?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
treewide: remove redundant IS_ERR() before error code check
[linux/fpc-iii.git] / tools / lib / bpf / libbpf.c
blob514b1a524abbc0ff49bfe25fa2244eb50b860e1c
1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2
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.
9 * Copyright (C) 2017 Nicira, Inc.
10 * Copyright (C) 2019 Isovalent, Inc.
11 */
12
13 #ifndef _GNU_SOURCE
14 #define _GNU_SOURCE
15 #endif
16 #include <stdlib.h>
17 #include <stdio.h>
18 #include <stdarg.h>
19 #include <libgen.h>
20 #include <inttypes.h>
21 #include <limits.h>
22 #include <string.h>
23 #include <unistd.h>
24 #include <endian.h>
25 #include <fcntl.h>
26 #include <errno.h>
27 #include <asm/unistd.h>
28 #include <linux/err.h>
29 #include <linux/kernel.h>
30 #include <linux/bpf.h>
31 #include <linux/btf.h>
32 #include <linux/filter.h>
33 #include <linux/list.h>
34 #include <linux/limits.h>
35 #include <linux/perf_event.h>
36 #include <linux/ring_buffer.h>
37 #include <linux/version.h>
38 #include <sys/epoll.h>
39 #include <sys/ioctl.h>
40 #include <sys/mman.h>
41 #include <sys/stat.h>
42 #include <sys/types.h>
43 #include <sys/vfs.h>
44 #include <sys/utsname.h>
45 #include <sys/resource.h>
46 #include <tools/libc_compat.h>
47 #include <libelf.h>
48 #include <gelf.h>
49 #include <zlib.h>
50
51 #include "libbpf.h"
52 #include "bpf.h"
53 #include "btf.h"
54 #include "str_error.h"
55 #include "libbpf_internal.h"
56 #include "hashmap.h"
57
58 /* make sure libbpf doesn't use kernel-only integer typedefs */
59 #pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64
60
61 #ifndef EM_BPF
62 #define EM_BPF 247
63 #endif
64
65 #ifndef BPF_FS_MAGIC
66 #define BPF_FS_MAGIC 0xcafe4a11
67 #endif
68
69 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
70 * compilation if user enables corresponding warning. Disable it explicitly.
71 */
72 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
73
74 #define __printf(a, b) __attribute__((format(printf, a, b)))
75
76 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
77 static struct bpf_program *bpf_object__find_prog_by_idx(struct bpf_object *obj,
78 int idx);
79 static const struct btf_type *
80 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id);
81
82 static int __base_pr(enum libbpf_print_level level, const char *format,
83 va_list args)
84 {
85 if (level == LIBBPF_DEBUG)
86 return 0;
87
88 return vfprintf(stderr, format, args);
89 }
90
91 static libbpf_print_fn_t __libbpf_pr = __base_pr;
92
93 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
94 {
95 libbpf_print_fn_t old_print_fn = __libbpf_pr;
96
97 __libbpf_pr = fn;
98 return old_print_fn;
99 }
100
101 __printf(2, 3)
102 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
103 {
104 va_list args;
105
106 if (!__libbpf_pr)
107 return;
108
109 va_start(args, format);
110 __libbpf_pr(level, format, args);
111 va_end(args);
112 }
113
114 static void pr_perm_msg(int err)
115 {
116 struct rlimit limit;
117 char buf[100];
118
119 if (err != -EPERM || geteuid() != 0)
120 return;
121
122 err = getrlimit(RLIMIT_MEMLOCK, &limit);
123 if (err)
124 return;
125
126 if (limit.rlim_cur == RLIM_INFINITY)
127 return;
128
129 if (limit.rlim_cur < 1024)
130 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
131 else if (limit.rlim_cur < 1024*1024)
132 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
133 else
134 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
135
136 pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
137 buf);
138 }
139
140 #define STRERR_BUFSIZE 128
141
142 /* Copied from tools/perf/util/util.h */
143 #ifndef zfree
144 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
145 #endif
146
147 #ifndef zclose
148 # define zclose(fd) ({ \
149 int ___err = 0; \
150 if ((fd) >= 0) \
151 ___err = close((fd)); \
152 fd = -1; \
153 ___err; })
154 #endif
155
156 #ifdef HAVE_LIBELF_MMAP_SUPPORT
157 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
158 #else
159 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
160 #endif
161
162 static inline __u64 ptr_to_u64(const void *ptr)
163 {
164 return (__u64) (unsigned long) ptr;
165 }
166
167 struct bpf_capabilities {
168 /* v4.14: kernel support for program & map names. */
169 __u32 name:1;
170 /* v5.2: kernel support for global data sections. */
171 __u32 global_data:1;
172 /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
173 __u32 btf_func:1;
174 /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
175 __u32 btf_datasec:1;
176 /* BPF_F_MMAPABLE is supported for arrays */
177 __u32 array_mmap:1;
178 /* BTF_FUNC_GLOBAL is supported */
179 __u32 btf_func_global:1;
180 };
181
182 enum reloc_type {
183 RELO_LD64,
184 RELO_CALL,
185 RELO_DATA,
186 RELO_EXTERN,
187 };
188
189 struct reloc_desc {
190 enum reloc_type type;
191 int insn_idx;
192 int map_idx;
193 int sym_off;
194 };
195
196 /*
197 * bpf_prog should be a better name but it has been used in
198 * linux/filter.h.
199 */
200 struct bpf_program {
201 /* Index in elf obj file, for relocation use. */
202 int idx;
203 char *name;
204 int prog_ifindex;
205 char *section_name;
206 /* section_name with / replaced by _; makes recursive pinning
207 * in bpf_object__pin_programs easier
208 */
209 char *pin_name;
210 struct bpf_insn *insns;
211 size_t insns_cnt, main_prog_cnt;
212 enum bpf_prog_type type;
213
214 struct reloc_desc *reloc_desc;
215 int nr_reloc;
216 int log_level;
217
218 struct {
219 int nr;
220 int *fds;
221 } instances;
222 bpf_program_prep_t preprocessor;
223
224 struct bpf_object *obj;
225 void *priv;
226 bpf_program_clear_priv_t clear_priv;
227
228 enum bpf_attach_type expected_attach_type;
229 __u32 attach_btf_id;
230 __u32 attach_prog_fd;
231 void *func_info;
232 __u32 func_info_rec_size;
233 __u32 func_info_cnt;
234
235 struct bpf_capabilities *caps;
236
237 void *line_info;
238 __u32 line_info_rec_size;
239 __u32 line_info_cnt;
240 __u32 prog_flags;
241 };
242
243 struct bpf_struct_ops {
244 const char *tname;
245 const struct btf_type *type;
246 struct bpf_program **progs;
247 __u32 *kern_func_off;
248 /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
249 void *data;
250 /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
251 * btf_vmlinux's format.
252 * struct bpf_struct_ops_tcp_congestion_ops {
253 * [... some other kernel fields ...]
254 * struct tcp_congestion_ops data;
255 * }
256 * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
257 * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
258 * from "data".
259 */
260 void *kern_vdata;
261 __u32 type_id;
262 };
263
264 #define DATA_SEC ".data"
265 #define BSS_SEC ".bss"
266 #define RODATA_SEC ".rodata"
267 #define KCONFIG_SEC ".kconfig"
268 #define STRUCT_OPS_SEC ".struct_ops"
269
270 enum libbpf_map_type {
271 LIBBPF_MAP_UNSPEC,
272 LIBBPF_MAP_DATA,
273 LIBBPF_MAP_BSS,
274 LIBBPF_MAP_RODATA,
275 LIBBPF_MAP_KCONFIG,
276 };
277
278 static const char * const libbpf_type_to_btf_name[] = {
279 [LIBBPF_MAP_DATA] = DATA_SEC,
280 [LIBBPF_MAP_BSS] = BSS_SEC,
281 [LIBBPF_MAP_RODATA] = RODATA_SEC,
282 [LIBBPF_MAP_KCONFIG] = KCONFIG_SEC,
283 };
284
285 struct bpf_map {
286 char *name;
287 int fd;
288 int sec_idx;
289 size_t sec_offset;
290 int map_ifindex;
291 int inner_map_fd;
292 struct bpf_map_def def;
293 __u32 btf_key_type_id;
294 __u32 btf_value_type_id;
295 __u32 btf_vmlinux_value_type_id;
296 void *priv;
297 bpf_map_clear_priv_t clear_priv;
298 enum libbpf_map_type libbpf_type;
299 void *mmaped;
300 struct bpf_struct_ops *st_ops;
301 char *pin_path;
302 bool pinned;
303 bool reused;
304 };
305
306 enum extern_type {
307 EXT_UNKNOWN,
308 EXT_CHAR,
309 EXT_BOOL,
310 EXT_INT,
311 EXT_TRISTATE,
312 EXT_CHAR_ARR,
313 };
314
315 struct extern_desc {
316 const char *name;
317 int sym_idx;
318 int btf_id;
319 enum extern_type type;
320 int sz;
321 int align;
322 int data_off;
323 bool is_signed;
324 bool is_weak;
325 bool is_set;
326 };
327
328 static LIST_HEAD(bpf_objects_list);
329
330 struct bpf_object {
331 char name[BPF_OBJ_NAME_LEN];
332 char license[64];
333 __u32 kern_version;
334
335 struct bpf_program *programs;
336 size_t nr_programs;
337 struct bpf_map *maps;
338 size_t nr_maps;
339 size_t maps_cap;
340
341 char *kconfig;
342 struct extern_desc *externs;
343 int nr_extern;
344 int kconfig_map_idx;
345
346 bool loaded;
347 bool has_pseudo_calls;
348
349 /*
350 * Information when doing elf related work. Only valid if fd
351 * is valid.
352 */
353 struct {
354 int fd;
355 const void *obj_buf;
356 size_t obj_buf_sz;
357 Elf *elf;
358 GElf_Ehdr ehdr;
359 Elf_Data *symbols;
360 Elf_Data *data;
361 Elf_Data *rodata;
362 Elf_Data *bss;
363 Elf_Data *st_ops_data;
364 size_t strtabidx;
365 struct {
366 GElf_Shdr shdr;
367 Elf_Data *data;
368 } *reloc_sects;
369 int nr_reloc_sects;
370 int maps_shndx;
371 int btf_maps_shndx;
372 int text_shndx;
373 int symbols_shndx;
374 int data_shndx;
375 int rodata_shndx;
376 int bss_shndx;
377 int st_ops_shndx;
378 } efile;
379 /*
380 * All loaded bpf_object is linked in a list, which is
381 * hidden to caller. bpf_objects__<func> handlers deal with
382 * all objects.
383 */
384 struct list_head list;
385
386 struct btf *btf;
387 /* Parse and load BTF vmlinux if any of the programs in the object need
388 * it at load time.
389 */
390 struct btf *btf_vmlinux;
391 struct btf_ext *btf_ext;
392
393 void *priv;
394 bpf_object_clear_priv_t clear_priv;
395
396 struct bpf_capabilities caps;
397
398 char path[];
399 };
400 #define obj_elf_valid(o) ((o)->efile.elf)
401
402 void bpf_program__unload(struct bpf_program *prog)
403 {
404 int i;
405
406 if (!prog)
407 return;
408
409 /*
410 * If the object is opened but the program was never loaded,
411 * it is possible that prog->instances.nr == -1.
412 */
413 if (prog->instances.nr > 0) {
414 for (i = 0; i < prog->instances.nr; i++)
415 zclose(prog->instances.fds[i]);
416 } else if (prog->instances.nr != -1) {
417 pr_warn("Internal error: instances.nr is %d\n",
418 prog->instances.nr);
419 }
420
421 prog->instances.nr = -1;
422 zfree(&prog->instances.fds);
423
424 zfree(&prog->func_info);
425 zfree(&prog->line_info);
426 }
427
428 static void bpf_program__exit(struct bpf_program *prog)
429 {
430 if (!prog)
431 return;
432
433 if (prog->clear_priv)
434 prog->clear_priv(prog, prog->priv);
435
436 prog->priv = NULL;
437 prog->clear_priv = NULL;
438
439 bpf_program__unload(prog);
440 zfree(&prog->name);
441 zfree(&prog->section_name);
442 zfree(&prog->pin_name);
443 zfree(&prog->insns);
444 zfree(&prog->reloc_desc);
445
446 prog->nr_reloc = 0;
447 prog->insns_cnt = 0;
448 prog->idx = -1;
449 }
450
451 static char *__bpf_program__pin_name(struct bpf_program *prog)
452 {
453 char *name, *p;
454
455 name = p = strdup(prog->section_name);
456 while ((p = strchr(p, '/')))
457 *p = '_';
458
459 return name;
460 }
461
462 static int
463 bpf_program__init(void *data, size_t size, char *section_name, int idx,
464 struct bpf_program *prog)
465 {
466 const size_t bpf_insn_sz = sizeof(struct bpf_insn);
467
468 if (size == 0 || size % bpf_insn_sz) {
469 pr_warn("corrupted section '%s', size: %zu\n",
470 section_name, size);
471 return -EINVAL;
472 }
473
474 memset(prog, 0, sizeof(*prog));
475
476 prog->section_name = strdup(section_name);
477 if (!prog->section_name) {
478 pr_warn("failed to alloc name for prog under section(%d) %s\n",
479 idx, section_name);
480 goto errout;
481 }
482
483 prog->pin_name = __bpf_program__pin_name(prog);
484 if (!prog->pin_name) {
485 pr_warn("failed to alloc pin name for prog under section(%d) %s\n",
486 idx, section_name);
487 goto errout;
488 }
489
490 prog->insns = malloc(size);
491 if (!prog->insns) {
492 pr_warn("failed to alloc insns for prog under section %s\n",
493 section_name);
494 goto errout;
495 }
496 prog->insns_cnt = size / bpf_insn_sz;
497 memcpy(prog->insns, data, size);
498 prog->idx = idx;
499 prog->instances.fds = NULL;
500 prog->instances.nr = -1;
501 prog->type = BPF_PROG_TYPE_UNSPEC;
502
503 return 0;
504 errout:
505 bpf_program__exit(prog);
506 return -ENOMEM;
507 }
508
509 static int
510 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
511 char *section_name, int idx)
512 {
513 struct bpf_program prog, *progs;
514 int nr_progs, err;
515
516 err = bpf_program__init(data, size, section_name, idx, &prog);
517 if (err)
518 return err;
519
520 prog.caps = &obj->caps;
521 progs = obj->programs;
522 nr_progs = obj->nr_programs;
523
524 progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
525 if (!progs) {
526 /*
527 * In this case the original obj->programs
528 * is still valid, so don't need special treat for
529 * bpf_close_object().
530 */
531 pr_warn("failed to alloc a new program under section '%s'\n",
532 section_name);
533 bpf_program__exit(&prog);
534 return -ENOMEM;
535 }
536
537 pr_debug("found program %s\n", prog.section_name);
538 obj->programs = progs;
539 obj->nr_programs = nr_progs + 1;
540 prog.obj = obj;
541 progs[nr_progs] = prog;
542 return 0;
543 }
544
545 static int
546 bpf_object__init_prog_names(struct bpf_object *obj)
547 {
548 Elf_Data *symbols = obj->efile.symbols;
549 struct bpf_program *prog;
550 size_t pi, si;
551
552 for (pi = 0; pi < obj->nr_programs; pi++) {
553 const char *name = NULL;
554
555 prog = &obj->programs[pi];
556
557 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
558 si++) {
559 GElf_Sym sym;
560
561 if (!gelf_getsym(symbols, si, &sym))
562 continue;
563 if (sym.st_shndx != prog->idx)
564 continue;
565 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
566 continue;
567
568 name = elf_strptr(obj->efile.elf,
569 obj->efile.strtabidx,
570 sym.st_name);
571 if (!name) {
572 pr_warn("failed to get sym name string for prog %s\n",
573 prog->section_name);
574 return -LIBBPF_ERRNO__LIBELF;
575 }
576 }
577
578 if (!name && prog->idx == obj->efile.text_shndx)
579 name = ".text";
580
581 if (!name) {
582 pr_warn("failed to find sym for prog %s\n",
583 prog->section_name);
584 return -EINVAL;
585 }
586
587 prog->name = strdup(name);
588 if (!prog->name) {
589 pr_warn("failed to allocate memory for prog sym %s\n",
590 name);
591 return -ENOMEM;
592 }
593 }
594
595 return 0;
596 }
597
598 static __u32 get_kernel_version(void)
599 {
600 __u32 major, minor, patch;
601 struct utsname info;
602
603 uname(&info);
604 if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
605 return 0;
606 return KERNEL_VERSION(major, minor, patch);
607 }
608
609 static const struct btf_member *
610 find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
611 {
612 struct btf_member *m;
613 int i;
614
615 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
616 if (btf_member_bit_offset(t, i) == bit_offset)
617 return m;
618 }
619
620 return NULL;
621 }
622
623 static const struct btf_member *
624 find_member_by_name(const struct btf *btf, const struct btf_type *t,
625 const char *name)
626 {
627 struct btf_member *m;
628 int i;
629
630 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
631 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
632 return m;
633 }
634
635 return NULL;
636 }
637
638 #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
639 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
640 const char *name, __u32 kind);
641
642 static int
643 find_struct_ops_kern_types(const struct btf *btf, const char *tname,
644 const struct btf_type **type, __u32 *type_id,
645 const struct btf_type **vtype, __u32 *vtype_id,
646 const struct btf_member **data_member)
647 {
648 const struct btf_type *kern_type, *kern_vtype;
649 const struct btf_member *kern_data_member;
650 __s32 kern_vtype_id, kern_type_id;
651 __u32 i;
652
653 kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
654 if (kern_type_id < 0) {
655 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
656 tname);
657 return kern_type_id;
658 }
659 kern_type = btf__type_by_id(btf, kern_type_id);
660
661 /* Find the corresponding "map_value" type that will be used
662 * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example,
663 * find "struct bpf_struct_ops_tcp_congestion_ops" from the
664 * btf_vmlinux.
665 */
666 kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
667 tname, BTF_KIND_STRUCT);
668 if (kern_vtype_id < 0) {
669 pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
670 STRUCT_OPS_VALUE_PREFIX, tname);
671 return kern_vtype_id;
672 }
673 kern_vtype = btf__type_by_id(btf, kern_vtype_id);
674
675 /* Find "struct tcp_congestion_ops" from
676 * struct bpf_struct_ops_tcp_congestion_ops {
677 * [ ... ]
678 * struct tcp_congestion_ops data;
679 * }
680 */
681 kern_data_member = btf_members(kern_vtype);
682 for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
683 if (kern_data_member->type == kern_type_id)
684 break;
685 }
686 if (i == btf_vlen(kern_vtype)) {
687 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
688 tname, STRUCT_OPS_VALUE_PREFIX, tname);
689 return -EINVAL;
690 }
691
692 *type = kern_type;
693 *type_id = kern_type_id;
694 *vtype = kern_vtype;
695 *vtype_id = kern_vtype_id;
696 *data_member = kern_data_member;
697
698 return 0;
699 }
700
701 static bool bpf_map__is_struct_ops(const struct bpf_map *map)
702 {
703 return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
704 }
705
706 /* Init the map's fields that depend on kern_btf */
707 static int bpf_map__init_kern_struct_ops(struct bpf_map *map,
708 const struct btf *btf,
709 const struct btf *kern_btf)
710 {
711 const struct btf_member *member, *kern_member, *kern_data_member;
712 const struct btf_type *type, *kern_type, *kern_vtype;
713 __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
714 struct bpf_struct_ops *st_ops;
715 void *data, *kern_data;
716 const char *tname;
717 int err;
718
719 st_ops = map->st_ops;
720 type = st_ops->type;
721 tname = st_ops->tname;
722 err = find_struct_ops_kern_types(kern_btf, tname,
723 &kern_type, &kern_type_id,
724 &kern_vtype, &kern_vtype_id,
725 &kern_data_member);
726 if (err)
727 return err;
728
729 pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
730 map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
731
732 map->def.value_size = kern_vtype->size;
733 map->btf_vmlinux_value_type_id = kern_vtype_id;
734
735 st_ops->kern_vdata = calloc(1, kern_vtype->size);
736 if (!st_ops->kern_vdata)
737 return -ENOMEM;
738
739 data = st_ops->data;
740 kern_data_off = kern_data_member->offset / 8;
741 kern_data = st_ops->kern_vdata + kern_data_off;
742
743 member = btf_members(type);
744 for (i = 0; i < btf_vlen(type); i++, member++) {
745 const struct btf_type *mtype, *kern_mtype;
746 __u32 mtype_id, kern_mtype_id;
747 void *mdata, *kern_mdata;
748 __s64 msize, kern_msize;
749 __u32 moff, kern_moff;
750 __u32 kern_member_idx;
751 const char *mname;
752
753 mname = btf__name_by_offset(btf, member->name_off);
754 kern_member = find_member_by_name(kern_btf, kern_type, mname);
755 if (!kern_member) {
756 pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
757 map->name, mname);
758 return -ENOTSUP;
759 }
760
761 kern_member_idx = kern_member - btf_members(kern_type);
762 if (btf_member_bitfield_size(type, i) ||
763 btf_member_bitfield_size(kern_type, kern_member_idx)) {
764 pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
765 map->name, mname);
766 return -ENOTSUP;
767 }
768
769 moff = member->offset / 8;
770 kern_moff = kern_member->offset / 8;
771
772 mdata = data + moff;
773 kern_mdata = kern_data + kern_moff;
774
775 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
776 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
777 &kern_mtype_id);
778 if (BTF_INFO_KIND(mtype->info) !=
779 BTF_INFO_KIND(kern_mtype->info)) {
780 pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
781 map->name, mname, BTF_INFO_KIND(mtype->info),
782 BTF_INFO_KIND(kern_mtype->info));
783 return -ENOTSUP;
784 }
785
786 if (btf_is_ptr(mtype)) {
787 struct bpf_program *prog;
788
789 mtype = skip_mods_and_typedefs(btf, mtype->type, &mtype_id);
790 kern_mtype = skip_mods_and_typedefs(kern_btf,
791 kern_mtype->type,
792 &kern_mtype_id);
793 if (!btf_is_func_proto(mtype) ||
794 !btf_is_func_proto(kern_mtype)) {
795 pr_warn("struct_ops init_kern %s: non func ptr %s is not supported\n",
796 map->name, mname);
797 return -ENOTSUP;
798 }
799
800 prog = st_ops->progs[i];
801 if (!prog) {
802 pr_debug("struct_ops init_kern %s: func ptr %s is not set\n",
803 map->name, mname);
804 continue;
805 }
806
807 prog->attach_btf_id = kern_type_id;
808 prog->expected_attach_type = kern_member_idx;
809
810 st_ops->kern_func_off[i] = kern_data_off + kern_moff;
811
812 pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
813 map->name, mname, prog->name, moff,
814 kern_moff);
815
816 continue;
817 }
818
819 msize = btf__resolve_size(btf, mtype_id);
820 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
821 if (msize < 0 || kern_msize < 0 || msize != kern_msize) {
822 pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
823 map->name, mname, (ssize_t)msize,
824 (ssize_t)kern_msize);
825 return -ENOTSUP;
826 }
827
828 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
829 map->name, mname, (unsigned int)msize,
830 moff, kern_moff);
831 memcpy(kern_mdata, mdata, msize);
832 }
833
834 return 0;
835 }
836
837 static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
838 {
839 struct bpf_map *map;
840 size_t i;
841 int err;
842
843 for (i = 0; i < obj->nr_maps; i++) {
844 map = &obj->maps[i];
845
846 if (!bpf_map__is_struct_ops(map))
847 continue;
848
849 err = bpf_map__init_kern_struct_ops(map, obj->btf,
850 obj->btf_vmlinux);
851 if (err)
852 return err;
853 }
854
855 return 0;
856 }
857
858 static int bpf_object__init_struct_ops_maps(struct bpf_object *obj)
859 {
860 const struct btf_type *type, *datasec;
861 const struct btf_var_secinfo *vsi;
862 struct bpf_struct_ops *st_ops;
863 const char *tname, *var_name;
864 __s32 type_id, datasec_id;
865 const struct btf *btf;
866 struct bpf_map *map;
867 __u32 i;
868
869 if (obj->efile.st_ops_shndx == -1)
870 return 0;
871
872 btf = obj->btf;
873 datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC,
874 BTF_KIND_DATASEC);
875 if (datasec_id < 0) {
876 pr_warn("struct_ops init: DATASEC %s not found\n",
877 STRUCT_OPS_SEC);
878 return -EINVAL;
879 }
880
881 datasec = btf__type_by_id(btf, datasec_id);
882 vsi = btf_var_secinfos(datasec);
883 for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
884 type = btf__type_by_id(obj->btf, vsi->type);
885 var_name = btf__name_by_offset(obj->btf, type->name_off);
886
887 type_id = btf__resolve_type(obj->btf, vsi->type);
888 if (type_id < 0) {
889 pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
890 vsi->type, STRUCT_OPS_SEC);
891 return -EINVAL;
892 }
893
894 type = btf__type_by_id(obj->btf, type_id);
895 tname = btf__name_by_offset(obj->btf, type->name_off);
896 if (!tname[0]) {
897 pr_warn("struct_ops init: anonymous type is not supported\n");
898 return -ENOTSUP;
899 }
900 if (!btf_is_struct(type)) {
901 pr_warn("struct_ops init: %s is not a struct\n", tname);
902 return -EINVAL;
903 }
904
905 map = bpf_object__add_map(obj);
906 if (IS_ERR(map))
907 return PTR_ERR(map);
908
909 map->sec_idx = obj->efile.st_ops_shndx;
910 map->sec_offset = vsi->offset;
911 map->name = strdup(var_name);
912 if (!map->name)
913 return -ENOMEM;
914
915 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
916 map->def.key_size = sizeof(int);
917 map->def.value_size = type->size;
918 map->def.max_entries = 1;
919
920 map->st_ops = calloc(1, sizeof(*map->st_ops));
921 if (!map->st_ops)
922 return -ENOMEM;
923 st_ops = map->st_ops;
924 st_ops->data = malloc(type->size);
925 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
926 st_ops->kern_func_off = malloc(btf_vlen(type) *
927 sizeof(*st_ops->kern_func_off));
928 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
929 return -ENOMEM;
930
931 if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) {
932 pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
933 var_name, STRUCT_OPS_SEC);
934 return -EINVAL;
935 }
936
937 memcpy(st_ops->data,
938 obj->efile.st_ops_data->d_buf + vsi->offset,
939 type->size);
940 st_ops->tname = tname;
941 st_ops->type = type;
942 st_ops->type_id = type_id;
943
944 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
945 tname, type_id, var_name, vsi->offset);
946 }
947
948 return 0;
949 }
950
951 static struct bpf_object *bpf_object__new(const char *path,
952 const void *obj_buf,
953 size_t obj_buf_sz,
954 const char *obj_name)
955 {
956 struct bpf_object *obj;
957 char *end;
958
959 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
960 if (!obj) {
961 pr_warn("alloc memory failed for %s\n", path);
962 return ERR_PTR(-ENOMEM);
963 }
964
965 strcpy(obj->path, path);
966 if (obj_name) {
967 strncpy(obj->name, obj_name, sizeof(obj->name) - 1);
968 obj->name[sizeof(obj->name) - 1] = 0;
969 } else {
970 /* Using basename() GNU version which doesn't modify arg. */
971 strncpy(obj->name, basename((void *)path),
972 sizeof(obj->name) - 1);
973 end = strchr(obj->name, '.');
974 if (end)
975 *end = 0;
976 }
977
978 obj->efile.fd = -1;
979 /*
980 * Caller of this function should also call
981 * bpf_object__elf_finish() after data collection to return
982 * obj_buf to user. If not, we should duplicate the buffer to
983 * avoid user freeing them before elf finish.
984 */
985 obj->efile.obj_buf = obj_buf;
986 obj->efile.obj_buf_sz = obj_buf_sz;
987 obj->efile.maps_shndx = -1;
988 obj->efile.btf_maps_shndx = -1;
989 obj->efile.data_shndx = -1;
990 obj->efile.rodata_shndx = -1;
991 obj->efile.bss_shndx = -1;
992 obj->efile.st_ops_shndx = -1;
993 obj->kconfig_map_idx = -1;
994
995 obj->kern_version = get_kernel_version();
996 obj->loaded = false;
997
998 INIT_LIST_HEAD(&obj->list);
999 list_add(&obj->list, &bpf_objects_list);
1000 return obj;
1001 }
1002
1003 static void bpf_object__elf_finish(struct bpf_object *obj)
1004 {
1005 if (!obj_elf_valid(obj))
1006 return;
1007
1008 if (obj->efile.elf) {
1009 elf_end(obj->efile.elf);
1010 obj->efile.elf = NULL;
1011 }
1012 obj->efile.symbols = NULL;
1013 obj->efile.data = NULL;
1014 obj->efile.rodata = NULL;
1015 obj->efile.bss = NULL;
1016 obj->efile.st_ops_data = NULL;
1017
1018 zfree(&obj->efile.reloc_sects);
1019 obj->efile.nr_reloc_sects = 0;
1020 zclose(obj->efile.fd);
1021 obj->efile.obj_buf = NULL;
1022 obj->efile.obj_buf_sz = 0;
1023 }
1024
1025 static int bpf_object__elf_init(struct bpf_object *obj)
1026 {
1027 int err = 0;
1028 GElf_Ehdr *ep;
1029
1030 if (obj_elf_valid(obj)) {
1031 pr_warn("elf init: internal error\n");
1032 return -LIBBPF_ERRNO__LIBELF;
1033 }
1034
1035 if (obj->efile.obj_buf_sz > 0) {
1036 /*
1037 * obj_buf should have been validated by
1038 * bpf_object__open_buffer().
1039 */
1040 obj->efile.elf = elf_memory((char *)obj->efile.obj_buf,
1041 obj->efile.obj_buf_sz);
1042 } else {
1043 obj->efile.fd = open(obj->path, O_RDONLY);
1044 if (obj->efile.fd < 0) {
1045 char errmsg[STRERR_BUFSIZE], *cp;
1046
1047 err = -errno;
1048 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
1049 pr_warn("failed to open %s: %s\n", obj->path, cp);
1050 return err;
1051 }
1052
1053 obj->efile.elf = elf_begin(obj->efile.fd,
1054 LIBBPF_ELF_C_READ_MMAP, NULL);
1055 }
1056
1057 if (!obj->efile.elf) {
1058 pr_warn("failed to open %s as ELF file\n", obj->path);
1059 err = -LIBBPF_ERRNO__LIBELF;
1060 goto errout;
1061 }
1062
1063 if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
1064 pr_warn("failed to get EHDR from %s\n", obj->path);
1065 err = -LIBBPF_ERRNO__FORMAT;
1066 goto errout;
1067 }
1068 ep = &obj->efile.ehdr;
1069
1070 /* Old LLVM set e_machine to EM_NONE */
1071 if (ep->e_type != ET_REL ||
1072 (ep->e_machine && ep->e_machine != EM_BPF)) {
1073 pr_warn("%s is not an eBPF object file\n", obj->path);
1074 err = -LIBBPF_ERRNO__FORMAT;
1075 goto errout;
1076 }
1077
1078 return 0;
1079 errout:
1080 bpf_object__elf_finish(obj);
1081 return err;
1082 }
1083
1084 static int bpf_object__check_endianness(struct bpf_object *obj)
1085 {
1086 #if __BYTE_ORDER == __LITTLE_ENDIAN
1087 if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
1088 return 0;
1089 #elif __BYTE_ORDER == __BIG_ENDIAN
1090 if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
1091 return 0;
1092 #else
1093 # error "Unrecognized __BYTE_ORDER__"
1094 #endif
1095 pr_warn("endianness mismatch.\n");
1096 return -LIBBPF_ERRNO__ENDIAN;
1097 }
1098
1099 static int
1100 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1101 {
1102 memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
1103 pr_debug("license of %s is %s\n", obj->path, obj->license);
1104 return 0;
1105 }
1106
1107 static int
1108 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1109 {
1110 __u32 kver;
1111
1112 if (size != sizeof(kver)) {
1113 pr_warn("invalid kver section in %s\n", obj->path);
1114 return -LIBBPF_ERRNO__FORMAT;
1115 }
1116 memcpy(&kver, data, sizeof(kver));
1117 obj->kern_version = kver;
1118 pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1119 return 0;
1120 }
1121
1122 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1123 {
1124 if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1125 type == BPF_MAP_TYPE_HASH_OF_MAPS)
1126 return true;
1127 return false;
1128 }
1129
1130 static int bpf_object_search_section_size(const struct bpf_object *obj,
1131 const char *name, size_t *d_size)
1132 {
1133 const GElf_Ehdr *ep = &obj->efile.ehdr;
1134 Elf *elf = obj->efile.elf;
1135 Elf_Scn *scn = NULL;
1136 int idx = 0;
1137
1138 while ((scn = elf_nextscn(elf, scn)) != NULL) {
1139 const char *sec_name;
1140 Elf_Data *data;
1141 GElf_Shdr sh;
1142
1143 idx++;
1144 if (gelf_getshdr(scn, &sh) != &sh) {
1145 pr_warn("failed to get section(%d) header from %s\n",
1146 idx, obj->path);
1147 return -EIO;
1148 }
1149
1150 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1151 if (!sec_name) {
1152 pr_warn("failed to get section(%d) name from %s\n",
1153 idx, obj->path);
1154 return -EIO;
1155 }
1156
1157 if (strcmp(name, sec_name))
1158 continue;
1159
1160 data = elf_getdata(scn, 0);
1161 if (!data) {
1162 pr_warn("failed to get section(%d) data from %s(%s)\n",
1163 idx, name, obj->path);
1164 return -EIO;
1165 }
1166
1167 *d_size = data->d_size;
1168 return 0;
1169 }
1170
1171 return -ENOENT;
1172 }
1173
1174 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
1175 __u32 *size)
1176 {
1177 int ret = -ENOENT;
1178 size_t d_size;
1179
1180 *size = 0;
1181 if (!name) {
1182 return -EINVAL;
1183 } else if (!strcmp(name, DATA_SEC)) {
1184 if (obj->efile.data)
1185 *size = obj->efile.data->d_size;
1186 } else if (!strcmp(name, BSS_SEC)) {
1187 if (obj->efile.bss)
1188 *size = obj->efile.bss->d_size;
1189 } else if (!strcmp(name, RODATA_SEC)) {
1190 if (obj->efile.rodata)
1191 *size = obj->efile.rodata->d_size;
1192 } else if (!strcmp(name, STRUCT_OPS_SEC)) {
1193 if (obj->efile.st_ops_data)
1194 *size = obj->efile.st_ops_data->d_size;
1195 } else {
1196 ret = bpf_object_search_section_size(obj, name, &d_size);
1197 if (!ret)
1198 *size = d_size;
1199 }
1200
1201 return *size ? 0 : ret;
1202 }
1203
1204 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
1205 __u32 *off)
1206 {
1207 Elf_Data *symbols = obj->efile.symbols;
1208 const char *sname;
1209 size_t si;
1210
1211 if (!name || !off)
1212 return -EINVAL;
1213
1214 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
1215 GElf_Sym sym;
1216
1217 if (!gelf_getsym(symbols, si, &sym))
1218 continue;
1219 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
1220 GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
1221 continue;
1222
1223 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1224 sym.st_name);
1225 if (!sname) {
1226 pr_warn("failed to get sym name string for var %s\n",
1227 name);
1228 return -EIO;
1229 }
1230 if (strcmp(name, sname) == 0) {
1231 *off = sym.st_value;
1232 return 0;
1233 }
1234 }
1235
1236 return -ENOENT;
1237 }
1238
1239 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1240 {
1241 struct bpf_map *new_maps;
1242 size_t new_cap;
1243 int i;
1244
1245 if (obj->nr_maps < obj->maps_cap)
1246 return &obj->maps[obj->nr_maps++];
1247
1248 new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
1249 new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
1250 if (!new_maps) {
1251 pr_warn("alloc maps for object failed\n");
1252 return ERR_PTR(-ENOMEM);
1253 }
1254
1255 obj->maps_cap = new_cap;
1256 obj->maps = new_maps;
1257
1258 /* zero out new maps */
1259 memset(obj->maps + obj->nr_maps, 0,
1260 (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
1261 /*
1262 * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
1263 * when failure (zclose won't close negative fd)).
1264 */
1265 for (i = obj->nr_maps; i < obj->maps_cap; i++) {
1266 obj->maps[i].fd = -1;
1267 obj->maps[i].inner_map_fd = -1;
1268 }
1269
1270 return &obj->maps[obj->nr_maps++];
1271 }
1272
1273 static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1274 {
1275 long page_sz = sysconf(_SC_PAGE_SIZE);
1276 size_t map_sz;
1277
1278 map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries;
1279 map_sz = roundup(map_sz, page_sz);
1280 return map_sz;
1281 }
1282
1283 static char *internal_map_name(struct bpf_object *obj,
1284 enum libbpf_map_type type)
1285 {
1286 char map_name[BPF_OBJ_NAME_LEN];
1287 const char *sfx = libbpf_type_to_btf_name[type];
1288 int sfx_len = max((size_t)7, strlen(sfx));
1289 int pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1,
1290 strlen(obj->name));
1291
1292 snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1293 sfx_len, libbpf_type_to_btf_name[type]);
1294
1295 return strdup(map_name);
1296 }
1297
1298 static int
1299 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1300 int sec_idx, void *data, size_t data_sz)
1301 {
1302 struct bpf_map_def *def;
1303 struct bpf_map *map;
1304 int err;
1305
1306 map = bpf_object__add_map(obj);
1307 if (IS_ERR(map))
1308 return PTR_ERR(map);
1309
1310 map->libbpf_type = type;
1311 map->sec_idx = sec_idx;
1312 map->sec_offset = 0;
1313 map->name = internal_map_name(obj, type);
1314 if (!map->name) {
1315 pr_warn("failed to alloc map name\n");
1316 return -ENOMEM;
1317 }
1318
1319 def = &map->def;
1320 def->type = BPF_MAP_TYPE_ARRAY;
1321 def->key_size = sizeof(int);
1322 def->value_size = data_sz;
1323 def->max_entries = 1;
1324 def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1325 ? BPF_F_RDONLY_PROG : 0;
1326 def->map_flags |= BPF_F_MMAPABLE;
1327
1328 pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1329 map->name, map->sec_idx, map->sec_offset, def->map_flags);
1330
1331 map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
1332 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1333 if (map->mmaped == MAP_FAILED) {
1334 err = -errno;
1335 map->mmaped = NULL;
1336 pr_warn("failed to alloc map '%s' content buffer: %d\n",
1337 map->name, err);
1338 zfree(&map->name);
1339 return err;
1340 }
1341
1342 if (data)
1343 memcpy(map->mmaped, data, data_sz);
1344
1345 pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
1346 return 0;
1347 }
1348
1349 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1350 {
1351 int err;
1352
1353 /*
1354 * Populate obj->maps with libbpf internal maps.
1355 */
1356 if (obj->efile.data_shndx >= 0) {
1357 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
1358 obj->efile.data_shndx,
1359 obj->efile.data->d_buf,
1360 obj->efile.data->d_size);
1361 if (err)
1362 return err;
1363 }
1364 if (obj->efile.rodata_shndx >= 0) {
1365 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
1366 obj->efile.rodata_shndx,
1367 obj->efile.rodata->d_buf,
1368 obj->efile.rodata->d_size);
1369 if (err)
1370 return err;
1371 }
1372 if (obj->efile.bss_shndx >= 0) {
1373 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
1374 obj->efile.bss_shndx,
1375 NULL,
1376 obj->efile.bss->d_size);
1377 if (err)
1378 return err;
1379 }
1380 return 0;
1381 }
1382
1383
1384 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1385 const void *name)
1386 {
1387 int i;
1388
1389 for (i = 0; i < obj->nr_extern; i++) {
1390 if (strcmp(obj->externs[i].name, name) == 0)
1391 return &obj->externs[i];
1392 }
1393 return NULL;
1394 }
1395
1396 static int set_ext_value_tri(struct extern_desc *ext, void *ext_val,
1397 char value)
1398 {
1399 switch (ext->type) {
1400 case EXT_BOOL:
1401 if (value == 'm') {
1402 pr_warn("extern %s=%c should be tristate or char\n",
1403 ext->name, value);
1404 return -EINVAL;
1405 }
1406 *(bool *)ext_val = value == 'y' ? true : false;
1407 break;
1408 case EXT_TRISTATE:
1409 if (value == 'y')
1410 *(enum libbpf_tristate *)ext_val = TRI_YES;
1411 else if (value == 'm')
1412 *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1413 else /* value == 'n' */
1414 *(enum libbpf_tristate *)ext_val = TRI_NO;
1415 break;
1416 case EXT_CHAR:
1417 *(char *)ext_val = value;
1418 break;
1419 case EXT_UNKNOWN:
1420 case EXT_INT:
1421 case EXT_CHAR_ARR:
1422 default:
1423 pr_warn("extern %s=%c should be bool, tristate, or char\n",
1424 ext->name, value);
1425 return -EINVAL;
1426 }
1427 ext->is_set = true;
1428 return 0;
1429 }
1430
1431 static int set_ext_value_str(struct extern_desc *ext, char *ext_val,
1432 const char *value)
1433 {
1434 size_t len;
1435
1436 if (ext->type != EXT_CHAR_ARR) {
1437 pr_warn("extern %s=%s should char array\n", ext->name, value);
1438 return -EINVAL;
1439 }
1440
1441 len = strlen(value);
1442 if (value[len - 1] != '"') {
1443 pr_warn("extern '%s': invalid string config '%s'\n",
1444 ext->name, value);
1445 return -EINVAL;
1446 }
1447
1448 /* strip quotes */
1449 len -= 2;
1450 if (len >= ext->sz) {
1451 pr_warn("extern '%s': long string config %s of (%zu bytes) truncated to %d bytes\n",
1452 ext->name, value, len, ext->sz - 1);
1453 len = ext->sz - 1;
1454 }
1455 memcpy(ext_val, value + 1, len);
1456 ext_val[len] = '\0';
1457 ext->is_set = true;
1458 return 0;
1459 }
1460
1461 static int parse_u64(const char *value, __u64 *res)
1462 {
1463 char *value_end;
1464 int err;
1465
1466 errno = 0;
1467 *res = strtoull(value, &value_end, 0);
1468 if (errno) {
1469 err = -errno;
1470 pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1471 return err;
1472 }
1473 if (*value_end) {
1474 pr_warn("failed to parse '%s' as integer completely\n", value);
1475 return -EINVAL;
1476 }
1477 return 0;
1478 }
1479
1480 static bool is_ext_value_in_range(const struct extern_desc *ext, __u64 v)
1481 {
1482 int bit_sz = ext->sz * 8;
1483
1484 if (ext->sz == 8)
1485 return true;
1486
1487 /* Validate that value stored in u64 fits in integer of `ext->sz`
1488 * bytes size without any loss of information. If the target integer
1489 * is signed, we rely on the following limits of integer type of
1490 * Y bits and subsequent transformation:
1491 *
1492 * -2^(Y-1) <= X <= 2^(Y-1) - 1
1493 * 0 <= X + 2^(Y-1) <= 2^Y - 1
1494 * 0 <= X + 2^(Y-1) < 2^Y
1495 *
1496 * For unsigned target integer, check that all the (64 - Y) bits are
1497 * zero.
1498 */
1499 if (ext->is_signed)
1500 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1501 else
1502 return (v >> bit_sz) == 0;
1503 }
1504
1505 static int set_ext_value_num(struct extern_desc *ext, void *ext_val,
1506 __u64 value)
1507 {
1508 if (ext->type != EXT_INT && ext->type != EXT_CHAR) {
1509 pr_warn("extern %s=%llu should be integer\n",
1510 ext->name, (unsigned long long)value);
1511 return -EINVAL;
1512 }
1513 if (!is_ext_value_in_range(ext, value)) {
1514 pr_warn("extern %s=%llu value doesn't fit in %d bytes\n",
1515 ext->name, (unsigned long long)value, ext->sz);
1516 return -ERANGE;
1517 }
1518 switch (ext->sz) {
1519 case 1: *(__u8 *)ext_val = value; break;
1520 case 2: *(__u16 *)ext_val = value; break;
1521 case 4: *(__u32 *)ext_val = value; break;
1522 case 8: *(__u64 *)ext_val = value; break;
1523 default:
1524 return -EINVAL;
1525 }
1526 ext->is_set = true;
1527 return 0;
1528 }
1529
1530 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1531 char *buf, void *data)
1532 {
1533 struct extern_desc *ext;
1534 char *sep, *value;
1535 int len, err = 0;
1536 void *ext_val;
1537 __u64 num;
1538
1539 if (strncmp(buf, "CONFIG_", 7))
1540 return 0;
1541
1542 sep = strchr(buf, '=');
1543 if (!sep) {
1544 pr_warn("failed to parse '%s': no separator\n", buf);
1545 return -EINVAL;
1546 }
1547
1548 /* Trim ending '\n' */
1549 len = strlen(buf);
1550 if (buf[len - 1] == '\n')
1551 buf[len - 1] = '\0';
1552 /* Split on '=' and ensure that a value is present. */
1553 *sep = '\0';
1554 if (!sep[1]) {
1555 *sep = '=';
1556 pr_warn("failed to parse '%s': no value\n", buf);
1557 return -EINVAL;
1558 }
1559
1560 ext = find_extern_by_name(obj, buf);
1561 if (!ext || ext->is_set)
1562 return 0;
1563
1564 ext_val = data + ext->data_off;
1565 value = sep + 1;
1566
1567 switch (*value) {
1568 case 'y': case 'n': case 'm':
1569 err = set_ext_value_tri(ext, ext_val, *value);
1570 break;
1571 case '"':
1572 err = set_ext_value_str(ext, ext_val, value);
1573 break;
1574 default:
1575 /* assume integer */
1576 err = parse_u64(value, &num);
1577 if (err) {
1578 pr_warn("extern %s=%s should be integer\n",
1579 ext->name, value);
1580 return err;
1581 }
1582 err = set_ext_value_num(ext, ext_val, num);
1583 break;
1584 }
1585 if (err)
1586 return err;
1587 pr_debug("extern %s=%s\n", ext->name, value);
1588 return 0;
1589 }
1590
1591 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
1592 {
1593 char buf[PATH_MAX];
1594 struct utsname uts;
1595 int len, err = 0;
1596 gzFile file;
1597
1598 uname(&uts);
1599 len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
1600 if (len < 0)
1601 return -EINVAL;
1602 else if (len >= PATH_MAX)
1603 return -ENAMETOOLONG;
1604
1605 /* gzopen also accepts uncompressed files. */
1606 file = gzopen(buf, "r");
1607 if (!file)
1608 file = gzopen("/proc/config.gz", "r");
1609
1610 if (!file) {
1611 pr_warn("failed to open system Kconfig\n");
1612 return -ENOENT;
1613 }
1614
1615 while (gzgets(file, buf, sizeof(buf))) {
1616 err = bpf_object__process_kconfig_line(obj, buf, data);
1617 if (err) {
1618 pr_warn("error parsing system Kconfig line '%s': %d\n",
1619 buf, err);
1620 goto out;
1621 }
1622 }
1623
1624 out:
1625 gzclose(file);
1626 return err;
1627 }
1628
1629 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
1630 const char *config, void *data)
1631 {
1632 char buf[PATH_MAX];
1633 int err = 0;
1634 FILE *file;
1635
1636 file = fmemopen((void *)config, strlen(config), "r");
1637 if (!file) {
1638 err = -errno;
1639 pr_warn("failed to open in-memory Kconfig: %d\n", err);
1640 return err;
1641 }
1642
1643 while (fgets(buf, sizeof(buf), file)) {
1644 err = bpf_object__process_kconfig_line(obj, buf, data);
1645 if (err) {
1646 pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
1647 buf, err);
1648 break;
1649 }
1650 }
1651
1652 fclose(file);
1653 return err;
1654 }
1655
1656 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
1657 {
1658 struct extern_desc *last_ext;
1659 size_t map_sz;
1660 int err;
1661
1662 if (obj->nr_extern == 0)
1663 return 0;
1664
1665 last_ext = &obj->externs[obj->nr_extern - 1];
1666 map_sz = last_ext->data_off + last_ext->sz;
1667
1668 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
1669 obj->efile.symbols_shndx,
1670 NULL, map_sz);
1671 if (err)
1672 return err;
1673
1674 obj->kconfig_map_idx = obj->nr_maps - 1;
1675
1676 return 0;
1677 }
1678
1679 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
1680 {
1681 Elf_Data *symbols = obj->efile.symbols;
1682 int i, map_def_sz = 0, nr_maps = 0, nr_syms;
1683 Elf_Data *data = NULL;
1684 Elf_Scn *scn;
1685
1686 if (obj->efile.maps_shndx < 0)
1687 return 0;
1688
1689 if (!symbols)
1690 return -EINVAL;
1691
1692 scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
1693 if (scn)
1694 data = elf_getdata(scn, NULL);
1695 if (!scn || !data) {
1696 pr_warn("failed to get Elf_Data from map section %d\n",
1697 obj->efile.maps_shndx);
1698 return -EINVAL;
1699 }
1700
1701 /*
1702 * Count number of maps. Each map has a name.
1703 * Array of maps is not supported: only the first element is
1704 * considered.
1705 *
1706 * TODO: Detect array of map and report error.
1707 */
1708 nr_syms = symbols->d_size / sizeof(GElf_Sym);
1709 for (i = 0; i < nr_syms; i++) {
1710 GElf_Sym sym;
1711
1712 if (!gelf_getsym(symbols, i, &sym))
1713 continue;
1714 if (sym.st_shndx != obj->efile.maps_shndx)
1715 continue;
1716 nr_maps++;
1717 }
1718 /* Assume equally sized map definitions */
1719 pr_debug("maps in %s: %d maps in %zd bytes\n",
1720 obj->path, nr_maps, data->d_size);
1721
1722 if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) {
1723 pr_warn("unable to determine map definition size section %s, %d maps in %zd bytes\n",
1724 obj->path, nr_maps, data->d_size);
1725 return -EINVAL;
1726 }
1727 map_def_sz = data->d_size / nr_maps;
1728
1729 /* Fill obj->maps using data in "maps" section. */
1730 for (i = 0; i < nr_syms; i++) {
1731 GElf_Sym sym;
1732 const char *map_name;
1733 struct bpf_map_def *def;
1734 struct bpf_map *map;
1735
1736 if (!gelf_getsym(symbols, i, &sym))
1737 continue;
1738 if (sym.st_shndx != obj->efile.maps_shndx)
1739 continue;
1740
1741 map = bpf_object__add_map(obj);
1742 if (IS_ERR(map))
1743 return PTR_ERR(map);
1744
1745 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1746 sym.st_name);
1747 if (!map_name) {
1748 pr_warn("failed to get map #%d name sym string for obj %s\n",
1749 i, obj->path);
1750 return -LIBBPF_ERRNO__FORMAT;
1751 }
1752
1753 map->libbpf_type = LIBBPF_MAP_UNSPEC;
1754 map->sec_idx = sym.st_shndx;
1755 map->sec_offset = sym.st_value;
1756 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
1757 map_name, map->sec_idx, map->sec_offset);
1758 if (sym.st_value + map_def_sz > data->d_size) {
1759 pr_warn("corrupted maps section in %s: last map \"%s\" too small\n",
1760 obj->path, map_name);
1761 return -EINVAL;
1762 }
1763
1764 map->name = strdup(map_name);
1765 if (!map->name) {
1766 pr_warn("failed to alloc map name\n");
1767 return -ENOMEM;
1768 }
1769 pr_debug("map %d is \"%s\"\n", i, map->name);
1770 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
1771 /*
1772 * If the definition of the map in the object file fits in
1773 * bpf_map_def, copy it. Any extra fields in our version
1774 * of bpf_map_def will default to zero as a result of the
1775 * calloc above.
1776 */
1777 if (map_def_sz <= sizeof(struct bpf_map_def)) {
1778 memcpy(&map->def, def, map_def_sz);
1779 } else {
1780 /*
1781 * Here the map structure being read is bigger than what
1782 * we expect, truncate if the excess bits are all zero.
1783 * If they are not zero, reject this map as
1784 * incompatible.
1785 */
1786 char *b;
1787
1788 for (b = ((char *)def) + sizeof(struct bpf_map_def);
1789 b < ((char *)def) + map_def_sz; b++) {
1790 if (*b != 0) {
1791 pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n",
1792 obj->path, map_name);
1793 if (strict)
1794 return -EINVAL;
1795 }
1796 }
1797 memcpy(&map->def, def, sizeof(struct bpf_map_def));
1798 }
1799 }
1800 return 0;
1801 }
1802
1803 static const struct btf_type *
1804 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1805 {
1806 const struct btf_type *t = btf__type_by_id(btf, id);
1807
1808 if (res_id)
1809 *res_id = id;
1810
1811 while (btf_is_mod(t) || btf_is_typedef(t)) {
1812 if (res_id)
1813 *res_id = t->type;
1814 t = btf__type_by_id(btf, t->type);
1815 }
1816
1817 return t;
1818 }
1819
1820 static const struct btf_type *
1821 resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
1822 {
1823 const struct btf_type *t;
1824
1825 t = skip_mods_and_typedefs(btf, id, NULL);
1826 if (!btf_is_ptr(t))
1827 return NULL;
1828
1829 t = skip_mods_and_typedefs(btf, t->type, res_id);
1830
1831 return btf_is_func_proto(t) ? t : NULL;
1832 }
1833
1834 /*
1835 * Fetch integer attribute of BTF map definition. Such attributes are
1836 * represented using a pointer to an array, in which dimensionality of array
1837 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1838 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1839 * type definition, while using only sizeof(void *) space in ELF data section.
1840 */
1841 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1842 const struct btf_type *def,
1843 const struct btf_member *m, __u32 *res)
1844 {
1845 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
1846 const char *name = btf__name_by_offset(btf, m->name_off);
1847 const struct btf_array *arr_info;
1848 const struct btf_type *arr_t;
1849
1850 if (!btf_is_ptr(t)) {
1851 pr_warn("map '%s': attr '%s': expected PTR, got %u.\n",
1852 map_name, name, btf_kind(t));
1853 return false;
1854 }
1855
1856 arr_t = btf__type_by_id(btf, t->type);
1857 if (!arr_t) {
1858 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
1859 map_name, name, t->type);
1860 return false;
1861 }
1862 if (!btf_is_array(arr_t)) {
1863 pr_warn("map '%s': attr '%s': expected ARRAY, got %u.\n",
1864 map_name, name, btf_kind(arr_t));
1865 return false;
1866 }
1867 arr_info = btf_array(arr_t);
1868 *res = arr_info->nelems;
1869 return true;
1870 }
1871
1872 static int build_map_pin_path(struct bpf_map *map, const char *path)
1873 {
1874 char buf[PATH_MAX];
1875 int err, len;
1876
1877 if (!path)
1878 path = "/sys/fs/bpf";
1879
1880 len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map));
1881 if (len < 0)
1882 return -EINVAL;
1883 else if (len >= PATH_MAX)
1884 return -ENAMETOOLONG;
1885
1886 err = bpf_map__set_pin_path(map, buf);
1887 if (err)
1888 return err;
1889
1890 return 0;
1891 }
1892
1893 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1894 const struct btf_type *sec,
1895 int var_idx, int sec_idx,
1896 const Elf_Data *data, bool strict,
1897 const char *pin_root_path)
1898 {
1899 const struct btf_type *var, *def, *t;
1900 const struct btf_var_secinfo *vi;
1901 const struct btf_var *var_extra;
1902 const struct btf_member *m;
1903 const char *map_name;
1904 struct bpf_map *map;
1905 int vlen, i;
1906
1907 vi = btf_var_secinfos(sec) + var_idx;
1908 var = btf__type_by_id(obj->btf, vi->type);
1909 var_extra = btf_var(var);
1910 map_name = btf__name_by_offset(obj->btf, var->name_off);
1911 vlen = btf_vlen(var);
1912
1913 if (map_name == NULL || map_name[0] == '\0') {
1914 pr_warn("map #%d: empty name.\n", var_idx);
1915 return -EINVAL;
1916 }
1917 if ((__u64)vi->offset + vi->size > data->d_size) {
1918 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
1919 return -EINVAL;
1920 }
1921 if (!btf_is_var(var)) {
1922 pr_warn("map '%s': unexpected var kind %u.\n",
1923 map_name, btf_kind(var));
1924 return -EINVAL;
1925 }
1926 if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1927 var_extra->linkage != BTF_VAR_STATIC) {
1928 pr_warn("map '%s': unsupported var linkage %u.\n",
1929 map_name, var_extra->linkage);
1930 return -EOPNOTSUPP;
1931 }
1932
1933 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
1934 if (!btf_is_struct(def)) {
1935 pr_warn("map '%s': unexpected def kind %u.\n",
1936 map_name, btf_kind(var));
1937 return -EINVAL;
1938 }
1939 if (def->size > vi->size) {
1940 pr_warn("map '%s': invalid def size.\n", map_name);
1941 return -EINVAL;
1942 }
1943
1944 map = bpf_object__add_map(obj);
1945 if (IS_ERR(map))
1946 return PTR_ERR(map);
1947 map->name = strdup(map_name);
1948 if (!map->name) {
1949 pr_warn("map '%s': failed to alloc map name.\n", map_name);
1950 return -ENOMEM;
1951 }
1952 map->libbpf_type = LIBBPF_MAP_UNSPEC;
1953 map->def.type = BPF_MAP_TYPE_UNSPEC;
1954 map->sec_idx = sec_idx;
1955 map->sec_offset = vi->offset;
1956 pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1957 map_name, map->sec_idx, map->sec_offset);
1958
1959 vlen = btf_vlen(def);
1960 m = btf_members(def);
1961 for (i = 0; i < vlen; i++, m++) {
1962 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1963
1964 if (!name) {
1965 pr_warn("map '%s': invalid field #%d.\n", map_name, i);
1966 return -EINVAL;
1967 }
1968 if (strcmp(name, "type") == 0) {
1969 if (!get_map_field_int(map_name, obj->btf, def, m,
1970 &map->def.type))
1971 return -EINVAL;
1972 pr_debug("map '%s': found type = %u.\n",
1973 map_name, map->def.type);
1974 } else if (strcmp(name, "max_entries") == 0) {
1975 if (!get_map_field_int(map_name, obj->btf, def, m,
1976 &map->def.max_entries))
1977 return -EINVAL;
1978 pr_debug("map '%s': found max_entries = %u.\n",
1979 map_name, map->def.max_entries);
1980 } else if (strcmp(name, "map_flags") == 0) {
1981 if (!get_map_field_int(map_name, obj->btf, def, m,
1982 &map->def.map_flags))
1983 return -EINVAL;
1984 pr_debug("map '%s': found map_flags = %u.\n",
1985 map_name, map->def.map_flags);
1986 } else if (strcmp(name, "key_size") == 0) {
1987 __u32 sz;
1988
1989 if (!get_map_field_int(map_name, obj->btf, def, m,
1990 &sz))
1991 return -EINVAL;
1992 pr_debug("map '%s': found key_size = %u.\n",
1993 map_name, sz);
1994 if (map->def.key_size && map->def.key_size != sz) {
1995 pr_warn("map '%s': conflicting key size %u != %u.\n",
1996 map_name, map->def.key_size, sz);
1997 return -EINVAL;
1998 }
1999 map->def.key_size = sz;
2000 } else if (strcmp(name, "key") == 0) {
2001 __s64 sz;
2002
2003 t = btf__type_by_id(obj->btf, m->type);
2004 if (!t) {
2005 pr_warn("map '%s': key type [%d] not found.\n",
2006 map_name, m->type);
2007 return -EINVAL;
2008 }
2009 if (!btf_is_ptr(t)) {
2010 pr_warn("map '%s': key spec is not PTR: %u.\n",
2011 map_name, btf_kind(t));
2012 return -EINVAL;
2013 }
2014 sz = btf__resolve_size(obj->btf, t->type);
2015 if (sz < 0) {
2016 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2017 map_name, t->type, (ssize_t)sz);
2018 return sz;
2019 }
2020 pr_debug("map '%s': found key [%u], sz = %zd.\n",
2021 map_name, t->type, (ssize_t)sz);
2022 if (map->def.key_size && map->def.key_size != sz) {
2023 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2024 map_name, map->def.key_size, (ssize_t)sz);
2025 return -EINVAL;
2026 }
2027 map->def.key_size = sz;
2028 map->btf_key_type_id = t->type;
2029 } else if (strcmp(name, "value_size") == 0) {
2030 __u32 sz;
2031
2032 if (!get_map_field_int(map_name, obj->btf, def, m,
2033 &sz))
2034 return -EINVAL;
2035 pr_debug("map '%s': found value_size = %u.\n",
2036 map_name, sz);
2037 if (map->def.value_size && map->def.value_size != sz) {
2038 pr_warn("map '%s': conflicting value size %u != %u.\n",
2039 map_name, map->def.value_size, sz);
2040 return -EINVAL;
2041 }
2042 map->def.value_size = sz;
2043 } else if (strcmp(name, "value") == 0) {
2044 __s64 sz;
2045
2046 t = btf__type_by_id(obj->btf, m->type);
2047 if (!t) {
2048 pr_warn("map '%s': value type [%d] not found.\n",
2049 map_name, m->type);
2050 return -EINVAL;
2051 }
2052 if (!btf_is_ptr(t)) {
2053 pr_warn("map '%s': value spec is not PTR: %u.\n",
2054 map_name, btf_kind(t));
2055 return -EINVAL;
2056 }
2057 sz = btf__resolve_size(obj->btf, t->type);
2058 if (sz < 0) {
2059 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2060 map_name, t->type, (ssize_t)sz);
2061 return sz;
2062 }
2063 pr_debug("map '%s': found value [%u], sz = %zd.\n",
2064 map_name, t->type, (ssize_t)sz);
2065 if (map->def.value_size && map->def.value_size != sz) {
2066 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2067 map_name, map->def.value_size, (ssize_t)sz);
2068 return -EINVAL;
2069 }
2070 map->def.value_size = sz;
2071 map->btf_value_type_id = t->type;
2072 } else if (strcmp(name, "pinning") == 0) {
2073 __u32 val;
2074 int err;
2075
2076 if (!get_map_field_int(map_name, obj->btf, def, m,
2077 &val))
2078 return -EINVAL;
2079 pr_debug("map '%s': found pinning = %u.\n",
2080 map_name, val);
2081
2082 if (val != LIBBPF_PIN_NONE &&
2083 val != LIBBPF_PIN_BY_NAME) {
2084 pr_warn("map '%s': invalid pinning value %u.\n",
2085 map_name, val);
2086 return -EINVAL;
2087 }
2088 if (val == LIBBPF_PIN_BY_NAME) {
2089 err = build_map_pin_path(map, pin_root_path);
2090 if (err) {
2091 pr_warn("map '%s': couldn't build pin path.\n",
2092 map_name);
2093 return err;
2094 }
2095 }
2096 } else {
2097 if (strict) {
2098 pr_warn("map '%s': unknown field '%s'.\n",
2099 map_name, name);
2100 return -ENOTSUP;
2101 }
2102 pr_debug("map '%s': ignoring unknown field '%s'.\n",
2103 map_name, name);
2104 }
2105 }
2106
2107 if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
2108 pr_warn("map '%s': map type isn't specified.\n", map_name);
2109 return -EINVAL;
2110 }
2111
2112 return 0;
2113 }
2114
2115 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2116 const char *pin_root_path)
2117 {
2118 const struct btf_type *sec = NULL;
2119 int nr_types, i, vlen, err;
2120 const struct btf_type *t;
2121 const char *name;
2122 Elf_Data *data;
2123 Elf_Scn *scn;
2124
2125 if (obj->efile.btf_maps_shndx < 0)
2126 return 0;
2127
2128 scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
2129 if (scn)
2130 data = elf_getdata(scn, NULL);
2131 if (!scn || !data) {
2132 pr_warn("failed to get Elf_Data from map section %d (%s)\n",
2133 obj->efile.maps_shndx, MAPS_ELF_SEC);
2134 return -EINVAL;
2135 }
2136
2137 nr_types = btf__get_nr_types(obj->btf);
2138 for (i = 1; i <= nr_types; i++) {
2139 t = btf__type_by_id(obj->btf, i);
2140 if (!btf_is_datasec(t))
2141 continue;
2142 name = btf__name_by_offset(obj->btf, t->name_off);
2143 if (strcmp(name, MAPS_ELF_SEC) == 0) {
2144 sec = t;
2145 break;
2146 }
2147 }
2148
2149 if (!sec) {
2150 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
2151 return -ENOENT;
2152 }
2153
2154 vlen = btf_vlen(sec);
2155 for (i = 0; i < vlen; i++) {
2156 err = bpf_object__init_user_btf_map(obj, sec, i,
2157 obj->efile.btf_maps_shndx,
2158 data, strict,
2159 pin_root_path);
2160 if (err)
2161 return err;
2162 }
2163
2164 return 0;
2165 }
2166
2167 static int bpf_object__init_maps(struct bpf_object *obj,
2168 const struct bpf_object_open_opts *opts)
2169 {
2170 const char *pin_root_path;
2171 bool strict;
2172 int err;
2173
2174 strict = !OPTS_GET(opts, relaxed_maps, false);
2175 pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
2176
2177 err = bpf_object__init_user_maps(obj, strict);
2178 err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
2179 err = err ?: bpf_object__init_global_data_maps(obj);
2180 err = err ?: bpf_object__init_kconfig_map(obj);
2181 err = err ?: bpf_object__init_struct_ops_maps(obj);
2182 if (err)
2183 return err;
2184
2185 return 0;
2186 }
2187
2188 static bool section_have_execinstr(struct bpf_object *obj, int idx)
2189 {
2190 Elf_Scn *scn;
2191 GElf_Shdr sh;
2192
2193 scn = elf_getscn(obj->efile.elf, idx);
2194 if (!scn)
2195 return false;
2196
2197 if (gelf_getshdr(scn, &sh) != &sh)
2198 return false;
2199
2200 if (sh.sh_flags & SHF_EXECINSTR)
2201 return true;
2202
2203 return false;
2204 }
2205
2206 static void bpf_object__sanitize_btf(struct bpf_object *obj)
2207 {
2208 bool has_func_global = obj->caps.btf_func_global;
2209 bool has_datasec = obj->caps.btf_datasec;
2210 bool has_func = obj->caps.btf_func;
2211 struct btf *btf = obj->btf;
2212 struct btf_type *t;
2213 int i, j, vlen;
2214
2215 if (!obj->btf || (has_func && has_datasec && has_func_global))
2216 return;
2217
2218 for (i = 1; i <= btf__get_nr_types(btf); i++) {
2219 t = (struct btf_type *)btf__type_by_id(btf, i);
2220
2221 if (!has_datasec && btf_is_var(t)) {
2222 /* replace VAR with INT */
2223 t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
2224 /*
2225 * using size = 1 is the safest choice, 4 will be too
2226 * big and cause kernel BTF validation failure if
2227 * original variable took less than 4 bytes
2228 */
2229 t->size = 1;
2230 *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
2231 } else if (!has_datasec && btf_is_datasec(t)) {
2232 /* replace DATASEC with STRUCT */
2233 const struct btf_var_secinfo *v = btf_var_secinfos(t);
2234 struct btf_member *m = btf_members(t);
2235 struct btf_type *vt;
2236 char *name;
2237
2238 name = (char *)btf__name_by_offset(btf, t->name_off);
2239 while (*name) {
2240 if (*name == '.')
2241 *name = '_';
2242 name++;
2243 }
2244
2245 vlen = btf_vlen(t);
2246 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
2247 for (j = 0; j < vlen; j++, v++, m++) {
2248 /* order of field assignments is important */
2249 m->offset = v->offset * 8;
2250 m->type = v->type;
2251 /* preserve variable name as member name */
2252 vt = (void *)btf__type_by_id(btf, v->type);
2253 m->name_off = vt->name_off;
2254 }
2255 } else if (!has_func && btf_is_func_proto(t)) {
2256 /* replace FUNC_PROTO with ENUM */
2257 vlen = btf_vlen(t);
2258 t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
2259 t->size = sizeof(__u32); /* kernel enforced */
2260 } else if (!has_func && btf_is_func(t)) {
2261 /* replace FUNC with TYPEDEF */
2262 t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
2263 } else if (!has_func_global && btf_is_func(t)) {
2264 /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
2265 t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
2266 }
2267 }
2268 }
2269
2270 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
2271 {
2272 if (!obj->btf_ext)
2273 return;
2274
2275 if (!obj->caps.btf_func) {
2276 btf_ext__free(obj->btf_ext);
2277 obj->btf_ext = NULL;
2278 }
2279 }
2280
2281 static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
2282 {
2283 return obj->efile.btf_maps_shndx >= 0 ||
2284 obj->efile.st_ops_shndx >= 0 ||
2285 obj->nr_extern > 0;
2286 }
2287
2288 static int bpf_object__init_btf(struct bpf_object *obj,
2289 Elf_Data *btf_data,
2290 Elf_Data *btf_ext_data)
2291 {
2292 int err = -ENOENT;
2293
2294 if (btf_data) {
2295 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
2296 if (IS_ERR(obj->btf)) {
2297 err = PTR_ERR(obj->btf);
2298 obj->btf = NULL;
2299 pr_warn("Error loading ELF section %s: %d.\n",
2300 BTF_ELF_SEC, err);
2301 goto out;
2302 }
2303 err = 0;
2304 }
2305 if (btf_ext_data) {
2306 if (!obj->btf) {
2307 pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2308 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2309 goto out;
2310 }
2311 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
2312 btf_ext_data->d_size);
2313 if (IS_ERR(obj->btf_ext)) {
2314 pr_warn("Error loading ELF section %s: %ld. Ignored and continue.\n",
2315 BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
2316 obj->btf_ext = NULL;
2317 goto out;
2318 }
2319 }
2320 out:
2321 if (err && bpf_object__is_btf_mandatory(obj)) {
2322 pr_warn("BTF is required, but is missing or corrupted.\n");
2323 return err;
2324 }
2325 return 0;
2326 }
2327
2328 static int bpf_object__finalize_btf(struct bpf_object *obj)
2329 {
2330 int err;
2331
2332 if (!obj->btf)
2333 return 0;
2334
2335 err = btf__finalize_data(obj, obj->btf);
2336 if (!err)
2337 return 0;
2338
2339 pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err);
2340 btf__free(obj->btf);
2341 obj->btf = NULL;
2342 btf_ext__free(obj->btf_ext);
2343 obj->btf_ext = NULL;
2344
2345 if (bpf_object__is_btf_mandatory(obj)) {
2346 pr_warn("BTF is required, but is missing or corrupted.\n");
2347 return -ENOENT;
2348 }
2349 return 0;
2350 }
2351
2352 static inline bool libbpf_prog_needs_vmlinux_btf(struct bpf_program *prog)
2353 {
2354 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS)
2355 return true;
2356
2357 /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
2358 * also need vmlinux BTF
2359 */
2360 if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
2361 return true;
2362
2363 return false;
2364 }
2365
2366 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj)
2367 {
2368 struct bpf_program *prog;
2369 int err;
2370
2371 bpf_object__for_each_program(prog, obj) {
2372 if (libbpf_prog_needs_vmlinux_btf(prog)) {
2373 obj->btf_vmlinux = libbpf_find_kernel_btf();
2374 if (IS_ERR(obj->btf_vmlinux)) {
2375 err = PTR_ERR(obj->btf_vmlinux);
2376 pr_warn("Error loading vmlinux BTF: %d\n", err);
2377 obj->btf_vmlinux = NULL;
2378 return err;
2379 }
2380 return 0;
2381 }
2382 }
2383
2384 return 0;
2385 }
2386
2387 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
2388 {
2389 int err = 0;
2390
2391 if (!obj->btf)
2392 return 0;
2393
2394 bpf_object__sanitize_btf(obj);
2395 bpf_object__sanitize_btf_ext(obj);
2396
2397 err = btf__load(obj->btf);
2398 if (err) {
2399 pr_warn("Error loading %s into kernel: %d.\n",
2400 BTF_ELF_SEC, err);
2401 btf__free(obj->btf);
2402 obj->btf = NULL;
2403 /* btf_ext can't exist without btf, so free it as well */
2404 if (obj->btf_ext) {
2405 btf_ext__free(obj->btf_ext);
2406 obj->btf_ext = NULL;
2407 }
2408
2409 if (bpf_object__is_btf_mandatory(obj))
2410 return err;
2411 }
2412 return 0;
2413 }
2414
2415 static int bpf_object__elf_collect(struct bpf_object *obj)
2416 {
2417 Elf *elf = obj->efile.elf;
2418 GElf_Ehdr *ep = &obj->efile.ehdr;
2419 Elf_Data *btf_ext_data = NULL;
2420 Elf_Data *btf_data = NULL;
2421 Elf_Scn *scn = NULL;
2422 int idx = 0, err = 0;
2423
2424 /* Elf is corrupted/truncated, avoid calling elf_strptr. */
2425 if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
2426 pr_warn("failed to get e_shstrndx from %s\n", obj->path);
2427 return -LIBBPF_ERRNO__FORMAT;
2428 }
2429
2430 while ((scn = elf_nextscn(elf, scn)) != NULL) {
2431 char *name;
2432 GElf_Shdr sh;
2433 Elf_Data *data;
2434
2435 idx++;
2436 if (gelf_getshdr(scn, &sh) != &sh) {
2437 pr_warn("failed to get section(%d) header from %s\n",
2438 idx, obj->path);
2439 return -LIBBPF_ERRNO__FORMAT;
2440 }
2441
2442 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
2443 if (!name) {
2444 pr_warn("failed to get section(%d) name from %s\n",
2445 idx, obj->path);
2446 return -LIBBPF_ERRNO__FORMAT;
2447 }
2448
2449 data = elf_getdata(scn, 0);
2450 if (!data) {
2451 pr_warn("failed to get section(%d) data from %s(%s)\n",
2452 idx, name, obj->path);
2453 return -LIBBPF_ERRNO__FORMAT;
2454 }
2455 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
2456 idx, name, (unsigned long)data->d_size,
2457 (int)sh.sh_link, (unsigned long)sh.sh_flags,
2458 (int)sh.sh_type);
2459
2460 if (strcmp(name, "license") == 0) {
2461 err = bpf_object__init_license(obj,
2462 data->d_buf,
2463 data->d_size);
2464 if (err)
2465 return err;
2466 } else if (strcmp(name, "version") == 0) {
2467 err = bpf_object__init_kversion(obj,
2468 data->d_buf,
2469 data->d_size);
2470 if (err)
2471 return err;
2472 } else if (strcmp(name, "maps") == 0) {
2473 obj->efile.maps_shndx = idx;
2474 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
2475 obj->efile.btf_maps_shndx = idx;
2476 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
2477 btf_data = data;
2478 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
2479 btf_ext_data = data;
2480 } else if (sh.sh_type == SHT_SYMTAB) {
2481 if (obj->efile.symbols) {
2482 pr_warn("bpf: multiple SYMTAB in %s\n",
2483 obj->path);
2484 return -LIBBPF_ERRNO__FORMAT;
2485 }
2486 obj->efile.symbols = data;
2487 obj->efile.symbols_shndx = idx;
2488 obj->efile.strtabidx = sh.sh_link;
2489 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
2490 if (sh.sh_flags & SHF_EXECINSTR) {
2491 if (strcmp(name, ".text") == 0)
2492 obj->efile.text_shndx = idx;
2493 err = bpf_object__add_program(obj, data->d_buf,
2494 data->d_size,
2495 name, idx);
2496 if (err) {
2497 char errmsg[STRERR_BUFSIZE];
2498 char *cp;
2499
2500 cp = libbpf_strerror_r(-err, errmsg,
2501 sizeof(errmsg));
2502 pr_warn("failed to alloc program %s (%s): %s",
2503 name, obj->path, cp);
2504 return err;
2505 }
2506 } else if (strcmp(name, DATA_SEC) == 0) {
2507 obj->efile.data = data;
2508 obj->efile.data_shndx = idx;
2509 } else if (strcmp(name, RODATA_SEC) == 0) {
2510 obj->efile.rodata = data;
2511 obj->efile.rodata_shndx = idx;
2512 } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
2513 obj->efile.st_ops_data = data;
2514 obj->efile.st_ops_shndx = idx;
2515 } else {
2516 pr_debug("skip section(%d) %s\n", idx, name);
2517 }
2518 } else if (sh.sh_type == SHT_REL) {
2519 int nr_sects = obj->efile.nr_reloc_sects;
2520 void *sects = obj->efile.reloc_sects;
2521 int sec = sh.sh_info; /* points to other section */
2522
2523 /* Only do relo for section with exec instructions */
2524 if (!section_have_execinstr(obj, sec) &&
2525 strcmp(name, ".rel" STRUCT_OPS_SEC)) {
2526 pr_debug("skip relo %s(%d) for section(%d)\n",
2527 name, idx, sec);
2528 continue;
2529 }
2530
2531 sects = reallocarray(sects, nr_sects + 1,
2532 sizeof(*obj->efile.reloc_sects));
2533 if (!sects) {
2534 pr_warn("reloc_sects realloc failed\n");
2535 return -ENOMEM;
2536 }
2537
2538 obj->efile.reloc_sects = sects;
2539 obj->efile.nr_reloc_sects++;
2540
2541 obj->efile.reloc_sects[nr_sects].shdr = sh;
2542 obj->efile.reloc_sects[nr_sects].data = data;
2543 } else if (sh.sh_type == SHT_NOBITS &&
2544 strcmp(name, BSS_SEC) == 0) {
2545 obj->efile.bss = data;
2546 obj->efile.bss_shndx = idx;
2547 } else {
2548 pr_debug("skip section(%d) %s\n", idx, name);
2549 }
2550 }
2551
2552 if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
2553 pr_warn("Corrupted ELF file: index of strtab invalid\n");
2554 return -LIBBPF_ERRNO__FORMAT;
2555 }
2556 return bpf_object__init_btf(obj, btf_data, btf_ext_data);
2557 }
2558
2559 static bool sym_is_extern(const GElf_Sym *sym)
2560 {
2561 int bind = GELF_ST_BIND(sym->st_info);
2562 /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
2563 return sym->st_shndx == SHN_UNDEF &&
2564 (bind == STB_GLOBAL || bind == STB_WEAK) &&
2565 GELF_ST_TYPE(sym->st_info) == STT_NOTYPE;
2566 }
2567
2568 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
2569 {
2570 const struct btf_type *t;
2571 const char *var_name;
2572 int i, n;
2573
2574 if (!btf)
2575 return -ESRCH;
2576
2577 n = btf__get_nr_types(btf);
2578 for (i = 1; i <= n; i++) {
2579 t = btf__type_by_id(btf, i);
2580
2581 if (!btf_is_var(t))
2582 continue;
2583
2584 var_name = btf__name_by_offset(btf, t->name_off);
2585 if (strcmp(var_name, ext_name))
2586 continue;
2587
2588 if (btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
2589 return -EINVAL;
2590
2591 return i;
2592 }
2593
2594 return -ENOENT;
2595 }
2596
2597 static enum extern_type find_extern_type(const struct btf *btf, int id,
2598 bool *is_signed)
2599 {
2600 const struct btf_type *t;
2601 const char *name;
2602
2603 t = skip_mods_and_typedefs(btf, id, NULL);
2604 name = btf__name_by_offset(btf, t->name_off);
2605
2606 if (is_signed)
2607 *is_signed = false;
2608 switch (btf_kind(t)) {
2609 case BTF_KIND_INT: {
2610 int enc = btf_int_encoding(t);
2611
2612 if (enc & BTF_INT_BOOL)
2613 return t->size == 1 ? EXT_BOOL : EXT_UNKNOWN;
2614 if (is_signed)
2615 *is_signed = enc & BTF_INT_SIGNED;
2616 if (t->size == 1)
2617 return EXT_CHAR;
2618 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
2619 return EXT_UNKNOWN;
2620 return EXT_INT;
2621 }
2622 case BTF_KIND_ENUM:
2623 if (t->size != 4)
2624 return EXT_UNKNOWN;
2625 if (strcmp(name, "libbpf_tristate"))
2626 return EXT_UNKNOWN;
2627 return EXT_TRISTATE;
2628 case BTF_KIND_ARRAY:
2629 if (btf_array(t)->nelems == 0)
2630 return EXT_UNKNOWN;
2631 if (find_extern_type(btf, btf_array(t)->type, NULL) != EXT_CHAR)
2632 return EXT_UNKNOWN;
2633 return EXT_CHAR_ARR;
2634 default:
2635 return EXT_UNKNOWN;
2636 }
2637 }
2638
2639 static int cmp_externs(const void *_a, const void *_b)
2640 {
2641 const struct extern_desc *a = _a;
2642 const struct extern_desc *b = _b;
2643
2644 /* descending order by alignment requirements */
2645 if (a->align != b->align)
2646 return a->align > b->align ? -1 : 1;
2647 /* ascending order by size, within same alignment class */
2648 if (a->sz != b->sz)
2649 return a->sz < b->sz ? -1 : 1;
2650 /* resolve ties by name */
2651 return strcmp(a->name, b->name);
2652 }
2653
2654 static int bpf_object__collect_externs(struct bpf_object *obj)
2655 {
2656 const struct btf_type *t;
2657 struct extern_desc *ext;
2658 int i, n, off, btf_id;
2659 struct btf_type *sec;
2660 const char *ext_name;
2661 Elf_Scn *scn;
2662 GElf_Shdr sh;
2663
2664 if (!obj->efile.symbols)
2665 return 0;
2666
2667 scn = elf_getscn(obj->efile.elf, obj->efile.symbols_shndx);
2668 if (!scn)
2669 return -LIBBPF_ERRNO__FORMAT;
2670 if (gelf_getshdr(scn, &sh) != &sh)
2671 return -LIBBPF_ERRNO__FORMAT;
2672 n = sh.sh_size / sh.sh_entsize;
2673
2674 pr_debug("looking for externs among %d symbols...\n", n);
2675 for (i = 0; i < n; i++) {
2676 GElf_Sym sym;
2677
2678 if (!gelf_getsym(obj->efile.symbols, i, &sym))
2679 return -LIBBPF_ERRNO__FORMAT;
2680 if (!sym_is_extern(&sym))
2681 continue;
2682 ext_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
2683 sym.st_name);
2684 if (!ext_name || !ext_name[0])
2685 continue;
2686
2687 ext = obj->externs;
2688 ext = reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
2689 if (!ext)
2690 return -ENOMEM;
2691 obj->externs = ext;
2692 ext = &ext[obj->nr_extern];
2693 memset(ext, 0, sizeof(*ext));
2694 obj->nr_extern++;
2695
2696 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
2697 if (ext->btf_id <= 0) {
2698 pr_warn("failed to find BTF for extern '%s': %d\n",
2699 ext_name, ext->btf_id);
2700 return ext->btf_id;
2701 }
2702 t = btf__type_by_id(obj->btf, ext->btf_id);
2703 ext->name = btf__name_by_offset(obj->btf, t->name_off);
2704 ext->sym_idx = i;
2705 ext->is_weak = GELF_ST_BIND(sym.st_info) == STB_WEAK;
2706 ext->sz = btf__resolve_size(obj->btf, t->type);
2707 if (ext->sz <= 0) {
2708 pr_warn("failed to resolve size of extern '%s': %d\n",
2709 ext_name, ext->sz);
2710 return ext->sz;
2711 }
2712 ext->align = btf__align_of(obj->btf, t->type);
2713 if (ext->align <= 0) {
2714 pr_warn("failed to determine alignment of extern '%s': %d\n",
2715 ext_name, ext->align);
2716 return -EINVAL;
2717 }
2718 ext->type = find_extern_type(obj->btf, t->type,
2719 &ext->is_signed);
2720 if (ext->type == EXT_UNKNOWN) {
2721 pr_warn("extern '%s' type is unsupported\n", ext_name);
2722 return -ENOTSUP;
2723 }
2724 }
2725 pr_debug("collected %d externs total\n", obj->nr_extern);
2726
2727 if (!obj->nr_extern)
2728 return 0;
2729
2730 /* sort externs by (alignment, size, name) and calculate their offsets
2731 * within a map */
2732 qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
2733 off = 0;
2734 for (i = 0; i < obj->nr_extern; i++) {
2735 ext = &obj->externs[i];
2736 ext->data_off = roundup(off, ext->align);
2737 off = ext->data_off + ext->sz;
2738 pr_debug("extern #%d: symbol %d, off %u, name %s\n",
2739 i, ext->sym_idx, ext->data_off, ext->name);
2740 }
2741
2742 btf_id = btf__find_by_name(obj->btf, KCONFIG_SEC);
2743 if (btf_id <= 0) {
2744 pr_warn("no BTF info found for '%s' datasec\n", KCONFIG_SEC);
2745 return -ESRCH;
2746 }
2747
2748 sec = (struct btf_type *)btf__type_by_id(obj->btf, btf_id);
2749 sec->size = off;
2750 n = btf_vlen(sec);
2751 for (i = 0; i < n; i++) {
2752 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
2753
2754 t = btf__type_by_id(obj->btf, vs->type);
2755 ext_name = btf__name_by_offset(obj->btf, t->name_off);
2756 ext = find_extern_by_name(obj, ext_name);
2757 if (!ext) {
2758 pr_warn("failed to find extern definition for BTF var '%s'\n",
2759 ext_name);
2760 return -ESRCH;
2761 }
2762 vs->offset = ext->data_off;
2763 btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
2764 }
2765
2766 return 0;
2767 }
2768
2769 static struct bpf_program *
2770 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
2771 {
2772 struct bpf_program *prog;
2773 size_t i;
2774
2775 for (i = 0; i < obj->nr_programs; i++) {
2776 prog = &obj->programs[i];
2777 if (prog->idx == idx)
2778 return prog;
2779 }
2780 return NULL;
2781 }
2782
2783 struct bpf_program *
2784 bpf_object__find_program_by_title(const struct bpf_object *obj,
2785 const char *title)
2786 {
2787 struct bpf_program *pos;
2788
2789 bpf_object__for_each_program(pos, obj) {
2790 if (pos->section_name && !strcmp(pos->section_name, title))
2791 return pos;
2792 }
2793 return NULL;
2794 }
2795
2796 struct bpf_program *
2797 bpf_object__find_program_by_name(const struct bpf_object *obj,
2798 const char *name)
2799 {
2800 struct bpf_program *prog;
2801
2802 bpf_object__for_each_program(prog, obj) {
2803 if (!strcmp(prog->name, name))
2804 return prog;
2805 }
2806 return NULL;
2807 }
2808
2809 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
2810 int shndx)
2811 {
2812 return shndx == obj->efile.data_shndx ||
2813 shndx == obj->efile.bss_shndx ||
2814 shndx == obj->efile.rodata_shndx;
2815 }
2816
2817 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
2818 int shndx)
2819 {
2820 return shndx == obj->efile.maps_shndx ||
2821 shndx == obj->efile.btf_maps_shndx;
2822 }
2823
2824 static enum libbpf_map_type
2825 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
2826 {
2827 if (shndx == obj->efile.data_shndx)
2828 return LIBBPF_MAP_DATA;
2829 else if (shndx == obj->efile.bss_shndx)
2830 return LIBBPF_MAP_BSS;
2831 else if (shndx == obj->efile.rodata_shndx)
2832 return LIBBPF_MAP_RODATA;
2833 else if (shndx == obj->efile.symbols_shndx)
2834 return LIBBPF_MAP_KCONFIG;
2835 else
2836 return LIBBPF_MAP_UNSPEC;
2837 }
2838
2839 static int bpf_program__record_reloc(struct bpf_program *prog,
2840 struct reloc_desc *reloc_desc,
2841 __u32 insn_idx, const char *name,
2842 const GElf_Sym *sym, const GElf_Rel *rel)
2843 {
2844 struct bpf_insn *insn = &prog->insns[insn_idx];
2845 size_t map_idx, nr_maps = prog->obj->nr_maps;
2846 struct bpf_object *obj = prog->obj;
2847 __u32 shdr_idx = sym->st_shndx;
2848 enum libbpf_map_type type;
2849 struct bpf_map *map;
2850
2851 /* sub-program call relocation */
2852 if (insn->code == (BPF_JMP | BPF_CALL)) {
2853 if (insn->src_reg != BPF_PSEUDO_CALL) {
2854 pr_warn("incorrect bpf_call opcode\n");
2855 return -LIBBPF_ERRNO__RELOC;
2856 }
2857 /* text_shndx can be 0, if no default "main" program exists */
2858 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
2859 pr_warn("bad call relo against section %u\n", shdr_idx);
2860 return -LIBBPF_ERRNO__RELOC;
2861 }
2862 if (sym->st_value % 8) {
2863 pr_warn("bad call relo offset: %zu\n",
2864 (size_t)sym->st_value);
2865 return -LIBBPF_ERRNO__RELOC;
2866 }
2867 reloc_desc->type = RELO_CALL;
2868 reloc_desc->insn_idx = insn_idx;
2869 reloc_desc->sym_off = sym->st_value;
2870 obj->has_pseudo_calls = true;
2871 return 0;
2872 }
2873
2874 if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) {
2875 pr_warn("invalid relo for insns[%d].code 0x%x\n",
2876 insn_idx, insn->code);
2877 return -LIBBPF_ERRNO__RELOC;
2878 }
2879
2880 if (sym_is_extern(sym)) {
2881 int sym_idx = GELF_R_SYM(rel->r_info);
2882 int i, n = obj->nr_extern;
2883 struct extern_desc *ext;
2884
2885 for (i = 0; i < n; i++) {
2886 ext = &obj->externs[i];
2887 if (ext->sym_idx == sym_idx)
2888 break;
2889 }
2890 if (i >= n) {
2891 pr_warn("extern relo failed to find extern for sym %d\n",
2892 sym_idx);
2893 return -LIBBPF_ERRNO__RELOC;
2894 }
2895 pr_debug("found extern #%d '%s' (sym %d, off %u) for insn %u\n",
2896 i, ext->name, ext->sym_idx, ext->data_off, insn_idx);
2897 reloc_desc->type = RELO_EXTERN;
2898 reloc_desc->insn_idx = insn_idx;
2899 reloc_desc->sym_off = ext->data_off;
2900 return 0;
2901 }
2902
2903 if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
2904 pr_warn("invalid relo for \'%s\' in special section 0x%x; forgot to initialize global var?..\n",
2905 name, shdr_idx);
2906 return -LIBBPF_ERRNO__RELOC;
2907 }
2908
2909 type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
2910
2911 /* generic map reference relocation */
2912 if (type == LIBBPF_MAP_UNSPEC) {
2913 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
2914 pr_warn("bad map relo against section %u\n",
2915 shdr_idx);
2916 return -LIBBPF_ERRNO__RELOC;
2917 }
2918 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
2919 map = &obj->maps[map_idx];
2920 if (map->libbpf_type != type ||
2921 map->sec_idx != sym->st_shndx ||
2922 map->sec_offset != sym->st_value)
2923 continue;
2924 pr_debug("found map %zd (%s, sec %d, off %zu) for insn %u\n",
2925 map_idx, map->name, map->sec_idx,
2926 map->sec_offset, insn_idx);
2927 break;
2928 }
2929 if (map_idx >= nr_maps) {
2930 pr_warn("map relo failed to find map for sec %u, off %zu\n",
2931 shdr_idx, (size_t)sym->st_value);
2932 return -LIBBPF_ERRNO__RELOC;
2933 }
2934 reloc_desc->type = RELO_LD64;
2935 reloc_desc->insn_idx = insn_idx;
2936 reloc_desc->map_idx = map_idx;
2937 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
2938 return 0;
2939 }
2940
2941 /* global data map relocation */
2942 if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
2943 pr_warn("bad data relo against section %u\n", shdr_idx);
2944 return -LIBBPF_ERRNO__RELOC;
2945 }
2946 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
2947 map = &obj->maps[map_idx];
2948 if (map->libbpf_type != type)
2949 continue;
2950 pr_debug("found data map %zd (%s, sec %d, off %zu) for insn %u\n",
2951 map_idx, map->name, map->sec_idx, map->sec_offset,
2952 insn_idx);
2953 break;
2954 }
2955 if (map_idx >= nr_maps) {
2956 pr_warn("data relo failed to find map for sec %u\n",
2957 shdr_idx);
2958 return -LIBBPF_ERRNO__RELOC;
2959 }
2960
2961 reloc_desc->type = RELO_DATA;
2962 reloc_desc->insn_idx = insn_idx;
2963 reloc_desc->map_idx = map_idx;
2964 reloc_desc->sym_off = sym->st_value;
2965 return 0;
2966 }
2967
2968 static int
2969 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
2970 Elf_Data *data, struct bpf_object *obj)
2971 {
2972 Elf_Data *symbols = obj->efile.symbols;
2973 int err, i, nrels;
2974
2975 pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
2976 nrels = shdr->sh_size / shdr->sh_entsize;
2977
2978 prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
2979 if (!prog->reloc_desc) {
2980 pr_warn("failed to alloc memory in relocation\n");
2981 return -ENOMEM;
2982 }
2983 prog->nr_reloc = nrels;
2984
2985 for (i = 0; i < nrels; i++) {
2986 const char *name;
2987 __u32 insn_idx;
2988 GElf_Sym sym;
2989 GElf_Rel rel;
2990
2991 if (!gelf_getrel(data, i, &rel)) {
2992 pr_warn("relocation: failed to get %d reloc\n", i);
2993 return -LIBBPF_ERRNO__FORMAT;
2994 }
2995 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
2996 pr_warn("relocation: symbol %"PRIx64" not found\n",
2997 GELF_R_SYM(rel.r_info));
2998 return -LIBBPF_ERRNO__FORMAT;
2999 }
3000 if (rel.r_offset % sizeof(struct bpf_insn))
3001 return -LIBBPF_ERRNO__FORMAT;
3002
3003 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
3004 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
3005 sym.st_name) ? : "<?>";
3006
3007 pr_debug("relo for shdr %u, symb %zu, value %zu, type %d, bind %d, name %d (\'%s\'), insn %u\n",
3008 (__u32)sym.st_shndx, (size_t)GELF_R_SYM(rel.r_info),
3009 (size_t)sym.st_value, GELF_ST_TYPE(sym.st_info),
3010 GELF_ST_BIND(sym.st_info), sym.st_name, name,
3011 insn_idx);
3012
3013 err = bpf_program__record_reloc(prog, &prog->reloc_desc[i],
3014 insn_idx, name, &sym, &rel);
3015 if (err)
3016 return err;
3017 }
3018 return 0;
3019 }
3020
3021 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
3022 {
3023 struct bpf_map_def *def = &map->def;
3024 __u32 key_type_id = 0, value_type_id = 0;
3025 int ret;
3026
3027 /* if it's BTF-defined map, we don't need to search for type IDs.
3028 * For struct_ops map, it does not need btf_key_type_id and
3029 * btf_value_type_id.
3030 */
3031 if (map->sec_idx == obj->efile.btf_maps_shndx ||
3032 bpf_map__is_struct_ops(map))
3033 return 0;
3034
3035 if (!bpf_map__is_internal(map)) {
3036 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
3037 def->value_size, &key_type_id,
3038 &value_type_id);
3039 } else {
3040 /*
3041 * LLVM annotates global data differently in BTF, that is,
3042 * only as '.data', '.bss' or '.rodata'.
3043 */
3044 ret = btf__find_by_name(obj->btf,
3045 libbpf_type_to_btf_name[map->libbpf_type]);
3046 }
3047 if (ret < 0)
3048 return ret;
3049
3050 map->btf_key_type_id = key_type_id;
3051 map->btf_value_type_id = bpf_map__is_internal(map) ?
3052 ret : value_type_id;
3053 return 0;
3054 }
3055
3056 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
3057 {
3058 struct bpf_map_info info = {};
3059 __u32 len = sizeof(info);
3060 int new_fd, err;
3061 char *new_name;
3062
3063 err = bpf_obj_get_info_by_fd(fd, &info, &len);
3064 if (err)
3065 return err;
3066
3067 new_name = strdup(info.name);
3068 if (!new_name)
3069 return -errno;
3070
3071 new_fd = open("/", O_RDONLY | O_CLOEXEC);
3072 if (new_fd < 0) {
3073 err = -errno;
3074 goto err_free_new_name;
3075 }
3076
3077 new_fd = dup3(fd, new_fd, O_CLOEXEC);
3078 if (new_fd < 0) {
3079 err = -errno;
3080 goto err_close_new_fd;
3081 }
3082
3083 err = zclose(map->fd);
3084 if (err) {
3085 err = -errno;
3086 goto err_close_new_fd;
3087 }
3088 free(map->name);
3089
3090 map->fd = new_fd;
3091 map->name = new_name;
3092 map->def.type = info.type;
3093 map->def.key_size = info.key_size;
3094 map->def.value_size = info.value_size;
3095 map->def.max_entries = info.max_entries;
3096 map->def.map_flags = info.map_flags;
3097 map->btf_key_type_id = info.btf_key_type_id;
3098 map->btf_value_type_id = info.btf_value_type_id;
3099 map->reused = true;
3100
3101 return 0;
3102
3103 err_close_new_fd:
3104 close(new_fd);
3105 err_free_new_name:
3106 free(new_name);
3107 return err;
3108 }
3109
3110 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
3111 {
3112 if (!map || !max_entries)
3113 return -EINVAL;
3114
3115 /* If map already created, its attributes can't be changed. */
3116 if (map->fd >= 0)
3117 return -EBUSY;
3118
3119 map->def.max_entries = max_entries;
3120
3121 return 0;
3122 }
3123
3124 static int
3125 bpf_object__probe_name(struct bpf_object *obj)
3126 {
3127 struct bpf_load_program_attr attr;
3128 char *cp, errmsg[STRERR_BUFSIZE];
3129 struct bpf_insn insns[] = {
3130 BPF_MOV64_IMM(BPF_REG_0, 0),
3131 BPF_EXIT_INSN(),
3132 };
3133 int ret;
3134
3135 /* make sure basic loading works */
3136
3137 memset(&attr, 0, sizeof(attr));
3138 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3139 attr.insns = insns;
3140 attr.insns_cnt = ARRAY_SIZE(insns);
3141 attr.license = "GPL";
3142
3143 ret = bpf_load_program_xattr(&attr, NULL, 0);
3144 if (ret < 0) {
3145 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3146 pr_warn("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
3147 __func__, cp, errno);
3148 return -errno;
3149 }
3150 close(ret);
3151
3152 /* now try the same program, but with the name */
3153
3154 attr.name = "test";
3155 ret = bpf_load_program_xattr(&attr, NULL, 0);
3156 if (ret >= 0) {
3157 obj->caps.name = 1;
3158 close(ret);
3159 }
3160
3161 return 0;
3162 }
3163
3164 static int
3165 bpf_object__probe_global_data(struct bpf_object *obj)
3166 {
3167 struct bpf_load_program_attr prg_attr;
3168 struct bpf_create_map_attr map_attr;
3169 char *cp, errmsg[STRERR_BUFSIZE];
3170 struct bpf_insn insns[] = {
3171 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
3172 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
3173 BPF_MOV64_IMM(BPF_REG_0, 0),
3174 BPF_EXIT_INSN(),
3175 };
3176 int ret, map;
3177
3178 memset(&map_attr, 0, sizeof(map_attr));
3179 map_attr.map_type = BPF_MAP_TYPE_ARRAY;
3180 map_attr.key_size = sizeof(int);
3181 map_attr.value_size = 32;
3182 map_attr.max_entries = 1;
3183
3184 map = bpf_create_map_xattr(&map_attr);
3185 if (map < 0) {
3186 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3187 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
3188 __func__, cp, errno);
3189 return -errno;
3190 }
3191
3192 insns[0].imm = map;
3193
3194 memset(&prg_attr, 0, sizeof(prg_attr));
3195 prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3196 prg_attr.insns = insns;
3197 prg_attr.insns_cnt = ARRAY_SIZE(insns);
3198 prg_attr.license = "GPL";
3199
3200 ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
3201 if (ret >= 0) {
3202 obj->caps.global_data = 1;
3203 close(ret);
3204 }
3205
3206 close(map);
3207 return 0;
3208 }
3209
3210 static int bpf_object__probe_btf_func(struct bpf_object *obj)
3211 {
3212 static const char strs[] = "\0int\0x\0a";
3213 /* void x(int a) {} */
3214 __u32 types[] = {
3215 /* int */
3216 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
3217 /* FUNC_PROTO */ /* [2] */
3218 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
3219 BTF_PARAM_ENC(7, 1),
3220 /* FUNC x */ /* [3] */
3221 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
3222 };
3223 int btf_fd;
3224
3225 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3226 strs, sizeof(strs));
3227 if (btf_fd >= 0) {
3228 obj->caps.btf_func = 1;
3229 close(btf_fd);
3230 return 1;
3231 }
3232
3233 return 0;
3234 }
3235
3236 static int bpf_object__probe_btf_func_global(struct bpf_object *obj)
3237 {
3238 static const char strs[] = "\0int\0x\0a";
3239 /* static void x(int a) {} */
3240 __u32 types[] = {
3241 /* int */
3242 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
3243 /* FUNC_PROTO */ /* [2] */
3244 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
3245 BTF_PARAM_ENC(7, 1),
3246 /* FUNC x BTF_FUNC_GLOBAL */ /* [3] */
3247 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
3248 };
3249 int btf_fd;
3250
3251 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3252 strs, sizeof(strs));
3253 if (btf_fd >= 0) {
3254 obj->caps.btf_func_global = 1;
3255 close(btf_fd);
3256 return 1;
3257 }
3258
3259 return 0;
3260 }
3261
3262 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
3263 {
3264 static const char strs[] = "\0x\0.data";
3265 /* static int a; */
3266 __u32 types[] = {
3267 /* int */
3268 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
3269 /* VAR x */ /* [2] */
3270 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
3271 BTF_VAR_STATIC,
3272 /* DATASEC val */ /* [3] */
3273 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
3274 BTF_VAR_SECINFO_ENC(2, 0, 4),
3275 };
3276 int btf_fd;
3277
3278 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3279 strs, sizeof(strs));
3280 if (btf_fd >= 0) {
3281 obj->caps.btf_datasec = 1;
3282 close(btf_fd);
3283 return 1;
3284 }
3285
3286 return 0;
3287 }
3288
3289 static int bpf_object__probe_array_mmap(struct bpf_object *obj)
3290 {
3291 struct bpf_create_map_attr attr = {
3292 .map_type = BPF_MAP_TYPE_ARRAY,
3293 .map_flags = BPF_F_MMAPABLE,
3294 .key_size = sizeof(int),
3295 .value_size = sizeof(int),
3296 .max_entries = 1,
3297 };
3298 int fd;
3299
3300 fd = bpf_create_map_xattr(&attr);
3301 if (fd >= 0) {
3302 obj->caps.array_mmap = 1;
3303 close(fd);
3304 return 1;
3305 }
3306
3307 return 0;
3308 }
3309
3310 static int
3311 bpf_object__probe_caps(struct bpf_object *obj)
3312 {
3313 int (*probe_fn[])(struct bpf_object *obj) = {
3314 bpf_object__probe_name,
3315 bpf_object__probe_global_data,
3316 bpf_object__probe_btf_func,
3317 bpf_object__probe_btf_func_global,
3318 bpf_object__probe_btf_datasec,
3319 bpf_object__probe_array_mmap,
3320 };
3321 int i, ret;
3322
3323 for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
3324 ret = probe_fn[i](obj);
3325 if (ret < 0)
3326 pr_debug("Probe #%d failed with %d.\n", i, ret);
3327 }
3328
3329 return 0;
3330 }
3331
3332 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
3333 {
3334 struct bpf_map_info map_info = {};
3335 char msg[STRERR_BUFSIZE];
3336 __u32 map_info_len;
3337
3338 map_info_len = sizeof(map_info);
3339
3340 if (bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len)) {
3341 pr_warn("failed to get map info for map FD %d: %s\n",
3342 map_fd, libbpf_strerror_r(errno, msg, sizeof(msg)));
3343 return false;
3344 }
3345
3346 return (map_info.type == map->def.type &&
3347 map_info.key_size == map->def.key_size &&
3348 map_info.value_size == map->def.value_size &&
3349 map_info.max_entries == map->def.max_entries &&
3350 map_info.map_flags == map->def.map_flags);
3351 }
3352
3353 static int
3354 bpf_object__reuse_map(struct bpf_map *map)
3355 {
3356 char *cp, errmsg[STRERR_BUFSIZE];
3357 int err, pin_fd;
3358
3359 pin_fd = bpf_obj_get(map->pin_path);
3360 if (pin_fd < 0) {
3361 err = -errno;
3362 if (err == -ENOENT) {
3363 pr_debug("found no pinned map to reuse at '%s'\n",
3364 map->pin_path);
3365 return 0;
3366 }
3367
3368 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
3369 pr_warn("couldn't retrieve pinned map '%s': %s\n",
3370 map->pin_path, cp);
3371 return err;
3372 }
3373
3374 if (!map_is_reuse_compat(map, pin_fd)) {
3375 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
3376 map->pin_path);
3377 close(pin_fd);
3378 return -EINVAL;
3379 }
3380
3381 err = bpf_map__reuse_fd(map, pin_fd);
3382 if (err) {
3383 close(pin_fd);
3384 return err;
3385 }
3386 map->pinned = true;
3387 pr_debug("reused pinned map at '%s'\n", map->pin_path);
3388
3389 return 0;
3390 }
3391
3392 static int
3393 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
3394 {
3395 enum libbpf_map_type map_type = map->libbpf_type;
3396 char *cp, errmsg[STRERR_BUFSIZE];
3397 int err, zero = 0;
3398
3399 /* kernel already zero-initializes .bss map. */
3400 if (map_type == LIBBPF_MAP_BSS)
3401 return 0;
3402
3403 err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
3404 if (err) {
3405 err = -errno;
3406 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3407 pr_warn("Error setting initial map(%s) contents: %s\n",
3408 map->name, cp);
3409 return err;
3410 }
3411
3412 /* Freeze .rodata and .kconfig map as read-only from syscall side. */
3413 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
3414 err = bpf_map_freeze(map->fd);
3415 if (err) {
3416 err = -errno;
3417 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3418 pr_warn("Error freezing map(%s) as read-only: %s\n",
3419 map->name, cp);
3420 return err;
3421 }
3422 }
3423 return 0;
3424 }
3425
3426 static int
3427 bpf_object__create_maps(struct bpf_object *obj)
3428 {
3429 struct bpf_create_map_attr create_attr = {};
3430 int nr_cpus = 0;
3431 unsigned int i;
3432 int err;
3433
3434 for (i = 0; i < obj->nr_maps; i++) {
3435 struct bpf_map *map = &obj->maps[i];
3436 struct bpf_map_def *def = &map->def;
3437 char *cp, errmsg[STRERR_BUFSIZE];
3438 int *pfd = &map->fd;
3439
3440 if (map->pin_path) {
3441 err = bpf_object__reuse_map(map);
3442 if (err) {
3443 pr_warn("error reusing pinned map %s\n",
3444 map->name);
3445 return err;
3446 }
3447 }
3448
3449 if (map->fd >= 0) {
3450 pr_debug("skip map create (preset) %s: fd=%d\n",
3451 map->name, map->fd);
3452 continue;
3453 }
3454
3455 if (obj->caps.name)
3456 create_attr.name = map->name;
3457 create_attr.map_ifindex = map->map_ifindex;
3458 create_attr.map_type = def->type;
3459 create_attr.map_flags = def->map_flags;
3460 create_attr.key_size = def->key_size;
3461 create_attr.value_size = def->value_size;
3462 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
3463 !def->max_entries) {
3464 if (!nr_cpus)
3465 nr_cpus = libbpf_num_possible_cpus();
3466 if (nr_cpus < 0) {
3467 pr_warn("failed to determine number of system CPUs: %d\n",
3468 nr_cpus);
3469 err = nr_cpus;
3470 goto err_out;
3471 }
3472 pr_debug("map '%s': setting size to %d\n",
3473 map->name, nr_cpus);
3474 create_attr.max_entries = nr_cpus;
3475 } else {
3476 create_attr.max_entries = def->max_entries;
3477 }
3478 create_attr.btf_fd = 0;
3479 create_attr.btf_key_type_id = 0;
3480 create_attr.btf_value_type_id = 0;
3481 if (bpf_map_type__is_map_in_map(def->type) &&
3482 map->inner_map_fd >= 0)
3483 create_attr.inner_map_fd = map->inner_map_fd;
3484 if (bpf_map__is_struct_ops(map))
3485 create_attr.btf_vmlinux_value_type_id =
3486 map->btf_vmlinux_value_type_id;
3487
3488 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
3489 create_attr.btf_fd = btf__fd(obj->btf);
3490 create_attr.btf_key_type_id = map->btf_key_type_id;
3491 create_attr.btf_value_type_id = map->btf_value_type_id;
3492 }
3493
3494 *pfd = bpf_create_map_xattr(&create_attr);
3495 if (*pfd < 0 && (create_attr.btf_key_type_id ||
3496 create_attr.btf_value_type_id)) {
3497 err = -errno;
3498 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3499 pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
3500 map->name, cp, err);
3501 create_attr.btf_fd = 0;
3502 create_attr.btf_key_type_id = 0;
3503 create_attr.btf_value_type_id = 0;
3504 map->btf_key_type_id = 0;
3505 map->btf_value_type_id = 0;
3506 *pfd = bpf_create_map_xattr(&create_attr);
3507 }
3508
3509 if (*pfd < 0) {
3510 size_t j;
3511
3512 err = -errno;
3513 err_out:
3514 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3515 pr_warn("failed to create map (name: '%s'): %s(%d)\n",
3516 map->name, cp, err);
3517 pr_perm_msg(err);
3518 for (j = 0; j < i; j++)
3519 zclose(obj->maps[j].fd);
3520 return err;
3521 }
3522
3523 if (bpf_map__is_internal(map)) {
3524 err = bpf_object__populate_internal_map(obj, map);
3525 if (err < 0) {
3526 zclose(*pfd);
3527 goto err_out;
3528 }
3529 }
3530
3531 if (map->pin_path && !map->pinned) {
3532 err = bpf_map__pin(map, NULL);
3533 if (err) {
3534 pr_warn("failed to auto-pin map name '%s' at '%s'\n",
3535 map->name, map->pin_path);
3536 return err;
3537 }
3538 }
3539
3540 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
3541 }
3542
3543 return 0;
3544 }
3545
3546 static int
3547 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
3548 void *btf_prog_info, const char *info_name)
3549 {
3550 if (err != -ENOENT) {
3551 pr_warn("Error in loading %s for sec %s.\n",
3552 info_name, prog->section_name);
3553 return err;
3554 }
3555
3556 /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
3557
3558 if (btf_prog_info) {
3559 /*
3560 * Some info has already been found but has problem
3561 * in the last btf_ext reloc. Must have to error out.
3562 */
3563 pr_warn("Error in relocating %s for sec %s.\n",
3564 info_name, prog->section_name);
3565 return err;
3566 }
3567
3568 /* Have problem loading the very first info. Ignore the rest. */
3569 pr_warn("Cannot find %s for main program sec %s. Ignore all %s.\n",
3570 info_name, prog->section_name, info_name);
3571 return 0;
3572 }
3573
3574 static int
3575 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
3576 const char *section_name, __u32 insn_offset)
3577 {
3578 int err;
3579
3580 if (!insn_offset || prog->func_info) {
3581 /*
3582 * !insn_offset => main program
3583 *
3584 * For sub prog, the main program's func_info has to
3585 * be loaded first (i.e. prog->func_info != NULL)
3586 */
3587 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
3588 section_name, insn_offset,
3589 &prog->func_info,
3590 &prog->func_info_cnt);
3591 if (err)
3592 return check_btf_ext_reloc_err(prog, err,
3593 prog->func_info,
3594 "bpf_func_info");
3595
3596 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
3597 }
3598
3599 if (!insn_offset || prog->line_info) {
3600 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
3601 section_name, insn_offset,
3602 &prog->line_info,
3603 &prog->line_info_cnt);
3604 if (err)
3605 return check_btf_ext_reloc_err(prog, err,
3606 prog->line_info,
3607 "bpf_line_info");
3608
3609 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
3610 }
3611
3612 return 0;
3613 }
3614
3615 #define BPF_CORE_SPEC_MAX_LEN 64
3616
3617 /* represents BPF CO-RE field or array element accessor */
3618 struct bpf_core_accessor {
3619 __u32 type_id; /* struct/union type or array element type */
3620 __u32 idx; /* field index or array index */
3621 const char *name; /* field name or NULL for array accessor */
3622 };
3623
3624 struct bpf_core_spec {
3625 const struct btf *btf;
3626 /* high-level spec: named fields and array indices only */
3627 struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
3628 /* high-level spec length */
3629 int len;
3630 /* raw, low-level spec: 1-to-1 with accessor spec string */
3631 int raw_spec[BPF_CORE_SPEC_MAX_LEN];
3632 /* raw spec length */
3633 int raw_len;
3634 /* field bit offset represented by spec */
3635 __u32 bit_offset;
3636 };
3637
3638 static bool str_is_empty(const char *s)
3639 {
3640 return !s || !s[0];
3641 }
3642
3643 static bool is_flex_arr(const struct btf *btf,
3644 const struct bpf_core_accessor *acc,
3645 const struct btf_array *arr)
3646 {
3647 const struct btf_type *t;
3648
3649 /* not a flexible array, if not inside a struct or has non-zero size */
3650 if (!acc->name || arr->nelems > 0)
3651 return false;
3652
3653 /* has to be the last member of enclosing struct */
3654 t = btf__type_by_id(btf, acc->type_id);
3655 return acc->idx == btf_vlen(t) - 1;
3656 }
3657
3658 /*
3659 * Turn bpf_field_reloc into a low- and high-level spec representation,
3660 * validating correctness along the way, as well as calculating resulting
3661 * field bit offset, specified by accessor string. Low-level spec captures
3662 * every single level of nestedness, including traversing anonymous
3663 * struct/union members. High-level one only captures semantically meaningful
3664 * "turning points": named fields and array indicies.
3665 * E.g., for this case:
3666 *
3667 * struct sample {
3668 * int __unimportant;
3669 * struct {
3670 * int __1;
3671 * int __2;
3672 * int a[7];
3673 * };
3674 * };
3675 *
3676 * struct sample *s = ...;
3677 *
3678 * int x = &s->a[3]; // access string = '0:1:2:3'
3679 *
3680 * Low-level spec has 1:1 mapping with each element of access string (it's
3681 * just a parsed access string representation): [0, 1, 2, 3].
3682 *
3683 * High-level spec will capture only 3 points:
3684 * - intial zero-index access by pointer (&s->... is the same as &s[0]...);
3685 * - field 'a' access (corresponds to '2' in low-level spec);
3686 * - array element #3 access (corresponds to '3' in low-level spec).
3687 *
3688 */
3689 static int bpf_core_spec_parse(const struct btf *btf,
3690 __u32 type_id,
3691 const char *spec_str,
3692 struct bpf_core_spec *spec)
3693 {
3694 int access_idx, parsed_len, i;
3695 struct bpf_core_accessor *acc;
3696 const struct btf_type *t;
3697 const char *name;
3698 __u32 id;
3699 __s64 sz;
3700
3701 if (str_is_empty(spec_str) || *spec_str == ':')
3702 return -EINVAL;
3703
3704 memset(spec, 0, sizeof(*spec));
3705 spec->btf = btf;
3706
3707 /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
3708 while (*spec_str) {
3709 if (*spec_str == ':')
3710 ++spec_str;
3711 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
3712 return -EINVAL;
3713 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
3714 return -E2BIG;
3715 spec_str += parsed_len;
3716 spec->raw_spec[spec->raw_len++] = access_idx;
3717 }
3718
3719 if (spec->raw_len == 0)
3720 return -EINVAL;
3721
3722 /* first spec value is always reloc type array index */
3723 t = skip_mods_and_typedefs(btf, type_id, &id);
3724 if (!t)
3725 return -EINVAL;
3726
3727 access_idx = spec->raw_spec[0];
3728 spec->spec[0].type_id = id;
3729 spec->spec[0].idx = access_idx;
3730 spec->len++;
3731
3732 sz = btf__resolve_size(btf, id);
3733 if (sz < 0)
3734 return sz;
3735 spec->bit_offset = access_idx * sz * 8;
3736
3737 for (i = 1; i < spec->raw_len; i++) {
3738 t = skip_mods_and_typedefs(btf, id, &id);
3739 if (!t)
3740 return -EINVAL;
3741
3742 access_idx = spec->raw_spec[i];
3743 acc = &spec->spec[spec->len];
3744
3745 if (btf_is_composite(t)) {
3746 const struct btf_member *m;
3747 __u32 bit_offset;
3748
3749 if (access_idx >= btf_vlen(t))
3750 return -EINVAL;
3751
3752 bit_offset = btf_member_bit_offset(t, access_idx);
3753 spec->bit_offset += bit_offset;
3754
3755 m = btf_members(t) + access_idx;
3756 if (m->name_off) {
3757 name = btf__name_by_offset(btf, m->name_off);
3758 if (str_is_empty(name))
3759 return -EINVAL;
3760
3761 acc->type_id = id;
3762 acc->idx = access_idx;
3763 acc->name = name;
3764 spec->len++;
3765 }
3766
3767 id = m->type;
3768 } else if (btf_is_array(t)) {
3769 const struct btf_array *a = btf_array(t);
3770 bool flex;
3771
3772 t = skip_mods_and_typedefs(btf, a->type, &id);
3773 if (!t)
3774 return -EINVAL;
3775
3776 flex = is_flex_arr(btf, acc - 1, a);
3777 if (!flex && access_idx >= a->nelems)
3778 return -EINVAL;
3779
3780 spec->spec[spec->len].type_id = id;
3781 spec->spec[spec->len].idx = access_idx;
3782 spec->len++;
3783
3784 sz = btf__resolve_size(btf, id);
3785 if (sz < 0)
3786 return sz;
3787 spec->bit_offset += access_idx * sz * 8;
3788 } else {
3789 pr_warn("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
3790 type_id, spec_str, i, id, btf_kind(t));
3791 return -EINVAL;
3792 }
3793 }
3794
3795 return 0;
3796 }
3797
3798 static bool bpf_core_is_flavor_sep(const char *s)
3799 {
3800 /* check X___Y name pattern, where X and Y are not underscores */
3801 return s[0] != '_' && /* X */
3802 s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */
3803 s[4] != '_'; /* Y */
3804 }
3805
3806 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
3807 * before last triple underscore. Struct name part after last triple
3808 * underscore is ignored by BPF CO-RE relocation during relocation matching.
3809 */
3810 static size_t bpf_core_essential_name_len(const char *name)
3811 {
3812 size_t n = strlen(name);
3813 int i;
3814
3815 for (i = n - 5; i >= 0; i--) {
3816 if (bpf_core_is_flavor_sep(name + i))
3817 return i + 1;
3818 }
3819 return n;
3820 }
3821
3822 /* dynamically sized list of type IDs */
3823 struct ids_vec {
3824 __u32 *data;
3825 int len;
3826 };
3827
3828 static void bpf_core_free_cands(struct ids_vec *cand_ids)
3829 {
3830 free(cand_ids->data);
3831 free(cand_ids);
3832 }
3833
3834 static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
3835 __u32 local_type_id,
3836 const struct btf *targ_btf)
3837 {
3838 size_t local_essent_len, targ_essent_len;
3839 const char *local_name, *targ_name;
3840 const struct btf_type *t;
3841 struct ids_vec *cand_ids;
3842 __u32 *new_ids;
3843 int i, err, n;
3844
3845 t = btf__type_by_id(local_btf, local_type_id);
3846 if (!t)
3847 return ERR_PTR(-EINVAL);
3848
3849 local_name = btf__name_by_offset(local_btf, t->name_off);
3850 if (str_is_empty(local_name))
3851 return ERR_PTR(-EINVAL);
3852 local_essent_len = bpf_core_essential_name_len(local_name);
3853
3854 cand_ids = calloc(1, sizeof(*cand_ids));
3855 if (!cand_ids)
3856 return ERR_PTR(-ENOMEM);
3857
3858 n = btf__get_nr_types(targ_btf);
3859 for (i = 1; i <= n; i++) {
3860 t = btf__type_by_id(targ_btf, i);
3861 targ_name = btf__name_by_offset(targ_btf, t->name_off);
3862 if (str_is_empty(targ_name))
3863 continue;
3864
3865 targ_essent_len = bpf_core_essential_name_len(targ_name);
3866 if (targ_essent_len != local_essent_len)
3867 continue;
3868
3869 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
3870 pr_debug("[%d] %s: found candidate [%d] %s\n",
3871 local_type_id, local_name, i, targ_name);
3872 new_ids = reallocarray(cand_ids->data,
3873 cand_ids->len + 1,
3874 sizeof(*cand_ids->data));
3875 if (!new_ids) {
3876 err = -ENOMEM;
3877 goto err_out;
3878 }
3879 cand_ids->data = new_ids;
3880 cand_ids->data[cand_ids->len++] = i;
3881 }
3882 }
3883 return cand_ids;
3884 err_out:
3885 bpf_core_free_cands(cand_ids);
3886 return ERR_PTR(err);
3887 }
3888
3889 /* Check two types for compatibility, skipping const/volatile/restrict and
3890 * typedefs, to ensure we are relocating compatible entities:
3891 * - any two STRUCTs/UNIONs are compatible and can be mixed;
3892 * - any two FWDs are compatible, if their names match (modulo flavor suffix);
3893 * - any two PTRs are always compatible;
3894 * - for ENUMs, names should be the same (ignoring flavor suffix) or at
3895 * least one of enums should be anonymous;
3896 * - for ENUMs, check sizes, names are ignored;
3897 * - for INT, size and signedness are ignored;
3898 * - for ARRAY, dimensionality is ignored, element types are checked for
3899 * compatibility recursively;
3900 * - everything else shouldn't be ever a target of relocation.
3901 * These rules are not set in stone and probably will be adjusted as we get
3902 * more experience with using BPF CO-RE relocations.
3903 */
3904 static int bpf_core_fields_are_compat(const struct btf *local_btf,
3905 __u32 local_id,
3906 const struct btf *targ_btf,
3907 __u32 targ_id)
3908 {
3909 const struct btf_type *local_type, *targ_type;
3910
3911 recur:
3912 local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
3913 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
3914 if (!local_type || !targ_type)
3915 return -EINVAL;
3916
3917 if (btf_is_composite(local_type) && btf_is_composite(targ_type))
3918 return 1;
3919 if (btf_kind(local_type) != btf_kind(targ_type))
3920 return 0;
3921
3922 switch (btf_kind(local_type)) {
3923 case BTF_KIND_PTR:
3924 return 1;
3925 case BTF_KIND_FWD:
3926 case BTF_KIND_ENUM: {
3927 const char *local_name, *targ_name;
3928 size_t local_len, targ_len;
3929
3930 local_name = btf__name_by_offset(local_btf,
3931 local_type->name_off);
3932 targ_name = btf__name_by_offset(targ_btf, targ_type->name_off);
3933 local_len = bpf_core_essential_name_len(local_name);
3934 targ_len = bpf_core_essential_name_len(targ_name);
3935 /* one of them is anonymous or both w/ same flavor-less names */
3936 return local_len == 0 || targ_len == 0 ||
3937 (local_len == targ_len &&
3938 strncmp(local_name, targ_name, local_len) == 0);
3939 }
3940 case BTF_KIND_INT:
3941 /* just reject deprecated bitfield-like integers; all other
3942 * integers are by default compatible between each other
3943 */
3944 return btf_int_offset(local_type) == 0 &&
3945 btf_int_offset(targ_type) == 0;
3946 case BTF_KIND_ARRAY:
3947 local_id = btf_array(local_type)->type;
3948 targ_id = btf_array(targ_type)->type;
3949 goto recur;
3950 default:
3951 pr_warn("unexpected kind %d relocated, local [%d], target [%d]\n",
3952 btf_kind(local_type), local_id, targ_id);
3953 return 0;
3954 }
3955 }
3956
3957 /*
3958 * Given single high-level named field accessor in local type, find
3959 * corresponding high-level accessor for a target type. Along the way,
3960 * maintain low-level spec for target as well. Also keep updating target
3961 * bit offset.
3962 *
3963 * Searching is performed through recursive exhaustive enumeration of all
3964 * fields of a struct/union. If there are any anonymous (embedded)
3965 * structs/unions, they are recursively searched as well. If field with
3966 * desired name is found, check compatibility between local and target types,
3967 * before returning result.
3968 *
3969 * 1 is returned, if field is found.
3970 * 0 is returned if no compatible field is found.
3971 * <0 is returned on error.
3972 */
3973 static int bpf_core_match_member(const struct btf *local_btf,
3974 const struct bpf_core_accessor *local_acc,
3975 const struct btf *targ_btf,
3976 __u32 targ_id,
3977 struct bpf_core_spec *spec,
3978 __u32 *next_targ_id)
3979 {
3980 const struct btf_type *local_type, *targ_type;
3981 const struct btf_member *local_member, *m;
3982 const char *local_name, *targ_name;
3983 __u32 local_id;
3984 int i, n, found;
3985
3986 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
3987 if (!targ_type)
3988 return -EINVAL;
3989 if (!btf_is_composite(targ_type))
3990 return 0;
3991
3992 local_id = local_acc->type_id;
3993 local_type = btf__type_by_id(local_btf, local_id);
3994 local_member = btf_members(local_type) + local_acc->idx;
3995 local_name = btf__name_by_offset(local_btf, local_member->name_off);
3996
3997 n = btf_vlen(targ_type);
3998 m = btf_members(targ_type);
3999 for (i = 0; i < n; i++, m++) {
4000 __u32 bit_offset;
4001
4002 bit_offset = btf_member_bit_offset(targ_type, i);
4003
4004 /* too deep struct/union/array nesting */
4005 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4006 return -E2BIG;
4007
4008 /* speculate this member will be the good one */
4009 spec->bit_offset += bit_offset;
4010 spec->raw_spec[spec->raw_len++] = i;
4011
4012 targ_name = btf__name_by_offset(targ_btf, m->name_off);
4013 if (str_is_empty(targ_name)) {
4014 /* embedded struct/union, we need to go deeper */
4015 found = bpf_core_match_member(local_btf, local_acc,
4016 targ_btf, m->type,
4017 spec, next_targ_id);
4018 if (found) /* either found or error */
4019 return found;
4020 } else if (strcmp(local_name, targ_name) == 0) {
4021 /* matching named field */
4022 struct bpf_core_accessor *targ_acc;
4023
4024 targ_acc = &spec->spec[spec->len++];
4025 targ_acc->type_id = targ_id;
4026 targ_acc->idx = i;
4027 targ_acc->name = targ_name;
4028
4029 *next_targ_id = m->type;
4030 found = bpf_core_fields_are_compat(local_btf,
4031 local_member->type,
4032 targ_btf, m->type);
4033 if (!found)
4034 spec->len--; /* pop accessor */
4035 return found;
4036 }
4037 /* member turned out not to be what we looked for */
4038 spec->bit_offset -= bit_offset;
4039 spec->raw_len--;
4040 }
4041
4042 return 0;
4043 }
4044
4045 /*
4046 * Try to match local spec to a target type and, if successful, produce full
4047 * target spec (high-level, low-level + bit offset).
4048 */
4049 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
4050 const struct btf *targ_btf, __u32 targ_id,
4051 struct bpf_core_spec *targ_spec)
4052 {
4053 const struct btf_type *targ_type;
4054 const struct bpf_core_accessor *local_acc;
4055 struct bpf_core_accessor *targ_acc;
4056 int i, sz, matched;
4057
4058 memset(targ_spec, 0, sizeof(*targ_spec));
4059 targ_spec->btf = targ_btf;
4060
4061 local_acc = &local_spec->spec[0];
4062 targ_acc = &targ_spec->spec[0];
4063
4064 for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
4065 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
4066 &targ_id);
4067 if (!targ_type)
4068 return -EINVAL;
4069
4070 if (local_acc->name) {
4071 matched = bpf_core_match_member(local_spec->btf,
4072 local_acc,
4073 targ_btf, targ_id,
4074 targ_spec, &targ_id);
4075 if (matched <= 0)
4076 return matched;
4077 } else {
4078 /* for i=0, targ_id is already treated as array element
4079 * type (because it's the original struct), for others
4080 * we should find array element type first
4081 */
4082 if (i > 0) {
4083 const struct btf_array *a;
4084 bool flex;
4085
4086 if (!btf_is_array(targ_type))
4087 return 0;
4088
4089 a = btf_array(targ_type);
4090 flex = is_flex_arr(targ_btf, targ_acc - 1, a);
4091 if (!flex && local_acc->idx >= a->nelems)
4092 return 0;
4093 if (!skip_mods_and_typedefs(targ_btf, a->type,
4094 &targ_id))
4095 return -EINVAL;
4096 }
4097
4098 /* too deep struct/union/array nesting */
4099 if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4100 return -E2BIG;
4101
4102 targ_acc->type_id = targ_id;
4103 targ_acc->idx = local_acc->idx;
4104 targ_acc->name = NULL;
4105 targ_spec->len++;
4106 targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
4107 targ_spec->raw_len++;
4108
4109 sz = btf__resolve_size(targ_btf, targ_id);
4110 if (sz < 0)
4111 return sz;
4112 targ_spec->bit_offset += local_acc->idx * sz * 8;
4113 }
4114 }
4115
4116 return 1;
4117 }
4118
4119 static int bpf_core_calc_field_relo(const struct bpf_program *prog,
4120 const struct bpf_field_reloc *relo,
4121 const struct bpf_core_spec *spec,
4122 __u32 *val, bool *validate)
4123 {
4124 const struct bpf_core_accessor *acc = &spec->spec[spec->len - 1];
4125 const struct btf_type *t = btf__type_by_id(spec->btf, acc->type_id);
4126 __u32 byte_off, byte_sz, bit_off, bit_sz;
4127 const struct btf_member *m;
4128 const struct btf_type *mt;
4129 bool bitfield;
4130 __s64 sz;
4131
4132 /* a[n] accessor needs special handling */
4133 if (!acc->name) {
4134 if (relo->kind == BPF_FIELD_BYTE_OFFSET) {
4135 *val = spec->bit_offset / 8;
4136 } else if (relo->kind == BPF_FIELD_BYTE_SIZE) {
4137 sz = btf__resolve_size(spec->btf, acc->type_id);
4138 if (sz < 0)
4139 return -EINVAL;
4140 *val = sz;
4141 } else {
4142 pr_warn("prog '%s': relo %d at insn #%d can't be applied to array access\n",
4143 bpf_program__title(prog, false),
4144 relo->kind, relo->insn_off / 8);
4145 return -EINVAL;
4146 }
4147 if (validate)
4148 *validate = true;
4149 return 0;
4150 }
4151
4152 m = btf_members(t) + acc->idx;
4153 mt = skip_mods_and_typedefs(spec->btf, m->type, NULL);
4154 bit_off = spec->bit_offset;
4155 bit_sz = btf_member_bitfield_size(t, acc->idx);
4156
4157 bitfield = bit_sz > 0;
4158 if (bitfield) {
4159 byte_sz = mt->size;
4160 byte_off = bit_off / 8 / byte_sz * byte_sz;
4161 /* figure out smallest int size necessary for bitfield load */
4162 while (bit_off + bit_sz - byte_off * 8 > byte_sz * 8) {
4163 if (byte_sz >= 8) {
4164 /* bitfield can't be read with 64-bit read */
4165 pr_warn("prog '%s': relo %d at insn #%d can't be satisfied for bitfield\n",
4166 bpf_program__title(prog, false),
4167 relo->kind, relo->insn_off / 8);
4168 return -E2BIG;
4169 }
4170 byte_sz *= 2;
4171 byte_off = bit_off / 8 / byte_sz * byte_sz;
4172 }
4173 } else {
4174 sz = btf__resolve_size(spec->btf, m->type);
4175 if (sz < 0)
4176 return -EINVAL;
4177 byte_sz = sz;
4178 byte_off = spec->bit_offset / 8;
4179 bit_sz = byte_sz * 8;
4180 }
4181
4182 /* for bitfields, all the relocatable aspects are ambiguous and we
4183 * might disagree with compiler, so turn off validation of expected
4184 * value, except for signedness
4185 */
4186 if (validate)
4187 *validate = !bitfield;
4188
4189 switch (relo->kind) {
4190 case BPF_FIELD_BYTE_OFFSET:
4191 *val = byte_off;
4192 break;
4193 case BPF_FIELD_BYTE_SIZE:
4194 *val = byte_sz;
4195 break;
4196 case BPF_FIELD_SIGNED:
4197 /* enums will be assumed unsigned */
4198 *val = btf_is_enum(mt) ||
4199 (btf_int_encoding(mt) & BTF_INT_SIGNED);
4200 if (validate)
4201 *validate = true; /* signedness is never ambiguous */
4202 break;
4203 case BPF_FIELD_LSHIFT_U64:
4204 #if __BYTE_ORDER == __LITTLE_ENDIAN
4205 *val = 64 - (bit_off + bit_sz - byte_off * 8);
4206 #else
4207 *val = (8 - byte_sz) * 8 + (bit_off - byte_off * 8);
4208 #endif
4209 break;
4210 case BPF_FIELD_RSHIFT_U64:
4211 *val = 64 - bit_sz;
4212 if (validate)
4213 *validate = true; /* right shift is never ambiguous */
4214 break;
4215 case BPF_FIELD_EXISTS:
4216 default:
4217 pr_warn("prog '%s': unknown relo %d at insn #%d\n",
4218 bpf_program__title(prog, false),
4219 relo->kind, relo->insn_off / 8);
4220 return -EINVAL;
4221 }
4222
4223 return 0;
4224 }
4225
4226 /*
4227 * Patch relocatable BPF instruction.
4228 *
4229 * Patched value is determined by relocation kind and target specification.
4230 * For field existence relocation target spec will be NULL if field is not
4231 * found.
4232 * Expected insn->imm value is determined using relocation kind and local
4233 * spec, and is checked before patching instruction. If actual insn->imm value
4234 * is wrong, bail out with error.
4235 *
4236 * Currently three kinds of BPF instructions are supported:
4237 * 1. rX = <imm> (assignment with immediate operand);
4238 * 2. rX += <imm> (arithmetic operations with immediate operand);
4239 */
4240 static int bpf_core_reloc_insn(struct bpf_program *prog,
4241 const struct bpf_field_reloc *relo,
4242 int relo_idx,
4243 const struct bpf_core_spec *local_spec,
4244 const struct bpf_core_spec *targ_spec)
4245 {
4246 __u32 orig_val, new_val;
4247 struct bpf_insn *insn;
4248 bool validate = true;
4249 int insn_idx, err;
4250 __u8 class;
4251
4252 if (relo->insn_off % sizeof(struct bpf_insn))
4253 return -EINVAL;
4254 insn_idx = relo->insn_off / sizeof(struct bpf_insn);
4255 insn = &prog->insns[insn_idx];
4256 class = BPF_CLASS(insn->code);
4257
4258 if (relo->kind == BPF_FIELD_EXISTS) {
4259 orig_val = 1; /* can't generate EXISTS relo w/o local field */
4260 new_val = targ_spec ? 1 : 0;
4261 } else if (!targ_spec) {
4262 pr_debug("prog '%s': relo #%d: substituting insn #%d w/ invalid insn\n",
4263 bpf_program__title(prog, false), relo_idx, insn_idx);
4264 insn->code = BPF_JMP | BPF_CALL;
4265 insn->dst_reg = 0;
4266 insn->src_reg = 0;
4267 insn->off = 0;
4268 /* if this instruction is reachable (not a dead code),
4269 * verifier will complain with the following message:
4270 * invalid func unknown#195896080
4271 */
4272 insn->imm = 195896080; /* => 0xbad2310 => "bad relo" */
4273 return 0;
4274 } else {
4275 err = bpf_core_calc_field_relo(prog, relo, local_spec,
4276 &orig_val, &validate);
4277 if (err)
4278 return err;
4279 err = bpf_core_calc_field_relo(prog, relo, targ_spec,
4280 &new_val, NULL);
4281 if (err)
4282 return err;
4283 }
4284
4285 switch (class) {
4286 case BPF_ALU:
4287 case BPF_ALU64:
4288 if (BPF_SRC(insn->code) != BPF_K)
4289 return -EINVAL;
4290 if (validate && insn->imm != orig_val) {
4291 pr_warn("prog '%s': relo #%d: unexpected insn #%d (ALU/ALU64) value: got %u, exp %u -> %u\n",
4292 bpf_program__title(prog, false), relo_idx,
4293 insn_idx, insn->imm, orig_val, new_val);
4294 return -EINVAL;
4295 }
4296 orig_val = insn->imm;
4297 insn->imm = new_val;
4298 pr_debug("prog '%s': relo #%d: patched insn #%d (ALU/ALU64) imm %u -> %u\n",
4299 bpf_program__title(prog, false), relo_idx, insn_idx,
4300 orig_val, new_val);
4301 break;
4302 case BPF_LDX:
4303 case BPF_ST:
4304 case BPF_STX:
4305 if (validate && insn->off != orig_val) {
4306 pr_warn("prog '%s': relo #%d: unexpected insn #%d (LD/LDX/ST/STX) value: got %u, exp %u -> %u\n",
4307 bpf_program__title(prog, false), relo_idx,
4308 insn_idx, insn->off, orig_val, new_val);
4309 return -EINVAL;
4310 }
4311 if (new_val > SHRT_MAX) {
4312 pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) value too big: %u\n",
4313 bpf_program__title(prog, false), relo_idx,
4314 insn_idx, new_val);
4315 return -ERANGE;
4316 }
4317 orig_val = insn->off;
4318 insn->off = new_val;
4319 pr_debug("prog '%s': relo #%d: patched insn #%d (LDX/ST/STX) off %u -> %u\n",
4320 bpf_program__title(prog, false), relo_idx, insn_idx,
4321 orig_val, new_val);
4322 break;
4323 default:
4324 pr_warn("prog '%s': relo #%d: trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
4325 bpf_program__title(prog, false), relo_idx,
4326 insn_idx, insn->code, insn->src_reg, insn->dst_reg,
4327 insn->off, insn->imm);
4328 return -EINVAL;
4329 }
4330
4331 return 0;
4332 }
4333
4334 /* Output spec definition in the format:
4335 * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
4336 * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
4337 */
4338 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
4339 {
4340 const struct btf_type *t;
4341 const char *s;
4342 __u32 type_id;
4343 int i;
4344
4345 type_id = spec->spec[0].type_id;
4346 t = btf__type_by_id(spec->btf, type_id);
4347 s = btf__name_by_offset(spec->btf, t->name_off);
4348 libbpf_print(level, "[%u] %s + ", type_id, s);
4349
4350 for (i = 0; i < spec->raw_len; i++)
4351 libbpf_print(level, "%d%s", spec->raw_spec[i],
4352 i == spec->raw_len - 1 ? " => " : ":");
4353
4354 libbpf_print(level, "%u.%u @ &x",
4355 spec->bit_offset / 8, spec->bit_offset % 8);
4356
4357 for (i = 0; i < spec->len; i++) {
4358 if (spec->spec[i].name)
4359 libbpf_print(level, ".%s", spec->spec[i].name);
4360 else
4361 libbpf_print(level, "[%u]", spec->spec[i].idx);
4362 }
4363
4364 }
4365
4366 static size_t bpf_core_hash_fn(const void *key, void *ctx)
4367 {
4368 return (size_t)key;
4369 }
4370
4371 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
4372 {
4373 return k1 == k2;
4374 }
4375
4376 static void *u32_as_hash_key(__u32 x)
4377 {
4378 return (void *)(uintptr_t)x;
4379 }
4380
4381 /*
4382 * CO-RE relocate single instruction.
4383 *
4384 * The outline and important points of the algorithm:
4385 * 1. For given local type, find corresponding candidate target types.
4386 * Candidate type is a type with the same "essential" name, ignoring
4387 * everything after last triple underscore (___). E.g., `sample`,
4388 * `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
4389 * for each other. Names with triple underscore are referred to as
4390 * "flavors" and are useful, among other things, to allow to
4391 * specify/support incompatible variations of the same kernel struct, which
4392 * might differ between different kernel versions and/or build
4393 * configurations.
4394 *
4395 * N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
4396 * converter, when deduplicated BTF of a kernel still contains more than
4397 * one different types with the same name. In that case, ___2, ___3, etc
4398 * are appended starting from second name conflict. But start flavors are
4399 * also useful to be defined "locally", in BPF program, to extract same
4400 * data from incompatible changes between different kernel
4401 * versions/configurations. For instance, to handle field renames between
4402 * kernel versions, one can use two flavors of the struct name with the
4403 * same common name and use conditional relocations to extract that field,
4404 * depending on target kernel version.
4405 * 2. For each candidate type, try to match local specification to this
4406 * candidate target type. Matching involves finding corresponding
4407 * high-level spec accessors, meaning that all named fields should match,
4408 * as well as all array accesses should be within the actual bounds. Also,
4409 * types should be compatible (see bpf_core_fields_are_compat for details).
4410 * 3. It is supported and expected that there might be multiple flavors
4411 * matching the spec. As long as all the specs resolve to the same set of
4412 * offsets across all candidates, there is no error. If there is any
4413 * ambiguity, CO-RE relocation will fail. This is necessary to accomodate
4414 * imprefection of BTF deduplication, which can cause slight duplication of
4415 * the same BTF type, if some directly or indirectly referenced (by
4416 * pointer) type gets resolved to different actual types in different
4417 * object files. If such situation occurs, deduplicated BTF will end up
4418 * with two (or more) structurally identical types, which differ only in
4419 * types they refer to through pointer. This should be OK in most cases and
4420 * is not an error.
4421 * 4. Candidate types search is performed by linearly scanning through all
4422 * types in target BTF. It is anticipated that this is overall more
4423 * efficient memory-wise and not significantly worse (if not better)
4424 * CPU-wise compared to prebuilding a map from all local type names to
4425 * a list of candidate type names. It's also sped up by caching resolved
4426 * list of matching candidates per each local "root" type ID, that has at
4427 * least one bpf_field_reloc associated with it. This list is shared
4428 * between multiple relocations for the same type ID and is updated as some
4429 * of the candidates are pruned due to structural incompatibility.
4430 */
4431 static int bpf_core_reloc_field(struct bpf_program *prog,
4432 const struct bpf_field_reloc *relo,
4433 int relo_idx,
4434 const struct btf *local_btf,
4435 const struct btf *targ_btf,
4436 struct hashmap *cand_cache)
4437 {
4438 const char *prog_name = bpf_program__title(prog, false);
4439 struct bpf_core_spec local_spec, cand_spec, targ_spec;
4440 const void *type_key = u32_as_hash_key(relo->type_id);
4441 const struct btf_type *local_type, *cand_type;
4442 const char *local_name, *cand_name;
4443 struct ids_vec *cand_ids;
4444 __u32 local_id, cand_id;
4445 const char *spec_str;
4446 int i, j, err;
4447
4448 local_id = relo->type_id;
4449 local_type = btf__type_by_id(local_btf, local_id);
4450 if (!local_type)
4451 return -EINVAL;
4452
4453 local_name = btf__name_by_offset(local_btf, local_type->name_off);
4454 if (str_is_empty(local_name))
4455 return -EINVAL;
4456
4457 spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
4458 if (str_is_empty(spec_str))
4459 return -EINVAL;
4460
4461 err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
4462 if (err) {
4463 pr_warn("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
4464 prog_name, relo_idx, local_id, local_name, spec_str,
4465 err);
4466 return -EINVAL;
4467 }
4468
4469 pr_debug("prog '%s': relo #%d: kind %d, spec is ", prog_name, relo_idx,
4470 relo->kind);
4471 bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
4472 libbpf_print(LIBBPF_DEBUG, "\n");
4473
4474 if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
4475 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
4476 if (IS_ERR(cand_ids)) {
4477 pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
4478 prog_name, relo_idx, local_id, local_name,
4479 PTR_ERR(cand_ids));
4480 return PTR_ERR(cand_ids);
4481 }
4482 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
4483 if (err) {
4484 bpf_core_free_cands(cand_ids);
4485 return err;
4486 }
4487 }
4488
4489 for (i = 0, j = 0; i < cand_ids->len; i++) {
4490 cand_id = cand_ids->data[i];
4491 cand_type = btf__type_by_id(targ_btf, cand_id);
4492 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
4493
4494 err = bpf_core_spec_match(&local_spec, targ_btf,
4495 cand_id, &cand_spec);
4496 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
4497 prog_name, relo_idx, i, cand_name);
4498 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
4499 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
4500 if (err < 0) {
4501 pr_warn("prog '%s': relo #%d: matching error: %d\n",
4502 prog_name, relo_idx, err);
4503 return err;
4504 }
4505 if (err == 0)
4506 continue;
4507
4508 if (j == 0) {
4509 targ_spec = cand_spec;
4510 } else if (cand_spec.bit_offset != targ_spec.bit_offset) {
4511 /* if there are many candidates, they should all
4512 * resolve to the same bit offset
4513 */
4514 pr_warn("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
4515 prog_name, relo_idx, cand_spec.bit_offset,
4516 targ_spec.bit_offset);
4517 return -EINVAL;
4518 }
4519
4520 cand_ids->data[j++] = cand_spec.spec[0].type_id;
4521 }
4522
4523 /*
4524 * For BPF_FIELD_EXISTS relo or when used BPF program has field
4525 * existence checks or kernel version/config checks, it's expected
4526 * that we might not find any candidates. In this case, if field
4527 * wasn't found in any candidate, the list of candidates shouldn't
4528 * change at all, we'll just handle relocating appropriately,
4529 * depending on relo's kind.
4530 */
4531 if (j > 0)
4532 cand_ids->len = j;
4533
4534 /*
4535 * If no candidates were found, it might be both a programmer error,
4536 * as well as expected case, depending whether instruction w/
4537 * relocation is guarded in some way that makes it unreachable (dead
4538 * code) if relocation can't be resolved. This is handled in
4539 * bpf_core_reloc_insn() uniformly by replacing that instruction with
4540 * BPF helper call insn (using invalid helper ID). If that instruction
4541 * is indeed unreachable, then it will be ignored and eliminated by
4542 * verifier. If it was an error, then verifier will complain and point
4543 * to a specific instruction number in its log.
4544 */
4545 if (j == 0)
4546 pr_debug("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
4547 prog_name, relo_idx, local_id, local_name, spec_str);
4548
4549 /* bpf_core_reloc_insn should know how to handle missing targ_spec */
4550 err = bpf_core_reloc_insn(prog, relo, relo_idx, &local_spec,
4551 j ? &targ_spec : NULL);
4552 if (err) {
4553 pr_warn("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
4554 prog_name, relo_idx, relo->insn_off, err);
4555 return -EINVAL;
4556 }
4557
4558 return 0;
4559 }
4560
4561 static int
4562 bpf_core_reloc_fields(struct bpf_object *obj, const char *targ_btf_path)
4563 {
4564 const struct btf_ext_info_sec *sec;
4565 const struct bpf_field_reloc *rec;
4566 const struct btf_ext_info *seg;
4567 struct hashmap_entry *entry;
4568 struct hashmap *cand_cache = NULL;
4569 struct bpf_program *prog;
4570 struct btf *targ_btf;
4571 const char *sec_name;
4572 int i, err = 0;
4573
4574 if (targ_btf_path)
4575 targ_btf = btf__parse_elf(targ_btf_path, NULL);
4576 else
4577 targ_btf = libbpf_find_kernel_btf();
4578 if (IS_ERR(targ_btf)) {
4579 pr_warn("failed to get target BTF: %ld\n", PTR_ERR(targ_btf));
4580 return PTR_ERR(targ_btf);
4581 }
4582
4583 cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
4584 if (IS_ERR(cand_cache)) {
4585 err = PTR_ERR(cand_cache);
4586 goto out;
4587 }
4588
4589 seg = &obj->btf_ext->field_reloc_info;
4590 for_each_btf_ext_sec(seg, sec) {
4591 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
4592 if (str_is_empty(sec_name)) {
4593 err = -EINVAL;
4594 goto out;
4595 }
4596 prog = bpf_object__find_program_by_title(obj, sec_name);
4597 if (!prog) {
4598 pr_warn("failed to find program '%s' for CO-RE offset relocation\n",
4599 sec_name);
4600 err = -EINVAL;
4601 goto out;
4602 }
4603
4604 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
4605 sec_name, sec->num_info);
4606
4607 for_each_btf_ext_rec(seg, sec, i, rec) {
4608 err = bpf_core_reloc_field(prog, rec, i, obj->btf,
4609 targ_btf, cand_cache);
4610 if (err) {
4611 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
4612 sec_name, i, err);
4613 goto out;
4614 }
4615 }
4616 }
4617
4618 out:
4619 btf__free(targ_btf);
4620 if (!IS_ERR_OR_NULL(cand_cache)) {
4621 hashmap__for_each_entry(cand_cache, entry, i) {
4622 bpf_core_free_cands(entry->value);
4623 }
4624 hashmap__free(cand_cache);
4625 }
4626 return err;
4627 }
4628
4629 static int
4630 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
4631 {
4632 int err = 0;
4633
4634 if (obj->btf_ext->field_reloc_info.len)
4635 err = bpf_core_reloc_fields(obj, targ_btf_path);
4636
4637 return err;
4638 }
4639
4640 static int
4641 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
4642 struct reloc_desc *relo)
4643 {
4644 struct bpf_insn *insn, *new_insn;
4645 struct bpf_program *text;
4646 size_t new_cnt;
4647 int err;
4648
4649 if (prog->idx != obj->efile.text_shndx && prog->main_prog_cnt == 0) {
4650 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
4651 if (!text) {
4652 pr_warn("no .text section found yet relo into text exist\n");
4653 return -LIBBPF_ERRNO__RELOC;
4654 }
4655 new_cnt = prog->insns_cnt + text->insns_cnt;
4656 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
4657 if (!new_insn) {
4658 pr_warn("oom in prog realloc\n");
4659 return -ENOMEM;
4660 }
4661 prog->insns = new_insn;
4662
4663 if (obj->btf_ext) {
4664 err = bpf_program_reloc_btf_ext(prog, obj,
4665 text->section_name,
4666 prog->insns_cnt);
4667 if (err)
4668 return err;
4669 }
4670
4671 memcpy(new_insn + prog->insns_cnt, text->insns,
4672 text->insns_cnt * sizeof(*insn));
4673 prog->main_prog_cnt = prog->insns_cnt;
4674 prog->insns_cnt = new_cnt;
4675 pr_debug("added %zd insn from %s to prog %s\n",
4676 text->insns_cnt, text->section_name,
4677 prog->section_name);
4678 }
4679
4680 insn = &prog->insns[relo->insn_idx];
4681 insn->imm += relo->sym_off / 8 + prog->main_prog_cnt - relo->insn_idx;
4682 return 0;
4683 }
4684
4685 static int
4686 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
4687 {
4688 int i, err;
4689
4690 if (!prog)
4691 return 0;
4692
4693 if (obj->btf_ext) {
4694 err = bpf_program_reloc_btf_ext(prog, obj,
4695 prog->section_name, 0);
4696 if (err)
4697 return err;
4698 }
4699
4700 if (!prog->reloc_desc)
4701 return 0;
4702
4703 for (i = 0; i < prog->nr_reloc; i++) {
4704 struct reloc_desc *relo = &prog->reloc_desc[i];
4705 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
4706
4707 if (relo->insn_idx + 1 >= (int)prog->insns_cnt) {
4708 pr_warn("relocation out of range: '%s'\n",
4709 prog->section_name);
4710 return -LIBBPF_ERRNO__RELOC;
4711 }
4712
4713 switch (relo->type) {
4714 case RELO_LD64:
4715 insn[0].src_reg = BPF_PSEUDO_MAP_FD;
4716 insn[0].imm = obj->maps[relo->map_idx].fd;
4717 break;
4718 case RELO_DATA:
4719 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
4720 insn[1].imm = insn[0].imm + relo->sym_off;
4721 insn[0].imm = obj->maps[relo->map_idx].fd;
4722 break;
4723 case RELO_EXTERN:
4724 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
4725 insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
4726 insn[1].imm = relo->sym_off;
4727 break;
4728 case RELO_CALL:
4729 err = bpf_program__reloc_text(prog, obj, relo);
4730 if (err)
4731 return err;
4732 break;
4733 default:
4734 pr_warn("relo #%d: bad relo type %d\n", i, relo->type);
4735 return -EINVAL;
4736 }
4737 }
4738
4739 zfree(&prog->reloc_desc);
4740 prog->nr_reloc = 0;
4741 return 0;
4742 }
4743
4744 static int
4745 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
4746 {
4747 struct bpf_program *prog;
4748 size_t i;
4749 int err;
4750
4751 if (obj->btf_ext) {
4752 err = bpf_object__relocate_core(obj, targ_btf_path);
4753 if (err) {
4754 pr_warn("failed to perform CO-RE relocations: %d\n",
4755 err);
4756 return err;
4757 }
4758 }
4759 /* ensure .text is relocated first, as it's going to be copied as-is
4760 * later for sub-program calls
4761 */
4762 for (i = 0; i < obj->nr_programs; i++) {
4763 prog = &obj->programs[i];
4764 if (prog->idx != obj->efile.text_shndx)
4765 continue;
4766
4767 err = bpf_program__relocate(prog, obj);
4768 if (err) {
4769 pr_warn("failed to relocate '%s'\n", prog->section_name);
4770 return err;
4771 }
4772 break;
4773 }
4774 /* now relocate everything but .text, which by now is relocated
4775 * properly, so we can copy raw sub-program instructions as is safely
4776 */
4777 for (i = 0; i < obj->nr_programs; i++) {
4778 prog = &obj->programs[i];
4779 if (prog->idx == obj->efile.text_shndx)
4780 continue;
4781
4782 err = bpf_program__relocate(prog, obj);
4783 if (err) {
4784 pr_warn("failed to relocate '%s'\n", prog->section_name);
4785 return err;
4786 }
4787 }
4788 return 0;
4789 }
4790
4791 static int bpf_object__collect_struct_ops_map_reloc(struct bpf_object *obj,
4792 GElf_Shdr *shdr,
4793 Elf_Data *data);
4794
4795 static int bpf_object__collect_reloc(struct bpf_object *obj)
4796 {
4797 int i, err;
4798
4799 if (!obj_elf_valid(obj)) {
4800 pr_warn("Internal error: elf object is closed\n");
4801 return -LIBBPF_ERRNO__INTERNAL;
4802 }
4803
4804 for (i = 0; i < obj->efile.nr_reloc_sects; i++) {
4805 GElf_Shdr *shdr = &obj->efile.reloc_sects[i].shdr;
4806 Elf_Data *data = obj->efile.reloc_sects[i].data;
4807 int idx = shdr->sh_info;
4808 struct bpf_program *prog;
4809
4810 if (shdr->sh_type != SHT_REL) {
4811 pr_warn("internal error at %d\n", __LINE__);
4812 return -LIBBPF_ERRNO__INTERNAL;
4813 }
4814
4815 if (idx == obj->efile.st_ops_shndx) {
4816 err = bpf_object__collect_struct_ops_map_reloc(obj,
4817 shdr,
4818 data);
4819 if (err)
4820 return err;
4821 continue;
4822 }
4823
4824 prog = bpf_object__find_prog_by_idx(obj, idx);
4825 if (!prog) {
4826 pr_warn("relocation failed: no section(%d)\n", idx);
4827 return -LIBBPF_ERRNO__RELOC;
4828 }
4829
4830 err = bpf_program__collect_reloc(prog, shdr, data, obj);
4831 if (err)
4832 return err;
4833 }
4834 return 0;
4835 }
4836
4837 static int
4838 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
4839 char *license, __u32 kern_version, int *pfd)
4840 {
4841 struct bpf_load_program_attr load_attr;
4842 char *cp, errmsg[STRERR_BUFSIZE];
4843 int log_buf_size = BPF_LOG_BUF_SIZE;
4844 char *log_buf;
4845 int btf_fd, ret;
4846
4847 if (!insns || !insns_cnt)
4848 return -EINVAL;
4849
4850 memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
4851 load_attr.prog_type = prog->type;
4852 load_attr.expected_attach_type = prog->expected_attach_type;
4853 if (prog->caps->name)
4854 load_attr.name = prog->name;
4855 load_attr.insns = insns;
4856 load_attr.insns_cnt = insns_cnt;
4857 load_attr.license = license;
4858 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) {
4859 load_attr.attach_btf_id = prog->attach_btf_id;
4860 } else if (prog->type == BPF_PROG_TYPE_TRACING ||
4861 prog->type == BPF_PROG_TYPE_EXT) {
4862 load_attr.attach_prog_fd = prog->attach_prog_fd;
4863 load_attr.attach_btf_id = prog->attach_btf_id;
4864 } else {
4865 load_attr.kern_version = kern_version;
4866 load_attr.prog_ifindex = prog->prog_ifindex;
4867 }
4868 /* if .BTF.ext was loaded, kernel supports associated BTF for prog */
4869 if (prog->obj->btf_ext)
4870 btf_fd = bpf_object__btf_fd(prog->obj);
4871 else
4872 btf_fd = -1;
4873 load_attr.prog_btf_fd = btf_fd >= 0 ? btf_fd : 0;
4874 load_attr.func_info = prog->func_info;
4875 load_attr.func_info_rec_size = prog->func_info_rec_size;
4876 load_attr.func_info_cnt = prog->func_info_cnt;
4877 load_attr.line_info = prog->line_info;
4878 load_attr.line_info_rec_size = prog->line_info_rec_size;
4879 load_attr.line_info_cnt = prog->line_info_cnt;
4880 load_attr.log_level = prog->log_level;
4881 load_attr.prog_flags = prog->prog_flags;
4882
4883 retry_load:
4884 log_buf = malloc(log_buf_size);
4885 if (!log_buf)
4886 pr_warn("Alloc log buffer for bpf loader error, continue without log\n");
4887
4888 ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
4889
4890 if (ret >= 0) {
4891 if (load_attr.log_level)
4892 pr_debug("verifier log:\n%s", log_buf);
4893 *pfd = ret;
4894 ret = 0;
4895 goto out;
4896 }
4897
4898 if (errno == ENOSPC) {
4899 log_buf_size <<= 1;
4900 free(log_buf);
4901 goto retry_load;
4902 }
4903 ret = -errno;
4904 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
4905 pr_warn("load bpf program failed: %s\n", cp);
4906 pr_perm_msg(ret);
4907
4908 if (log_buf && log_buf[0] != '\0') {
4909 ret = -LIBBPF_ERRNO__VERIFY;
4910 pr_warn("-- BEGIN DUMP LOG ---\n");
4911 pr_warn("\n%s\n", log_buf);
4912 pr_warn("-- END LOG --\n");
4913 } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
4914 pr_warn("Program too large (%zu insns), at most %d insns\n",
4915 load_attr.insns_cnt, BPF_MAXINSNS);
4916 ret = -LIBBPF_ERRNO__PROG2BIG;
4917 } else if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
4918 /* Wrong program type? */
4919 int fd;
4920
4921 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
4922 load_attr.expected_attach_type = 0;
4923 fd = bpf_load_program_xattr(&load_attr, NULL, 0);
4924 if (fd >= 0) {
4925 close(fd);
4926 ret = -LIBBPF_ERRNO__PROGTYPE;
4927 goto out;
4928 }
4929 }
4930
4931 out:
4932 free(log_buf);
4933 return ret;
4934 }
4935
4936 static int libbpf_find_attach_btf_id(struct bpf_program *prog);
4937
4938 int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver)
4939 {
4940 int err = 0, fd, i, btf_id;
4941
4942 if (prog->type == BPF_PROG_TYPE_TRACING ||
4943 prog->type == BPF_PROG_TYPE_EXT) {
4944 btf_id = libbpf_find_attach_btf_id(prog);
4945 if (btf_id <= 0)
4946 return btf_id;
4947 prog->attach_btf_id = btf_id;
4948 }
4949
4950 if (prog->instances.nr < 0 || !prog->instances.fds) {
4951 if (prog->preprocessor) {
4952 pr_warn("Internal error: can't load program '%s'\n",
4953 prog->section_name);
4954 return -LIBBPF_ERRNO__INTERNAL;
4955 }
4956
4957 prog->instances.fds = malloc(sizeof(int));
4958 if (!prog->instances.fds) {
4959 pr_warn("Not enough memory for BPF fds\n");
4960 return -ENOMEM;
4961 }
4962 prog->instances.nr = 1;
4963 prog->instances.fds[0] = -1;
4964 }
4965
4966 if (!prog->preprocessor) {
4967 if (prog->instances.nr != 1) {
4968 pr_warn("Program '%s' is inconsistent: nr(%d) != 1\n",
4969 prog->section_name, prog->instances.nr);
4970 }
4971 err = load_program(prog, prog->insns, prog->insns_cnt,
4972 license, kern_ver, &fd);
4973 if (!err)
4974 prog->instances.fds[0] = fd;
4975 goto out;
4976 }
4977
4978 for (i = 0; i < prog->instances.nr; i++) {
4979 struct bpf_prog_prep_result result;
4980 bpf_program_prep_t preprocessor = prog->preprocessor;
4981
4982 memset(&result, 0, sizeof(result));
4983 err = preprocessor(prog, i, prog->insns,
4984 prog->insns_cnt, &result);
4985 if (err) {
4986 pr_warn("Preprocessing the %dth instance of program '%s' failed\n",
4987 i, prog->section_name);
4988 goto out;
4989 }
4990
4991 if (!result.new_insn_ptr || !result.new_insn_cnt) {
4992 pr_debug("Skip loading the %dth instance of program '%s'\n",
4993 i, prog->section_name);
4994 prog->instances.fds[i] = -1;
4995 if (result.pfd)
4996 *result.pfd = -1;
4997 continue;
4998 }
4999
5000 err = load_program(prog, result.new_insn_ptr,
5001 result.new_insn_cnt, license, kern_ver, &fd);
5002 if (err) {
5003 pr_warn("Loading the %dth instance of program '%s' failed\n",
5004 i, prog->section_name);
5005 goto out;
5006 }
5007
5008 if (result.pfd)
5009 *result.pfd = fd;
5010 prog->instances.fds[i] = fd;
5011 }
5012 out:
5013 if (err)
5014 pr_warn("failed to load program '%s'\n", prog->section_name);
5015 zfree(&prog->insns);
5016 prog->insns_cnt = 0;
5017 return err;
5018 }
5019
5020 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
5021 const struct bpf_object *obj)
5022 {
5023 return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
5024 }
5025
5026 static int
5027 bpf_object__load_progs(struct bpf_object *obj, int log_level)
5028 {
5029 size_t i;
5030 int err;
5031
5032 for (i = 0; i < obj->nr_programs; i++) {
5033 if (bpf_program__is_function_storage(&obj->programs[i], obj))
5034 continue;
5035 obj->programs[i].log_level |= log_level;
5036 err = bpf_program__load(&obj->programs[i],
5037 obj->license,
5038 obj->kern_version);
5039 if (err)
5040 return err;
5041 }
5042 return 0;
5043 }
5044
5045 static struct bpf_object *
5046 __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
5047 const struct bpf_object_open_opts *opts)
5048 {
5049 const char *obj_name, *kconfig;
5050 struct bpf_program *prog;
5051 struct bpf_object *obj;
5052 char tmp_name[64];
5053 int err;
5054
5055 if (elf_version(EV_CURRENT) == EV_NONE) {
5056 pr_warn("failed to init libelf for %s\n",
5057 path ? : "(mem buf)");
5058 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
5059 }
5060
5061 if (!OPTS_VALID(opts, bpf_object_open_opts))
5062 return ERR_PTR(-EINVAL);
5063
5064 obj_name = OPTS_GET(opts, object_name, NULL);
5065 if (obj_buf) {
5066 if (!obj_name) {
5067 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
5068 (unsigned long)obj_buf,
5069 (unsigned long)obj_buf_sz);
5070 obj_name = tmp_name;
5071 }
5072 path = obj_name;
5073 pr_debug("loading object '%s' from buffer\n", obj_name);
5074 }
5075
5076 obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
5077 if (IS_ERR(obj))
5078 return obj;
5079
5080 kconfig = OPTS_GET(opts, kconfig, NULL);
5081 if (kconfig) {
5082 obj->kconfig = strdup(kconfig);
5083 if (!obj->kconfig)
5084 return ERR_PTR(-ENOMEM);
5085 }
5086
5087 err = bpf_object__elf_init(obj);
5088 err = err ? : bpf_object__check_endianness(obj);
5089 err = err ? : bpf_object__elf_collect(obj);
5090 err = err ? : bpf_object__collect_externs(obj);
5091 err = err ? : bpf_object__finalize_btf(obj);
5092 err = err ? : bpf_object__init_maps(obj, opts);
5093 err = err ? : bpf_object__init_prog_names(obj);
5094 err = err ? : bpf_object__collect_reloc(obj);
5095 if (err)
5096 goto out;
5097 bpf_object__elf_finish(obj);
5098
5099 bpf_object__for_each_program(prog, obj) {
5100 enum bpf_prog_type prog_type;
5101 enum bpf_attach_type attach_type;
5102
5103 if (prog->type != BPF_PROG_TYPE_UNSPEC)
5104 continue;
5105
5106 err = libbpf_prog_type_by_name(prog->section_name, &prog_type,
5107 &attach_type);
5108 if (err == -ESRCH)
5109 /* couldn't guess, but user might manually specify */
5110 continue;
5111 if (err)
5112 goto out;
5113
5114 bpf_program__set_type(prog, prog_type);
5115 bpf_program__set_expected_attach_type(prog, attach_type);
5116 if (prog_type == BPF_PROG_TYPE_TRACING ||
5117 prog_type == BPF_PROG_TYPE_EXT)
5118 prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
5119 }
5120
5121 return obj;
5122 out:
5123 bpf_object__close(obj);
5124 return ERR_PTR(err);
5125 }
5126
5127 static struct bpf_object *
5128 __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
5129 {
5130 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
5131 .relaxed_maps = flags & MAPS_RELAX_COMPAT,
5132 );
5133
5134 /* param validation */
5135 if (!attr->file)
5136 return NULL;
5137
5138 pr_debug("loading %s\n", attr->file);
5139 return __bpf_object__open(attr->file, NULL, 0, &opts);
5140 }
5141
5142 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
5143 {
5144 return __bpf_object__open_xattr(attr, 0);
5145 }
5146
5147 struct bpf_object *bpf_object__open(const char *path)
5148 {
5149 struct bpf_object_open_attr attr = {
5150 .file = path,
5151 .prog_type = BPF_PROG_TYPE_UNSPEC,
5152 };
5153
5154 return bpf_object__open_xattr(&attr);
5155 }
5156
5157 struct bpf_object *
5158 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
5159 {
5160 if (!path)
5161 return ERR_PTR(-EINVAL);
5162
5163 pr_debug("loading %s\n", path);
5164
5165 return __bpf_object__open(path, NULL, 0, opts);
5166 }
5167
5168 struct bpf_object *
5169 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
5170 const struct bpf_object_open_opts *opts)
5171 {
5172 if (!obj_buf || obj_buf_sz == 0)
5173 return ERR_PTR(-EINVAL);
5174
5175 return __bpf_object__open(NULL, obj_buf, obj_buf_sz, opts);
5176 }
5177
5178 struct bpf_object *
5179 bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
5180 const char *name)
5181 {
5182 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
5183 .object_name = name,
5184 /* wrong default, but backwards-compatible */
5185 .relaxed_maps = true,
5186 );
5187
5188 /* returning NULL is wrong, but backwards-compatible */
5189 if (!obj_buf || obj_buf_sz == 0)
5190 return NULL;
5191
5192 return bpf_object__open_mem(obj_buf, obj_buf_sz, &opts);
5193 }
5194
5195 int bpf_object__unload(struct bpf_object *obj)
5196 {
5197 size_t i;
5198
5199 if (!obj)
5200 return -EINVAL;
5201
5202 for (i = 0; i < obj->nr_maps; i++) {
5203 zclose(obj->maps[i].fd);
5204 if (obj->maps[i].st_ops)
5205 zfree(&obj->maps[i].st_ops->kern_vdata);
5206 }
5207
5208 for (i = 0; i < obj->nr_programs; i++)
5209 bpf_program__unload(&obj->programs[i]);
5210
5211 return 0;
5212 }
5213
5214 static int bpf_object__sanitize_maps(struct bpf_object *obj)
5215 {
5216 struct bpf_map *m;
5217
5218 bpf_object__for_each_map(m, obj) {
5219 if (!bpf_map__is_internal(m))
5220 continue;
5221 if (!obj->caps.global_data) {
5222 pr_warn("kernel doesn't support global data\n");
5223 return -ENOTSUP;
5224 }
5225 if (!obj->caps.array_mmap)
5226 m->def.map_flags ^= BPF_F_MMAPABLE;
5227 }
5228
5229 return 0;
5230 }
5231
5232 static int bpf_object__resolve_externs(struct bpf_object *obj,
5233 const char *extra_kconfig)
5234 {
5235 bool need_config = false;
5236 struct extern_desc *ext;
5237 int err, i;
5238 void *data;
5239
5240 if (obj->nr_extern == 0)
5241 return 0;
5242
5243 data = obj->maps[obj->kconfig_map_idx].mmaped;
5244
5245 for (i = 0; i < obj->nr_extern; i++) {
5246 ext = &obj->externs[i];
5247
5248 if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
5249 void *ext_val = data + ext->data_off;
5250 __u32 kver = get_kernel_version();
5251
5252 if (!kver) {
5253 pr_warn("failed to get kernel version\n");
5254 return -EINVAL;
5255 }
5256 err = set_ext_value_num(ext, ext_val, kver);
5257 if (err)
5258 return err;
5259 pr_debug("extern %s=0x%x\n", ext->name, kver);
5260 } else if (strncmp(ext->name, "CONFIG_", 7) == 0) {
5261 need_config = true;
5262 } else {
5263 pr_warn("unrecognized extern '%s'\n", ext->name);
5264 return -EINVAL;
5265 }
5266 }
5267 if (need_config && extra_kconfig) {
5268 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, data);
5269 if (err)
5270 return -EINVAL;
5271 need_config = false;
5272 for (i = 0; i < obj->nr_extern; i++) {
5273 ext = &obj->externs[i];
5274 if (!ext->is_set) {
5275 need_config = true;
5276 break;
5277 }
5278 }
5279 }
5280 if (need_config) {
5281 err = bpf_object__read_kconfig_file(obj, data);
5282 if (err)
5283 return -EINVAL;
5284 }
5285 for (i = 0; i < obj->nr_extern; i++) {
5286 ext = &obj->externs[i];
5287
5288 if (!ext->is_set && !ext->is_weak) {
5289 pr_warn("extern %s (strong) not resolved\n", ext->name);
5290 return -ESRCH;
5291 } else if (!ext->is_set) {
5292 pr_debug("extern %s (weak) not resolved, defaulting to zero\n",
5293 ext->name);
5294 }
5295 }
5296
5297 return 0;
5298 }
5299
5300 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
5301 {
5302 struct bpf_object *obj;
5303 int err, i;
5304
5305 if (!attr)
5306 return -EINVAL;
5307 obj = attr->obj;
5308 if (!obj)
5309 return -EINVAL;
5310
5311 if (obj->loaded) {
5312 pr_warn("object should not be loaded twice\n");
5313 return -EINVAL;
5314 }
5315
5316 obj->loaded = true;
5317
5318 err = bpf_object__probe_caps(obj);
5319 err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
5320 err = err ? : bpf_object__sanitize_and_load_btf(obj);
5321 err = err ? : bpf_object__sanitize_maps(obj);
5322 err = err ? : bpf_object__load_vmlinux_btf(obj);
5323 err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
5324 err = err ? : bpf_object__create_maps(obj);
5325 err = err ? : bpf_object__relocate(obj, attr->target_btf_path);
5326 err = err ? : bpf_object__load_progs(obj, attr->log_level);
5327
5328 btf__free(obj->btf_vmlinux);
5329 obj->btf_vmlinux = NULL;
5330
5331 if (err)
5332 goto out;
5333
5334 return 0;
5335 out:
5336 /* unpin any maps that were auto-pinned during load */
5337 for (i = 0; i < obj->nr_maps; i++)
5338 if (obj->maps[i].pinned && !obj->maps[i].reused)
5339 bpf_map__unpin(&obj->maps[i], NULL);
5340
5341 bpf_object__unload(obj);
5342 pr_warn("failed to load object '%s'\n", obj->path);
5343 return err;
5344 }
5345
5346 int bpf_object__load(struct bpf_object *obj)
5347 {
5348 struct bpf_object_load_attr attr = {
5349 .obj = obj,
5350 };
5351
5352 return bpf_object__load_xattr(&attr);
5353 }
5354
5355 static int make_parent_dir(const char *path)
5356 {
5357 char *cp, errmsg[STRERR_BUFSIZE];
5358 char *dname, *dir;
5359 int err = 0;
5360
5361 dname = strdup(path);
5362 if (dname == NULL)
5363 return -ENOMEM;
5364
5365 dir = dirname(dname);
5366 if (mkdir(dir, 0700) && errno != EEXIST)
5367 err = -errno;
5368
5369 free(dname);
5370 if (err) {
5371 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
5372 pr_warn("failed to mkdir %s: %s\n", path, cp);
5373 }
5374 return err;
5375 }
5376
5377 static int check_path(const char *path)
5378 {
5379 char *cp, errmsg[STRERR_BUFSIZE];
5380 struct statfs st_fs;
5381 char *dname, *dir;
5382 int err = 0;
5383
5384 if (path == NULL)
5385 return -EINVAL;
5386
5387 dname = strdup(path);
5388 if (dname == NULL)
5389 return -ENOMEM;
5390
5391 dir = dirname(dname);
5392 if (statfs(dir, &st_fs)) {
5393 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
5394 pr_warn("failed to statfs %s: %s\n", dir, cp);
5395 err = -errno;
5396 }
5397 free(dname);
5398
5399 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
5400 pr_warn("specified path %s is not on BPF FS\n", path);
5401 err = -EINVAL;
5402 }
5403
5404 return err;
5405 }
5406
5407 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
5408 int instance)
5409 {
5410 char *cp, errmsg[STRERR_BUFSIZE];
5411 int err;
5412
5413 err = make_parent_dir(path);
5414 if (err)
5415 return err;
5416
5417 err = check_path(path);
5418 if (err)
5419 return err;
5420
5421 if (prog == NULL) {
5422 pr_warn("invalid program pointer\n");
5423 return -EINVAL;
5424 }
5425
5426 if (instance < 0 || instance >= prog->instances.nr) {
5427 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
5428 instance, prog->section_name, prog->instances.nr);
5429 return -EINVAL;
5430 }
5431
5432 if (bpf_obj_pin(prog->instances.fds[instance], path)) {
5433 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
5434 pr_warn("failed to pin program: %s\n", cp);
5435 return -errno;
5436 }
5437 pr_debug("pinned program '%s'\n", path);
5438
5439 return 0;
5440 }
5441
5442 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
5443 int instance)
5444 {
5445 int err;
5446
5447 err = check_path(path);
5448 if (err)
5449 return err;
5450
5451 if (prog == NULL) {
5452 pr_warn("invalid program pointer\n");
5453 return -EINVAL;
5454 }
5455
5456 if (instance < 0 || instance >= prog->instances.nr) {
5457 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
5458 instance, prog->section_name, prog->instances.nr);
5459 return -EINVAL;
5460 }
5461
5462 err = unlink(path);
5463 if (err != 0)
5464 return -errno;
5465 pr_debug("unpinned program '%s'\n", path);
5466
5467 return 0;
5468 }
5469
5470 int bpf_program__pin(struct bpf_program *prog, const char *path)
5471 {
5472 int i, err;
5473
5474 err = make_parent_dir(path);
5475 if (err)
5476 return err;
5477
5478 err = check_path(path);
5479 if (err)
5480 return err;
5481
5482 if (prog == NULL) {
5483 pr_warn("invalid program pointer\n");
5484 return -EINVAL;
5485 }
5486
5487 if (prog->instances.nr <= 0) {
5488 pr_warn("no instances of prog %s to pin\n",
5489 prog->section_name);
5490 return -EINVAL;
5491 }
5492
5493 if (prog->instances.nr == 1) {
5494 /* don't create subdirs when pinning single instance */
5495 return bpf_program__pin_instance(prog, path, 0);
5496 }
5497
5498 for (i = 0; i < prog->instances.nr; i++) {
5499 char buf[PATH_MAX];
5500 int len;
5501
5502 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
5503 if (len < 0) {
5504 err = -EINVAL;
5505 goto err_unpin;
5506 } else if (len >= PATH_MAX) {
5507 err = -ENAMETOOLONG;
5508 goto err_unpin;
5509 }
5510
5511 err = bpf_program__pin_instance(prog, buf, i);
5512 if (err)
5513 goto err_unpin;
5514 }
5515
5516 return 0;
5517
5518 err_unpin:
5519 for (i = i - 1; i >= 0; i--) {
5520 char buf[PATH_MAX];
5521 int len;
5522
5523 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
5524 if (len < 0)
5525 continue;
5526 else if (len >= PATH_MAX)
5527 continue;
5528
5529 bpf_program__unpin_instance(prog, buf, i);
5530 }
5531
5532 rmdir(path);
5533
5534 return err;
5535 }
5536
5537 int bpf_program__unpin(struct bpf_program *prog, const char *path)
5538 {
5539 int i, err;
5540
5541 err = check_path(path);
5542 if (err)
5543 return err;
5544
5545 if (prog == NULL) {
5546 pr_warn("invalid program pointer\n");
5547 return -EINVAL;
5548 }
5549
5550 if (prog->instances.nr <= 0) {
5551 pr_warn("no instances of prog %s to pin\n",
5552 prog->section_name);
5553 return -EINVAL;
5554 }
5555
5556 if (prog->instances.nr == 1) {
5557 /* don't create subdirs when pinning single instance */
5558 return bpf_program__unpin_instance(prog, path, 0);
5559 }
5560
5561 for (i = 0; i < prog->instances.nr; i++) {
5562 char buf[PATH_MAX];
5563 int len;
5564
5565 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
5566 if (len < 0)
5567 return -EINVAL;
5568 else if (len >= PATH_MAX)
5569 return -ENAMETOOLONG;
5570
5571 err = bpf_program__unpin_instance(prog, buf, i);
5572 if (err)
5573 return err;
5574 }
5575
5576 err = rmdir(path);
5577 if (err)
5578 return -errno;
5579
5580 return 0;
5581 }
5582
5583 int bpf_map__pin(struct bpf_map *map, const char *path)
5584 {
5585 char *cp, errmsg[STRERR_BUFSIZE];
5586 int err;
5587
5588 if (map == NULL) {
5589 pr_warn("invalid map pointer\n");
5590 return -EINVAL;
5591 }
5592
5593 if (map->pin_path) {
5594 if (path && strcmp(path, map->pin_path)) {
5595 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
5596 bpf_map__name(map), map->pin_path, path);
5597 return -EINVAL;
5598 } else if (map->pinned) {
5599 pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
5600 bpf_map__name(map), map->pin_path);
5601 return 0;
5602 }
5603 } else {
5604 if (!path) {
5605 pr_warn("missing a path to pin map '%s' at\n",
5606 bpf_map__name(map));
5607 return -EINVAL;
5608 } else if (map->pinned) {
5609 pr_warn("map '%s' already pinned\n", bpf_map__name(map));
5610 return -EEXIST;
5611 }
5612
5613 map->pin_path = strdup(path);
5614 if (!map->pin_path) {
5615 err = -errno;
5616 goto out_err;
5617 }
5618 }
5619
5620 err = make_parent_dir(map->pin_path);
5621 if (err)
5622 return err;
5623
5624 err = check_path(map->pin_path);
5625 if (err)
5626 return err;
5627
5628 if (bpf_obj_pin(map->fd, map->pin_path)) {
5629 err = -errno;
5630 goto out_err;
5631 }
5632
5633 map->pinned = true;
5634 pr_debug("pinned map '%s'\n", map->pin_path);
5635
5636 return 0;
5637
5638 out_err:
5639 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
5640 pr_warn("failed to pin map: %s\n", cp);
5641 return err;
5642 }
5643
5644 int bpf_map__unpin(struct bpf_map *map, const char *path)
5645 {
5646 int err;
5647
5648 if (map == NULL) {
5649 pr_warn("invalid map pointer\n");
5650 return -EINVAL;
5651 }
5652
5653 if (map->pin_path) {
5654 if (path && strcmp(path, map->pin_path)) {
5655 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
5656 bpf_map__name(map), map->pin_path, path);
5657 return -EINVAL;
5658 }
5659 path = map->pin_path;
5660 } else if (!path) {
5661 pr_warn("no path to unpin map '%s' from\n",
5662 bpf_map__name(map));
5663 return -EINVAL;
5664 }
5665
5666 err = check_path(path);
5667 if (err)
5668 return err;
5669
5670 err = unlink(path);
5671 if (err != 0)
5672 return -errno;
5673
5674 map->pinned = false;
5675 pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
5676
5677 return 0;
5678 }
5679
5680 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
5681 {
5682 char *new = NULL;
5683
5684 if (path) {
5685 new = strdup(path);
5686 if (!new)
5687 return -errno;
5688 }
5689
5690 free(map->pin_path);
5691 map->pin_path = new;
5692 return 0;
5693 }
5694
5695 const char *bpf_map__get_pin_path(const struct bpf_map *map)
5696 {
5697 return map->pin_path;
5698 }
5699
5700 bool bpf_map__is_pinned(const struct bpf_map *map)
5701 {
5702 return map->pinned;
5703 }
5704
5705 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
5706 {
5707 struct bpf_map *map;
5708 int err;
5709
5710 if (!obj)
5711 return -ENOENT;
5712
5713 if (!obj->loaded) {
5714 pr_warn("object not yet loaded; load it first\n");
5715 return -ENOENT;
5716 }
5717
5718 bpf_object__for_each_map(map, obj) {
5719 char *pin_path = NULL;
5720 char buf[PATH_MAX];
5721
5722 if (path) {
5723 int len;
5724
5725 len = snprintf(buf, PATH_MAX, "%s/%s", path,
5726 bpf_map__name(map));
5727 if (len < 0) {
5728 err = -EINVAL;
5729 goto err_unpin_maps;
5730 } else if (len >= PATH_MAX) {
5731 err = -ENAMETOOLONG;
5732 goto err_unpin_maps;
5733 }
5734 pin_path = buf;
5735 } else if (!map->pin_path) {
5736 continue;
5737 }
5738
5739 err = bpf_map__pin(map, pin_path);
5740 if (err)
5741 goto err_unpin_maps;
5742 }
5743
5744 return 0;
5745
5746 err_unpin_maps:
5747 while ((map = bpf_map__prev(map, obj))) {
5748 if (!map->pin_path)
5749 continue;
5750
5751 bpf_map__unpin(map, NULL);
5752 }
5753
5754 return err;
5755 }
5756
5757 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
5758 {
5759 struct bpf_map *map;
5760 int err;
5761
5762 if (!obj)
5763 return -ENOENT;
5764
5765 bpf_object__for_each_map(map, obj) {
5766 char *pin_path = NULL;
5767 char buf[PATH_MAX];
5768
5769 if (path) {
5770 int len;
5771
5772 len = snprintf(buf, PATH_MAX, "%s/%s", path,
5773 bpf_map__name(map));
5774 if (len < 0)
5775 return -EINVAL;
5776 else if (len >= PATH_MAX)
5777 return -ENAMETOOLONG;
5778 pin_path = buf;
5779 } else if (!map->pin_path) {
5780 continue;
5781 }
5782
5783 err = bpf_map__unpin(map, pin_path);
5784 if (err)
5785 return err;
5786 }
5787
5788 return 0;
5789 }
5790
5791 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
5792 {
5793 struct bpf_program *prog;
5794 int err;
5795
5796 if (!obj)
5797 return -ENOENT;
5798
5799 if (!obj->loaded) {
5800 pr_warn("object not yet loaded; load it first\n");
5801 return -ENOENT;
5802 }
5803
5804 bpf_object__for_each_program(prog, obj) {
5805 char buf[PATH_MAX];
5806 int len;
5807
5808 len = snprintf(buf, PATH_MAX, "%s/%s", path,
5809 prog->pin_name);
5810 if (len < 0) {
5811 err = -EINVAL;
5812 goto err_unpin_programs;
5813 } else if (len >= PATH_MAX) {
5814 err = -ENAMETOOLONG;
5815 goto err_unpin_programs;
5816 }
5817
5818 err = bpf_program__pin(prog, buf);
5819 if (err)
5820 goto err_unpin_programs;
5821 }
5822
5823 return 0;
5824
5825 err_unpin_programs:
5826 while ((prog = bpf_program__prev(prog, obj))) {
5827 char buf[PATH_MAX];
5828 int len;
5829
5830 len = snprintf(buf, PATH_MAX, "%s/%s", path,
5831 prog->pin_name);
5832 if (len < 0)
5833 continue;
5834 else if (len >= PATH_MAX)
5835 continue;
5836
5837 bpf_program__unpin(prog, buf);
5838 }
5839
5840 return err;
5841 }
5842
5843 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
5844 {
5845 struct bpf_program *prog;
5846 int err;
5847
5848 if (!obj)
5849 return -ENOENT;
5850
5851 bpf_object__for_each_program(prog, obj) {
5852 char buf[PATH_MAX];
5853 int len;
5854
5855 len = snprintf(buf, PATH_MAX, "%s/%s", path,
5856 prog->pin_name);
5857 if (len < 0)
5858 return -EINVAL;
5859 else if (len >= PATH_MAX)
5860 return -ENAMETOOLONG;
5861
5862 err = bpf_program__unpin(prog, buf);
5863 if (err)
5864 return err;
5865 }
5866
5867 return 0;
5868 }
5869
5870 int bpf_object__pin(struct bpf_object *obj, const char *path)
5871 {
5872 int err;
5873
5874 err = bpf_object__pin_maps(obj, path);
5875 if (err)
5876 return err;
5877
5878 err = bpf_object__pin_programs(obj, path);
5879 if (err) {
5880 bpf_object__unpin_maps(obj, path);
5881 return err;
5882 }
5883
5884 return 0;
5885 }
5886
5887 void bpf_object__close(struct bpf_object *obj)
5888 {
5889 size_t i;
5890
5891 if (!obj)
5892 return;
5893
5894 if (obj->clear_priv)
5895 obj->clear_priv(obj, obj->priv);
5896
5897 bpf_object__elf_finish(obj);
5898 bpf_object__unload(obj);
5899 btf__free(obj->btf);
5900 btf_ext__free(obj->btf_ext);
5901
5902 for (i = 0; i < obj->nr_maps; i++) {
5903 struct bpf_map *map = &obj->maps[i];
5904
5905 if (map->clear_priv)
5906 map->clear_priv(map, map->priv);
5907 map->priv = NULL;
5908 map->clear_priv = NULL;
5909
5910 if (map->mmaped) {
5911 munmap(map->mmaped, bpf_map_mmap_sz(map));
5912 map->mmaped = NULL;
5913 }
5914
5915 if (map->st_ops) {
5916 zfree(&map->st_ops->data);
5917 zfree(&map->st_ops->progs);
5918 zfree(&map->st_ops->kern_func_off);
5919 zfree(&map->st_ops);
5920 }
5921
5922 zfree(&map->name);
5923 zfree(&map->pin_path);
5924 }
5925
5926 zfree(&obj->kconfig);
5927 zfree(&obj->externs);
5928 obj->nr_extern = 0;
5929
5930 zfree(&obj->maps);
5931 obj->nr_maps = 0;
5932
5933 if (obj->programs && obj->nr_programs) {
5934 for (i = 0; i < obj->nr_programs; i++)
5935 bpf_program__exit(&obj->programs[i]);
5936 }
5937 zfree(&obj->programs);
5938
5939 list_del(&obj->list);
5940 free(obj);
5941 }
5942
5943 struct bpf_object *
5944 bpf_object__next(struct bpf_object *prev)
5945 {
5946 struct bpf_object *next;
5947
5948 if (!prev)
5949 next = list_first_entry(&bpf_objects_list,
5950 struct bpf_object,
5951 list);
5952 else
5953 next = list_next_entry(prev, list);
5954
5955 /* Empty list is noticed here so don't need checking on entry. */
5956 if (&next->list == &bpf_objects_list)
5957 return NULL;
5958
5959 return next;
5960 }
5961
5962 const char *bpf_object__name(const struct bpf_object *obj)
5963 {
5964 return obj ? obj->name : ERR_PTR(-EINVAL);
5965 }
5966
5967 unsigned int bpf_object__kversion(const struct bpf_object *obj)
5968 {
5969 return obj ? obj->kern_version : 0;
5970 }
5971
5972 struct btf *bpf_object__btf(const struct bpf_object *obj)
5973 {
5974 return obj ? obj->btf : NULL;
5975 }
5976
5977 int bpf_object__btf_fd(const struct bpf_object *obj)
5978 {
5979 return obj->btf ? btf__fd(obj->btf) : -1;
5980 }
5981
5982 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
5983 bpf_object_clear_priv_t clear_priv)
5984 {
5985 if (obj->priv && obj->clear_priv)
5986 obj->clear_priv(obj, obj->priv);
5987
5988 obj->priv = priv;
5989 obj->clear_priv = clear_priv;
5990 return 0;
5991 }
5992
5993 void *bpf_object__priv(const struct bpf_object *obj)
5994 {
5995 return obj ? obj->priv : ERR_PTR(-EINVAL);
5996 }
5997
5998 static struct bpf_program *
5999 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
6000 bool forward)
6001 {
6002 size_t nr_programs = obj->nr_programs;
6003 ssize_t idx;
6004
6005 if (!nr_programs)
6006 return NULL;
6007
6008 if (!p)
6009 /* Iter from the beginning */
6010 return forward ? &obj->programs[0] :
6011 &obj->programs[nr_programs - 1];
6012
6013 if (p->obj != obj) {
6014 pr_warn("error: program handler doesn't match object\n");
6015 return NULL;
6016 }
6017
6018 idx = (p - obj->programs) + (forward ? 1 : -1);
6019 if (idx >= obj->nr_programs || idx < 0)
6020 return NULL;
6021 return &obj->programs[idx];
6022 }
6023
6024 struct bpf_program *
6025 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
6026 {
6027 struct bpf_program *prog = prev;
6028
6029 do {
6030 prog = __bpf_program__iter(prog, obj, true);
6031 } while (prog && bpf_program__is_function_storage(prog, obj));
6032
6033 return prog;
6034 }
6035
6036 struct bpf_program *
6037 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
6038 {
6039 struct bpf_program *prog = next;
6040
6041 do {
6042 prog = __bpf_program__iter(prog, obj, false);
6043 } while (prog && bpf_program__is_function_storage(prog, obj));
6044
6045 return prog;
6046 }
6047
6048 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
6049 bpf_program_clear_priv_t clear_priv)
6050 {
6051 if (prog->priv && prog->clear_priv)
6052 prog->clear_priv(prog, prog->priv);
6053
6054 prog->priv = priv;
6055 prog->clear_priv = clear_priv;
6056 return 0;
6057 }
6058
6059 void *bpf_program__priv(const struct bpf_program *prog)
6060 {
6061 return prog ? prog->priv : ERR_PTR(-EINVAL);
6062 }
6063
6064 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
6065 {
6066 prog->prog_ifindex = ifindex;
6067 }
6068
6069 const char *bpf_program__name(const struct bpf_program *prog)
6070 {
6071 return prog->name;
6072 }
6073
6074 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
6075 {
6076 const char *title;
6077
6078 title = prog->section_name;
6079 if (needs_copy) {
6080 title = strdup(title);
6081 if (!title) {
6082 pr_warn("failed to strdup program title\n");
6083 return ERR_PTR(-ENOMEM);
6084 }
6085 }
6086
6087 return title;
6088 }
6089
6090 int bpf_program__fd(const struct bpf_program *prog)
6091 {
6092 return bpf_program__nth_fd(prog, 0);
6093 }
6094
6095 size_t bpf_program__size(const struct bpf_program *prog)
6096 {
6097 return prog->insns_cnt * sizeof(struct bpf_insn);
6098 }
6099
6100 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
6101 bpf_program_prep_t prep)
6102 {
6103 int *instances_fds;
6104
6105 if (nr_instances <= 0 || !prep)
6106 return -EINVAL;
6107
6108 if (prog->instances.nr > 0 || prog->instances.fds) {
6109 pr_warn("Can't set pre-processor after loading\n");
6110 return -EINVAL;
6111 }
6112
6113 instances_fds = malloc(sizeof(int) * nr_instances);
6114 if (!instances_fds) {
6115 pr_warn("alloc memory failed for fds\n");
6116 return -ENOMEM;
6117 }
6118
6119 /* fill all fd with -1 */
6120 memset(instances_fds, -1, sizeof(int) * nr_instances);
6121
6122 prog->instances.nr = nr_instances;
6123 prog->instances.fds = instances_fds;
6124 prog->preprocessor = prep;
6125 return 0;
6126 }
6127
6128 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
6129 {
6130 int fd;
6131
6132 if (!prog)
6133 return -EINVAL;
6134
6135 if (n >= prog->instances.nr || n < 0) {
6136 pr_warn("Can't get the %dth fd from program %s: only %d instances\n",
6137 n, prog->section_name, prog->instances.nr);
6138 return -EINVAL;
6139 }
6140
6141 fd = prog->instances.fds[n];
6142 if (fd < 0) {
6143 pr_warn("%dth instance of program '%s' is invalid\n",
6144 n, prog->section_name);
6145 return -ENOENT;
6146 }
6147
6148 return fd;
6149 }
6150
6151 enum bpf_prog_type bpf_program__get_type(struct bpf_program *prog)
6152 {
6153 return prog->type;
6154 }
6155
6156 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
6157 {
6158 prog->type = type;
6159 }
6160
6161 static bool bpf_program__is_type(const struct bpf_program *prog,
6162 enum bpf_prog_type type)
6163 {
6164 return prog ? (prog->type == type) : false;
6165 }
6166
6167 #define BPF_PROG_TYPE_FNS(NAME, TYPE) \
6168 int bpf_program__set_##NAME(struct bpf_program *prog) \
6169 { \
6170 if (!prog) \
6171 return -EINVAL; \
6172 bpf_program__set_type(prog, TYPE); \
6173 return 0; \
6174 } \
6175 \
6176 bool bpf_program__is_##NAME(const struct bpf_program *prog) \
6177 { \
6178 return bpf_program__is_type(prog, TYPE); \
6179 } \
6180
6181 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
6182 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
6183 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
6184 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
6185 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
6186 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
6187 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
6188 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
6189 BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
6190 BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS);
6191 BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT);
6192
6193 enum bpf_attach_type
6194 bpf_program__get_expected_attach_type(struct bpf_program *prog)
6195 {
6196 return prog->expected_attach_type;
6197 }
6198
6199 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
6200 enum bpf_attach_type type)
6201 {
6202 prog->expected_attach_type = type;
6203 }
6204
6205 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, btf, atype) \
6206 { string, sizeof(string) - 1, ptype, eatype, is_attachable, btf, atype }
6207
6208 /* Programs that can NOT be attached. */
6209 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0, 0)
6210
6211 /* Programs that can be attached. */
6212 #define BPF_APROG_SEC(string, ptype, atype) \
6213 BPF_PROG_SEC_IMPL(string, ptype, 0, 1, 0, atype)
6214
6215 /* Programs that must specify expected attach type at load time. */
6216 #define BPF_EAPROG_SEC(string, ptype, eatype) \
6217 BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, 0, eatype)
6218
6219 /* Programs that use BTF to identify attach point */
6220 #define BPF_PROG_BTF(string, ptype, eatype) \
6221 BPF_PROG_SEC_IMPL(string, ptype, eatype, 0, 1, 0)
6222
6223 /* Programs that can be attached but attach type can't be identified by section
6224 * name. Kept for backward compatibility.
6225 */
6226 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
6227
6228 #define SEC_DEF(sec_pfx, ptype, ...) { \
6229 .sec = sec_pfx, \
6230 .len = sizeof(sec_pfx) - 1, \
6231 .prog_type = BPF_PROG_TYPE_##ptype, \
6232 __VA_ARGS__ \
6233 }
6234
6235 struct bpf_sec_def;
6236
6237 typedef struct bpf_link *(*attach_fn_t)(const struct bpf_sec_def *sec,
6238 struct bpf_program *prog);
6239
6240 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
6241 struct bpf_program *prog);
6242 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
6243 struct bpf_program *prog);
6244 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
6245 struct bpf_program *prog);
6246 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
6247 struct bpf_program *prog);
6248
6249 struct bpf_sec_def {
6250 const char *sec;
6251 size_t len;
6252 enum bpf_prog_type prog_type;
6253 enum bpf_attach_type expected_attach_type;
6254 bool is_attachable;
6255 bool is_attach_btf;
6256 enum bpf_attach_type attach_type;
6257 attach_fn_t attach_fn;
6258 };
6259
6260 static const struct bpf_sec_def section_defs[] = {
6261 BPF_PROG_SEC("socket", BPF_PROG_TYPE_SOCKET_FILTER),
6262 BPF_PROG_SEC("sk_reuseport", BPF_PROG_TYPE_SK_REUSEPORT),
6263 SEC_DEF("kprobe/", KPROBE,
6264 .attach_fn = attach_kprobe),
6265 BPF_PROG_SEC("uprobe/", BPF_PROG_TYPE_KPROBE),
6266 SEC_DEF("kretprobe/", KPROBE,
6267 .attach_fn = attach_kprobe),
6268 BPF_PROG_SEC("uretprobe/", BPF_PROG_TYPE_KPROBE),
6269 BPF_PROG_SEC("classifier", BPF_PROG_TYPE_SCHED_CLS),
6270 BPF_PROG_SEC("action", BPF_PROG_TYPE_SCHED_ACT),
6271 SEC_DEF("tracepoint/", TRACEPOINT,
6272 .attach_fn = attach_tp),
6273 SEC_DEF("tp/", TRACEPOINT,
6274 .attach_fn = attach_tp),
6275 SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT,
6276 .attach_fn = attach_raw_tp),
6277 SEC_DEF("raw_tp/", RAW_TRACEPOINT,
6278 .attach_fn = attach_raw_tp),
6279 SEC_DEF("tp_btf/", TRACING,
6280 .expected_attach_type = BPF_TRACE_RAW_TP,
6281 .is_attach_btf = true,
6282 .attach_fn = attach_trace),
6283 SEC_DEF("fentry/", TRACING,
6284 .expected_attach_type = BPF_TRACE_FENTRY,
6285 .is_attach_btf = true,
6286 .attach_fn = attach_trace),
6287 SEC_DEF("fexit/", TRACING,
6288 .expected_attach_type = BPF_TRACE_FEXIT,
6289 .is_attach_btf = true,
6290 .attach_fn = attach_trace),
6291 SEC_DEF("freplace/", EXT,
6292 .is_attach_btf = true,
6293 .attach_fn = attach_trace),
6294 BPF_PROG_SEC("xdp", BPF_PROG_TYPE_XDP),
6295 BPF_PROG_SEC("perf_event", BPF_PROG_TYPE_PERF_EVENT),
6296 BPF_PROG_SEC("lwt_in", BPF_PROG_TYPE_LWT_IN),
6297 BPF_PROG_SEC("lwt_out", BPF_PROG_TYPE_LWT_OUT),
6298 BPF_PROG_SEC("lwt_xmit", BPF_PROG_TYPE_LWT_XMIT),
6299 BPF_PROG_SEC("lwt_seg6local", BPF_PROG_TYPE_LWT_SEG6LOCAL),
6300 BPF_APROG_SEC("cgroup_skb/ingress", BPF_PROG_TYPE_CGROUP_SKB,
6301 BPF_CGROUP_INET_INGRESS),
6302 BPF_APROG_SEC("cgroup_skb/egress", BPF_PROG_TYPE_CGROUP_SKB,
6303 BPF_CGROUP_INET_EGRESS),
6304 BPF_APROG_COMPAT("cgroup/skb", BPF_PROG_TYPE_CGROUP_SKB),
6305 BPF_APROG_SEC("cgroup/sock", BPF_PROG_TYPE_CGROUP_SOCK,
6306 BPF_CGROUP_INET_SOCK_CREATE),
6307 BPF_EAPROG_SEC("cgroup/post_bind4", BPF_PROG_TYPE_CGROUP_SOCK,
6308 BPF_CGROUP_INET4_POST_BIND),
6309 BPF_EAPROG_SEC("cgroup/post_bind6", BPF_PROG_TYPE_CGROUP_SOCK,
6310 BPF_CGROUP_INET6_POST_BIND),
6311 BPF_APROG_SEC("cgroup/dev", BPF_PROG_TYPE_CGROUP_DEVICE,
6312 BPF_CGROUP_DEVICE),
6313 BPF_APROG_SEC("sockops", BPF_PROG_TYPE_SOCK_OPS,
6314 BPF_CGROUP_SOCK_OPS),
6315 BPF_APROG_SEC("sk_skb/stream_parser", BPF_PROG_TYPE_SK_SKB,
6316 BPF_SK_SKB_STREAM_PARSER),
6317 BPF_APROG_SEC("sk_skb/stream_verdict", BPF_PROG_TYPE_SK_SKB,
6318 BPF_SK_SKB_STREAM_VERDICT),
6319 BPF_APROG_COMPAT("sk_skb", BPF_PROG_TYPE_SK_SKB),
6320 BPF_APROG_SEC("sk_msg", BPF_PROG_TYPE_SK_MSG,
6321 BPF_SK_MSG_VERDICT),
6322 BPF_APROG_SEC("lirc_mode2", BPF_PROG_TYPE_LIRC_MODE2,
6323 BPF_LIRC_MODE2),
6324 BPF_APROG_SEC("flow_dissector", BPF_PROG_TYPE_FLOW_DISSECTOR,
6325 BPF_FLOW_DISSECTOR),
6326 BPF_EAPROG_SEC("cgroup/bind4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6327 BPF_CGROUP_INET4_BIND),
6328 BPF_EAPROG_SEC("cgroup/bind6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6329 BPF_CGROUP_INET6_BIND),
6330 BPF_EAPROG_SEC("cgroup/connect4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6331 BPF_CGROUP_INET4_CONNECT),
6332 BPF_EAPROG_SEC("cgroup/connect6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6333 BPF_CGROUP_INET6_CONNECT),
6334 BPF_EAPROG_SEC("cgroup/sendmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6335 BPF_CGROUP_UDP4_SENDMSG),
6336 BPF_EAPROG_SEC("cgroup/sendmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6337 BPF_CGROUP_UDP6_SENDMSG),
6338 BPF_EAPROG_SEC("cgroup/recvmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6339 BPF_CGROUP_UDP4_RECVMSG),
6340 BPF_EAPROG_SEC("cgroup/recvmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6341 BPF_CGROUP_UDP6_RECVMSG),
6342 BPF_EAPROG_SEC("cgroup/sysctl", BPF_PROG_TYPE_CGROUP_SYSCTL,
6343 BPF_CGROUP_SYSCTL),
6344 BPF_EAPROG_SEC("cgroup/getsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT,
6345 BPF_CGROUP_GETSOCKOPT),
6346 BPF_EAPROG_SEC("cgroup/setsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT,
6347 BPF_CGROUP_SETSOCKOPT),
6348 BPF_PROG_SEC("struct_ops", BPF_PROG_TYPE_STRUCT_OPS),
6349 };
6350
6351 #undef BPF_PROG_SEC_IMPL
6352 #undef BPF_PROG_SEC
6353 #undef BPF_APROG_SEC
6354 #undef BPF_EAPROG_SEC
6355 #undef BPF_APROG_COMPAT
6356 #undef SEC_DEF
6357
6358 #define MAX_TYPE_NAME_SIZE 32
6359
6360 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
6361 {
6362 int i, n = ARRAY_SIZE(section_defs);
6363
6364 for (i = 0; i < n; i++) {
6365 if (strncmp(sec_name,
6366 section_defs[i].sec, section_defs[i].len))
6367 continue;
6368 return &section_defs[i];
6369 }
6370 return NULL;
6371 }
6372
6373 static char *libbpf_get_type_names(bool attach_type)
6374 {
6375 int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
6376 char *buf;
6377
6378 buf = malloc(len);
6379 if (!buf)
6380 return NULL;
6381
6382 buf[0] = '\0';
6383 /* Forge string buf with all available names */
6384 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
6385 if (attach_type && !section_defs[i].is_attachable)
6386 continue;
6387
6388 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
6389 free(buf);
6390 return NULL;
6391 }
6392 strcat(buf, " ");
6393 strcat(buf, section_defs[i].sec);
6394 }
6395
6396 return buf;
6397 }
6398
6399 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
6400 enum bpf_attach_type *expected_attach_type)
6401 {
6402 const struct bpf_sec_def *sec_def;
6403 char *type_names;
6404
6405 if (!name)
6406 return -EINVAL;
6407
6408 sec_def = find_sec_def(name);
6409 if (sec_def) {
6410 *prog_type = sec_def->prog_type;
6411 *expected_attach_type = sec_def->expected_attach_type;
6412 return 0;
6413 }
6414
6415 pr_debug("failed to guess program type from ELF section '%s'\n", name);
6416 type_names = libbpf_get_type_names(false);
6417 if (type_names != NULL) {
6418 pr_debug("supported section(type) names are:%s\n", type_names);
6419 free(type_names);
6420 }
6421
6422 return -ESRCH;
6423 }
6424
6425 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
6426 size_t offset)
6427 {
6428 struct bpf_map *map;
6429 size_t i;
6430
6431 for (i = 0; i < obj->nr_maps; i++) {
6432 map = &obj->maps[i];
6433 if (!bpf_map__is_struct_ops(map))
6434 continue;
6435 if (map->sec_offset <= offset &&
6436 offset - map->sec_offset < map->def.value_size)
6437 return map;
6438 }
6439
6440 return NULL;
6441 }
6442
6443 /* Collect the reloc from ELF and populate the st_ops->progs[] */
6444 static int bpf_object__collect_struct_ops_map_reloc(struct bpf_object *obj,
6445 GElf_Shdr *shdr,
6446 Elf_Data *data)
6447 {
6448 const struct btf_member *member;
6449 struct bpf_struct_ops *st_ops;
6450 struct bpf_program *prog;
6451 unsigned int shdr_idx;
6452 const struct btf *btf;
6453 struct bpf_map *map;
6454 Elf_Data *symbols;
6455 unsigned int moff;
6456 const char *name;
6457 __u32 member_idx;
6458 GElf_Sym sym;
6459 GElf_Rel rel;
6460 int i, nrels;
6461
6462 symbols = obj->efile.symbols;
6463 btf = obj->btf;
6464 nrels = shdr->sh_size / shdr->sh_entsize;
6465 for (i = 0; i < nrels; i++) {
6466 if (!gelf_getrel(data, i, &rel)) {
6467 pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
6468 return -LIBBPF_ERRNO__FORMAT;
6469 }
6470
6471 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
6472 pr_warn("struct_ops reloc: symbol %zx not found\n",
6473 (size_t)GELF_R_SYM(rel.r_info));
6474 return -LIBBPF_ERRNO__FORMAT;
6475 }
6476
6477 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
6478 sym.st_name) ? : "<?>";
6479 map = find_struct_ops_map_by_offset(obj, rel.r_offset);
6480 if (!map) {
6481 pr_warn("struct_ops reloc: cannot find map at rel.r_offset %zu\n",
6482 (size_t)rel.r_offset);
6483 return -EINVAL;
6484 }
6485
6486 moff = rel.r_offset - map->sec_offset;
6487 shdr_idx = sym.st_shndx;
6488 st_ops = map->st_ops;
6489 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",
6490 map->name,
6491 (long long)(rel.r_info >> 32),
6492 (long long)sym.st_value,
6493 shdr_idx, (size_t)rel.r_offset,
6494 map->sec_offset, sym.st_name, name);
6495
6496 if (shdr_idx >= SHN_LORESERVE) {
6497 pr_warn("struct_ops reloc %s: rel.r_offset %zu shdr_idx %u unsupported non-static function\n",
6498 map->name, (size_t)rel.r_offset, shdr_idx);
6499 return -LIBBPF_ERRNO__RELOC;
6500 }
6501
6502 member = find_member_by_offset(st_ops->type, moff * 8);
6503 if (!member) {
6504 pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
6505 map->name, moff);
6506 return -EINVAL;
6507 }
6508 member_idx = member - btf_members(st_ops->type);
6509 name = btf__name_by_offset(btf, member->name_off);
6510
6511 if (!resolve_func_ptr(btf, member->type, NULL)) {
6512 pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
6513 map->name, name);
6514 return -EINVAL;
6515 }
6516
6517 prog = bpf_object__find_prog_by_idx(obj, shdr_idx);
6518 if (!prog) {
6519 pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
6520 map->name, shdr_idx, name);
6521 return -EINVAL;
6522 }
6523
6524 if (prog->type == BPF_PROG_TYPE_UNSPEC) {
6525 const struct bpf_sec_def *sec_def;
6526
6527 sec_def = find_sec_def(prog->section_name);
6528 if (sec_def &&
6529 sec_def->prog_type != BPF_PROG_TYPE_STRUCT_OPS) {
6530 /* for pr_warn */
6531 prog->type = sec_def->prog_type;
6532 goto invalid_prog;
6533 }
6534
6535 prog->type = BPF_PROG_TYPE_STRUCT_OPS;
6536 prog->attach_btf_id = st_ops->type_id;
6537 prog->expected_attach_type = member_idx;
6538 } else if (prog->type != BPF_PROG_TYPE_STRUCT_OPS ||
6539 prog->attach_btf_id != st_ops->type_id ||
6540 prog->expected_attach_type != member_idx) {
6541 goto invalid_prog;
6542 }
6543 st_ops->progs[member_idx] = prog;
6544 }
6545
6546 return 0;
6547
6548 invalid_prog:
6549 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",
6550 map->name, prog->name, prog->section_name, prog->type,
6551 prog->attach_btf_id, prog->expected_attach_type, name);
6552 return -EINVAL;
6553 }
6554
6555 #define BTF_TRACE_PREFIX "btf_trace_"
6556 #define BTF_MAX_NAME_SIZE 128
6557
6558 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
6559 const char *name, __u32 kind)
6560 {
6561 char btf_type_name[BTF_MAX_NAME_SIZE];
6562 int ret;
6563
6564 ret = snprintf(btf_type_name, sizeof(btf_type_name),
6565 "%s%s", prefix, name);
6566 /* snprintf returns the number of characters written excluding the
6567 * the terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
6568 * indicates truncation.
6569 */
6570 if (ret < 0 || ret >= sizeof(btf_type_name))
6571 return -ENAMETOOLONG;
6572 return btf__find_by_name_kind(btf, btf_type_name, kind);
6573 }
6574
6575 static inline int __find_vmlinux_btf_id(struct btf *btf, const char *name,
6576 enum bpf_attach_type attach_type)
6577 {
6578 int err;
6579
6580 if (attach_type == BPF_TRACE_RAW_TP)
6581 err = find_btf_by_prefix_kind(btf, BTF_TRACE_PREFIX, name,
6582 BTF_KIND_TYPEDEF);
6583 else
6584 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
6585
6586 return err;
6587 }
6588
6589 int libbpf_find_vmlinux_btf_id(const char *name,
6590 enum bpf_attach_type attach_type)
6591 {
6592 struct btf *btf;
6593
6594 btf = libbpf_find_kernel_btf();
6595 if (IS_ERR(btf)) {
6596 pr_warn("vmlinux BTF is not found\n");
6597 return -EINVAL;
6598 }
6599
6600 return __find_vmlinux_btf_id(btf, name, attach_type);
6601 }
6602
6603 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
6604 {
6605 struct bpf_prog_info_linear *info_linear;
6606 struct bpf_prog_info *info;
6607 struct btf *btf = NULL;
6608 int err = -EINVAL;
6609
6610 info_linear = bpf_program__get_prog_info_linear(attach_prog_fd, 0);
6611 if (IS_ERR_OR_NULL(info_linear)) {
6612 pr_warn("failed get_prog_info_linear for FD %d\n",
6613 attach_prog_fd);
6614 return -EINVAL;
6615 }
6616 info = &info_linear->info;
6617 if (!info->btf_id) {
6618 pr_warn("The target program doesn't have BTF\n");
6619 goto out;
6620 }
6621 if (btf__get_from_id(info->btf_id, &btf)) {
6622 pr_warn("Failed to get BTF of the program\n");
6623 goto out;
6624 }
6625 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
6626 btf__free(btf);
6627 if (err <= 0) {
6628 pr_warn("%s is not found in prog's BTF\n", name);
6629 goto out;
6630 }
6631 out:
6632 free(info_linear);
6633 return err;
6634 }
6635
6636 static int libbpf_find_attach_btf_id(struct bpf_program *prog)
6637 {
6638 enum bpf_attach_type attach_type = prog->expected_attach_type;
6639 __u32 attach_prog_fd = prog->attach_prog_fd;
6640 const char *name = prog->section_name;
6641 int i, err;
6642
6643 if (!name)
6644 return -EINVAL;
6645
6646 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
6647 if (!section_defs[i].is_attach_btf)
6648 continue;
6649 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
6650 continue;
6651 if (attach_prog_fd)
6652 err = libbpf_find_prog_btf_id(name + section_defs[i].len,
6653 attach_prog_fd);
6654 else
6655 err = __find_vmlinux_btf_id(prog->obj->btf_vmlinux,
6656 name + section_defs[i].len,
6657 attach_type);
6658 if (err <= 0)
6659 pr_warn("%s is not found in vmlinux BTF\n", name);
6660 return err;
6661 }
6662 pr_warn("failed to identify btf_id based on ELF section name '%s'\n", name);
6663 return -ESRCH;
6664 }
6665
6666 int libbpf_attach_type_by_name(const char *name,
6667 enum bpf_attach_type *attach_type)
6668 {
6669 char *type_names;
6670 int i;
6671
6672 if (!name)
6673 return -EINVAL;
6674
6675 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
6676 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
6677 continue;
6678 if (!section_defs[i].is_attachable)
6679 return -EINVAL;
6680 *attach_type = section_defs[i].attach_type;
6681 return 0;
6682 }
6683 pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
6684 type_names = libbpf_get_type_names(true);
6685 if (type_names != NULL) {
6686 pr_debug("attachable section(type) names are:%s\n", type_names);
6687 free(type_names);
6688 }
6689
6690 return -EINVAL;
6691 }
6692
6693 int bpf_map__fd(const struct bpf_map *map)
6694 {
6695 return map ? map->fd : -EINVAL;
6696 }
6697
6698 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
6699 {
6700 return map ? &map->def : ERR_PTR(-EINVAL);
6701 }
6702
6703 const char *bpf_map__name(const struct bpf_map *map)
6704 {
6705 return map ? map->name : NULL;
6706 }
6707
6708 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
6709 {
6710 return map ? map->btf_key_type_id : 0;
6711 }
6712
6713 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
6714 {
6715 return map ? map->btf_value_type_id : 0;
6716 }
6717
6718 int bpf_map__set_priv(struct bpf_map *map, void *priv,
6719 bpf_map_clear_priv_t clear_priv)
6720 {
6721 if (!map)
6722 return -EINVAL;
6723
6724 if (map->priv) {
6725 if (map->clear_priv)
6726 map->clear_priv(map, map->priv);
6727 }
6728
6729 map->priv = priv;
6730 map->clear_priv = clear_priv;
6731 return 0;
6732 }
6733
6734 void *bpf_map__priv(const struct bpf_map *map)
6735 {
6736 return map ? map->priv : ERR_PTR(-EINVAL);
6737 }
6738
6739 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
6740 {
6741 return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
6742 }
6743
6744 bool bpf_map__is_internal(const struct bpf_map *map)
6745 {
6746 return map->libbpf_type != LIBBPF_MAP_UNSPEC;
6747 }
6748
6749 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
6750 {
6751 map->map_ifindex = ifindex;
6752 }
6753
6754 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
6755 {
6756 if (!bpf_map_type__is_map_in_map(map->def.type)) {
6757 pr_warn("error: unsupported map type\n");
6758 return -EINVAL;
6759 }
6760 if (map->inner_map_fd != -1) {
6761 pr_warn("error: inner_map_fd already specified\n");
6762 return -EINVAL;
6763 }
6764 map->inner_map_fd = fd;
6765 return 0;
6766 }
6767
6768 static struct bpf_map *
6769 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
6770 {
6771 ssize_t idx;
6772 struct bpf_map *s, *e;
6773
6774 if (!obj || !obj->maps)
6775 return NULL;
6776
6777 s = obj->maps;
6778 e = obj->maps + obj->nr_maps;
6779
6780 if ((m < s) || (m >= e)) {
6781 pr_warn("error in %s: map handler doesn't belong to object\n",
6782 __func__);
6783 return NULL;
6784 }
6785
6786 idx = (m - obj->maps) + i;
6787 if (idx >= obj->nr_maps || idx < 0)
6788 return NULL;
6789 return &obj->maps[idx];
6790 }
6791
6792 struct bpf_map *
6793 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
6794 {
6795 if (prev == NULL)
6796 return obj->maps;
6797
6798 return __bpf_map__iter(prev, obj, 1);
6799 }
6800
6801 struct bpf_map *
6802 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
6803 {
6804 if (next == NULL) {
6805 if (!obj->nr_maps)
6806 return NULL;
6807 return obj->maps + obj->nr_maps - 1;
6808 }
6809
6810 return __bpf_map__iter(next, obj, -1);
6811 }
6812
6813 struct bpf_map *
6814 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
6815 {
6816 struct bpf_map *pos;
6817
6818 bpf_object__for_each_map(pos, obj) {
6819 if (pos->name && !strcmp(pos->name, name))
6820 return pos;
6821 }
6822 return NULL;
6823 }
6824
6825 int
6826 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
6827 {
6828 return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
6829 }
6830
6831 struct bpf_map *
6832 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
6833 {
6834 return ERR_PTR(-ENOTSUP);
6835 }
6836
6837 long libbpf_get_error(const void *ptr)
6838 {
6839 return PTR_ERR_OR_ZERO(ptr);
6840 }
6841
6842 int bpf_prog_load(const char *file, enum bpf_prog_type type,
6843 struct bpf_object **pobj, int *prog_fd)
6844 {
6845 struct bpf_prog_load_attr attr;
6846
6847 memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
6848 attr.file = file;
6849 attr.prog_type = type;
6850 attr.expected_attach_type = 0;
6851
6852 return bpf_prog_load_xattr(&attr, pobj, prog_fd);
6853 }
6854
6855 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
6856 struct bpf_object **pobj, int *prog_fd)
6857 {
6858 struct bpf_object_open_attr open_attr = {};
6859 struct bpf_program *prog, *first_prog = NULL;
6860 struct bpf_object *obj;
6861 struct bpf_map *map;
6862 int err;
6863
6864 if (!attr)
6865 return -EINVAL;
6866 if (!attr->file)
6867 return -EINVAL;
6868
6869 open_attr.file = attr->file;
6870 open_attr.prog_type = attr->prog_type;
6871
6872 obj = bpf_object__open_xattr(&open_attr);
6873 if (IS_ERR_OR_NULL(obj))
6874 return -ENOENT;
6875
6876 bpf_object__for_each_program(prog, obj) {
6877 enum bpf_attach_type attach_type = attr->expected_attach_type;
6878 /*
6879 * to preserve backwards compatibility, bpf_prog_load treats
6880 * attr->prog_type, if specified, as an override to whatever
6881 * bpf_object__open guessed
6882 */
6883 if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) {
6884 bpf_program__set_type(prog, attr->prog_type);
6885 bpf_program__set_expected_attach_type(prog,
6886 attach_type);
6887 }
6888 if (bpf_program__get_type(prog) == BPF_PROG_TYPE_UNSPEC) {
6889 /*
6890 * we haven't guessed from section name and user
6891 * didn't provide a fallback type, too bad...
6892 */
6893 bpf_object__close(obj);
6894 return -EINVAL;
6895 }
6896
6897 prog->prog_ifindex = attr->ifindex;
6898 prog->log_level = attr->log_level;
6899 prog->prog_flags = attr->prog_flags;
6900 if (!first_prog)
6901 first_prog = prog;
6902 }
6903
6904 bpf_object__for_each_map(map, obj) {
6905 if (!bpf_map__is_offload_neutral(map))
6906 map->map_ifindex = attr->ifindex;
6907 }
6908
6909 if (!first_prog) {
6910 pr_warn("object file doesn't contain bpf program\n");
6911 bpf_object__close(obj);
6912 return -ENOENT;
6913 }
6914
6915 err = bpf_object__load(obj);
6916 if (err) {
6917 bpf_object__close(obj);
6918 return -EINVAL;
6919 }
6920
6921 *pobj = obj;
6922 *prog_fd = bpf_program__fd(first_prog);
6923 return 0;
6924 }
6925
6926 struct bpf_link {
6927 int (*detach)(struct bpf_link *link);
6928 int (*destroy)(struct bpf_link *link);
6929 bool disconnected;
6930 };
6931
6932 /* Release "ownership" of underlying BPF resource (typically, BPF program
6933 * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
6934 * link, when destructed through bpf_link__destroy() call won't attempt to
6935 * detach/unregisted that BPF resource. This is useful in situations where,
6936 * say, attached BPF program has to outlive userspace program that attached it
6937 * in the system. Depending on type of BPF program, though, there might be
6938 * additional steps (like pinning BPF program in BPF FS) necessary to ensure
6939 * exit of userspace program doesn't trigger automatic detachment and clean up
6940 * inside the kernel.
6941 */
6942 void bpf_link__disconnect(struct bpf_link *link)
6943 {
6944 link->disconnected = true;
6945 }
6946
6947 int bpf_link__destroy(struct bpf_link *link)
6948 {
6949 int err = 0;
6950
6951 if (!link)
6952 return 0;
6953
6954 if (!link->disconnected && link->detach)
6955 err = link->detach(link);
6956 if (link->destroy)
6957 link->destroy(link);
6958 free(link);
6959
6960 return err;
6961 }
6962
6963 struct bpf_link_fd {
6964 struct bpf_link link; /* has to be at the top of struct */
6965 int fd; /* hook FD */
6966 };
6967
6968 static int bpf_link__detach_perf_event(struct bpf_link *link)
6969 {
6970 struct bpf_link_fd *l = (void *)link;
6971 int err;
6972
6973 err = ioctl(l->fd, PERF_EVENT_IOC_DISABLE, 0);
6974 if (err)
6975 err = -errno;
6976
6977 close(l->fd);
6978 return err;
6979 }
6980
6981 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
6982 int pfd)
6983 {
6984 char errmsg[STRERR_BUFSIZE];
6985 struct bpf_link_fd *link;
6986 int prog_fd, err;
6987
6988 if (pfd < 0) {
6989 pr_warn("program '%s': invalid perf event FD %d\n",
6990 bpf_program__title(prog, false), pfd);
6991 return ERR_PTR(-EINVAL);
6992 }
6993 prog_fd = bpf_program__fd(prog);
6994 if (prog_fd < 0) {
6995 pr_warn("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
6996 bpf_program__title(prog, false));
6997 return ERR_PTR(-EINVAL);
6998 }
6999
7000 link = calloc(1, sizeof(*link));
7001 if (!link)
7002 return ERR_PTR(-ENOMEM);
7003 link->link.detach = &bpf_link__detach_perf_event;
7004 link->fd = pfd;
7005
7006 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
7007 err = -errno;
7008 free(link);
7009 pr_warn("program '%s': failed to attach to pfd %d: %s\n",
7010 bpf_program__title(prog, false), pfd,
7011 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7012 return ERR_PTR(err);
7013 }
7014 if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
7015 err = -errno;
7016 free(link);
7017 pr_warn("program '%s': failed to enable pfd %d: %s\n",
7018 bpf_program__title(prog, false), pfd,
7019 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7020 return ERR_PTR(err);
7021 }
7022 return (struct bpf_link *)link;
7023 }
7024
7025 /*
7026 * this function is expected to parse integer in the range of [0, 2^31-1] from
7027 * given file using scanf format string fmt. If actual parsed value is
7028 * negative, the result might be indistinguishable from error
7029 */
7030 static int parse_uint_from_file(const char *file, const char *fmt)
7031 {
7032 char buf[STRERR_BUFSIZE];
7033 int err, ret;
7034 FILE *f;
7035
7036 f = fopen(file, "r");
7037 if (!f) {
7038 err = -errno;
7039 pr_debug("failed to open '%s': %s\n", file,
7040 libbpf_strerror_r(err, buf, sizeof(buf)));
7041 return err;
7042 }
7043 err = fscanf(f, fmt, &ret);
7044 if (err != 1) {
7045 err = err == EOF ? -EIO : -errno;
7046 pr_debug("failed to parse '%s': %s\n", file,
7047 libbpf_strerror_r(err, buf, sizeof(buf)));
7048 fclose(f);
7049 return err;
7050 }
7051 fclose(f);
7052 return ret;
7053 }
7054
7055 static int determine_kprobe_perf_type(void)
7056 {
7057 const char *file = "/sys/bus/event_source/devices/kprobe/type";
7058
7059 return parse_uint_from_file(file, "%d\n");
7060 }
7061
7062 static int determine_uprobe_perf_type(void)
7063 {
7064 const char *file = "/sys/bus/event_source/devices/uprobe/type";
7065
7066 return parse_uint_from_file(file, "%d\n");
7067 }
7068
7069 static int determine_kprobe_retprobe_bit(void)
7070 {
7071 const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
7072
7073 return parse_uint_from_file(file, "config:%d\n");
7074 }
7075
7076 static int determine_uprobe_retprobe_bit(void)
7077 {
7078 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
7079
7080 return parse_uint_from_file(file, "config:%d\n");
7081 }
7082
7083 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
7084 uint64_t offset, int pid)
7085 {
7086 struct perf_event_attr attr = {};
7087 char errmsg[STRERR_BUFSIZE];
7088 int type, pfd, err;
7089
7090 type = uprobe ? determine_uprobe_perf_type()
7091 : determine_kprobe_perf_type();
7092 if (type < 0) {
7093 pr_warn("failed to determine %s perf type: %s\n",
7094 uprobe ? "uprobe" : "kprobe",
7095 libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
7096 return type;
7097 }
7098 if (retprobe) {
7099 int bit = uprobe ? determine_uprobe_retprobe_bit()
7100 : determine_kprobe_retprobe_bit();
7101
7102 if (bit < 0) {
7103 pr_warn("failed to determine %s retprobe bit: %s\n",
7104 uprobe ? "uprobe" : "kprobe",
7105 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
7106 return bit;
7107 }
7108 attr.config |= 1 << bit;
7109 }
7110 attr.size = sizeof(attr);
7111 attr.type = type;
7112 attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
7113 attr.config2 = offset; /* kprobe_addr or probe_offset */
7114
7115 /* pid filter is meaningful only for uprobes */
7116 pfd = syscall(__NR_perf_event_open, &attr,
7117 pid < 0 ? -1 : pid /* pid */,
7118 pid == -1 ? 0 : -1 /* cpu */,
7119 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
7120 if (pfd < 0) {
7121 err = -errno;
7122 pr_warn("%s perf_event_open() failed: %s\n",
7123 uprobe ? "uprobe" : "kprobe",
7124 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7125 return err;
7126 }
7127 return pfd;
7128 }
7129
7130 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
7131 bool retprobe,
7132 const char *func_name)
7133 {
7134 char errmsg[STRERR_BUFSIZE];
7135 struct bpf_link *link;
7136 int pfd, err;
7137
7138 pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
7139 0 /* offset */, -1 /* pid */);
7140 if (pfd < 0) {
7141 pr_warn("program '%s': failed to create %s '%s' perf event: %s\n",
7142 bpf_program__title(prog, false),
7143 retprobe ? "kretprobe" : "kprobe", func_name,
7144 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
7145 return ERR_PTR(pfd);
7146 }
7147 link = bpf_program__attach_perf_event(prog, pfd);
7148 if (IS_ERR(link)) {
7149 close(pfd);
7150 err = PTR_ERR(link);
7151 pr_warn("program '%s': failed to attach to %s '%s': %s\n",
7152 bpf_program__title(prog, false),
7153 retprobe ? "kretprobe" : "kprobe", func_name,
7154 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7155 return link;
7156 }
7157 return link;
7158 }
7159
7160 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
7161 struct bpf_program *prog)
7162 {
7163 const char *func_name;
7164 bool retprobe;
7165
7166 func_name = bpf_program__title(prog, false) + sec->len;
7167 retprobe = strcmp(sec->sec, "kretprobe/") == 0;
7168
7169 return bpf_program__attach_kprobe(prog, retprobe, func_name);
7170 }
7171
7172 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
7173 bool retprobe, pid_t pid,
7174 const char *binary_path,
7175 size_t func_offset)
7176 {
7177 char errmsg[STRERR_BUFSIZE];
7178 struct bpf_link *link;
7179 int pfd, err;
7180
7181 pfd = perf_event_open_probe(true /* uprobe */, retprobe,
7182 binary_path, func_offset, pid);
7183 if (pfd < 0) {
7184 pr_warn("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
7185 bpf_program__title(prog, false),
7186 retprobe ? "uretprobe" : "uprobe",
7187 binary_path, func_offset,
7188 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
7189 return ERR_PTR(pfd);
7190 }
7191 link = bpf_program__attach_perf_event(prog, pfd);
7192 if (IS_ERR(link)) {
7193 close(pfd);
7194 err = PTR_ERR(link);
7195 pr_warn("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
7196 bpf_program__title(prog, false),
7197 retprobe ? "uretprobe" : "uprobe",
7198 binary_path, func_offset,
7199 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7200 return link;
7201 }
7202 return link;
7203 }
7204
7205 static int determine_tracepoint_id(const char *tp_category,
7206 const char *tp_name)
7207 {
7208 char file[PATH_MAX];
7209 int ret;
7210
7211 ret = snprintf(file, sizeof(file),
7212 "/sys/kernel/debug/tracing/events/%s/%s/id",
7213 tp_category, tp_name);
7214 if (ret < 0)
7215 return -errno;
7216 if (ret >= sizeof(file)) {
7217 pr_debug("tracepoint %s/%s path is too long\n",
7218 tp_category, tp_name);
7219 return -E2BIG;
7220 }
7221 return parse_uint_from_file(file, "%d\n");
7222 }
7223
7224 static int perf_event_open_tracepoint(const char *tp_category,
7225 const char *tp_name)
7226 {
7227 struct perf_event_attr attr = {};
7228 char errmsg[STRERR_BUFSIZE];
7229 int tp_id, pfd, err;
7230
7231 tp_id = determine_tracepoint_id(tp_category, tp_name);
7232 if (tp_id < 0) {
7233 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
7234 tp_category, tp_name,
7235 libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
7236 return tp_id;
7237 }
7238
7239 attr.type = PERF_TYPE_TRACEPOINT;
7240 attr.size = sizeof(attr);
7241 attr.config = tp_id;
7242
7243 pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
7244 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
7245 if (pfd < 0) {
7246 err = -errno;
7247 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
7248 tp_category, tp_name,
7249 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7250 return err;
7251 }
7252 return pfd;
7253 }
7254
7255 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
7256 const char *tp_category,
7257 const char *tp_name)
7258 {
7259 char errmsg[STRERR_BUFSIZE];
7260 struct bpf_link *link;
7261 int pfd, err;
7262
7263 pfd = perf_event_open_tracepoint(tp_category, tp_name);
7264 if (pfd < 0) {
7265 pr_warn("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
7266 bpf_program__title(prog, false),
7267 tp_category, tp_name,
7268 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
7269 return ERR_PTR(pfd);
7270 }
7271 link = bpf_program__attach_perf_event(prog, pfd);
7272 if (IS_ERR(link)) {
7273 close(pfd);
7274 err = PTR_ERR(link);
7275 pr_warn("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
7276 bpf_program__title(prog, false),
7277 tp_category, tp_name,
7278 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7279 return link;
7280 }
7281 return link;
7282 }
7283
7284 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
7285 struct bpf_program *prog)
7286 {
7287 char *sec_name, *tp_cat, *tp_name;
7288 struct bpf_link *link;
7289
7290 sec_name = strdup(bpf_program__title(prog, false));
7291 if (!sec_name)
7292 return ERR_PTR(-ENOMEM);
7293
7294 /* extract "tp/<category>/<name>" */
7295 tp_cat = sec_name + sec->len;
7296 tp_name = strchr(tp_cat, '/');
7297 if (!tp_name) {
7298 link = ERR_PTR(-EINVAL);
7299 goto out;
7300 }
7301 *tp_name = '\0';
7302 tp_name++;
7303
7304 link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
7305 out:
7306 free(sec_name);
7307 return link;
7308 }
7309
7310 static int bpf_link__detach_fd(struct bpf_link *link)
7311 {
7312 struct bpf_link_fd *l = (void *)link;
7313
7314 return close(l->fd);
7315 }
7316
7317 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
7318 const char *tp_name)
7319 {
7320 char errmsg[STRERR_BUFSIZE];
7321 struct bpf_link_fd *link;
7322 int prog_fd, pfd;
7323
7324 prog_fd = bpf_program__fd(prog);
7325 if (prog_fd < 0) {
7326 pr_warn("program '%s': can't attach before loaded\n",
7327 bpf_program__title(prog, false));
7328 return ERR_PTR(-EINVAL);
7329 }
7330
7331 link = calloc(1, sizeof(*link));
7332 if (!link)
7333 return ERR_PTR(-ENOMEM);
7334 link->link.detach = &bpf_link__detach_fd;
7335
7336 pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
7337 if (pfd < 0) {
7338 pfd = -errno;
7339 free(link);
7340 pr_warn("program '%s': failed to attach to raw tracepoint '%s': %s\n",
7341 bpf_program__title(prog, false), tp_name,
7342 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
7343 return ERR_PTR(pfd);
7344 }
7345 link->fd = pfd;
7346 return (struct bpf_link *)link;
7347 }
7348
7349 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
7350 struct bpf_program *prog)
7351 {
7352 const char *tp_name = bpf_program__title(prog, false) + sec->len;
7353
7354 return bpf_program__attach_raw_tracepoint(prog, tp_name);
7355 }
7356
7357 struct bpf_link *bpf_program__attach_trace(struct bpf_program *prog)
7358 {
7359 char errmsg[STRERR_BUFSIZE];
7360 struct bpf_link_fd *link;
7361 int prog_fd, pfd;
7362
7363 prog_fd = bpf_program__fd(prog);
7364 if (prog_fd < 0) {
7365 pr_warn("program '%s': can't attach before loaded\n",
7366 bpf_program__title(prog, false));
7367 return ERR_PTR(-EINVAL);
7368 }
7369
7370 link = calloc(1, sizeof(*link));
7371 if (!link)
7372 return ERR_PTR(-ENOMEM);
7373 link->link.detach = &bpf_link__detach_fd;
7374
7375 pfd = bpf_raw_tracepoint_open(NULL, prog_fd);
7376 if (pfd < 0) {
7377 pfd = -errno;
7378 free(link);
7379 pr_warn("program '%s': failed to attach to trace: %s\n",
7380 bpf_program__title(prog, false),
7381 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
7382 return ERR_PTR(pfd);
7383 }
7384 link->fd = pfd;
7385 return (struct bpf_link *)link;
7386 }
7387
7388 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
7389 struct bpf_program *prog)
7390 {
7391 return bpf_program__attach_trace(prog);
7392 }
7393
7394 struct bpf_link *bpf_program__attach(struct bpf_program *prog)
7395 {
7396 const struct bpf_sec_def *sec_def;
7397
7398 sec_def = find_sec_def(bpf_program__title(prog, false));
7399 if (!sec_def || !sec_def->attach_fn)
7400 return ERR_PTR(-ESRCH);
7401
7402 return sec_def->attach_fn(sec_def, prog);
7403 }
7404
7405 static int bpf_link__detach_struct_ops(struct bpf_link *link)
7406 {
7407 struct bpf_link_fd *l = (void *)link;
7408 __u32 zero = 0;
7409
7410 if (bpf_map_delete_elem(l->fd, &zero))
7411 return -errno;
7412
7413 return 0;
7414 }
7415
7416 struct bpf_link *bpf_map__attach_struct_ops(struct bpf_map *map)
7417 {
7418 struct bpf_struct_ops *st_ops;
7419 struct bpf_link_fd *link;
7420 __u32 i, zero = 0;
7421 int err;
7422
7423 if (!bpf_map__is_struct_ops(map) || map->fd == -1)
7424 return ERR_PTR(-EINVAL);
7425
7426 link = calloc(1, sizeof(*link));
7427 if (!link)
7428 return ERR_PTR(-EINVAL);
7429
7430 st_ops = map->st_ops;
7431 for (i = 0; i < btf_vlen(st_ops->type); i++) {
7432 struct bpf_program *prog = st_ops->progs[i];
7433 void *kern_data;
7434 int prog_fd;
7435
7436 if (!prog)
7437 continue;
7438
7439 prog_fd = bpf_program__fd(prog);
7440 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
7441 *(unsigned long *)kern_data = prog_fd;
7442 }
7443
7444 err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0);
7445 if (err) {
7446 err = -errno;
7447 free(link);
7448 return ERR_PTR(err);
7449 }
7450
7451 link->link.detach = bpf_link__detach_struct_ops;
7452 link->fd = map->fd;
7453
7454 return (struct bpf_link *)link;
7455 }
7456
7457 enum bpf_perf_event_ret
7458 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
7459 void **copy_mem, size_t *copy_size,
7460 bpf_perf_event_print_t fn, void *private_data)
7461 {
7462 struct perf_event_mmap_page *header = mmap_mem;
7463 __u64 data_head = ring_buffer_read_head(header);
7464 __u64 data_tail = header->data_tail;
7465 void *base = ((__u8 *)header) + page_size;
7466 int ret = LIBBPF_PERF_EVENT_CONT;
7467 struct perf_event_header *ehdr;
7468 size_t ehdr_size;
7469
7470 while (data_head != data_tail) {
7471 ehdr = base + (data_tail & (mmap_size - 1));
7472 ehdr_size = ehdr->size;
7473
7474 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
7475 void *copy_start = ehdr;
7476 size_t len_first = base + mmap_size - copy_start;
7477 size_t len_secnd = ehdr_size - len_first;
7478
7479 if (*copy_size < ehdr_size) {
7480 free(*copy_mem);
7481 *copy_mem = malloc(ehdr_size);
7482 if (!*copy_mem) {
7483 *copy_size = 0;
7484 ret = LIBBPF_PERF_EVENT_ERROR;
7485 break;
7486 }
7487 *copy_size = ehdr_size;
7488 }
7489
7490 memcpy(*copy_mem, copy_start, len_first);
7491 memcpy(*copy_mem + len_first, base, len_secnd);
7492 ehdr = *copy_mem;
7493 }
7494
7495 ret = fn(ehdr, private_data);
7496 data_tail += ehdr_size;
7497 if (ret != LIBBPF_PERF_EVENT_CONT)
7498 break;
7499 }
7500
7501 ring_buffer_write_tail(header, data_tail);
7502 return ret;
7503 }
7504
7505 struct perf_buffer;
7506
7507 struct perf_buffer_params {
7508 struct perf_event_attr *attr;
7509 /* if event_cb is specified, it takes precendence */
7510 perf_buffer_event_fn event_cb;
7511 /* sample_cb and lost_cb are higher-level common-case callbacks */
7512 perf_buffer_sample_fn sample_cb;
7513 perf_buffer_lost_fn lost_cb;
7514 void *ctx;
7515 int cpu_cnt;
7516 int *cpus;
7517 int *map_keys;
7518 };
7519
7520 struct perf_cpu_buf {
7521 struct perf_buffer *pb;
7522 void *base; /* mmap()'ed memory */
7523 void *buf; /* for reconstructing segmented data */
7524 size_t buf_size;
7525 int fd;
7526 int cpu;
7527 int map_key;
7528 };
7529
7530 struct perf_buffer {
7531 perf_buffer_event_fn event_cb;
7532 perf_buffer_sample_fn sample_cb;
7533 perf_buffer_lost_fn lost_cb;
7534 void *ctx; /* passed into callbacks */
7535
7536 size_t page_size;
7537 size_t mmap_size;
7538 struct perf_cpu_buf **cpu_bufs;
7539 struct epoll_event *events;
7540 int cpu_cnt; /* number of allocated CPU buffers */
7541 int epoll_fd; /* perf event FD */
7542 int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
7543 };
7544
7545 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
7546 struct perf_cpu_buf *cpu_buf)
7547 {
7548 if (!cpu_buf)
7549 return;
7550 if (cpu_buf->base &&
7551 munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
7552 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
7553 if (cpu_buf->fd >= 0) {
7554 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
7555 close(cpu_buf->fd);
7556 }
7557 free(cpu_buf->buf);
7558 free(cpu_buf);
7559 }
7560
7561 void perf_buffer__free(struct perf_buffer *pb)
7562 {
7563 int i;
7564
7565 if (!pb)
7566 return;
7567 if (pb->cpu_bufs) {
7568 for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
7569 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
7570
7571 bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
7572 perf_buffer__free_cpu_buf(pb, cpu_buf);
7573 }
7574 free(pb->cpu_bufs);
7575 }
7576 if (pb->epoll_fd >= 0)
7577 close(pb->epoll_fd);
7578 free(pb->events);
7579 free(pb);
7580 }
7581
7582 static struct perf_cpu_buf *
7583 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
7584 int cpu, int map_key)
7585 {
7586 struct perf_cpu_buf *cpu_buf;
7587 char msg[STRERR_BUFSIZE];
7588 int err;
7589
7590 cpu_buf = calloc(1, sizeof(*cpu_buf));
7591 if (!cpu_buf)
7592 return ERR_PTR(-ENOMEM);
7593
7594 cpu_buf->pb = pb;
7595 cpu_buf->cpu = cpu;
7596 cpu_buf->map_key = map_key;
7597
7598 cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
7599 -1, PERF_FLAG_FD_CLOEXEC);
7600 if (cpu_buf->fd < 0) {
7601 err = -errno;
7602 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
7603 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
7604 goto error;
7605 }
7606
7607 cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
7608 PROT_READ | PROT_WRITE, MAP_SHARED,
7609 cpu_buf->fd, 0);
7610 if (cpu_buf->base == MAP_FAILED) {
7611 cpu_buf->base = NULL;
7612 err = -errno;
7613 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
7614 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
7615 goto error;
7616 }
7617
7618 if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
7619 err = -errno;
7620 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
7621 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
7622 goto error;
7623 }
7624
7625 return cpu_buf;
7626
7627 error:
7628 perf_buffer__free_cpu_buf(pb, cpu_buf);
7629 return (struct perf_cpu_buf *)ERR_PTR(err);
7630 }
7631
7632 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
7633 struct perf_buffer_params *p);
7634
7635 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
7636 const struct perf_buffer_opts *opts)
7637 {
7638 struct perf_buffer_params p = {};
7639 struct perf_event_attr attr = { 0, };
7640
7641 attr.config = PERF_COUNT_SW_BPF_OUTPUT,
7642 attr.type = PERF_TYPE_SOFTWARE;
7643 attr.sample_type = PERF_SAMPLE_RAW;
7644 attr.sample_period = 1;
7645 attr.wakeup_events = 1;
7646
7647 p.attr = &attr;
7648 p.sample_cb = opts ? opts->sample_cb : NULL;
7649 p.lost_cb = opts ? opts->lost_cb : NULL;
7650 p.ctx = opts ? opts->ctx : NULL;
7651
7652 return __perf_buffer__new(map_fd, page_cnt, &p);
7653 }
7654
7655 struct perf_buffer *
7656 perf_buffer__new_raw(int map_fd, size_t page_cnt,
7657 const struct perf_buffer_raw_opts *opts)
7658 {
7659 struct perf_buffer_params p = {};
7660
7661 p.attr = opts->attr;
7662 p.event_cb = opts->event_cb;
7663 p.ctx = opts->ctx;
7664 p.cpu_cnt = opts->cpu_cnt;
7665 p.cpus = opts->cpus;
7666 p.map_keys = opts->map_keys;
7667
7668 return __perf_buffer__new(map_fd, page_cnt, &p);
7669 }
7670
7671 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
7672 struct perf_buffer_params *p)
7673 {
7674 const char *online_cpus_file = "/sys/devices/system/cpu/online";
7675 struct bpf_map_info map = {};
7676 char msg[STRERR_BUFSIZE];
7677 struct perf_buffer *pb;
7678 bool *online = NULL;
7679 __u32 map_info_len;
7680 int err, i, j, n;
7681
7682 if (page_cnt & (page_cnt - 1)) {
7683 pr_warn("page count should be power of two, but is %zu\n",
7684 page_cnt);
7685 return ERR_PTR(-EINVAL);
7686 }
7687
7688 map_info_len = sizeof(map);
7689 err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
7690 if (err) {
7691 err = -errno;
7692 pr_warn("failed to get map info for map FD %d: %s\n",
7693 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
7694 return ERR_PTR(err);
7695 }
7696
7697 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
7698 pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
7699 map.name);
7700 return ERR_PTR(-EINVAL);
7701 }
7702
7703 pb = calloc(1, sizeof(*pb));
7704 if (!pb)
7705 return ERR_PTR(-ENOMEM);
7706
7707 pb->event_cb = p->event_cb;
7708 pb->sample_cb = p->sample_cb;
7709 pb->lost_cb = p->lost_cb;
7710 pb->ctx = p->ctx;
7711
7712 pb->page_size = getpagesize();
7713 pb->mmap_size = pb->page_size * page_cnt;
7714 pb->map_fd = map_fd;
7715
7716 pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
7717 if (pb->epoll_fd < 0) {
7718 err = -errno;
7719 pr_warn("failed to create epoll instance: %s\n",
7720 libbpf_strerror_r(err, msg, sizeof(msg)));
7721 goto error;
7722 }
7723
7724 if (p->cpu_cnt > 0) {
7725 pb->cpu_cnt = p->cpu_cnt;
7726 } else {
7727 pb->cpu_cnt = libbpf_num_possible_cpus();
7728 if (pb->cpu_cnt < 0) {
7729 err = pb->cpu_cnt;
7730 goto error;
7731 }
7732 if (map.max_entries < pb->cpu_cnt)
7733 pb->cpu_cnt = map.max_entries;
7734 }
7735
7736 pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
7737 if (!pb->events) {
7738 err = -ENOMEM;
7739 pr_warn("failed to allocate events: out of memory\n");
7740 goto error;
7741 }
7742 pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
7743 if (!pb->cpu_bufs) {
7744 err = -ENOMEM;
7745 pr_warn("failed to allocate buffers: out of memory\n");
7746 goto error;
7747 }
7748
7749 err = parse_cpu_mask_file(online_cpus_file, &online, &n);
7750 if (err) {
7751 pr_warn("failed to get online CPU mask: %d\n", err);
7752 goto error;
7753 }
7754
7755 for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
7756 struct perf_cpu_buf *cpu_buf;
7757 int cpu, map_key;
7758
7759 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
7760 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
7761
7762 /* in case user didn't explicitly requested particular CPUs to
7763 * be attached to, skip offline/not present CPUs
7764 */
7765 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
7766 continue;
7767
7768 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
7769 if (IS_ERR(cpu_buf)) {
7770 err = PTR_ERR(cpu_buf);
7771 goto error;
7772 }
7773
7774 pb->cpu_bufs[j] = cpu_buf;
7775
7776 err = bpf_map_update_elem(pb->map_fd, &map_key,
7777 &cpu_buf->fd, 0);
7778 if (err) {
7779 err = -errno;
7780 pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
7781 cpu, map_key, cpu_buf->fd,
7782 libbpf_strerror_r(err, msg, sizeof(msg)));
7783 goto error;
7784 }
7785
7786 pb->events[j].events = EPOLLIN;
7787 pb->events[j].data.ptr = cpu_buf;
7788 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
7789 &pb->events[j]) < 0) {
7790 err = -errno;
7791 pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
7792 cpu, cpu_buf->fd,
7793 libbpf_strerror_r(err, msg, sizeof(msg)));
7794 goto error;
7795 }
7796 j++;
7797 }
7798 pb->cpu_cnt = j;
7799 free(online);
7800
7801 return pb;
7802
7803 error:
7804 free(online);
7805 if (pb)
7806 perf_buffer__free(pb);
7807 return ERR_PTR(err);
7808 }
7809
7810 struct perf_sample_raw {
7811 struct perf_event_header header;
7812 uint32_t size;
7813 char data[0];
7814 };
7815
7816 struct perf_sample_lost {
7817 struct perf_event_header header;
7818 uint64_t id;
7819 uint64_t lost;
7820 uint64_t sample_id;
7821 };
7822
7823 static enum bpf_perf_event_ret
7824 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
7825 {
7826 struct perf_cpu_buf *cpu_buf = ctx;
7827 struct perf_buffer *pb = cpu_buf->pb;
7828 void *data = e;
7829
7830 /* user wants full control over parsing perf event */
7831 if (pb->event_cb)
7832 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
7833
7834 switch (e->type) {
7835 case PERF_RECORD_SAMPLE: {
7836 struct perf_sample_raw *s = data;
7837
7838 if (pb->sample_cb)
7839 pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
7840 break;
7841 }
7842 case PERF_RECORD_LOST: {
7843 struct perf_sample_lost *s = data;
7844
7845 if (pb->lost_cb)
7846 pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
7847 break;
7848 }
7849 default:
7850 pr_warn("unknown perf sample type %d\n", e->type);
7851 return LIBBPF_PERF_EVENT_ERROR;
7852 }
7853 return LIBBPF_PERF_EVENT_CONT;
7854 }
7855
7856 static int perf_buffer__process_records(struct perf_buffer *pb,
7857 struct perf_cpu_buf *cpu_buf)
7858 {
7859 enum bpf_perf_event_ret ret;
7860
7861 ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
7862 pb->page_size, &cpu_buf->buf,
7863 &cpu_buf->buf_size,
7864 perf_buffer__process_record, cpu_buf);
7865 if (ret != LIBBPF_PERF_EVENT_CONT)
7866 return ret;
7867 return 0;
7868 }
7869
7870 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
7871 {
7872 int i, cnt, err;
7873
7874 cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
7875 for (i = 0; i < cnt; i++) {
7876 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
7877
7878 err = perf_buffer__process_records(pb, cpu_buf);
7879 if (err) {
7880 pr_warn("error while processing records: %d\n", err);
7881 return err;
7882 }
7883 }
7884 return cnt < 0 ? -errno : cnt;
7885 }
7886
7887 struct bpf_prog_info_array_desc {
7888 int array_offset; /* e.g. offset of jited_prog_insns */
7889 int count_offset; /* e.g. offset of jited_prog_len */
7890 int size_offset; /* > 0: offset of rec size,
7891 * < 0: fix size of -size_offset
7892 */
7893 };
7894
7895 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
7896 [BPF_PROG_INFO_JITED_INSNS] = {
7897 offsetof(struct bpf_prog_info, jited_prog_insns),
7898 offsetof(struct bpf_prog_info, jited_prog_len),
7899 -1,
7900 },
7901 [BPF_PROG_INFO_XLATED_INSNS] = {
7902 offsetof(struct bpf_prog_info, xlated_prog_insns),
7903 offsetof(struct bpf_prog_info, xlated_prog_len),
7904 -1,
7905 },
7906 [BPF_PROG_INFO_MAP_IDS] = {
7907 offsetof(struct bpf_prog_info, map_ids),
7908 offsetof(struct bpf_prog_info, nr_map_ids),
7909 -(int)sizeof(__u32),
7910 },
7911 [BPF_PROG_INFO_JITED_KSYMS] = {
7912 offsetof(struct bpf_prog_info, jited_ksyms),
7913 offsetof(struct bpf_prog_info, nr_jited_ksyms),
7914 -(int)sizeof(__u64),
7915 },
7916 [BPF_PROG_INFO_JITED_FUNC_LENS] = {
7917 offsetof(struct bpf_prog_info, jited_func_lens),
7918 offsetof(struct bpf_prog_info, nr_jited_func_lens),
7919 -(int)sizeof(__u32),
7920 },
7921 [BPF_PROG_INFO_FUNC_INFO] = {
7922 offsetof(struct bpf_prog_info, func_info),
7923 offsetof(struct bpf_prog_info, nr_func_info),
7924 offsetof(struct bpf_prog_info, func_info_rec_size),
7925 },
7926 [BPF_PROG_INFO_LINE_INFO] = {
7927 offsetof(struct bpf_prog_info, line_info),
7928 offsetof(struct bpf_prog_info, nr_line_info),
7929 offsetof(struct bpf_prog_info, line_info_rec_size),
7930 },
7931 [BPF_PROG_INFO_JITED_LINE_INFO] = {
7932 offsetof(struct bpf_prog_info, jited_line_info),
7933 offsetof(struct bpf_prog_info, nr_jited_line_info),
7934 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
7935 },
7936 [BPF_PROG_INFO_PROG_TAGS] = {
7937 offsetof(struct bpf_prog_info, prog_tags),
7938 offsetof(struct bpf_prog_info, nr_prog_tags),
7939 -(int)sizeof(__u8) * BPF_TAG_SIZE,
7940 },
7941
7942 };
7943
7944 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info,
7945 int offset)
7946 {
7947 __u32 *array = (__u32 *)info;
7948
7949 if (offset >= 0)
7950 return array[offset / sizeof(__u32)];
7951 return -(int)offset;
7952 }
7953
7954 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info,
7955 int offset)
7956 {
7957 __u64 *array = (__u64 *)info;
7958
7959 if (offset >= 0)
7960 return array[offset / sizeof(__u64)];
7961 return -(int)offset;
7962 }
7963
7964 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
7965 __u32 val)
7966 {
7967 __u32 *array = (__u32 *)info;
7968
7969 if (offset >= 0)
7970 array[offset / sizeof(__u32)] = val;
7971 }
7972
7973 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
7974 __u64 val)
7975 {
7976 __u64 *array = (__u64 *)info;
7977
7978 if (offset >= 0)
7979 array[offset / sizeof(__u64)] = val;
7980 }
7981
7982 struct bpf_prog_info_linear *
7983 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
7984 {
7985 struct bpf_prog_info_linear *info_linear;
7986 struct bpf_prog_info info = {};
7987 __u32 info_len = sizeof(info);
7988 __u32 data_len = 0;
7989 int i, err;
7990 void *ptr;
7991
7992 if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
7993 return ERR_PTR(-EINVAL);
7994
7995 /* step 1: get array dimensions */
7996 err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
7997 if (err) {
7998 pr_debug("can't get prog info: %s", strerror(errno));
7999 return ERR_PTR(-EFAULT);
8000 }
8001
8002 /* step 2: calculate total size of all arrays */
8003 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8004 bool include_array = (arrays & (1UL << i)) > 0;
8005 struct bpf_prog_info_array_desc *desc;
8006 __u32 count, size;
8007
8008 desc = bpf_prog_info_array_desc + i;
8009
8010 /* kernel is too old to support this field */
8011 if (info_len < desc->array_offset + sizeof(__u32) ||
8012 info_len < desc->count_offset + sizeof(__u32) ||
8013 (desc->size_offset > 0 && info_len < desc->size_offset))
8014 include_array = false;
8015
8016 if (!include_array) {
8017 arrays &= ~(1UL << i); /* clear the bit */
8018 continue;
8019 }
8020
8021 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
8022 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
8023
8024 data_len += count * size;
8025 }
8026
8027 /* step 3: allocate continuous memory */
8028 data_len = roundup(data_len, sizeof(__u64));
8029 info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
8030 if (!info_linear)
8031 return ERR_PTR(-ENOMEM);
8032
8033 /* step 4: fill data to info_linear->info */
8034 info_linear->arrays = arrays;
8035 memset(&info_linear->info, 0, sizeof(info));
8036 ptr = info_linear->data;
8037
8038 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8039 struct bpf_prog_info_array_desc *desc;
8040 __u32 count, size;
8041
8042 if ((arrays & (1UL << i)) == 0)
8043 continue;
8044
8045 desc = bpf_prog_info_array_desc + i;
8046 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
8047 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
8048 bpf_prog_info_set_offset_u32(&info_linear->info,
8049 desc->count_offset, count);
8050 bpf_prog_info_set_offset_u32(&info_linear->info,
8051 desc->size_offset, size);
8052 bpf_prog_info_set_offset_u64(&info_linear->info,
8053 desc->array_offset,
8054 ptr_to_u64(ptr));
8055 ptr += count * size;
8056 }
8057
8058 /* step 5: call syscall again to get required arrays */
8059 err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
8060 if (err) {
8061 pr_debug("can't get prog info: %s", strerror(errno));
8062 free(info_linear);
8063 return ERR_PTR(-EFAULT);
8064 }
8065
8066 /* step 6: verify the data */
8067 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8068 struct bpf_prog_info_array_desc *desc;
8069 __u32 v1, v2;
8070
8071 if ((arrays & (1UL << i)) == 0)
8072 continue;
8073
8074 desc = bpf_prog_info_array_desc + i;
8075 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
8076 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
8077 desc->count_offset);
8078 if (v1 != v2)
8079 pr_warn("%s: mismatch in element count\n", __func__);
8080
8081 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
8082 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
8083 desc->size_offset);
8084 if (v1 != v2)
8085 pr_warn("%s: mismatch in rec size\n", __func__);
8086 }
8087
8088 /* step 7: update info_len and data_len */
8089 info_linear->info_len = sizeof(struct bpf_prog_info);
8090 info_linear->data_len = data_len;
8091
8092 return info_linear;
8093 }
8094
8095 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
8096 {
8097 int i;
8098
8099 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8100 struct bpf_prog_info_array_desc *desc;
8101 __u64 addr, offs;
8102
8103 if ((info_linear->arrays & (1UL << i)) == 0)
8104 continue;
8105
8106 desc = bpf_prog_info_array_desc + i;
8107 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
8108 desc->array_offset);
8109 offs = addr - ptr_to_u64(info_linear->data);
8110 bpf_prog_info_set_offset_u64(&info_linear->info,
8111 desc->array_offset, offs);
8112 }
8113 }
8114
8115 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
8116 {
8117 int i;
8118
8119 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8120 struct bpf_prog_info_array_desc *desc;
8121 __u64 addr, offs;
8122
8123 if ((info_linear->arrays & (1UL << i)) == 0)
8124 continue;
8125
8126 desc = bpf_prog_info_array_desc + i;
8127 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
8128 desc->array_offset);
8129 addr = offs + ptr_to_u64(info_linear->data);
8130 bpf_prog_info_set_offset_u64(&info_linear->info,
8131 desc->array_offset, addr);
8132 }
8133 }
8134
8135 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
8136 {
8137 int err = 0, n, len, start, end = -1;
8138 bool *tmp;
8139
8140 *mask = NULL;
8141 *mask_sz = 0;
8142
8143 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
8144 while (*s) {
8145 if (*s == ',' || *s == '\n') {
8146 s++;
8147 continue;
8148 }
8149 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
8150 if (n <= 0 || n > 2) {
8151 pr_warn("Failed to get CPU range %s: %d\n", s, n);
8152 err = -EINVAL;
8153 goto cleanup;
8154 } else if (n == 1) {
8155 end = start;
8156 }
8157 if (start < 0 || start > end) {
8158 pr_warn("Invalid CPU range [%d,%d] in %s\n",
8159 start, end, s);
8160 err = -EINVAL;
8161 goto cleanup;
8162 }
8163 tmp = realloc(*mask, end + 1);
8164 if (!tmp) {
8165 err = -ENOMEM;
8166 goto cleanup;
8167 }
8168 *mask = tmp;
8169 memset(tmp + *mask_sz, 0, start - *mask_sz);
8170 memset(tmp + start, 1, end - start + 1);
8171 *mask_sz = end + 1;
8172 s += len;
8173 }
8174 if (!*mask_sz) {
8175 pr_warn("Empty CPU range\n");
8176 return -EINVAL;
8177 }
8178 return 0;
8179 cleanup:
8180 free(*mask);
8181 *mask = NULL;
8182 return err;
8183 }
8184
8185 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
8186 {
8187 int fd, err = 0, len;
8188 char buf[128];
8189
8190 fd = open(fcpu, O_RDONLY);
8191 if (fd < 0) {
8192 err = -errno;
8193 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
8194 return err;
8195 }
8196 len = read(fd, buf, sizeof(buf));
8197 close(fd);
8198 if (len <= 0) {
8199 err = len ? -errno : -EINVAL;
8200 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
8201 return err;
8202 }
8203 if (len >= sizeof(buf)) {
8204 pr_warn("CPU mask is too big in file %s\n", fcpu);
8205 return -E2BIG;
8206 }
8207 buf[len] = '\0';
8208
8209 return parse_cpu_mask_str(buf, mask, mask_sz);
8210 }
8211
8212 int libbpf_num_possible_cpus(void)
8213 {
8214 static const char *fcpu = "/sys/devices/system/cpu/possible";
8215 static int cpus;
8216 int err, n, i, tmp_cpus;
8217 bool *mask;
8218
8219 tmp_cpus = READ_ONCE(cpus);
8220 if (tmp_cpus > 0)
8221 return tmp_cpus;
8222
8223 err = parse_cpu_mask_file(fcpu, &mask, &n);
8224 if (err)
8225 return err;
8226
8227 tmp_cpus = 0;
8228 for (i = 0; i < n; i++) {
8229 if (mask[i])
8230 tmp_cpus++;
8231 }
8232 free(mask);
8233
8234 WRITE_ONCE(cpus, tmp_cpus);
8235 return tmp_cpus;
8236 }
8237
8238 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
8239 const struct bpf_object_open_opts *opts)
8240 {
8241 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
8242 .object_name = s->name,
8243 );
8244 struct bpf_object *obj;
8245 int i;
8246
8247 /* Attempt to preserve opts->object_name, unless overriden by user
8248 * explicitly. Overwriting object name for skeletons is discouraged,
8249 * as it breaks global data maps, because they contain object name
8250 * prefix as their own map name prefix. When skeleton is generated,
8251 * bpftool is making an assumption that this name will stay the same.
8252 */
8253 if (opts) {
8254 memcpy(&skel_opts, opts, sizeof(*opts));
8255 if (!opts->object_name)
8256 skel_opts.object_name = s->name;
8257 }
8258
8259 obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
8260 if (IS_ERR(obj)) {
8261 pr_warn("failed to initialize skeleton BPF object '%s': %ld\n",
8262 s->name, PTR_ERR(obj));
8263 return PTR_ERR(obj);
8264 }
8265
8266 *s->obj = obj;
8267
8268 for (i = 0; i < s->map_cnt; i++) {
8269 struct bpf_map **map = s->maps[i].map;
8270 const char *name = s->maps[i].name;
8271 void **mmaped = s->maps[i].mmaped;
8272
8273 *map = bpf_object__find_map_by_name(obj, name);
8274 if (!*map) {
8275 pr_warn("failed to find skeleton map '%s'\n", name);
8276 return -ESRCH;
8277 }
8278
8279 /* externs shouldn't be pre-setup from user code */
8280 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
8281 *mmaped = (*map)->mmaped;
8282 }
8283
8284 for (i = 0; i < s->prog_cnt; i++) {
8285 struct bpf_program **prog = s->progs[i].prog;
8286 const char *name = s->progs[i].name;
8287
8288 *prog = bpf_object__find_program_by_name(obj, name);
8289 if (!*prog) {
8290 pr_warn("failed to find skeleton program '%s'\n", name);
8291 return -ESRCH;
8292 }
8293 }
8294
8295 return 0;
8296 }
8297
8298 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
8299 {
8300 int i, err;
8301
8302 err = bpf_object__load(*s->obj);
8303 if (err) {
8304 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
8305 return err;
8306 }
8307
8308 for (i = 0; i < s->map_cnt; i++) {
8309 struct bpf_map *map = *s->maps[i].map;
8310 size_t mmap_sz = bpf_map_mmap_sz(map);
8311 int prot, map_fd = bpf_map__fd(map);
8312 void **mmaped = s->maps[i].mmaped;
8313
8314 if (!mmaped)
8315 continue;
8316
8317 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
8318 *mmaped = NULL;
8319 continue;
8320 }
8321
8322 if (map->def.map_flags & BPF_F_RDONLY_PROG)
8323 prot = PROT_READ;
8324 else
8325 prot = PROT_READ | PROT_WRITE;
8326
8327 /* Remap anonymous mmap()-ed "map initialization image" as
8328 * a BPF map-backed mmap()-ed memory, but preserving the same
8329 * memory address. This will cause kernel to change process'
8330 * page table to point to a different piece of kernel memory,
8331 * but from userspace point of view memory address (and its
8332 * contents, being identical at this point) will stay the
8333 * same. This mapping will be released by bpf_object__close()
8334 * as per normal clean up procedure, so we don't need to worry
8335 * about it from skeleton's clean up perspective.
8336 */
8337 *mmaped = mmap(map->mmaped, mmap_sz, prot,
8338 MAP_SHARED | MAP_FIXED, map_fd, 0);
8339 if (*mmaped == MAP_FAILED) {
8340 err = -errno;
8341 *mmaped = NULL;
8342 pr_warn("failed to re-mmap() map '%s': %d\n",
8343 bpf_map__name(map), err);
8344 return err;
8345 }
8346 }
8347
8348 return 0;
8349 }
8350
8351 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
8352 {
8353 int i;
8354
8355 for (i = 0; i < s->prog_cnt; i++) {
8356 struct bpf_program *prog = *s->progs[i].prog;
8357 struct bpf_link **link = s->progs[i].link;
8358 const struct bpf_sec_def *sec_def;
8359 const char *sec_name = bpf_program__title(prog, false);
8360
8361 sec_def = find_sec_def(sec_name);
8362 if (!sec_def || !sec_def->attach_fn)
8363 continue;
8364
8365 *link = sec_def->attach_fn(sec_def, prog);
8366 if (IS_ERR(*link)) {
8367 pr_warn("failed to auto-attach program '%s': %ld\n",
8368 bpf_program__name(prog), PTR_ERR(*link));
8369 return PTR_ERR(*link);
8370 }
8371 }
8372
8373 return 0;
8374 }
8375
8376 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
8377 {
8378 int i;
8379
8380 for (i = 0; i < s->prog_cnt; i++) {
8381 struct bpf_link **link = s->progs[i].link;
8382
8383 if (!IS_ERR_OR_NULL(*link))
8384 bpf_link__destroy(*link);
8385 *link = NULL;
8386 }
8387 }
8388
8389 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
8390 {
8391 if (s->progs)
8392 bpf_object__detach_skeleton(s);
8393 if (s->obj)
8394 bpf_object__close(*s->obj);
8395 free(s->maps);
8396 free(s->progs);
8397 free(s);
8398 }
Linux with added FPC-III support