_acc->idx;
4108 targ_acc->name = NULL;
4109 targ_spec->len++;
4110 targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
4111 targ_spec->raw_len++;
4112
4113 sz = btf__resolve_size(targ_btf, targ_id);
4114 if (sz < 0)
4115 return sz;
ee26dade 4116 targ_spec->bit_offset += local_acc->idx * sz * 8;
ddc7c304
AN		 4117 }
4118 }
4119
4120 return 1;
4121}
4122
ee26dade
AN		 4123static int bpf_core_calc_field_relo(const struct bpf_program *prog,
4124 const struct bpf_field_reloc *relo,
4125 const struct bpf_core_spec *spec,
4126 __u32 *val, bool *validate)
4127{
4128 const struct bpf_core_accessor *acc = &spec->spec[spec->len - 1];
4129 const struct btf_type *t = btf__type_by_id(spec->btf, acc->type_id);
4130 __u32 byte_off, byte_sz, bit_off, bit_sz;
4131 const struct btf_member *m;
4132 const struct btf_type *mt;
4133 bool bitfield;
94f060e9 4134 __s64 sz;
ee26dade
AN		 4135
4136 /* a[n] accessor needs special handling */
4137 if (!acc->name) {
94f060e9
AN		 4138 if (relo->kind == BPF_FIELD_BYTE_OFFSET) {
4139 *val = spec->bit_offset / 8;
4140 } else if (relo->kind == BPF_FIELD_BYTE_SIZE) {
4141 sz = btf__resolve_size(spec->btf, acc->type_id);
4142 if (sz < 0)
4143 return -EINVAL;
4144 *val = sz;
4145 } else {
4146 pr_warn("prog '%s': relo %d at insn #%d can't be applied to array access\n",
ee26dade
AN		 4147 bpf_program__title(prog, false),
4148 relo->kind, relo->insn_off / 8);
4149 return -EINVAL;
4150 }
ee26dade
AN		 4151 if (validate)
4152 *validate = true;
4153 return 0;
4154 }
4155
4156 m = btf_members(t) + acc->idx;
4157 mt = skip_mods_and_typedefs(spec->btf, m->type, NULL);
4158 bit_off = spec->bit_offset;
4159 bit_sz = btf_member_bitfield_size(t, acc->idx);
4160
4161 bitfield = bit_sz > 0;
4162 if (bitfield) {
4163 byte_sz = mt->size;
4164 byte_off = bit_off / 8 / byte_sz * byte_sz;
4165 /* figure out smallest int size necessary for bitfield load */
4166 while (bit_off + bit_sz - byte_off * 8 > byte_sz * 8) {
4167 if (byte_sz >= 8) {
4168 /* bitfield can't be read with 64-bit read */
4169 pr_warn("prog '%s': relo %d at insn #%d can't be satisfied for bitfield\n",
4170 bpf_program__title(prog, false),
4171 relo->kind, relo->insn_off / 8);
4172 return -E2BIG;
4173 }
4174 byte_sz *= 2;
4175 byte_off = bit_off / 8 / byte_sz * byte_sz;
4176 }
4177 } else {
94f060e9
AN		 4178 sz = btf__resolve_size(spec->btf, m->type);
4179 if (sz < 0)
4180 return -EINVAL;
4181 byte_sz = sz;
ee26dade
AN		 4182 byte_off = spec->bit_offset / 8;
4183 bit_sz = byte_sz * 8;
4184 }
4185
4186 /* for bitfields, all the relocatable aspects are ambiguous and we
4187 * might disagree with compiler, so turn off validation of expected
4188 * value, except for signedness
4189 */
4190 if (validate)
4191 *validate = !bitfield;
4192
4193 switch (relo->kind) {
4194 case BPF_FIELD_BYTE_OFFSET:
4195 *val = byte_off;
4196 break;
4197 case BPF_FIELD_BYTE_SIZE:
4198 *val = byte_sz;
4199 break;
4200 case BPF_FIELD_SIGNED:
4201 /* enums will be assumed unsigned */
4202 *val = btf_is_enum(mt) ||
4203 (btf_int_encoding(mt) & BTF_INT_SIGNED);
4204 if (validate)
4205 *validate = true; /* signedness is never ambiguous */
4206 break;
4207 case BPF_FIELD_LSHIFT_U64:
4208#if __BYTE_ORDER == __LITTLE_ENDIAN
4209 *val = 64 - (bit_off + bit_sz - byte_off * 8);
4210#else
4211 *val = (8 - byte_sz) * 8 + (bit_off - byte_off * 8);
4212#endif
4213 break;
4214 case BPF_FIELD_RSHIFT_U64:
4215 *val = 64 - bit_sz;
4216 if (validate)
4217 *validate = true; /* right shift is never ambiguous */
4218 break;
4219 case BPF_FIELD_EXISTS:
4220 default:
4221 pr_warn("prog '%s': unknown relo %d at insn #%d\n",
4222 bpf_program__title(prog, false),
4223 relo->kind, relo->insn_off / 8);
4224 return -EINVAL;
4225 }
4226
4227 return 0;
4228}
4229
ddc7c304
AN		 4230/*
4231 * Patch relocatable BPF instruction.
62561eb4
AN		 4232 *
4233 * Patched value is determined by relocation kind and target specification.
4234 * For field existence relocation target spec will be NULL if field is not
4235 * found.
4236 * Expected insn->imm value is determined using relocation kind and local
4237 * spec, and is checked before patching instruction. If actual insn->imm value
4238 * is wrong, bail out with error.
ddc7c304
AN		 4239 *
4240 * Currently three kinds of BPF instructions are supported:
4241 * 1. rX = <imm> (assignment with immediate operand);
4242 * 2. rX += <imm> (arithmetic operations with immediate operand);
ddc7c304 4243 */
62561eb4
AN		 4244static int bpf_core_reloc_insn(struct bpf_program *prog,
4245 const struct bpf_field_reloc *relo,
d7a25270 4246 int relo_idx,
62561eb4
AN		 4247 const struct bpf_core_spec *local_spec,
4248 const struct bpf_core_spec *targ_spec)
ddc7c304 4249{
62561eb4 4250 __u32 orig_val, new_val;
ddc7c304 4251 struct bpf_insn *insn;
d7a25270 4252 bool validate = true;
ee26dade 4253 int insn_idx, err;
ddc7c304
AN		 4254 __u8 class;
4255
62561eb4
AN		 4256 if (relo->insn_off % sizeof(struct bpf_insn))
4257 return -EINVAL;
4258 insn_idx = relo->insn_off / sizeof(struct bpf_insn);
d7a25270
AN		 4259 insn = &prog->insns[insn_idx];
4260 class = BPF_CLASS(insn->code);
62561eb4 4261
ee26dade 4262 if (relo->kind == BPF_FIELD_EXISTS) {
62561eb4
AN		 4263 orig_val = 1; /* can't generate EXISTS relo w/o local field */
4264 new_val = targ_spec ? 1 : 0;
ee26dade 4265 } else if (!targ_spec) {
d7a25270
AN		 4266 pr_debug("prog '%s': relo #%d: substituting insn #%d w/ invalid insn\n",
4267 bpf_program__title(prog, false), relo_idx, insn_idx);
4268 insn->code = BPF_JMP | BPF_CALL;
4269 insn->dst_reg = 0;
4270 insn->src_reg = 0;
4271 insn->off = 0;
4272 /* if this instruction is reachable (not a dead code),
4273 * verifier will complain with the following message:
4274 * invalid func unknown#195896080
4275 */
4276 insn->imm = 195896080; /* => 0xbad2310 => "bad relo" */
4277 return 0;
ee26dade
AN		 4278 } else {
4279 err = bpf_core_calc_field_relo(prog, relo, local_spec,
4280 &orig_val, &validate);
4281 if (err)
4282 return err;
4283 err = bpf_core_calc_field_relo(prog, relo, targ_spec,
4284 &new_val, NULL);
4285 if (err)
4286 return err;
62561eb4 4287 }
ddc7c304 4288
8ab9da57
AN		 4289 switch (class) {
4290 case BPF_ALU:
4291 case BPF_ALU64:
ddc7c304
AN		 4292 if (BPF_SRC(insn->code) != BPF_K)
4293 return -EINVAL;
d7a25270
AN		 4294 if (validate && insn->imm != orig_val) {
4295 pr_warn("prog '%s': relo #%d: unexpected insn #%d (ALU/ALU64) value: got %u, exp %u -> %u\n",
4296 bpf_program__title(prog, false), relo_idx,
4297 insn_idx, insn->imm, orig_val, new_val);
ddc7c304 4298 return -EINVAL;
ee26dade
AN		 4299 }
4300 orig_val = insn->imm;
62561eb4 4301 insn->imm = new_val;
d7a25270
AN		 4302 pr_debug("prog '%s': relo #%d: patched insn #%d (ALU/ALU64) imm %u -> %u\n",
4303 bpf_program__title(prog, false), relo_idx, insn_idx,
4304 orig_val, new_val);
8ab9da57
AN		 4305 break;
4306 case BPF_LDX:
4307 case BPF_ST:
4308 case BPF_STX:
d7a25270
AN		 4309 if (validate && insn->off != orig_val) {
4310 pr_warn("prog '%s': relo #%d: unexpected insn #%d (LD/LDX/ST/STX) value: got %u, exp %u -> %u\n",
4311 bpf_program__title(prog, false), relo_idx,
4312 insn_idx, insn->off, orig_val, new_val);
8ab9da57
AN		 4313 return -EINVAL;
4314 }
4315 if (new_val > SHRT_MAX) {
d7a25270
AN		 4316 pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) value too big: %u\n",
4317 bpf_program__title(prog, false), relo_idx,
4318 insn_idx, new_val);
8ab9da57
AN		 4319 return -ERANGE;
4320 }
4321 orig_val = insn->off;
4322 insn->off = new_val;
d7a25270
AN		 4323 pr_debug("prog '%s': relo #%d: patched insn #%d (LDX/ST/STX) off %u -> %u\n",
4324 bpf_program__title(prog, false), relo_idx, insn_idx,
4325 orig_val, new_val);
8ab9da57
AN		 4326 break;
4327 default:
d7a25270
AN		 4328 pr_warn("prog '%s': relo #%d: trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
4329 bpf_program__title(prog, false), relo_idx,
be18010e
KW		 4330 insn_idx, insn->code, insn->src_reg, insn->dst_reg,
4331 insn->off, insn->imm);
ddc7c304
AN		 4332 return -EINVAL;
4333 }
62561eb4 4334
ddc7c304
AN		 4335 return 0;
4336}
4337
ddc7c304
AN		 4338/* Output spec definition in the format:
4339 * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
4340 * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
4341 */
4342static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
4343{
4344 const struct btf_type *t;
4345 const char *s;
4346 __u32 type_id;
4347 int i;
4348
4349 type_id = spec->spec[0].type_id;
4350 t = btf__type_by_id(spec->btf, type_id);
4351 s = btf__name_by_offset(spec->btf, t->name_off);
4352 libbpf_print(level, "[%u] %s + ", type_id, s);
4353
4354 for (i = 0; i < spec->raw_len; i++)
4355 libbpf_print(level, "%d%s", spec->raw_spec[i],
4356 i == spec->raw_len - 1 ? " => " : ":");
4357
ee26dade
AN		 4358 libbpf_print(level, "%u.%u @ &x",
4359 spec->bit_offset / 8, spec->bit_offset % 8);
ddc7c304
AN		 4360
4361 for (i = 0; i < spec->len; i++) {
4362 if (spec->spec[i].name)
4363 libbpf_print(level, ".%s", spec->spec[i].name);
4364 else
4365 libbpf_print(level, "[%u]", spec->spec[i].idx);
4366 }
4367
4368}
4369
4370static size_t bpf_core_hash_fn(const void *key, void *ctx)
4371{
4372 return (size_t)key;
4373}
4374
4375static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
4376{
4377 return k1 == k2;
4378}
4379
4380static void *u32_as_hash_key(__u32 x)
4381{
4382 return (void *)(uintptr_t)x;
4383}
4384
4385/*
4386 * CO-RE relocate single instruction.
4387 *
4388 * The outline and important points of the algorithm:
4389 * 1. For given local type, find corresponding candidate target types.
4390 * Candidate type is a type with the same "essential" name, ignoring
4391 * everything after last triple underscore (___). E.g., `sample`,
4392 * `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
4393 * for each other. Names with triple underscore are referred to as
4394 * "flavors" and are useful, among other things, to allow to
4395 * specify/support incompatible variations of the same kernel struct, which
4396 * might differ between different kernel versions and/or build
4397 * configurations.
4398 *
4399 * N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
4400 * converter, when deduplicated BTF of a kernel still contains more than
4401 * one different types with the same name. In that case, ___2, ___3, etc
4402 * are appended starting from second name conflict. But start flavors are
4403 * also useful to be defined "locally", in BPF program, to extract same
4404 * data from incompatible changes between different kernel
4405 * versions/configurations. For instance, to handle field renames between
4406 * kernel versions, one can use two flavors of the struct name with the
4407 * same common name and use conditional relocations to extract that field,
4408 * depending on target kernel version.
4409 * 2. For each candidate type, try to match local specification to this
4410 * candidate target type. Matching involves finding corresponding
4411 * high-level spec accessors, meaning that all named fields should match,
4412 * as well as all array accesses should be within the actual bounds. Also,
4413 * types should be compatible (see bpf_core_fields_are_compat for details).
4414 * 3. It is supported and expected that there might be multiple flavors
4415 * matching the spec. As long as all the specs resolve to the same set of
511bb008 4416 * offsets across all candidates, there is no error. If there is any
ddc7c304
AN		 4417 * ambiguity, CO-RE relocation will fail. This is necessary to accomodate
4418 * imprefection of BTF deduplication, which can cause slight duplication of
4419 * the same BTF type, if some directly or indirectly referenced (by
4420 * pointer) type gets resolved to different actual types in different
4421 * object files. If such situation occurs, deduplicated BTF will end up
4422 * with two (or more) structurally identical types, which differ only in
4423 * types they refer to through pointer. This should be OK in most cases and
4424 * is not an error.
4425 * 4. Candidate types search is performed by linearly scanning through all
4426 * types in target BTF. It is anticipated that this is overall more
4427 * efficient memory-wise and not significantly worse (if not better)
4428 * CPU-wise compared to prebuilding a map from all local type names to
4429 * a list of candidate type names. It's also sped up by caching resolved
4430 * list of matching candidates per each local "root" type ID, that has at
511bb008 4431 * least one bpf_field_reloc associated with it. This list is shared
ddc7c304
AN		 4432 * between multiple relocations for the same type ID and is updated as some
4433 * of the candidates are pruned due to structural incompatibility.
4434 */
511bb008
AN		 4435static int bpf_core_reloc_field(struct bpf_program *prog,
4436 const struct bpf_field_reloc *relo,
ddc7c304
AN		 4437 int relo_idx,
4438 const struct btf *local_btf,
4439 const struct btf *targ_btf,
4440 struct hashmap *cand_cache)
4441{
4442 const char *prog_name = bpf_program__title(prog, false);
4443 struct bpf_core_spec local_spec, cand_spec, targ_spec;
4444 const void *type_key = u32_as_hash_key(relo->type_id);
4445 const struct btf_type *local_type, *cand_type;
4446 const char *local_name, *cand_name;
4447 struct ids_vec *cand_ids;
4448 __u32 local_id, cand_id;
4449 const char *spec_str;
4450 int i, j, err;
4451
4452 local_id = relo->type_id;
4453 local_type = btf__type_by_id(local_btf, local_id);
4454 if (!local_type)
4455 return -EINVAL;
4456
4457 local_name = btf__name_by_offset(local_btf, local_type->name_off);
4458 if (str_is_empty(local_name))
4459 return -EINVAL;
4460
4461 spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
4462 if (str_is_empty(spec_str))
4463 return -EINVAL;
4464
4465 err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
4466 if (err) {
be18010e
KW		 4467 pr_warn("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
4468 prog_name, relo_idx, local_id, local_name, spec_str,
4469 err);
ddc7c304
AN		 4470 return -EINVAL;
4471 }
4472
ee26dade
AN		 4473 pr_debug("prog '%s': relo #%d: kind %d, spec is ", prog_name, relo_idx,
4474 relo->kind);
ddc7c304
AN		 4475 bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
4476 libbpf_print(LIBBPF_DEBUG, "\n");
4477
4478 if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
4479 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
4480 if (IS_ERR(cand_ids)) {
be18010e
KW		 4481 pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
4482 prog_name, relo_idx, local_id, local_name,
4483 PTR_ERR(cand_ids));
ddc7c304
AN		 4484 return PTR_ERR(cand_ids);
4485 }
4486 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
4487 if (err) {
4488 bpf_core_free_cands(cand_ids);
4489 return err;
4490 }
4491 }
4492
4493 for (i = 0, j = 0; i < cand_ids->len; i++) {
4494 cand_id = cand_ids->data[i];
4495 cand_type = btf__type_by_id(targ_btf, cand_id);
4496 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
4497
4498 err = bpf_core_spec_match(&local_spec, targ_btf,
4499 cand_id, &cand_spec);
4500 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
4501 prog_name, relo_idx, i, cand_name);
4502 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
4503 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
4504 if (err < 0) {
be18010e
KW		 4505 pr_warn("prog '%s': relo #%d: matching error: %d\n",
4506 prog_name, relo_idx, err);
ddc7c304
AN		 4507 return err;
4508 }
4509 if (err == 0)
4510 continue;
4511
4512 if (j == 0) {
4513 targ_spec = cand_spec;
ee26dade 4514 } else if (cand_spec.bit_offset != targ_spec.bit_offset) {
ddc7c304 4515 /* if there are many candidates, they should all
ee26dade 4516 * resolve to the same bit offset
ddc7c304 4517 */
be18010e 4518 pr_warn("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
ee26dade
AN		 4519 prog_name, relo_idx, cand_spec.bit_offset,
4520 targ_spec.bit_offset);
ddc7c304
AN		 4521 return -EINVAL;
4522 }
4523
4524 cand_ids->data[j++] = cand_spec.spec[0].type_id;
4525 }
4526
62561eb4 4527 /*
d7a25270
AN		 4528 * For BPF_FIELD_EXISTS relo or when used BPF program has field
4529 * existence checks or kernel version/config checks, it's expected
4530 * that we might not find any candidates. In this case, if field
4531 * wasn't found in any candidate, the list of candidates shouldn't
4532 * change at all, we'll just handle relocating appropriately,
4533 * depending on relo's kind.
62561eb4
AN		 4534 */
4535 if (j > 0)
4536 cand_ids->len = j;
4537
d7a25270
AN		 4538 /*
4539 * If no candidates were found, it might be both a programmer error,
4540 * as well as expected case, depending whether instruction w/
4541 * relocation is guarded in some way that makes it unreachable (dead
4542 * code) if relocation can't be resolved. This is handled in
4543 * bpf_core_reloc_insn() uniformly by replacing that instruction with
4544 * BPF helper call insn (using invalid helper ID). If that instruction
4545 * is indeed unreachable, then it will be ignored and eliminated by
4546 * verifier. If it was an error, then verifier will complain and point
4547 * to a specific instruction number in its log.
4548 */
4549 if (j == 0)
4550 pr_debug("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
4551 prog_name, relo_idx, local_id, local_name, spec_str);
ddc7c304 4552
62561eb4 4553 /* bpf_core_reloc_insn should know how to handle missing targ_spec */
d7a25270 4554 err = bpf_core_reloc_insn(prog, relo, relo_idx, &local_spec,
62561eb4 4555 j ? &targ_spec : NULL);
ddc7c304 4556 if (err) {
be18010e
KW		 4557 pr_warn("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
4558 prog_name, relo_idx, relo->insn_off, err);
ddc7c304
AN		 4559 return -EINVAL;
4560 }
4561
4562 return 0;
4563}
4564
4565static int
511bb008 4566bpf_core_reloc_fields(struct bpf_object *obj, const char *targ_btf_path)
ddc7c304
AN		 4567{
4568 const struct btf_ext_info_sec *sec;
511bb008 4569 const struct bpf_field_reloc *rec;
ddc7c304
AN		 4570 const struct btf_ext_info *seg;
4571 struct hashmap_entry *entry;
4572 struct hashmap *cand_cache = NULL;
4573 struct bpf_program *prog;
4574 struct btf *targ_btf;
4575 const char *sec_name;
4576 int i, err = 0;
4577
4578 if (targ_btf_path)
4579 targ_btf = btf__parse_elf(targ_btf_path, NULL);
4580 else
fb2426ad 4581 targ_btf = libbpf_find_kernel_btf();
ddc7c304 4582 if (IS_ERR(targ_btf)) {
be18010e 4583 pr_warn("failed to get target BTF: %ld\n", PTR_ERR(targ_btf));
ddc7c304
AN		 4584 return PTR_ERR(targ_btf);
4585 }
4586
4587 cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
4588 if (IS_ERR(cand_cache)) {
4589 err = PTR_ERR(cand_cache);
4590 goto out;
4591 }
4592
511bb008 4593 seg = &obj->btf_ext->field_reloc_info;
ddc7c304
AN		 4594 for_each_btf_ext_sec(seg, sec) {
4595 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
4596 if (str_is_empty(sec_name)) {
4597 err = -EINVAL;
4598 goto out;
4599 }
4600 prog = bpf_object__find_program_by_title(obj, sec_name);
4601 if (!prog) {
be18010e
KW		 4602 pr_warn("failed to find program '%s' for CO-RE offset relocation\n",
4603 sec_name);
ddc7c304
AN		 4604 err = -EINVAL;
4605 goto out;
4606 }
4607
4608 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
4609 sec_name, sec->num_info);
4610
4611 for_each_btf_ext_rec(seg, sec, i, rec) {
511bb008
AN		 4612 err = bpf_core_reloc_field(prog, rec, i, obj->btf,
4613 targ_btf, cand_cache);
ddc7c304 4614 if (err) {
be18010e
KW		 4615 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
4616 sec_name, i, err);
ddc7c304
AN		 4617 goto out;
4618 }
4619 }
4620 }
4621
4622out:
4623 btf__free(targ_btf);
4624 if (!IS_ERR_OR_NULL(cand_cache)) {
4625 hashmap__for_each_entry(cand_cache, entry, i) {
4626 bpf_core_free_cands(entry->value);
4627 }
4628 hashmap__free(cand_cache);
4629 }
4630 return err;
4631}
4632
4633static int
4634bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
4635{
4636 int err = 0;
4637
511bb008
AN		 4638 if (obj->btf_ext->field_reloc_info.len)
4639 err = bpf_core_reloc_fields(obj, targ_btf_path);
ddc7c304
AN		 4640
4641 return err;
4642}
4643
48cca7e4
AS		 4644static int
4645bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
4646 struct reloc_desc *relo)
4647{
4648 struct bpf_insn *insn, *new_insn;
4649 struct bpf_program *text;
4650 size_t new_cnt;
2993e051 4651 int err;
48cca7e4 4652
9173cac3 4653 if (prog->idx != obj->efile.text_shndx && prog->main_prog_cnt == 0) {
48cca7e4
AS		 4654 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
4655 if (!text) {
be18010e 4656 pr_warn("no .text section found yet relo into text exist\n");
48cca7e4
AS		 4657 return -LIBBPF_ERRNO__RELOC;
4658 }
4659 new_cnt = prog->insns_cnt + text->insns_cnt;
531b014e 4660 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
48cca7e4 4661 if (!new_insn) {
be18010e 4662 pr_warn("oom in prog realloc\n");
48cca7e4
AS		 4663 return -ENOMEM;
4664 }
3dc5e059 4665 prog->insns = new_insn;
2993e051 4666
f0187f0b
MKL		 4667 if (obj->btf_ext) {
4668 err = bpf_program_reloc_btf_ext(prog, obj,
4669 text->section_name,
4670 prog->insns_cnt);
4671 if (err)
2993e051 4672 return err;
2993e051
YS		 4673 }
4674
48cca7e4
AS		 4675 memcpy(new_insn + prog->insns_cnt, text->insns,
4676 text->insns_cnt * sizeof(*insn));
48cca7e4
AS		 4677 prog->main_prog_cnt = prog->insns_cnt;
4678 prog->insns_cnt = new_cnt;
b1a2ce82
JC		 4679 pr_debug("added %zd insn from %s to prog %s\n",
4680 text->insns_cnt, text->section_name,
4681 prog->section_name);
48cca7e4 4682 }
9173cac3 4683
48cca7e4 4684 insn = &prog->insns[relo->insn_idx];
53f8dd43 4685 insn->imm += relo->sym_off / 8 + prog->main_prog_cnt - relo->insn_idx;
48cca7e4
AS		 4686 return 0;
4687}
4688
8a47a6c5 4689static int
9d759a9b 4690bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
8a47a6c5 4691{
48cca7e4 4692 int i, err;
8a47a6c5 4693
2993e051
YS		 4694 if (!prog)
4695 return 0;
4696
f0187f0b
MKL		 4697 if (obj->btf_ext) {
4698 err = bpf_program_reloc_btf_ext(prog, obj,
4699 prog->section_name, 0);
4700 if (err)
2993e051 4701 return err;
2993e051
YS		 4702 }
4703
4704 if (!prog->reloc_desc)
8a47a6c5
WN		 4705 return 0;
4706
4707 for (i = 0; i < prog->nr_reloc; i++) {
53f8dd43 4708 struct reloc_desc *relo = &prog->reloc_desc[i];
166750bc 4709 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
8a47a6c5 4710
166750bc
AN		 4711 if (relo->insn_idx + 1 >= (int)prog->insns_cnt) {
4712 pr_warn("relocation out of range: '%s'\n",
4713 prog->section_name);
4714 return -LIBBPF_ERRNO__RELOC;
4715 }
d859900c 4716
166750bc
AN		 4717 switch (relo->type) {
4718 case RELO_LD64:
4719 insn[0].src_reg = BPF_PSEUDO_MAP_FD;
4720 insn[0].imm = obj->maps[relo->map_idx].fd;
4721 break;
4722 case RELO_DATA:
4723 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
4724 insn[1].imm = insn[0].imm + relo->sym_off;
53f8dd43 4725 insn[0].imm = obj->maps[relo->map_idx].fd;
166750bc
AN		 4726 break;
4727 case RELO_EXTERN:
4728 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
81bfdd08 4729 insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
166750bc
AN		 4730 insn[1].imm = relo->sym_off;
4731 break;
4732 case RELO_CALL:
53f8dd43 4733 err = bpf_program__reloc_text(prog, obj, relo);
48cca7e4
AS		 4734 if (err)
4735 return err;
166750bc
AN		 4736 break;
4737 default:
4738 pr_warn("relo #%d: bad relo type %d\n", i, relo->type);
4739 return -EINVAL;
8a47a6c5 4740 }
8a47a6c5
WN		 4741 }
4742
4743 zfree(&prog->reloc_desc);
4744 prog->nr_reloc = 0;
4745 return 0;
4746}
4747
8a47a6c5 4748static int
ddc7c304 4749bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
8a47a6c5
WN		 4750{
4751 struct bpf_program *prog;
4752 size_t i;
4753 int err;
4754
ddc7c304
AN		 4755 if (obj->btf_ext) {
4756 err = bpf_object__relocate_core(obj, targ_btf_path);
4757 if (err) {
be18010e
KW		 4758 pr_warn("failed to perform CO-RE relocations: %d\n",
4759 err);
ddc7c304
AN		 4760 return err;
4761 }
4762 }
9173cac3
AN		 4763 /* ensure .text is relocated first, as it's going to be copied as-is
4764 * later for sub-program calls
4765 */
4766 for (i = 0; i < obj->nr_programs; i++) {
4767 prog = &obj->programs[i];
4768 if (prog->idx != obj->efile.text_shndx)
4769 continue;
4770
4771 err = bpf_program__relocate(prog, obj);
4772 if (err) {
4773 pr_warn("failed to relocate '%s'\n", prog->section_name);
4774 return err;
4775 }
4776 break;
4777 }
4778 /* now relocate everything but .text, which by now is relocated
4779 * properly, so we can copy raw sub-program instructions as is safely
4780 */
8a47a6c5
WN		 4781 for (i = 0; i < obj->nr_programs; i++) {
4782 prog = &obj->programs[i];
9173cac3
AN		 4783 if (prog->idx == obj->efile.text_shndx)
4784 continue;
8a47a6c5 4785
9d759a9b 4786 err = bpf_program__relocate(prog, obj);
8a47a6c5 4787 if (err) {
be18010e 4788 pr_warn("failed to relocate '%s'\n", prog->section_name);
8a47a6c5
WN		 4789 return err;
4790 }
4791 }
4792 return 0;
4793}
4794
590a0088
MKL		 4795static int bpf_object__collect_struct_ops_map_reloc(struct bpf_object *obj,
4796 GElf_Shdr *shdr,
4797 Elf_Data *data);
4798
34090915
WN		 4799static int bpf_object__collect_reloc(struct bpf_object *obj)
4800{
4801 int i, err;
4802
4803 if (!obj_elf_valid(obj)) {
be18010e 4804 pr_warn("Internal error: elf object is closed\n");
6371ca3b 4805 return -LIBBPF_ERRNO__INTERNAL;
34090915
WN		 4806 }
4807
1f8e2bcb
AN		 4808 for (i = 0; i < obj->efile.nr_reloc_sects; i++) {
4809 GElf_Shdr *shdr = &obj->efile.reloc_sects[i].shdr;
4810 Elf_Data *data = obj->efile.reloc_sects[i].data;
34090915
WN		 4811 int idx = shdr->sh_info;
4812 struct bpf_program *prog;
34090915
WN		 4813
4814 if (shdr->sh_type != SHT_REL) {
be18010e 4815 pr_warn("internal error at %d\n", __LINE__);
6371ca3b 4816 return -LIBBPF_ERRNO__INTERNAL;
34090915
WN		 4817 }
4818
590a0088
MKL		 4819 if (idx == obj->efile.st_ops_shndx) {
4820 err = bpf_object__collect_struct_ops_map_reloc(obj,
4821 shdr,
4822 data);
4823 if (err)
4824 return err;
4825 continue;
4826 }
4827
34090915
WN		 4828 prog = bpf_object__find_prog_by_idx(obj, idx);
4829 if (!prog) {
be18010e 4830 pr_warn("relocation failed: no section(%d)\n", idx);
6371ca3b 4831 return -LIBBPF_ERRNO__RELOC;
34090915
WN		 4832 }
4833
399dc65e 4834 err = bpf_program__collect_reloc(prog, shdr, data, obj);
34090915 4835 if (err)
6371ca3b 4836 return err;
34090915
WN		 4837 }
4838 return 0;
4839}
4840
55cffde2 4841static int
2993e051 4842load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
f0187f0b 4843 char *license, __u32 kern_version, int *pfd)
55cffde2 4844{
d7be143b 4845 struct bpf_load_program_attr load_attr;
1ce6a9fc 4846 char *cp, errmsg[STRERR_BUFSIZE];
da11b417 4847 int log_buf_size = BPF_LOG_BUF_SIZE;
55cffde2 4848 char *log_buf;
5d01ab7b 4849 int btf_fd, ret;
55cffde2 4850
fba01a06
AN		 4851 if (!insns || !insns_cnt)
4852 return -EINVAL;
4853
d7be143b 4854 memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
2993e051
YS		 4855 load_attr.prog_type = prog->type;
4856 load_attr.expected_attach_type = prog->expected_attach_type;
5b32a23e
SF		 4857 if (prog->caps->name)
4858 load_attr.name = prog->name;
d7be143b
AI		 4859 load_attr.insns = insns;
4860 load_attr.insns_cnt = insns_cnt;
4861 load_attr.license = license;
590a0088
MKL		 4862 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) {
4863 load_attr.attach_btf_id = prog->attach_btf_id;
2db6eab1
AS		 4864 } else if (prog->type == BPF_PROG_TYPE_TRACING ||
4865 prog->type == BPF_PROG_TYPE_EXT) {
e7bf94db
AS		 4866 load_attr.attach_prog_fd = prog->attach_prog_fd;
4867 load_attr.attach_btf_id = prog->attach_btf_id;
4868 } else {
4869 load_attr.kern_version = kern_version;
4870 load_attr.prog_ifindex = prog->prog_ifindex;
4871 }
3415ec64
AN		 4872 /* if .BTF.ext was loaded, kernel supports associated BTF for prog */
4873 if (prog->obj->btf_ext)
4874 btf_fd = bpf_object__btf_fd(prog->obj);
4875 else
4876 btf_fd = -1;
5d01ab7b 4877 load_attr.prog_btf_fd = btf_fd >= 0 ? btf_fd : 0;
2993e051
YS		 4878 load_attr.func_info = prog->func_info;
4879 load_attr.func_info_rec_size = prog->func_info_rec_size;
f0187f0b 4880 load_attr.func_info_cnt = prog->func_info_cnt;
3d650141
MKL		 4881 load_attr.line_info = prog->line_info;
4882 load_attr.line_info_rec_size = prog->line_info_rec_size;
4883 load_attr.line_info_cnt = prog->line_info_cnt;
da11b417 4884 load_attr.log_level = prog->log_level;
04656198 4885 load_attr.prog_flags = prog->prog_flags;
55cffde2 4886
da11b417
AS		 4887retry_load:
4888 log_buf = malloc(log_buf_size);
55cffde2 4889 if (!log_buf)
be18010e 4890 pr_warn("Alloc log buffer for bpf loader error, continue without log\n");
55cffde2 4891
da11b417 4892 ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
55cffde2
WN		 4893
4894 if (ret >= 0) {
da11b417
AS		 4895 if (load_attr.log_level)
4896 pr_debug("verifier log:\n%s", log_buf);
55cffde2
WN		 4897 *pfd = ret;
4898 ret = 0;
4899 goto out;
4900 }
4901
da11b417
AS		 4902 if (errno == ENOSPC) {
4903 log_buf_size <<= 1;
4904 free(log_buf);
4905 goto retry_load;
4906 }
4f33ddb4 4907 ret = -errno;
24d6a808 4908 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
be18010e 4909 pr_warn("load bpf program failed: %s\n", cp);
dc3a2d25 4910 pr_perm_msg(ret);
55cffde2 4911
6371ca3b
WN		 4912 if (log_buf && log_buf[0] != '\0') {
4913 ret = -LIBBPF_ERRNO__VERIFY;
be18010e
KW		 4914 pr_warn("-- BEGIN DUMP LOG ---\n");
4915 pr_warn("\n%s\n", log_buf);
4916 pr_warn("-- END LOG --\n");
d7be143b 4917 } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
be18010e
KW		 4918 pr_warn("Program too large (%zu insns), at most %d insns\n",
4919 load_attr.insns_cnt, BPF_MAXINSNS);
705fa219 4920 ret = -LIBBPF_ERRNO__PROG2BIG;
4f33ddb4 4921 } else if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
705fa219 4922 /* Wrong program type? */
4f33ddb4 4923 int fd;
705fa219 4924
4f33ddb4
THJ		 4925 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
4926 load_attr.expected_attach_type = 0;
4927 fd = bpf_load_program_xattr(&load_attr, NULL, 0);
4928 if (fd >= 0) {
4929 close(fd);
4930 ret = -LIBBPF_ERRNO__PROGTYPE;
4931 goto out;
4932 }
55cffde2
WN		 4933 }
4934
4935out:
4936 free(log_buf);
4937 return ret;
4938}
4939
a6ed02ca 4940static int libbpf_find_attach_btf_id(struct bpf_program *prog);
13acb508
AN		 4941
4942int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver)
55cffde2 4943{
13acb508
AN		 4944 int err = 0, fd, i, btf_id;
4945
ff26ce5c
EC		 4946 if ((prog->type == BPF_PROG_TYPE_TRACING ||
4947 prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) {
a6ed02ca 4948 btf_id = libbpf_find_attach_btf_id(prog);
13acb508
AN		 4949 if (btf_id <= 0)
4950 return btf_id;
4951 prog->attach_btf_id = btf_id;
4952 }
55cffde2 4953
b580563e
WN		 4954 if (prog->instances.nr < 0 || !prog->instances.fds) {
4955 if (prog->preprocessor) {
be18010e
KW		 4956 pr_warn("Internal error: can't load program '%s'\n",
4957 prog->section_name);
b580563e
WN		 4958 return -LIBBPF_ERRNO__INTERNAL;
4959 }
55cffde2 4960
b580563e
WN		 4961 prog->instances.fds = malloc(sizeof(int));
4962 if (!prog->instances.fds) {
be18010e 4963 pr_warn("Not enough memory for BPF fds\n");
b580563e
WN		 4964 return -ENOMEM;
4965 }
4966 prog->instances.nr = 1;
4967 prog->instances.fds[0] = -1;
4968 }
4969
4970 if (!prog->preprocessor) {
4971 if (prog->instances.nr != 1) {
be18010e
KW		 4972 pr_warn("Program '%s' is inconsistent: nr(%d) != 1\n",
4973 prog->section_name, prog->instances.nr);
b580563e 4974 }
2993e051 4975 err = load_program(prog, prog->insns, prog->insns_cnt,
13acb508 4976 license, kern_ver, &fd);
b580563e
WN		 4977 if (!err)
4978 prog->instances.fds[0] = fd;
4979 goto out;
4980 }
4981
4982 for (i = 0; i < prog->instances.nr; i++) {
4983 struct bpf_prog_prep_result result;
4984 bpf_program_prep_t preprocessor = prog->preprocessor;
4985
1ad9cbb8 4986 memset(&result, 0, sizeof(result));
b580563e
WN		 4987 err = preprocessor(prog, i, prog->insns,
4988 prog->insns_cnt, &result);
4989 if (err) {
be18010e
KW		 4990 pr_warn("Preprocessing the %dth instance of program '%s' failed\n",
4991 i, prog->section_name);
b580563e
WN		 4992 goto out;
4993 }
4994
4995 if (!result.new_insn_ptr || !result.new_insn_cnt) {
4996 pr_debug("Skip loading the %dth instance of program '%s'\n",
4997 i, prog->section_name);
4998 prog->instances.fds[i] = -1;
4999 if (result.pfd)
5000 *result.pfd = -1;
5001 continue;
5002 }
5003
2993e051 5004 err = load_program(prog, result.new_insn_ptr,
13acb508 5005 result.new_insn_cnt, license, kern_ver, &fd);
b580563e 5006 if (err) {
be18010e
KW		 5007 pr_warn("Loading the %dth instance of program '%s' failed\n",
5008 i, prog->section_name);
b580563e
WN		 5009 goto out;
5010 }
5011
5012 if (result.pfd)
5013 *result.pfd = fd;
5014 prog->instances.fds[i] = fd;
5015 }
5016out:
55cffde2 5017 if (err)
be18010e 5018 pr_warn("failed to load program '%s'\n", prog->section_name);
55cffde2
WN		 5019 zfree(&prog->insns);
5020 prog->insns_cnt = 0;
5021 return err;
5022}
5023
a324aae3
AN		 5024static bool bpf_program__is_function_storage(const struct bpf_program *prog,
5025 const struct bpf_object *obj)
9a94f277
JK		 5026{
5027 return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
5028}
5029
55cffde2 5030static int
60276f98 5031bpf_object__load_progs(struct bpf_object *obj, int log_level)
55cffde2
WN		 5032{
5033 size_t i;
5034 int err;
5035
5036 for (i = 0; i < obj->nr_programs; i++) {
9a94f277 5037 if (bpf_program__is_function_storage(&obj->programs[i], obj))
48cca7e4 5038 continue;
501b125a 5039 obj->programs[i].log_level |= log_level;
55cffde2
WN		 5040 err = bpf_program__load(&obj->programs[i],
5041 obj->license,
5042 obj->kern_version);
5043 if (err)
5044 return err;
5045 }
5046 return 0;
5047}
5048
1a5e3fb1 5049static struct bpf_object *
5e61f270 5050__bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
01af3bf0 5051 const struct bpf_object_open_opts *opts)
1a5e3fb1 5052{
8601fd42 5053 const char *obj_name, *kconfig;
dd4436bb 5054 struct bpf_program *prog;
1a5e3fb1 5055 struct bpf_object *obj;
291ee02b 5056 char tmp_name[64];
6371ca3b 5057 int err;
1a5e3fb1
WN		 5058
5059 if (elf_version(EV_CURRENT) == EV_NONE) {
be18010e
KW		 5060 pr_warn("failed to init libelf for %s\n",
5061 path ? : "(mem buf)");
6371ca3b 5062 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1a5e3fb1
WN		 5063 }
5064
291ee02b
AN		 5065 if (!OPTS_VALID(opts, bpf_object_open_opts))
5066 return ERR_PTR(-EINVAL);
5067
1aace10f 5068 obj_name = OPTS_GET(opts, object_name, NULL);
291ee02b
AN		 5069 if (obj_buf) {
5070 if (!obj_name) {
5071 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
5072 (unsigned long)obj_buf,
5073 (unsigned long)obj_buf_sz);
5074 obj_name = tmp_name;
5075 }
5076 path = obj_name;
5077 pr_debug("loading object '%s' from buffer\n", obj_name);
5078 }
5079
2ce8450e 5080 obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
6371ca3b
WN		 5081 if (IS_ERR(obj))
5082 return obj;
1a5e3fb1 5083
8601fd42
AN		 5084 kconfig = OPTS_GET(opts, kconfig, NULL);
5085 if (kconfig) {
5086 obj->kconfig = strdup(kconfig);
5087 if (!obj->kconfig)
166750bc
AN		 5088 return ERR_PTR(-ENOMEM);
5089 }
291ee02b 5090
0d13bfce
AN		 5091 err = bpf_object__elf_init(obj);
5092 err = err ? : bpf_object__check_endianness(obj);
5093 err = err ? : bpf_object__elf_collect(obj);
166750bc
AN		 5094 err = err ? : bpf_object__collect_externs(obj);
5095 err = err ? : bpf_object__finalize_btf(obj);
0d13bfce
AN		 5096 err = err ? : bpf_object__init_maps(obj, opts);
5097 err = err ? : bpf_object__init_prog_names(obj);
5098 err = err ? : bpf_object__collect_reloc(obj);
5099 if (err)
5100 goto out;
1a5e3fb1 5101 bpf_object__elf_finish(obj);
dd4436bb
AN		 5102
5103 bpf_object__for_each_program(prog, obj) {
5104 enum bpf_prog_type prog_type;
5105 enum bpf_attach_type attach_type;
5106
590a0088
MKL		 5107 if (prog->type != BPF_PROG_TYPE_UNSPEC)
5108 continue;
5109
dd4436bb
AN		 5110 err = libbpf_prog_type_by_name(prog->section_name, &prog_type,
5111 &attach_type);
5112 if (err == -ESRCH)
5113 /* couldn't guess, but user might manually specify */
5114 continue;
5115 if (err)
5116 goto out;
5117
5118 bpf_program__set_type(prog, prog_type);
5119 bpf_program__set_expected_attach_type(prog, attach_type);
2db6eab1
AS		 5120 if (prog_type == BPF_PROG_TYPE_TRACING ||
5121 prog_type == BPF_PROG_TYPE_EXT)
166750bc 5122 prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
dd4436bb
AN		 5123 }
5124
1a5e3fb1
WN		 5125 return obj;
5126out:
5127 bpf_object__close(obj);
6371ca3b 5128 return ERR_PTR(err);
1a5e3fb1
WN		 5129}
5130
5e61f270
AN		 5131static struct bpf_object *
5132__bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
1a5e3fb1 5133{
e00aca65 5134 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
291ee02b
AN		 5135 .relaxed_maps = flags & MAPS_RELAX_COMPAT,
5136 );
5137
1a5e3fb1 5138 /* param validation */
07f2d4ea 5139 if (!attr->file)
1a5e3fb1
WN		 5140 return NULL;
5141
07f2d4ea 5142 pr_debug("loading %s\n", attr->file);
291ee02b 5143 return __bpf_object__open(attr->file, NULL, 0, &opts);
c034a177
JF		 5144}
5145
5146struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
5147{
5148 return __bpf_object__open_xattr(attr, 0);
07f2d4ea
JK		 5149}
5150
5151struct bpf_object *bpf_object__open(const char *path)
5152{
5153 struct bpf_object_open_attr attr = {
5154 .file = path,
5155 .prog_type = BPF_PROG_TYPE_UNSPEC,
5156 };
1a5e3fb1 5157
07f2d4ea 5158 return bpf_object__open_xattr(&attr);
6c956392
WN		 5159}
5160
2ce8450e 5161struct bpf_object *
01af3bf0 5162bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
2ce8450e 5163{
2ce8450e
AN		 5164 if (!path)
5165 return ERR_PTR(-EINVAL);
5166
5167 pr_debug("loading %s\n", path);
5168
291ee02b 5169 return __bpf_object__open(path, NULL, 0, opts);
2ce8450e
AN		 5170}
5171
5172struct bpf_object *
5173bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
01af3bf0 5174 const struct bpf_object_open_opts *opts)
6c956392 5175{
2ce8450e
AN		 5176 if (!obj_buf || obj_buf_sz == 0)
5177 return ERR_PTR(-EINVAL);
6c956392 5178
291ee02b 5179 return __bpf_object__open(NULL, obj_buf, obj_buf_sz, opts);
2ce8450e
AN		 5180}
5181
5182struct bpf_object *
5183bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
5184 const char *name)
5185{
e00aca65 5186 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
2ce8450e
AN		 5187 .object_name = name,
5188 /* wrong default, but backwards-compatible */
5189 .relaxed_maps = true,
5190 );
5191
5192 /* returning NULL is wrong, but backwards-compatible */
5193 if (!obj_buf || obj_buf_sz == 0)
5194 return NULL;
6c956392 5195
2ce8450e 5196 return bpf_object__open_mem(obj_buf, obj_buf_sz, &opts);
1a5e3fb1
WN		 5197}
5198
52d3352e
WN		 5199int bpf_object__unload(struct bpf_object *obj)
5200{
5201 size_t i;
5202
5203 if (!obj)
5204 return -EINVAL;
5205
590a0088 5206 for (i = 0; i < obj->nr_maps; i++) {
9d759a9b 5207 zclose(obj->maps[i].fd);
590a0088
MKL		 5208 if (obj->maps[i].st_ops)
5209 zfree(&obj->maps[i].st_ops->kern_vdata);
5210 }
52d3352e 5211
55cffde2
WN		 5212 for (i = 0; i < obj->nr_programs; i++)
5213 bpf_program__unload(&obj->programs[i]);
5214
52d3352e
WN		 5215 return 0;
5216}
5217
0d13bfce
AN		 5218static int bpf_object__sanitize_maps(struct bpf_object *obj)
5219{
5220 struct bpf_map *m;
5221
5222 bpf_object__for_each_map(m, obj) {
5223 if (!bpf_map__is_internal(m))
5224 continue;
5225 if (!obj->caps.global_data) {
5226 pr_warn("kernel doesn't support global data\n");
5227 return -ENOTSUP;
5228 }
5229 if (!obj->caps.array_mmap)
5230 m->def.map_flags ^= BPF_F_MMAPABLE;
5231 }
5232
5233 return 0;
5234}
5235
166750bc 5236static int bpf_object__resolve_externs(struct bpf_object *obj,
8601fd42 5237 const char *extra_kconfig)
166750bc
AN		 5238{
5239 bool need_config = false;
5240 struct extern_desc *ext;
5241 int err, i;
5242 void *data;
5243
5244 if (obj->nr_extern == 0)
5245 return 0;
5246
81bfdd08 5247 data = obj->maps[obj->kconfig_map_idx].mmaped;
166750bc
AN		 5248
5249 for (i = 0; i < obj->nr_extern; i++) {
5250 ext = &obj->externs[i];
5251
5252 if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
5253 void *ext_val = data + ext->data_off;
5254 __u32 kver = get_kernel_version();
5255
5256 if (!kver) {
5257 pr_warn("failed to get kernel version\n");
5258 return -EINVAL;
5259 }
5260 err = set_ext_value_num(ext, ext_val, kver);
5261 if (err)
5262 return err;
5263 pr_debug("extern %s=0x%x\n", ext->name, kver);
5264 } else if (strncmp(ext->name, "CONFIG_", 7) == 0) {
5265 need_config = true;
5266 } else {
5267 pr_warn("unrecognized extern '%s'\n", ext->name);
5268 return -EINVAL;
5269 }
5270 }
8601fd42
AN		 5271 if (need_config && extra_kconfig) {
5272 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, data);
5273 if (err)
5274 return -EINVAL;
5275 need_config = false;
5276 for (i = 0; i < obj->nr_extern; i++) {
5277 ext = &obj->externs[i];
5278 if (!ext->is_set) {
5279 need_config = true;
5280 break;
5281 }
5282 }
5283 }
166750bc 5284 if (need_config) {
8601fd42 5285 err = bpf_object__read_kconfig_file(obj, data);
166750bc
AN		 5286 if (err)
5287 return -EINVAL;
5288 }
5289 for (i = 0; i < obj->nr_extern; i++) {
5290 ext = &obj->externs[i];
5291
5292 if (!ext->is_set && !ext->is_weak) {
5293 pr_warn("extern %s (strong) not resolved\n", ext->name);
5294 return -ESRCH;
5295 } else if (!ext->is_set) {
5296 pr_debug("extern %s (weak) not resolved, defaulting to zero\n",
5297 ext->name);
5298 }
5299 }
5300
5301 return 0;
5302}
5303
60276f98 5304int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
52d3352e 5305{
60276f98 5306 struct bpf_object *obj;
ec6d5f47 5307 int err, i;
6371ca3b 5308
60276f98
QM		 5309 if (!attr)
5310 return -EINVAL;
5311 obj = attr->obj;
52d3352e
WN		 5312 if (!obj)
5313 return -EINVAL;
5314
5315 if (obj->loaded) {
be18010e 5316 pr_warn("object should not be loaded twice\n");
52d3352e
WN		 5317 return -EINVAL;
5318 }
5319
5320 obj->loaded = true;
6371ca3b 5321
0d13bfce 5322 err = bpf_object__probe_caps(obj);
8601fd42 5323 err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
0d13bfce
AN		 5324 err = err ? : bpf_object__sanitize_and_load_btf(obj);
5325 err = err ? : bpf_object__sanitize_maps(obj);
a6ed02ca 5326 err = err ? : bpf_object__load_vmlinux_btf(obj);
590a0088 5327 err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
0d13bfce
AN		 5328 err = err ? : bpf_object__create_maps(obj);
5329 err = err ? : bpf_object__relocate(obj, attr->target_btf_path);
5330 err = err ? : bpf_object__load_progs(obj, attr->log_level);
a6ed02ca
KS		 5331
5332 btf__free(obj->btf_vmlinux);
5333 obj->btf_vmlinux = NULL;
5334
0d13bfce
AN		 5335 if (err)
5336 goto out;
52d3352e
WN		 5337
5338 return 0;
5339out:
ec6d5f47
THJ		 5340 /* unpin any maps that were auto-pinned during load */
5341 for (i = 0; i < obj->nr_maps; i++)
5342 if (obj->maps[i].pinned && !obj->maps[i].reused)
5343 bpf_map__unpin(&obj->maps[i], NULL);
5344
52d3352e 5345 bpf_object__unload(obj);
be18010e 5346 pr_warn("failed to load object '%s'\n", obj->path);
6371ca3b 5347 return err;
52d3352e
WN		 5348}
5349
60276f98
QM		 5350int bpf_object__load(struct bpf_object *obj)
5351{
5352 struct bpf_object_load_attr attr = {
5353 .obj = obj,
5354 };
5355
5356 return bpf_object__load_xattr(&attr);
5357}
5358
196f8487
THJ		 5359static int make_parent_dir(const char *path)
5360{
5361 char *cp, errmsg[STRERR_BUFSIZE];
5362 char *dname, *dir;
5363 int err = 0;
5364
5365 dname = strdup(path);
5366 if (dname == NULL)
5367 return -ENOMEM;
5368
5369 dir = dirname(dname);
5370 if (mkdir(dir, 0700) && errno != EEXIST)
5371 err = -errno;
5372
5373 free(dname);
5374 if (err) {
5375 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
5376 pr_warn("failed to mkdir %s: %s\n", path, cp);
5377 }
5378 return err;
5379}
5380
f367540c
JS		 5381static int check_path(const char *path)
5382{
1ce6a9fc 5383 char *cp, errmsg[STRERR_BUFSIZE];
f367540c
JS		 5384 struct statfs st_fs;
5385 char *dname, *dir;
5386 int err = 0;
5387
5388 if (path == NULL)
5389 return -EINVAL;
5390
5391 dname = strdup(path);
5392 if (dname == NULL)
5393 return -ENOMEM;
5394
5395 dir = dirname(dname);
5396 if (statfs(dir, &st_fs)) {
24d6a808 5397 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
be18010e 5398 pr_warn("failed to statfs %s: %s\n", dir, cp);
f367540c
JS		 5399 err = -errno;
5400 }
5401 free(dname);
5402
5403 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
be18010e 5404 pr_warn("specified path %s is not on BPF FS\n", path);
f367540c
JS		 5405 err = -EINVAL;
5406 }
5407
5408 return err;
5409}
5410
5411int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
5412 int instance)
5413{
1ce6a9fc 5414 char *cp, errmsg[STRERR_BUFSIZE];
f367540c
JS		 5415 int err;
5416
196f8487
THJ		 5417 err = make_parent_dir(path);
5418 if (err)
5419 return err;
5420
f367540c
JS		 5421 err = check_path(path);
5422 if (err)
5423 return err;
5424
5425 if (prog == NULL) {
be18010e 5426 pr_warn("invalid program pointer\n");
f367540c
JS		 5427 return -EINVAL;
5428 }
5429
5430 if (instance < 0 || instance >= prog->instances.nr) {
be18010e
KW		 5431 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
5432 instance, prog->section_name, prog->instances.nr);
f367540c
JS		 5433 return -EINVAL;
5434 }
5435
5436 if (bpf_obj_pin(prog->instances.fds[instance], path)) {
24d6a808 5437 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
be18010e 5438 pr_warn("failed to pin program: %s\n", cp);
f367540c
JS		 5439 return -errno;
5440 }
5441 pr_debug("pinned program '%s'\n", path);
5442
5443 return 0;
5444}
5445
0c19a9fb
SF		 5446int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
5447 int instance)
5448{
5449 int err;
5450
5451 err = check_path(path);
5452 if (err)
5453 return err;
5454
5455 if (prog == NULL) {
be18010e 5456 pr_warn("invalid program pointer\n");
0c19a9fb
SF		 5457 return -EINVAL;
5458 }
5459
5460 if (instance < 0 || instance >= prog->instances.nr) {
be18010e
KW		 5461 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
5462 instance, prog->section_name, prog->instances.nr);
0c19a9fb
SF		 5463 return -EINVAL;
5464 }
5465
5466 err = unlink(path);
5467 if (err != 0)
5468 return -errno;
5469 pr_debug("unpinned program '%s'\n", path);
5470
5471 return 0;
5472}
5473
f367540c
JS		 5474int bpf_program__pin(struct bpf_program *prog, const char *path)
5475{
5476 int i, err;
5477
196f8487
THJ		 5478 err = make_parent_dir(path);
5479 if (err)
5480 return err;
5481
f367540c
JS		 5482 err = check_path(path);
5483 if (err)
5484 return err;
5485
5486 if (prog == NULL) {
be18010e 5487 pr_warn("invalid program pointer\n");
f367540c
JS		 5488 return -EINVAL;
5489 }
5490
5491 if (prog->instances.nr <= 0) {
be18010e 5492 pr_warn("no instances of prog %s to pin\n",
f367540c
JS		 5493 prog->section_name);
5494 return -EINVAL;
5495 }
5496
fd734c5c
SF		 5497 if (prog->instances.nr == 1) {
5498 /* don't create subdirs when pinning single instance */
5499 return bpf_program__pin_instance(prog, path, 0);
5500 }
5501
0c19a9fb
SF		 5502 for (i = 0; i < prog->instances.nr; i++) {
5503 char buf[PATH_MAX];
5504 int len;
5505
5506 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
5507 if (len < 0) {
5508 err = -EINVAL;
5509 goto err_unpin;
5510 } else if (len >= PATH_MAX) {
5511 err = -ENAMETOOLONG;
5512 goto err_unpin;
5513 }
5514
5515 err = bpf_program__pin_instance(prog, buf, i);
5516 if (err)
5517 goto err_unpin;
5518 }
5519
5520 return 0;
5521
5522err_unpin:
5523 for (i = i - 1; i >= 0; i--) {
5524 char buf[PATH_MAX];
5525 int len;
5526
5527 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
5528 if (len < 0)
5529 continue;
5530 else if (len >= PATH_MAX)
5531 continue;
5532
5533 bpf_program__unpin_instance(prog, buf, i);
5534 }
5535
5536 rmdir(path);
5537
5538 return err;
5539}
5540
5541int bpf_program__unpin(struct bpf_program *prog, const char *path)
5542{
5543 int i, err;
5544
5545 err = check_path(path);
5546 if (err)
5547 return err;
5548
5549 if (prog == NULL) {
be18010e 5550 pr_warn("invalid program pointer\n");
0c19a9fb
SF		 5551 return -EINVAL;
5552 }
5553
5554 if (prog->instances.nr <= 0) {
be18010e 5555 pr_warn("no instances of prog %s to pin\n",
0c19a9fb
SF		 5556 prog->section_name);
5557 return -EINVAL;
fd734c5c
SF		 5558 }
5559
5560 if (prog->instances.nr == 1) {
5561 /* don't create subdirs when pinning single instance */
5562 return bpf_program__unpin_instance(prog, path, 0);
0c19a9fb
SF		 5563 }
5564
f367540c
JS		 5565 for (i = 0; i < prog->instances.nr; i++) {
5566 char buf[PATH_MAX];
5567 int len;
5568
5569 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
5570 if (len < 0)
5571 return -EINVAL;
5572 else if (len >= PATH_MAX)
5573 return -ENAMETOOLONG;
5574
0c19a9fb 5575 err = bpf_program__unpin_instance(prog, buf, i);
f367540c
JS		 5576 if (err)
5577 return err;
5578 }
5579
0c19a9fb
SF		 5580 err = rmdir(path);
5581 if (err)
5582 return -errno;
5583
f367540c
JS		 5584 return 0;
5585}
5586
b6989f35
JS		 5587int bpf_map__pin(struct bpf_map *map, const char *path)
5588{
1ce6a9fc 5589 char *cp, errmsg[STRERR_BUFSIZE];
b6989f35
JS		 5590 int err;
5591
b6989f35 5592 if (map == NULL) {
be18010e 5593 pr_warn("invalid map pointer\n");
b6989f35
JS		 5594 return -EINVAL;
5595 }
5596
4580b25f
THJ		 5597 if (map->pin_path) {
5598 if (path && strcmp(path, map->pin_path)) {
5599 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
5600 bpf_map__name(map), map->pin_path, path);
5601 return -EINVAL;
5602 } else if (map->pinned) {
5603 pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
5604 bpf_map__name(map), map->pin_path);
5605 return 0;
5606 }
5607 } else {
5608 if (!path) {
5609 pr_warn("missing a path to pin map '%s' at\n",
5610 bpf_map__name(map));
5611 return -EINVAL;
5612 } else if (map->pinned) {
5613 pr_warn("map '%s' already pinned\n", bpf_map__name(map));
5614 return -EEXIST;
5615 }
5616
5617 map->pin_path = strdup(path);
5618 if (!map->pin_path) {
5619 err = -errno;
5620 goto out_err;
5621 }
b6989f35
JS		 5622 }
5623
196f8487
THJ		 5624 err = make_parent_dir(map->pin_path);
5625 if (err)
5626 return err;
5627
4580b25f
THJ		 5628 err = check_path(map->pin_path);
5629 if (err)
5630 return err;
5631
5632 if (bpf_obj_pin(map->fd, map->pin_path)) {
5633 err = -errno;
5634 goto out_err;
5635 }
5636
5637 map->pinned = true;
5638 pr_debug("pinned map '%s'\n", map->pin_path);
0c19a9fb 5639
b6989f35 5640 return 0;
4580b25f
THJ		 5641
5642out_err:
5643 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
5644 pr_warn("failed to pin map: %s\n", cp);
5645 return err;
b6989f35
JS		 5646}
5647
0c19a9fb
SF		 5648int bpf_map__unpin(struct bpf_map *map, const char *path)
5649{
5650 int err;
5651
0c19a9fb 5652 if (map == NULL) {
be18010e 5653 pr_warn("invalid map pointer\n");
0c19a9fb
SF		 5654 return -EINVAL;
5655 }
5656
4580b25f
THJ		 5657 if (map->pin_path) {
5658 if (path && strcmp(path, map->pin_path)) {
5659 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
5660 bpf_map__name(map), map->pin_path, path);
5661 return -EINVAL;
5662 }
5663 path = map->pin_path;
5664 } else if (!path) {
5665 pr_warn("no path to unpin map '%s' from\n",
5666 bpf_map__name(map));
5667 return -EINVAL;
5668 }
5669
5670 err = check_path(path);
5671 if (err)
5672 return err;
5673
0c19a9fb
SF		 5674 err = unlink(path);
5675 if (err != 0)
5676 return -errno;
4580b25f
THJ		 5677
5678 map->pinned = false;
5679 pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
0c19a9fb
SF		 5680
5681 return 0;
5682}
5683
4580b25f
THJ		 5684int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
5685{
5686 char *new = NULL;
5687
5688 if (path) {
5689 new = strdup(path);
5690 if (!new)
5691 return -errno;
5692 }
5693
5694 free(map->pin_path);
5695 map->pin_path = new;
5696 return 0;
5697}
5698
5699const char *bpf_map__get_pin_path(const struct bpf_map *map)
5700{
5701 return map->pin_path;
5702}
5703
5704bool bpf_map__is_pinned(const struct bpf_map *map)
5705{
5706 return map->pinned;
5707}
5708
0c19a9fb 5709int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
d5148d85 5710{
d5148d85
JS		 5711 struct bpf_map *map;
5712 int err;
5713
5714 if (!obj)
5715 return -ENOENT;
5716
5717 if (!obj->loaded) {
be18010e 5718 pr_warn("object not yet loaded; load it first\n");
d5148d85
JS		 5719 return -ENOENT;
5720 }
5721
f74a53d9 5722 bpf_object__for_each_map(map, obj) {
4580b25f 5723 char *pin_path = NULL;
0c19a9fb 5724 char buf[PATH_MAX];
0c19a9fb 5725
4580b25f
THJ		 5726 if (path) {
5727 int len;
5728
5729 len = snprintf(buf, PATH_MAX, "%s/%s", path,
5730 bpf_map__name(map));
5731 if (len < 0) {
5732 err = -EINVAL;
5733 goto err_unpin_maps;
5734 } else if (len >= PATH_MAX) {
5735 err = -ENAMETOOLONG;
5736 goto err_unpin_maps;
5737 }
5738 pin_path = buf;
5739 } else if (!map->pin_path) {
5740 continue;
0c19a9fb
SF		 5741 }
5742
4580b25f 5743 err = bpf_map__pin(map, pin_path);
0c19a9fb
SF		 5744 if (err)
5745 goto err_unpin_maps;
5746 }
5747
5748 return 0;
5749
5750err_unpin_maps:
5751 while ((map = bpf_map__prev(map, obj))) {
4580b25f 5752 if (!map->pin_path)
0c19a9fb
SF		 5753 continue;
5754
4580b25f 5755 bpf_map__unpin(map, NULL);
0c19a9fb
SF		 5756 }
5757
5758 return err;
5759}
5760
5761int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
5762{
5763 struct bpf_map *map;
5764 int err;
5765
5766 if (!obj)
5767 return -ENOENT;
5768
f74a53d9 5769 bpf_object__for_each_map(map, obj) {
4580b25f 5770 char *pin_path = NULL;
d5148d85 5771 char buf[PATH_MAX];
d5148d85 5772
4580b25f
THJ		 5773 if (path) {
5774 int len;
5775
5776 len = snprintf(buf, PATH_MAX, "%s/%s", path,
5777 bpf_map__name(map));
5778 if (len < 0)
5779 return -EINVAL;
5780 else if (len >= PATH_MAX)
5781 return -ENAMETOOLONG;
5782 pin_path = buf;
5783 } else if (!map->pin_path) {
5784 continue;
5785 }
d5148d85 5786
4580b25f 5787 err = bpf_map__unpin(map, pin_path);
d5148d85
JS		 5788 if (err)
5789 return err;
5790 }
5791
0c19a9fb
SF		 5792 return 0;
5793}
5794
5795int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
5796{
5797 struct bpf_program *prog;
5798 int err;
5799
5800 if (!obj)
5801 return -ENOENT;
5802
5803 if (!obj->loaded) {
be18010e 5804 pr_warn("object not yet loaded; load it first\n");
0c19a9fb
SF		 5805 return -ENOENT;
5806 }
5807
0c19a9fb
SF		 5808 bpf_object__for_each_program(prog, obj) {
5809 char buf[PATH_MAX];
5810 int len;
5811
5812 len = snprintf(buf, PATH_MAX, "%s/%s", path,
33a2c75c 5813 prog->pin_name);
0c19a9fb
SF		 5814 if (len < 0) {
5815 err = -EINVAL;
5816 goto err_unpin_programs;
5817 } else if (len >= PATH_MAX) {
5818 err = -ENAMETOOLONG;
5819 goto err_unpin_programs;
5820 }
5821
5822 err = bpf_program__pin(prog, buf);
5823 if (err)
5824 goto err_unpin_programs;
5825 }
5826
5827 return 0;
5828
5829err_unpin_programs:
5830 while ((prog = bpf_program__prev(prog, obj))) {
5831 char buf[PATH_MAX];
5832 int len;
5833
5834 len = snprintf(buf, PATH_MAX, "%s/%s", path,
33a2c75c 5835 prog->pin_name);
0c19a9fb
SF		 5836 if (len < 0)
5837 continue;
5838 else if (len >= PATH_MAX)
5839 continue;
5840
5841 bpf_program__unpin(prog, buf);
5842 }
5843
5844 return err;
5845}
5846
5847int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
5848{
5849 struct bpf_program *prog;
5850 int err;
5851
5852 if (!obj)
5853 return -ENOENT;
5854
d5148d85
JS		 5855 bpf_object__for_each_program(prog, obj) {
5856 char buf[PATH_MAX];
5857 int len;
5858
5859 len = snprintf(buf, PATH_MAX, "%s/%s", path,
33a2c75c 5860 prog->pin_name);
d5148d85
JS		 5861 if (len < 0)
5862 return -EINVAL;
5863 else if (len >= PATH_MAX)
5864 return -ENAMETOOLONG;
5865
0c19a9fb 5866 err = bpf_program__unpin(prog, buf);
d5148d85
JS		 5867 if (err)
5868 return err;
5869 }
5870
5871 return 0;
5872}
5873
0c19a9fb
SF		 5874int bpf_object__pin(struct bpf_object *obj, const char *path)
5875{
5876 int err;
5877
5878 err = bpf_object__pin_maps(obj, path);
5879 if (err)
5880 return err;
5881
5882 err = bpf_object__pin_programs(obj, path);
5883 if (err) {
5884 bpf_object__unpin_maps(obj, path);
5885 return err;
5886 }
5887
5888 return 0;
5889}
5890
1a5e3fb1
WN		 5891void bpf_object__close(struct bpf_object *obj)
5892{
a5b8bd47
WN		 5893 size_t i;
5894
1a5e3fb1
WN		 5895 if (!obj)
5896 return;
5897
10931d24
WN		 5898 if (obj->clear_priv)
5899 obj->clear_priv(obj, obj->priv);
5900
1a5e3fb1 5901 bpf_object__elf_finish(obj);
52d3352e 5902 bpf_object__unload(obj);
8a138aed 5903 btf__free(obj->btf);
2993e051 5904 btf_ext__free(obj->btf_ext);
1a5e3fb1 5905
9d759a9b 5906 for (i = 0; i < obj->nr_maps; i++) {
eba9c5f4
AN		 5907 struct bpf_map *map = &obj->maps[i];
5908
5909 if (map->clear_priv)
5910 map->clear_priv(map, map->priv);
5911 map->priv = NULL;
5912 map->clear_priv = NULL;
5913
5914 if (map->mmaped) {
5915 munmap(map->mmaped, bpf_map_mmap_sz(map));
5916 map->mmaped = NULL;
5917 }
5918
590a0088
MKL		 5919 if (map->st_ops) {
5920 zfree(&map->st_ops->data);
5921 zfree(&map->st_ops->progs);
5922 zfree(&map->st_ops->kern_func_off);
5923 zfree(&map->st_ops);
5924 }
5925
eba9c5f4
AN		 5926 zfree(&map->name);
5927 zfree(&map->pin_path);
9d759a9b 5928 }
d859900c 5929
8601fd42 5930 zfree(&obj->kconfig);
166750bc
AN		 5931 zfree(&obj->externs);
5932 obj->nr_extern = 0;
5933
9d759a9b
WN		 5934 zfree(&obj->maps);
5935 obj->nr_maps = 0;
a5b8bd47
WN		 5936
5937 if (obj->programs && obj->nr_programs) {
5938 for (i = 0; i < obj->nr_programs; i++)
5939 bpf_program__exit(&obj->programs[i]);
5940 }
5941 zfree(&obj->programs);
5942
9a208eff 5943 list_del(&obj->list);
1a5e3fb1
WN		 5944 free(obj);
5945}
aa9b1ac3 5946
9a208eff
WN		 5947struct bpf_object *
5948bpf_object__next(struct bpf_object *prev)
5949{
5950 struct bpf_object *next;
5951
5952 if (!prev)
5953 next = list_first_entry(&bpf_objects_list,
5954 struct bpf_object,
5955 list);
5956 else
5957 next = list_next_entry(prev, list);
5958
5959 /* Empty list is noticed here so don't need checking on entry. */
5960 if (&next->list == &bpf_objects_list)
5961 return NULL;
5962
5963 return next;
5964}
5965
a324aae3 5966const char *bpf_object__name(const struct bpf_object *obj)
acf860ae 5967{
c9e4c301 5968 return obj ? obj->name : ERR_PTR(-EINVAL);
acf860ae
WN		 5969}
5970
a324aae3 5971unsigned int bpf_object__kversion(const struct bpf_object *obj)
45825d8a 5972{
a7fe0450 5973 return obj ? obj->kern_version : 0;
45825d8a
WN		 5974}
5975
a324aae3 5976struct btf *bpf_object__btf(const struct bpf_object *obj)
789f6bab
AI		 5977{
5978 return obj ? obj->btf : NULL;
5979}
5980
8a138aed
MKL		 5981int bpf_object__btf_fd(const struct bpf_object *obj)
5982{
5983 return obj->btf ? btf__fd(obj->btf) : -1;
5984}
5985
10931d24
WN		 5986int bpf_object__set_priv(struct bpf_object *obj, void *priv,
5987 bpf_object_clear_priv_t clear_priv)
5988{
5989 if (obj->priv && obj->clear_priv)
5990 obj->clear_priv(obj, obj->priv);
5991
5992 obj->priv = priv;
5993 obj->clear_priv = clear_priv;
5994 return 0;
5995}
5996
a324aae3 5997void *bpf_object__priv(const struct bpf_object *obj)
10931d24
WN		 5998{
5999 return obj ? obj->priv : ERR_PTR(-EINVAL);
6000}
6001
eac7d845 6002static struct bpf_program *
a324aae3
AN		 6003__bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
6004 bool forward)
aa9b1ac3 6005{
a83d6e76 6006 size_t nr_programs = obj->nr_programs;
0c19a9fb 6007 ssize_t idx;
aa9b1ac3 6008
a83d6e76 6009 if (!nr_programs)
aa9b1ac3 6010 return NULL;
aa9b1ac3 6011
a83d6e76
MKL		 6012 if (!p)
6013 /* Iter from the beginning */
6014 return forward ? &obj->programs[0] :
6015 &obj->programs[nr_programs - 1];
6016
0c19a9fb 6017 if (p->obj != obj) {
be18010e 6018 pr_warn("error: program handler doesn't match object\n");
aa9b1ac3
WN		 6019 return NULL;
6020 }
6021
a83d6e76 6022 idx = (p - obj->programs) + (forward ? 1 : -1);
0c19a9fb 6023 if (idx >= obj->nr_programs || idx < 0)
aa9b1ac3
WN		 6024 return NULL;
6025 return &obj->programs[idx];
6026}
6027
eac7d845 6028struct bpf_program *
a324aae3 6029bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
eac7d845
JK		 6030{
6031 struct bpf_program *prog = prev;
6032
6033 do {
a83d6e76 6034 prog = __bpf_program__iter(prog, obj, true);
0c19a9fb
SF		 6035 } while (prog && bpf_program__is_function_storage(prog, obj));
6036
6037 return prog;
6038}
6039
6040struct bpf_program *
a324aae3 6041bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
0c19a9fb
SF		 6042{
6043 struct bpf_program *prog = next;
6044
0c19a9fb 6045 do {
a83d6e76 6046 prog = __bpf_program__iter(prog, obj, false);
eac7d845
JK		 6047 } while (prog && bpf_program__is_function_storage(prog, obj));
6048
6049 return prog;
6050}
6051
edb13ed4
ACM		 6052int bpf_program__set_priv(struct bpf_program *prog, void *priv,
6053 bpf_program_clear_priv_t clear_priv)
aa9b1ac3
WN		 6054{
6055 if (prog->priv && prog->clear_priv)
6056 prog->clear_priv(prog, prog->priv);
6057
6058 prog->priv = priv;
6059 prog->clear_priv = clear_priv;
6060 return 0;
6061}
6062
a324aae3 6063void *bpf_program__priv(const struct bpf_program *prog)
aa9b1ac3 6064{
be834ffb 6065 return prog ? prog->priv : ERR_PTR(-EINVAL);
aa9b1ac3
WN		 6066}
6067
9aba3613
JK		 6068void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
6069{
6070 prog->prog_ifindex = ifindex;
6071}
6072
01af3bf0
AN		 6073const char *bpf_program__name(const struct bpf_program *prog)
6074{
6075 return prog->name;
6076}
6077
a324aae3 6078const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
aa9b1ac3
WN		 6079{
6080 const char *title;
6081
6082 title = prog->section_name;
715f8db9 6083 if (needs_copy) {
aa9b1ac3
WN		 6084 title = strdup(title);
6085 if (!title) {
be18010e 6086 pr_warn("failed to strdup program title\n");
6371ca3b 6087 return ERR_PTR(-ENOMEM);
aa9b1ac3
WN		 6088 }
6089 }
6090
6091 return title;
6092}
6093
a324aae3 6094int bpf_program__fd(const struct bpf_program *prog)
aa9b1ac3 6095{
b580563e
WN		 6096 return bpf_program__nth_fd(prog, 0);
6097}
6098
1a734efe
THJ		 6099size_t bpf_program__size(const struct bpf_program *prog)
6100{
6101 return prog->insns_cnt * sizeof(struct bpf_insn);
6102}
6103
b580563e
WN		 6104int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
6105 bpf_program_prep_t prep)
6106{
6107 int *instances_fds;
6108
6109 if (nr_instances <= 0 || !prep)
6110 return -EINVAL;
6111
6112 if (prog->instances.nr > 0 || prog->instances.fds) {
be18010e 6113 pr_warn("Can't set pre-processor after loading\n");
b580563e
WN		 6114 return -EINVAL;
6115 }
6116
6117 instances_fds = malloc(sizeof(int) * nr_instances);
6118 if (!instances_fds) {
be18010e 6119 pr_warn("alloc memory failed for fds\n");
b580563e
WN		 6120 return -ENOMEM;
6121 }
6122
6123 /* fill all fd with -1 */
6124 memset(instances_fds, -1, sizeof(int) * nr_instances);
6125
6126 prog->instances.nr = nr_instances;
6127 prog->instances.fds = instances_fds;
6128 prog->preprocessor = prep;
6129 return 0;
6130}
6131
a324aae3 6132int bpf_program__nth_fd(const struct bpf_program *prog, int n)
b580563e
WN		 6133{
6134 int fd;
6135
1e960043
JK		 6136 if (!prog)
6137 return -EINVAL;
6138
b580563e 6139 if (n >= prog->instances.nr || n < 0) {
be18010e
KW		 6140 pr_warn("Can't get the %dth fd from program %s: only %d instances\n",
6141 n, prog->section_name, prog->instances.nr);
b580563e
WN		 6142 return -EINVAL;
6143 }
6144
6145 fd = prog->instances.fds[n];
6146 if (fd < 0) {
be18010e
KW		 6147 pr_warn("%dth instance of program '%s' is invalid\n",
6148 n, prog->section_name);
b580563e
WN		 6149 return -ENOENT;
6150 }
6151
6152 return fd;
aa9b1ac3 6153}
9d759a9b 6154
f1eead9e
AN		 6155enum bpf_prog_type bpf_program__get_type(struct bpf_program *prog)
6156{
6157 return prog->type;
6158}
6159
dd26b7f5 6160void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
5f44e4c8
WN		 6161{
6162 prog->type = type;
6163}
6164
a324aae3 6165static bool bpf_program__is_type(const struct bpf_program *prog,
5f44e4c8
WN		 6166 enum bpf_prog_type type)
6167{
6168 return prog ? (prog->type == type) : false;
6169}
6170
a324aae3
AN		 6171#define BPF_PROG_TYPE_FNS(NAME, TYPE) \
6172int bpf_program__set_##NAME(struct bpf_program *prog) \
6173{ \
6174 if (!prog) \
6175 return -EINVAL; \
6176 bpf_program__set_type(prog, TYPE); \
6177 return 0; \
6178} \
6179 \
6180bool bpf_program__is_##NAME(const struct bpf_program *prog) \
6181{ \
6182 return bpf_program__is_type(prog, TYPE); \
6183} \
ed794073 6184
7803ba73 6185BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
ed794073 6186BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
7803ba73
JS		 6187BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
6188BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
ed794073 6189BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
e14c93fd 6190BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
7803ba73
JS		 6191BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
6192BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
12a8654b 6193BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
590a0088 6194BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS);
2db6eab1 6195BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT);
5f44e4c8 6196
f1eead9e
AN		 6197enum bpf_attach_type
6198bpf_program__get_expected_attach_type(struct bpf_program *prog)
6199{
6200 return prog->expected_attach_type;
6201}
6202
16962b24
JF		 6203void bpf_program__set_expected_attach_type(struct bpf_program *prog,
6204 enum bpf_attach_type type)
d7be143b
AI		 6205{
6206 prog->expected_attach_type = type;
6207}
6208
f75a697e
AS		 6209#define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, btf, atype) \
6210 { string, sizeof(string) - 1, ptype, eatype, is_attachable, btf, atype }
d7be143b 6211
956b620f 6212/* Programs that can NOT be attached. */
f75a697e 6213#define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0, 0)
d7be143b 6214
956b620f
AI		 6215/* Programs that can be attached. */
6216#define BPF_APROG_SEC(string, ptype, atype) \
f75a697e 6217 BPF_PROG_SEC_IMPL(string, ptype, 0, 1, 0, atype)
81efee75 6218
956b620f
AI		 6219/* Programs that must specify expected attach type at load time. */
6220#define BPF_EAPROG_SEC(string, ptype, eatype) \
f75a697e
AS		 6221 BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, 0, eatype)
6222
6223/* Programs that use BTF to identify attach point */
12a8654b
AS		 6224#define BPF_PROG_BTF(string, ptype, eatype) \
6225 BPF_PROG_SEC_IMPL(string, ptype, eatype, 0, 1, 0)
956b620f
AI		 6226
6227/* Programs that can be attached but attach type can't be identified by section
6228 * name. Kept for backward compatibility.
6229 */
6230#define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
e50b0a6f 6231
d7a18ea7
AN		 6232#define SEC_DEF(sec_pfx, ptype, ...) { \
6233 .sec = sec_pfx, \
6234 .len = sizeof(sec_pfx) - 1, \
6235 .prog_type = BPF_PROG_TYPE_##ptype, \
6236 __VA_ARGS__ \
6237}
6238
6239struct bpf_sec_def;
6240
6241typedef struct bpf_link *(*attach_fn_t)(const struct bpf_sec_def *sec,
6242 struct bpf_program *prog);
6243
6244static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
6245 struct bpf_program *prog);
6246static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
6247 struct bpf_program *prog);
6248static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
6249 struct bpf_program *prog);
6250static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
6251 struct bpf_program *prog);
6252
6253struct bpf_sec_def {
583c9009
RG		 6254 const char *sec;
6255 size_t len;
6256 enum bpf_prog_type prog_type;
d7be143b 6257 enum bpf_attach_type expected_attach_type;
f75a697e
AS		 6258 bool is_attachable;
6259 bool is_attach_btf;
956b620f 6260 enum bpf_attach_type attach_type;
d7a18ea7
AN		 6261 attach_fn_t attach_fn;
6262};
6263
6264static const struct bpf_sec_def section_defs[] = {
956b620f 6265 BPF_PROG_SEC("socket", BPF_PROG_TYPE_SOCKET_FILTER),
67d69ccd 6266 BPF_PROG_SEC("sk_reuseport", BPF_PROG_TYPE_SK_REUSEPORT),
d7a18ea7
AN		 6267 SEC_DEF("kprobe/", KPROBE,
6268 .attach_fn = attach_kprobe),
32dff6db 6269 BPF_PROG_SEC("uprobe/", BPF_PROG_TYPE_KPROBE),
d7a18ea7
AN		 6270 SEC_DEF("kretprobe/", KPROBE,
6271 .attach_fn = attach_kprobe),
32dff6db 6272 BPF_PROG_SEC("uretprobe/", BPF_PROG_TYPE_KPROBE),
956b620f
AI		 6273 BPF_PROG_SEC("classifier", BPF_PROG_TYPE_SCHED_CLS),
6274 BPF_PROG_SEC("action", BPF_PROG_TYPE_SCHED_ACT),
d7a18ea7
AN		 6275 SEC_DEF("tracepoint/", TRACEPOINT,
6276 .attach_fn = attach_tp),
6277 SEC_DEF("tp/", TRACEPOINT,
6278 .attach_fn = attach_tp),
6279 SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT,
6280 .attach_fn = attach_raw_tp),
6281 SEC_DEF("raw_tp/", RAW_TRACEPOINT,
6282 .attach_fn = attach_raw_tp),
6283 SEC_DEF("tp_btf/", TRACING,
6284 .expected_attach_type = BPF_TRACE_RAW_TP,
6285 .is_attach_btf = true,
6286 .attach_fn = attach_trace),
6287 SEC_DEF("fentry/", TRACING,
6288 .expected_attach_type = BPF_TRACE_FENTRY,
6289 .is_attach_btf = true,
6290 .attach_fn = attach_trace),
6291 SEC_DEF("fexit/", TRACING,
6292 .expected_attach_type = BPF_TRACE_FEXIT,
6293 .is_attach_btf = true,
6294 .attach_fn = attach_trace),
2db6eab1
AS		 6295 SEC_DEF("freplace/", EXT,
6296 .is_attach_btf = true,
6297 .attach_fn = attach_trace),
956b620f
AI		 6298 BPF_PROG_SEC("xdp", BPF_PROG_TYPE_XDP),
6299 BPF_PROG_SEC("perf_event", BPF_PROG_TYPE_PERF_EVENT),
6300 BPF_PROG_SEC("lwt_in", BPF_PROG_TYPE_LWT_IN),
6301 BPF_PROG_SEC("lwt_out", BPF_PROG_TYPE_LWT_OUT),
6302 BPF_PROG_SEC("lwt_xmit", BPF_PROG_TYPE_LWT_XMIT),
6303 BPF_PROG_SEC("lwt_seg6local", BPF_PROG_TYPE_LWT_SEG6LOCAL),
bafa7afe
AI		 6304 BPF_APROG_SEC("cgroup_skb/ingress", BPF_PROG_TYPE_CGROUP_SKB,
6305 BPF_CGROUP_INET_INGRESS),
6306 BPF_APROG_SEC("cgroup_skb/egress", BPF_PROG_TYPE_CGROUP_SKB,
6307 BPF_CGROUP_INET_EGRESS),
956b620f
AI		 6308 BPF_APROG_COMPAT("cgroup/skb", BPF_PROG_TYPE_CGROUP_SKB),
6309 BPF_APROG_SEC("cgroup/sock", BPF_PROG_TYPE_CGROUP_SOCK,
6310 BPF_CGROUP_INET_SOCK_CREATE),
6311 BPF_EAPROG_SEC("cgroup/post_bind4", BPF_PROG_TYPE_CGROUP_SOCK,
6312 BPF_CGROUP_INET4_POST_BIND),
6313 BPF_EAPROG_SEC("cgroup/post_bind6", BPF_PROG_TYPE_CGROUP_SOCK,
6314 BPF_CGROUP_INET6_POST_BIND),
6315 BPF_APROG_SEC("cgroup/dev", BPF_PROG_TYPE_CGROUP_DEVICE,
6316 BPF_CGROUP_DEVICE),
6317 BPF_APROG_SEC("sockops", BPF_PROG_TYPE_SOCK_OPS,
6318 BPF_CGROUP_SOCK_OPS),
c6f6851b
AI		 6319 BPF_APROG_SEC("sk_skb/stream_parser", BPF_PROG_TYPE_SK_SKB,
6320 BPF_SK_SKB_STREAM_PARSER),
6321 BPF_APROG_SEC("sk_skb/stream_verdict", BPF_PROG_TYPE_SK_SKB,
6322 BPF_SK_SKB_STREAM_VERDICT),
956b620f
AI		 6323 BPF_APROG_COMPAT("sk_skb", BPF_PROG_TYPE_SK_SKB),
6324 BPF_APROG_SEC("sk_msg", BPF_PROG_TYPE_SK_MSG,
6325 BPF_SK_MSG_VERDICT),
6326 BPF_APROG_SEC("lirc_mode2", BPF_PROG_TYPE_LIRC_MODE2,
6327 BPF_LIRC_MODE2),
6328 BPF_APROG_SEC("flow_dissector", BPF_PROG_TYPE_FLOW_DISSECTOR,
6329 BPF_FLOW_DISSECTOR),
6330 BPF_EAPROG_SEC("cgroup/bind4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6331 BPF_CGROUP_INET4_BIND),
6332 BPF_EAPROG_SEC("cgroup/bind6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6333 BPF_CGROUP_INET6_BIND),
6334 BPF_EAPROG_SEC("cgroup/connect4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6335 BPF_CGROUP_INET4_CONNECT),
6336 BPF_EAPROG_SEC("cgroup/connect6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6337 BPF_CGROUP_INET6_CONNECT),
6338 BPF_EAPROG_SEC("cgroup/sendmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6339 BPF_CGROUP_UDP4_SENDMSG),
6340 BPF_EAPROG_SEC("cgroup/sendmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6341 BPF_CGROUP_UDP6_SENDMSG),
9bb59ac1
DB		 6342 BPF_EAPROG_SEC("cgroup/recvmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6343 BPF_CGROUP_UDP4_RECVMSG),
6344 BPF_EAPROG_SEC("cgroup/recvmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6345 BPF_CGROUP_UDP6_RECVMSG),
063cc9f0
AI		 6346 BPF_EAPROG_SEC("cgroup/sysctl", BPF_PROG_TYPE_CGROUP_SYSCTL,
6347 BPF_CGROUP_SYSCTL),
4cdbfb59
SF		 6348 BPF_EAPROG_SEC("cgroup/getsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT,
6349 BPF_CGROUP_GETSOCKOPT),
6350 BPF_EAPROG_SEC("cgroup/setsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT,
6351 BPF_CGROUP_SETSOCKOPT),
590a0088 6352 BPF_PROG_SEC("struct_ops", BPF_PROG_TYPE_STRUCT_OPS),
583c9009 6353};
d7be143b 6354
956b620f 6355#undef BPF_PROG_SEC_IMPL
583c9009 6356#undef BPF_PROG_SEC
956b620f
AI		 6357#undef BPF_APROG_SEC
6358#undef BPF_EAPROG_SEC
6359#undef BPF_APROG_COMPAT
d7a18ea7 6360#undef SEC_DEF
583c9009 6361
c76e4c22
TS		 6362#define MAX_TYPE_NAME_SIZE 32
6363
d7a18ea7
AN		 6364static const struct bpf_sec_def *find_sec_def(const char *sec_name)
6365{
6366 int i, n = ARRAY_SIZE(section_defs);
6367
6368 for (i = 0; i < n; i++) {
6369 if (strncmp(sec_name,
6370 section_defs[i].sec, section_defs[i].len))
6371 continue;
6372 return &section_defs[i];
6373 }
6374 return NULL;
6375}
6376
c76e4c22
TS		 6377static char *libbpf_get_type_names(bool attach_type)
6378{
d7a18ea7 6379 int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
c76e4c22
TS		 6380 char *buf;
6381
6382 buf = malloc(len);
6383 if (!buf)
6384 return NULL;
6385
6386 buf[0] = '\0';
6387 /* Forge string buf with all available names */
d7a18ea7
AN		 6388 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
6389 if (attach_type && !section_defs[i].is_attachable)
c76e4c22
TS		 6390 continue;
6391
d7a18ea7 6392 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
c76e4c22
TS		 6393 free(buf);
6394 return NULL;
6395 }
6396 strcat(buf, " ");
d7a18ea7 6397 strcat(buf, section_defs[i].sec);
c76e4c22
TS		 6398 }
6399
6400 return buf;
6401}
6402
b60df2a0
JK		 6403int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
6404 enum bpf_attach_type *expected_attach_type)
583c9009 6405{
d7a18ea7 6406 const struct bpf_sec_def *sec_def;
c76e4c22 6407 char *type_names;
583c9009 6408
b60df2a0
JK		 6409 if (!name)
6410 return -EINVAL;
583c9009 6411
d7a18ea7
AN		 6412 sec_def = find_sec_def(name);
6413 if (sec_def) {
6414 *prog_type = sec_def->prog_type;
6415 *expected_attach_type = sec_def->expected_attach_type;
b60df2a0
JK		 6416 return 0;
6417 }
d7a18ea7 6418
4a3d6c6a 6419 pr_debug("failed to guess program type from ELF section '%s'\n", name);
c76e4c22
TS		 6420 type_names = libbpf_get_type_names(false);
6421 if (type_names != NULL) {
3f519353 6422 pr_debug("supported section(type) names are:%s\n", type_names);
c76e4c22
TS		 6423 free(type_names);
6424 }
6425
dd4436bb 6426 return -ESRCH;
b60df2a0 6427}
583c9009 6428
590a0088
MKL		 6429static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
6430 size_t offset)
6431{
6432 struct bpf_map *map;
6433 size_t i;
6434
6435 for (i = 0; i < obj->nr_maps; i++) {
6436 map = &obj->maps[i];
6437 if (!bpf_map__is_struct_ops(map))
6438 continue;
6439 if (map->sec_offset <= offset &&
6440 offset - map->sec_offset < map->def.value_size)
6441 return map;
6442 }
6443
6444 return NULL;
6445}
6446
6447/* Collect the reloc from ELF and populate the st_ops->progs[] */
6448static int bpf_object__collect_struct_ops_map_reloc(struct bpf_object *obj,
6449 GElf_Shdr *shdr,
6450 Elf_Data *data)
6451{
6452 const struct btf_member *member;
6453 struct bpf_struct_ops *st_ops;
6454 struct bpf_program *prog;
6455 unsigned int shdr_idx;
6456 const struct btf *btf;
6457 struct bpf_map *map;
6458 Elf_Data *symbols;
6459 unsigned int moff;
6460 const char *name;
1d1a3bcf 6461 __u32 member_idx;
590a0088
MKL		 6462 GElf_Sym sym;
6463 GElf_Rel rel;
6464 int i, nrels;
6465
6466 symbols = obj->efile.symbols;
6467 btf = obj->btf;
6468 nrels = shdr->sh_size / shdr->sh_entsize;
6469 for (i = 0; i < nrels; i++) {
6470 if (!gelf_getrel(data, i, &rel)) {
6471 pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
6472 return -LIBBPF_ERRNO__FORMAT;
6473 }
6474
6475 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
6476 pr_warn("struct_ops reloc: symbol %zx not found\n",
6477 (size_t)GELF_R_SYM(rel.r_info));
6478 return -LIBBPF_ERRNO__FORMAT;
6479 }
6480
6481 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
6482 sym.st_name) ? : "<?>";
6483 map = find_struct_ops_map_by_offset(obj, rel.r_offset);
6484 if (!map) {
6485 pr_warn("struct_ops reloc: cannot find map at rel.r_offset %zu\n",
6486 (size_t)rel.r_offset);
6487 return -EINVAL;
6488 }
6489
6490 moff = rel.r_offset - map->sec_offset;
6491 shdr_idx = sym.st_shndx;
6492 st_ops = map->st_ops;
6493 pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel.r_offset %zu map->sec_offset %zu name %d (\'%s\')\n",
6494 map->name,
6495 (long long)(rel.r_info >> 32),
6496 (long long)sym.st_value,
6497 shdr_idx, (size_t)rel.r_offset,
6498 map->sec_offset, sym.st_name, name);
6499
6500 if (shdr_idx >= SHN_LORESERVE) {
6501 pr_warn("struct_ops reloc %s: rel.r_offset %zu shdr_idx %u unsupported non-static function\n",
6502 map->name, (size_t)rel.r_offset, shdr_idx);
6503 return -LIBBPF_ERRNO__RELOC;
6504 }
6505
6506 member = find_member_by_offset(st_ops->type, moff * 8);
6507 if (!member) {
6508 pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
6509 map->name, moff);
6510 return -EINVAL;
6511 }
6512 member_idx = member - btf_members(st_ops->type);
6513 name = btf__name_by_offset(btf, member->name_off);
6514
6515 if (!resolve_func_ptr(btf, member->type, NULL)) {
6516 pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
6517 map->name, name);
6518 return -EINVAL;
6519 }
6520
6521 prog = bpf_object__find_prog_by_idx(obj, shdr_idx);
6522 if (!prog) {
6523 pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
6524 map->name, shdr_idx, name);
6525 return -EINVAL;
6526 }
6527
6528 if (prog->type == BPF_PROG_TYPE_UNSPEC) {
6529 const struct bpf_sec_def *sec_def;
6530
6531 sec_def = find_sec_def(prog->section_name);
6532 if (sec_def &&
6533 sec_def->prog_type != BPF_PROG_TYPE_STRUCT_OPS) {
6534 /* for pr_warn */
6535 prog->type = sec_def->prog_type;
6536 goto invalid_prog;
6537 }
6538
6539 prog->type = BPF_PROG_TYPE_STRUCT_OPS;
6540 prog->attach_btf_id = st_ops->type_id;
6541 prog->expected_attach_type = member_idx;
6542 } else if (prog->type != BPF_PROG_TYPE_STRUCT_OPS ||
6543 prog->attach_btf_id != st_ops->type_id ||
6544 prog->expected_attach_type != member_idx) {
6545 goto invalid_prog;
6546 }
6547 st_ops->progs[member_idx] = prog;
6548 }
6549
6550 return 0;
6551
6552invalid_prog:
6553 pr_warn("struct_ops reloc %s: cannot use prog %s in sec %s with type %u attach_btf_id %u expected_attach_type %u for func ptr %s\n",
6554 map->name, prog->name, prog->section_name, prog->type,
6555 prog->attach_btf_id, prog->expected_attach_type, name);
6556 return -EINVAL;
6557}
6558
a6ed02ca
KS		 6559#define BTF_TRACE_PREFIX "btf_trace_"
6560#define BTF_MAX_NAME_SIZE 128
6561
6562static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
6563 const char *name, __u32 kind)
6564{
6565 char btf_type_name[BTF_MAX_NAME_SIZE];
6566 int ret;
6567
6568 ret = snprintf(btf_type_name, sizeof(btf_type_name),
6569 "%s%s", prefix, name);
6570 /* snprintf returns the number of characters written excluding the
6571 * the terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
6572 * indicates truncation.
6573 */
6574 if (ret < 0 || ret >= sizeof(btf_type_name))
6575 return -ENAMETOOLONG;
6576 return btf__find_by_name_kind(btf, btf_type_name, kind);
6577}
6578
6579static inline int __find_vmlinux_btf_id(struct btf *btf, const char *name,
6580 enum bpf_attach_type attach_type)
6581{
6582 int err;
6583
6584 if (attach_type == BPF_TRACE_RAW_TP)
6585 err = find_btf_by_prefix_kind(btf, BTF_TRACE_PREFIX, name,
6586 BTF_KIND_TYPEDEF);
6587 else
6588 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
6589
ff26ce5c
EC		 6590 if (err <= 0)
6591 pr_warn("%s is not found in vmlinux BTF\n", name);
6592
a6ed02ca
KS		 6593 return err;
6594}
6595
b8c54ea4
AS		 6596int libbpf_find_vmlinux_btf_id(const char *name,
6597 enum bpf_attach_type attach_type)
12a8654b 6598{
a6ed02ca 6599 struct btf *btf;
12a8654b 6600
a6ed02ca 6601 btf = libbpf_find_kernel_btf();
12a8654b
AS		 6602 if (IS_ERR(btf)) {
6603 pr_warn("vmlinux BTF is not found\n");
6604 return -EINVAL;
6605 }
6606
a6ed02ca 6607 return __find_vmlinux_btf_id(btf, name, attach_type);
b8c54ea4
AS		 6608}
6609
e7bf94db
AS		 6610static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
6611{
6612 struct bpf_prog_info_linear *info_linear;
6613 struct bpf_prog_info *info;
6614 struct btf *btf = NULL;
6615 int err = -EINVAL;
6616
6617 info_linear = bpf_program__get_prog_info_linear(attach_prog_fd, 0);
6618 if (IS_ERR_OR_NULL(info_linear)) {
6619 pr_warn("failed get_prog_info_linear for FD %d\n",
6620 attach_prog_fd);
6621 return -EINVAL;
6622 }
6623 info = &info_linear->info;
6624 if (!info->btf_id) {
6625 pr_warn("The target program doesn't have BTF\n");
6626 goto out;
6627 }
6628 if (btf__get_from_id(info->btf_id, &btf)) {
6629 pr_warn("Failed to get BTF of the program\n");
6630 goto out;
6631 }
6632 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
6633 btf__free(btf);
6634 if (err <= 0) {
6635 pr_warn("%s is not found in prog's BTF\n", name);
6636 goto out;
6637 }
6638out:
6639 free(info_linear);
6640 return err;
6641}
6642
a6ed02ca 6643static int libbpf_find_attach_btf_id(struct bpf_program *prog)
b8c54ea4 6644{
a6ed02ca
KS		 6645 enum bpf_attach_type attach_type = prog->expected_attach_type;
6646 __u32 attach_prog_fd = prog->attach_prog_fd;
6647 const char *name = prog->section_name;
b8c54ea4
AS		 6648 int i, err;
6649
12a8654b 6650 if (!name)
b8c54ea4 6651 return -EINVAL;
12a8654b 6652
d7a18ea7
AN		 6653 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
6654 if (!section_defs[i].is_attach_btf)
12a8654b 6655 continue;
d7a18ea7 6656 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
12a8654b 6657 continue;
e7bf94db 6658 if (attach_prog_fd)
d7a18ea7 6659 err = libbpf_find_prog_btf_id(name + section_defs[i].len,
e7bf94db
AS		 6660 attach_prog_fd);
6661 else
a6ed02ca
KS		 6662 err = __find_vmlinux_btf_id(prog->obj->btf_vmlinux,
6663 name + section_defs[i].len,
6664 attach_type);
b8c54ea4 6665 return err;
12a8654b
AS		 6666 }
6667 pr_warn("failed to identify btf_id based on ELF section name '%s'\n", name);
b8c54ea4 6668 return -ESRCH;
12a8654b
AS		 6669}
6670
956b620f
AI		 6671int libbpf_attach_type_by_name(const char *name,
6672 enum bpf_attach_type *attach_type)
6673{
c76e4c22 6674 char *type_names;
956b620f
AI		 6675 int i;
6676
6677 if (!name)
6678 return -EINVAL;
6679
d7a18ea7
AN		 6680 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
6681 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
956b620f 6682 continue;
d7a18ea7 6683 if (!section_defs[i].is_attachable)
956b620f 6684 return -EINVAL;
d7a18ea7 6685 *attach_type = section_defs[i].attach_type;
956b620f
AI		 6686 return 0;
6687 }
4a3d6c6a 6688 pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
c76e4c22
TS		 6689 type_names = libbpf_get_type_names(true);
6690 if (type_names != NULL) {
4a3d6c6a 6691 pr_debug("attachable section(type) names are:%s\n", type_names);
c76e4c22
TS		 6692 free(type_names);
6693 }
6694
956b620f
AI		 6695 return -EINVAL;
6696}
6697
a324aae3 6698int bpf_map__fd(const struct bpf_map *map)
9d759a9b 6699{
6e009e65 6700 return map ? map->fd : -EINVAL;
9d759a9b
WN		 6701}
6702
a324aae3 6703const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
9d759a9b 6704{
53897a78 6705 return map ? &map->def : ERR_PTR(-EINVAL);
9d759a9b
WN		 6706}
6707
a324aae3 6708const char *bpf_map__name(const struct bpf_map *map)
561bbcca 6709{
009ad5d5 6710 return map ? map->name : NULL;
561bbcca
WN		 6711}
6712
5b891af7 6713__u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
8a138aed 6714{
61746dbe 6715 return map ? map->btf_key_type_id : 0;
8a138aed
MKL		 6716}
6717
5b891af7 6718__u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
8a138aed 6719{
61746dbe 6720 return map ? map->btf_value_type_id : 0;
8a138aed
MKL		 6721}
6722
edb13ed4
ACM		 6723int bpf_map__set_priv(struct bpf_map *map, void *priv,
6724 bpf_map_clear_priv_t clear_priv)
9d759a9b
WN		 6725{
6726 if (!map)
6727 return -EINVAL;
6728
6729 if (map->priv) {
6730 if (map->clear_priv)
6731 map->clear_priv(map, map->priv);
6732 }
6733
6734 map->priv = priv;
6735 map->clear_priv = clear_priv;
6736 return 0;
6737}
6738
a324aae3 6739void *bpf_map__priv(const struct bpf_map *map)
9d759a9b 6740{
b4cbfa56 6741 return map ? map->priv : ERR_PTR(-EINVAL);
9d759a9b
WN		 6742}
6743
a324aae3 6744bool bpf_map__is_offload_neutral(const struct bpf_map *map)
f83fb22c
JK		 6745{
6746 return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
6747}
6748
a324aae3 6749bool bpf_map__is_internal(const struct bpf_map *map)
d859900c
DB		 6750{
6751 return map->libbpf_type != LIBBPF_MAP_UNSPEC;
6752}
6753
9aba3613
JK		 6754void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
6755{
6756 map->map_ifindex = ifindex;
6757}
6758
addb9fc9
NS		 6759int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
6760{
6761 if (!bpf_map_type__is_map_in_map(map->def.type)) {
be18010e 6762 pr_warn("error: unsupported map type\n");
addb9fc9
NS		 6763 return -EINVAL;
6764 }
6765 if (map->inner_map_fd != -1) {
be18010e 6766 pr_warn("error: inner_map_fd already specified\n");
addb9fc9
NS		 6767 return -EINVAL;
6768 }
6769 map->inner_map_fd = fd;
6770 return 0;
6771}
6772
0c19a9fb 6773static struct bpf_map *
a324aae3 6774__bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
9d759a9b 6775{
0c19a9fb 6776 ssize_t idx;
9d759a9b
WN		 6777 struct bpf_map *s, *e;
6778
6779 if (!obj || !obj->maps)
6780 return NULL;
6781
6782 s = obj->maps;
6783 e = obj->maps + obj->nr_maps;
6784
0c19a9fb 6785 if ((m < s) || (m >= e)) {
be18010e
KW		 6786 pr_warn("error in %s: map handler doesn't belong to object\n",
6787 __func__);
9d759a9b
WN		 6788 return NULL;
6789 }
6790
0c19a9fb
SF		 6791 idx = (m - obj->maps) + i;
6792 if (idx >= obj->nr_maps || idx < 0)
9d759a9b
WN		 6793 return NULL;
6794 return &obj->maps[idx];
6795}
561bbcca 6796
0c19a9fb 6797struct bpf_map *
a324aae3 6798bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
0c19a9fb
SF		 6799{
6800 if (prev == NULL)
6801 return obj->maps;
6802
6803 return __bpf_map__iter(prev, obj, 1);
6804}
6805
6806struct bpf_map *
a324aae3 6807bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
0c19a9fb
SF		 6808{
6809 if (next == NULL) {
6810 if (!obj->nr_maps)
6811 return NULL;
6812 return obj->maps + obj->nr_maps - 1;
6813 }
6814
6815 return __bpf_map__iter(next, obj, -1);
6816}
6817
561bbcca 6818struct bpf_map *
a324aae3 6819bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
561bbcca
WN		 6820{
6821 struct bpf_map *pos;
6822
f74a53d9 6823 bpf_object__for_each_map(pos, obj) {
973170e6 6824 if (pos->name && !strcmp(pos->name, name))
561bbcca
WN		 6825 return pos;
6826 }
6827 return NULL;
6828}
5a6acad1 6829
f3cea32d 6830int
a324aae3 6831bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
f3cea32d
MF		 6832{
6833 return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
6834}
6835
5a6acad1
WN		 6836struct bpf_map *
6837bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
6838{
db48814b 6839 return ERR_PTR(-ENOTSUP);
5a6acad1 6840}
e28ff1a8
JS		 6841
6842long libbpf_get_error(const void *ptr)
6843{
d98363b5 6844 return PTR_ERR_OR_ZERO(ptr);
e28ff1a8 6845}
6f6d33f3
JF		 6846
6847int bpf_prog_load(const char *file, enum bpf_prog_type type,
6848 struct bpf_object **pobj, int *prog_fd)
d7be143b
AI		 6849{
6850 struct bpf_prog_load_attr attr;
6851
6852 memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
6853 attr.file = file;
6854 attr.prog_type = type;
6855 attr.expected_attach_type = 0;
6856
6857 return bpf_prog_load_xattr(&attr, pobj, prog_fd);
6858}
6859
6860int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
6861 struct bpf_object **pobj, int *prog_fd)
6f6d33f3 6862{
33bae185 6863 struct bpf_object_open_attr open_attr = {};
48cca7e4 6864 struct bpf_program *prog, *first_prog = NULL;
6f6d33f3 6865 struct bpf_object *obj;
f0307a7e 6866 struct bpf_map *map;
6f6d33f3
JF		 6867 int err;
6868
d7be143b
AI		 6869 if (!attr)
6870 return -EINVAL;
17387dd5
JK		 6871 if (!attr->file)
6872 return -EINVAL;
d7be143b 6873
33bae185
LY		 6874 open_attr.file = attr->file;
6875 open_attr.prog_type = attr->prog_type;
6876
07f2d4ea 6877 obj = bpf_object__open_xattr(&open_attr);
3597683c 6878 if (IS_ERR_OR_NULL(obj))
6f6d33f3
JF		 6879 return -ENOENT;
6880
48cca7e4 6881 bpf_object__for_each_program(prog, obj) {
dd4436bb 6882 enum bpf_attach_type attach_type = attr->expected_attach_type;
48cca7e4 6883 /*
dd4436bb
AN		 6884 * to preserve backwards compatibility, bpf_prog_load treats
6885 * attr->prog_type, if specified, as an override to whatever
6886 * bpf_object__open guessed
48cca7e4 6887 */
dd4436bb
AN		 6888 if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) {
6889 bpf_program__set_type(prog, attr->prog_type);
6890 bpf_program__set_expected_attach_type(prog,
6891 attach_type);
6892 }
6893 if (bpf_program__get_type(prog) == BPF_PROG_TYPE_UNSPEC) {
6894 /*
6895 * we haven't guessed from section name and user
6896 * didn't provide a fallback type, too bad...
6897 */
6898 bpf_object__close(obj);
6899 return -EINVAL;
583c9009 6900 }
48cca7e4 6901
dd4436bb 6902 prog->prog_ifindex = attr->ifindex;
da11b417 6903 prog->log_level = attr->log_level;
04656198 6904 prog->prog_flags = attr->prog_flags;
69495d2a 6905 if (!first_prog)
48cca7e4
AS		 6906 first_prog = prog;
6907 }
6908
f74a53d9 6909 bpf_object__for_each_map(map, obj) {
f83fb22c
JK		 6910 if (!bpf_map__is_offload_neutral(map))
6911 map->map_ifindex = attr->ifindex;
f0307a7e
DB		 6912 }
6913
48cca7e4 6914 if (!first_prog) {
be18010e 6915 pr_warn("object file doesn't contain bpf program\n");
48cca7e4
AS		 6916 bpf_object__close(obj);
6917 return -ENOENT;
583c9009
RG		 6918 }
6919
6f6d33f3
JF		 6920 err = bpf_object__load(obj);
6921 if (err) {
6922 bpf_object__close(obj);
6923 return -EINVAL;
6924 }
6925
6926 *pobj = obj;
48cca7e4 6927 *prog_fd = bpf_program__fd(first_prog);
6f6d33f3
JF		 6928 return 0;
6929}
d0cabbb0 6930
1c2e9efc 6931struct bpf_link {
d6958706 6932 int (*detach)(struct bpf_link *link);
1c2e9efc 6933 int (*destroy)(struct bpf_link *link);
c016b68e
AN		 6934 char *pin_path; /* NULL, if not pinned */
6935 int fd; /* hook FD, -1 if not applicable */
d6958706 6936 bool disconnected;
1c2e9efc
AN		 6937};
6938
d6958706
AN		 6939/* Release "ownership" of underlying BPF resource (typically, BPF program
6940 * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
6941 * link, when destructed through bpf_link__destroy() call won't attempt to
6942 * detach/unregisted that BPF resource. This is useful in situations where,
6943 * say, attached BPF program has to outlive userspace program that attached it
6944 * in the system. Depending on type of BPF program, though, there might be
6945 * additional steps (like pinning BPF program in BPF FS) necessary to ensure
6946 * exit of userspace program doesn't trigger automatic detachment and clean up
6947 * inside the kernel.
6948 */
6949void bpf_link__disconnect(struct bpf_link *link)
6950{
6951 link->disconnected = true;
6952}
6953
1c2e9efc
AN		 6954int bpf_link__destroy(struct bpf_link *link)
6955{
d6958706 6956 int err = 0;
1c2e9efc
AN		 6957
6958 if (!link)
6959 return 0;
6960
d6958706
AN		 6961 if (!link->disconnected && link->detach)
6962 err = link->detach(link);
6963 if (link->destroy)
6964 link->destroy(link);
c016b68e
AN		 6965 if (link->pin_path)
6966 free(link->pin_path);
1c2e9efc
AN		 6967 free(link);
6968
6969 return err;
6970}
6971
c016b68e
AN		 6972int bpf_link__fd(const struct bpf_link *link)
6973{
6974 return link->fd;
6975}
6976
6977const char *bpf_link__pin_path(const struct bpf_link *link)
6978{
6979 return link->pin_path;
6980}
6981
6982static int bpf_link__detach_fd(struct bpf_link *link)
6983{
6984 return close(link->fd);
6985}
6986
6987struct bpf_link *bpf_link__open(const char *path)
6988{
6989 struct bpf_link *link;
6990 int fd;
6991
6992 fd = bpf_obj_get(path);
6993 if (fd < 0) {
6994 fd = -errno;
6995 pr_warn("failed to open link at %s: %d\n", path, fd);
6996 return ERR_PTR(fd);
6997 }
6998
6999 link = calloc(1, sizeof(*link));
7000 if (!link) {
7001 close(fd);
7002 return ERR_PTR(-ENOMEM);
7003 }
7004 link->detach = &bpf_link__detach_fd;
7005 link->fd = fd;
7006
7007 link->pin_path = strdup(path);
7008 if (!link->pin_path) {
7009 bpf_link__destroy(link);
7010 return ERR_PTR(-ENOMEM);
7011 }
7012
7013 return link;
7014}
7015
7016int bpf_link__pin(struct bpf_link *link, const char *path)
7017{
7018 int err;
7019
7020 if (link->pin_path)
7021 return -EBUSY;
7022 err = make_parent_dir(path);
7023 if (err)
7024 return err;
7025 err = check_path(path);
7026 if (err)
7027 return err;
7028
7029 link->pin_path = strdup(path);
7030 if (!link->pin_path)
7031 return -ENOMEM;
7032
7033 if (bpf_obj_pin(link->fd, link->pin_path)) {
7034 err = -errno;
7035 zfree(&link->pin_path);
7036 return err;
7037 }
7038
7039 pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
7040 return 0;
7041}
7042
7043int bpf_link__unpin(struct bpf_link *link)
7044{
7045 int err;
7046
7047 if (!link->pin_path)
7048 return -EINVAL;
7049
7050 err = unlink(link->pin_path);
7051 if (err != 0)
7052 return -errno;
7053
7054 pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
7055 zfree(&link->pin_path);
7056 return 0;
7057}
63f2f5ee 7058
d6958706 7059static int bpf_link__detach_perf_event(struct bpf_link *link)
63f2f5ee 7060{
63f2f5ee
AN		 7061 int err;
7062
c016b68e 7063 err = ioctl(link->fd, PERF_EVENT_IOC_DISABLE, 0);
63f2f5ee
AN		 7064 if (err)
7065 err = -errno;
7066
c016b68e 7067 close(link->fd);
63f2f5ee
AN		 7068 return err;
7069}
7070
7071struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
7072 int pfd)
7073{
7074 char errmsg[STRERR_BUFSIZE];
c016b68e 7075 struct bpf_link *link;
63f2f5ee
AN		 7076 int prog_fd, err;
7077
7078 if (pfd < 0) {
be18010e
KW		 7079 pr_warn("program '%s': invalid perf event FD %d\n",
7080 bpf_program__title(prog, false), pfd);
63f2f5ee
AN		 7081 return ERR_PTR(-EINVAL);
7082 }
7083 prog_fd = bpf_program__fd(prog);
7084 if (prog_fd < 0) {
be18010e
KW		 7085 pr_warn("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
7086 bpf_program__title(prog, false));
63f2f5ee
AN		 7087 return ERR_PTR(-EINVAL);
7088 }
7089
d6958706 7090 link = calloc(1, sizeof(*link));
63f2f5ee
AN		 7091 if (!link)
7092 return ERR_PTR(-ENOMEM);
c016b68e 7093 link->detach = &bpf_link__detach_perf_event;
63f2f5ee
AN		 7094 link->fd = pfd;
7095
7096 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
7097 err = -errno;
7098 free(link);
be18010e
KW		 7099 pr_warn("program '%s': failed to attach to pfd %d: %s\n",
7100 bpf_program__title(prog, false), pfd,
63f2f5ee
AN		 7101 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7102 return ERR_PTR(err);
7103 }
7104 if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
7105 err = -errno;
7106 free(link);
be18010e
KW		 7107 pr_warn("program '%s': failed to enable pfd %d: %s\n",
7108 bpf_program__title(prog, false), pfd,
63f2f5ee
AN		 7109 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7110 return ERR_PTR(err);
7111 }
c016b68e 7112 return link;
63f2f5ee
AN		 7113}
7114
b2650027
AN		 7115/*
7116 * this function is expected to parse integer in the range of [0, 2^31-1] from
7117 * given file using scanf format string fmt. If actual parsed value is
7118 * negative, the result might be indistinguishable from error
7119 */
7120static int parse_uint_from_file(const char *file, const char *fmt)
7121{
7122 char buf[STRERR_BUFSIZE];
7123 int err, ret;
7124 FILE *f;
7125
7126 f = fopen(file, "r");
7127 if (!f) {
7128 err = -errno;
7129 pr_debug("failed to open '%s': %s\n", file,
7130 libbpf_strerror_r(err, buf, sizeof(buf)));
7131 return err;
7132 }
7133 err = fscanf(f, fmt, &ret);
7134 if (err != 1) {
7135 err = err == EOF ? -EIO : -errno;
7136 pr_debug("failed to parse '%s': %s\n", file,
7137 libbpf_strerror_r(err, buf, sizeof(buf)));
7138 fclose(f);
7139 return err;
7140 }
7141 fclose(f);
7142 return ret;
7143}
7144
7145static int determine_kprobe_perf_type(void)
7146{
7147 const char *file = "/sys/bus/event_source/devices/kprobe/type";
7148
7149 return parse_uint_from_file(file, "%d\n");
7150}
7151
7152static int determine_uprobe_perf_type(void)
7153{
7154 const char *file = "/sys/bus/event_source/devices/uprobe/type";
7155
7156 return parse_uint_from_file(file, "%d\n");
7157}
7158
7159static int determine_kprobe_retprobe_bit(void)
7160{
7161 const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
7162
7163 return parse_uint_from_file(file, "config:%d\n");
7164}
7165
7166static int determine_uprobe_retprobe_bit(void)
7167{
7168 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
7169
7170 return parse_uint_from_file(file, "config:%d\n");
7171}
7172
7173static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
7174 uint64_t offset, int pid)
7175{
7176 struct perf_event_attr attr = {};
7177 char errmsg[STRERR_BUFSIZE];
7178 int type, pfd, err;
7179
7180 type = uprobe ? determine_uprobe_perf_type()
7181 : determine_kprobe_perf_type();
7182 if (type < 0) {
be18010e
KW		 7183 pr_warn("failed to determine %s perf type: %s\n",
7184 uprobe ? "uprobe" : "kprobe",
7185 libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
b2650027
AN		 7186 return type;
7187 }
7188 if (retprobe) {
7189 int bit = uprobe ? determine_uprobe_retprobe_bit()
7190 : determine_kprobe_retprobe_bit();
7191
7192 if (bit < 0) {
be18010e
KW		 7193 pr_warn("failed to determine %s retprobe bit: %s\n",
7194 uprobe ? "uprobe" : "kprobe",
7195 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
b2650027
AN		 7196 return bit;
7197 }
7198 attr.config |= 1 << bit;
7199 }
7200 attr.size = sizeof(attr);
7201 attr.type = type;
36db2a94
AN		 7202 attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
7203 attr.config2 = offset; /* kprobe_addr or probe_offset */
b2650027
AN		 7204
7205 /* pid filter is meaningful only for uprobes */
7206 pfd = syscall(__NR_perf_event_open, &attr,
7207 pid < 0 ? -1 : pid /* pid */,
7208 pid == -1 ? 0 : -1 /* cpu */,
7209 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
7210 if (pfd < 0) {
7211 err = -errno;
be18010e
KW		 7212 pr_warn("%s perf_event_open() failed: %s\n",
7213 uprobe ? "uprobe" : "kprobe",
7214 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
b2650027
AN		 7215 return err;
7216 }
7217 return pfd;
7218}
7219
7220struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
7221 bool retprobe,
7222 const char *func_name)
7223{
7224 char errmsg[STRERR_BUFSIZE];
7225 struct bpf_link *link;
7226 int pfd, err;
7227
7228 pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
7229 0 /* offset */, -1 /* pid */);
7230 if (pfd < 0) {
be18010e
KW		 7231 pr_warn("program '%s': failed to create %s '%s' perf event: %s\n",
7232 bpf_program__title(prog, false),
7233 retprobe ? "kretprobe" : "kprobe", func_name,
7234 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
b2650027
AN		 7235 return ERR_PTR(pfd);
7236 }
7237 link = bpf_program__attach_perf_event(prog, pfd);
7238 if (IS_ERR(link)) {
7239 close(pfd);
7240 err = PTR_ERR(link);
be18010e
KW		 7241 pr_warn("program '%s': failed to attach to %s '%s': %s\n",
7242 bpf_program__title(prog, false),
7243 retprobe ? "kretprobe" : "kprobe", func_name,
7244 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
b2650027
AN		 7245 return link;
7246 }
7247 return link;
7248}
7249
d7a18ea7
AN		 7250static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
7251 struct bpf_program *prog)
7252{
7253 const char *func_name;
7254 bool retprobe;
7255
7256 func_name = bpf_program__title(prog, false) + sec->len;
7257 retprobe = strcmp(sec->sec, "kretprobe/") == 0;
7258
7259 return bpf_program__attach_kprobe(prog, retprobe, func_name);
7260}
7261
b2650027
AN		 7262struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
7263 bool retprobe, pid_t pid,
7264 const char *binary_path,
7265 size_t func_offset)
7266{
7267 char errmsg[STRERR_BUFSIZE];
7268 struct bpf_link *link;
7269 int pfd, err;
7270
7271 pfd = perf_event_open_probe(true /* uprobe */, retprobe,
7272 binary_path, func_offset, pid);
7273 if (pfd < 0) {
be18010e
KW		 7274 pr_warn("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
7275 bpf_program__title(prog, false),
7276 retprobe ? "uretprobe" : "uprobe",
7277 binary_path, func_offset,
7278 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
b2650027
AN		 7279 return ERR_PTR(pfd);
7280 }
7281 link = bpf_program__attach_perf_event(prog, pfd);
7282 if (IS_ERR(link)) {
7283 close(pfd);
7284 err = PTR_ERR(link);
be18010e
KW		 7285 pr_warn("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
7286 bpf_program__title(prog, false),
7287 retprobe ? "uretprobe" : "uprobe",
7288 binary_path, func_offset,
7289 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
b2650027
AN		 7290 return link;
7291 }
7292 return link;
7293}
7294
f6de59c1
AN		 7295static int determine_tracepoint_id(const char *tp_category,
7296 const char *tp_name)
7297{
7298 char file[PATH_MAX];
7299 int ret;
7300
7301 ret = snprintf(file, sizeof(file),
7302 "/sys/kernel/debug/tracing/events/%s/%s/id",
7303 tp_category, tp_name);
7304 if (ret < 0)
7305 return -errno;
7306 if (ret >= sizeof(file)) {
7307 pr_debug("tracepoint %s/%s path is too long\n",
7308 tp_category, tp_name);
7309 return -E2BIG;
7310 }
7311 return parse_uint_from_file(file, "%d\n");
7312}
7313
7314static int perf_event_open_tracepoint(const char *tp_category,
7315 const char *tp_name)
7316{
7317 struct perf_event_attr attr = {};
7318 char errmsg[STRERR_BUFSIZE];
7319 int tp_id, pfd, err;
7320
7321 tp_id = determine_tracepoint_id(tp_category, tp_name);
7322 if (tp_id < 0) {
be18010e
KW		 7323 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
7324 tp_category, tp_name,
7325 libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
f6de59c1
AN		 7326 return tp_id;
7327 }
7328
7329 attr.type = PERF_TYPE_TRACEPOINT;
7330 attr.size = sizeof(attr);
7331 attr.config = tp_id;
7332
7333 pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
7334 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
7335 if (pfd < 0) {
7336 err = -errno;
be18010e
KW		 7337 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
7338 tp_category, tp_name,
7339 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
f6de59c1
AN		 7340 return err;
7341 }
7342 return pfd;
7343}
7344
7345struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
7346 const char *tp_category,
7347 const char *tp_name)
7348{
7349 char errmsg[STRERR_BUFSIZE];
7350 struct bpf_link *link;
7351 int pfd, err;
7352
7353 pfd = perf_event_open_tracepoint(tp_category, tp_name);
7354 if (pfd < 0) {
be18010e
KW		 7355 pr_warn("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
7356 bpf_program__title(prog, false),
7357 tp_category, tp_name,
7358 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
f6de59c1
AN		 7359 return ERR_PTR(pfd);
7360 }
7361 link = bpf_program__attach_perf_event(prog, pfd);
7362 if (IS_ERR(link)) {
7363 close(pfd);
7364 err = PTR_ERR(link);
be18010e
KW		 7365 pr_warn("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
7366 bpf_program__title(prog, false),
7367 tp_category, tp_name,
7368 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
f6de59c1
AN		 7369 return link;
7370 }
7371 return link;
7372}
7373
d7a18ea7
AN		 7374static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
7375 struct bpf_program *prog)
7376{
7377 char *sec_name, *tp_cat, *tp_name;
7378 struct bpf_link *link;
7379
7380 sec_name = strdup(bpf_program__title(prog, false));
7381 if (!sec_name)
7382 return ERR_PTR(-ENOMEM);
7383
7384 /* extract "tp/<category>/<name>" */
7385 tp_cat = sec_name + sec->len;
7386 tp_name = strchr(tp_cat, '/');
7387 if (!tp_name) {
7388 link = ERR_PTR(-EINVAL);
7389 goto out;
7390 }
7391 *tp_name = '\0';
7392 tp_name++;
7393
7394 link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
7395out:
7396 free(sec_name);
7397 return link;
7398}
7399
84bf5e1f
AN		 7400struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
7401 const char *tp_name)
7402{
7403 char errmsg[STRERR_BUFSIZE];
c016b68e 7404 struct bpf_link *link;
84bf5e1f
AN		 7405 int prog_fd, pfd;
7406
7407 prog_fd = bpf_program__fd(prog);
7408 if (prog_fd < 0) {
be18010e
KW		 7409 pr_warn("program '%s': can't attach before loaded\n",
7410 bpf_program__title(prog, false));
84bf5e1f
AN		 7411 return ERR_PTR(-EINVAL);
7412 }
7413
d6958706 7414 link = calloc(1, sizeof(*link));
84bf5e1f
AN		 7415 if (!link)
7416 return ERR_PTR(-ENOMEM);
c016b68e 7417 link->detach = &bpf_link__detach_fd;
84bf5e1f
AN		 7418
7419 pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
7420 if (pfd < 0) {
7421 pfd = -errno;
7422 free(link);
be18010e
KW		 7423 pr_warn("program '%s': failed to attach to raw tracepoint '%s': %s\n",
7424 bpf_program__title(prog, false), tp_name,
7425 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
84bf5e1f
AN		 7426 return ERR_PTR(pfd);
7427 }
7428 link->fd = pfd;
c016b68e 7429 return link;
84bf5e1f
AN		 7430}
7431
d7a18ea7
AN		 7432static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
7433 struct bpf_program *prog)
7434{
7435 const char *tp_name = bpf_program__title(prog, false) + sec->len;
7436
7437 return bpf_program__attach_raw_tracepoint(prog, tp_name);
7438}
7439
b8c54ea4
AS		 7440struct bpf_link *bpf_program__attach_trace(struct bpf_program *prog)
7441{
7442 char errmsg[STRERR_BUFSIZE];
c016b68e 7443 struct bpf_link *link;
b8c54ea4
AS		 7444 int prog_fd, pfd;
7445
7446 prog_fd = bpf_program__fd(prog);
7447 if (prog_fd < 0) {
7448 pr_warn("program '%s': can't attach before loaded\n",
7449 bpf_program__title(prog, false));
7450 return ERR_PTR(-EINVAL);
7451 }
7452
d6958706 7453 link = calloc(1, sizeof(*link));
b8c54ea4
AS		 7454 if (!link)
7455 return ERR_PTR(-ENOMEM);
c016b68e 7456 link->detach = &bpf_link__detach_fd;
b8c54ea4
AS		 7457
7458 pfd = bpf_raw_tracepoint_open(NULL, prog_fd);
7459 if (pfd < 0) {
7460 pfd = -errno;
7461 free(link);
7462 pr_warn("program '%s': failed to attach to trace: %s\n",
7463 bpf_program__title(prog, false),
7464 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
7465 return ERR_PTR(pfd);
7466 }
7467 link->fd = pfd;
7468 return (struct bpf_link *)link;
7469}
7470
d7a18ea7
AN		 7471static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
7472 struct bpf_program *prog)
7473{
7474 return bpf_program__attach_trace(prog);
7475}
7476
7477struct bpf_link *bpf_program__attach(struct bpf_program *prog)
7478{
7479 const struct bpf_sec_def *sec_def;
7480
7481 sec_def = find_sec_def(bpf_program__title(prog, false));
7482 if (!sec_def || !sec_def->attach_fn)
7483 return ERR_PTR(-ESRCH);
7484
7485 return sec_def->attach_fn(sec_def, prog);
7486}
7487
590a0088
MKL		 7488static int bpf_link__detach_struct_ops(struct bpf_link *link)
7489{
590a0088
MKL		 7490 __u32 zero = 0;
7491
c016b68e 7492 if (bpf_map_delete_elem(link->fd, &zero))
590a0088
MKL		 7493 return -errno;
7494
7495 return 0;
7496}
7497
7498struct bpf_link *bpf_map__attach_struct_ops(struct bpf_map *map)
7499{
7500 struct bpf_struct_ops *st_ops;
c016b68e 7501 struct bpf_link *link;
590a0088
MKL		 7502 __u32 i, zero = 0;
7503 int err;
7504
7505 if (!bpf_map__is_struct_ops(map) || map->fd == -1)
7506 return ERR_PTR(-EINVAL);
7507
7508 link = calloc(1, sizeof(*link));
7509 if (!link)
7510 return ERR_PTR(-EINVAL);
7511
7512 st_ops = map->st_ops;
7513 for (i = 0; i < btf_vlen(st_ops->type); i++) {
7514 struct bpf_program *prog = st_ops->progs[i];
7515 void *kern_data;
7516 int prog_fd;
7517
7518 if (!prog)
7519 continue;
7520
7521 prog_fd = bpf_program__fd(prog);
7522 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
7523 *(unsigned long *)kern_data = prog_fd;
7524 }
7525
7526 err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0);
7527 if (err) {
7528 err = -errno;
7529 free(link);
7530 return ERR_PTR(err);
7531 }
7532
c016b68e 7533 link->detach = bpf_link__detach_struct_ops;
590a0088
MKL		 7534 link->fd = map->fd;
7535
c016b68e 7536 return link;
590a0088
MKL		 7537}
7538
d0cabbb0 7539enum bpf_perf_event_ret
3dca2115
DB		 7540bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
7541 void **copy_mem, size_t *copy_size,
7542 bpf_perf_event_print_t fn, void *private_data)
d0cabbb0 7543{
3dca2115 7544 struct perf_event_mmap_page *header = mmap_mem;
a64af0ef 7545 __u64 data_head = ring_buffer_read_head(header);
d0cabbb0 7546 __u64 data_tail = header->data_tail;
3dca2115
DB		 7547 void *base = ((__u8 *)header) + page_size;
7548 int ret = LIBBPF_PERF_EVENT_CONT;
7549 struct perf_event_header *ehdr;
7550 size_t ehdr_size;
7551
7552 while (data_head != data_tail) {
7553 ehdr = base + (data_tail & (mmap_size - 1));
7554 ehdr_size = ehdr->size;
7555
7556 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
7557 void *copy_start = ehdr;
7558 size_t len_first = base + mmap_size - copy_start;
7559 size_t len_secnd = ehdr_size - len_first;
7560
7561 if (*copy_size < ehdr_size) {
7562 free(*copy_mem);
7563 *copy_mem = malloc(ehdr_size);
7564 if (!*copy_mem) {
7565 *copy_size = 0;
d0cabbb0
JK		 7566 ret = LIBBPF_PERF_EVENT_ERROR;
7567 break;
7568 }
3dca2115 7569 *copy_size = ehdr_size;
d0cabbb0
JK		 7570 }
7571
3dca2115
DB		 7572 memcpy(*copy_mem, copy_start, len_first);
7573 memcpy(*copy_mem + len_first, base, len_secnd);
7574 ehdr = *copy_mem;
d0cabbb0
JK		 7575 }
7576
3dca2115
DB		 7577 ret = fn(ehdr, private_data);
7578 data_tail += ehdr_size;
d0cabbb0
JK		 7579 if (ret != LIBBPF_PERF_EVENT_CONT)
7580 break;
d0cabbb0
JK		 7581 }
7582
a64af0ef 7583 ring_buffer_write_tail(header, data_tail);
d0cabbb0
JK		 7584 return ret;
7585}
34be1646 7586
fb84b822
AN		 7587struct perf_buffer;
7588
7589struct perf_buffer_params {
7590 struct perf_event_attr *attr;
7591 /* if event_cb is specified, it takes precendence */
7592 perf_buffer_event_fn event_cb;
7593 /* sample_cb and lost_cb are higher-level common-case callbacks */
7594 perf_buffer_sample_fn sample_cb;
7595 perf_buffer_lost_fn lost_cb;
7596 void *ctx;
7597 int cpu_cnt;
7598 int *cpus;
7599 int *map_keys;
7600};
7601
7602struct perf_cpu_buf {
7603 struct perf_buffer *pb;
7604 void *base; /* mmap()'ed memory */
7605 void *buf; /* for reconstructing segmented data */
7606 size_t buf_size;
7607 int fd;
7608 int cpu;
7609 int map_key;
7610};
7611
7612struct perf_buffer {
7613 perf_buffer_event_fn event_cb;
7614 perf_buffer_sample_fn sample_cb;
7615 perf_buffer_lost_fn lost_cb;
7616 void *ctx; /* passed into callbacks */
7617
7618 size_t page_size;
7619 size_t mmap_size;
7620 struct perf_cpu_buf **cpu_bufs;
7621 struct epoll_event *events;
783b8f01 7622 int cpu_cnt; /* number of allocated CPU buffers */
fb84b822
AN		 7623 int epoll_fd; /* perf event FD */
7624 int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
7625};
7626
7627static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
7628 struct perf_cpu_buf *cpu_buf)
7629{
7630 if (!cpu_buf)
7631 return;
7632 if (cpu_buf->base &&
7633 munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
be18010e 7634 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
fb84b822
AN		 7635 if (cpu_buf->fd >= 0) {
7636 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
7637 close(cpu_buf->fd);
7638 }
7639 free(cpu_buf->buf);
7640 free(cpu_buf);
7641}
7642
7643void perf_buffer__free(struct perf_buffer *pb)
7644{
7645 int i;
7646
7647 if (!pb)
7648 return;
7649 if (pb->cpu_bufs) {
7650 for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
7651 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
7652
7653 bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
7654 perf_buffer__free_cpu_buf(pb, cpu_buf);
7655 }
7656 free(pb->cpu_bufs);
7657 }
7658 if (pb->epoll_fd >= 0)
7659 close(pb->epoll_fd);
7660 free(pb->events);
7661 free(pb);
7662}
7663
7664static struct perf_cpu_buf *
7665perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
7666 int cpu, int map_key)
7667{
7668 struct perf_cpu_buf *cpu_buf;
7669 char msg[STRERR_BUFSIZE];
7670 int err;
7671
7672 cpu_buf = calloc(1, sizeof(*cpu_buf));
7673 if (!cpu_buf)
7674 return ERR_PTR(-ENOMEM);
7675
7676 cpu_buf->pb = pb;
7677 cpu_buf->cpu = cpu;
7678 cpu_buf->map_key = map_key;
7679
7680 cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
7681 -1, PERF_FLAG_FD_CLOEXEC);
7682 if (cpu_buf->fd < 0) {
7683 err = -errno;
be18010e
KW		 7684 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
7685 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
fb84b822
AN		 7686 goto error;
7687 }
7688
7689 cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
7690 PROT_READ | PROT_WRITE, MAP_SHARED,
7691 cpu_buf->fd, 0);
7692 if (cpu_buf->base == MAP_FAILED) {
7693 cpu_buf->base = NULL;
7694 err = -errno;
be18010e
KW		 7695 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
7696 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
fb84b822
AN		 7697 goto error;
7698 }
7699
7700 if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
7701 err = -errno;
be18010e
KW		 7702 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
7703 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
fb84b822
AN		 7704 goto error;
7705 }
7706
7707 return cpu_buf;
7708
7709error:
7710 perf_buffer__free_cpu_buf(pb, cpu_buf);
7711 return (struct perf_cpu_buf *)ERR_PTR(err);
7712}
7713
7714static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
7715 struct perf_buffer_params *p);
7716
7717struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
7718 const struct perf_buffer_opts *opts)
7719{
7720 struct perf_buffer_params p = {};
4be6e05c
ACM		 7721 struct perf_event_attr attr = { 0, };
7722
7723 attr.config = PERF_COUNT_SW_BPF_OUTPUT,
7724 attr.type = PERF_TYPE_SOFTWARE;
7725 attr.sample_type = PERF_SAMPLE_RAW;
7726 attr.sample_period = 1;
7727 attr.wakeup_events = 1;
fb84b822
AN		 7728
7729 p.attr = &attr;
7730 p.sample_cb = opts ? opts->sample_cb : NULL;
7731 p.lost_cb = opts ? opts->lost_cb : NULL;
7732 p.ctx = opts ? opts->ctx : NULL;
7733
7734 return __perf_buffer__new(map_fd, page_cnt, &p);
7735}
7736
7737struct perf_buffer *
7738perf_buffer__new_raw(int map_fd, size_t page_cnt,
7739 const struct perf_buffer_raw_opts *opts)
7740{
7741 struct perf_buffer_params p = {};
7742
7743 p.attr = opts->attr;
7744 p.event_cb = opts->event_cb;
7745 p.ctx = opts->ctx;
7746 p.cpu_cnt = opts->cpu_cnt;
7747 p.cpus = opts->cpus;
7748 p.map_keys = opts->map_keys;
7749
7750 return __perf_buffer__new(map_fd, page_cnt, &p);
7751}
7752
7753static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
7754 struct perf_buffer_params *p)
7755{
783b8f01 7756 const char *online_cpus_file = "/sys/devices/system/cpu/online";
fb84b822
AN		 7757 struct bpf_map_info map = {};
7758 char msg[STRERR_BUFSIZE];
7759 struct perf_buffer *pb;
783b8f01 7760 bool *online = NULL;
fb84b822 7761 __u32 map_info_len;
783b8f01 7762 int err, i, j, n;
fb84b822
AN		 7763
7764 if (page_cnt & (page_cnt - 1)) {
be18010e
KW		 7765 pr_warn("page count should be power of two, but is %zu\n",
7766 page_cnt);
fb84b822
AN		 7767 return ERR_PTR(-EINVAL);
7768 }
7769
7770 map_info_len = sizeof(map);
7771 err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
7772 if (err) {
7773 err = -errno;
be18010e
KW		 7774 pr_warn("failed to get map info for map FD %d: %s\n",
7775 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
fb84b822
AN		 7776 return ERR_PTR(err);
7777 }
7778
7779 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
be18010e
KW		 7780 pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
7781 map.name);
fb84b822
AN		 7782 return ERR_PTR(-EINVAL);
7783 }
7784
7785 pb = calloc(1, sizeof(*pb));
7786 if (!pb)
7787 return ERR_PTR(-ENOMEM);
7788
7789 pb->event_cb = p->event_cb;
7790 pb->sample_cb = p->sample_cb;
7791 pb->lost_cb = p->lost_cb;
7792 pb->ctx = p->ctx;
7793
7794 pb->page_size = getpagesize();
7795 pb->mmap_size = pb->page_size * page_cnt;
7796 pb->map_fd = map_fd;
7797
7798 pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
7799 if (pb->epoll_fd < 0) {
7800 err = -errno;
be18010e
KW		 7801 pr_warn("failed to create epoll instance: %s\n",
7802 libbpf_strerror_r(err, msg, sizeof(msg)));
fb84b822
AN		 7803 goto error;
7804 }
7805
7806 if (p->cpu_cnt > 0) {
7807 pb->cpu_cnt = p->cpu_cnt;
7808 } else {
7809 pb->cpu_cnt = libbpf_num_possible_cpus();
7810 if (pb->cpu_cnt < 0) {
7811 err = pb->cpu_cnt;
7812 goto error;
7813 }
7814 if (map.max_entries < pb->cpu_cnt)
7815 pb->cpu_cnt = map.max_entries;
7816 }
7817
7818 pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
7819 if (!pb->events) {
7820 err = -ENOMEM;
be18010e 7821 pr_warn("failed to allocate events: out of memory\n");
fb84b822
AN		 7822 goto error;
7823 }
7824 pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
7825 if (!pb->cpu_bufs) {
7826 err = -ENOMEM;
be18010e 7827 pr_warn("failed to allocate buffers: out of memory\n");
fb84b822
AN		 7828 goto error;
7829 }
7830
783b8f01
AN		 7831 err = parse_cpu_mask_file(online_cpus_file, &online, &n);
7832 if (err) {
7833 pr_warn("failed to get online CPU mask: %d\n", err);
7834 goto error;
7835 }
7836
7837 for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
fb84b822
AN		 7838 struct perf_cpu_buf *cpu_buf;
7839 int cpu, map_key;
7840
7841 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
7842 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
7843
783b8f01
AN		 7844 /* in case user didn't explicitly requested particular CPUs to
7845 * be attached to, skip offline/not present CPUs
7846 */
7847 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
7848 continue;
7849
fb84b822
AN		 7850 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
7851 if (IS_ERR(cpu_buf)) {
7852 err = PTR_ERR(cpu_buf);
7853 goto error;
7854 }
7855
783b8f01 7856 pb->cpu_bufs[j] = cpu_buf;
fb84b822
AN		 7857
7858 err = bpf_map_update_elem(pb->map_fd, &map_key,
7859 &cpu_buf->fd, 0);
7860 if (err) {
7861 err = -errno;
be18010e
KW		 7862 pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
7863 cpu, map_key, cpu_buf->fd,
7864 libbpf_strerror_r(err, msg, sizeof(msg)));
fb84b822
AN		 7865 goto error;
7866 }
7867
783b8f01
AN		 7868 pb->events[j].events = EPOLLIN;
7869 pb->events[j].data.ptr = cpu_buf;
fb84b822 7870 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
783b8f01 7871 &pb->events[j]) < 0) {
fb84b822 7872 err = -errno;
be18010e
KW		 7873 pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
7874 cpu, cpu_buf->fd,
7875 libbpf_strerror_r(err, msg, sizeof(msg)));
fb84b822
AN		 7876 goto error;
7877 }
783b8f01 7878 j++;
fb84b822 7879 }
783b8f01
AN		 7880 pb->cpu_cnt = j;
7881 free(online);
fb84b822
AN		 7882
7883 return pb;
7884
7885error:
783b8f01 7886 free(online);
fb84b822
AN		 7887 if (pb)
7888 perf_buffer__free(pb);
7889 return ERR_PTR(err);
7890}
7891
7892struct perf_sample_raw {
7893 struct perf_event_header header;
7894 uint32_t size;
7895 char data[0];
7896};
7897
7898struct perf_sample_lost {
7899 struct perf_event_header header;
7900 uint64_t id;
7901 uint64_t lost;
7902 uint64_t sample_id;
7903};
7904
7905static enum bpf_perf_event_ret
7906perf_buffer__process_record(struct perf_event_header *e, void *ctx)
7907{
7908 struct perf_cpu_buf *cpu_buf = ctx;
7909 struct perf_buffer *pb = cpu_buf->pb;
7910 void *data = e;
7911
7912 /* user wants full control over parsing perf event */
7913 if (pb->event_cb)
7914 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
7915
7916 switch (e->type) {
7917 case PERF_RECORD_SAMPLE: {
7918 struct perf_sample_raw *s = data;
7919
7920 if (pb->sample_cb)
7921 pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
7922 break;
7923 }
7924 case PERF_RECORD_LOST: {
7925 struct perf_sample_lost *s = data;
7926
7927 if (pb->lost_cb)
7928 pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
7929 break;
7930 }
7931 default:
be18010e 7932 pr_warn("unknown perf sample type %d\n", e->type);
fb84b822
AN		 7933 return LIBBPF_PERF_EVENT_ERROR;
7934 }
7935 return LIBBPF_PERF_EVENT_CONT;
7936}
7937
7938static int perf_buffer__process_records(struct perf_buffer *pb,
7939 struct perf_cpu_buf *cpu_buf)
7940{
7941 enum bpf_perf_event_ret ret;
7942
7943 ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
7944 pb->page_size, &cpu_buf->buf,
7945 &cpu_buf->buf_size,
7946 perf_buffer__process_record, cpu_buf);
7947 if (ret != LIBBPF_PERF_EVENT_CONT)
7948 return ret;
7949 return 0;
7950}
7951
7952int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
7953{
7954 int i, cnt, err;
7955
7956 cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
7957 for (i = 0; i < cnt; i++) {
7958 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
7959
7960 err = perf_buffer__process_records(pb, cpu_buf);
7961 if (err) {
be18010e 7962 pr_warn("error while processing records: %d\n", err);
fb84b822
AN		 7963 return err;
7964 }
7965 }
7966 return cnt < 0 ? -errno : cnt;
7967}
7968
34be1646
SL		 7969struct bpf_prog_info_array_desc {
7970 int array_offset; /* e.g. offset of jited_prog_insns */
7971 int count_offset; /* e.g. offset of jited_prog_len */
7972 int size_offset; /* > 0: offset of rec size,
7973 * < 0: fix size of -size_offset
7974 */
7975};
7976
7977static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
7978 [BPF_PROG_INFO_JITED_INSNS] = {
7979 offsetof(struct bpf_prog_info, jited_prog_insns),
7980 offsetof(struct bpf_prog_info, jited_prog_len),
7981 -1,
7982 },
7983 [BPF_PROG_INFO_XLATED_INSNS] = {
7984 offsetof(struct bpf_prog_info, xlated_prog_insns),
7985 offsetof(struct bpf_prog_info, xlated_prog_len),
7986 -1,
7987 },
7988 [BPF_PROG_INFO_MAP_IDS] = {
7989 offsetof(struct bpf_prog_info, map_ids),
7990 offsetof(struct bpf_prog_info, nr_map_ids),
7991 -(int)sizeof(__u32),
7992 },
7993 [BPF_PROG_INFO_JITED_KSYMS] = {
7994 offsetof(struct bpf_prog_info, jited_ksyms),
7995 offsetof(struct bpf_prog_info, nr_jited_ksyms),
7996 -(int)sizeof(__u64),
7997 },
7998 [BPF_PROG_INFO_JITED_FUNC_LENS] = {
7999 offsetof(struct bpf_prog_info, jited_func_lens),
8000 offsetof(struct bpf_prog_info, nr_jited_func_lens),
8001 -(int)sizeof(__u32),
8002 },
8003 [BPF_PROG_INFO_FUNC_INFO] = {
8004 offsetof(struct bpf_prog_info, func_info),
8005 offsetof(struct bpf_prog_info, nr_func_info),
8006 offsetof(struct bpf_prog_info, func_info_rec_size),
8007 },
8008 [BPF_PROG_INFO_LINE_INFO] = {
8009 offsetof(struct bpf_prog_info, line_info),
8010 offsetof(struct bpf_prog_info, nr_line_info),
8011 offsetof(struct bpf_prog_info, line_info_rec_size),
8012 },
8013 [BPF_PROG_INFO_JITED_LINE_INFO] = {
8014 offsetof(struct bpf_prog_info, jited_line_info),
8015 offsetof(struct bpf_prog_info, nr_jited_line_info),
8016 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
8017 },
8018 [BPF_PROG_INFO_PROG_TAGS] = {
8019 offsetof(struct bpf_prog_info, prog_tags),
8020 offsetof(struct bpf_prog_info, nr_prog_tags),
8021 -(int)sizeof(__u8) * BPF_TAG_SIZE,
8022 },
8023
8024};
8025
8983b731
AN		 8026static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info,
8027 int offset)
34be1646
SL		 8028{
8029 __u32 *array = (__u32 *)info;
8030
8031 if (offset >= 0)
8032 return array[offset / sizeof(__u32)];
8033 return -(int)offset;
8034}
8035
8983b731
AN		 8036static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info,
8037 int offset)
34be1646
SL		 8038{
8039 __u64 *array = (__u64 *)info;
8040
8041 if (offset >= 0)
8042 return array[offset / sizeof(__u64)];
8043 return -(int)offset;
8044}
8045
8046static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
8047 __u32 val)
8048{
8049 __u32 *array = (__u32 *)info;
8050
8051 if (offset >= 0)
8052 array[offset / sizeof(__u32)] = val;
8053}
8054
8055static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
8056 __u64 val)
8057{
8058 __u64 *array = (__u64 *)info;
8059
8060 if (offset >= 0)
8061 array[offset / sizeof(__u64)] = val;
8062}
8063
8064struct bpf_prog_info_linear *
8065bpf_program__get_prog_info_linear(int fd, __u64 arrays)
8066{
8067 struct bpf_prog_info_linear *info_linear;
8068 struct bpf_prog_info info = {};
8069 __u32 info_len = sizeof(info);
8070 __u32 data_len = 0;
8071 int i, err;
8072 void *ptr;
8073
8074 if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
8075 return ERR_PTR(-EINVAL);
8076
8077 /* step 1: get array dimensions */
8078 err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
8079 if (err) {
8080 pr_debug("can't get prog info: %s", strerror(errno));
8081 return ERR_PTR(-EFAULT);
8082 }
8083
8084 /* step 2: calculate total size of all arrays */
8085 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8086 bool include_array = (arrays & (1UL << i)) > 0;
8087 struct bpf_prog_info_array_desc *desc;
8088 __u32 count, size;
8089
8090 desc = bpf_prog_info_array_desc + i;
8091
8092 /* kernel is too old to support this field */
8093 if (info_len < desc->array_offset + sizeof(__u32) ||
8094 info_len < desc->count_offset + sizeof(__u32) ||
8095 (desc->size_offset > 0 && info_len < desc->size_offset))
8096 include_array = false;
8097
8098 if (!include_array) {
8099 arrays &= ~(1UL << i); /* clear the bit */
8100 continue;
8101 }
8102
8103 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
8104 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
8105
8106 data_len += count * size;
8107 }
8108
8109 /* step 3: allocate continuous memory */
8110 data_len = roundup(data_len, sizeof(__u64));
8111 info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
8112 if (!info_linear)
8113 return ERR_PTR(-ENOMEM);
8114
8115 /* step 4: fill data to info_linear->info */
8116 info_linear->arrays = arrays;
8117 memset(&info_linear->info, 0, sizeof(info));
8118 ptr = info_linear->data;
8119
8120 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8121 struct bpf_prog_info_array_desc *desc;
8122 __u32 count, size;
8123
8124 if ((arrays & (1UL << i)) == 0)
8125 continue;
8126
8127 desc = bpf_prog_info_array_desc + i;
8128 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
8129 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
8130 bpf_prog_info_set_offset_u32(&info_linear->info,
8131 desc->count_offset, count);
8132 bpf_prog_info_set_offset_u32(&info_linear->info,
8133 desc->size_offset, size);
8134 bpf_prog_info_set_offset_u64(&info_linear->info,
8135 desc->array_offset,
8136 ptr_to_u64(ptr));
8137 ptr += count * size;
8138 }
8139
8140 /* step 5: call syscall again to get required arrays */
8141 err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
8142 if (err) {
8143 pr_debug("can't get prog info: %s", strerror(errno));
8144 free(info_linear);
8145 return ERR_PTR(-EFAULT);
8146 }
8147
8148 /* step 6: verify the data */
8149 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8150 struct bpf_prog_info_array_desc *desc;
8151 __u32 v1, v2;
8152
8153 if ((arrays & (1UL << i)) == 0)
8154 continue;
8155
8156 desc = bpf_prog_info_array_desc + i;
8157 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
8158 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
8159 desc->count_offset);
8160 if (v1 != v2)
be18010e 8161 pr_warn("%s: mismatch in element count\n", __func__);
34be1646
SL		 8162
8163 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
8164 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
8165 desc->size_offset);
8166 if (v1 != v2)
be18010e 8167 pr_warn("%s: mismatch in rec size\n", __func__);
34be1646
SL		 8168 }
8169
8170 /* step 7: update info_len and data_len */
8171 info_linear->info_len = sizeof(struct bpf_prog_info);
8172 info_linear->data_len = data_len;
8173
8174 return info_linear;
8175}
8176
8177void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
8178{
8179 int i;
8180
8181 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8182 struct bpf_prog_info_array_desc *desc;
8183 __u64 addr, offs;
8184
8185 if ((info_linear->arrays & (1UL << i)) == 0)
8186 continue;
8187
8188 desc = bpf_prog_info_array_desc + i;
8189 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
8190 desc->array_offset);
8191 offs = addr - ptr_to_u64(info_linear->data);
8192 bpf_prog_info_set_offset_u64(&info_linear->info,
8193 desc->array_offset, offs);
8194 }
8195}
8196
8197void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
8198{
8199 int i;
8200
8201 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8202 struct bpf_prog_info_array_desc *desc;
8203 __u64 addr, offs;
8204
8205 if ((info_linear->arrays & (1UL << i)) == 0)
8206 continue;
8207
8208 desc = bpf_prog_info_array_desc + i;
8209 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
8210 desc->array_offset);
8211 addr = offs + ptr_to_u64(info_linear->data);
8212 bpf_prog_info_set_offset_u64(&info_linear->info,
8213 desc->array_offset, addr);
8214 }
8215}
6446b315 8216
ff26ce5c
EC		 8217int bpf_program__set_attach_target(struct bpf_program *prog,
8218 int attach_prog_fd,
8219 const char *attach_func_name)
8220{
8221 int btf_id;
8222
8223 if (!prog || attach_prog_fd < 0 || !attach_func_name)
8224 return -EINVAL;
8225
8226 if (attach_prog_fd)
8227 btf_id = libbpf_find_prog_btf_id(attach_func_name,
8228 attach_prog_fd);
8229 else
8230 btf_id = __find_vmlinux_btf_id(prog->obj->btf_vmlinux,
8231 attach_func_name,
8232 prog->expected_attach_type);
8233
8234 if (btf_id < 0)
8235 return btf_id;
8236
8237 prog->attach_btf_id = btf_id;
8238 prog->attach_prog_fd = attach_prog_fd;
8239 return 0;
8240}
8241
6803ee25 8242int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
6446b315 8243{
6803ee25
AN		 8244 int err = 0, n, len, start, end = -1;
8245 bool *tmp;
6446b315 8246
6803ee25
AN		 8247 *mask = NULL;
8248 *mask_sz = 0;
8249
8250 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
8251 while (*s) {
8252 if (*s == ',' || *s == '\n') {
8253 s++;
8254 continue;
8255 }
8256 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
8257 if (n <= 0 || n > 2) {
8258 pr_warn("Failed to get CPU range %s: %d\n", s, n);
8259 err = -EINVAL;
8260 goto cleanup;
8261 } else if (n == 1) {
8262 end = start;
8263 }
8264 if (start < 0 || start > end) {
8265 pr_warn("Invalid CPU range [%d,%d] in %s\n",
8266 start, end, s);
8267 err = -EINVAL;
8268 goto cleanup;
8269 }
8270 tmp = realloc(*mask, end + 1);
8271 if (!tmp) {
8272 err = -ENOMEM;
8273 goto cleanup;
8274 }
8275 *mask = tmp;
8276 memset(tmp + *mask_sz, 0, start - *mask_sz);
8277 memset(tmp + start, 1, end - start + 1);
8278 *mask_sz = end + 1;
8279 s += len;
8280 }
8281 if (!*mask_sz) {
8282 pr_warn("Empty CPU range\n");
8283 return -EINVAL;
8284 }
8285 return 0;
8286cleanup:
8287 free(*mask);
8288 *mask = NULL;
8289 return err;
8290}
8291
8292int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
8293{
8294 int fd, err = 0, len;
8295 char buf[128];
6446b315
HL		 8296
8297 fd = open(fcpu, O_RDONLY);
8298 if (fd < 0) {
6803ee25
AN		 8299 err = -errno;
8300 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
8301 return err;
6446b315
HL		 8302 }
8303 len = read(fd, buf, sizeof(buf));
8304 close(fd);
8305 if (len <= 0) {
6803ee25
AN		 8306 err = len ? -errno : -EINVAL;
8307 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
8308 return err;
6446b315 8309 }
6803ee25
AN		 8310 if (len >= sizeof(buf)) {
8311 pr_warn("CPU mask is too big in file %s\n", fcpu);
8312 return -E2BIG;
6446b315
HL		 8313 }
8314 buf[len] = '\0';
8315
6803ee25
AN		 8316 return parse_cpu_mask_str(buf, mask, mask_sz);
8317}
8318
8319int libbpf_num_possible_cpus(void)
8320{
8321 static const char *fcpu = "/sys/devices/system/cpu/possible";
8322 static int cpus;
8323 int err, n, i, tmp_cpus;
8324 bool *mask;
8325
8326 tmp_cpus = READ_ONCE(cpus);
8327 if (tmp_cpus > 0)
8328 return tmp_cpus;
8329
8330 err = parse_cpu_mask_file(fcpu, &mask, &n);
8331 if (err)
8332 return err;
8333
8334 tmp_cpus = 0;
8335 for (i = 0; i < n; i++) {
8336 if (mask[i])
8337 tmp_cpus++;
6446b315 8338 }
6803ee25 8339 free(mask);
56fbc241
TC		 8340
8341 WRITE_ONCE(cpus, tmp_cpus);
8342 return tmp_cpus;
6446b315 8343}
d66562fb
AN		 8344
8345int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
8346 const struct bpf_object_open_opts *opts)
8347{
8348 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
8349 .object_name = s->name,
8350 );
8351 struct bpf_object *obj;
8352 int i;
8353
8354 /* Attempt to preserve opts->object_name, unless overriden by user
8355 * explicitly. Overwriting object name for skeletons is discouraged,
8356 * as it breaks global data maps, because they contain object name
8357 * prefix as their own map name prefix. When skeleton is generated,
8358 * bpftool is making an assumption that this name will stay the same.
8359 */
8360 if (opts) {
8361 memcpy(&skel_opts, opts, sizeof(*opts));
8362 if (!opts->object_name)
8363 skel_opts.object_name = s->name;
8364 }
8365
8366 obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
8367 if (IS_ERR(obj)) {
8368 pr_warn("failed to initialize skeleton BPF object '%s': %ld\n",
8369 s->name, PTR_ERR(obj));
8370 return PTR_ERR(obj);
8371 }
8372
8373 *s->obj = obj;
8374
8375 for (i = 0; i < s->map_cnt; i++) {
8376 struct bpf_map **map = s->maps[i].map;
8377 const char *name = s->maps[i].name;
8378 void **mmaped = s->maps[i].mmaped;
8379
8380 *map = bpf_object__find_map_by_name(obj, name);
8381 if (!*map) {
8382 pr_warn("failed to find skeleton map '%s'\n", name);
8383 return -ESRCH;
8384 }
8385
2ad97d47 8386 /* externs shouldn't be pre-setup from user code */
81bfdd08 8387 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
d66562fb
AN		 8388 *mmaped = (*map)->mmaped;
8389 }
8390
8391 for (i = 0; i < s->prog_cnt; i++) {
8392 struct bpf_program **prog = s->progs[i].prog;
8393 const char *name = s->progs[i].name;
8394
8395 *prog = bpf_object__find_program_by_name(obj, name);
8396 if (!*prog) {
8397 pr_warn("failed to find skeleton program '%s'\n", name);
8398 return -ESRCH;
8399 }
8400 }
8401
8402 return 0;
8403}
8404
8405int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
8406{
8407 int i, err;
8408
8409 err = bpf_object__load(*s->obj);
8410 if (err) {
8411 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
8412 return err;
8413 }
8414
8415 for (i = 0; i < s->map_cnt; i++) {
8416 struct bpf_map *map = *s->maps[i].map;
8417 size_t mmap_sz = bpf_map_mmap_sz(map);
8418 int prot, map_fd = bpf_map__fd(map);
8419 void **mmaped = s->maps[i].mmaped;
d66562fb
AN		 8420
8421 if (!mmaped)
8422 continue;
8423
8424 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
8425 *mmaped = NULL;
8426 continue;
8427 }
8428
8429 if (map->def.map_flags & BPF_F_RDONLY_PROG)
8430 prot = PROT_READ;
8431 else
8432 prot = PROT_READ | PROT_WRITE;
8433
8434 /* Remap anonymous mmap()-ed "map initialization image" as
8435 * a BPF map-backed mmap()-ed memory, but preserving the same
8436 * memory address. This will cause kernel to change process'
8437 * page table to point to a different piece of kernel memory,
8438 * but from userspace point of view memory address (and its
8439 * contents, being identical at this point) will stay the
8440 * same. This mapping will be released by bpf_object__close()
8441 * as per normal clean up procedure, so we don't need to worry
8442 * about it from skeleton's clean up perspective.
8443 */
2ad97d47
AN		 8444 *mmaped = mmap(map->mmaped, mmap_sz, prot,
8445 MAP_SHARED | MAP_FIXED, map_fd, 0);
8446 if (*mmaped == MAP_FAILED) {
d66562fb
AN		 8447 err = -errno;
8448 *mmaped = NULL;
8449 pr_warn("failed to re-mmap() map '%s': %d\n",
8450 bpf_map__name(map), err);
8451 return err;
8452 }
8453 }
8454
8455 return 0;
8456}
8457
8458int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
8459{
8460 int i;
8461
8462 for (i = 0; i < s->prog_cnt; i++) {
8463 struct bpf_program *prog = *s->progs[i].prog;
8464 struct bpf_link **link = s->progs[i].link;
8465 const struct bpf_sec_def *sec_def;
8466 const char *sec_name = bpf_program__title(prog, false);
8467
8468 sec_def = find_sec_def(sec_name);
8469 if (!sec_def || !sec_def->attach_fn)
8470 continue;
8471
8472 *link = sec_def->attach_fn(sec_def, prog);
8473 if (IS_ERR(*link)) {
8474 pr_warn("failed to auto-attach program '%s': %ld\n",
8475 bpf_program__name(prog), PTR_ERR(*link));
8476 return PTR_ERR(*link);
8477 }
8478 }
8479
8480 return 0;
8481}
8482
8483void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
8484{
8485 int i;
8486
8487 for (i = 0; i < s->prog_cnt; i++) {
8488 struct bpf_link **link = s->progs[i].link;
8489
8490 if (!IS_ERR_OR_NULL(*link))
8491 bpf_link__destroy(*link);
8492 *link = NULL;
8493 }
8494}
8495
8496void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
8497{
8498 if (s->progs)
8499 bpf_object__detach_skeleton(s);
8500 if (s->obj)
8501 bpf_object__close(*s->obj);
8502 free(s->maps);
8503 free(s->progs);
8504 free(s);
8505}
Linux 6.x block layer and io_uring tree(s)