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 <string.h>
22 #include <unistd.h>
23 #include <endian.h>
24 #include <fcntl.h>
25 #include <errno.h>
26 #include <asm/unistd.h>
27 #include <linux/err.h>
28 #include <linux/kernel.h>
29 #include <linux/bpf.h>
30 #include <linux/btf.h>
31 #include <linux/filter.h>
32 #include <linux/list.h>
33 #include <linux/limits.h>
34 #include <linux/perf_event.h>
35 #include <linux/ring_buffer.h>
36 #include <sys/epoll.h>
37 #include <sys/ioctl.h>
38 #include <sys/mman.h>
39 #include <sys/stat.h>
40 #include <sys/types.h>
41 #include <sys/vfs.h>
42 #include <sys/utsname.h>
43 #include <tools/libc_compat.h>
44 #include <libelf.h>
45 #include <gelf.h>
46
47 #include "libbpf.h"
48 #include "bpf.h"
49 #include "btf.h"
50 #include "str_error.h"
51 #include "libbpf_internal.h"
52 #include "hashmap.h"
53
54 #ifndef EM_BPF
55 #define EM_BPF 247
56 #endif
57
58 #ifndef BPF_FS_MAGIC
59 #define BPF_FS_MAGIC 0xcafe4a11
60 #endif
61
62 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
63 * compilation if user enables corresponding warning. Disable it explicitly.
64 */
65 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
66
67 #define __printf(a, b) __attribute__((format(printf, a, b)))
68
__base_pr(enum libbpf_print_level level,const char * format,va_list args)69 static int __base_pr(enum libbpf_print_level level, const char *format,
70 va_list args)
71 {
72 if (level == LIBBPF_DEBUG)
73 return 0;
74
75 return vfprintf(stderr, format, args);
76 }
77
78 static libbpf_print_fn_t __libbpf_pr = __base_pr;
79
libbpf_set_print(libbpf_print_fn_t fn)80 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
81 {
82 libbpf_print_fn_t old_print_fn = __libbpf_pr;
83
84 __libbpf_pr = fn;
85 return old_print_fn;
86 }
87
88 __printf(2, 3)
libbpf_print(enum libbpf_print_level level,const char * format,...)89 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
90 {
91 va_list args;
92
93 if (!__libbpf_pr)
94 return;
95
96 va_start(args, format);
97 __libbpf_pr(level, format, args);
98 va_end(args);
99 }
100
101 #define STRERR_BUFSIZE 128
102
103 #define CHECK_ERR(action, err, out) do { \
104 err = action; \
105 if (err) \
106 goto out; \
107 } while(0)
108
109
110 /* Copied from tools/perf/util/util.h */
111 #ifndef zfree
112 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
113 #endif
114
115 #ifndef zclose
116 # define zclose(fd) ({ \
117 int ___err = 0; \
118 if ((fd) >= 0) \
119 ___err = close((fd)); \
120 fd = -1; \
121 ___err; })
122 #endif
123
124 #ifdef HAVE_LIBELF_MMAP_SUPPORT
125 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
126 #else
127 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
128 #endif
129
ptr_to_u64(const void * ptr)130 static inline __u64 ptr_to_u64(const void *ptr)
131 {
132 return (__u64) (unsigned long) ptr;
133 }
134
135 struct bpf_capabilities {
136 /* v4.14: kernel support for program & map names. */
137 __u32 name:1;
138 /* v5.2: kernel support for global data sections. */
139 __u32 global_data:1;
140 /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
141 __u32 btf_func:1;
142 /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
143 __u32 btf_datasec:1;
144 };
145
146 /*
147 * bpf_prog should be a better name but it has been used in
148 * linux/filter.h.
149 */
150 struct bpf_program {
151 /* Index in elf obj file, for relocation use. */
152 int idx;
153 char *name;
154 int prog_ifindex;
155 char *section_name;
156 /* section_name with / replaced by _; makes recursive pinning
157 * in bpf_object__pin_programs easier
158 */
159 char *pin_name;
160 struct bpf_insn *insns;
161 size_t insns_cnt, main_prog_cnt;
162 enum bpf_prog_type type;
163
164 struct reloc_desc {
165 enum {
166 RELO_LD64,
167 RELO_CALL,
168 RELO_DATA,
169 } type;
170 int insn_idx;
171 union {
172 int map_idx;
173 int text_off;
174 };
175 } *reloc_desc;
176 int nr_reloc;
177 int log_level;
178
179 struct {
180 int nr;
181 int *fds;
182 } instances;
183 bpf_program_prep_t preprocessor;
184
185 struct bpf_object *obj;
186 void *priv;
187 bpf_program_clear_priv_t clear_priv;
188
189 enum bpf_attach_type expected_attach_type;
190 void *func_info;
191 __u32 func_info_rec_size;
192 __u32 func_info_cnt;
193
194 struct bpf_capabilities *caps;
195
196 void *line_info;
197 __u32 line_info_rec_size;
198 __u32 line_info_cnt;
199 __u32 prog_flags;
200 };
201
202 enum libbpf_map_type {
203 LIBBPF_MAP_UNSPEC,
204 LIBBPF_MAP_DATA,
205 LIBBPF_MAP_BSS,
206 LIBBPF_MAP_RODATA,
207 };
208
209 static const char * const libbpf_type_to_btf_name[] = {
210 [LIBBPF_MAP_DATA] = ".data",
211 [LIBBPF_MAP_BSS] = ".bss",
212 [LIBBPF_MAP_RODATA] = ".rodata",
213 };
214
215 struct bpf_map {
216 int fd;
217 char *name;
218 int sec_idx;
219 size_t sec_offset;
220 int map_ifindex;
221 int inner_map_fd;
222 struct bpf_map_def def;
223 __u32 btf_key_type_id;
224 __u32 btf_value_type_id;
225 void *priv;
226 bpf_map_clear_priv_t clear_priv;
227 enum libbpf_map_type libbpf_type;
228 };
229
230 struct bpf_secdata {
231 void *rodata;
232 void *data;
233 };
234
235 static LIST_HEAD(bpf_objects_list);
236
237 struct bpf_object {
238 char name[BPF_OBJ_NAME_LEN];
239 char license[64];
240 __u32 kern_version;
241
242 struct bpf_program *programs;
243 size_t nr_programs;
244 struct bpf_map *maps;
245 size_t nr_maps;
246 size_t maps_cap;
247 struct bpf_secdata sections;
248
249 bool loaded;
250 bool has_pseudo_calls;
251
252 /*
253 * Information when doing elf related work. Only valid if fd
254 * is valid.
255 */
256 struct {
257 int fd;
258 void *obj_buf;
259 size_t obj_buf_sz;
260 Elf *elf;
261 GElf_Ehdr ehdr;
262 Elf_Data *symbols;
263 Elf_Data *data;
264 Elf_Data *rodata;
265 Elf_Data *bss;
266 size_t strtabidx;
267 struct {
268 GElf_Shdr shdr;
269 Elf_Data *data;
270 } *reloc;
271 int nr_reloc;
272 int maps_shndx;
273 int btf_maps_shndx;
274 int text_shndx;
275 int data_shndx;
276 int rodata_shndx;
277 int bss_shndx;
278 } efile;
279 /*
280 * All loaded bpf_object is linked in a list, which is
281 * hidden to caller. bpf_objects__<func> handlers deal with
282 * all objects.
283 */
284 struct list_head list;
285
286 struct btf *btf;
287 struct btf_ext *btf_ext;
288
289 void *priv;
290 bpf_object_clear_priv_t clear_priv;
291
292 struct bpf_capabilities caps;
293
294 char path[];
295 };
296 #define obj_elf_valid(o) ((o)->efile.elf)
297
bpf_program__unload(struct bpf_program * prog)298 void bpf_program__unload(struct bpf_program *prog)
299 {
300 int i;
301
302 if (!prog)
303 return;
304
305 /*
306 * If the object is opened but the program was never loaded,
307 * it is possible that prog->instances.nr == -1.
308 */
309 if (prog->instances.nr > 0) {
310 for (i = 0; i < prog->instances.nr; i++)
311 zclose(prog->instances.fds[i]);
312 } else if (prog->instances.nr != -1) {
313 pr_warning("Internal error: instances.nr is %d\n",
314 prog->instances.nr);
315 }
316
317 prog->instances.nr = -1;
318 zfree(&prog->instances.fds);
319
320 zfree(&prog->func_info);
321 zfree(&prog->line_info);
322 }
323
bpf_program__exit(struct bpf_program * prog)324 static void bpf_program__exit(struct bpf_program *prog)
325 {
326 if (!prog)
327 return;
328
329 if (prog->clear_priv)
330 prog->clear_priv(prog, prog->priv);
331
332 prog->priv = NULL;
333 prog->clear_priv = NULL;
334
335 bpf_program__unload(prog);
336 zfree(&prog->name);
337 zfree(&prog->section_name);
338 zfree(&prog->pin_name);
339 zfree(&prog->insns);
340 zfree(&prog->reloc_desc);
341
342 prog->nr_reloc = 0;
343 prog->insns_cnt = 0;
344 prog->idx = -1;
345 }
346
__bpf_program__pin_name(struct bpf_program * prog)347 static char *__bpf_program__pin_name(struct bpf_program *prog)
348 {
349 char *name, *p;
350
351 name = p = strdup(prog->section_name);
352 while ((p = strchr(p, '/')))
353 *p = '_';
354
355 return name;
356 }
357
358 static int
bpf_program__init(void * data,size_t size,char * section_name,int idx,struct bpf_program * prog)359 bpf_program__init(void *data, size_t size, char *section_name, int idx,
360 struct bpf_program *prog)
361 {
362 const size_t bpf_insn_sz = sizeof(struct bpf_insn);
363
364 if (size == 0 || size % bpf_insn_sz) {
365 pr_warning("corrupted section '%s', size: %zu\n",
366 section_name, size);
367 return -EINVAL;
368 }
369
370 memset(prog, 0, sizeof(*prog));
371
372 prog->section_name = strdup(section_name);
373 if (!prog->section_name) {
374 pr_warning("failed to alloc name for prog under section(%d) %s\n",
375 idx, section_name);
376 goto errout;
377 }
378
379 prog->pin_name = __bpf_program__pin_name(prog);
380 if (!prog->pin_name) {
381 pr_warning("failed to alloc pin name for prog under section(%d) %s\n",
382 idx, section_name);
383 goto errout;
384 }
385
386 prog->insns = malloc(size);
387 if (!prog->insns) {
388 pr_warning("failed to alloc insns for prog under section %s\n",
389 section_name);
390 goto errout;
391 }
392 prog->insns_cnt = size / bpf_insn_sz;
393 memcpy(prog->insns, data, size);
394 prog->idx = idx;
395 prog->instances.fds = NULL;
396 prog->instances.nr = -1;
397 prog->type = BPF_PROG_TYPE_UNSPEC;
398
399 return 0;
400 errout:
401 bpf_program__exit(prog);
402 return -ENOMEM;
403 }
404
405 static int
bpf_object__add_program(struct bpf_object * obj,void * data,size_t size,char * section_name,int idx)406 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
407 char *section_name, int idx)
408 {
409 struct bpf_program prog, *progs;
410 int nr_progs, err;
411
412 err = bpf_program__init(data, size, section_name, idx, &prog);
413 if (err)
414 return err;
415
416 prog.caps = &obj->caps;
417 progs = obj->programs;
418 nr_progs = obj->nr_programs;
419
420 progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
421 if (!progs) {
422 /*
423 * In this case the original obj->programs
424 * is still valid, so don't need special treat for
425 * bpf_close_object().
426 */
427 pr_warning("failed to alloc a new program under section '%s'\n",
428 section_name);
429 bpf_program__exit(&prog);
430 return -ENOMEM;
431 }
432
433 pr_debug("found program %s\n", prog.section_name);
434 obj->programs = progs;
435 obj->nr_programs = nr_progs + 1;
436 prog.obj = obj;
437 progs[nr_progs] = prog;
438 return 0;
439 }
440
441 static int
bpf_object__init_prog_names(struct bpf_object * obj)442 bpf_object__init_prog_names(struct bpf_object *obj)
443 {
444 Elf_Data *symbols = obj->efile.symbols;
445 struct bpf_program *prog;
446 size_t pi, si;
447
448 for (pi = 0; pi < obj->nr_programs; pi++) {
449 const char *name = NULL;
450
451 prog = &obj->programs[pi];
452
453 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
454 si++) {
455 GElf_Sym sym;
456
457 if (!gelf_getsym(symbols, si, &sym))
458 continue;
459 if (sym.st_shndx != prog->idx)
460 continue;
461 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
462 continue;
463
464 name = elf_strptr(obj->efile.elf,
465 obj->efile.strtabidx,
466 sym.st_name);
467 if (!name) {
468 pr_warning("failed to get sym name string for prog %s\n",
469 prog->section_name);
470 return -LIBBPF_ERRNO__LIBELF;
471 }
472 }
473
474 if (!name && prog->idx == obj->efile.text_shndx)
475 name = ".text";
476
477 if (!name) {
478 pr_warning("failed to find sym for prog %s\n",
479 prog->section_name);
480 return -EINVAL;
481 }
482
483 prog->name = strdup(name);
484 if (!prog->name) {
485 pr_warning("failed to allocate memory for prog sym %s\n",
486 name);
487 return -ENOMEM;
488 }
489 }
490
491 return 0;
492 }
493
bpf_object__new(const char * path,void * obj_buf,size_t obj_buf_sz)494 static struct bpf_object *bpf_object__new(const char *path,
495 void *obj_buf,
496 size_t obj_buf_sz)
497 {
498 struct bpf_object *obj;
499 char *end;
500
501 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
502 if (!obj) {
503 pr_warning("alloc memory failed for %s\n", path);
504 return ERR_PTR(-ENOMEM);
505 }
506
507 strcpy(obj->path, path);
508 /* Using basename() GNU version which doesn't modify arg. */
509 strncpy(obj->name, basename((void *)path), sizeof(obj->name) - 1);
510 end = strchr(obj->name, '.');
511 if (end)
512 *end = 0;
513
514 obj->efile.fd = -1;
515 /*
516 * Caller of this function should also call
517 * bpf_object__elf_finish() after data collection to return
518 * obj_buf to user. If not, we should duplicate the buffer to
519 * avoid user freeing them before elf finish.
520 */
521 obj->efile.obj_buf = obj_buf;
522 obj->efile.obj_buf_sz = obj_buf_sz;
523 obj->efile.maps_shndx = -1;
524 obj->efile.btf_maps_shndx = -1;
525 obj->efile.data_shndx = -1;
526 obj->efile.rodata_shndx = -1;
527 obj->efile.bss_shndx = -1;
528
529 obj->loaded = false;
530
531 INIT_LIST_HEAD(&obj->list);
532 list_add(&obj->list, &bpf_objects_list);
533 return obj;
534 }
535
bpf_object__elf_finish(struct bpf_object * obj)536 static void bpf_object__elf_finish(struct bpf_object *obj)
537 {
538 if (!obj_elf_valid(obj))
539 return;
540
541 if (obj->efile.elf) {
542 elf_end(obj->efile.elf);
543 obj->efile.elf = NULL;
544 }
545 obj->efile.symbols = NULL;
546 obj->efile.data = NULL;
547 obj->efile.rodata = NULL;
548 obj->efile.bss = NULL;
549
550 zfree(&obj->efile.reloc);
551 obj->efile.nr_reloc = 0;
552 zclose(obj->efile.fd);
553 obj->efile.obj_buf = NULL;
554 obj->efile.obj_buf_sz = 0;
555 }
556
bpf_object__elf_init(struct bpf_object * obj)557 static int bpf_object__elf_init(struct bpf_object *obj)
558 {
559 int err = 0;
560 GElf_Ehdr *ep;
561
562 if (obj_elf_valid(obj)) {
563 pr_warning("elf init: internal error\n");
564 return -LIBBPF_ERRNO__LIBELF;
565 }
566
567 if (obj->efile.obj_buf_sz > 0) {
568 /*
569 * obj_buf should have been validated by
570 * bpf_object__open_buffer().
571 */
572 obj->efile.elf = elf_memory(obj->efile.obj_buf,
573 obj->efile.obj_buf_sz);
574 } else {
575 obj->efile.fd = open(obj->path, O_RDONLY);
576 if (obj->efile.fd < 0) {
577 char errmsg[STRERR_BUFSIZE], *cp;
578
579 err = -errno;
580 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
581 pr_warning("failed to open %s: %s\n", obj->path, cp);
582 return err;
583 }
584
585 obj->efile.elf = elf_begin(obj->efile.fd,
586 LIBBPF_ELF_C_READ_MMAP, NULL);
587 }
588
589 if (!obj->efile.elf) {
590 pr_warning("failed to open %s as ELF file\n", obj->path);
591 err = -LIBBPF_ERRNO__LIBELF;
592 goto errout;
593 }
594
595 if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
596 pr_warning("failed to get EHDR from %s\n", obj->path);
597 err = -LIBBPF_ERRNO__FORMAT;
598 goto errout;
599 }
600 ep = &obj->efile.ehdr;
601
602 /* Old LLVM set e_machine to EM_NONE */
603 if (ep->e_type != ET_REL ||
604 (ep->e_machine && ep->e_machine != EM_BPF)) {
605 pr_warning("%s is not an eBPF object file\n", obj->path);
606 err = -LIBBPF_ERRNO__FORMAT;
607 goto errout;
608 }
609
610 return 0;
611 errout:
612 bpf_object__elf_finish(obj);
613 return err;
614 }
615
bpf_object__check_endianness(struct bpf_object * obj)616 static int bpf_object__check_endianness(struct bpf_object *obj)
617 {
618 #if __BYTE_ORDER == __LITTLE_ENDIAN
619 if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
620 return 0;
621 #elif __BYTE_ORDER == __BIG_ENDIAN
622 if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
623 return 0;
624 #else
625 # error "Unrecognized __BYTE_ORDER__"
626 #endif
627 pr_warning("endianness mismatch.\n");
628 return -LIBBPF_ERRNO__ENDIAN;
629 }
630
631 static int
bpf_object__init_license(struct bpf_object * obj,void * data,size_t size)632 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
633 {
634 memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
635 pr_debug("license of %s is %s\n", obj->path, obj->license);
636 return 0;
637 }
638
639 static int
bpf_object__init_kversion(struct bpf_object * obj,void * data,size_t size)640 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
641 {
642 __u32 kver;
643
644 if (size != sizeof(kver)) {
645 pr_warning("invalid kver section in %s\n", obj->path);
646 return -LIBBPF_ERRNO__FORMAT;
647 }
648 memcpy(&kver, data, sizeof(kver));
649 obj->kern_version = kver;
650 pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
651 return 0;
652 }
653
compare_bpf_map(const void * _a,const void * _b)654 static int compare_bpf_map(const void *_a, const void *_b)
655 {
656 const struct bpf_map *a = _a;
657 const struct bpf_map *b = _b;
658
659 if (a->sec_idx != b->sec_idx)
660 return a->sec_idx - b->sec_idx;
661 return a->sec_offset - b->sec_offset;
662 }
663
bpf_map_type__is_map_in_map(enum bpf_map_type type)664 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
665 {
666 if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
667 type == BPF_MAP_TYPE_HASH_OF_MAPS)
668 return true;
669 return false;
670 }
671
bpf_object_search_section_size(const struct bpf_object * obj,const char * name,size_t * d_size)672 static int bpf_object_search_section_size(const struct bpf_object *obj,
673 const char *name, size_t *d_size)
674 {
675 const GElf_Ehdr *ep = &obj->efile.ehdr;
676 Elf *elf = obj->efile.elf;
677 Elf_Scn *scn = NULL;
678 int idx = 0;
679
680 while ((scn = elf_nextscn(elf, scn)) != NULL) {
681 const char *sec_name;
682 Elf_Data *data;
683 GElf_Shdr sh;
684
685 idx++;
686 if (gelf_getshdr(scn, &sh) != &sh) {
687 pr_warning("failed to get section(%d) header from %s\n",
688 idx, obj->path);
689 return -EIO;
690 }
691
692 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
693 if (!sec_name) {
694 pr_warning("failed to get section(%d) name from %s\n",
695 idx, obj->path);
696 return -EIO;
697 }
698
699 if (strcmp(name, sec_name))
700 continue;
701
702 data = elf_getdata(scn, 0);
703 if (!data) {
704 pr_warning("failed to get section(%d) data from %s(%s)\n",
705 idx, name, obj->path);
706 return -EIO;
707 }
708
709 *d_size = data->d_size;
710 return 0;
711 }
712
713 return -ENOENT;
714 }
715
bpf_object__section_size(const struct bpf_object * obj,const char * name,__u32 * size)716 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
717 __u32 *size)
718 {
719 int ret = -ENOENT;
720 size_t d_size;
721
722 *size = 0;
723 if (!name) {
724 return -EINVAL;
725 } else if (!strcmp(name, ".data")) {
726 if (obj->efile.data)
727 *size = obj->efile.data->d_size;
728 } else if (!strcmp(name, ".bss")) {
729 if (obj->efile.bss)
730 *size = obj->efile.bss->d_size;
731 } else if (!strcmp(name, ".rodata")) {
732 if (obj->efile.rodata)
733 *size = obj->efile.rodata->d_size;
734 } else {
735 ret = bpf_object_search_section_size(obj, name, &d_size);
736 if (!ret)
737 *size = d_size;
738 }
739
740 return *size ? 0 : ret;
741 }
742
bpf_object__variable_offset(const struct bpf_object * obj,const char * name,__u32 * off)743 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
744 __u32 *off)
745 {
746 Elf_Data *symbols = obj->efile.symbols;
747 const char *sname;
748 size_t si;
749
750 if (!name || !off)
751 return -EINVAL;
752
753 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
754 GElf_Sym sym;
755
756 if (!gelf_getsym(symbols, si, &sym))
757 continue;
758 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
759 GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
760 continue;
761
762 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
763 sym.st_name);
764 if (!sname) {
765 pr_warning("failed to get sym name string for var %s\n",
766 name);
767 return -EIO;
768 }
769 if (strcmp(name, sname) == 0) {
770 *off = sym.st_value;
771 return 0;
772 }
773 }
774
775 return -ENOENT;
776 }
777
bpf_object__add_map(struct bpf_object * obj)778 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
779 {
780 struct bpf_map *new_maps;
781 size_t new_cap;
782 int i;
783
784 if (obj->nr_maps < obj->maps_cap)
785 return &obj->maps[obj->nr_maps++];
786
787 new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
788 new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
789 if (!new_maps) {
790 pr_warning("alloc maps for object failed\n");
791 return ERR_PTR(-ENOMEM);
792 }
793
794 obj->maps_cap = new_cap;
795 obj->maps = new_maps;
796
797 /* zero out new maps */
798 memset(obj->maps + obj->nr_maps, 0,
799 (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
800 /*
801 * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
802 * when failure (zclose won't close negative fd)).
803 */
804 for (i = obj->nr_maps; i < obj->maps_cap; i++) {
805 obj->maps[i].fd = -1;
806 obj->maps[i].inner_map_fd = -1;
807 }
808
809 return &obj->maps[obj->nr_maps++];
810 }
811
812 static int
bpf_object__init_internal_map(struct bpf_object * obj,enum libbpf_map_type type,int sec_idx,Elf_Data * data,void ** data_buff)813 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
814 int sec_idx, Elf_Data *data, void **data_buff)
815 {
816 char map_name[BPF_OBJ_NAME_LEN];
817 struct bpf_map_def *def;
818 struct bpf_map *map;
819
820 map = bpf_object__add_map(obj);
821 if (IS_ERR(map))
822 return PTR_ERR(map);
823
824 map->libbpf_type = type;
825 map->sec_idx = sec_idx;
826 map->sec_offset = 0;
827 snprintf(map_name, sizeof(map_name), "%.8s%.7s", obj->name,
828 libbpf_type_to_btf_name[type]);
829 map->name = strdup(map_name);
830 if (!map->name) {
831 pr_warning("failed to alloc map name\n");
832 return -ENOMEM;
833 }
834 pr_debug("map '%s' (global data): at sec_idx %d, offset %zu.\n",
835 map_name, map->sec_idx, map->sec_offset);
836
837 def = &map->def;
838 def->type = BPF_MAP_TYPE_ARRAY;
839 def->key_size = sizeof(int);
840 def->value_size = data->d_size;
841 def->max_entries = 1;
842 def->map_flags = type == LIBBPF_MAP_RODATA ? BPF_F_RDONLY_PROG : 0;
843 if (data_buff) {
844 *data_buff = malloc(data->d_size);
845 if (!*data_buff) {
846 zfree(&map->name);
847 pr_warning("failed to alloc map content buffer\n");
848 return -ENOMEM;
849 }
850 memcpy(*data_buff, data->d_buf, data->d_size);
851 }
852
853 pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
854 return 0;
855 }
856
bpf_object__init_global_data_maps(struct bpf_object * obj)857 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
858 {
859 int err;
860
861 if (!obj->caps.global_data)
862 return 0;
863 /*
864 * Populate obj->maps with libbpf internal maps.
865 */
866 if (obj->efile.data_shndx >= 0) {
867 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
868 obj->efile.data_shndx,
869 obj->efile.data,
870 &obj->sections.data);
871 if (err)
872 return err;
873 }
874 if (obj->efile.rodata_shndx >= 0) {
875 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
876 obj->efile.rodata_shndx,
877 obj->efile.rodata,
878 &obj->sections.rodata);
879 if (err)
880 return err;
881 }
882 if (obj->efile.bss_shndx >= 0) {
883 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
884 obj->efile.bss_shndx,
885 obj->efile.bss, NULL);
886 if (err)
887 return err;
888 }
889 return 0;
890 }
891
bpf_object__init_user_maps(struct bpf_object * obj,bool strict)892 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
893 {
894 Elf_Data *symbols = obj->efile.symbols;
895 int i, map_def_sz = 0, nr_maps = 0, nr_syms;
896 Elf_Data *data = NULL;
897 Elf_Scn *scn;
898
899 if (obj->efile.maps_shndx < 0)
900 return 0;
901
902 if (!symbols)
903 return -EINVAL;
904
905 scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
906 if (scn)
907 data = elf_getdata(scn, NULL);
908 if (!scn || !data) {
909 pr_warning("failed to get Elf_Data from map section %d\n",
910 obj->efile.maps_shndx);
911 return -EINVAL;
912 }
913
914 /*
915 * Count number of maps. Each map has a name.
916 * Array of maps is not supported: only the first element is
917 * considered.
918 *
919 * TODO: Detect array of map and report error.
920 */
921 nr_syms = symbols->d_size / sizeof(GElf_Sym);
922 for (i = 0; i < nr_syms; i++) {
923 GElf_Sym sym;
924
925 if (!gelf_getsym(symbols, i, &sym))
926 continue;
927 if (sym.st_shndx != obj->efile.maps_shndx)
928 continue;
929 nr_maps++;
930 }
931 /* Assume equally sized map definitions */
932 pr_debug("maps in %s: %d maps in %zd bytes\n",
933 obj->path, nr_maps, data->d_size);
934
935 map_def_sz = data->d_size / nr_maps;
936 if (!data->d_size || (data->d_size % nr_maps) != 0) {
937 pr_warning("unable to determine map definition size "
938 "section %s, %d maps in %zd bytes\n",
939 obj->path, nr_maps, data->d_size);
940 return -EINVAL;
941 }
942
943 /* Fill obj->maps using data in "maps" section. */
944 for (i = 0; i < nr_syms; i++) {
945 GElf_Sym sym;
946 const char *map_name;
947 struct bpf_map_def *def;
948 struct bpf_map *map;
949
950 if (!gelf_getsym(symbols, i, &sym))
951 continue;
952 if (sym.st_shndx != obj->efile.maps_shndx)
953 continue;
954
955 map = bpf_object__add_map(obj);
956 if (IS_ERR(map))
957 return PTR_ERR(map);
958
959 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
960 sym.st_name);
961 if (!map_name) {
962 pr_warning("failed to get map #%d name sym string for obj %s\n",
963 i, obj->path);
964 return -LIBBPF_ERRNO__FORMAT;
965 }
966
967 map->libbpf_type = LIBBPF_MAP_UNSPEC;
968 map->sec_idx = sym.st_shndx;
969 map->sec_offset = sym.st_value;
970 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
971 map_name, map->sec_idx, map->sec_offset);
972 if (sym.st_value + map_def_sz > data->d_size) {
973 pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
974 obj->path, map_name);
975 return -EINVAL;
976 }
977
978 map->name = strdup(map_name);
979 if (!map->name) {
980 pr_warning("failed to alloc map name\n");
981 return -ENOMEM;
982 }
983 pr_debug("map %d is \"%s\"\n", i, map->name);
984 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
985 /*
986 * If the definition of the map in the object file fits in
987 * bpf_map_def, copy it. Any extra fields in our version
988 * of bpf_map_def will default to zero as a result of the
989 * calloc above.
990 */
991 if (map_def_sz <= sizeof(struct bpf_map_def)) {
992 memcpy(&map->def, def, map_def_sz);
993 } else {
994 /*
995 * Here the map structure being read is bigger than what
996 * we expect, truncate if the excess bits are all zero.
997 * If they are not zero, reject this map as
998 * incompatible.
999 */
1000 char *b;
1001 for (b = ((char *)def) + sizeof(struct bpf_map_def);
1002 b < ((char *)def) + map_def_sz; b++) {
1003 if (*b != 0) {
1004 pr_warning("maps section in %s: \"%s\" "
1005 "has unrecognized, non-zero "
1006 "options\n",
1007 obj->path, map_name);
1008 if (strict)
1009 return -EINVAL;
1010 }
1011 }
1012 memcpy(&map->def, def, sizeof(struct bpf_map_def));
1013 }
1014 }
1015 return 0;
1016 }
1017
1018 static const struct btf_type *
skip_mods_and_typedefs(const struct btf * btf,__u32 id,__u32 * res_id)1019 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1020 {
1021 const struct btf_type *t = btf__type_by_id(btf, id);
1022
1023 if (res_id)
1024 *res_id = id;
1025
1026 while (btf_is_mod(t) || btf_is_typedef(t)) {
1027 if (res_id)
1028 *res_id = t->type;
1029 t = btf__type_by_id(btf, t->type);
1030 }
1031
1032 return t;
1033 }
1034
1035 /*
1036 * Fetch integer attribute of BTF map definition. Such attributes are
1037 * represented using a pointer to an array, in which dimensionality of array
1038 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1039 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1040 * type definition, while using only sizeof(void *) space in ELF data section.
1041 */
get_map_field_int(const char * map_name,const struct btf * btf,const struct btf_type * def,const struct btf_member * m,__u32 * res)1042 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1043 const struct btf_type *def,
1044 const struct btf_member *m, __u32 *res) {
1045 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
1046 const char *name = btf__name_by_offset(btf, m->name_off);
1047 const struct btf_array *arr_info;
1048 const struct btf_type *arr_t;
1049
1050 if (!btf_is_ptr(t)) {
1051 pr_warning("map '%s': attr '%s': expected PTR, got %u.\n",
1052 map_name, name, btf_kind(t));
1053 return false;
1054 }
1055
1056 arr_t = btf__type_by_id(btf, t->type);
1057 if (!arr_t) {
1058 pr_warning("map '%s': attr '%s': type [%u] not found.\n",
1059 map_name, name, t->type);
1060 return false;
1061 }
1062 if (!btf_is_array(arr_t)) {
1063 pr_warning("map '%s': attr '%s': expected ARRAY, got %u.\n",
1064 map_name, name, btf_kind(arr_t));
1065 return false;
1066 }
1067 arr_info = btf_array(arr_t);
1068 *res = arr_info->nelems;
1069 return true;
1070 }
1071
bpf_object__init_user_btf_map(struct bpf_object * obj,const struct btf_type * sec,int var_idx,int sec_idx,const Elf_Data * data,bool strict)1072 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1073 const struct btf_type *sec,
1074 int var_idx, int sec_idx,
1075 const Elf_Data *data, bool strict)
1076 {
1077 const struct btf_type *var, *def, *t;
1078 const struct btf_var_secinfo *vi;
1079 const struct btf_var *var_extra;
1080 const struct btf_member *m;
1081 const char *map_name;
1082 struct bpf_map *map;
1083 int vlen, i;
1084
1085 vi = btf_var_secinfos(sec) + var_idx;
1086 var = btf__type_by_id(obj->btf, vi->type);
1087 var_extra = btf_var(var);
1088 map_name = btf__name_by_offset(obj->btf, var->name_off);
1089 vlen = btf_vlen(var);
1090
1091 if (map_name == NULL || map_name[0] == '\0') {
1092 pr_warning("map #%d: empty name.\n", var_idx);
1093 return -EINVAL;
1094 }
1095 if ((__u64)vi->offset + vi->size > data->d_size) {
1096 pr_warning("map '%s' BTF data is corrupted.\n", map_name);
1097 return -EINVAL;
1098 }
1099 if (!btf_is_var(var)) {
1100 pr_warning("map '%s': unexpected var kind %u.\n",
1101 map_name, btf_kind(var));
1102 return -EINVAL;
1103 }
1104 if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1105 var_extra->linkage != BTF_VAR_STATIC) {
1106 pr_warning("map '%s': unsupported var linkage %u.\n",
1107 map_name, var_extra->linkage);
1108 return -EOPNOTSUPP;
1109 }
1110
1111 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
1112 if (!btf_is_struct(def)) {
1113 pr_warning("map '%s': unexpected def kind %u.\n",
1114 map_name, btf_kind(var));
1115 return -EINVAL;
1116 }
1117 if (def->size > vi->size) {
1118 pr_warning("map '%s': invalid def size.\n", map_name);
1119 return -EINVAL;
1120 }
1121
1122 map = bpf_object__add_map(obj);
1123 if (IS_ERR(map))
1124 return PTR_ERR(map);
1125 map->name = strdup(map_name);
1126 if (!map->name) {
1127 pr_warning("map '%s': failed to alloc map name.\n", map_name);
1128 return -ENOMEM;
1129 }
1130 map->libbpf_type = LIBBPF_MAP_UNSPEC;
1131 map->def.type = BPF_MAP_TYPE_UNSPEC;
1132 map->sec_idx = sec_idx;
1133 map->sec_offset = vi->offset;
1134 pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1135 map_name, map->sec_idx, map->sec_offset);
1136
1137 vlen = btf_vlen(def);
1138 m = btf_members(def);
1139 for (i = 0; i < vlen; i++, m++) {
1140 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1141
1142 if (!name) {
1143 pr_warning("map '%s': invalid field #%d.\n",
1144 map_name, i);
1145 return -EINVAL;
1146 }
1147 if (strcmp(name, "type") == 0) {
1148 if (!get_map_field_int(map_name, obj->btf, def, m,
1149 &map->def.type))
1150 return -EINVAL;
1151 pr_debug("map '%s': found type = %u.\n",
1152 map_name, map->def.type);
1153 } else if (strcmp(name, "max_entries") == 0) {
1154 if (!get_map_field_int(map_name, obj->btf, def, m,
1155 &map->def.max_entries))
1156 return -EINVAL;
1157 pr_debug("map '%s': found max_entries = %u.\n",
1158 map_name, map->def.max_entries);
1159 } else if (strcmp(name, "map_flags") == 0) {
1160 if (!get_map_field_int(map_name, obj->btf, def, m,
1161 &map->def.map_flags))
1162 return -EINVAL;
1163 pr_debug("map '%s': found map_flags = %u.\n",
1164 map_name, map->def.map_flags);
1165 } else if (strcmp(name, "key_size") == 0) {
1166 __u32 sz;
1167
1168 if (!get_map_field_int(map_name, obj->btf, def, m,
1169 &sz))
1170 return -EINVAL;
1171 pr_debug("map '%s': found key_size = %u.\n",
1172 map_name, sz);
1173 if (map->def.key_size && map->def.key_size != sz) {
1174 pr_warning("map '%s': conflicting key size %u != %u.\n",
1175 map_name, map->def.key_size, sz);
1176 return -EINVAL;
1177 }
1178 map->def.key_size = sz;
1179 } else if (strcmp(name, "key") == 0) {
1180 __s64 sz;
1181
1182 t = btf__type_by_id(obj->btf, m->type);
1183 if (!t) {
1184 pr_warning("map '%s': key type [%d] not found.\n",
1185 map_name, m->type);
1186 return -EINVAL;
1187 }
1188 if (!btf_is_ptr(t)) {
1189 pr_warning("map '%s': key spec is not PTR: %u.\n",
1190 map_name, btf_kind(t));
1191 return -EINVAL;
1192 }
1193 sz = btf__resolve_size(obj->btf, t->type);
1194 if (sz < 0) {
1195 pr_warning("map '%s': can't determine key size for type [%u]: %lld.\n",
1196 map_name, t->type, sz);
1197 return sz;
1198 }
1199 pr_debug("map '%s': found key [%u], sz = %lld.\n",
1200 map_name, t->type, sz);
1201 if (map->def.key_size && map->def.key_size != sz) {
1202 pr_warning("map '%s': conflicting key size %u != %lld.\n",
1203 map_name, map->def.key_size, sz);
1204 return -EINVAL;
1205 }
1206 map->def.key_size = sz;
1207 map->btf_key_type_id = t->type;
1208 } else if (strcmp(name, "value_size") == 0) {
1209 __u32 sz;
1210
1211 if (!get_map_field_int(map_name, obj->btf, def, m,
1212 &sz))
1213 return -EINVAL;
1214 pr_debug("map '%s': found value_size = %u.\n",
1215 map_name, sz);
1216 if (map->def.value_size && map->def.value_size != sz) {
1217 pr_warning("map '%s': conflicting value size %u != %u.\n",
1218 map_name, map->def.value_size, sz);
1219 return -EINVAL;
1220 }
1221 map->def.value_size = sz;
1222 } else if (strcmp(name, "value") == 0) {
1223 __s64 sz;
1224
1225 t = btf__type_by_id(obj->btf, m->type);
1226 if (!t) {
1227 pr_warning("map '%s': value type [%d] not found.\n",
1228 map_name, m->type);
1229 return -EINVAL;
1230 }
1231 if (!btf_is_ptr(t)) {
1232 pr_warning("map '%s': value spec is not PTR: %u.\n",
1233 map_name, btf_kind(t));
1234 return -EINVAL;
1235 }
1236 sz = btf__resolve_size(obj->btf, t->type);
1237 if (sz < 0) {
1238 pr_warning("map '%s': can't determine value size for type [%u]: %lld.\n",
1239 map_name, t->type, sz);
1240 return sz;
1241 }
1242 pr_debug("map '%s': found value [%u], sz = %lld.\n",
1243 map_name, t->type, sz);
1244 if (map->def.value_size && map->def.value_size != sz) {
1245 pr_warning("map '%s': conflicting value size %u != %lld.\n",
1246 map_name, map->def.value_size, sz);
1247 return -EINVAL;
1248 }
1249 map->def.value_size = sz;
1250 map->btf_value_type_id = t->type;
1251 } else {
1252 if (strict) {
1253 pr_warning("map '%s': unknown field '%s'.\n",
1254 map_name, name);
1255 return -ENOTSUP;
1256 }
1257 pr_debug("map '%s': ignoring unknown field '%s'.\n",
1258 map_name, name);
1259 }
1260 }
1261
1262 if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
1263 pr_warning("map '%s': map type isn't specified.\n", map_name);
1264 return -EINVAL;
1265 }
1266
1267 return 0;
1268 }
1269
bpf_object__init_user_btf_maps(struct bpf_object * obj,bool strict)1270 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict)
1271 {
1272 const struct btf_type *sec = NULL;
1273 int nr_types, i, vlen, err;
1274 const struct btf_type *t;
1275 const char *name;
1276 Elf_Data *data;
1277 Elf_Scn *scn;
1278
1279 if (obj->efile.btf_maps_shndx < 0)
1280 return 0;
1281
1282 scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
1283 if (scn)
1284 data = elf_getdata(scn, NULL);
1285 if (!scn || !data) {
1286 pr_warning("failed to get Elf_Data from map section %d (%s)\n",
1287 obj->efile.maps_shndx, MAPS_ELF_SEC);
1288 return -EINVAL;
1289 }
1290
1291 nr_types = btf__get_nr_types(obj->btf);
1292 for (i = 1; i <= nr_types; i++) {
1293 t = btf__type_by_id(obj->btf, i);
1294 if (!btf_is_datasec(t))
1295 continue;
1296 name = btf__name_by_offset(obj->btf, t->name_off);
1297 if (strcmp(name, MAPS_ELF_SEC) == 0) {
1298 sec = t;
1299 break;
1300 }
1301 }
1302
1303 if (!sec) {
1304 pr_warning("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
1305 return -ENOENT;
1306 }
1307
1308 vlen = btf_vlen(sec);
1309 for (i = 0; i < vlen; i++) {
1310 err = bpf_object__init_user_btf_map(obj, sec, i,
1311 obj->efile.btf_maps_shndx,
1312 data, strict);
1313 if (err)
1314 return err;
1315 }
1316
1317 return 0;
1318 }
1319
bpf_object__init_maps(struct bpf_object * obj,int flags)1320 static int bpf_object__init_maps(struct bpf_object *obj, int flags)
1321 {
1322 bool strict = !(flags & MAPS_RELAX_COMPAT);
1323 int err;
1324
1325 err = bpf_object__init_user_maps(obj, strict);
1326 if (err)
1327 return err;
1328
1329 err = bpf_object__init_user_btf_maps(obj, strict);
1330 if (err)
1331 return err;
1332
1333 err = bpf_object__init_global_data_maps(obj);
1334 if (err)
1335 return err;
1336
1337 if (obj->nr_maps) {
1338 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]),
1339 compare_bpf_map);
1340 }
1341 return 0;
1342 }
1343
section_have_execinstr(struct bpf_object * obj,int idx)1344 static bool section_have_execinstr(struct bpf_object *obj, int idx)
1345 {
1346 Elf_Scn *scn;
1347 GElf_Shdr sh;
1348
1349 scn = elf_getscn(obj->efile.elf, idx);
1350 if (!scn)
1351 return false;
1352
1353 if (gelf_getshdr(scn, &sh) != &sh)
1354 return false;
1355
1356 if (sh.sh_flags & SHF_EXECINSTR)
1357 return true;
1358
1359 return false;
1360 }
1361
bpf_object__sanitize_btf(struct bpf_object * obj)1362 static void bpf_object__sanitize_btf(struct bpf_object *obj)
1363 {
1364 bool has_datasec = obj->caps.btf_datasec;
1365 bool has_func = obj->caps.btf_func;
1366 struct btf *btf = obj->btf;
1367 struct btf_type *t;
1368 int i, j, vlen;
1369
1370 if (!obj->btf || (has_func && has_datasec))
1371 return;
1372
1373 for (i = 1; i <= btf__get_nr_types(btf); i++) {
1374 t = (struct btf_type *)btf__type_by_id(btf, i);
1375
1376 if (!has_datasec && btf_is_var(t)) {
1377 /* replace VAR with INT */
1378 t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1379 /*
1380 * using size = 1 is the safest choice, 4 will be too
1381 * big and cause kernel BTF validation failure if
1382 * original variable took less than 4 bytes
1383 */
1384 t->size = 1;
1385 *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
1386 } else if (!has_datasec && btf_is_datasec(t)) {
1387 /* replace DATASEC with STRUCT */
1388 const struct btf_var_secinfo *v = btf_var_secinfos(t);
1389 struct btf_member *m = btf_members(t);
1390 struct btf_type *vt;
1391 char *name;
1392
1393 name = (char *)btf__name_by_offset(btf, t->name_off);
1394 while (*name) {
1395 if (*name == '.')
1396 *name = '_';
1397 name++;
1398 }
1399
1400 vlen = btf_vlen(t);
1401 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
1402 for (j = 0; j < vlen; j++, v++, m++) {
1403 /* order of field assignments is important */
1404 m->offset = v->offset * 8;
1405 m->type = v->type;
1406 /* preserve variable name as member name */
1407 vt = (void *)btf__type_by_id(btf, v->type);
1408 m->name_off = vt->name_off;
1409 }
1410 } else if (!has_func && btf_is_func_proto(t)) {
1411 /* replace FUNC_PROTO with ENUM */
1412 vlen = btf_vlen(t);
1413 t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
1414 t->size = sizeof(__u32); /* kernel enforced */
1415 } else if (!has_func && btf_is_func(t)) {
1416 /* replace FUNC with TYPEDEF */
1417 t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
1418 }
1419 }
1420 }
1421
bpf_object__sanitize_btf_ext(struct bpf_object * obj)1422 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
1423 {
1424 if (!obj->btf_ext)
1425 return;
1426
1427 if (!obj->caps.btf_func) {
1428 btf_ext__free(obj->btf_ext);
1429 obj->btf_ext = NULL;
1430 }
1431 }
1432
bpf_object__is_btf_mandatory(const struct bpf_object * obj)1433 static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
1434 {
1435 return obj->efile.btf_maps_shndx >= 0;
1436 }
1437
bpf_object__init_btf(struct bpf_object * obj,Elf_Data * btf_data,Elf_Data * btf_ext_data)1438 static int bpf_object__init_btf(struct bpf_object *obj,
1439 Elf_Data *btf_data,
1440 Elf_Data *btf_ext_data)
1441 {
1442 bool btf_required = bpf_object__is_btf_mandatory(obj);
1443 int err = 0;
1444
1445 if (btf_data) {
1446 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
1447 if (IS_ERR(obj->btf)) {
1448 pr_warning("Error loading ELF section %s: %d.\n",
1449 BTF_ELF_SEC, err);
1450 goto out;
1451 }
1452 err = btf__finalize_data(obj, obj->btf);
1453 if (err) {
1454 pr_warning("Error finalizing %s: %d.\n",
1455 BTF_ELF_SEC, err);
1456 goto out;
1457 }
1458 }
1459 if (btf_ext_data) {
1460 if (!obj->btf) {
1461 pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
1462 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
1463 goto out;
1464 }
1465 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
1466 btf_ext_data->d_size);
1467 if (IS_ERR(obj->btf_ext)) {
1468 pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
1469 BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
1470 obj->btf_ext = NULL;
1471 goto out;
1472 }
1473 }
1474 out:
1475 if (err || IS_ERR(obj->btf)) {
1476 if (btf_required)
1477 err = err ? : PTR_ERR(obj->btf);
1478 else
1479 err = 0;
1480 if (!IS_ERR_OR_NULL(obj->btf))
1481 btf__free(obj->btf);
1482 obj->btf = NULL;
1483 }
1484 if (btf_required && !obj->btf) {
1485 pr_warning("BTF is required, but is missing or corrupted.\n");
1486 return err == 0 ? -ENOENT : err;
1487 }
1488 return 0;
1489 }
1490
bpf_object__sanitize_and_load_btf(struct bpf_object * obj)1491 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
1492 {
1493 int err = 0;
1494
1495 if (!obj->btf)
1496 return 0;
1497
1498 bpf_object__sanitize_btf(obj);
1499 bpf_object__sanitize_btf_ext(obj);
1500
1501 err = btf__load(obj->btf);
1502 if (err) {
1503 pr_warning("Error loading %s into kernel: %d.\n",
1504 BTF_ELF_SEC, err);
1505 btf__free(obj->btf);
1506 obj->btf = NULL;
1507 /* btf_ext can't exist without btf, so free it as well */
1508 if (obj->btf_ext) {
1509 btf_ext__free(obj->btf_ext);
1510 obj->btf_ext = NULL;
1511 }
1512
1513 if (bpf_object__is_btf_mandatory(obj))
1514 return err;
1515 }
1516 return 0;
1517 }
1518
bpf_object__elf_collect(struct bpf_object * obj,int flags)1519 static int bpf_object__elf_collect(struct bpf_object *obj, int flags)
1520 {
1521 Elf *elf = obj->efile.elf;
1522 GElf_Ehdr *ep = &obj->efile.ehdr;
1523 Elf_Data *btf_ext_data = NULL;
1524 Elf_Data *btf_data = NULL;
1525 Elf_Scn *scn = NULL;
1526 int idx = 0, err = 0;
1527
1528 /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1529 if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
1530 pr_warning("failed to get e_shstrndx from %s\n", obj->path);
1531 return -LIBBPF_ERRNO__FORMAT;
1532 }
1533
1534 while ((scn = elf_nextscn(elf, scn)) != NULL) {
1535 char *name;
1536 GElf_Shdr sh;
1537 Elf_Data *data;
1538
1539 idx++;
1540 if (gelf_getshdr(scn, &sh) != &sh) {
1541 pr_warning("failed to get section(%d) header from %s\n",
1542 idx, obj->path);
1543 return -LIBBPF_ERRNO__FORMAT;
1544 }
1545
1546 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1547 if (!name) {
1548 pr_warning("failed to get section(%d) name from %s\n",
1549 idx, obj->path);
1550 return -LIBBPF_ERRNO__FORMAT;
1551 }
1552
1553 data = elf_getdata(scn, 0);
1554 if (!data) {
1555 pr_warning("failed to get section(%d) data from %s(%s)\n",
1556 idx, name, obj->path);
1557 return -LIBBPF_ERRNO__FORMAT;
1558 }
1559 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
1560 idx, name, (unsigned long)data->d_size,
1561 (int)sh.sh_link, (unsigned long)sh.sh_flags,
1562 (int)sh.sh_type);
1563
1564 if (strcmp(name, "license") == 0) {
1565 err = bpf_object__init_license(obj,
1566 data->d_buf,
1567 data->d_size);
1568 if (err)
1569 return err;
1570 } else if (strcmp(name, "version") == 0) {
1571 err = bpf_object__init_kversion(obj,
1572 data->d_buf,
1573 data->d_size);
1574 if (err)
1575 return err;
1576 } else if (strcmp(name, "maps") == 0) {
1577 obj->efile.maps_shndx = idx;
1578 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
1579 obj->efile.btf_maps_shndx = idx;
1580 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
1581 btf_data = data;
1582 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
1583 btf_ext_data = data;
1584 } else if (sh.sh_type == SHT_SYMTAB) {
1585 if (obj->efile.symbols) {
1586 pr_warning("bpf: multiple SYMTAB in %s\n",
1587 obj->path);
1588 return -LIBBPF_ERRNO__FORMAT;
1589 }
1590 obj->efile.symbols = data;
1591 obj->efile.strtabidx = sh.sh_link;
1592 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
1593 if (sh.sh_flags & SHF_EXECINSTR) {
1594 if (strcmp(name, ".text") == 0)
1595 obj->efile.text_shndx = idx;
1596 err = bpf_object__add_program(obj, data->d_buf,
1597 data->d_size, name, idx);
1598 if (err) {
1599 char errmsg[STRERR_BUFSIZE];
1600 char *cp = libbpf_strerror_r(-err, errmsg,
1601 sizeof(errmsg));
1602
1603 pr_warning("failed to alloc program %s (%s): %s",
1604 name, obj->path, cp);
1605 return err;
1606 }
1607 } else if (strcmp(name, ".data") == 0) {
1608 obj->efile.data = data;
1609 obj->efile.data_shndx = idx;
1610 } else if (strcmp(name, ".rodata") == 0) {
1611 obj->efile.rodata = data;
1612 obj->efile.rodata_shndx = idx;
1613 } else {
1614 pr_debug("skip section(%d) %s\n", idx, name);
1615 }
1616 } else if (sh.sh_type == SHT_REL) {
1617 int nr_reloc = obj->efile.nr_reloc;
1618 void *reloc = obj->efile.reloc;
1619 int sec = sh.sh_info; /* points to other section */
1620
1621 /* Only do relo for section with exec instructions */
1622 if (!section_have_execinstr(obj, sec)) {
1623 pr_debug("skip relo %s(%d) for section(%d)\n",
1624 name, idx, sec);
1625 continue;
1626 }
1627
1628 reloc = reallocarray(reloc, nr_reloc + 1,
1629 sizeof(*obj->efile.reloc));
1630 if (!reloc) {
1631 pr_warning("realloc failed\n");
1632 return -ENOMEM;
1633 }
1634
1635 obj->efile.reloc = reloc;
1636 obj->efile.nr_reloc++;
1637
1638 obj->efile.reloc[nr_reloc].shdr = sh;
1639 obj->efile.reloc[nr_reloc].data = data;
1640 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, ".bss") == 0) {
1641 obj->efile.bss = data;
1642 obj->efile.bss_shndx = idx;
1643 } else {
1644 pr_debug("skip section(%d) %s\n", idx, name);
1645 }
1646 }
1647
1648 if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
1649 pr_warning("Corrupted ELF file: index of strtab invalid\n");
1650 return -LIBBPF_ERRNO__FORMAT;
1651 }
1652 err = bpf_object__init_btf(obj, btf_data, btf_ext_data);
1653 if (!err)
1654 err = bpf_object__init_maps(obj, flags);
1655 if (!err)
1656 err = bpf_object__sanitize_and_load_btf(obj);
1657 if (!err)
1658 err = bpf_object__init_prog_names(obj);
1659 return err;
1660 }
1661
1662 static struct bpf_program *
bpf_object__find_prog_by_idx(struct bpf_object * obj,int idx)1663 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
1664 {
1665 struct bpf_program *prog;
1666 size_t i;
1667
1668 for (i = 0; i < obj->nr_programs; i++) {
1669 prog = &obj->programs[i];
1670 if (prog->idx == idx)
1671 return prog;
1672 }
1673 return NULL;
1674 }
1675
1676 struct bpf_program *
bpf_object__find_program_by_title(const struct bpf_object * obj,const char * title)1677 bpf_object__find_program_by_title(const struct bpf_object *obj,
1678 const char *title)
1679 {
1680 struct bpf_program *pos;
1681
1682 bpf_object__for_each_program(pos, obj) {
1683 if (pos->section_name && !strcmp(pos->section_name, title))
1684 return pos;
1685 }
1686 return NULL;
1687 }
1688
bpf_object__shndx_is_data(const struct bpf_object * obj,int shndx)1689 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
1690 int shndx)
1691 {
1692 return shndx == obj->efile.data_shndx ||
1693 shndx == obj->efile.bss_shndx ||
1694 shndx == obj->efile.rodata_shndx;
1695 }
1696
bpf_object__shndx_is_maps(const struct bpf_object * obj,int shndx)1697 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
1698 int shndx)
1699 {
1700 return shndx == obj->efile.maps_shndx ||
1701 shndx == obj->efile.btf_maps_shndx;
1702 }
1703
bpf_object__relo_in_known_section(const struct bpf_object * obj,int shndx)1704 static bool bpf_object__relo_in_known_section(const struct bpf_object *obj,
1705 int shndx)
1706 {
1707 return shndx == obj->efile.text_shndx ||
1708 bpf_object__shndx_is_maps(obj, shndx) ||
1709 bpf_object__shndx_is_data(obj, shndx);
1710 }
1711
1712 static enum libbpf_map_type
bpf_object__section_to_libbpf_map_type(const struct bpf_object * obj,int shndx)1713 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
1714 {
1715 if (shndx == obj->efile.data_shndx)
1716 return LIBBPF_MAP_DATA;
1717 else if (shndx == obj->efile.bss_shndx)
1718 return LIBBPF_MAP_BSS;
1719 else if (shndx == obj->efile.rodata_shndx)
1720 return LIBBPF_MAP_RODATA;
1721 else
1722 return LIBBPF_MAP_UNSPEC;
1723 }
1724
1725 static int
bpf_program__collect_reloc(struct bpf_program * prog,GElf_Shdr * shdr,Elf_Data * data,struct bpf_object * obj)1726 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
1727 Elf_Data *data, struct bpf_object *obj)
1728 {
1729 Elf_Data *symbols = obj->efile.symbols;
1730 struct bpf_map *maps = obj->maps;
1731 size_t nr_maps = obj->nr_maps;
1732 int i, nrels;
1733
1734 pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
1735 nrels = shdr->sh_size / shdr->sh_entsize;
1736
1737 prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
1738 if (!prog->reloc_desc) {
1739 pr_warning("failed to alloc memory in relocation\n");
1740 return -ENOMEM;
1741 }
1742 prog->nr_reloc = nrels;
1743
1744 for (i = 0; i < nrels; i++) {
1745 struct bpf_insn *insns = prog->insns;
1746 enum libbpf_map_type type;
1747 unsigned int insn_idx;
1748 unsigned int shdr_idx;
1749 const char *name;
1750 size_t map_idx;
1751 GElf_Sym sym;
1752 GElf_Rel rel;
1753
1754 if (!gelf_getrel(data, i, &rel)) {
1755 pr_warning("relocation: failed to get %d reloc\n", i);
1756 return -LIBBPF_ERRNO__FORMAT;
1757 }
1758
1759 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
1760 pr_warning("relocation: symbol %"PRIx64" not found\n",
1761 GELF_R_SYM(rel.r_info));
1762 return -LIBBPF_ERRNO__FORMAT;
1763 }
1764
1765 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1766 sym.st_name) ? : "<?>";
1767
1768 pr_debug("relo for %lld value %lld name %d (\'%s\')\n",
1769 (long long) (rel.r_info >> 32),
1770 (long long) sym.st_value, sym.st_name, name);
1771
1772 shdr_idx = sym.st_shndx;
1773 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
1774 pr_debug("relocation: insn_idx=%u, shdr_idx=%u\n",
1775 insn_idx, shdr_idx);
1776
1777 if (shdr_idx >= SHN_LORESERVE) {
1778 pr_warning("relocation: not yet supported relo for non-static global \'%s\' variable in special section (0x%x) found in insns[%d].code 0x%x\n",
1779 name, shdr_idx, insn_idx,
1780 insns[insn_idx].code);
1781 return -LIBBPF_ERRNO__RELOC;
1782 }
1783 if (!bpf_object__relo_in_known_section(obj, shdr_idx)) {
1784 pr_warning("Program '%s' contains unrecognized relo data pointing to section %u\n",
1785 prog->section_name, shdr_idx);
1786 return -LIBBPF_ERRNO__RELOC;
1787 }
1788
1789 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
1790 if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
1791 pr_warning("incorrect bpf_call opcode\n");
1792 return -LIBBPF_ERRNO__RELOC;
1793 }
1794 prog->reloc_desc[i].type = RELO_CALL;
1795 prog->reloc_desc[i].insn_idx = insn_idx;
1796 prog->reloc_desc[i].text_off = sym.st_value;
1797 obj->has_pseudo_calls = true;
1798 continue;
1799 }
1800
1801 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
1802 pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
1803 insn_idx, insns[insn_idx].code);
1804 return -LIBBPF_ERRNO__RELOC;
1805 }
1806
1807 if (bpf_object__shndx_is_maps(obj, shdr_idx) ||
1808 bpf_object__shndx_is_data(obj, shdr_idx)) {
1809 type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
1810 if (type != LIBBPF_MAP_UNSPEC) {
1811 if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL) {
1812 pr_warning("bpf: relocation: not yet supported relo for non-static global \'%s\' variable found in insns[%d].code 0x%x\n",
1813 name, insn_idx, insns[insn_idx].code);
1814 return -LIBBPF_ERRNO__RELOC;
1815 }
1816 if (!obj->caps.global_data) {
1817 pr_warning("bpf: relocation: kernel does not support global \'%s\' variable access in insns[%d]\n",
1818 name, insn_idx);
1819 return -LIBBPF_ERRNO__RELOC;
1820 }
1821 }
1822
1823 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
1824 if (maps[map_idx].libbpf_type != type)
1825 continue;
1826 if (type != LIBBPF_MAP_UNSPEC ||
1827 (maps[map_idx].sec_idx == sym.st_shndx &&
1828 maps[map_idx].sec_offset == sym.st_value)) {
1829 pr_debug("relocation: found map %zd (%s, sec_idx %d, offset %zu) for insn %u\n",
1830 map_idx, maps[map_idx].name,
1831 maps[map_idx].sec_idx,
1832 maps[map_idx].sec_offset,
1833 insn_idx);
1834 break;
1835 }
1836 }
1837
1838 if (map_idx >= nr_maps) {
1839 pr_warning("bpf relocation: map_idx %d larger than %d\n",
1840 (int)map_idx, (int)nr_maps - 1);
1841 return -LIBBPF_ERRNO__RELOC;
1842 }
1843
1844 prog->reloc_desc[i].type = type != LIBBPF_MAP_UNSPEC ?
1845 RELO_DATA : RELO_LD64;
1846 prog->reloc_desc[i].insn_idx = insn_idx;
1847 prog->reloc_desc[i].map_idx = map_idx;
1848 }
1849 }
1850 return 0;
1851 }
1852
bpf_map_find_btf_info(struct bpf_object * obj,struct bpf_map * map)1853 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
1854 {
1855 struct bpf_map_def *def = &map->def;
1856 __u32 key_type_id = 0, value_type_id = 0;
1857 int ret;
1858
1859 /* if it's BTF-defined map, we don't need to search for type IDs */
1860 if (map->sec_idx == obj->efile.btf_maps_shndx)
1861 return 0;
1862
1863 if (!bpf_map__is_internal(map)) {
1864 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
1865 def->value_size, &key_type_id,
1866 &value_type_id);
1867 } else {
1868 /*
1869 * LLVM annotates global data differently in BTF, that is,
1870 * only as '.data', '.bss' or '.rodata'.
1871 */
1872 ret = btf__find_by_name(obj->btf,
1873 libbpf_type_to_btf_name[map->libbpf_type]);
1874 }
1875 if (ret < 0)
1876 return ret;
1877
1878 map->btf_key_type_id = key_type_id;
1879 map->btf_value_type_id = bpf_map__is_internal(map) ?
1880 ret : value_type_id;
1881 return 0;
1882 }
1883
bpf_map__reuse_fd(struct bpf_map * map,int fd)1884 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1885 {
1886 struct bpf_map_info info = {};
1887 __u32 len = sizeof(info);
1888 int new_fd, err;
1889 char *new_name;
1890
1891 err = bpf_obj_get_info_by_fd(fd, &info, &len);
1892 if (err)
1893 return err;
1894
1895 new_name = strdup(info.name);
1896 if (!new_name)
1897 return -errno;
1898
1899 new_fd = open("/", O_RDONLY | O_CLOEXEC);
1900 if (new_fd < 0) {
1901 err = -errno;
1902 goto err_free_new_name;
1903 }
1904
1905 new_fd = dup3(fd, new_fd, O_CLOEXEC);
1906 if (new_fd < 0) {
1907 err = -errno;
1908 goto err_close_new_fd;
1909 }
1910
1911 err = zclose(map->fd);
1912 if (err) {
1913 err = -errno;
1914 goto err_close_new_fd;
1915 }
1916 free(map->name);
1917
1918 map->fd = new_fd;
1919 map->name = new_name;
1920 map->def.type = info.type;
1921 map->def.key_size = info.key_size;
1922 map->def.value_size = info.value_size;
1923 map->def.max_entries = info.max_entries;
1924 map->def.map_flags = info.map_flags;
1925 map->btf_key_type_id = info.btf_key_type_id;
1926 map->btf_value_type_id = info.btf_value_type_id;
1927
1928 return 0;
1929
1930 err_close_new_fd:
1931 close(new_fd);
1932 err_free_new_name:
1933 free(new_name);
1934 return err;
1935 }
1936
bpf_map__resize(struct bpf_map * map,__u32 max_entries)1937 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
1938 {
1939 if (!map || !max_entries)
1940 return -EINVAL;
1941
1942 /* If map already created, its attributes can't be changed. */
1943 if (map->fd >= 0)
1944 return -EBUSY;
1945
1946 map->def.max_entries = max_entries;
1947
1948 return 0;
1949 }
1950
1951 static int
bpf_object__probe_name(struct bpf_object * obj)1952 bpf_object__probe_name(struct bpf_object *obj)
1953 {
1954 struct bpf_load_program_attr attr;
1955 char *cp, errmsg[STRERR_BUFSIZE];
1956 struct bpf_insn insns[] = {
1957 BPF_MOV64_IMM(BPF_REG_0, 0),
1958 BPF_EXIT_INSN(),
1959 };
1960 int ret;
1961
1962 /* make sure basic loading works */
1963
1964 memset(&attr, 0, sizeof(attr));
1965 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
1966 attr.insns = insns;
1967 attr.insns_cnt = ARRAY_SIZE(insns);
1968 attr.license = "GPL";
1969
1970 ret = bpf_load_program_xattr(&attr, NULL, 0);
1971 if (ret < 0) {
1972 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1973 pr_warning("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
1974 __func__, cp, errno);
1975 return -errno;
1976 }
1977 close(ret);
1978
1979 /* now try the same program, but with the name */
1980
1981 attr.name = "test";
1982 ret = bpf_load_program_xattr(&attr, NULL, 0);
1983 if (ret >= 0) {
1984 obj->caps.name = 1;
1985 close(ret);
1986 }
1987
1988 return 0;
1989 }
1990
1991 static int
bpf_object__probe_global_data(struct bpf_object * obj)1992 bpf_object__probe_global_data(struct bpf_object *obj)
1993 {
1994 struct bpf_load_program_attr prg_attr;
1995 struct bpf_create_map_attr map_attr;
1996 char *cp, errmsg[STRERR_BUFSIZE];
1997 struct bpf_insn insns[] = {
1998 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
1999 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
2000 BPF_MOV64_IMM(BPF_REG_0, 0),
2001 BPF_EXIT_INSN(),
2002 };
2003 int ret, map;
2004
2005 memset(&map_attr, 0, sizeof(map_attr));
2006 map_attr.map_type = BPF_MAP_TYPE_ARRAY;
2007 map_attr.key_size = sizeof(int);
2008 map_attr.value_size = 32;
2009 map_attr.max_entries = 1;
2010
2011 map = bpf_create_map_xattr(&map_attr);
2012 if (map < 0) {
2013 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2014 pr_warning("Error in %s():%s(%d). Couldn't create simple array map.\n",
2015 __func__, cp, errno);
2016 return -errno;
2017 }
2018
2019 insns[0].imm = map;
2020
2021 memset(&prg_attr, 0, sizeof(prg_attr));
2022 prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
2023 prg_attr.insns = insns;
2024 prg_attr.insns_cnt = ARRAY_SIZE(insns);
2025 prg_attr.license = "GPL";
2026
2027 ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
2028 if (ret >= 0) {
2029 obj->caps.global_data = 1;
2030 close(ret);
2031 }
2032
2033 close(map);
2034 return 0;
2035 }
2036
bpf_object__probe_btf_func(struct bpf_object * obj)2037 static int bpf_object__probe_btf_func(struct bpf_object *obj)
2038 {
2039 const char strs[] = "\0int\0x\0a";
2040 /* void x(int a) {} */
2041 __u32 types[] = {
2042 /* int */
2043 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
2044 /* FUNC_PROTO */ /* [2] */
2045 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
2046 BTF_PARAM_ENC(7, 1),
2047 /* FUNC x */ /* [3] */
2048 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
2049 };
2050 int btf_fd;
2051
2052 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2053 strs, sizeof(strs));
2054 if (btf_fd >= 0) {
2055 obj->caps.btf_func = 1;
2056 close(btf_fd);
2057 return 1;
2058 }
2059
2060 return 0;
2061 }
2062
bpf_object__probe_btf_datasec(struct bpf_object * obj)2063 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
2064 {
2065 const char strs[] = "\0x\0.data";
2066 /* static int a; */
2067 __u32 types[] = {
2068 /* int */
2069 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
2070 /* VAR x */ /* [2] */
2071 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
2072 BTF_VAR_STATIC,
2073 /* DATASEC val */ /* [3] */
2074 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
2075 BTF_VAR_SECINFO_ENC(2, 0, 4),
2076 };
2077 int btf_fd;
2078
2079 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2080 strs, sizeof(strs));
2081 if (btf_fd >= 0) {
2082 obj->caps.btf_datasec = 1;
2083 close(btf_fd);
2084 return 1;
2085 }
2086
2087 return 0;
2088 }
2089
2090 static int
bpf_object__probe_caps(struct bpf_object * obj)2091 bpf_object__probe_caps(struct bpf_object *obj)
2092 {
2093 int (*probe_fn[])(struct bpf_object *obj) = {
2094 bpf_object__probe_name,
2095 bpf_object__probe_global_data,
2096 bpf_object__probe_btf_func,
2097 bpf_object__probe_btf_datasec,
2098 };
2099 int i, ret;
2100
2101 for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
2102 ret = probe_fn[i](obj);
2103 if (ret < 0)
2104 pr_debug("Probe #%d failed with %d.\n", i, ret);
2105 }
2106
2107 return 0;
2108 }
2109
2110 static int
bpf_object__populate_internal_map(struct bpf_object * obj,struct bpf_map * map)2111 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
2112 {
2113 char *cp, errmsg[STRERR_BUFSIZE];
2114 int err, zero = 0;
2115 __u8 *data;
2116
2117 /* Nothing to do here since kernel already zero-initializes .bss map. */
2118 if (map->libbpf_type == LIBBPF_MAP_BSS)
2119 return 0;
2120
2121 data = map->libbpf_type == LIBBPF_MAP_DATA ?
2122 obj->sections.data : obj->sections.rodata;
2123
2124 err = bpf_map_update_elem(map->fd, &zero, data, 0);
2125 /* Freeze .rodata map as read-only from syscall side. */
2126 if (!err && map->libbpf_type == LIBBPF_MAP_RODATA) {
2127 err = bpf_map_freeze(map->fd);
2128 if (err) {
2129 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2130 pr_warning("Error freezing map(%s) as read-only: %s\n",
2131 map->name, cp);
2132 err = 0;
2133 }
2134 }
2135 return err;
2136 }
2137
2138 static int
bpf_object__create_maps(struct bpf_object * obj)2139 bpf_object__create_maps(struct bpf_object *obj)
2140 {
2141 struct bpf_create_map_attr create_attr = {};
2142 int nr_cpus = 0;
2143 unsigned int i;
2144 int err;
2145
2146 for (i = 0; i < obj->nr_maps; i++) {
2147 struct bpf_map *map = &obj->maps[i];
2148 struct bpf_map_def *def = &map->def;
2149 char *cp, errmsg[STRERR_BUFSIZE];
2150 int *pfd = &map->fd;
2151
2152 if (map->fd >= 0) {
2153 pr_debug("skip map create (preset) %s: fd=%d\n",
2154 map->name, map->fd);
2155 continue;
2156 }
2157
2158 if (obj->caps.name)
2159 create_attr.name = map->name;
2160 create_attr.map_ifindex = map->map_ifindex;
2161 create_attr.map_type = def->type;
2162 create_attr.map_flags = def->map_flags;
2163 create_attr.key_size = def->key_size;
2164 create_attr.value_size = def->value_size;
2165 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
2166 !def->max_entries) {
2167 if (!nr_cpus)
2168 nr_cpus = libbpf_num_possible_cpus();
2169 if (nr_cpus < 0) {
2170 pr_warning("failed to determine number of system CPUs: %d\n",
2171 nr_cpus);
2172 err = nr_cpus;
2173 goto err_out;
2174 }
2175 pr_debug("map '%s': setting size to %d\n",
2176 map->name, nr_cpus);
2177 create_attr.max_entries = nr_cpus;
2178 } else {
2179 create_attr.max_entries = def->max_entries;
2180 }
2181 create_attr.btf_fd = 0;
2182 create_attr.btf_key_type_id = 0;
2183 create_attr.btf_value_type_id = 0;
2184 if (bpf_map_type__is_map_in_map(def->type) &&
2185 map->inner_map_fd >= 0)
2186 create_attr.inner_map_fd = map->inner_map_fd;
2187
2188 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
2189 create_attr.btf_fd = btf__fd(obj->btf);
2190 create_attr.btf_key_type_id = map->btf_key_type_id;
2191 create_attr.btf_value_type_id = map->btf_value_type_id;
2192 }
2193
2194 *pfd = bpf_create_map_xattr(&create_attr);
2195 if (*pfd < 0 && (create_attr.btf_key_type_id ||
2196 create_attr.btf_value_type_id)) {
2197 err = -errno;
2198 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2199 pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
2200 map->name, cp, err);
2201 create_attr.btf_fd = 0;
2202 create_attr.btf_key_type_id = 0;
2203 create_attr.btf_value_type_id = 0;
2204 map->btf_key_type_id = 0;
2205 map->btf_value_type_id = 0;
2206 *pfd = bpf_create_map_xattr(&create_attr);
2207 }
2208
2209 if (*pfd < 0) {
2210 size_t j;
2211
2212 err = -errno;
2213 err_out:
2214 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2215 pr_warning("failed to create map (name: '%s'): %s(%d)\n",
2216 map->name, cp, err);
2217 for (j = 0; j < i; j++)
2218 zclose(obj->maps[j].fd);
2219 return err;
2220 }
2221
2222 if (bpf_map__is_internal(map)) {
2223 err = bpf_object__populate_internal_map(obj, map);
2224 if (err < 0) {
2225 zclose(*pfd);
2226 goto err_out;
2227 }
2228 }
2229
2230 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
2231 }
2232
2233 return 0;
2234 }
2235
2236 static int
check_btf_ext_reloc_err(struct bpf_program * prog,int err,void * btf_prog_info,const char * info_name)2237 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
2238 void *btf_prog_info, const char *info_name)
2239 {
2240 if (err != -ENOENT) {
2241 pr_warning("Error in loading %s for sec %s.\n",
2242 info_name, prog->section_name);
2243 return err;
2244 }
2245
2246 /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
2247
2248 if (btf_prog_info) {
2249 /*
2250 * Some info has already been found but has problem
2251 * in the last btf_ext reloc. Must have to error out.
2252 */
2253 pr_warning("Error in relocating %s for sec %s.\n",
2254 info_name, prog->section_name);
2255 return err;
2256 }
2257
2258 /* Have problem loading the very first info. Ignore the rest. */
2259 pr_warning("Cannot find %s for main program sec %s. Ignore all %s.\n",
2260 info_name, prog->section_name, info_name);
2261 return 0;
2262 }
2263
2264 static int
bpf_program_reloc_btf_ext(struct bpf_program * prog,struct bpf_object * obj,const char * section_name,__u32 insn_offset)2265 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
2266 const char *section_name, __u32 insn_offset)
2267 {
2268 int err;
2269
2270 if (!insn_offset || prog->func_info) {
2271 /*
2272 * !insn_offset => main program
2273 *
2274 * For sub prog, the main program's func_info has to
2275 * be loaded first (i.e. prog->func_info != NULL)
2276 */
2277 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
2278 section_name, insn_offset,
2279 &prog->func_info,
2280 &prog->func_info_cnt);
2281 if (err)
2282 return check_btf_ext_reloc_err(prog, err,
2283 prog->func_info,
2284 "bpf_func_info");
2285
2286 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
2287 }
2288
2289 if (!insn_offset || prog->line_info) {
2290 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
2291 section_name, insn_offset,
2292 &prog->line_info,
2293 &prog->line_info_cnt);
2294 if (err)
2295 return check_btf_ext_reloc_err(prog, err,
2296 prog->line_info,
2297 "bpf_line_info");
2298
2299 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
2300 }
2301
2302 return 0;
2303 }
2304
2305 #define BPF_CORE_SPEC_MAX_LEN 64
2306
2307 /* represents BPF CO-RE field or array element accessor */
2308 struct bpf_core_accessor {
2309 __u32 type_id; /* struct/union type or array element type */
2310 __u32 idx; /* field index or array index */
2311 const char *name; /* field name or NULL for array accessor */
2312 };
2313
2314 struct bpf_core_spec {
2315 const struct btf *btf;
2316 /* high-level spec: named fields and array indices only */
2317 struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
2318 /* high-level spec length */
2319 int len;
2320 /* raw, low-level spec: 1-to-1 with accessor spec string */
2321 int raw_spec[BPF_CORE_SPEC_MAX_LEN];
2322 /* raw spec length */
2323 int raw_len;
2324 /* field byte offset represented by spec */
2325 __u32 offset;
2326 };
2327
str_is_empty(const char * s)2328 static bool str_is_empty(const char *s)
2329 {
2330 return !s || !s[0];
2331 }
2332
2333 /*
2334 * Turn bpf_offset_reloc into a low- and high-level spec representation,
2335 * validating correctness along the way, as well as calculating resulting
2336 * field offset (in bytes), specified by accessor string. Low-level spec
2337 * captures every single level of nestedness, including traversing anonymous
2338 * struct/union members. High-level one only captures semantically meaningful
2339 * "turning points": named fields and array indicies.
2340 * E.g., for this case:
2341 *
2342 * struct sample {
2343 * int __unimportant;
2344 * struct {
2345 * int __1;
2346 * int __2;
2347 * int a[7];
2348 * };
2349 * };
2350 *
2351 * struct sample *s = ...;
2352 *
2353 * int x = &s->a[3]; // access string = '0:1:2:3'
2354 *
2355 * Low-level spec has 1:1 mapping with each element of access string (it's
2356 * just a parsed access string representation): [0, 1, 2, 3].
2357 *
2358 * High-level spec will capture only 3 points:
2359 * - intial zero-index access by pointer (&s->... is the same as &s[0]...);
2360 * - field 'a' access (corresponds to '2' in low-level spec);
2361 * - array element #3 access (corresponds to '3' in low-level spec).
2362 *
2363 */
bpf_core_spec_parse(const struct btf * btf,__u32 type_id,const char * spec_str,struct bpf_core_spec * spec)2364 static int bpf_core_spec_parse(const struct btf *btf,
2365 __u32 type_id,
2366 const char *spec_str,
2367 struct bpf_core_spec *spec)
2368 {
2369 int access_idx, parsed_len, i;
2370 const struct btf_type *t;
2371 const char *name;
2372 __u32 id;
2373 __s64 sz;
2374
2375 if (str_is_empty(spec_str) || *spec_str == ':')
2376 return -EINVAL;
2377
2378 memset(spec, 0, sizeof(*spec));
2379 spec->btf = btf;
2380
2381 /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
2382 while (*spec_str) {
2383 if (*spec_str == ':')
2384 ++spec_str;
2385 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
2386 return -EINVAL;
2387 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2388 return -E2BIG;
2389 spec_str += parsed_len;
2390 spec->raw_spec[spec->raw_len++] = access_idx;
2391 }
2392
2393 if (spec->raw_len == 0)
2394 return -EINVAL;
2395
2396 /* first spec value is always reloc type array index */
2397 t = skip_mods_and_typedefs(btf, type_id, &id);
2398 if (!t)
2399 return -EINVAL;
2400
2401 access_idx = spec->raw_spec[0];
2402 spec->spec[0].type_id = id;
2403 spec->spec[0].idx = access_idx;
2404 spec->len++;
2405
2406 sz = btf__resolve_size(btf, id);
2407 if (sz < 0)
2408 return sz;
2409 spec->offset = access_idx * sz;
2410
2411 for (i = 1; i < spec->raw_len; i++) {
2412 t = skip_mods_and_typedefs(btf, id, &id);
2413 if (!t)
2414 return -EINVAL;
2415
2416 access_idx = spec->raw_spec[i];
2417
2418 if (btf_is_composite(t)) {
2419 const struct btf_member *m;
2420 __u32 offset;
2421
2422 if (access_idx >= btf_vlen(t))
2423 return -EINVAL;
2424 if (btf_member_bitfield_size(t, access_idx))
2425 return -EINVAL;
2426
2427 offset = btf_member_bit_offset(t, access_idx);
2428 if (offset % 8)
2429 return -EINVAL;
2430 spec->offset += offset / 8;
2431
2432 m = btf_members(t) + access_idx;
2433 if (m->name_off) {
2434 name = btf__name_by_offset(btf, m->name_off);
2435 if (str_is_empty(name))
2436 return -EINVAL;
2437
2438 spec->spec[spec->len].type_id = id;
2439 spec->spec[spec->len].idx = access_idx;
2440 spec->spec[spec->len].name = name;
2441 spec->len++;
2442 }
2443
2444 id = m->type;
2445 } else if (btf_is_array(t)) {
2446 const struct btf_array *a = btf_array(t);
2447
2448 t = skip_mods_and_typedefs(btf, a->type, &id);
2449 if (!t || access_idx >= a->nelems)
2450 return -EINVAL;
2451
2452 spec->spec[spec->len].type_id = id;
2453 spec->spec[spec->len].idx = access_idx;
2454 spec->len++;
2455
2456 sz = btf__resolve_size(btf, id);
2457 if (sz < 0)
2458 return sz;
2459 spec->offset += access_idx * sz;
2460 } else {
2461 pr_warning("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
2462 type_id, spec_str, i, id, btf_kind(t));
2463 return -EINVAL;
2464 }
2465 }
2466
2467 return 0;
2468 }
2469
bpf_core_is_flavor_sep(const char * s)2470 static bool bpf_core_is_flavor_sep(const char *s)
2471 {
2472 /* check X___Y name pattern, where X and Y are not underscores */
2473 return s[0] != '_' && /* X */
2474 s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */
2475 s[4] != '_'; /* Y */
2476 }
2477
2478 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
2479 * before last triple underscore. Struct name part after last triple
2480 * underscore is ignored by BPF CO-RE relocation during relocation matching.
2481 */
bpf_core_essential_name_len(const char * name)2482 static size_t bpf_core_essential_name_len(const char *name)
2483 {
2484 size_t n = strlen(name);
2485 int i;
2486
2487 for (i = n - 5; i >= 0; i--) {
2488 if (bpf_core_is_flavor_sep(name + i))
2489 return i + 1;
2490 }
2491 return n;
2492 }
2493
2494 /* dynamically sized list of type IDs */
2495 struct ids_vec {
2496 __u32 *data;
2497 int len;
2498 };
2499
bpf_core_free_cands(struct ids_vec * cand_ids)2500 static void bpf_core_free_cands(struct ids_vec *cand_ids)
2501 {
2502 free(cand_ids->data);
2503 free(cand_ids);
2504 }
2505
bpf_core_find_cands(const struct btf * local_btf,__u32 local_type_id,const struct btf * targ_btf)2506 static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
2507 __u32 local_type_id,
2508 const struct btf *targ_btf)
2509 {
2510 size_t local_essent_len, targ_essent_len;
2511 const char *local_name, *targ_name;
2512 const struct btf_type *t;
2513 struct ids_vec *cand_ids;
2514 __u32 *new_ids;
2515 int i, err, n;
2516
2517 t = btf__type_by_id(local_btf, local_type_id);
2518 if (!t)
2519 return ERR_PTR(-EINVAL);
2520
2521 local_name = btf__name_by_offset(local_btf, t->name_off);
2522 if (str_is_empty(local_name))
2523 return ERR_PTR(-EINVAL);
2524 local_essent_len = bpf_core_essential_name_len(local_name);
2525
2526 cand_ids = calloc(1, sizeof(*cand_ids));
2527 if (!cand_ids)
2528 return ERR_PTR(-ENOMEM);
2529
2530 n = btf__get_nr_types(targ_btf);
2531 for (i = 1; i <= n; i++) {
2532 t = btf__type_by_id(targ_btf, i);
2533 targ_name = btf__name_by_offset(targ_btf, t->name_off);
2534 if (str_is_empty(targ_name))
2535 continue;
2536
2537 targ_essent_len = bpf_core_essential_name_len(targ_name);
2538 if (targ_essent_len != local_essent_len)
2539 continue;
2540
2541 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
2542 pr_debug("[%d] %s: found candidate [%d] %s\n",
2543 local_type_id, local_name, i, targ_name);
2544 new_ids = realloc(cand_ids->data, cand_ids->len + 1);
2545 if (!new_ids) {
2546 err = -ENOMEM;
2547 goto err_out;
2548 }
2549 cand_ids->data = new_ids;
2550 cand_ids->data[cand_ids->len++] = i;
2551 }
2552 }
2553 return cand_ids;
2554 err_out:
2555 bpf_core_free_cands(cand_ids);
2556 return ERR_PTR(err);
2557 }
2558
2559 /* Check two types for compatibility, skipping const/volatile/restrict and
2560 * typedefs, to ensure we are relocating offset to the compatible entities:
2561 * - any two STRUCTs/UNIONs are compatible and can be mixed;
2562 * - any two FWDs are compatible;
2563 * - any two PTRs are always compatible;
2564 * - for ENUMs, check sizes, names are ignored;
2565 * - for INT, size and bitness should match, signedness is ignored;
2566 * - for ARRAY, dimensionality is ignored, element types are checked for
2567 * compatibility recursively;
2568 * - everything else shouldn't be ever a target of relocation.
2569 * These rules are not set in stone and probably will be adjusted as we get
2570 * more experience with using BPF CO-RE relocations.
2571 */
bpf_core_fields_are_compat(const struct btf * local_btf,__u32 local_id,const struct btf * targ_btf,__u32 targ_id)2572 static int bpf_core_fields_are_compat(const struct btf *local_btf,
2573 __u32 local_id,
2574 const struct btf *targ_btf,
2575 __u32 targ_id)
2576 {
2577 const struct btf_type *local_type, *targ_type;
2578
2579 recur:
2580 local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
2581 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2582 if (!local_type || !targ_type)
2583 return -EINVAL;
2584
2585 if (btf_is_composite(local_type) && btf_is_composite(targ_type))
2586 return 1;
2587 if (btf_kind(local_type) != btf_kind(targ_type))
2588 return 0;
2589
2590 switch (btf_kind(local_type)) {
2591 case BTF_KIND_FWD:
2592 case BTF_KIND_PTR:
2593 return 1;
2594 case BTF_KIND_ENUM:
2595 return local_type->size == targ_type->size;
2596 case BTF_KIND_INT:
2597 return btf_int_offset(local_type) == 0 &&
2598 btf_int_offset(targ_type) == 0 &&
2599 local_type->size == targ_type->size &&
2600 btf_int_bits(local_type) == btf_int_bits(targ_type);
2601 case BTF_KIND_ARRAY:
2602 local_id = btf_array(local_type)->type;
2603 targ_id = btf_array(targ_type)->type;
2604 goto recur;
2605 default:
2606 pr_warning("unexpected kind %d relocated, local [%d], target [%d]\n",
2607 btf_kind(local_type), local_id, targ_id);
2608 return 0;
2609 }
2610 }
2611
2612 /*
2613 * Given single high-level named field accessor in local type, find
2614 * corresponding high-level accessor for a target type. Along the way,
2615 * maintain low-level spec for target as well. Also keep updating target
2616 * offset.
2617 *
2618 * Searching is performed through recursive exhaustive enumeration of all
2619 * fields of a struct/union. If there are any anonymous (embedded)
2620 * structs/unions, they are recursively searched as well. If field with
2621 * desired name is found, check compatibility between local and target types,
2622 * before returning result.
2623 *
2624 * 1 is returned, if field is found.
2625 * 0 is returned if no compatible field is found.
2626 * <0 is returned on error.
2627 */
bpf_core_match_member(const struct btf * local_btf,const struct bpf_core_accessor * local_acc,const struct btf * targ_btf,__u32 targ_id,struct bpf_core_spec * spec,__u32 * next_targ_id)2628 static int bpf_core_match_member(const struct btf *local_btf,
2629 const struct bpf_core_accessor *local_acc,
2630 const struct btf *targ_btf,
2631 __u32 targ_id,
2632 struct bpf_core_spec *spec,
2633 __u32 *next_targ_id)
2634 {
2635 const struct btf_type *local_type, *targ_type;
2636 const struct btf_member *local_member, *m;
2637 const char *local_name, *targ_name;
2638 __u32 local_id;
2639 int i, n, found;
2640
2641 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2642 if (!targ_type)
2643 return -EINVAL;
2644 if (!btf_is_composite(targ_type))
2645 return 0;
2646
2647 local_id = local_acc->type_id;
2648 local_type = btf__type_by_id(local_btf, local_id);
2649 local_member = btf_members(local_type) + local_acc->idx;
2650 local_name = btf__name_by_offset(local_btf, local_member->name_off);
2651
2652 n = btf_vlen(targ_type);
2653 m = btf_members(targ_type);
2654 for (i = 0; i < n; i++, m++) {
2655 __u32 offset;
2656
2657 /* bitfield relocations not supported */
2658 if (btf_member_bitfield_size(targ_type, i))
2659 continue;
2660 offset = btf_member_bit_offset(targ_type, i);
2661 if (offset % 8)
2662 continue;
2663
2664 /* too deep struct/union/array nesting */
2665 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2666 return -E2BIG;
2667
2668 /* speculate this member will be the good one */
2669 spec->offset += offset / 8;
2670 spec->raw_spec[spec->raw_len++] = i;
2671
2672 targ_name = btf__name_by_offset(targ_btf, m->name_off);
2673 if (str_is_empty(targ_name)) {
2674 /* embedded struct/union, we need to go deeper */
2675 found = bpf_core_match_member(local_btf, local_acc,
2676 targ_btf, m->type,
2677 spec, next_targ_id);
2678 if (found) /* either found or error */
2679 return found;
2680 } else if (strcmp(local_name, targ_name) == 0) {
2681 /* matching named field */
2682 struct bpf_core_accessor *targ_acc;
2683
2684 targ_acc = &spec->spec[spec->len++];
2685 targ_acc->type_id = targ_id;
2686 targ_acc->idx = i;
2687 targ_acc->name = targ_name;
2688
2689 *next_targ_id = m->type;
2690 found = bpf_core_fields_are_compat(local_btf,
2691 local_member->type,
2692 targ_btf, m->type);
2693 if (!found)
2694 spec->len--; /* pop accessor */
2695 return found;
2696 }
2697 /* member turned out not to be what we looked for */
2698 spec->offset -= offset / 8;
2699 spec->raw_len--;
2700 }
2701
2702 return 0;
2703 }
2704
2705 /*
2706 * Try to match local spec to a target type and, if successful, produce full
2707 * target spec (high-level, low-level + offset).
2708 */
bpf_core_spec_match(struct bpf_core_spec * local_spec,const struct btf * targ_btf,__u32 targ_id,struct bpf_core_spec * targ_spec)2709 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
2710 const struct btf *targ_btf, __u32 targ_id,
2711 struct bpf_core_spec *targ_spec)
2712 {
2713 const struct btf_type *targ_type;
2714 const struct bpf_core_accessor *local_acc;
2715 struct bpf_core_accessor *targ_acc;
2716 int i, sz, matched;
2717
2718 memset(targ_spec, 0, sizeof(*targ_spec));
2719 targ_spec->btf = targ_btf;
2720
2721 local_acc = &local_spec->spec[0];
2722 targ_acc = &targ_spec->spec[0];
2723
2724 for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
2725 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
2726 &targ_id);
2727 if (!targ_type)
2728 return -EINVAL;
2729
2730 if (local_acc->name) {
2731 matched = bpf_core_match_member(local_spec->btf,
2732 local_acc,
2733 targ_btf, targ_id,
2734 targ_spec, &targ_id);
2735 if (matched <= 0)
2736 return matched;
2737 } else {
2738 /* for i=0, targ_id is already treated as array element
2739 * type (because it's the original struct), for others
2740 * we should find array element type first
2741 */
2742 if (i > 0) {
2743 const struct btf_array *a;
2744
2745 if (!btf_is_array(targ_type))
2746 return 0;
2747
2748 a = btf_array(targ_type);
2749 if (local_acc->idx >= a->nelems)
2750 return 0;
2751 if (!skip_mods_and_typedefs(targ_btf, a->type,
2752 &targ_id))
2753 return -EINVAL;
2754 }
2755
2756 /* too deep struct/union/array nesting */
2757 if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2758 return -E2BIG;
2759
2760 targ_acc->type_id = targ_id;
2761 targ_acc->idx = local_acc->idx;
2762 targ_acc->name = NULL;
2763 targ_spec->len++;
2764 targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
2765 targ_spec->raw_len++;
2766
2767 sz = btf__resolve_size(targ_btf, targ_id);
2768 if (sz < 0)
2769 return sz;
2770 targ_spec->offset += local_acc->idx * sz;
2771 }
2772 }
2773
2774 return 1;
2775 }
2776
2777 /*
2778 * Patch relocatable BPF instruction.
2779 * Expected insn->imm value is provided for validation, as well as the new
2780 * relocated value.
2781 *
2782 * Currently three kinds of BPF instructions are supported:
2783 * 1. rX = <imm> (assignment with immediate operand);
2784 * 2. rX += <imm> (arithmetic operations with immediate operand);
2785 * 3. *(rX) = <imm> (indirect memory assignment with immediate operand).
2786 *
2787 * If actual insn->imm value is wrong, bail out.
2788 */
bpf_core_reloc_insn(struct bpf_program * prog,int insn_off,__u32 orig_off,__u32 new_off)2789 static int bpf_core_reloc_insn(struct bpf_program *prog, int insn_off,
2790 __u32 orig_off, __u32 new_off)
2791 {
2792 struct bpf_insn *insn;
2793 int insn_idx;
2794 __u8 class;
2795
2796 if (insn_off % sizeof(struct bpf_insn))
2797 return -EINVAL;
2798 insn_idx = insn_off / sizeof(struct bpf_insn);
2799
2800 insn = &prog->insns[insn_idx];
2801 class = BPF_CLASS(insn->code);
2802
2803 if (class == BPF_ALU || class == BPF_ALU64) {
2804 if (BPF_SRC(insn->code) != BPF_K)
2805 return -EINVAL;
2806 if (insn->imm != orig_off)
2807 return -EINVAL;
2808 insn->imm = new_off;
2809 pr_debug("prog '%s': patched insn #%d (ALU/ALU64) imm %d -> %d\n",
2810 bpf_program__title(prog, false),
2811 insn_idx, orig_off, new_off);
2812 } else {
2813 pr_warning("prog '%s': trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
2814 bpf_program__title(prog, false),
2815 insn_idx, insn->code, insn->src_reg, insn->dst_reg,
2816 insn->off, insn->imm);
2817 return -EINVAL;
2818 }
2819 return 0;
2820 }
2821
btf_load_raw(const char * path)2822 static struct btf *btf_load_raw(const char *path)
2823 {
2824 struct btf *btf;
2825 size_t read_cnt;
2826 struct stat st;
2827 void *data;
2828 FILE *f;
2829
2830 if (stat(path, &st))
2831 return ERR_PTR(-errno);
2832
2833 data = malloc(st.st_size);
2834 if (!data)
2835 return ERR_PTR(-ENOMEM);
2836
2837 f = fopen(path, "rb");
2838 if (!f) {
2839 btf = ERR_PTR(-errno);
2840 goto cleanup;
2841 }
2842
2843 read_cnt = fread(data, 1, st.st_size, f);
2844 fclose(f);
2845 if (read_cnt < st.st_size) {
2846 btf = ERR_PTR(-EBADF);
2847 goto cleanup;
2848 }
2849
2850 btf = btf__new(data, read_cnt);
2851
2852 cleanup:
2853 free(data);
2854 return btf;
2855 }
2856
2857 /*
2858 * Probe few well-known locations for vmlinux kernel image and try to load BTF
2859 * data out of it to use for target BTF.
2860 */
bpf_core_find_kernel_btf(void)2861 static struct btf *bpf_core_find_kernel_btf(void)
2862 {
2863 struct {
2864 const char *path_fmt;
2865 bool raw_btf;
2866 } locations[] = {
2867 /* try canonical vmlinux BTF through sysfs first */
2868 { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
2869 /* fall back to trying to find vmlinux ELF on disk otherwise */
2870 { "/boot/vmlinux-%1$s" },
2871 { "/lib/modules/%1$s/vmlinux-%1$s" },
2872 { "/lib/modules/%1$s/build/vmlinux" },
2873 { "/usr/lib/modules/%1$s/kernel/vmlinux" },
2874 { "/usr/lib/debug/boot/vmlinux-%1$s" },
2875 { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
2876 { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
2877 };
2878 char path[PATH_MAX + 1];
2879 struct utsname buf;
2880 struct btf *btf;
2881 int i;
2882
2883 uname(&buf);
2884
2885 for (i = 0; i < ARRAY_SIZE(locations); i++) {
2886 snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
2887
2888 if (access(path, R_OK))
2889 continue;
2890
2891 if (locations[i].raw_btf)
2892 btf = btf_load_raw(path);
2893 else
2894 btf = btf__parse_elf(path, NULL);
2895
2896 pr_debug("loading kernel BTF '%s': %ld\n",
2897 path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
2898 if (IS_ERR(btf))
2899 continue;
2900
2901 return btf;
2902 }
2903
2904 pr_warning("failed to find valid kernel BTF\n");
2905 return ERR_PTR(-ESRCH);
2906 }
2907
2908 /* Output spec definition in the format:
2909 * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
2910 * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
2911 */
bpf_core_dump_spec(int level,const struct bpf_core_spec * spec)2912 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
2913 {
2914 const struct btf_type *t;
2915 const char *s;
2916 __u32 type_id;
2917 int i;
2918
2919 type_id = spec->spec[0].type_id;
2920 t = btf__type_by_id(spec->btf, type_id);
2921 s = btf__name_by_offset(spec->btf, t->name_off);
2922 libbpf_print(level, "[%u] %s + ", type_id, s);
2923
2924 for (i = 0; i < spec->raw_len; i++)
2925 libbpf_print(level, "%d%s", spec->raw_spec[i],
2926 i == spec->raw_len - 1 ? " => " : ":");
2927
2928 libbpf_print(level, "%u @ &x", spec->offset);
2929
2930 for (i = 0; i < spec->len; i++) {
2931 if (spec->spec[i].name)
2932 libbpf_print(level, ".%s", spec->spec[i].name);
2933 else
2934 libbpf_print(level, "[%u]", spec->spec[i].idx);
2935 }
2936
2937 }
2938
bpf_core_hash_fn(const void * key,void * ctx)2939 static size_t bpf_core_hash_fn(const void *key, void *ctx)
2940 {
2941 return (size_t)key;
2942 }
2943
bpf_core_equal_fn(const void * k1,const void * k2,void * ctx)2944 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
2945 {
2946 return k1 == k2;
2947 }
2948
u32_as_hash_key(__u32 x)2949 static void *u32_as_hash_key(__u32 x)
2950 {
2951 return (void *)(uintptr_t)x;
2952 }
2953
2954 /*
2955 * CO-RE relocate single instruction.
2956 *
2957 * The outline and important points of the algorithm:
2958 * 1. For given local type, find corresponding candidate target types.
2959 * Candidate type is a type with the same "essential" name, ignoring
2960 * everything after last triple underscore (___). E.g., `sample`,
2961 * `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
2962 * for each other. Names with triple underscore are referred to as
2963 * "flavors" and are useful, among other things, to allow to
2964 * specify/support incompatible variations of the same kernel struct, which
2965 * might differ between different kernel versions and/or build
2966 * configurations.
2967 *
2968 * N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
2969 * converter, when deduplicated BTF of a kernel still contains more than
2970 * one different types with the same name. In that case, ___2, ___3, etc
2971 * are appended starting from second name conflict. But start flavors are
2972 * also useful to be defined "locally", in BPF program, to extract same
2973 * data from incompatible changes between different kernel
2974 * versions/configurations. For instance, to handle field renames between
2975 * kernel versions, one can use two flavors of the struct name with the
2976 * same common name and use conditional relocations to extract that field,
2977 * depending on target kernel version.
2978 * 2. For each candidate type, try to match local specification to this
2979 * candidate target type. Matching involves finding corresponding
2980 * high-level spec accessors, meaning that all named fields should match,
2981 * as well as all array accesses should be within the actual bounds. Also,
2982 * types should be compatible (see bpf_core_fields_are_compat for details).
2983 * 3. It is supported and expected that there might be multiple flavors
2984 * matching the spec. As long as all the specs resolve to the same set of
2985 * offsets across all candidates, there is not error. If there is any
2986 * ambiguity, CO-RE relocation will fail. This is necessary to accomodate
2987 * imprefection of BTF deduplication, which can cause slight duplication of
2988 * the same BTF type, if some directly or indirectly referenced (by
2989 * pointer) type gets resolved to different actual types in different
2990 * object files. If such situation occurs, deduplicated BTF will end up
2991 * with two (or more) structurally identical types, which differ only in
2992 * types they refer to through pointer. This should be OK in most cases and
2993 * is not an error.
2994 * 4. Candidate types search is performed by linearly scanning through all
2995 * types in target BTF. It is anticipated that this is overall more
2996 * efficient memory-wise and not significantly worse (if not better)
2997 * CPU-wise compared to prebuilding a map from all local type names to
2998 * a list of candidate type names. It's also sped up by caching resolved
2999 * list of matching candidates per each local "root" type ID, that has at
3000 * least one bpf_offset_reloc associated with it. This list is shared
3001 * between multiple relocations for the same type ID and is updated as some
3002 * of the candidates are pruned due to structural incompatibility.
3003 */
bpf_core_reloc_offset(struct bpf_program * prog,const struct bpf_offset_reloc * relo,int relo_idx,const struct btf * local_btf,const struct btf * targ_btf,struct hashmap * cand_cache)3004 static int bpf_core_reloc_offset(struct bpf_program *prog,
3005 const struct bpf_offset_reloc *relo,
3006 int relo_idx,
3007 const struct btf *local_btf,
3008 const struct btf *targ_btf,
3009 struct hashmap *cand_cache)
3010 {
3011 const char *prog_name = bpf_program__title(prog, false);
3012 struct bpf_core_spec local_spec, cand_spec, targ_spec;
3013 const void *type_key = u32_as_hash_key(relo->type_id);
3014 const struct btf_type *local_type, *cand_type;
3015 const char *local_name, *cand_name;
3016 struct ids_vec *cand_ids;
3017 __u32 local_id, cand_id;
3018 const char *spec_str;
3019 int i, j, err;
3020
3021 local_id = relo->type_id;
3022 local_type = btf__type_by_id(local_btf, local_id);
3023 if (!local_type)
3024 return -EINVAL;
3025
3026 local_name = btf__name_by_offset(local_btf, local_type->name_off);
3027 if (str_is_empty(local_name))
3028 return -EINVAL;
3029
3030 spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
3031 if (str_is_empty(spec_str))
3032 return -EINVAL;
3033
3034 err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
3035 if (err) {
3036 pr_warning("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
3037 prog_name, relo_idx, local_id, local_name, spec_str,
3038 err);
3039 return -EINVAL;
3040 }
3041
3042 pr_debug("prog '%s': relo #%d: spec is ", prog_name, relo_idx);
3043 bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
3044 libbpf_print(LIBBPF_DEBUG, "\n");
3045
3046 if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
3047 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
3048 if (IS_ERR(cand_ids)) {
3049 pr_warning("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
3050 prog_name, relo_idx, local_id, local_name,
3051 PTR_ERR(cand_ids));
3052 return PTR_ERR(cand_ids);
3053 }
3054 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
3055 if (err) {
3056 bpf_core_free_cands(cand_ids);
3057 return err;
3058 }
3059 }
3060
3061 for (i = 0, j = 0; i < cand_ids->len; i++) {
3062 cand_id = cand_ids->data[i];
3063 cand_type = btf__type_by_id(targ_btf, cand_id);
3064 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
3065
3066 err = bpf_core_spec_match(&local_spec, targ_btf,
3067 cand_id, &cand_spec);
3068 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
3069 prog_name, relo_idx, i, cand_name);
3070 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
3071 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
3072 if (err < 0) {
3073 pr_warning("prog '%s': relo #%d: matching error: %d\n",
3074 prog_name, relo_idx, err);
3075 return err;
3076 }
3077 if (err == 0)
3078 continue;
3079
3080 if (j == 0) {
3081 targ_spec = cand_spec;
3082 } else if (cand_spec.offset != targ_spec.offset) {
3083 /* if there are many candidates, they should all
3084 * resolve to the same offset
3085 */
3086 pr_warning("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
3087 prog_name, relo_idx, cand_spec.offset,
3088 targ_spec.offset);
3089 return -EINVAL;
3090 }
3091
3092 cand_ids->data[j++] = cand_spec.spec[0].type_id;
3093 }
3094
3095 cand_ids->len = j;
3096 if (cand_ids->len == 0) {
3097 pr_warning("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
3098 prog_name, relo_idx, local_id, local_name, spec_str);
3099 return -ESRCH;
3100 }
3101
3102 err = bpf_core_reloc_insn(prog, relo->insn_off,
3103 local_spec.offset, targ_spec.offset);
3104 if (err) {
3105 pr_warning("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
3106 prog_name, relo_idx, relo->insn_off, err);
3107 return -EINVAL;
3108 }
3109
3110 return 0;
3111 }
3112
3113 static int
bpf_core_reloc_offsets(struct bpf_object * obj,const char * targ_btf_path)3114 bpf_core_reloc_offsets(struct bpf_object *obj, const char *targ_btf_path)
3115 {
3116 const struct btf_ext_info_sec *sec;
3117 const struct bpf_offset_reloc *rec;
3118 const struct btf_ext_info *seg;
3119 struct hashmap_entry *entry;
3120 struct hashmap *cand_cache = NULL;
3121 struct bpf_program *prog;
3122 struct btf *targ_btf;
3123 const char *sec_name;
3124 int i, err = 0;
3125
3126 if (targ_btf_path)
3127 targ_btf = btf__parse_elf(targ_btf_path, NULL);
3128 else
3129 targ_btf = bpf_core_find_kernel_btf();
3130 if (IS_ERR(targ_btf)) {
3131 pr_warning("failed to get target BTF: %ld\n",
3132 PTR_ERR(targ_btf));
3133 return PTR_ERR(targ_btf);
3134 }
3135
3136 cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
3137 if (IS_ERR(cand_cache)) {
3138 err = PTR_ERR(cand_cache);
3139 goto out;
3140 }
3141
3142 seg = &obj->btf_ext->offset_reloc_info;
3143 for_each_btf_ext_sec(seg, sec) {
3144 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
3145 if (str_is_empty(sec_name)) {
3146 err = -EINVAL;
3147 goto out;
3148 }
3149 prog = bpf_object__find_program_by_title(obj, sec_name);
3150 if (!prog) {
3151 pr_warning("failed to find program '%s' for CO-RE offset relocation\n",
3152 sec_name);
3153 err = -EINVAL;
3154 goto out;
3155 }
3156
3157 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
3158 sec_name, sec->num_info);
3159
3160 for_each_btf_ext_rec(seg, sec, i, rec) {
3161 err = bpf_core_reloc_offset(prog, rec, i, obj->btf,
3162 targ_btf, cand_cache);
3163 if (err) {
3164 pr_warning("prog '%s': relo #%d: failed to relocate: %d\n",
3165 sec_name, i, err);
3166 goto out;
3167 }
3168 }
3169 }
3170
3171 out:
3172 btf__free(targ_btf);
3173 if (!IS_ERR_OR_NULL(cand_cache)) {
3174 hashmap__for_each_entry(cand_cache, entry, i) {
3175 bpf_core_free_cands(entry->value);
3176 }
3177 hashmap__free(cand_cache);
3178 }
3179 return err;
3180 }
3181
3182 static int
bpf_object__relocate_core(struct bpf_object * obj,const char * targ_btf_path)3183 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
3184 {
3185 int err = 0;
3186
3187 if (obj->btf_ext->offset_reloc_info.len)
3188 err = bpf_core_reloc_offsets(obj, targ_btf_path);
3189
3190 return err;
3191 }
3192
3193 static int
bpf_program__reloc_text(struct bpf_program * prog,struct bpf_object * obj,struct reloc_desc * relo)3194 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
3195 struct reloc_desc *relo)
3196 {
3197 struct bpf_insn *insn, *new_insn;
3198 struct bpf_program *text;
3199 size_t new_cnt;
3200 int err;
3201
3202 if (relo->type != RELO_CALL)
3203 return -LIBBPF_ERRNO__RELOC;
3204
3205 if (prog->idx == obj->efile.text_shndx) {
3206 pr_warning("relo in .text insn %d into off %d\n",
3207 relo->insn_idx, relo->text_off);
3208 return -LIBBPF_ERRNO__RELOC;
3209 }
3210
3211 if (prog->main_prog_cnt == 0) {
3212 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
3213 if (!text) {
3214 pr_warning("no .text section found yet relo into text exist\n");
3215 return -LIBBPF_ERRNO__RELOC;
3216 }
3217 new_cnt = prog->insns_cnt + text->insns_cnt;
3218 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
3219 if (!new_insn) {
3220 pr_warning("oom in prog realloc\n");
3221 return -ENOMEM;
3222 }
3223 prog->insns = new_insn;
3224
3225 if (obj->btf_ext) {
3226 err = bpf_program_reloc_btf_ext(prog, obj,
3227 text->section_name,
3228 prog->insns_cnt);
3229 if (err)
3230 return err;
3231 }
3232
3233 memcpy(new_insn + prog->insns_cnt, text->insns,
3234 text->insns_cnt * sizeof(*insn));
3235 prog->main_prog_cnt = prog->insns_cnt;
3236 prog->insns_cnt = new_cnt;
3237 pr_debug("added %zd insn from %s to prog %s\n",
3238 text->insns_cnt, text->section_name,
3239 prog->section_name);
3240 }
3241 insn = &prog->insns[relo->insn_idx];
3242 insn->imm += prog->main_prog_cnt - relo->insn_idx;
3243 return 0;
3244 }
3245
3246 static int
bpf_program__relocate(struct bpf_program * prog,struct bpf_object * obj)3247 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
3248 {
3249 int i, err;
3250
3251 if (!prog)
3252 return 0;
3253
3254 if (obj->btf_ext) {
3255 err = bpf_program_reloc_btf_ext(prog, obj,
3256 prog->section_name, 0);
3257 if (err)
3258 return err;
3259 }
3260
3261 if (!prog->reloc_desc)
3262 return 0;
3263
3264 for (i = 0; i < prog->nr_reloc; i++) {
3265 if (prog->reloc_desc[i].type == RELO_LD64 ||
3266 prog->reloc_desc[i].type == RELO_DATA) {
3267 bool relo_data = prog->reloc_desc[i].type == RELO_DATA;
3268 struct bpf_insn *insns = prog->insns;
3269 int insn_idx, map_idx;
3270
3271 insn_idx = prog->reloc_desc[i].insn_idx;
3272 map_idx = prog->reloc_desc[i].map_idx;
3273
3274 if (insn_idx + 1 >= (int)prog->insns_cnt) {
3275 pr_warning("relocation out of range: '%s'\n",
3276 prog->section_name);
3277 return -LIBBPF_ERRNO__RELOC;
3278 }
3279
3280 if (!relo_data) {
3281 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
3282 } else {
3283 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_VALUE;
3284 insns[insn_idx + 1].imm = insns[insn_idx].imm;
3285 }
3286 insns[insn_idx].imm = obj->maps[map_idx].fd;
3287 } else if (prog->reloc_desc[i].type == RELO_CALL) {
3288 err = bpf_program__reloc_text(prog, obj,
3289 &prog->reloc_desc[i]);
3290 if (err)
3291 return err;
3292 }
3293 }
3294
3295 zfree(&prog->reloc_desc);
3296 prog->nr_reloc = 0;
3297 return 0;
3298 }
3299
3300 static int
bpf_object__relocate(struct bpf_object * obj,const char * targ_btf_path)3301 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
3302 {
3303 struct bpf_program *prog;
3304 size_t i;
3305 int err;
3306
3307 if (obj->btf_ext) {
3308 err = bpf_object__relocate_core(obj, targ_btf_path);
3309 if (err) {
3310 pr_warning("failed to perform CO-RE relocations: %d\n",
3311 err);
3312 return err;
3313 }
3314 }
3315 for (i = 0; i < obj->nr_programs; i++) {
3316 prog = &obj->programs[i];
3317
3318 err = bpf_program__relocate(prog, obj);
3319 if (err) {
3320 pr_warning("failed to relocate '%s'\n",
3321 prog->section_name);
3322 return err;
3323 }
3324 }
3325 return 0;
3326 }
3327
bpf_object__collect_reloc(struct bpf_object * obj)3328 static int bpf_object__collect_reloc(struct bpf_object *obj)
3329 {
3330 int i, err;
3331
3332 if (!obj_elf_valid(obj)) {
3333 pr_warning("Internal error: elf object is closed\n");
3334 return -LIBBPF_ERRNO__INTERNAL;
3335 }
3336
3337 for (i = 0; i < obj->efile.nr_reloc; i++) {
3338 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
3339 Elf_Data *data = obj->efile.reloc[i].data;
3340 int idx = shdr->sh_info;
3341 struct bpf_program *prog;
3342
3343 if (shdr->sh_type != SHT_REL) {
3344 pr_warning("internal error at %d\n", __LINE__);
3345 return -LIBBPF_ERRNO__INTERNAL;
3346 }
3347
3348 prog = bpf_object__find_prog_by_idx(obj, idx);
3349 if (!prog) {
3350 pr_warning("relocation failed: no section(%d)\n", idx);
3351 return -LIBBPF_ERRNO__RELOC;
3352 }
3353
3354 err = bpf_program__collect_reloc(prog, shdr, data, obj);
3355 if (err)
3356 return err;
3357 }
3358 return 0;
3359 }
3360
3361 static int
load_program(struct bpf_program * prog,struct bpf_insn * insns,int insns_cnt,char * license,__u32 kern_version,int * pfd)3362 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
3363 char *license, __u32 kern_version, int *pfd)
3364 {
3365 struct bpf_load_program_attr load_attr;
3366 char *cp, errmsg[STRERR_BUFSIZE];
3367 int log_buf_size = BPF_LOG_BUF_SIZE;
3368 char *log_buf;
3369 int btf_fd, ret;
3370
3371 if (!insns || !insns_cnt)
3372 return -EINVAL;
3373
3374 memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
3375 load_attr.prog_type = prog->type;
3376 load_attr.expected_attach_type = prog->expected_attach_type;
3377 if (prog->caps->name)
3378 load_attr.name = prog->name;
3379 load_attr.insns = insns;
3380 load_attr.insns_cnt = insns_cnt;
3381 load_attr.license = license;
3382 load_attr.kern_version = kern_version;
3383 load_attr.prog_ifindex = prog->prog_ifindex;
3384 /* if .BTF.ext was loaded, kernel supports associated BTF for prog */
3385 if (prog->obj->btf_ext)
3386 btf_fd = bpf_object__btf_fd(prog->obj);
3387 else
3388 btf_fd = -1;
3389 load_attr.prog_btf_fd = btf_fd >= 0 ? btf_fd : 0;
3390 load_attr.func_info = prog->func_info;
3391 load_attr.func_info_rec_size = prog->func_info_rec_size;
3392 load_attr.func_info_cnt = prog->func_info_cnt;
3393 load_attr.line_info = prog->line_info;
3394 load_attr.line_info_rec_size = prog->line_info_rec_size;
3395 load_attr.line_info_cnt = prog->line_info_cnt;
3396 load_attr.log_level = prog->log_level;
3397 load_attr.prog_flags = prog->prog_flags;
3398
3399 retry_load:
3400 log_buf = malloc(log_buf_size);
3401 if (!log_buf)
3402 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
3403
3404 ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
3405
3406 if (ret >= 0) {
3407 if (load_attr.log_level)
3408 pr_debug("verifier log:\n%s", log_buf);
3409 *pfd = ret;
3410 ret = 0;
3411 goto out;
3412 }
3413
3414 if (errno == ENOSPC) {
3415 log_buf_size <<= 1;
3416 free(log_buf);
3417 goto retry_load;
3418 }
3419 ret = -LIBBPF_ERRNO__LOAD;
3420 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3421 pr_warning("load bpf program failed: %s\n", cp);
3422
3423 if (log_buf && log_buf[0] != '\0') {
3424 ret = -LIBBPF_ERRNO__VERIFY;
3425 pr_warning("-- BEGIN DUMP LOG ---\n");
3426 pr_warning("\n%s\n", log_buf);
3427 pr_warning("-- END LOG --\n");
3428 } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
3429 pr_warning("Program too large (%zu insns), at most %d insns\n",
3430 load_attr.insns_cnt, BPF_MAXINSNS);
3431 ret = -LIBBPF_ERRNO__PROG2BIG;
3432 } else {
3433 /* Wrong program type? */
3434 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
3435 int fd;
3436
3437 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
3438 load_attr.expected_attach_type = 0;
3439 fd = bpf_load_program_xattr(&load_attr, NULL, 0);
3440 if (fd >= 0) {
3441 close(fd);
3442 ret = -LIBBPF_ERRNO__PROGTYPE;
3443 goto out;
3444 }
3445 }
3446
3447 if (log_buf)
3448 ret = -LIBBPF_ERRNO__KVER;
3449 }
3450
3451 out:
3452 free(log_buf);
3453 return ret;
3454 }
3455
3456 int
bpf_program__load(struct bpf_program * prog,char * license,__u32 kern_version)3457 bpf_program__load(struct bpf_program *prog,
3458 char *license, __u32 kern_version)
3459 {
3460 int err = 0, fd, i;
3461
3462 if (prog->instances.nr < 0 || !prog->instances.fds) {
3463 if (prog->preprocessor) {
3464 pr_warning("Internal error: can't load program '%s'\n",
3465 prog->section_name);
3466 return -LIBBPF_ERRNO__INTERNAL;
3467 }
3468
3469 prog->instances.fds = malloc(sizeof(int));
3470 if (!prog->instances.fds) {
3471 pr_warning("Not enough memory for BPF fds\n");
3472 return -ENOMEM;
3473 }
3474 prog->instances.nr = 1;
3475 prog->instances.fds[0] = -1;
3476 }
3477
3478 if (!prog->preprocessor) {
3479 if (prog->instances.nr != 1) {
3480 pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
3481 prog->section_name, prog->instances.nr);
3482 }
3483 err = load_program(prog, prog->insns, prog->insns_cnt,
3484 license, kern_version, &fd);
3485 if (!err)
3486 prog->instances.fds[0] = fd;
3487 goto out;
3488 }
3489
3490 for (i = 0; i < prog->instances.nr; i++) {
3491 struct bpf_prog_prep_result result;
3492 bpf_program_prep_t preprocessor = prog->preprocessor;
3493
3494 memset(&result, 0, sizeof(result));
3495 err = preprocessor(prog, i, prog->insns,
3496 prog->insns_cnt, &result);
3497 if (err) {
3498 pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
3499 i, prog->section_name);
3500 goto out;
3501 }
3502
3503 if (!result.new_insn_ptr || !result.new_insn_cnt) {
3504 pr_debug("Skip loading the %dth instance of program '%s'\n",
3505 i, prog->section_name);
3506 prog->instances.fds[i] = -1;
3507 if (result.pfd)
3508 *result.pfd = -1;
3509 continue;
3510 }
3511
3512 err = load_program(prog, result.new_insn_ptr,
3513 result.new_insn_cnt,
3514 license, kern_version, &fd);
3515
3516 if (err) {
3517 pr_warning("Loading the %dth instance of program '%s' failed\n",
3518 i, prog->section_name);
3519 goto out;
3520 }
3521
3522 if (result.pfd)
3523 *result.pfd = fd;
3524 prog->instances.fds[i] = fd;
3525 }
3526 out:
3527 if (err)
3528 pr_warning("failed to load program '%s'\n",
3529 prog->section_name);
3530 zfree(&prog->insns);
3531 prog->insns_cnt = 0;
3532 return err;
3533 }
3534
bpf_program__is_function_storage(const struct bpf_program * prog,const struct bpf_object * obj)3535 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
3536 const struct bpf_object *obj)
3537 {
3538 return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
3539 }
3540
3541 static int
bpf_object__load_progs(struct bpf_object * obj,int log_level)3542 bpf_object__load_progs(struct bpf_object *obj, int log_level)
3543 {
3544 size_t i;
3545 int err;
3546
3547 for (i = 0; i < obj->nr_programs; i++) {
3548 if (bpf_program__is_function_storage(&obj->programs[i], obj))
3549 continue;
3550 obj->programs[i].log_level |= log_level;
3551 err = bpf_program__load(&obj->programs[i],
3552 obj->license,
3553 obj->kern_version);
3554 if (err)
3555 return err;
3556 }
3557 return 0;
3558 }
3559
bpf_prog_type__needs_kver(enum bpf_prog_type type)3560 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
3561 {
3562 switch (type) {
3563 case BPF_PROG_TYPE_SOCKET_FILTER:
3564 case BPF_PROG_TYPE_SCHED_CLS:
3565 case BPF_PROG_TYPE_SCHED_ACT:
3566 case BPF_PROG_TYPE_XDP:
3567 case BPF_PROG_TYPE_CGROUP_SKB:
3568 case BPF_PROG_TYPE_CGROUP_SOCK:
3569 case BPF_PROG_TYPE_LWT_IN:
3570 case BPF_PROG_TYPE_LWT_OUT:
3571 case BPF_PROG_TYPE_LWT_XMIT:
3572 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
3573 case BPF_PROG_TYPE_SOCK_OPS:
3574 case BPF_PROG_TYPE_SK_SKB:
3575 case BPF_PROG_TYPE_CGROUP_DEVICE:
3576 case BPF_PROG_TYPE_SK_MSG:
3577 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3578 case BPF_PROG_TYPE_LIRC_MODE2:
3579 case BPF_PROG_TYPE_SK_REUSEPORT:
3580 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3581 case BPF_PROG_TYPE_UNSPEC:
3582 case BPF_PROG_TYPE_TRACEPOINT:
3583 case BPF_PROG_TYPE_RAW_TRACEPOINT:
3584 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3585 case BPF_PROG_TYPE_PERF_EVENT:
3586 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3587 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3588 return false;
3589 case BPF_PROG_TYPE_KPROBE:
3590 default:
3591 return true;
3592 }
3593 }
3594
bpf_object__validate(struct bpf_object * obj,bool needs_kver)3595 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
3596 {
3597 if (needs_kver && obj->kern_version == 0) {
3598 pr_warning("%s doesn't provide kernel version\n",
3599 obj->path);
3600 return -LIBBPF_ERRNO__KVERSION;
3601 }
3602 return 0;
3603 }
3604
3605 static struct bpf_object *
__bpf_object__open(const char * path,void * obj_buf,size_t obj_buf_sz,bool needs_kver,int flags)3606 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
3607 bool needs_kver, int flags)
3608 {
3609 struct bpf_object *obj;
3610 int err;
3611
3612 if (elf_version(EV_CURRENT) == EV_NONE) {
3613 pr_warning("failed to init libelf for %s\n", path);
3614 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
3615 }
3616
3617 obj = bpf_object__new(path, obj_buf, obj_buf_sz);
3618 if (IS_ERR(obj))
3619 return obj;
3620
3621 CHECK_ERR(bpf_object__elf_init(obj), err, out);
3622 CHECK_ERR(bpf_object__check_endianness(obj), err, out);
3623 CHECK_ERR(bpf_object__probe_caps(obj), err, out);
3624 CHECK_ERR(bpf_object__elf_collect(obj, flags), err, out);
3625 CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
3626 CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
3627
3628 bpf_object__elf_finish(obj);
3629 return obj;
3630 out:
3631 bpf_object__close(obj);
3632 return ERR_PTR(err);
3633 }
3634
__bpf_object__open_xattr(struct bpf_object_open_attr * attr,int flags)3635 struct bpf_object *__bpf_object__open_xattr(struct bpf_object_open_attr *attr,
3636 int flags)
3637 {
3638 /* param validation */
3639 if (!attr->file)
3640 return NULL;
3641
3642 pr_debug("loading %s\n", attr->file);
3643
3644 return __bpf_object__open(attr->file, NULL, 0,
3645 bpf_prog_type__needs_kver(attr->prog_type),
3646 flags);
3647 }
3648
bpf_object__open_xattr(struct bpf_object_open_attr * attr)3649 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
3650 {
3651 return __bpf_object__open_xattr(attr, 0);
3652 }
3653
bpf_object__open(const char * path)3654 struct bpf_object *bpf_object__open(const char *path)
3655 {
3656 struct bpf_object_open_attr attr = {
3657 .file = path,
3658 .prog_type = BPF_PROG_TYPE_UNSPEC,
3659 };
3660
3661 return bpf_object__open_xattr(&attr);
3662 }
3663
bpf_object__open_buffer(void * obj_buf,size_t obj_buf_sz,const char * name)3664 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
3665 size_t obj_buf_sz,
3666 const char *name)
3667 {
3668 char tmp_name[64];
3669
3670 /* param validation */
3671 if (!obj_buf || obj_buf_sz <= 0)
3672 return NULL;
3673
3674 if (!name) {
3675 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
3676 (unsigned long)obj_buf,
3677 (unsigned long)obj_buf_sz);
3678 name = tmp_name;
3679 }
3680 pr_debug("loading object '%s' from buffer\n", name);
3681
3682 return __bpf_object__open(name, obj_buf, obj_buf_sz, true, true);
3683 }
3684
bpf_object__unload(struct bpf_object * obj)3685 int bpf_object__unload(struct bpf_object *obj)
3686 {
3687 size_t i;
3688
3689 if (!obj)
3690 return -EINVAL;
3691
3692 for (i = 0; i < obj->nr_maps; i++)
3693 zclose(obj->maps[i].fd);
3694
3695 for (i = 0; i < obj->nr_programs; i++)
3696 bpf_program__unload(&obj->programs[i]);
3697
3698 return 0;
3699 }
3700
bpf_object__load_xattr(struct bpf_object_load_attr * attr)3701 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
3702 {
3703 struct bpf_object *obj;
3704 int err;
3705
3706 if (!attr)
3707 return -EINVAL;
3708 obj = attr->obj;
3709 if (!obj)
3710 return -EINVAL;
3711
3712 if (obj->loaded) {
3713 pr_warning("object should not be loaded twice\n");
3714 return -EINVAL;
3715 }
3716
3717 obj->loaded = true;
3718
3719 CHECK_ERR(bpf_object__create_maps(obj), err, out);
3720 CHECK_ERR(bpf_object__relocate(obj, attr->target_btf_path), err, out);
3721 CHECK_ERR(bpf_object__load_progs(obj, attr->log_level), err, out);
3722
3723 return 0;
3724 out:
3725 bpf_object__unload(obj);
3726 pr_warning("failed to load object '%s'\n", obj->path);
3727 return err;
3728 }
3729
bpf_object__load(struct bpf_object * obj)3730 int bpf_object__load(struct bpf_object *obj)
3731 {
3732 struct bpf_object_load_attr attr = {
3733 .obj = obj,
3734 };
3735
3736 return bpf_object__load_xattr(&attr);
3737 }
3738
check_path(const char * path)3739 static int check_path(const char *path)
3740 {
3741 char *cp, errmsg[STRERR_BUFSIZE];
3742 struct statfs st_fs;
3743 char *dname, *dir;
3744 int err = 0;
3745
3746 if (path == NULL)
3747 return -EINVAL;
3748
3749 dname = strdup(path);
3750 if (dname == NULL)
3751 return -ENOMEM;
3752
3753 dir = dirname(dname);
3754 if (statfs(dir, &st_fs)) {
3755 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3756 pr_warning("failed to statfs %s: %s\n", dir, cp);
3757 err = -errno;
3758 }
3759 free(dname);
3760
3761 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
3762 pr_warning("specified path %s is not on BPF FS\n", path);
3763 err = -EINVAL;
3764 }
3765
3766 return err;
3767 }
3768
bpf_program__pin_instance(struct bpf_program * prog,const char * path,int instance)3769 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
3770 int instance)
3771 {
3772 char *cp, errmsg[STRERR_BUFSIZE];
3773 int err;
3774
3775 err = check_path(path);
3776 if (err)
3777 return err;
3778
3779 if (prog == NULL) {
3780 pr_warning("invalid program pointer\n");
3781 return -EINVAL;
3782 }
3783
3784 if (instance < 0 || instance >= prog->instances.nr) {
3785 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
3786 instance, prog->section_name, prog->instances.nr);
3787 return -EINVAL;
3788 }
3789
3790 if (bpf_obj_pin(prog->instances.fds[instance], path)) {
3791 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3792 pr_warning("failed to pin program: %s\n", cp);
3793 return -errno;
3794 }
3795 pr_debug("pinned program '%s'\n", path);
3796
3797 return 0;
3798 }
3799
bpf_program__unpin_instance(struct bpf_program * prog,const char * path,int instance)3800 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
3801 int instance)
3802 {
3803 int err;
3804
3805 err = check_path(path);
3806 if (err)
3807 return err;
3808
3809 if (prog == NULL) {
3810 pr_warning("invalid program pointer\n");
3811 return -EINVAL;
3812 }
3813
3814 if (instance < 0 || instance >= prog->instances.nr) {
3815 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
3816 instance, prog->section_name, prog->instances.nr);
3817 return -EINVAL;
3818 }
3819
3820 err = unlink(path);
3821 if (err != 0)
3822 return -errno;
3823 pr_debug("unpinned program '%s'\n", path);
3824
3825 return 0;
3826 }
3827
make_dir(const char * path)3828 static int make_dir(const char *path)
3829 {
3830 char *cp, errmsg[STRERR_BUFSIZE];
3831 int err = 0;
3832
3833 if (mkdir(path, 0700) && errno != EEXIST)
3834 err = -errno;
3835
3836 if (err) {
3837 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
3838 pr_warning("failed to mkdir %s: %s\n", path, cp);
3839 }
3840 return err;
3841 }
3842
bpf_program__pin(struct bpf_program * prog,const char * path)3843 int bpf_program__pin(struct bpf_program *prog, const char *path)
3844 {
3845 int i, err;
3846
3847 err = check_path(path);
3848 if (err)
3849 return err;
3850
3851 if (prog == NULL) {
3852 pr_warning("invalid program pointer\n");
3853 return -EINVAL;
3854 }
3855
3856 if (prog->instances.nr <= 0) {
3857 pr_warning("no instances of prog %s to pin\n",
3858 prog->section_name);
3859 return -EINVAL;
3860 }
3861
3862 if (prog->instances.nr == 1) {
3863 /* don't create subdirs when pinning single instance */
3864 return bpf_program__pin_instance(prog, path, 0);
3865 }
3866
3867 err = make_dir(path);
3868 if (err)
3869 return err;
3870
3871 for (i = 0; i < prog->instances.nr; i++) {
3872 char buf[PATH_MAX];
3873 int len;
3874
3875 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3876 if (len < 0) {
3877 err = -EINVAL;
3878 goto err_unpin;
3879 } else if (len >= PATH_MAX) {
3880 err = -ENAMETOOLONG;
3881 goto err_unpin;
3882 }
3883
3884 err = bpf_program__pin_instance(prog, buf, i);
3885 if (err)
3886 goto err_unpin;
3887 }
3888
3889 return 0;
3890
3891 err_unpin:
3892 for (i = i - 1; i >= 0; i--) {
3893 char buf[PATH_MAX];
3894 int len;
3895
3896 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3897 if (len < 0)
3898 continue;
3899 else if (len >= PATH_MAX)
3900 continue;
3901
3902 bpf_program__unpin_instance(prog, buf, i);
3903 }
3904
3905 rmdir(path);
3906
3907 return err;
3908 }
3909
bpf_program__unpin(struct bpf_program * prog,const char * path)3910 int bpf_program__unpin(struct bpf_program *prog, const char *path)
3911 {
3912 int i, err;
3913
3914 err = check_path(path);
3915 if (err)
3916 return err;
3917
3918 if (prog == NULL) {
3919 pr_warning("invalid program pointer\n");
3920 return -EINVAL;
3921 }
3922
3923 if (prog->instances.nr <= 0) {
3924 pr_warning("no instances of prog %s to pin\n",
3925 prog->section_name);
3926 return -EINVAL;
3927 }
3928
3929 if (prog->instances.nr == 1) {
3930 /* don't create subdirs when pinning single instance */
3931 return bpf_program__unpin_instance(prog, path, 0);
3932 }
3933
3934 for (i = 0; i < prog->instances.nr; i++) {
3935 char buf[PATH_MAX];
3936 int len;
3937
3938 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3939 if (len < 0)
3940 return -EINVAL;
3941 else if (len >= PATH_MAX)
3942 return -ENAMETOOLONG;
3943
3944 err = bpf_program__unpin_instance(prog, buf, i);
3945 if (err)
3946 return err;
3947 }
3948
3949 err = rmdir(path);
3950 if (err)
3951 return -errno;
3952
3953 return 0;
3954 }
3955
bpf_map__pin(struct bpf_map * map,const char * path)3956 int bpf_map__pin(struct bpf_map *map, const char *path)
3957 {
3958 char *cp, errmsg[STRERR_BUFSIZE];
3959 int err;
3960
3961 err = check_path(path);
3962 if (err)
3963 return err;
3964
3965 if (map == NULL) {
3966 pr_warning("invalid map pointer\n");
3967 return -EINVAL;
3968 }
3969
3970 if (bpf_obj_pin(map->fd, path)) {
3971 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3972 pr_warning("failed to pin map: %s\n", cp);
3973 return -errno;
3974 }
3975
3976 pr_debug("pinned map '%s'\n", path);
3977
3978 return 0;
3979 }
3980
bpf_map__unpin(struct bpf_map * map,const char * path)3981 int bpf_map__unpin(struct bpf_map *map, const char *path)
3982 {
3983 int err;
3984
3985 err = check_path(path);
3986 if (err)
3987 return err;
3988
3989 if (map == NULL) {
3990 pr_warning("invalid map pointer\n");
3991 return -EINVAL;
3992 }
3993
3994 err = unlink(path);
3995 if (err != 0)
3996 return -errno;
3997 pr_debug("unpinned map '%s'\n", path);
3998
3999 return 0;
4000 }
4001
bpf_object__pin_maps(struct bpf_object * obj,const char * path)4002 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
4003 {
4004 struct bpf_map *map;
4005 int err;
4006
4007 if (!obj)
4008 return -ENOENT;
4009
4010 if (!obj->loaded) {
4011 pr_warning("object not yet loaded; load it first\n");
4012 return -ENOENT;
4013 }
4014
4015 err = make_dir(path);
4016 if (err)
4017 return err;
4018
4019 bpf_object__for_each_map(map, obj) {
4020 char buf[PATH_MAX];
4021 int len;
4022
4023 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4024 bpf_map__name(map));
4025 if (len < 0) {
4026 err = -EINVAL;
4027 goto err_unpin_maps;
4028 } else if (len >= PATH_MAX) {
4029 err = -ENAMETOOLONG;
4030 goto err_unpin_maps;
4031 }
4032
4033 err = bpf_map__pin(map, buf);
4034 if (err)
4035 goto err_unpin_maps;
4036 }
4037
4038 return 0;
4039
4040 err_unpin_maps:
4041 while ((map = bpf_map__prev(map, obj))) {
4042 char buf[PATH_MAX];
4043 int len;
4044
4045 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4046 bpf_map__name(map));
4047 if (len < 0)
4048 continue;
4049 else if (len >= PATH_MAX)
4050 continue;
4051
4052 bpf_map__unpin(map, buf);
4053 }
4054
4055 return err;
4056 }
4057
bpf_object__unpin_maps(struct bpf_object * obj,const char * path)4058 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
4059 {
4060 struct bpf_map *map;
4061 int err;
4062
4063 if (!obj)
4064 return -ENOENT;
4065
4066 bpf_object__for_each_map(map, obj) {
4067 char buf[PATH_MAX];
4068 int len;
4069
4070 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4071 bpf_map__name(map));
4072 if (len < 0)
4073 return -EINVAL;
4074 else if (len >= PATH_MAX)
4075 return -ENAMETOOLONG;
4076
4077 err = bpf_map__unpin(map, buf);
4078 if (err)
4079 return err;
4080 }
4081
4082 return 0;
4083 }
4084
bpf_object__pin_programs(struct bpf_object * obj,const char * path)4085 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
4086 {
4087 struct bpf_program *prog;
4088 int err;
4089
4090 if (!obj)
4091 return -ENOENT;
4092
4093 if (!obj->loaded) {
4094 pr_warning("object not yet loaded; load it first\n");
4095 return -ENOENT;
4096 }
4097
4098 err = make_dir(path);
4099 if (err)
4100 return err;
4101
4102 bpf_object__for_each_program(prog, obj) {
4103 char buf[PATH_MAX];
4104 int len;
4105
4106 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4107 prog->pin_name);
4108 if (len < 0) {
4109 err = -EINVAL;
4110 goto err_unpin_programs;
4111 } else if (len >= PATH_MAX) {
4112 err = -ENAMETOOLONG;
4113 goto err_unpin_programs;
4114 }
4115
4116 err = bpf_program__pin(prog, buf);
4117 if (err)
4118 goto err_unpin_programs;
4119 }
4120
4121 return 0;
4122
4123 err_unpin_programs:
4124 while ((prog = bpf_program__prev(prog, obj))) {
4125 char buf[PATH_MAX];
4126 int len;
4127
4128 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4129 prog->pin_name);
4130 if (len < 0)
4131 continue;
4132 else if (len >= PATH_MAX)
4133 continue;
4134
4135 bpf_program__unpin(prog, buf);
4136 }
4137
4138 return err;
4139 }
4140
bpf_object__unpin_programs(struct bpf_object * obj,const char * path)4141 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
4142 {
4143 struct bpf_program *prog;
4144 int err;
4145
4146 if (!obj)
4147 return -ENOENT;
4148
4149 bpf_object__for_each_program(prog, obj) {
4150 char buf[PATH_MAX];
4151 int len;
4152
4153 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4154 prog->pin_name);
4155 if (len < 0)
4156 return -EINVAL;
4157 else if (len >= PATH_MAX)
4158 return -ENAMETOOLONG;
4159
4160 err = bpf_program__unpin(prog, buf);
4161 if (err)
4162 return err;
4163 }
4164
4165 return 0;
4166 }
4167
bpf_object__pin(struct bpf_object * obj,const char * path)4168 int bpf_object__pin(struct bpf_object *obj, const char *path)
4169 {
4170 int err;
4171
4172 err = bpf_object__pin_maps(obj, path);
4173 if (err)
4174 return err;
4175
4176 err = bpf_object__pin_programs(obj, path);
4177 if (err) {
4178 bpf_object__unpin_maps(obj, path);
4179 return err;
4180 }
4181
4182 return 0;
4183 }
4184
bpf_object__close(struct bpf_object * obj)4185 void bpf_object__close(struct bpf_object *obj)
4186 {
4187 size_t i;
4188
4189 if (!obj)
4190 return;
4191
4192 if (obj->clear_priv)
4193 obj->clear_priv(obj, obj->priv);
4194
4195 bpf_object__elf_finish(obj);
4196 bpf_object__unload(obj);
4197 btf__free(obj->btf);
4198 btf_ext__free(obj->btf_ext);
4199
4200 for (i = 0; i < obj->nr_maps; i++) {
4201 zfree(&obj->maps[i].name);
4202 if (obj->maps[i].clear_priv)
4203 obj->maps[i].clear_priv(&obj->maps[i],
4204 obj->maps[i].priv);
4205 obj->maps[i].priv = NULL;
4206 obj->maps[i].clear_priv = NULL;
4207 }
4208
4209 zfree(&obj->sections.rodata);
4210 zfree(&obj->sections.data);
4211 zfree(&obj->maps);
4212 obj->nr_maps = 0;
4213
4214 if (obj->programs && obj->nr_programs) {
4215 for (i = 0; i < obj->nr_programs; i++)
4216 bpf_program__exit(&obj->programs[i]);
4217 }
4218 zfree(&obj->programs);
4219
4220 list_del(&obj->list);
4221 free(obj);
4222 }
4223
4224 struct bpf_object *
bpf_object__next(struct bpf_object * prev)4225 bpf_object__next(struct bpf_object *prev)
4226 {
4227 struct bpf_object *next;
4228
4229 if (!prev)
4230 next = list_first_entry(&bpf_objects_list,
4231 struct bpf_object,
4232 list);
4233 else
4234 next = list_next_entry(prev, list);
4235
4236 /* Empty list is noticed here so don't need checking on entry. */
4237 if (&next->list == &bpf_objects_list)
4238 return NULL;
4239
4240 return next;
4241 }
4242
bpf_object__name(const struct bpf_object * obj)4243 const char *bpf_object__name(const struct bpf_object *obj)
4244 {
4245 return obj ? obj->path : ERR_PTR(-EINVAL);
4246 }
4247
bpf_object__kversion(const struct bpf_object * obj)4248 unsigned int bpf_object__kversion(const struct bpf_object *obj)
4249 {
4250 return obj ? obj->kern_version : 0;
4251 }
4252
bpf_object__btf(const struct bpf_object * obj)4253 struct btf *bpf_object__btf(const struct bpf_object *obj)
4254 {
4255 return obj ? obj->btf : NULL;
4256 }
4257
bpf_object__btf_fd(const struct bpf_object * obj)4258 int bpf_object__btf_fd(const struct bpf_object *obj)
4259 {
4260 return obj->btf ? btf__fd(obj->btf) : -1;
4261 }
4262
bpf_object__set_priv(struct bpf_object * obj,void * priv,bpf_object_clear_priv_t clear_priv)4263 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
4264 bpf_object_clear_priv_t clear_priv)
4265 {
4266 if (obj->priv && obj->clear_priv)
4267 obj->clear_priv(obj, obj->priv);
4268
4269 obj->priv = priv;
4270 obj->clear_priv = clear_priv;
4271 return 0;
4272 }
4273
bpf_object__priv(const struct bpf_object * obj)4274 void *bpf_object__priv(const struct bpf_object *obj)
4275 {
4276 return obj ? obj->priv : ERR_PTR(-EINVAL);
4277 }
4278
4279 static struct bpf_program *
__bpf_program__iter(const struct bpf_program * p,const struct bpf_object * obj,bool forward)4280 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
4281 bool forward)
4282 {
4283 size_t nr_programs = obj->nr_programs;
4284 ssize_t idx;
4285
4286 if (!nr_programs)
4287 return NULL;
4288
4289 if (!p)
4290 /* Iter from the beginning */
4291 return forward ? &obj->programs[0] :
4292 &obj->programs[nr_programs - 1];
4293
4294 if (p->obj != obj) {
4295 pr_warning("error: program handler doesn't match object\n");
4296 return NULL;
4297 }
4298
4299 idx = (p - obj->programs) + (forward ? 1 : -1);
4300 if (idx >= obj->nr_programs || idx < 0)
4301 return NULL;
4302 return &obj->programs[idx];
4303 }
4304
4305 struct bpf_program *
bpf_program__next(struct bpf_program * prev,const struct bpf_object * obj)4306 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
4307 {
4308 struct bpf_program *prog = prev;
4309
4310 do {
4311 prog = __bpf_program__iter(prog, obj, true);
4312 } while (prog && bpf_program__is_function_storage(prog, obj));
4313
4314 return prog;
4315 }
4316
4317 struct bpf_program *
bpf_program__prev(struct bpf_program * next,const struct bpf_object * obj)4318 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
4319 {
4320 struct bpf_program *prog = next;
4321
4322 do {
4323 prog = __bpf_program__iter(prog, obj, false);
4324 } while (prog && bpf_program__is_function_storage(prog, obj));
4325
4326 return prog;
4327 }
4328
bpf_program__set_priv(struct bpf_program * prog,void * priv,bpf_program_clear_priv_t clear_priv)4329 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
4330 bpf_program_clear_priv_t clear_priv)
4331 {
4332 if (prog->priv && prog->clear_priv)
4333 prog->clear_priv(prog, prog->priv);
4334
4335 prog->priv = priv;
4336 prog->clear_priv = clear_priv;
4337 return 0;
4338 }
4339
bpf_program__priv(const struct bpf_program * prog)4340 void *bpf_program__priv(const struct bpf_program *prog)
4341 {
4342 return prog ? prog->priv : ERR_PTR(-EINVAL);
4343 }
4344
bpf_program__set_ifindex(struct bpf_program * prog,__u32 ifindex)4345 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
4346 {
4347 prog->prog_ifindex = ifindex;
4348 }
4349
bpf_program__title(const struct bpf_program * prog,bool needs_copy)4350 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
4351 {
4352 const char *title;
4353
4354 title = prog->section_name;
4355 if (needs_copy) {
4356 title = strdup(title);
4357 if (!title) {
4358 pr_warning("failed to strdup program title\n");
4359 return ERR_PTR(-ENOMEM);
4360 }
4361 }
4362
4363 return title;
4364 }
4365
bpf_program__fd(const struct bpf_program * prog)4366 int bpf_program__fd(const struct bpf_program *prog)
4367 {
4368 return bpf_program__nth_fd(prog, 0);
4369 }
4370
bpf_program__set_prep(struct bpf_program * prog,int nr_instances,bpf_program_prep_t prep)4371 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
4372 bpf_program_prep_t prep)
4373 {
4374 int *instances_fds;
4375
4376 if (nr_instances <= 0 || !prep)
4377 return -EINVAL;
4378
4379 if (prog->instances.nr > 0 || prog->instances.fds) {
4380 pr_warning("Can't set pre-processor after loading\n");
4381 return -EINVAL;
4382 }
4383
4384 instances_fds = malloc(sizeof(int) * nr_instances);
4385 if (!instances_fds) {
4386 pr_warning("alloc memory failed for fds\n");
4387 return -ENOMEM;
4388 }
4389
4390 /* fill all fd with -1 */
4391 memset(instances_fds, -1, sizeof(int) * nr_instances);
4392
4393 prog->instances.nr = nr_instances;
4394 prog->instances.fds = instances_fds;
4395 prog->preprocessor = prep;
4396 return 0;
4397 }
4398
bpf_program__nth_fd(const struct bpf_program * prog,int n)4399 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
4400 {
4401 int fd;
4402
4403 if (!prog)
4404 return -EINVAL;
4405
4406 if (n >= prog->instances.nr || n < 0) {
4407 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
4408 n, prog->section_name, prog->instances.nr);
4409 return -EINVAL;
4410 }
4411
4412 fd = prog->instances.fds[n];
4413 if (fd < 0) {
4414 pr_warning("%dth instance of program '%s' is invalid\n",
4415 n, prog->section_name);
4416 return -ENOENT;
4417 }
4418
4419 return fd;
4420 }
4421
bpf_program__set_type(struct bpf_program * prog,enum bpf_prog_type type)4422 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
4423 {
4424 prog->type = type;
4425 }
4426
bpf_program__is_type(const struct bpf_program * prog,enum bpf_prog_type type)4427 static bool bpf_program__is_type(const struct bpf_program *prog,
4428 enum bpf_prog_type type)
4429 {
4430 return prog ? (prog->type == type) : false;
4431 }
4432
4433 #define BPF_PROG_TYPE_FNS(NAME, TYPE) \
4434 int bpf_program__set_##NAME(struct bpf_program *prog) \
4435 { \
4436 if (!prog) \
4437 return -EINVAL; \
4438 bpf_program__set_type(prog, TYPE); \
4439 return 0; \
4440 } \
4441 \
4442 bool bpf_program__is_##NAME(const struct bpf_program *prog) \
4443 { \
4444 return bpf_program__is_type(prog, TYPE); \
4445 } \
4446
4447 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
4448 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
4449 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
4450 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
4451 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
4452 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
4453 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
4454 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
4455
bpf_program__set_expected_attach_type(struct bpf_program * prog,enum bpf_attach_type type)4456 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
4457 enum bpf_attach_type type)
4458 {
4459 prog->expected_attach_type = type;
4460 }
4461
4462 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, atype) \
4463 { string, sizeof(string) - 1, ptype, eatype, is_attachable, atype }
4464
4465 /* Programs that can NOT be attached. */
4466 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0)
4467
4468 /* Programs that can be attached. */
4469 #define BPF_APROG_SEC(string, ptype, atype) \
4470 BPF_PROG_SEC_IMPL(string, ptype, 0, 1, atype)
4471
4472 /* Programs that must specify expected attach type at load time. */
4473 #define BPF_EAPROG_SEC(string, ptype, eatype) \
4474 BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, eatype)
4475
4476 /* Programs that can be attached but attach type can't be identified by section
4477 * name. Kept for backward compatibility.
4478 */
4479 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
4480
4481 static const struct {
4482 const char *sec;
4483 size_t len;
4484 enum bpf_prog_type prog_type;
4485 enum bpf_attach_type expected_attach_type;
4486 int is_attachable;
4487 enum bpf_attach_type attach_type;
4488 } section_names[] = {
4489 BPF_PROG_SEC("socket", BPF_PROG_TYPE_SOCKET_FILTER),
4490 BPF_PROG_SEC("kprobe/", BPF_PROG_TYPE_KPROBE),
4491 BPF_PROG_SEC("kretprobe/", BPF_PROG_TYPE_KPROBE),
4492 BPF_PROG_SEC("classifier", BPF_PROG_TYPE_SCHED_CLS),
4493 BPF_PROG_SEC("action", BPF_PROG_TYPE_SCHED_ACT),
4494 BPF_PROG_SEC("tracepoint/", BPF_PROG_TYPE_TRACEPOINT),
4495 BPF_PROG_SEC("raw_tracepoint/", BPF_PROG_TYPE_RAW_TRACEPOINT),
4496 BPF_PROG_SEC("xdp", BPF_PROG_TYPE_XDP),
4497 BPF_PROG_SEC("perf_event", BPF_PROG_TYPE_PERF_EVENT),
4498 BPF_PROG_SEC("lwt_in", BPF_PROG_TYPE_LWT_IN),
4499 BPF_PROG_SEC("lwt_out", BPF_PROG_TYPE_LWT_OUT),
4500 BPF_PROG_SEC("lwt_xmit", BPF_PROG_TYPE_LWT_XMIT),
4501 BPF_PROG_SEC("lwt_seg6local", BPF_PROG_TYPE_LWT_SEG6LOCAL),
4502 BPF_APROG_SEC("cgroup_skb/ingress", BPF_PROG_TYPE_CGROUP_SKB,
4503 BPF_CGROUP_INET_INGRESS),
4504 BPF_APROG_SEC("cgroup_skb/egress", BPF_PROG_TYPE_CGROUP_SKB,
4505 BPF_CGROUP_INET_EGRESS),
4506 BPF_APROG_COMPAT("cgroup/skb", BPF_PROG_TYPE_CGROUP_SKB),
4507 BPF_APROG_SEC("cgroup/sock", BPF_PROG_TYPE_CGROUP_SOCK,
4508 BPF_CGROUP_INET_SOCK_CREATE),
4509 BPF_EAPROG_SEC("cgroup/post_bind4", BPF_PROG_TYPE_CGROUP_SOCK,
4510 BPF_CGROUP_INET4_POST_BIND),
4511 BPF_EAPROG_SEC("cgroup/post_bind6", BPF_PROG_TYPE_CGROUP_SOCK,
4512 BPF_CGROUP_INET6_POST_BIND),
4513 BPF_APROG_SEC("cgroup/dev", BPF_PROG_TYPE_CGROUP_DEVICE,
4514 BPF_CGROUP_DEVICE),
4515 BPF_APROG_SEC("sockops", BPF_PROG_TYPE_SOCK_OPS,
4516 BPF_CGROUP_SOCK_OPS),
4517 BPF_APROG_SEC("sk_skb/stream_parser", BPF_PROG_TYPE_SK_SKB,
4518 BPF_SK_SKB_STREAM_PARSER),
4519 BPF_APROG_SEC("sk_skb/stream_verdict", BPF_PROG_TYPE_SK_SKB,
4520 BPF_SK_SKB_STREAM_VERDICT),
4521 BPF_APROG_COMPAT("sk_skb", BPF_PROG_TYPE_SK_SKB),
4522 BPF_APROG_SEC("sk_msg", BPF_PROG_TYPE_SK_MSG,
4523 BPF_SK_MSG_VERDICT),
4524 BPF_APROG_SEC("lirc_mode2", BPF_PROG_TYPE_LIRC_MODE2,
4525 BPF_LIRC_MODE2),
4526 BPF_APROG_SEC("flow_dissector", BPF_PROG_TYPE_FLOW_DISSECTOR,
4527 BPF_FLOW_DISSECTOR),
4528 BPF_EAPROG_SEC("cgroup/bind4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4529 BPF_CGROUP_INET4_BIND),
4530 BPF_EAPROG_SEC("cgroup/bind6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4531 BPF_CGROUP_INET6_BIND),
4532 BPF_EAPROG_SEC("cgroup/connect4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4533 BPF_CGROUP_INET4_CONNECT),
4534 BPF_EAPROG_SEC("cgroup/connect6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4535 BPF_CGROUP_INET6_CONNECT),
4536 BPF_EAPROG_SEC("cgroup/sendmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4537 BPF_CGROUP_UDP4_SENDMSG),
4538 BPF_EAPROG_SEC("cgroup/sendmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4539 BPF_CGROUP_UDP6_SENDMSG),
4540 BPF_EAPROG_SEC("cgroup/recvmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4541 BPF_CGROUP_UDP4_RECVMSG),
4542 BPF_EAPROG_SEC("cgroup/recvmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4543 BPF_CGROUP_UDP6_RECVMSG),
4544 BPF_EAPROG_SEC("cgroup/sysctl", BPF_PROG_TYPE_CGROUP_SYSCTL,
4545 BPF_CGROUP_SYSCTL),
4546 BPF_EAPROG_SEC("cgroup/getsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT,
4547 BPF_CGROUP_GETSOCKOPT),
4548 BPF_EAPROG_SEC("cgroup/setsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT,
4549 BPF_CGROUP_SETSOCKOPT),
4550 };
4551
4552 #undef BPF_PROG_SEC_IMPL
4553 #undef BPF_PROG_SEC
4554 #undef BPF_APROG_SEC
4555 #undef BPF_EAPROG_SEC
4556 #undef BPF_APROG_COMPAT
4557
4558 #define MAX_TYPE_NAME_SIZE 32
4559
libbpf_get_type_names(bool attach_type)4560 static char *libbpf_get_type_names(bool attach_type)
4561 {
4562 int i, len = ARRAY_SIZE(section_names) * MAX_TYPE_NAME_SIZE;
4563 char *buf;
4564
4565 buf = malloc(len);
4566 if (!buf)
4567 return NULL;
4568
4569 buf[0] = '\0';
4570 /* Forge string buf with all available names */
4571 for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4572 if (attach_type && !section_names[i].is_attachable)
4573 continue;
4574
4575 if (strlen(buf) + strlen(section_names[i].sec) + 2 > len) {
4576 free(buf);
4577 return NULL;
4578 }
4579 strcat(buf, " ");
4580 strcat(buf, section_names[i].sec);
4581 }
4582
4583 return buf;
4584 }
4585
libbpf_prog_type_by_name(const char * name,enum bpf_prog_type * prog_type,enum bpf_attach_type * expected_attach_type)4586 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
4587 enum bpf_attach_type *expected_attach_type)
4588 {
4589 char *type_names;
4590 int i;
4591
4592 if (!name)
4593 return -EINVAL;
4594
4595 for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4596 if (strncmp(name, section_names[i].sec, section_names[i].len))
4597 continue;
4598 *prog_type = section_names[i].prog_type;
4599 *expected_attach_type = section_names[i].expected_attach_type;
4600 return 0;
4601 }
4602 pr_warning("failed to guess program type based on ELF section name '%s'\n", name);
4603 type_names = libbpf_get_type_names(false);
4604 if (type_names != NULL) {
4605 pr_info("supported section(type) names are:%s\n", type_names);
4606 free(type_names);
4607 }
4608
4609 return -EINVAL;
4610 }
4611
libbpf_attach_type_by_name(const char * name,enum bpf_attach_type * attach_type)4612 int libbpf_attach_type_by_name(const char *name,
4613 enum bpf_attach_type *attach_type)
4614 {
4615 char *type_names;
4616 int i;
4617
4618 if (!name)
4619 return -EINVAL;
4620
4621 for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4622 if (strncmp(name, section_names[i].sec, section_names[i].len))
4623 continue;
4624 if (!section_names[i].is_attachable)
4625 return -EINVAL;
4626 *attach_type = section_names[i].attach_type;
4627 return 0;
4628 }
4629 pr_warning("failed to guess attach type based on ELF section name '%s'\n", name);
4630 type_names = libbpf_get_type_names(true);
4631 if (type_names != NULL) {
4632 pr_info("attachable section(type) names are:%s\n", type_names);
4633 free(type_names);
4634 }
4635
4636 return -EINVAL;
4637 }
4638
4639 static int
bpf_program__identify_section(struct bpf_program * prog,enum bpf_prog_type * prog_type,enum bpf_attach_type * expected_attach_type)4640 bpf_program__identify_section(struct bpf_program *prog,
4641 enum bpf_prog_type *prog_type,
4642 enum bpf_attach_type *expected_attach_type)
4643 {
4644 return libbpf_prog_type_by_name(prog->section_name, prog_type,
4645 expected_attach_type);
4646 }
4647
bpf_map__fd(const struct bpf_map * map)4648 int bpf_map__fd(const struct bpf_map *map)
4649 {
4650 return map ? map->fd : -EINVAL;
4651 }
4652
bpf_map__def(const struct bpf_map * map)4653 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
4654 {
4655 return map ? &map->def : ERR_PTR(-EINVAL);
4656 }
4657
bpf_map__name(const struct bpf_map * map)4658 const char *bpf_map__name(const struct bpf_map *map)
4659 {
4660 return map ? map->name : NULL;
4661 }
4662
bpf_map__btf_key_type_id(const struct bpf_map * map)4663 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
4664 {
4665 return map ? map->btf_key_type_id : 0;
4666 }
4667
bpf_map__btf_value_type_id(const struct bpf_map * map)4668 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
4669 {
4670 return map ? map->btf_value_type_id : 0;
4671 }
4672
bpf_map__set_priv(struct bpf_map * map,void * priv,bpf_map_clear_priv_t clear_priv)4673 int bpf_map__set_priv(struct bpf_map *map, void *priv,
4674 bpf_map_clear_priv_t clear_priv)
4675 {
4676 if (!map)
4677 return -EINVAL;
4678
4679 if (map->priv) {
4680 if (map->clear_priv)
4681 map->clear_priv(map, map->priv);
4682 }
4683
4684 map->priv = priv;
4685 map->clear_priv = clear_priv;
4686 return 0;
4687 }
4688
bpf_map__priv(const struct bpf_map * map)4689 void *bpf_map__priv(const struct bpf_map *map)
4690 {
4691 return map ? map->priv : ERR_PTR(-EINVAL);
4692 }
4693
bpf_map__is_offload_neutral(const struct bpf_map * map)4694 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
4695 {
4696 return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
4697 }
4698
bpf_map__is_internal(const struct bpf_map * map)4699 bool bpf_map__is_internal(const struct bpf_map *map)
4700 {
4701 return map->libbpf_type != LIBBPF_MAP_UNSPEC;
4702 }
4703
bpf_map__set_ifindex(struct bpf_map * map,__u32 ifindex)4704 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
4705 {
4706 map->map_ifindex = ifindex;
4707 }
4708
bpf_map__set_inner_map_fd(struct bpf_map * map,int fd)4709 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
4710 {
4711 if (!bpf_map_type__is_map_in_map(map->def.type)) {
4712 pr_warning("error: unsupported map type\n");
4713 return -EINVAL;
4714 }
4715 if (map->inner_map_fd != -1) {
4716 pr_warning("error: inner_map_fd already specified\n");
4717 return -EINVAL;
4718 }
4719 map->inner_map_fd = fd;
4720 return 0;
4721 }
4722
4723 static struct bpf_map *
__bpf_map__iter(const struct bpf_map * m,const struct bpf_object * obj,int i)4724 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
4725 {
4726 ssize_t idx;
4727 struct bpf_map *s, *e;
4728
4729 if (!obj || !obj->maps)
4730 return NULL;
4731
4732 s = obj->maps;
4733 e = obj->maps + obj->nr_maps;
4734
4735 if ((m < s) || (m >= e)) {
4736 pr_warning("error in %s: map handler doesn't belong to object\n",
4737 __func__);
4738 return NULL;
4739 }
4740
4741 idx = (m - obj->maps) + i;
4742 if (idx >= obj->nr_maps || idx < 0)
4743 return NULL;
4744 return &obj->maps[idx];
4745 }
4746
4747 struct bpf_map *
bpf_map__next(const struct bpf_map * prev,const struct bpf_object * obj)4748 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
4749 {
4750 if (prev == NULL)
4751 return obj->maps;
4752
4753 return __bpf_map__iter(prev, obj, 1);
4754 }
4755
4756 struct bpf_map *
bpf_map__prev(const struct bpf_map * next,const struct bpf_object * obj)4757 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
4758 {
4759 if (next == NULL) {
4760 if (!obj->nr_maps)
4761 return NULL;
4762 return obj->maps + obj->nr_maps - 1;
4763 }
4764
4765 return __bpf_map__iter(next, obj, -1);
4766 }
4767
4768 struct bpf_map *
bpf_object__find_map_by_name(const struct bpf_object * obj,const char * name)4769 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
4770 {
4771 struct bpf_map *pos;
4772
4773 bpf_object__for_each_map(pos, obj) {
4774 if (pos->name && !strcmp(pos->name, name))
4775 return pos;
4776 }
4777 return NULL;
4778 }
4779
4780 int
bpf_object__find_map_fd_by_name(const struct bpf_object * obj,const char * name)4781 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
4782 {
4783 return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
4784 }
4785
4786 struct bpf_map *
bpf_object__find_map_by_offset(struct bpf_object * obj,size_t offset)4787 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
4788 {
4789 return ERR_PTR(-ENOTSUP);
4790 }
4791
libbpf_get_error(const void * ptr)4792 long libbpf_get_error(const void *ptr)
4793 {
4794 return PTR_ERR_OR_ZERO(ptr);
4795 }
4796
bpf_prog_load(const char * file,enum bpf_prog_type type,struct bpf_object ** pobj,int * prog_fd)4797 int bpf_prog_load(const char *file, enum bpf_prog_type type,
4798 struct bpf_object **pobj, int *prog_fd)
4799 {
4800 struct bpf_prog_load_attr attr;
4801
4802 memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
4803 attr.file = file;
4804 attr.prog_type = type;
4805 attr.expected_attach_type = 0;
4806
4807 return bpf_prog_load_xattr(&attr, pobj, prog_fd);
4808 }
4809
bpf_prog_load_xattr(const struct bpf_prog_load_attr * attr,struct bpf_object ** pobj,int * prog_fd)4810 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
4811 struct bpf_object **pobj, int *prog_fd)
4812 {
4813 struct bpf_object_open_attr open_attr = {};
4814 struct bpf_program *prog, *first_prog = NULL;
4815 enum bpf_attach_type expected_attach_type;
4816 enum bpf_prog_type prog_type;
4817 struct bpf_object *obj;
4818 struct bpf_map *map;
4819 int err;
4820
4821 if (!attr)
4822 return -EINVAL;
4823 if (!attr->file)
4824 return -EINVAL;
4825
4826 open_attr.file = attr->file;
4827 open_attr.prog_type = attr->prog_type;
4828
4829 obj = bpf_object__open_xattr(&open_attr);
4830 if (IS_ERR_OR_NULL(obj))
4831 return -ENOENT;
4832
4833 bpf_object__for_each_program(prog, obj) {
4834 /*
4835 * If type is not specified, try to guess it based on
4836 * section name.
4837 */
4838 prog_type = attr->prog_type;
4839 prog->prog_ifindex = attr->ifindex;
4840 expected_attach_type = attr->expected_attach_type;
4841 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
4842 err = bpf_program__identify_section(prog, &prog_type,
4843 &expected_attach_type);
4844 if (err < 0) {
4845 bpf_object__close(obj);
4846 return -EINVAL;
4847 }
4848 }
4849
4850 bpf_program__set_type(prog, prog_type);
4851 bpf_program__set_expected_attach_type(prog,
4852 expected_attach_type);
4853
4854 prog->log_level = attr->log_level;
4855 prog->prog_flags = attr->prog_flags;
4856 if (!first_prog)
4857 first_prog = prog;
4858 }
4859
4860 bpf_object__for_each_map(map, obj) {
4861 if (!bpf_map__is_offload_neutral(map))
4862 map->map_ifindex = attr->ifindex;
4863 }
4864
4865 if (!first_prog) {
4866 pr_warning("object file doesn't contain bpf program\n");
4867 bpf_object__close(obj);
4868 return -ENOENT;
4869 }
4870
4871 err = bpf_object__load(obj);
4872 if (err) {
4873 bpf_object__close(obj);
4874 return -EINVAL;
4875 }
4876
4877 *pobj = obj;
4878 *prog_fd = bpf_program__fd(first_prog);
4879 return 0;
4880 }
4881
4882 struct bpf_link {
4883 int (*destroy)(struct bpf_link *link);
4884 };
4885
bpf_link__destroy(struct bpf_link * link)4886 int bpf_link__destroy(struct bpf_link *link)
4887 {
4888 int err;
4889
4890 if (!link)
4891 return 0;
4892
4893 err = link->destroy(link);
4894 free(link);
4895
4896 return err;
4897 }
4898
4899 struct bpf_link_fd {
4900 struct bpf_link link; /* has to be at the top of struct */
4901 int fd; /* hook FD */
4902 };
4903
bpf_link__destroy_perf_event(struct bpf_link * link)4904 static int bpf_link__destroy_perf_event(struct bpf_link *link)
4905 {
4906 struct bpf_link_fd *l = (void *)link;
4907 int err;
4908
4909 err = ioctl(l->fd, PERF_EVENT_IOC_DISABLE, 0);
4910 if (err)
4911 err = -errno;
4912
4913 close(l->fd);
4914 return err;
4915 }
4916
bpf_program__attach_perf_event(struct bpf_program * prog,int pfd)4917 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
4918 int pfd)
4919 {
4920 char errmsg[STRERR_BUFSIZE];
4921 struct bpf_link_fd *link;
4922 int prog_fd, err;
4923
4924 if (pfd < 0) {
4925 pr_warning("program '%s': invalid perf event FD %d\n",
4926 bpf_program__title(prog, false), pfd);
4927 return ERR_PTR(-EINVAL);
4928 }
4929 prog_fd = bpf_program__fd(prog);
4930 if (prog_fd < 0) {
4931 pr_warning("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
4932 bpf_program__title(prog, false));
4933 return ERR_PTR(-EINVAL);
4934 }
4935
4936 link = malloc(sizeof(*link));
4937 if (!link)
4938 return ERR_PTR(-ENOMEM);
4939 link->link.destroy = &bpf_link__destroy_perf_event;
4940 link->fd = pfd;
4941
4942 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
4943 err = -errno;
4944 free(link);
4945 pr_warning("program '%s': failed to attach to pfd %d: %s\n",
4946 bpf_program__title(prog, false), pfd,
4947 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4948 return ERR_PTR(err);
4949 }
4950 if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
4951 err = -errno;
4952 free(link);
4953 pr_warning("program '%s': failed to enable pfd %d: %s\n",
4954 bpf_program__title(prog, false), pfd,
4955 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4956 return ERR_PTR(err);
4957 }
4958 return (struct bpf_link *)link;
4959 }
4960
4961 /*
4962 * this function is expected to parse integer in the range of [0, 2^31-1] from
4963 * given file using scanf format string fmt. If actual parsed value is
4964 * negative, the result might be indistinguishable from error
4965 */
parse_uint_from_file(const char * file,const char * fmt)4966 static int parse_uint_from_file(const char *file, const char *fmt)
4967 {
4968 char buf[STRERR_BUFSIZE];
4969 int err, ret;
4970 FILE *f;
4971
4972 f = fopen(file, "r");
4973 if (!f) {
4974 err = -errno;
4975 pr_debug("failed to open '%s': %s\n", file,
4976 libbpf_strerror_r(err, buf, sizeof(buf)));
4977 return err;
4978 }
4979 err = fscanf(f, fmt, &ret);
4980 if (err != 1) {
4981 err = err == EOF ? -EIO : -errno;
4982 pr_debug("failed to parse '%s': %s\n", file,
4983 libbpf_strerror_r(err, buf, sizeof(buf)));
4984 fclose(f);
4985 return err;
4986 }
4987 fclose(f);
4988 return ret;
4989 }
4990
determine_kprobe_perf_type(void)4991 static int determine_kprobe_perf_type(void)
4992 {
4993 const char *file = "/sys/bus/event_source/devices/kprobe/type";
4994
4995 return parse_uint_from_file(file, "%d\n");
4996 }
4997
determine_uprobe_perf_type(void)4998 static int determine_uprobe_perf_type(void)
4999 {
5000 const char *file = "/sys/bus/event_source/devices/uprobe/type";
5001
5002 return parse_uint_from_file(file, "%d\n");
5003 }
5004
determine_kprobe_retprobe_bit(void)5005 static int determine_kprobe_retprobe_bit(void)
5006 {
5007 const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
5008
5009 return parse_uint_from_file(file, "config:%d\n");
5010 }
5011
determine_uprobe_retprobe_bit(void)5012 static int determine_uprobe_retprobe_bit(void)
5013 {
5014 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
5015
5016 return parse_uint_from_file(file, "config:%d\n");
5017 }
5018
perf_event_open_probe(bool uprobe,bool retprobe,const char * name,uint64_t offset,int pid)5019 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
5020 uint64_t offset, int pid)
5021 {
5022 struct perf_event_attr attr = {};
5023 char errmsg[STRERR_BUFSIZE];
5024 int type, pfd, err;
5025
5026 type = uprobe ? determine_uprobe_perf_type()
5027 : determine_kprobe_perf_type();
5028 if (type < 0) {
5029 pr_warning("failed to determine %s perf type: %s\n",
5030 uprobe ? "uprobe" : "kprobe",
5031 libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
5032 return type;
5033 }
5034 if (retprobe) {
5035 int bit = uprobe ? determine_uprobe_retprobe_bit()
5036 : determine_kprobe_retprobe_bit();
5037
5038 if (bit < 0) {
5039 pr_warning("failed to determine %s retprobe bit: %s\n",
5040 uprobe ? "uprobe" : "kprobe",
5041 libbpf_strerror_r(bit, errmsg,
5042 sizeof(errmsg)));
5043 return bit;
5044 }
5045 attr.config |= 1 << bit;
5046 }
5047 attr.size = sizeof(attr);
5048 attr.type = type;
5049 attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
5050 attr.config2 = offset; /* kprobe_addr or probe_offset */
5051
5052 /* pid filter is meaningful only for uprobes */
5053 pfd = syscall(__NR_perf_event_open, &attr,
5054 pid < 0 ? -1 : pid /* pid */,
5055 pid == -1 ? 0 : -1 /* cpu */,
5056 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5057 if (pfd < 0) {
5058 err = -errno;
5059 pr_warning("%s perf_event_open() failed: %s\n",
5060 uprobe ? "uprobe" : "kprobe",
5061 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5062 return err;
5063 }
5064 return pfd;
5065 }
5066
bpf_program__attach_kprobe(struct bpf_program * prog,bool retprobe,const char * func_name)5067 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
5068 bool retprobe,
5069 const char *func_name)
5070 {
5071 char errmsg[STRERR_BUFSIZE];
5072 struct bpf_link *link;
5073 int pfd, err;
5074
5075 pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
5076 0 /* offset */, -1 /* pid */);
5077 if (pfd < 0) {
5078 pr_warning("program '%s': failed to create %s '%s' perf event: %s\n",
5079 bpf_program__title(prog, false),
5080 retprobe ? "kretprobe" : "kprobe", func_name,
5081 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5082 return ERR_PTR(pfd);
5083 }
5084 link = bpf_program__attach_perf_event(prog, pfd);
5085 if (IS_ERR(link)) {
5086 close(pfd);
5087 err = PTR_ERR(link);
5088 pr_warning("program '%s': failed to attach to %s '%s': %s\n",
5089 bpf_program__title(prog, false),
5090 retprobe ? "kretprobe" : "kprobe", func_name,
5091 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5092 return link;
5093 }
5094 return link;
5095 }
5096
bpf_program__attach_uprobe(struct bpf_program * prog,bool retprobe,pid_t pid,const char * binary_path,size_t func_offset)5097 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
5098 bool retprobe, pid_t pid,
5099 const char *binary_path,
5100 size_t func_offset)
5101 {
5102 char errmsg[STRERR_BUFSIZE];
5103 struct bpf_link *link;
5104 int pfd, err;
5105
5106 pfd = perf_event_open_probe(true /* uprobe */, retprobe,
5107 binary_path, func_offset, pid);
5108 if (pfd < 0) {
5109 pr_warning("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
5110 bpf_program__title(prog, false),
5111 retprobe ? "uretprobe" : "uprobe",
5112 binary_path, func_offset,
5113 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5114 return ERR_PTR(pfd);
5115 }
5116 link = bpf_program__attach_perf_event(prog, pfd);
5117 if (IS_ERR(link)) {
5118 close(pfd);
5119 err = PTR_ERR(link);
5120 pr_warning("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
5121 bpf_program__title(prog, false),
5122 retprobe ? "uretprobe" : "uprobe",
5123 binary_path, func_offset,
5124 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5125 return link;
5126 }
5127 return link;
5128 }
5129
determine_tracepoint_id(const char * tp_category,const char * tp_name)5130 static int determine_tracepoint_id(const char *tp_category,
5131 const char *tp_name)
5132 {
5133 char file[PATH_MAX];
5134 int ret;
5135
5136 ret = snprintf(file, sizeof(file),
5137 "/sys/kernel/debug/tracing/events/%s/%s/id",
5138 tp_category, tp_name);
5139 if (ret < 0)
5140 return -errno;
5141 if (ret >= sizeof(file)) {
5142 pr_debug("tracepoint %s/%s path is too long\n",
5143 tp_category, tp_name);
5144 return -E2BIG;
5145 }
5146 return parse_uint_from_file(file, "%d\n");
5147 }
5148
perf_event_open_tracepoint(const char * tp_category,const char * tp_name)5149 static int perf_event_open_tracepoint(const char *tp_category,
5150 const char *tp_name)
5151 {
5152 struct perf_event_attr attr = {};
5153 char errmsg[STRERR_BUFSIZE];
5154 int tp_id, pfd, err;
5155
5156 tp_id = determine_tracepoint_id(tp_category, tp_name);
5157 if (tp_id < 0) {
5158 pr_warning("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
5159 tp_category, tp_name,
5160 libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
5161 return tp_id;
5162 }
5163
5164 attr.type = PERF_TYPE_TRACEPOINT;
5165 attr.size = sizeof(attr);
5166 attr.config = tp_id;
5167
5168 pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
5169 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5170 if (pfd < 0) {
5171 err = -errno;
5172 pr_warning("tracepoint '%s/%s' perf_event_open() failed: %s\n",
5173 tp_category, tp_name,
5174 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5175 return err;
5176 }
5177 return pfd;
5178 }
5179
bpf_program__attach_tracepoint(struct bpf_program * prog,const char * tp_category,const char * tp_name)5180 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
5181 const char *tp_category,
5182 const char *tp_name)
5183 {
5184 char errmsg[STRERR_BUFSIZE];
5185 struct bpf_link *link;
5186 int pfd, err;
5187
5188 pfd = perf_event_open_tracepoint(tp_category, tp_name);
5189 if (pfd < 0) {
5190 pr_warning("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
5191 bpf_program__title(prog, false),
5192 tp_category, tp_name,
5193 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5194 return ERR_PTR(pfd);
5195 }
5196 link = bpf_program__attach_perf_event(prog, pfd);
5197 if (IS_ERR(link)) {
5198 close(pfd);
5199 err = PTR_ERR(link);
5200 pr_warning("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
5201 bpf_program__title(prog, false),
5202 tp_category, tp_name,
5203 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5204 return link;
5205 }
5206 return link;
5207 }
5208
bpf_link__destroy_fd(struct bpf_link * link)5209 static int bpf_link__destroy_fd(struct bpf_link *link)
5210 {
5211 struct bpf_link_fd *l = (void *)link;
5212
5213 return close(l->fd);
5214 }
5215
bpf_program__attach_raw_tracepoint(struct bpf_program * prog,const char * tp_name)5216 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
5217 const char *tp_name)
5218 {
5219 char errmsg[STRERR_BUFSIZE];
5220 struct bpf_link_fd *link;
5221 int prog_fd, pfd;
5222
5223 prog_fd = bpf_program__fd(prog);
5224 if (prog_fd < 0) {
5225 pr_warning("program '%s': can't attach before loaded\n",
5226 bpf_program__title(prog, false));
5227 return ERR_PTR(-EINVAL);
5228 }
5229
5230 link = malloc(sizeof(*link));
5231 if (!link)
5232 return ERR_PTR(-ENOMEM);
5233 link->link.destroy = &bpf_link__destroy_fd;
5234
5235 pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
5236 if (pfd < 0) {
5237 pfd = -errno;
5238 free(link);
5239 pr_warning("program '%s': failed to attach to raw tracepoint '%s': %s\n",
5240 bpf_program__title(prog, false), tp_name,
5241 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5242 return ERR_PTR(pfd);
5243 }
5244 link->fd = pfd;
5245 return (struct bpf_link *)link;
5246 }
5247
5248 enum bpf_perf_event_ret
bpf_perf_event_read_simple(void * mmap_mem,size_t mmap_size,size_t page_size,void ** copy_mem,size_t * copy_size,bpf_perf_event_print_t fn,void * private_data)5249 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
5250 void **copy_mem, size_t *copy_size,
5251 bpf_perf_event_print_t fn, void *private_data)
5252 {
5253 struct perf_event_mmap_page *header = mmap_mem;
5254 __u64 data_head = ring_buffer_read_head(header);
5255 __u64 data_tail = header->data_tail;
5256 void *base = ((__u8 *)header) + page_size;
5257 int ret = LIBBPF_PERF_EVENT_CONT;
5258 struct perf_event_header *ehdr;
5259 size_t ehdr_size;
5260
5261 while (data_head != data_tail) {
5262 ehdr = base + (data_tail & (mmap_size - 1));
5263 ehdr_size = ehdr->size;
5264
5265 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
5266 void *copy_start = ehdr;
5267 size_t len_first = base + mmap_size - copy_start;
5268 size_t len_secnd = ehdr_size - len_first;
5269
5270 if (*copy_size < ehdr_size) {
5271 free(*copy_mem);
5272 *copy_mem = malloc(ehdr_size);
5273 if (!*copy_mem) {
5274 *copy_size = 0;
5275 ret = LIBBPF_PERF_EVENT_ERROR;
5276 break;
5277 }
5278 *copy_size = ehdr_size;
5279 }
5280
5281 memcpy(*copy_mem, copy_start, len_first);
5282 memcpy(*copy_mem + len_first, base, len_secnd);
5283 ehdr = *copy_mem;
5284 }
5285
5286 ret = fn(ehdr, private_data);
5287 data_tail += ehdr_size;
5288 if (ret != LIBBPF_PERF_EVENT_CONT)
5289 break;
5290 }
5291
5292 ring_buffer_write_tail(header, data_tail);
5293 return ret;
5294 }
5295
5296 struct perf_buffer;
5297
5298 struct perf_buffer_params {
5299 struct perf_event_attr *attr;
5300 /* if event_cb is specified, it takes precendence */
5301 perf_buffer_event_fn event_cb;
5302 /* sample_cb and lost_cb are higher-level common-case callbacks */
5303 perf_buffer_sample_fn sample_cb;
5304 perf_buffer_lost_fn lost_cb;
5305 void *ctx;
5306 int cpu_cnt;
5307 int *cpus;
5308 int *map_keys;
5309 };
5310
5311 struct perf_cpu_buf {
5312 struct perf_buffer *pb;
5313 void *base; /* mmap()'ed memory */
5314 void *buf; /* for reconstructing segmented data */
5315 size_t buf_size;
5316 int fd;
5317 int cpu;
5318 int map_key;
5319 };
5320
5321 struct perf_buffer {
5322 perf_buffer_event_fn event_cb;
5323 perf_buffer_sample_fn sample_cb;
5324 perf_buffer_lost_fn lost_cb;
5325 void *ctx; /* passed into callbacks */
5326
5327 size_t page_size;
5328 size_t mmap_size;
5329 struct perf_cpu_buf **cpu_bufs;
5330 struct epoll_event *events;
5331 int cpu_cnt;
5332 int epoll_fd; /* perf event FD */
5333 int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
5334 };
5335
perf_buffer__free_cpu_buf(struct perf_buffer * pb,struct perf_cpu_buf * cpu_buf)5336 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
5337 struct perf_cpu_buf *cpu_buf)
5338 {
5339 if (!cpu_buf)
5340 return;
5341 if (cpu_buf->base &&
5342 munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
5343 pr_warning("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
5344 if (cpu_buf->fd >= 0) {
5345 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
5346 close(cpu_buf->fd);
5347 }
5348 free(cpu_buf->buf);
5349 free(cpu_buf);
5350 }
5351
perf_buffer__free(struct perf_buffer * pb)5352 void perf_buffer__free(struct perf_buffer *pb)
5353 {
5354 int i;
5355
5356 if (!pb)
5357 return;
5358 if (pb->cpu_bufs) {
5359 for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
5360 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
5361
5362 bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
5363 perf_buffer__free_cpu_buf(pb, cpu_buf);
5364 }
5365 free(pb->cpu_bufs);
5366 }
5367 if (pb->epoll_fd >= 0)
5368 close(pb->epoll_fd);
5369 free(pb->events);
5370 free(pb);
5371 }
5372
5373 static struct perf_cpu_buf *
perf_buffer__open_cpu_buf(struct perf_buffer * pb,struct perf_event_attr * attr,int cpu,int map_key)5374 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
5375 int cpu, int map_key)
5376 {
5377 struct perf_cpu_buf *cpu_buf;
5378 char msg[STRERR_BUFSIZE];
5379 int err;
5380
5381 cpu_buf = calloc(1, sizeof(*cpu_buf));
5382 if (!cpu_buf)
5383 return ERR_PTR(-ENOMEM);
5384
5385 cpu_buf->pb = pb;
5386 cpu_buf->cpu = cpu;
5387 cpu_buf->map_key = map_key;
5388
5389 cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
5390 -1, PERF_FLAG_FD_CLOEXEC);
5391 if (cpu_buf->fd < 0) {
5392 err = -errno;
5393 pr_warning("failed to open perf buffer event on cpu #%d: %s\n",
5394 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5395 goto error;
5396 }
5397
5398 cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
5399 PROT_READ | PROT_WRITE, MAP_SHARED,
5400 cpu_buf->fd, 0);
5401 if (cpu_buf->base == MAP_FAILED) {
5402 cpu_buf->base = NULL;
5403 err = -errno;
5404 pr_warning("failed to mmap perf buffer on cpu #%d: %s\n",
5405 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5406 goto error;
5407 }
5408
5409 if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
5410 err = -errno;
5411 pr_warning("failed to enable perf buffer event on cpu #%d: %s\n",
5412 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5413 goto error;
5414 }
5415
5416 return cpu_buf;
5417
5418 error:
5419 perf_buffer__free_cpu_buf(pb, cpu_buf);
5420 return (struct perf_cpu_buf *)ERR_PTR(err);
5421 }
5422
5423 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
5424 struct perf_buffer_params *p);
5425
perf_buffer__new(int map_fd,size_t page_cnt,const struct perf_buffer_opts * opts)5426 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
5427 const struct perf_buffer_opts *opts)
5428 {
5429 struct perf_buffer_params p = {};
5430 struct perf_event_attr attr = { 0, };
5431
5432 attr.config = PERF_COUNT_SW_BPF_OUTPUT,
5433 attr.type = PERF_TYPE_SOFTWARE;
5434 attr.sample_type = PERF_SAMPLE_RAW;
5435 attr.sample_period = 1;
5436 attr.wakeup_events = 1;
5437
5438 p.attr = &attr;
5439 p.sample_cb = opts ? opts->sample_cb : NULL;
5440 p.lost_cb = opts ? opts->lost_cb : NULL;
5441 p.ctx = opts ? opts->ctx : NULL;
5442
5443 return __perf_buffer__new(map_fd, page_cnt, &p);
5444 }
5445
5446 struct perf_buffer *
perf_buffer__new_raw(int map_fd,size_t page_cnt,const struct perf_buffer_raw_opts * opts)5447 perf_buffer__new_raw(int map_fd, size_t page_cnt,
5448 const struct perf_buffer_raw_opts *opts)
5449 {
5450 struct perf_buffer_params p = {};
5451
5452 p.attr = opts->attr;
5453 p.event_cb = opts->event_cb;
5454 p.ctx = opts->ctx;
5455 p.cpu_cnt = opts->cpu_cnt;
5456 p.cpus = opts->cpus;
5457 p.map_keys = opts->map_keys;
5458
5459 return __perf_buffer__new(map_fd, page_cnt, &p);
5460 }
5461
__perf_buffer__new(int map_fd,size_t page_cnt,struct perf_buffer_params * p)5462 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
5463 struct perf_buffer_params *p)
5464 {
5465 struct bpf_map_info map = {};
5466 char msg[STRERR_BUFSIZE];
5467 struct perf_buffer *pb;
5468 __u32 map_info_len;
5469 int err, i;
5470
5471 if (page_cnt & (page_cnt - 1)) {
5472 pr_warning("page count should be power of two, but is %zu\n",
5473 page_cnt);
5474 return ERR_PTR(-EINVAL);
5475 }
5476
5477 map_info_len = sizeof(map);
5478 err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
5479 if (err) {
5480 err = -errno;
5481 pr_warning("failed to get map info for map FD %d: %s\n",
5482 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
5483 return ERR_PTR(err);
5484 }
5485
5486 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
5487 pr_warning("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
5488 map.name);
5489 return ERR_PTR(-EINVAL);
5490 }
5491
5492 pb = calloc(1, sizeof(*pb));
5493 if (!pb)
5494 return ERR_PTR(-ENOMEM);
5495
5496 pb->event_cb = p->event_cb;
5497 pb->sample_cb = p->sample_cb;
5498 pb->lost_cb = p->lost_cb;
5499 pb->ctx = p->ctx;
5500
5501 pb->page_size = getpagesize();
5502 pb->mmap_size = pb->page_size * page_cnt;
5503 pb->map_fd = map_fd;
5504
5505 pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
5506 if (pb->epoll_fd < 0) {
5507 err = -errno;
5508 pr_warning("failed to create epoll instance: %s\n",
5509 libbpf_strerror_r(err, msg, sizeof(msg)));
5510 goto error;
5511 }
5512
5513 if (p->cpu_cnt > 0) {
5514 pb->cpu_cnt = p->cpu_cnt;
5515 } else {
5516 pb->cpu_cnt = libbpf_num_possible_cpus();
5517 if (pb->cpu_cnt < 0) {
5518 err = pb->cpu_cnt;
5519 goto error;
5520 }
5521 if (map.max_entries < pb->cpu_cnt)
5522 pb->cpu_cnt = map.max_entries;
5523 }
5524
5525 pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
5526 if (!pb->events) {
5527 err = -ENOMEM;
5528 pr_warning("failed to allocate events: out of memory\n");
5529 goto error;
5530 }
5531 pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
5532 if (!pb->cpu_bufs) {
5533 err = -ENOMEM;
5534 pr_warning("failed to allocate buffers: out of memory\n");
5535 goto error;
5536 }
5537
5538 for (i = 0; i < pb->cpu_cnt; i++) {
5539 struct perf_cpu_buf *cpu_buf;
5540 int cpu, map_key;
5541
5542 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
5543 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
5544
5545 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
5546 if (IS_ERR(cpu_buf)) {
5547 err = PTR_ERR(cpu_buf);
5548 goto error;
5549 }
5550
5551 pb->cpu_bufs[i] = cpu_buf;
5552
5553 err = bpf_map_update_elem(pb->map_fd, &map_key,
5554 &cpu_buf->fd, 0);
5555 if (err) {
5556 err = -errno;
5557 pr_warning("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
5558 cpu, map_key, cpu_buf->fd,
5559 libbpf_strerror_r(err, msg, sizeof(msg)));
5560 goto error;
5561 }
5562
5563 pb->events[i].events = EPOLLIN;
5564 pb->events[i].data.ptr = cpu_buf;
5565 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
5566 &pb->events[i]) < 0) {
5567 err = -errno;
5568 pr_warning("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
5569 cpu, cpu_buf->fd,
5570 libbpf_strerror_r(err, msg, sizeof(msg)));
5571 goto error;
5572 }
5573 }
5574
5575 return pb;
5576
5577 error:
5578 if (pb)
5579 perf_buffer__free(pb);
5580 return ERR_PTR(err);
5581 }
5582
5583 struct perf_sample_raw {
5584 struct perf_event_header header;
5585 uint32_t size;
5586 char data[0];
5587 };
5588
5589 struct perf_sample_lost {
5590 struct perf_event_header header;
5591 uint64_t id;
5592 uint64_t lost;
5593 uint64_t sample_id;
5594 };
5595
5596 static enum bpf_perf_event_ret
perf_buffer__process_record(struct perf_event_header * e,void * ctx)5597 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
5598 {
5599 struct perf_cpu_buf *cpu_buf = ctx;
5600 struct perf_buffer *pb = cpu_buf->pb;
5601 void *data = e;
5602
5603 /* user wants full control over parsing perf event */
5604 if (pb->event_cb)
5605 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
5606
5607 switch (e->type) {
5608 case PERF_RECORD_SAMPLE: {
5609 struct perf_sample_raw *s = data;
5610
5611 if (pb->sample_cb)
5612 pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
5613 break;
5614 }
5615 case PERF_RECORD_LOST: {
5616 struct perf_sample_lost *s = data;
5617
5618 if (pb->lost_cb)
5619 pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
5620 break;
5621 }
5622 default:
5623 pr_warning("unknown perf sample type %d\n", e->type);
5624 return LIBBPF_PERF_EVENT_ERROR;
5625 }
5626 return LIBBPF_PERF_EVENT_CONT;
5627 }
5628
perf_buffer__process_records(struct perf_buffer * pb,struct perf_cpu_buf * cpu_buf)5629 static int perf_buffer__process_records(struct perf_buffer *pb,
5630 struct perf_cpu_buf *cpu_buf)
5631 {
5632 enum bpf_perf_event_ret ret;
5633
5634 ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
5635 pb->page_size, &cpu_buf->buf,
5636 &cpu_buf->buf_size,
5637 perf_buffer__process_record, cpu_buf);
5638 if (ret != LIBBPF_PERF_EVENT_CONT)
5639 return ret;
5640 return 0;
5641 }
5642
perf_buffer__poll(struct perf_buffer * pb,int timeout_ms)5643 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
5644 {
5645 int i, cnt, err;
5646
5647 cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
5648 for (i = 0; i < cnt; i++) {
5649 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
5650
5651 err = perf_buffer__process_records(pb, cpu_buf);
5652 if (err) {
5653 pr_warning("error while processing records: %d\n", err);
5654 return err;
5655 }
5656 }
5657 return cnt < 0 ? -errno : cnt;
5658 }
5659
5660 struct bpf_prog_info_array_desc {
5661 int array_offset; /* e.g. offset of jited_prog_insns */
5662 int count_offset; /* e.g. offset of jited_prog_len */
5663 int size_offset; /* > 0: offset of rec size,
5664 * < 0: fix size of -size_offset
5665 */
5666 };
5667
5668 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
5669 [BPF_PROG_INFO_JITED_INSNS] = {
5670 offsetof(struct bpf_prog_info, jited_prog_insns),
5671 offsetof(struct bpf_prog_info, jited_prog_len),
5672 -1,
5673 },
5674 [BPF_PROG_INFO_XLATED_INSNS] = {
5675 offsetof(struct bpf_prog_info, xlated_prog_insns),
5676 offsetof(struct bpf_prog_info, xlated_prog_len),
5677 -1,
5678 },
5679 [BPF_PROG_INFO_MAP_IDS] = {
5680 offsetof(struct bpf_prog_info, map_ids),
5681 offsetof(struct bpf_prog_info, nr_map_ids),
5682 -(int)sizeof(__u32),
5683 },
5684 [BPF_PROG_INFO_JITED_KSYMS] = {
5685 offsetof(struct bpf_prog_info, jited_ksyms),
5686 offsetof(struct bpf_prog_info, nr_jited_ksyms),
5687 -(int)sizeof(__u64),
5688 },
5689 [BPF_PROG_INFO_JITED_FUNC_LENS] = {
5690 offsetof(struct bpf_prog_info, jited_func_lens),
5691 offsetof(struct bpf_prog_info, nr_jited_func_lens),
5692 -(int)sizeof(__u32),
5693 },
5694 [BPF_PROG_INFO_FUNC_INFO] = {
5695 offsetof(struct bpf_prog_info, func_info),
5696 offsetof(struct bpf_prog_info, nr_func_info),
5697 offsetof(struct bpf_prog_info, func_info_rec_size),
5698 },
5699 [BPF_PROG_INFO_LINE_INFO] = {
5700 offsetof(struct bpf_prog_info, line_info),
5701 offsetof(struct bpf_prog_info, nr_line_info),
5702 offsetof(struct bpf_prog_info, line_info_rec_size),
5703 },
5704 [BPF_PROG_INFO_JITED_LINE_INFO] = {
5705 offsetof(struct bpf_prog_info, jited_line_info),
5706 offsetof(struct bpf_prog_info, nr_jited_line_info),
5707 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
5708 },
5709 [BPF_PROG_INFO_PROG_TAGS] = {
5710 offsetof(struct bpf_prog_info, prog_tags),
5711 offsetof(struct bpf_prog_info, nr_prog_tags),
5712 -(int)sizeof(__u8) * BPF_TAG_SIZE,
5713 },
5714
5715 };
5716
bpf_prog_info_read_offset_u32(struct bpf_prog_info * info,int offset)5717 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, int offset)
5718 {
5719 __u32 *array = (__u32 *)info;
5720
5721 if (offset >= 0)
5722 return array[offset / sizeof(__u32)];
5723 return -(int)offset;
5724 }
5725
bpf_prog_info_read_offset_u64(struct bpf_prog_info * info,int offset)5726 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, int offset)
5727 {
5728 __u64 *array = (__u64 *)info;
5729
5730 if (offset >= 0)
5731 return array[offset / sizeof(__u64)];
5732 return -(int)offset;
5733 }
5734
bpf_prog_info_set_offset_u32(struct bpf_prog_info * info,int offset,__u32 val)5735 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
5736 __u32 val)
5737 {
5738 __u32 *array = (__u32 *)info;
5739
5740 if (offset >= 0)
5741 array[offset / sizeof(__u32)] = val;
5742 }
5743
bpf_prog_info_set_offset_u64(struct bpf_prog_info * info,int offset,__u64 val)5744 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
5745 __u64 val)
5746 {
5747 __u64 *array = (__u64 *)info;
5748
5749 if (offset >= 0)
5750 array[offset / sizeof(__u64)] = val;
5751 }
5752
5753 struct bpf_prog_info_linear *
bpf_program__get_prog_info_linear(int fd,__u64 arrays)5754 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
5755 {
5756 struct bpf_prog_info_linear *info_linear;
5757 struct bpf_prog_info info = {};
5758 __u32 info_len = sizeof(info);
5759 __u32 data_len = 0;
5760 int i, err;
5761 void *ptr;
5762
5763 if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
5764 return ERR_PTR(-EINVAL);
5765
5766 /* step 1: get array dimensions */
5767 err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
5768 if (err) {
5769 pr_debug("can't get prog info: %s", strerror(errno));
5770 return ERR_PTR(-EFAULT);
5771 }
5772
5773 /* step 2: calculate total size of all arrays */
5774 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5775 bool include_array = (arrays & (1UL << i)) > 0;
5776 struct bpf_prog_info_array_desc *desc;
5777 __u32 count, size;
5778
5779 desc = bpf_prog_info_array_desc + i;
5780
5781 /* kernel is too old to support this field */
5782 if (info_len < desc->array_offset + sizeof(__u32) ||
5783 info_len < desc->count_offset + sizeof(__u32) ||
5784 (desc->size_offset > 0 && info_len < desc->size_offset))
5785 include_array = false;
5786
5787 if (!include_array) {
5788 arrays &= ~(1UL << i); /* clear the bit */
5789 continue;
5790 }
5791
5792 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5793 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5794
5795 data_len += count * size;
5796 }
5797
5798 /* step 3: allocate continuous memory */
5799 data_len = roundup(data_len, sizeof(__u64));
5800 info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
5801 if (!info_linear)
5802 return ERR_PTR(-ENOMEM);
5803
5804 /* step 4: fill data to info_linear->info */
5805 info_linear->arrays = arrays;
5806 memset(&info_linear->info, 0, sizeof(info));
5807 ptr = info_linear->data;
5808
5809 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5810 struct bpf_prog_info_array_desc *desc;
5811 __u32 count, size;
5812
5813 if ((arrays & (1UL << i)) == 0)
5814 continue;
5815
5816 desc = bpf_prog_info_array_desc + i;
5817 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5818 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5819 bpf_prog_info_set_offset_u32(&info_linear->info,
5820 desc->count_offset, count);
5821 bpf_prog_info_set_offset_u32(&info_linear->info,
5822 desc->size_offset, size);
5823 bpf_prog_info_set_offset_u64(&info_linear->info,
5824 desc->array_offset,
5825 ptr_to_u64(ptr));
5826 ptr += count * size;
5827 }
5828
5829 /* step 5: call syscall again to get required arrays */
5830 err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
5831 if (err) {
5832 pr_debug("can't get prog info: %s", strerror(errno));
5833 free(info_linear);
5834 return ERR_PTR(-EFAULT);
5835 }
5836
5837 /* step 6: verify the data */
5838 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5839 struct bpf_prog_info_array_desc *desc;
5840 __u32 v1, v2;
5841
5842 if ((arrays & (1UL << i)) == 0)
5843 continue;
5844
5845 desc = bpf_prog_info_array_desc + i;
5846 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5847 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
5848 desc->count_offset);
5849 if (v1 != v2)
5850 pr_warning("%s: mismatch in element count\n", __func__);
5851
5852 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5853 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
5854 desc->size_offset);
5855 if (v1 != v2)
5856 pr_warning("%s: mismatch in rec size\n", __func__);
5857 }
5858
5859 /* step 7: update info_len and data_len */
5860 info_linear->info_len = sizeof(struct bpf_prog_info);
5861 info_linear->data_len = data_len;
5862
5863 return info_linear;
5864 }
5865
bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear * info_linear)5866 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
5867 {
5868 int i;
5869
5870 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5871 struct bpf_prog_info_array_desc *desc;
5872 __u64 addr, offs;
5873
5874 if ((info_linear->arrays & (1UL << i)) == 0)
5875 continue;
5876
5877 desc = bpf_prog_info_array_desc + i;
5878 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
5879 desc->array_offset);
5880 offs = addr - ptr_to_u64(info_linear->data);
5881 bpf_prog_info_set_offset_u64(&info_linear->info,
5882 desc->array_offset, offs);
5883 }
5884 }
5885
bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear * info_linear)5886 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
5887 {
5888 int i;
5889
5890 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5891 struct bpf_prog_info_array_desc *desc;
5892 __u64 addr, offs;
5893
5894 if ((info_linear->arrays & (1UL << i)) == 0)
5895 continue;
5896
5897 desc = bpf_prog_info_array_desc + i;
5898 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
5899 desc->array_offset);
5900 addr = offs + ptr_to_u64(info_linear->data);
5901 bpf_prog_info_set_offset_u64(&info_linear->info,
5902 desc->array_offset, addr);
5903 }
5904 }
5905
libbpf_num_possible_cpus(void)5906 int libbpf_num_possible_cpus(void)
5907 {
5908 static const char *fcpu = "/sys/devices/system/cpu/possible";
5909 int len = 0, n = 0, il = 0, ir = 0;
5910 unsigned int start = 0, end = 0;
5911 int tmp_cpus = 0;
5912 static int cpus;
5913 char buf[128];
5914 int error = 0;
5915 int fd = -1;
5916
5917 tmp_cpus = READ_ONCE(cpus);
5918 if (tmp_cpus > 0)
5919 return tmp_cpus;
5920
5921 fd = open(fcpu, O_RDONLY);
5922 if (fd < 0) {
5923 error = errno;
5924 pr_warning("Failed to open file %s: %s\n",
5925 fcpu, strerror(error));
5926 return -error;
5927 }
5928 len = read(fd, buf, sizeof(buf));
5929 close(fd);
5930 if (len <= 0) {
5931 error = len ? errno : EINVAL;
5932 pr_warning("Failed to read # of possible cpus from %s: %s\n",
5933 fcpu, strerror(error));
5934 return -error;
5935 }
5936 if (len == sizeof(buf)) {
5937 pr_warning("File %s size overflow\n", fcpu);
5938 return -EOVERFLOW;
5939 }
5940 buf[len] = '\0';
5941
5942 for (ir = 0, tmp_cpus = 0; ir <= len; ir++) {
5943 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
5944 if (buf[ir] == ',' || buf[ir] == '\0') {
5945 buf[ir] = '\0';
5946 n = sscanf(&buf[il], "%u-%u", &start, &end);
5947 if (n <= 0) {
5948 pr_warning("Failed to get # CPUs from %s\n",
5949 &buf[il]);
5950 return -EINVAL;
5951 } else if (n == 1) {
5952 end = start;
5953 }
5954 tmp_cpus += end - start + 1;
5955 il = ir + 1;
5956 }
5957 }
5958 if (tmp_cpus <= 0) {
5959 pr_warning("Invalid #CPUs %d from %s\n", tmp_cpus, fcpu);
5960 return -EINVAL;
5961 }
5962
5963 WRITE_ONCE(cpus, tmp_cpus);
5964 return tmp_cpus;
5965 }
5966