1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2
3 /*
4 * Common eBPF ELF object loading operations.
5 *
6 * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
7 * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
8 * Copyright (C) 2015 Huawei Inc.
9 * Copyright (C) 2017 Nicira, Inc.
10 * Copyright (C) 2019 Isovalent, Inc.
11 */
12
13 #ifndef _GNU_SOURCE
14 #define _GNU_SOURCE
15 #endif
16 #include <stdlib.h>
17 #include <stdio.h>
18 #include <stdarg.h>
19 #include <libgen.h>
20 #include <inttypes.h>
21 #include <limits.h>
22 #include <string.h>
23 #include <unistd.h>
24 #include <endian.h>
25 #include <fcntl.h>
26 #include <errno.h>
27 #include <ctype.h>
28 #include <asm/unistd.h>
29 #include <linux/err.h>
30 #include <linux/kernel.h>
31 #include <linux/bpf.h>
32 #include <linux/btf.h>
33 #include <linux/filter.h>
34 #include <linux/list.h>
35 #include <linux/limits.h>
36 #include <linux/perf_event.h>
37 #include <linux/ring_buffer.h>
38 #include <linux/version.h>
39 #include <sys/epoll.h>
40 #include <sys/ioctl.h>
41 #include <sys/mman.h>
42 #include <sys/stat.h>
43 #include <sys/types.h>
44 #include <sys/vfs.h>
45 #include <sys/utsname.h>
46 #include <sys/resource.h>
47 #include <libelf.h>
48 #include <gelf.h>
49 #include <zlib.h>
50
51 #include "libbpf.h"
52 #include "bpf.h"
53 #include "btf.h"
54 #include "str_error.h"
55 #include "libbpf_internal.h"
56 #include "hashmap.h"
57 #include "bpf_gen_internal.h"
58
59 #ifndef BPF_FS_MAGIC
60 #define BPF_FS_MAGIC 0xcafe4a11
61 #endif
62
63 #define BPF_INSN_SZ (sizeof(struct bpf_insn))
64
65 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
66 * compilation if user enables corresponding warning. Disable it explicitly.
67 */
68 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
69
70 #define __printf(a, b) __attribute__((format(printf, a, b)))
71
72 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
73 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog);
74
__base_pr(enum libbpf_print_level level,const char * format,va_list args)75 static int __base_pr(enum libbpf_print_level level, const char *format,
76 va_list args)
77 {
78 if (level == LIBBPF_DEBUG)
79 return 0;
80
81 return vfprintf(stderr, format, args);
82 }
83
84 static libbpf_print_fn_t __libbpf_pr = __base_pr;
85
libbpf_set_print(libbpf_print_fn_t fn)86 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
87 {
88 libbpf_print_fn_t old_print_fn = __libbpf_pr;
89
90 __libbpf_pr = fn;
91 return old_print_fn;
92 }
93
94 __printf(2, 3)
libbpf_print(enum libbpf_print_level level,const char * format,...)95 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
96 {
97 va_list args;
98
99 if (!__libbpf_pr)
100 return;
101
102 va_start(args, format);
103 __libbpf_pr(level, format, args);
104 va_end(args);
105 }
106
pr_perm_msg(int err)107 static void pr_perm_msg(int err)
108 {
109 struct rlimit limit;
110 char buf[100];
111
112 if (err != -EPERM || geteuid() != 0)
113 return;
114
115 err = getrlimit(RLIMIT_MEMLOCK, &limit);
116 if (err)
117 return;
118
119 if (limit.rlim_cur == RLIM_INFINITY)
120 return;
121
122 if (limit.rlim_cur < 1024)
123 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
124 else if (limit.rlim_cur < 1024*1024)
125 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
126 else
127 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
128
129 pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
130 buf);
131 }
132
133 #define STRERR_BUFSIZE 128
134
135 /* Copied from tools/perf/util/util.h */
136 #ifndef zfree
137 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
138 #endif
139
140 #ifndef zclose
141 # define zclose(fd) ({ \
142 int ___err = 0; \
143 if ((fd) >= 0) \
144 ___err = close((fd)); \
145 fd = -1; \
146 ___err; })
147 #endif
148
ptr_to_u64(const void * ptr)149 static inline __u64 ptr_to_u64(const void *ptr)
150 {
151 return (__u64) (unsigned long) ptr;
152 }
153
154 /* this goes away in libbpf 1.0 */
155 enum libbpf_strict_mode libbpf_mode = LIBBPF_STRICT_NONE;
156
libbpf_set_strict_mode(enum libbpf_strict_mode mode)157 int libbpf_set_strict_mode(enum libbpf_strict_mode mode)
158 {
159 libbpf_mode = mode;
160 return 0;
161 }
162
libbpf_major_version(void)163 __u32 libbpf_major_version(void)
164 {
165 return LIBBPF_MAJOR_VERSION;
166 }
167
libbpf_minor_version(void)168 __u32 libbpf_minor_version(void)
169 {
170 return LIBBPF_MINOR_VERSION;
171 }
172
libbpf_version_string(void)173 const char *libbpf_version_string(void)
174 {
175 #define __S(X) #X
176 #define _S(X) __S(X)
177 return "v" _S(LIBBPF_MAJOR_VERSION) "." _S(LIBBPF_MINOR_VERSION);
178 #undef _S
179 #undef __S
180 }
181
182 enum reloc_type {
183 RELO_LD64,
184 RELO_CALL,
185 RELO_DATA,
186 RELO_EXTERN_VAR,
187 RELO_EXTERN_FUNC,
188 RELO_SUBPROG_ADDR,
189 RELO_CORE,
190 };
191
192 struct reloc_desc {
193 enum reloc_type type;
194 int insn_idx;
195 union {
196 const struct bpf_core_relo *core_relo; /* used when type == RELO_CORE */
197 struct {
198 int map_idx;
199 int sym_off;
200 };
201 };
202 };
203
204 struct bpf_sec_def;
205
206 typedef int (*init_fn_t)(struct bpf_program *prog, long cookie);
207 typedef int (*preload_fn_t)(struct bpf_program *prog, struct bpf_prog_load_opts *opts, long cookie);
208 typedef struct bpf_link *(*attach_fn_t)(const struct bpf_program *prog, long cookie);
209
210 /* stored as sec_def->cookie for all libbpf-supported SEC()s */
211 enum sec_def_flags {
212 SEC_NONE = 0,
213 /* expected_attach_type is optional, if kernel doesn't support that */
214 SEC_EXP_ATTACH_OPT = 1,
215 /* legacy, only used by libbpf_get_type_names() and
216 * libbpf_attach_type_by_name(), not used by libbpf itself at all.
217 * This used to be associated with cgroup (and few other) BPF programs
218 * that were attachable through BPF_PROG_ATTACH command. Pretty
219 * meaningless nowadays, though.
220 */
221 SEC_ATTACHABLE = 2,
222 SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT,
223 /* attachment target is specified through BTF ID in either kernel or
224 * other BPF program's BTF object */
225 SEC_ATTACH_BTF = 4,
226 /* BPF program type allows sleeping/blocking in kernel */
227 SEC_SLEEPABLE = 8,
228 /* allow non-strict prefix matching */
229 SEC_SLOPPY_PFX = 16,
230 /* BPF program support non-linear XDP buffer */
231 SEC_XDP_FRAGS = 32,
232 /* deprecated sec definitions not supposed to be used */
233 SEC_DEPRECATED = 64,
234 };
235
236 struct bpf_sec_def {
237 const char *sec;
238 enum bpf_prog_type prog_type;
239 enum bpf_attach_type expected_attach_type;
240 long cookie;
241
242 init_fn_t init_fn;
243 preload_fn_t preload_fn;
244 attach_fn_t attach_fn;
245 };
246
247 /*
248 * bpf_prog should be a better name but it has been used in
249 * linux/filter.h.
250 */
251 struct bpf_program {
252 const struct bpf_sec_def *sec_def;
253 char *sec_name;
254 size_t sec_idx;
255 /* this program's instruction offset (in number of instructions)
256 * within its containing ELF section
257 */
258 size_t sec_insn_off;
259 /* number of original instructions in ELF section belonging to this
260 * program, not taking into account subprogram instructions possible
261 * appended later during relocation
262 */
263 size_t sec_insn_cnt;
264 /* Offset (in number of instructions) of the start of instruction
265 * belonging to this BPF program within its containing main BPF
266 * program. For the entry-point (main) BPF program, this is always
267 * zero. For a sub-program, this gets reset before each of main BPF
268 * programs are processed and relocated and is used to determined
269 * whether sub-program was already appended to the main program, and
270 * if yes, at which instruction offset.
271 */
272 size_t sub_insn_off;
273
274 char *name;
275 /* name with / replaced by _; makes recursive pinning
276 * in bpf_object__pin_programs easier
277 */
278 char *pin_name;
279
280 /* instructions that belong to BPF program; insns[0] is located at
281 * sec_insn_off instruction within its ELF section in ELF file, so
282 * when mapping ELF file instruction index to the local instruction,
283 * one needs to subtract sec_insn_off; and vice versa.
284 */
285 struct bpf_insn *insns;
286 /* actual number of instruction in this BPF program's image; for
287 * entry-point BPF programs this includes the size of main program
288 * itself plus all the used sub-programs, appended at the end
289 */
290 size_t insns_cnt;
291
292 struct reloc_desc *reloc_desc;
293 int nr_reloc;
294
295 /* BPF verifier log settings */
296 char *log_buf;
297 size_t log_size;
298 __u32 log_level;
299
300 struct {
301 int nr;
302 int *fds;
303 } instances;
304 bpf_program_prep_t preprocessor;
305
306 struct bpf_object *obj;
307 void *priv;
308 bpf_program_clear_priv_t clear_priv;
309
310 bool load;
311 bool mark_btf_static;
312 enum bpf_prog_type type;
313 enum bpf_attach_type expected_attach_type;
314 int prog_ifindex;
315 __u32 attach_btf_obj_fd;
316 __u32 attach_btf_id;
317 __u32 attach_prog_fd;
318 void *func_info;
319 __u32 func_info_rec_size;
320 __u32 func_info_cnt;
321
322 void *line_info;
323 __u32 line_info_rec_size;
324 __u32 line_info_cnt;
325 __u32 prog_flags;
326 };
327
328 struct bpf_struct_ops {
329 const char *tname;
330 const struct btf_type *type;
331 struct bpf_program **progs;
332 __u32 *kern_func_off;
333 /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
334 void *data;
335 /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
336 * btf_vmlinux's format.
337 * struct bpf_struct_ops_tcp_congestion_ops {
338 * [... some other kernel fields ...]
339 * struct tcp_congestion_ops data;
340 * }
341 * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
342 * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
343 * from "data".
344 */
345 void *kern_vdata;
346 __u32 type_id;
347 };
348
349 #define DATA_SEC ".data"
350 #define BSS_SEC ".bss"
351 #define RODATA_SEC ".rodata"
352 #define KCONFIG_SEC ".kconfig"
353 #define KSYMS_SEC ".ksyms"
354 #define STRUCT_OPS_SEC ".struct_ops"
355
356 enum libbpf_map_type {
357 LIBBPF_MAP_UNSPEC,
358 LIBBPF_MAP_DATA,
359 LIBBPF_MAP_BSS,
360 LIBBPF_MAP_RODATA,
361 LIBBPF_MAP_KCONFIG,
362 };
363
364 struct bpf_map {
365 char *name;
366 /* real_name is defined for special internal maps (.rodata*,
367 * .data*, .bss, .kconfig) and preserves their original ELF section
368 * name. This is important to be be able to find corresponding BTF
369 * DATASEC information.
370 */
371 char *real_name;
372 int fd;
373 int sec_idx;
374 size_t sec_offset;
375 int map_ifindex;
376 int inner_map_fd;
377 struct bpf_map_def def;
378 __u32 numa_node;
379 __u32 btf_var_idx;
380 __u32 btf_key_type_id;
381 __u32 btf_value_type_id;
382 __u32 btf_vmlinux_value_type_id;
383 void *priv;
384 bpf_map_clear_priv_t clear_priv;
385 enum libbpf_map_type libbpf_type;
386 void *mmaped;
387 struct bpf_struct_ops *st_ops;
388 struct bpf_map *inner_map;
389 void **init_slots;
390 int init_slots_sz;
391 char *pin_path;
392 bool pinned;
393 bool reused;
394 bool skipped;
395 __u64 map_extra;
396 };
397
398 enum extern_type {
399 EXT_UNKNOWN,
400 EXT_KCFG,
401 EXT_KSYM,
402 };
403
404 enum kcfg_type {
405 KCFG_UNKNOWN,
406 KCFG_CHAR,
407 KCFG_BOOL,
408 KCFG_INT,
409 KCFG_TRISTATE,
410 KCFG_CHAR_ARR,
411 };
412
413 struct extern_desc {
414 enum extern_type type;
415 int sym_idx;
416 int btf_id;
417 int sec_btf_id;
418 const char *name;
419 bool is_set;
420 bool is_weak;
421 union {
422 struct {
423 enum kcfg_type type;
424 int sz;
425 int align;
426 int data_off;
427 bool is_signed;
428 } kcfg;
429 struct {
430 unsigned long long addr;
431
432 /* target btf_id of the corresponding kernel var. */
433 int kernel_btf_obj_fd;
434 int kernel_btf_id;
435
436 /* local btf_id of the ksym extern's type. */
437 __u32 type_id;
438 /* BTF fd index to be patched in for insn->off, this is
439 * 0 for vmlinux BTF, index in obj->fd_array for module
440 * BTF
441 */
442 __s16 btf_fd_idx;
443 } ksym;
444 };
445 };
446
447 static LIST_HEAD(bpf_objects_list);
448
449 struct module_btf {
450 struct btf *btf;
451 char *name;
452 __u32 id;
453 int fd;
454 int fd_array_idx;
455 };
456
457 enum sec_type {
458 SEC_UNUSED = 0,
459 SEC_RELO,
460 SEC_BSS,
461 SEC_DATA,
462 SEC_RODATA,
463 };
464
465 struct elf_sec_desc {
466 enum sec_type sec_type;
467 Elf64_Shdr *shdr;
468 Elf_Data *data;
469 };
470
471 struct elf_state {
472 int fd;
473 const void *obj_buf;
474 size_t obj_buf_sz;
475 Elf *elf;
476 Elf64_Ehdr *ehdr;
477 Elf_Data *symbols;
478 Elf_Data *st_ops_data;
479 size_t shstrndx; /* section index for section name strings */
480 size_t strtabidx;
481 struct elf_sec_desc *secs;
482 int sec_cnt;
483 int maps_shndx;
484 int btf_maps_shndx;
485 __u32 btf_maps_sec_btf_id;
486 int text_shndx;
487 int symbols_shndx;
488 int st_ops_shndx;
489 };
490
491 struct bpf_object {
492 char name[BPF_OBJ_NAME_LEN];
493 char license[64];
494 __u32 kern_version;
495
496 struct bpf_program *programs;
497 size_t nr_programs;
498 struct bpf_map *maps;
499 size_t nr_maps;
500 size_t maps_cap;
501
502 char *kconfig;
503 struct extern_desc *externs;
504 int nr_extern;
505 int kconfig_map_idx;
506
507 bool loaded;
508 bool has_subcalls;
509 bool has_rodata;
510
511 struct bpf_gen *gen_loader;
512
513 /* Information when doing ELF related work. Only valid if efile.elf is not NULL */
514 struct elf_state efile;
515 /*
516 * All loaded bpf_object are linked in a list, which is
517 * hidden to caller. bpf_objects__<func> handlers deal with
518 * all objects.
519 */
520 struct list_head list;
521
522 struct btf *btf;
523 struct btf_ext *btf_ext;
524
525 /* Parse and load BTF vmlinux if any of the programs in the object need
526 * it at load time.
527 */
528 struct btf *btf_vmlinux;
529 /* Path to the custom BTF to be used for BPF CO-RE relocations as an
530 * override for vmlinux BTF.
531 */
532 char *btf_custom_path;
533 /* vmlinux BTF override for CO-RE relocations */
534 struct btf *btf_vmlinux_override;
535 /* Lazily initialized kernel module BTFs */
536 struct module_btf *btf_modules;
537 bool btf_modules_loaded;
538 size_t btf_module_cnt;
539 size_t btf_module_cap;
540
541 /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */
542 char *log_buf;
543 size_t log_size;
544 __u32 log_level;
545
546 void *priv;
547 bpf_object_clear_priv_t clear_priv;
548
549 int *fd_array;
550 size_t fd_array_cap;
551 size_t fd_array_cnt;
552
553 char path[];
554 };
555
556 static const char *elf_sym_str(const struct bpf_object *obj, size_t off);
557 static const char *elf_sec_str(const struct bpf_object *obj, size_t off);
558 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx);
559 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name);
560 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn);
561 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn);
562 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn);
563 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx);
564 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx);
565
bpf_program__unload(struct bpf_program * prog)566 void bpf_program__unload(struct bpf_program *prog)
567 {
568 int i;
569
570 if (!prog)
571 return;
572
573 /*
574 * If the object is opened but the program was never loaded,
575 * it is possible that prog->instances.nr == -1.
576 */
577 if (prog->instances.nr > 0) {
578 for (i = 0; i < prog->instances.nr; i++)
579 zclose(prog->instances.fds[i]);
580 } else if (prog->instances.nr != -1) {
581 pr_warn("Internal error: instances.nr is %d\n",
582 prog->instances.nr);
583 }
584
585 prog->instances.nr = -1;
586 zfree(&prog->instances.fds);
587
588 zfree(&prog->func_info);
589 zfree(&prog->line_info);
590 }
591
bpf_program__exit(struct bpf_program * prog)592 static void bpf_program__exit(struct bpf_program *prog)
593 {
594 if (!prog)
595 return;
596
597 if (prog->clear_priv)
598 prog->clear_priv(prog, prog->priv);
599
600 prog->priv = NULL;
601 prog->clear_priv = NULL;
602
603 bpf_program__unload(prog);
604 zfree(&prog->name);
605 zfree(&prog->sec_name);
606 zfree(&prog->pin_name);
607 zfree(&prog->insns);
608 zfree(&prog->reloc_desc);
609
610 prog->nr_reloc = 0;
611 prog->insns_cnt = 0;
612 prog->sec_idx = -1;
613 }
614
__bpf_program__pin_name(struct bpf_program * prog)615 static char *__bpf_program__pin_name(struct bpf_program *prog)
616 {
617 char *name, *p;
618
619 if (libbpf_mode & LIBBPF_STRICT_SEC_NAME)
620 name = strdup(prog->name);
621 else
622 name = strdup(prog->sec_name);
623
624 if (!name)
625 return NULL;
626
627 p = name;
628
629 while ((p = strchr(p, '/')))
630 *p = '_';
631
632 return name;
633 }
634
insn_is_subprog_call(const struct bpf_insn * insn)635 static bool insn_is_subprog_call(const struct bpf_insn *insn)
636 {
637 return BPF_CLASS(insn->code) == BPF_JMP &&
638 BPF_OP(insn->code) == BPF_CALL &&
639 BPF_SRC(insn->code) == BPF_K &&
640 insn->src_reg == BPF_PSEUDO_CALL &&
641 insn->dst_reg == 0 &&
642 insn->off == 0;
643 }
644
is_call_insn(const struct bpf_insn * insn)645 static bool is_call_insn(const struct bpf_insn *insn)
646 {
647 return insn->code == (BPF_JMP | BPF_CALL);
648 }
649
insn_is_pseudo_func(struct bpf_insn * insn)650 static bool insn_is_pseudo_func(struct bpf_insn *insn)
651 {
652 return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC;
653 }
654
655 static int
bpf_object__init_prog(struct bpf_object * obj,struct bpf_program * prog,const char * name,size_t sec_idx,const char * sec_name,size_t sec_off,void * insn_data,size_t insn_data_sz)656 bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog,
657 const char *name, size_t sec_idx, const char *sec_name,
658 size_t sec_off, void *insn_data, size_t insn_data_sz)
659 {
660 if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) {
661 pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n",
662 sec_name, name, sec_off, insn_data_sz);
663 return -EINVAL;
664 }
665
666 memset(prog, 0, sizeof(*prog));
667 prog->obj = obj;
668
669 prog->sec_idx = sec_idx;
670 prog->sec_insn_off = sec_off / BPF_INSN_SZ;
671 prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ;
672 /* insns_cnt can later be increased by appending used subprograms */
673 prog->insns_cnt = prog->sec_insn_cnt;
674
675 prog->type = BPF_PROG_TYPE_UNSPEC;
676 prog->load = true;
677
678 prog->instances.fds = NULL;
679 prog->instances.nr = -1;
680
681 /* inherit object's log_level */
682 prog->log_level = obj->log_level;
683
684 prog->sec_name = strdup(sec_name);
685 if (!prog->sec_name)
686 goto errout;
687
688 prog->name = strdup(name);
689 if (!prog->name)
690 goto errout;
691
692 prog->pin_name = __bpf_program__pin_name(prog);
693 if (!prog->pin_name)
694 goto errout;
695
696 prog->insns = malloc(insn_data_sz);
697 if (!prog->insns)
698 goto errout;
699 memcpy(prog->insns, insn_data, insn_data_sz);
700
701 return 0;
702 errout:
703 pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name);
704 bpf_program__exit(prog);
705 return -ENOMEM;
706 }
707
708 static int
bpf_object__add_programs(struct bpf_object * obj,Elf_Data * sec_data,const char * sec_name,int sec_idx)709 bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data,
710 const char *sec_name, int sec_idx)
711 {
712 Elf_Data *symbols = obj->efile.symbols;
713 struct bpf_program *prog, *progs;
714 void *data = sec_data->d_buf;
715 size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms;
716 int nr_progs, err, i;
717 const char *name;
718 Elf64_Sym *sym;
719
720 progs = obj->programs;
721 nr_progs = obj->nr_programs;
722 nr_syms = symbols->d_size / sizeof(Elf64_Sym);
723 sec_off = 0;
724
725 for (i = 0; i < nr_syms; i++) {
726 sym = elf_sym_by_idx(obj, i);
727
728 if (sym->st_shndx != sec_idx)
729 continue;
730 if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC)
731 continue;
732
733 prog_sz = sym->st_size;
734 sec_off = sym->st_value;
735
736 name = elf_sym_str(obj, sym->st_name);
737 if (!name) {
738 pr_warn("sec '%s': failed to get symbol name for offset %zu\n",
739 sec_name, sec_off);
740 return -LIBBPF_ERRNO__FORMAT;
741 }
742
743 if (sec_off + prog_sz > sec_sz) {
744 pr_warn("sec '%s': program at offset %zu crosses section boundary\n",
745 sec_name, sec_off);
746 return -LIBBPF_ERRNO__FORMAT;
747 }
748
749 if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) {
750 pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name);
751 return -ENOTSUP;
752 }
753
754 pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n",
755 sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz);
756
757 progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs));
758 if (!progs) {
759 /*
760 * In this case the original obj->programs
761 * is still valid, so don't need special treat for
762 * bpf_close_object().
763 */
764 pr_warn("sec '%s': failed to alloc memory for new program '%s'\n",
765 sec_name, name);
766 return -ENOMEM;
767 }
768 obj->programs = progs;
769
770 prog = &progs[nr_progs];
771
772 err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name,
773 sec_off, data + sec_off, prog_sz);
774 if (err)
775 return err;
776
777 /* if function is a global/weak symbol, but has restricted
778 * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC
779 * as static to enable more permissive BPF verification mode
780 * with more outside context available to BPF verifier
781 */
782 if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL
783 && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
784 || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL))
785 prog->mark_btf_static = true;
786
787 nr_progs++;
788 obj->nr_programs = nr_progs;
789 }
790
791 return 0;
792 }
793
get_kernel_version(void)794 __u32 get_kernel_version(void)
795 {
796 /* On Ubuntu LINUX_VERSION_CODE doesn't correspond to info.release,
797 * but Ubuntu provides /proc/version_signature file, as described at
798 * https://ubuntu.com/kernel, with an example contents below, which we
799 * can use to get a proper LINUX_VERSION_CODE.
800 *
801 * Ubuntu 5.4.0-12.15-generic 5.4.8
802 *
803 * In the above, 5.4.8 is what kernel is actually expecting, while
804 * uname() call will return 5.4.0 in info.release.
805 */
806 const char *ubuntu_kver_file = "/proc/version_signature";
807 __u32 major, minor, patch;
808 struct utsname info;
809
810 if (access(ubuntu_kver_file, R_OK) == 0) {
811 FILE *f;
812
813 f = fopen(ubuntu_kver_file, "r");
814 if (f) {
815 if (fscanf(f, "%*s %*s %d.%d.%d\n", &major, &minor, &patch) == 3) {
816 fclose(f);
817 return KERNEL_VERSION(major, minor, patch);
818 }
819 fclose(f);
820 }
821 /* something went wrong, fall back to uname() approach */
822 }
823
824 uname(&info);
825 if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
826 return 0;
827 return KERNEL_VERSION(major, minor, patch);
828 }
829
830 static const struct btf_member *
find_member_by_offset(const struct btf_type * t,__u32 bit_offset)831 find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
832 {
833 struct btf_member *m;
834 int i;
835
836 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
837 if (btf_member_bit_offset(t, i) == bit_offset)
838 return m;
839 }
840
841 return NULL;
842 }
843
844 static const struct btf_member *
find_member_by_name(const struct btf * btf,const struct btf_type * t,const char * name)845 find_member_by_name(const struct btf *btf, const struct btf_type *t,
846 const char *name)
847 {
848 struct btf_member *m;
849 int i;
850
851 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
852 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
853 return m;
854 }
855
856 return NULL;
857 }
858
859 #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
860 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
861 const char *name, __u32 kind);
862
863 static int
find_struct_ops_kern_types(const struct btf * btf,const char * tname,const struct btf_type ** type,__u32 * type_id,const struct btf_type ** vtype,__u32 * vtype_id,const struct btf_member ** data_member)864 find_struct_ops_kern_types(const struct btf *btf, const char *tname,
865 const struct btf_type **type, __u32 *type_id,
866 const struct btf_type **vtype, __u32 *vtype_id,
867 const struct btf_member **data_member)
868 {
869 const struct btf_type *kern_type, *kern_vtype;
870 const struct btf_member *kern_data_member;
871 __s32 kern_vtype_id, kern_type_id;
872 __u32 i;
873
874 kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
875 if (kern_type_id < 0) {
876 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
877 tname);
878 return kern_type_id;
879 }
880 kern_type = btf__type_by_id(btf, kern_type_id);
881
882 /* Find the corresponding "map_value" type that will be used
883 * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example,
884 * find "struct bpf_struct_ops_tcp_congestion_ops" from the
885 * btf_vmlinux.
886 */
887 kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
888 tname, BTF_KIND_STRUCT);
889 if (kern_vtype_id < 0) {
890 pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
891 STRUCT_OPS_VALUE_PREFIX, tname);
892 return kern_vtype_id;
893 }
894 kern_vtype = btf__type_by_id(btf, kern_vtype_id);
895
896 /* Find "struct tcp_congestion_ops" from
897 * struct bpf_struct_ops_tcp_congestion_ops {
898 * [ ... ]
899 * struct tcp_congestion_ops data;
900 * }
901 */
902 kern_data_member = btf_members(kern_vtype);
903 for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
904 if (kern_data_member->type == kern_type_id)
905 break;
906 }
907 if (i == btf_vlen(kern_vtype)) {
908 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
909 tname, STRUCT_OPS_VALUE_PREFIX, tname);
910 return -EINVAL;
911 }
912
913 *type = kern_type;
914 *type_id = kern_type_id;
915 *vtype = kern_vtype;
916 *vtype_id = kern_vtype_id;
917 *data_member = kern_data_member;
918
919 return 0;
920 }
921
bpf_map__is_struct_ops(const struct bpf_map * map)922 static bool bpf_map__is_struct_ops(const struct bpf_map *map)
923 {
924 return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
925 }
926
927 /* Init the map's fields that depend on kern_btf */
bpf_map__init_kern_struct_ops(struct bpf_map * map,const struct btf * btf,const struct btf * kern_btf)928 static int bpf_map__init_kern_struct_ops(struct bpf_map *map,
929 const struct btf *btf,
930 const struct btf *kern_btf)
931 {
932 const struct btf_member *member, *kern_member, *kern_data_member;
933 const struct btf_type *type, *kern_type, *kern_vtype;
934 __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
935 struct bpf_struct_ops *st_ops;
936 void *data, *kern_data;
937 const char *tname;
938 int err;
939
940 st_ops = map->st_ops;
941 type = st_ops->type;
942 tname = st_ops->tname;
943 err = find_struct_ops_kern_types(kern_btf, tname,
944 &kern_type, &kern_type_id,
945 &kern_vtype, &kern_vtype_id,
946 &kern_data_member);
947 if (err)
948 return err;
949
950 pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
951 map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
952
953 map->def.value_size = kern_vtype->size;
954 map->btf_vmlinux_value_type_id = kern_vtype_id;
955
956 st_ops->kern_vdata = calloc(1, kern_vtype->size);
957 if (!st_ops->kern_vdata)
958 return -ENOMEM;
959
960 data = st_ops->data;
961 kern_data_off = kern_data_member->offset / 8;
962 kern_data = st_ops->kern_vdata + kern_data_off;
963
964 member = btf_members(type);
965 for (i = 0; i < btf_vlen(type); i++, member++) {
966 const struct btf_type *mtype, *kern_mtype;
967 __u32 mtype_id, kern_mtype_id;
968 void *mdata, *kern_mdata;
969 __s64 msize, kern_msize;
970 __u32 moff, kern_moff;
971 __u32 kern_member_idx;
972 const char *mname;
973
974 mname = btf__name_by_offset(btf, member->name_off);
975 kern_member = find_member_by_name(kern_btf, kern_type, mname);
976 if (!kern_member) {
977 pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
978 map->name, mname);
979 return -ENOTSUP;
980 }
981
982 kern_member_idx = kern_member - btf_members(kern_type);
983 if (btf_member_bitfield_size(type, i) ||
984 btf_member_bitfield_size(kern_type, kern_member_idx)) {
985 pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
986 map->name, mname);
987 return -ENOTSUP;
988 }
989
990 moff = member->offset / 8;
991 kern_moff = kern_member->offset / 8;
992
993 mdata = data + moff;
994 kern_mdata = kern_data + kern_moff;
995
996 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
997 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
998 &kern_mtype_id);
999 if (BTF_INFO_KIND(mtype->info) !=
1000 BTF_INFO_KIND(kern_mtype->info)) {
1001 pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
1002 map->name, mname, BTF_INFO_KIND(mtype->info),
1003 BTF_INFO_KIND(kern_mtype->info));
1004 return -ENOTSUP;
1005 }
1006
1007 if (btf_is_ptr(mtype)) {
1008 struct bpf_program *prog;
1009
1010 prog = st_ops->progs[i];
1011 if (!prog)
1012 continue;
1013
1014 kern_mtype = skip_mods_and_typedefs(kern_btf,
1015 kern_mtype->type,
1016 &kern_mtype_id);
1017
1018 /* mtype->type must be a func_proto which was
1019 * guaranteed in bpf_object__collect_st_ops_relos(),
1020 * so only check kern_mtype for func_proto here.
1021 */
1022 if (!btf_is_func_proto(kern_mtype)) {
1023 pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n",
1024 map->name, mname);
1025 return -ENOTSUP;
1026 }
1027
1028 prog->attach_btf_id = kern_type_id;
1029 prog->expected_attach_type = kern_member_idx;
1030
1031 st_ops->kern_func_off[i] = kern_data_off + kern_moff;
1032
1033 pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
1034 map->name, mname, prog->name, moff,
1035 kern_moff);
1036
1037 continue;
1038 }
1039
1040 msize = btf__resolve_size(btf, mtype_id);
1041 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
1042 if (msize < 0 || kern_msize < 0 || msize != kern_msize) {
1043 pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
1044 map->name, mname, (ssize_t)msize,
1045 (ssize_t)kern_msize);
1046 return -ENOTSUP;
1047 }
1048
1049 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
1050 map->name, mname, (unsigned int)msize,
1051 moff, kern_moff);
1052 memcpy(kern_mdata, mdata, msize);
1053 }
1054
1055 return 0;
1056 }
1057
bpf_object__init_kern_struct_ops_maps(struct bpf_object * obj)1058 static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
1059 {
1060 struct bpf_map *map;
1061 size_t i;
1062 int err;
1063
1064 for (i = 0; i < obj->nr_maps; i++) {
1065 map = &obj->maps[i];
1066
1067 if (!bpf_map__is_struct_ops(map))
1068 continue;
1069
1070 err = bpf_map__init_kern_struct_ops(map, obj->btf,
1071 obj->btf_vmlinux);
1072 if (err)
1073 return err;
1074 }
1075
1076 return 0;
1077 }
1078
bpf_object__init_struct_ops_maps(struct bpf_object * obj)1079 static int bpf_object__init_struct_ops_maps(struct bpf_object *obj)
1080 {
1081 const struct btf_type *type, *datasec;
1082 const struct btf_var_secinfo *vsi;
1083 struct bpf_struct_ops *st_ops;
1084 const char *tname, *var_name;
1085 __s32 type_id, datasec_id;
1086 const struct btf *btf;
1087 struct bpf_map *map;
1088 __u32 i;
1089
1090 if (obj->efile.st_ops_shndx == -1)
1091 return 0;
1092
1093 btf = obj->btf;
1094 datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC,
1095 BTF_KIND_DATASEC);
1096 if (datasec_id < 0) {
1097 pr_warn("struct_ops init: DATASEC %s not found\n",
1098 STRUCT_OPS_SEC);
1099 return -EINVAL;
1100 }
1101
1102 datasec = btf__type_by_id(btf, datasec_id);
1103 vsi = btf_var_secinfos(datasec);
1104 for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
1105 type = btf__type_by_id(obj->btf, vsi->type);
1106 var_name = btf__name_by_offset(obj->btf, type->name_off);
1107
1108 type_id = btf__resolve_type(obj->btf, vsi->type);
1109 if (type_id < 0) {
1110 pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
1111 vsi->type, STRUCT_OPS_SEC);
1112 return -EINVAL;
1113 }
1114
1115 type = btf__type_by_id(obj->btf, type_id);
1116 tname = btf__name_by_offset(obj->btf, type->name_off);
1117 if (!tname[0]) {
1118 pr_warn("struct_ops init: anonymous type is not supported\n");
1119 return -ENOTSUP;
1120 }
1121 if (!btf_is_struct(type)) {
1122 pr_warn("struct_ops init: %s is not a struct\n", tname);
1123 return -EINVAL;
1124 }
1125
1126 map = bpf_object__add_map(obj);
1127 if (IS_ERR(map))
1128 return PTR_ERR(map);
1129
1130 map->sec_idx = obj->efile.st_ops_shndx;
1131 map->sec_offset = vsi->offset;
1132 map->name = strdup(var_name);
1133 if (!map->name)
1134 return -ENOMEM;
1135
1136 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
1137 map->def.key_size = sizeof(int);
1138 map->def.value_size = type->size;
1139 map->def.max_entries = 1;
1140
1141 map->st_ops = calloc(1, sizeof(*map->st_ops));
1142 if (!map->st_ops)
1143 return -ENOMEM;
1144 st_ops = map->st_ops;
1145 st_ops->data = malloc(type->size);
1146 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
1147 st_ops->kern_func_off = malloc(btf_vlen(type) *
1148 sizeof(*st_ops->kern_func_off));
1149 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
1150 return -ENOMEM;
1151
1152 if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) {
1153 pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
1154 var_name, STRUCT_OPS_SEC);
1155 return -EINVAL;
1156 }
1157
1158 memcpy(st_ops->data,
1159 obj->efile.st_ops_data->d_buf + vsi->offset,
1160 type->size);
1161 st_ops->tname = tname;
1162 st_ops->type = type;
1163 st_ops->type_id = type_id;
1164
1165 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
1166 tname, type_id, var_name, vsi->offset);
1167 }
1168
1169 return 0;
1170 }
1171
bpf_object__new(const char * path,const void * obj_buf,size_t obj_buf_sz,const char * obj_name)1172 static struct bpf_object *bpf_object__new(const char *path,
1173 const void *obj_buf,
1174 size_t obj_buf_sz,
1175 const char *obj_name)
1176 {
1177 bool strict = (libbpf_mode & LIBBPF_STRICT_NO_OBJECT_LIST);
1178 struct bpf_object *obj;
1179 char *end;
1180
1181 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
1182 if (!obj) {
1183 pr_warn("alloc memory failed for %s\n", path);
1184 return ERR_PTR(-ENOMEM);
1185 }
1186
1187 strcpy(obj->path, path);
1188 if (obj_name) {
1189 libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name));
1190 } else {
1191 /* Using basename() GNU version which doesn't modify arg. */
1192 libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name));
1193 end = strchr(obj->name, '.');
1194 if (end)
1195 *end = 0;
1196 }
1197
1198 obj->efile.fd = -1;
1199 /*
1200 * Caller of this function should also call
1201 * bpf_object__elf_finish() after data collection to return
1202 * obj_buf to user. If not, we should duplicate the buffer to
1203 * avoid user freeing them before elf finish.
1204 */
1205 obj->efile.obj_buf = obj_buf;
1206 obj->efile.obj_buf_sz = obj_buf_sz;
1207 obj->efile.maps_shndx = -1;
1208 obj->efile.btf_maps_shndx = -1;
1209 obj->efile.st_ops_shndx = -1;
1210 obj->kconfig_map_idx = -1;
1211
1212 obj->kern_version = get_kernel_version();
1213 obj->loaded = false;
1214
1215 INIT_LIST_HEAD(&obj->list);
1216 if (!strict)
1217 list_add(&obj->list, &bpf_objects_list);
1218 return obj;
1219 }
1220
bpf_object__elf_finish(struct bpf_object * obj)1221 static void bpf_object__elf_finish(struct bpf_object *obj)
1222 {
1223 if (!obj->efile.elf)
1224 return;
1225
1226 if (obj->efile.elf) {
1227 elf_end(obj->efile.elf);
1228 obj->efile.elf = NULL;
1229 }
1230 obj->efile.symbols = NULL;
1231 obj->efile.st_ops_data = NULL;
1232
1233 zfree(&obj->efile.secs);
1234 obj->efile.sec_cnt = 0;
1235 zclose(obj->efile.fd);
1236 obj->efile.obj_buf = NULL;
1237 obj->efile.obj_buf_sz = 0;
1238 }
1239
bpf_object__elf_init(struct bpf_object * obj)1240 static int bpf_object__elf_init(struct bpf_object *obj)
1241 {
1242 Elf64_Ehdr *ehdr;
1243 int err = 0;
1244 Elf *elf;
1245
1246 if (obj->efile.elf) {
1247 pr_warn("elf: init internal error\n");
1248 return -LIBBPF_ERRNO__LIBELF;
1249 }
1250
1251 if (obj->efile.obj_buf_sz > 0) {
1252 /*
1253 * obj_buf should have been validated by
1254 * bpf_object__open_buffer().
1255 */
1256 elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz);
1257 } else {
1258 obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC);
1259 if (obj->efile.fd < 0) {
1260 char errmsg[STRERR_BUFSIZE], *cp;
1261
1262 err = -errno;
1263 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
1264 pr_warn("elf: failed to open %s: %s\n", obj->path, cp);
1265 return err;
1266 }
1267
1268 elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL);
1269 }
1270
1271 if (!elf) {
1272 pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1));
1273 err = -LIBBPF_ERRNO__LIBELF;
1274 goto errout;
1275 }
1276
1277 obj->efile.elf = elf;
1278
1279 if (elf_kind(elf) != ELF_K_ELF) {
1280 err = -LIBBPF_ERRNO__FORMAT;
1281 pr_warn("elf: '%s' is not a proper ELF object\n", obj->path);
1282 goto errout;
1283 }
1284
1285 if (gelf_getclass(elf) != ELFCLASS64) {
1286 err = -LIBBPF_ERRNO__FORMAT;
1287 pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path);
1288 goto errout;
1289 }
1290
1291 obj->efile.ehdr = ehdr = elf64_getehdr(elf);
1292 if (!obj->efile.ehdr) {
1293 pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1));
1294 err = -LIBBPF_ERRNO__FORMAT;
1295 goto errout;
1296 }
1297
1298 if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) {
1299 pr_warn("elf: failed to get section names section index for %s: %s\n",
1300 obj->path, elf_errmsg(-1));
1301 err = -LIBBPF_ERRNO__FORMAT;
1302 goto errout;
1303 }
1304
1305 /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1306 if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) {
1307 pr_warn("elf: failed to get section names strings from %s: %s\n",
1308 obj->path, elf_errmsg(-1));
1309 err = -LIBBPF_ERRNO__FORMAT;
1310 goto errout;
1311 }
1312
1313 /* Old LLVM set e_machine to EM_NONE */
1314 if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) {
1315 pr_warn("elf: %s is not a valid eBPF object file\n", obj->path);
1316 err = -LIBBPF_ERRNO__FORMAT;
1317 goto errout;
1318 }
1319
1320 return 0;
1321 errout:
1322 bpf_object__elf_finish(obj);
1323 return err;
1324 }
1325
bpf_object__check_endianness(struct bpf_object * obj)1326 static int bpf_object__check_endianness(struct bpf_object *obj)
1327 {
1328 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1329 if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB)
1330 return 0;
1331 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1332 if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB)
1333 return 0;
1334 #else
1335 # error "Unrecognized __BYTE_ORDER__"
1336 #endif
1337 pr_warn("elf: endianness mismatch in %s.\n", obj->path);
1338 return -LIBBPF_ERRNO__ENDIAN;
1339 }
1340
1341 static int
bpf_object__init_license(struct bpf_object * obj,void * data,size_t size)1342 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1343 {
1344 /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't
1345 * go over allowed ELF data section buffer
1346 */
1347 libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license)));
1348 pr_debug("license of %s is %s\n", obj->path, obj->license);
1349 return 0;
1350 }
1351
1352 static int
bpf_object__init_kversion(struct bpf_object * obj,void * data,size_t size)1353 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1354 {
1355 __u32 kver;
1356
1357 if (size != sizeof(kver)) {
1358 pr_warn("invalid kver section in %s\n", obj->path);
1359 return -LIBBPF_ERRNO__FORMAT;
1360 }
1361 memcpy(&kver, data, sizeof(kver));
1362 obj->kern_version = kver;
1363 pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1364 return 0;
1365 }
1366
bpf_map_type__is_map_in_map(enum bpf_map_type type)1367 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1368 {
1369 if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1370 type == BPF_MAP_TYPE_HASH_OF_MAPS)
1371 return true;
1372 return false;
1373 }
1374
find_elf_sec_sz(const struct bpf_object * obj,const char * name,__u32 * size)1375 static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size)
1376 {
1377 int ret = -ENOENT;
1378 Elf_Data *data;
1379 Elf_Scn *scn;
1380
1381 *size = 0;
1382 if (!name)
1383 return -EINVAL;
1384
1385 scn = elf_sec_by_name(obj, name);
1386 data = elf_sec_data(obj, scn);
1387 if (data) {
1388 ret = 0; /* found it */
1389 *size = data->d_size;
1390 }
1391
1392 return *size ? 0 : ret;
1393 }
1394
find_elf_var_offset(const struct bpf_object * obj,const char * name,__u32 * off)1395 static int find_elf_var_offset(const struct bpf_object *obj, const char *name, __u32 *off)
1396 {
1397 Elf_Data *symbols = obj->efile.symbols;
1398 const char *sname;
1399 size_t si;
1400
1401 if (!name || !off)
1402 return -EINVAL;
1403
1404 for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) {
1405 Elf64_Sym *sym = elf_sym_by_idx(obj, si);
1406
1407 if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL ||
1408 ELF64_ST_TYPE(sym->st_info) != STT_OBJECT)
1409 continue;
1410
1411 sname = elf_sym_str(obj, sym->st_name);
1412 if (!sname) {
1413 pr_warn("failed to get sym name string for var %s\n", name);
1414 return -EIO;
1415 }
1416 if (strcmp(name, sname) == 0) {
1417 *off = sym->st_value;
1418 return 0;
1419 }
1420 }
1421
1422 return -ENOENT;
1423 }
1424
bpf_object__add_map(struct bpf_object * obj)1425 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1426 {
1427 struct bpf_map *new_maps;
1428 size_t new_cap;
1429 int i;
1430
1431 if (obj->nr_maps < obj->maps_cap)
1432 return &obj->maps[obj->nr_maps++];
1433
1434 new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
1435 new_maps = libbpf_reallocarray(obj->maps, new_cap, sizeof(*obj->maps));
1436 if (!new_maps) {
1437 pr_warn("alloc maps for object failed\n");
1438 return ERR_PTR(-ENOMEM);
1439 }
1440
1441 obj->maps_cap = new_cap;
1442 obj->maps = new_maps;
1443
1444 /* zero out new maps */
1445 memset(obj->maps + obj->nr_maps, 0,
1446 (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
1447 /*
1448 * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
1449 * when failure (zclose won't close negative fd)).
1450 */
1451 for (i = obj->nr_maps; i < obj->maps_cap; i++) {
1452 obj->maps[i].fd = -1;
1453 obj->maps[i].inner_map_fd = -1;
1454 }
1455
1456 return &obj->maps[obj->nr_maps++];
1457 }
1458
bpf_map_mmap_sz(const struct bpf_map * map)1459 static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1460 {
1461 long page_sz = sysconf(_SC_PAGE_SIZE);
1462 size_t map_sz;
1463
1464 map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries;
1465 map_sz = roundup(map_sz, page_sz);
1466 return map_sz;
1467 }
1468
internal_map_name(struct bpf_object * obj,const char * real_name)1469 static char *internal_map_name(struct bpf_object *obj, const char *real_name)
1470 {
1471 char map_name[BPF_OBJ_NAME_LEN], *p;
1472 int pfx_len, sfx_len = max((size_t)7, strlen(real_name));
1473
1474 /* This is one of the more confusing parts of libbpf for various
1475 * reasons, some of which are historical. The original idea for naming
1476 * internal names was to include as much of BPF object name prefix as
1477 * possible, so that it can be distinguished from similar internal
1478 * maps of a different BPF object.
1479 * As an example, let's say we have bpf_object named 'my_object_name'
1480 * and internal map corresponding to '.rodata' ELF section. The final
1481 * map name advertised to user and to the kernel will be
1482 * 'my_objec.rodata', taking first 8 characters of object name and
1483 * entire 7 characters of '.rodata'.
1484 * Somewhat confusingly, if internal map ELF section name is shorter
1485 * than 7 characters, e.g., '.bss', we still reserve 7 characters
1486 * for the suffix, even though we only have 4 actual characters, and
1487 * resulting map will be called 'my_objec.bss', not even using all 15
1488 * characters allowed by the kernel. Oh well, at least the truncated
1489 * object name is somewhat consistent in this case. But if the map
1490 * name is '.kconfig', we'll still have entirety of '.kconfig' added
1491 * (8 chars) and thus will be left with only first 7 characters of the
1492 * object name ('my_obje'). Happy guessing, user, that the final map
1493 * name will be "my_obje.kconfig".
1494 * Now, with libbpf starting to support arbitrarily named .rodata.*
1495 * and .data.* data sections, it's possible that ELF section name is
1496 * longer than allowed 15 chars, so we now need to be careful to take
1497 * only up to 15 first characters of ELF name, taking no BPF object
1498 * name characters at all. So '.rodata.abracadabra' will result in
1499 * '.rodata.abracad' kernel and user-visible name.
1500 * We need to keep this convoluted logic intact for .data, .bss and
1501 * .rodata maps, but for new custom .data.custom and .rodata.custom
1502 * maps we use their ELF names as is, not prepending bpf_object name
1503 * in front. We still need to truncate them to 15 characters for the
1504 * kernel. Full name can be recovered for such maps by using DATASEC
1505 * BTF type associated with such map's value type, though.
1506 */
1507 if (sfx_len >= BPF_OBJ_NAME_LEN)
1508 sfx_len = BPF_OBJ_NAME_LEN - 1;
1509
1510 /* if there are two or more dots in map name, it's a custom dot map */
1511 if (strchr(real_name + 1, '.') != NULL)
1512 pfx_len = 0;
1513 else
1514 pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name));
1515
1516 snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1517 sfx_len, real_name);
1518
1519 /* sanitise map name to characters allowed by kernel */
1520 for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1521 if (!isalnum(*p) && *p != '_' && *p != '.')
1522 *p = '_';
1523
1524 return strdup(map_name);
1525 }
1526
1527 static int
bpf_object__init_internal_map(struct bpf_object * obj,enum libbpf_map_type type,const char * real_name,int sec_idx,void * data,size_t data_sz)1528 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1529 const char *real_name, int sec_idx, void *data, size_t data_sz)
1530 {
1531 struct bpf_map_def *def;
1532 struct bpf_map *map;
1533 int err;
1534
1535 map = bpf_object__add_map(obj);
1536 if (IS_ERR(map))
1537 return PTR_ERR(map);
1538
1539 map->libbpf_type = type;
1540 map->sec_idx = sec_idx;
1541 map->sec_offset = 0;
1542 map->real_name = strdup(real_name);
1543 map->name = internal_map_name(obj, real_name);
1544 if (!map->real_name || !map->name) {
1545 zfree(&map->real_name);
1546 zfree(&map->name);
1547 return -ENOMEM;
1548 }
1549
1550 def = &map->def;
1551 def->type = BPF_MAP_TYPE_ARRAY;
1552 def->key_size = sizeof(int);
1553 def->value_size = data_sz;
1554 def->max_entries = 1;
1555 def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1556 ? BPF_F_RDONLY_PROG : 0;
1557 def->map_flags |= BPF_F_MMAPABLE;
1558
1559 pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1560 map->name, map->sec_idx, map->sec_offset, def->map_flags);
1561
1562 map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
1563 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1564 if (map->mmaped == MAP_FAILED) {
1565 err = -errno;
1566 map->mmaped = NULL;
1567 pr_warn("failed to alloc map '%s' content buffer: %d\n",
1568 map->name, err);
1569 zfree(&map->real_name);
1570 zfree(&map->name);
1571 return err;
1572 }
1573
1574 if (data)
1575 memcpy(map->mmaped, data, data_sz);
1576
1577 pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
1578 return 0;
1579 }
1580
bpf_object__init_global_data_maps(struct bpf_object * obj)1581 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1582 {
1583 struct elf_sec_desc *sec_desc;
1584 const char *sec_name;
1585 int err = 0, sec_idx;
1586
1587 /*
1588 * Populate obj->maps with libbpf internal maps.
1589 */
1590 for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) {
1591 sec_desc = &obj->efile.secs[sec_idx];
1592
1593 switch (sec_desc->sec_type) {
1594 case SEC_DATA:
1595 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1596 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
1597 sec_name, sec_idx,
1598 sec_desc->data->d_buf,
1599 sec_desc->data->d_size);
1600 break;
1601 case SEC_RODATA:
1602 obj->has_rodata = true;
1603 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1604 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
1605 sec_name, sec_idx,
1606 sec_desc->data->d_buf,
1607 sec_desc->data->d_size);
1608 break;
1609 case SEC_BSS:
1610 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1611 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
1612 sec_name, sec_idx,
1613 NULL,
1614 sec_desc->data->d_size);
1615 break;
1616 default:
1617 /* skip */
1618 break;
1619 }
1620 if (err)
1621 return err;
1622 }
1623 return 0;
1624 }
1625
1626
find_extern_by_name(const struct bpf_object * obj,const void * name)1627 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1628 const void *name)
1629 {
1630 int i;
1631
1632 for (i = 0; i < obj->nr_extern; i++) {
1633 if (strcmp(obj->externs[i].name, name) == 0)
1634 return &obj->externs[i];
1635 }
1636 return NULL;
1637 }
1638
set_kcfg_value_tri(struct extern_desc * ext,void * ext_val,char value)1639 static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val,
1640 char value)
1641 {
1642 switch (ext->kcfg.type) {
1643 case KCFG_BOOL:
1644 if (value == 'm') {
1645 pr_warn("extern (kcfg) %s=%c should be tristate or char\n",
1646 ext->name, value);
1647 return -EINVAL;
1648 }
1649 *(bool *)ext_val = value == 'y' ? true : false;
1650 break;
1651 case KCFG_TRISTATE:
1652 if (value == 'y')
1653 *(enum libbpf_tristate *)ext_val = TRI_YES;
1654 else if (value == 'm')
1655 *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1656 else /* value == 'n' */
1657 *(enum libbpf_tristate *)ext_val = TRI_NO;
1658 break;
1659 case KCFG_CHAR:
1660 *(char *)ext_val = value;
1661 break;
1662 case KCFG_UNKNOWN:
1663 case KCFG_INT:
1664 case KCFG_CHAR_ARR:
1665 default:
1666 pr_warn("extern (kcfg) %s=%c should be bool, tristate, or char\n",
1667 ext->name, value);
1668 return -EINVAL;
1669 }
1670 ext->is_set = true;
1671 return 0;
1672 }
1673
set_kcfg_value_str(struct extern_desc * ext,char * ext_val,const char * value)1674 static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val,
1675 const char *value)
1676 {
1677 size_t len;
1678
1679 if (ext->kcfg.type != KCFG_CHAR_ARR) {
1680 pr_warn("extern (kcfg) %s=%s should be char array\n", ext->name, value);
1681 return -EINVAL;
1682 }
1683
1684 len = strlen(value);
1685 if (value[len - 1] != '"') {
1686 pr_warn("extern (kcfg) '%s': invalid string config '%s'\n",
1687 ext->name, value);
1688 return -EINVAL;
1689 }
1690
1691 /* strip quotes */
1692 len -= 2;
1693 if (len >= ext->kcfg.sz) {
1694 pr_warn("extern (kcfg) '%s': long string config %s of (%zu bytes) truncated to %d bytes\n",
1695 ext->name, value, len, ext->kcfg.sz - 1);
1696 len = ext->kcfg.sz - 1;
1697 }
1698 memcpy(ext_val, value + 1, len);
1699 ext_val[len] = '\0';
1700 ext->is_set = true;
1701 return 0;
1702 }
1703
parse_u64(const char * value,__u64 * res)1704 static int parse_u64(const char *value, __u64 *res)
1705 {
1706 char *value_end;
1707 int err;
1708
1709 errno = 0;
1710 *res = strtoull(value, &value_end, 0);
1711 if (errno) {
1712 err = -errno;
1713 pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1714 return err;
1715 }
1716 if (*value_end) {
1717 pr_warn("failed to parse '%s' as integer completely\n", value);
1718 return -EINVAL;
1719 }
1720 return 0;
1721 }
1722
is_kcfg_value_in_range(const struct extern_desc * ext,__u64 v)1723 static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v)
1724 {
1725 int bit_sz = ext->kcfg.sz * 8;
1726
1727 if (ext->kcfg.sz == 8)
1728 return true;
1729
1730 /* Validate that value stored in u64 fits in integer of `ext->sz`
1731 * bytes size without any loss of information. If the target integer
1732 * is signed, we rely on the following limits of integer type of
1733 * Y bits and subsequent transformation:
1734 *
1735 * -2^(Y-1) <= X <= 2^(Y-1) - 1
1736 * 0 <= X + 2^(Y-1) <= 2^Y - 1
1737 * 0 <= X + 2^(Y-1) < 2^Y
1738 *
1739 * For unsigned target integer, check that all the (64 - Y) bits are
1740 * zero.
1741 */
1742 if (ext->kcfg.is_signed)
1743 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1744 else
1745 return (v >> bit_sz) == 0;
1746 }
1747
set_kcfg_value_num(struct extern_desc * ext,void * ext_val,__u64 value)1748 static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val,
1749 __u64 value)
1750 {
1751 if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) {
1752 pr_warn("extern (kcfg) %s=%llu should be integer\n",
1753 ext->name, (unsigned long long)value);
1754 return -EINVAL;
1755 }
1756 if (!is_kcfg_value_in_range(ext, value)) {
1757 pr_warn("extern (kcfg) %s=%llu value doesn't fit in %d bytes\n",
1758 ext->name, (unsigned long long)value, ext->kcfg.sz);
1759 return -ERANGE;
1760 }
1761 switch (ext->kcfg.sz) {
1762 case 1: *(__u8 *)ext_val = value; break;
1763 case 2: *(__u16 *)ext_val = value; break;
1764 case 4: *(__u32 *)ext_val = value; break;
1765 case 8: *(__u64 *)ext_val = value; break;
1766 default:
1767 return -EINVAL;
1768 }
1769 ext->is_set = true;
1770 return 0;
1771 }
1772
bpf_object__process_kconfig_line(struct bpf_object * obj,char * buf,void * data)1773 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1774 char *buf, void *data)
1775 {
1776 struct extern_desc *ext;
1777 char *sep, *value;
1778 int len, err = 0;
1779 void *ext_val;
1780 __u64 num;
1781
1782 if (!str_has_pfx(buf, "CONFIG_"))
1783 return 0;
1784
1785 sep = strchr(buf, '=');
1786 if (!sep) {
1787 pr_warn("failed to parse '%s': no separator\n", buf);
1788 return -EINVAL;
1789 }
1790
1791 /* Trim ending '\n' */
1792 len = strlen(buf);
1793 if (buf[len - 1] == '\n')
1794 buf[len - 1] = '\0';
1795 /* Split on '=' and ensure that a value is present. */
1796 *sep = '\0';
1797 if (!sep[1]) {
1798 *sep = '=';
1799 pr_warn("failed to parse '%s': no value\n", buf);
1800 return -EINVAL;
1801 }
1802
1803 ext = find_extern_by_name(obj, buf);
1804 if (!ext || ext->is_set)
1805 return 0;
1806
1807 ext_val = data + ext->kcfg.data_off;
1808 value = sep + 1;
1809
1810 switch (*value) {
1811 case 'y': case 'n': case 'm':
1812 err = set_kcfg_value_tri(ext, ext_val, *value);
1813 break;
1814 case '"':
1815 err = set_kcfg_value_str(ext, ext_val, value);
1816 break;
1817 default:
1818 /* assume integer */
1819 err = parse_u64(value, &num);
1820 if (err) {
1821 pr_warn("extern (kcfg) %s=%s should be integer\n",
1822 ext->name, value);
1823 return err;
1824 }
1825 err = set_kcfg_value_num(ext, ext_val, num);
1826 break;
1827 }
1828 if (err)
1829 return err;
1830 pr_debug("extern (kcfg) %s=%s\n", ext->name, value);
1831 return 0;
1832 }
1833
bpf_object__read_kconfig_file(struct bpf_object * obj,void * data)1834 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
1835 {
1836 char buf[PATH_MAX];
1837 struct utsname uts;
1838 int len, err = 0;
1839 gzFile file;
1840
1841 uname(&uts);
1842 len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
1843 if (len < 0)
1844 return -EINVAL;
1845 else if (len >= PATH_MAX)
1846 return -ENAMETOOLONG;
1847
1848 /* gzopen also accepts uncompressed files. */
1849 file = gzopen(buf, "r");
1850 if (!file)
1851 file = gzopen("/proc/config.gz", "r");
1852
1853 if (!file) {
1854 pr_warn("failed to open system Kconfig\n");
1855 return -ENOENT;
1856 }
1857
1858 while (gzgets(file, buf, sizeof(buf))) {
1859 err = bpf_object__process_kconfig_line(obj, buf, data);
1860 if (err) {
1861 pr_warn("error parsing system Kconfig line '%s': %d\n",
1862 buf, err);
1863 goto out;
1864 }
1865 }
1866
1867 out:
1868 gzclose(file);
1869 return err;
1870 }
1871
bpf_object__read_kconfig_mem(struct bpf_object * obj,const char * config,void * data)1872 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
1873 const char *config, void *data)
1874 {
1875 char buf[PATH_MAX];
1876 int err = 0;
1877 FILE *file;
1878
1879 file = fmemopen((void *)config, strlen(config), "r");
1880 if (!file) {
1881 err = -errno;
1882 pr_warn("failed to open in-memory Kconfig: %d\n", err);
1883 return err;
1884 }
1885
1886 while (fgets(buf, sizeof(buf), file)) {
1887 err = bpf_object__process_kconfig_line(obj, buf, data);
1888 if (err) {
1889 pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
1890 buf, err);
1891 break;
1892 }
1893 }
1894
1895 fclose(file);
1896 return err;
1897 }
1898
bpf_object__init_kconfig_map(struct bpf_object * obj)1899 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
1900 {
1901 struct extern_desc *last_ext = NULL, *ext;
1902 size_t map_sz;
1903 int i, err;
1904
1905 for (i = 0; i < obj->nr_extern; i++) {
1906 ext = &obj->externs[i];
1907 if (ext->type == EXT_KCFG)
1908 last_ext = ext;
1909 }
1910
1911 if (!last_ext)
1912 return 0;
1913
1914 map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz;
1915 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
1916 ".kconfig", obj->efile.symbols_shndx,
1917 NULL, map_sz);
1918 if (err)
1919 return err;
1920
1921 obj->kconfig_map_idx = obj->nr_maps - 1;
1922
1923 return 0;
1924 }
1925
bpf_object__init_user_maps(struct bpf_object * obj,bool strict)1926 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
1927 {
1928 Elf_Data *symbols = obj->efile.symbols;
1929 int i, map_def_sz = 0, nr_maps = 0, nr_syms;
1930 Elf_Data *data = NULL;
1931 Elf_Scn *scn;
1932
1933 if (obj->efile.maps_shndx < 0)
1934 return 0;
1935
1936 if (libbpf_mode & LIBBPF_STRICT_MAP_DEFINITIONS) {
1937 pr_warn("legacy map definitions in SEC(\"maps\") are not supported\n");
1938 return -EOPNOTSUPP;
1939 }
1940
1941 if (!symbols)
1942 return -EINVAL;
1943
1944 scn = elf_sec_by_idx(obj, obj->efile.maps_shndx);
1945 data = elf_sec_data(obj, scn);
1946 if (!scn || !data) {
1947 pr_warn("elf: failed to get legacy map definitions for %s\n",
1948 obj->path);
1949 return -EINVAL;
1950 }
1951
1952 /*
1953 * Count number of maps. Each map has a name.
1954 * Array of maps is not supported: only the first element is
1955 * considered.
1956 *
1957 * TODO: Detect array of map and report error.
1958 */
1959 nr_syms = symbols->d_size / sizeof(Elf64_Sym);
1960 for (i = 0; i < nr_syms; i++) {
1961 Elf64_Sym *sym = elf_sym_by_idx(obj, i);
1962
1963 if (sym->st_shndx != obj->efile.maps_shndx)
1964 continue;
1965 if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION)
1966 continue;
1967 nr_maps++;
1968 }
1969 /* Assume equally sized map definitions */
1970 pr_debug("elf: found %d legacy map definitions (%zd bytes) in %s\n",
1971 nr_maps, data->d_size, obj->path);
1972
1973 if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) {
1974 pr_warn("elf: unable to determine legacy map definition size in %s\n",
1975 obj->path);
1976 return -EINVAL;
1977 }
1978 map_def_sz = data->d_size / nr_maps;
1979
1980 /* Fill obj->maps using data in "maps" section. */
1981 for (i = 0; i < nr_syms; i++) {
1982 Elf64_Sym *sym = elf_sym_by_idx(obj, i);
1983 const char *map_name;
1984 struct bpf_map_def *def;
1985 struct bpf_map *map;
1986
1987 if (sym->st_shndx != obj->efile.maps_shndx)
1988 continue;
1989 if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION)
1990 continue;
1991
1992 map = bpf_object__add_map(obj);
1993 if (IS_ERR(map))
1994 return PTR_ERR(map);
1995
1996 map_name = elf_sym_str(obj, sym->st_name);
1997 if (!map_name) {
1998 pr_warn("failed to get map #%d name sym string for obj %s\n",
1999 i, obj->path);
2000 return -LIBBPF_ERRNO__FORMAT;
2001 }
2002
2003 pr_warn("map '%s' (legacy): legacy map definitions are deprecated, use BTF-defined maps instead\n", map_name);
2004
2005 if (ELF64_ST_BIND(sym->st_info) == STB_LOCAL) {
2006 pr_warn("map '%s' (legacy): static maps are not supported\n", map_name);
2007 return -ENOTSUP;
2008 }
2009
2010 map->libbpf_type = LIBBPF_MAP_UNSPEC;
2011 map->sec_idx = sym->st_shndx;
2012 map->sec_offset = sym->st_value;
2013 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
2014 map_name, map->sec_idx, map->sec_offset);
2015 if (sym->st_value + map_def_sz > data->d_size) {
2016 pr_warn("corrupted maps section in %s: last map \"%s\" too small\n",
2017 obj->path, map_name);
2018 return -EINVAL;
2019 }
2020
2021 map->name = strdup(map_name);
2022 if (!map->name) {
2023 pr_warn("map '%s': failed to alloc map name\n", map_name);
2024 return -ENOMEM;
2025 }
2026 pr_debug("map %d is \"%s\"\n", i, map->name);
2027 def = (struct bpf_map_def *)(data->d_buf + sym->st_value);
2028 /*
2029 * If the definition of the map in the object file fits in
2030 * bpf_map_def, copy it. Any extra fields in our version
2031 * of bpf_map_def will default to zero as a result of the
2032 * calloc above.
2033 */
2034 if (map_def_sz <= sizeof(struct bpf_map_def)) {
2035 memcpy(&map->def, def, map_def_sz);
2036 } else {
2037 /*
2038 * Here the map structure being read is bigger than what
2039 * we expect, truncate if the excess bits are all zero.
2040 * If they are not zero, reject this map as
2041 * incompatible.
2042 */
2043 char *b;
2044
2045 for (b = ((char *)def) + sizeof(struct bpf_map_def);
2046 b < ((char *)def) + map_def_sz; b++) {
2047 if (*b != 0) {
2048 pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n",
2049 obj->path, map_name);
2050 if (strict)
2051 return -EINVAL;
2052 }
2053 }
2054 memcpy(&map->def, def, sizeof(struct bpf_map_def));
2055 }
2056 }
2057 return 0;
2058 }
2059
2060 const struct btf_type *
skip_mods_and_typedefs(const struct btf * btf,__u32 id,__u32 * res_id)2061 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
2062 {
2063 const struct btf_type *t = btf__type_by_id(btf, id);
2064
2065 if (res_id)
2066 *res_id = id;
2067
2068 while (btf_is_mod(t) || btf_is_typedef(t)) {
2069 if (res_id)
2070 *res_id = t->type;
2071 t = btf__type_by_id(btf, t->type);
2072 }
2073
2074 return t;
2075 }
2076
2077 static const struct btf_type *
resolve_func_ptr(const struct btf * btf,__u32 id,__u32 * res_id)2078 resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
2079 {
2080 const struct btf_type *t;
2081
2082 t = skip_mods_and_typedefs(btf, id, NULL);
2083 if (!btf_is_ptr(t))
2084 return NULL;
2085
2086 t = skip_mods_and_typedefs(btf, t->type, res_id);
2087
2088 return btf_is_func_proto(t) ? t : NULL;
2089 }
2090
__btf_kind_str(__u16 kind)2091 static const char *__btf_kind_str(__u16 kind)
2092 {
2093 switch (kind) {
2094 case BTF_KIND_UNKN: return "void";
2095 case BTF_KIND_INT: return "int";
2096 case BTF_KIND_PTR: return "ptr";
2097 case BTF_KIND_ARRAY: return "array";
2098 case BTF_KIND_STRUCT: return "struct";
2099 case BTF_KIND_UNION: return "union";
2100 case BTF_KIND_ENUM: return "enum";
2101 case BTF_KIND_FWD: return "fwd";
2102 case BTF_KIND_TYPEDEF: return "typedef";
2103 case BTF_KIND_VOLATILE: return "volatile";
2104 case BTF_KIND_CONST: return "const";
2105 case BTF_KIND_RESTRICT: return "restrict";
2106 case BTF_KIND_FUNC: return "func";
2107 case BTF_KIND_FUNC_PROTO: return "func_proto";
2108 case BTF_KIND_VAR: return "var";
2109 case BTF_KIND_DATASEC: return "datasec";
2110 case BTF_KIND_FLOAT: return "float";
2111 case BTF_KIND_DECL_TAG: return "decl_tag";
2112 case BTF_KIND_TYPE_TAG: return "type_tag";
2113 default: return "unknown";
2114 }
2115 }
2116
btf_kind_str(const struct btf_type * t)2117 const char *btf_kind_str(const struct btf_type *t)
2118 {
2119 return __btf_kind_str(btf_kind(t));
2120 }
2121
2122 /*
2123 * Fetch integer attribute of BTF map definition. Such attributes are
2124 * represented using a pointer to an array, in which dimensionality of array
2125 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
2126 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
2127 * type definition, while using only sizeof(void *) space in ELF data section.
2128 */
get_map_field_int(const char * map_name,const struct btf * btf,const struct btf_member * m,__u32 * res)2129 static bool get_map_field_int(const char *map_name, const struct btf *btf,
2130 const struct btf_member *m, __u32 *res)
2131 {
2132 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
2133 const char *name = btf__name_by_offset(btf, m->name_off);
2134 const struct btf_array *arr_info;
2135 const struct btf_type *arr_t;
2136
2137 if (!btf_is_ptr(t)) {
2138 pr_warn("map '%s': attr '%s': expected PTR, got %s.\n",
2139 map_name, name, btf_kind_str(t));
2140 return false;
2141 }
2142
2143 arr_t = btf__type_by_id(btf, t->type);
2144 if (!arr_t) {
2145 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
2146 map_name, name, t->type);
2147 return false;
2148 }
2149 if (!btf_is_array(arr_t)) {
2150 pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n",
2151 map_name, name, btf_kind_str(arr_t));
2152 return false;
2153 }
2154 arr_info = btf_array(arr_t);
2155 *res = arr_info->nelems;
2156 return true;
2157 }
2158
build_map_pin_path(struct bpf_map * map,const char * path)2159 static int build_map_pin_path(struct bpf_map *map, const char *path)
2160 {
2161 char buf[PATH_MAX];
2162 int len;
2163
2164 if (!path)
2165 path = "/sys/fs/bpf";
2166
2167 len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map));
2168 if (len < 0)
2169 return -EINVAL;
2170 else if (len >= PATH_MAX)
2171 return -ENAMETOOLONG;
2172
2173 return bpf_map__set_pin_path(map, buf);
2174 }
2175
parse_btf_map_def(const char * map_name,struct btf * btf,const struct btf_type * def_t,bool strict,struct btf_map_def * map_def,struct btf_map_def * inner_def)2176 int parse_btf_map_def(const char *map_name, struct btf *btf,
2177 const struct btf_type *def_t, bool strict,
2178 struct btf_map_def *map_def, struct btf_map_def *inner_def)
2179 {
2180 const struct btf_type *t;
2181 const struct btf_member *m;
2182 bool is_inner = inner_def == NULL;
2183 int vlen, i;
2184
2185 vlen = btf_vlen(def_t);
2186 m = btf_members(def_t);
2187 for (i = 0; i < vlen; i++, m++) {
2188 const char *name = btf__name_by_offset(btf, m->name_off);
2189
2190 if (!name) {
2191 pr_warn("map '%s': invalid field #%d.\n", map_name, i);
2192 return -EINVAL;
2193 }
2194 if (strcmp(name, "type") == 0) {
2195 if (!get_map_field_int(map_name, btf, m, &map_def->map_type))
2196 return -EINVAL;
2197 map_def->parts |= MAP_DEF_MAP_TYPE;
2198 } else if (strcmp(name, "max_entries") == 0) {
2199 if (!get_map_field_int(map_name, btf, m, &map_def->max_entries))
2200 return -EINVAL;
2201 map_def->parts |= MAP_DEF_MAX_ENTRIES;
2202 } else if (strcmp(name, "map_flags") == 0) {
2203 if (!get_map_field_int(map_name, btf, m, &map_def->map_flags))
2204 return -EINVAL;
2205 map_def->parts |= MAP_DEF_MAP_FLAGS;
2206 } else if (strcmp(name, "numa_node") == 0) {
2207 if (!get_map_field_int(map_name, btf, m, &map_def->numa_node))
2208 return -EINVAL;
2209 map_def->parts |= MAP_DEF_NUMA_NODE;
2210 } else if (strcmp(name, "key_size") == 0) {
2211 __u32 sz;
2212
2213 if (!get_map_field_int(map_name, btf, m, &sz))
2214 return -EINVAL;
2215 if (map_def->key_size && map_def->key_size != sz) {
2216 pr_warn("map '%s': conflicting key size %u != %u.\n",
2217 map_name, map_def->key_size, sz);
2218 return -EINVAL;
2219 }
2220 map_def->key_size = sz;
2221 map_def->parts |= MAP_DEF_KEY_SIZE;
2222 } else if (strcmp(name, "key") == 0) {
2223 __s64 sz;
2224
2225 t = btf__type_by_id(btf, m->type);
2226 if (!t) {
2227 pr_warn("map '%s': key type [%d] not found.\n",
2228 map_name, m->type);
2229 return -EINVAL;
2230 }
2231 if (!btf_is_ptr(t)) {
2232 pr_warn("map '%s': key spec is not PTR: %s.\n",
2233 map_name, btf_kind_str(t));
2234 return -EINVAL;
2235 }
2236 sz = btf__resolve_size(btf, t->type);
2237 if (sz < 0) {
2238 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2239 map_name, t->type, (ssize_t)sz);
2240 return sz;
2241 }
2242 if (map_def->key_size && map_def->key_size != sz) {
2243 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2244 map_name, map_def->key_size, (ssize_t)sz);
2245 return -EINVAL;
2246 }
2247 map_def->key_size = sz;
2248 map_def->key_type_id = t->type;
2249 map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE;
2250 } else if (strcmp(name, "value_size") == 0) {
2251 __u32 sz;
2252
2253 if (!get_map_field_int(map_name, btf, m, &sz))
2254 return -EINVAL;
2255 if (map_def->value_size && map_def->value_size != sz) {
2256 pr_warn("map '%s': conflicting value size %u != %u.\n",
2257 map_name, map_def->value_size, sz);
2258 return -EINVAL;
2259 }
2260 map_def->value_size = sz;
2261 map_def->parts |= MAP_DEF_VALUE_SIZE;
2262 } else if (strcmp(name, "value") == 0) {
2263 __s64 sz;
2264
2265 t = btf__type_by_id(btf, m->type);
2266 if (!t) {
2267 pr_warn("map '%s': value type [%d] not found.\n",
2268 map_name, m->type);
2269 return -EINVAL;
2270 }
2271 if (!btf_is_ptr(t)) {
2272 pr_warn("map '%s': value spec is not PTR: %s.\n",
2273 map_name, btf_kind_str(t));
2274 return -EINVAL;
2275 }
2276 sz = btf__resolve_size(btf, t->type);
2277 if (sz < 0) {
2278 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2279 map_name, t->type, (ssize_t)sz);
2280 return sz;
2281 }
2282 if (map_def->value_size && map_def->value_size != sz) {
2283 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2284 map_name, map_def->value_size, (ssize_t)sz);
2285 return -EINVAL;
2286 }
2287 map_def->value_size = sz;
2288 map_def->value_type_id = t->type;
2289 map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE;
2290 }
2291 else if (strcmp(name, "values") == 0) {
2292 bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type);
2293 bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY;
2294 const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value";
2295 char inner_map_name[128];
2296 int err;
2297
2298 if (is_inner) {
2299 pr_warn("map '%s': multi-level inner maps not supported.\n",
2300 map_name);
2301 return -ENOTSUP;
2302 }
2303 if (i != vlen - 1) {
2304 pr_warn("map '%s': '%s' member should be last.\n",
2305 map_name, name);
2306 return -EINVAL;
2307 }
2308 if (!is_map_in_map && !is_prog_array) {
2309 pr_warn("map '%s': should be map-in-map or prog-array.\n",
2310 map_name);
2311 return -ENOTSUP;
2312 }
2313 if (map_def->value_size && map_def->value_size != 4) {
2314 pr_warn("map '%s': conflicting value size %u != 4.\n",
2315 map_name, map_def->value_size);
2316 return -EINVAL;
2317 }
2318 map_def->value_size = 4;
2319 t = btf__type_by_id(btf, m->type);
2320 if (!t) {
2321 pr_warn("map '%s': %s type [%d] not found.\n",
2322 map_name, desc, m->type);
2323 return -EINVAL;
2324 }
2325 if (!btf_is_array(t) || btf_array(t)->nelems) {
2326 pr_warn("map '%s': %s spec is not a zero-sized array.\n",
2327 map_name, desc);
2328 return -EINVAL;
2329 }
2330 t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL);
2331 if (!btf_is_ptr(t)) {
2332 pr_warn("map '%s': %s def is of unexpected kind %s.\n",
2333 map_name, desc, btf_kind_str(t));
2334 return -EINVAL;
2335 }
2336 t = skip_mods_and_typedefs(btf, t->type, NULL);
2337 if (is_prog_array) {
2338 if (!btf_is_func_proto(t)) {
2339 pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n",
2340 map_name, btf_kind_str(t));
2341 return -EINVAL;
2342 }
2343 continue;
2344 }
2345 if (!btf_is_struct(t)) {
2346 pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2347 map_name, btf_kind_str(t));
2348 return -EINVAL;
2349 }
2350
2351 snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name);
2352 err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL);
2353 if (err)
2354 return err;
2355
2356 map_def->parts |= MAP_DEF_INNER_MAP;
2357 } else if (strcmp(name, "pinning") == 0) {
2358 __u32 val;
2359
2360 if (is_inner) {
2361 pr_warn("map '%s': inner def can't be pinned.\n", map_name);
2362 return -EINVAL;
2363 }
2364 if (!get_map_field_int(map_name, btf, m, &val))
2365 return -EINVAL;
2366 if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) {
2367 pr_warn("map '%s': invalid pinning value %u.\n",
2368 map_name, val);
2369 return -EINVAL;
2370 }
2371 map_def->pinning = val;
2372 map_def->parts |= MAP_DEF_PINNING;
2373 } else if (strcmp(name, "map_extra") == 0) {
2374 __u32 map_extra;
2375
2376 if (!get_map_field_int(map_name, btf, m, &map_extra))
2377 return -EINVAL;
2378 map_def->map_extra = map_extra;
2379 map_def->parts |= MAP_DEF_MAP_EXTRA;
2380 } else {
2381 if (strict) {
2382 pr_warn("map '%s': unknown field '%s'.\n", map_name, name);
2383 return -ENOTSUP;
2384 }
2385 pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name);
2386 }
2387 }
2388
2389 if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) {
2390 pr_warn("map '%s': map type isn't specified.\n", map_name);
2391 return -EINVAL;
2392 }
2393
2394 return 0;
2395 }
2396
fill_map_from_def(struct bpf_map * map,const struct btf_map_def * def)2397 static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def)
2398 {
2399 map->def.type = def->map_type;
2400 map->def.key_size = def->key_size;
2401 map->def.value_size = def->value_size;
2402 map->def.max_entries = def->max_entries;
2403 map->def.map_flags = def->map_flags;
2404 map->map_extra = def->map_extra;
2405
2406 map->numa_node = def->numa_node;
2407 map->btf_key_type_id = def->key_type_id;
2408 map->btf_value_type_id = def->value_type_id;
2409
2410 if (def->parts & MAP_DEF_MAP_TYPE)
2411 pr_debug("map '%s': found type = %u.\n", map->name, def->map_type);
2412
2413 if (def->parts & MAP_DEF_KEY_TYPE)
2414 pr_debug("map '%s': found key [%u], sz = %u.\n",
2415 map->name, def->key_type_id, def->key_size);
2416 else if (def->parts & MAP_DEF_KEY_SIZE)
2417 pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size);
2418
2419 if (def->parts & MAP_DEF_VALUE_TYPE)
2420 pr_debug("map '%s': found value [%u], sz = %u.\n",
2421 map->name, def->value_type_id, def->value_size);
2422 else if (def->parts & MAP_DEF_VALUE_SIZE)
2423 pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size);
2424
2425 if (def->parts & MAP_DEF_MAX_ENTRIES)
2426 pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries);
2427 if (def->parts & MAP_DEF_MAP_FLAGS)
2428 pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags);
2429 if (def->parts & MAP_DEF_MAP_EXTRA)
2430 pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name,
2431 (unsigned long long)def->map_extra);
2432 if (def->parts & MAP_DEF_PINNING)
2433 pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning);
2434 if (def->parts & MAP_DEF_NUMA_NODE)
2435 pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node);
2436
2437 if (def->parts & MAP_DEF_INNER_MAP)
2438 pr_debug("map '%s': found inner map definition.\n", map->name);
2439 }
2440
btf_var_linkage_str(__u32 linkage)2441 static const char *btf_var_linkage_str(__u32 linkage)
2442 {
2443 switch (linkage) {
2444 case BTF_VAR_STATIC: return "static";
2445 case BTF_VAR_GLOBAL_ALLOCATED: return "global";
2446 case BTF_VAR_GLOBAL_EXTERN: return "extern";
2447 default: return "unknown";
2448 }
2449 }
2450
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,const char * pin_root_path)2451 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
2452 const struct btf_type *sec,
2453 int var_idx, int sec_idx,
2454 const Elf_Data *data, bool strict,
2455 const char *pin_root_path)
2456 {
2457 struct btf_map_def map_def = {}, inner_def = {};
2458 const struct btf_type *var, *def;
2459 const struct btf_var_secinfo *vi;
2460 const struct btf_var *var_extra;
2461 const char *map_name;
2462 struct bpf_map *map;
2463 int err;
2464
2465 vi = btf_var_secinfos(sec) + var_idx;
2466 var = btf__type_by_id(obj->btf, vi->type);
2467 var_extra = btf_var(var);
2468 map_name = btf__name_by_offset(obj->btf, var->name_off);
2469
2470 if (map_name == NULL || map_name[0] == '\0') {
2471 pr_warn("map #%d: empty name.\n", var_idx);
2472 return -EINVAL;
2473 }
2474 if ((__u64)vi->offset + vi->size > data->d_size) {
2475 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
2476 return -EINVAL;
2477 }
2478 if (!btf_is_var(var)) {
2479 pr_warn("map '%s': unexpected var kind %s.\n",
2480 map_name, btf_kind_str(var));
2481 return -EINVAL;
2482 }
2483 if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) {
2484 pr_warn("map '%s': unsupported map linkage %s.\n",
2485 map_name, btf_var_linkage_str(var_extra->linkage));
2486 return -EOPNOTSUPP;
2487 }
2488
2489 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
2490 if (!btf_is_struct(def)) {
2491 pr_warn("map '%s': unexpected def kind %s.\n",
2492 map_name, btf_kind_str(var));
2493 return -EINVAL;
2494 }
2495 if (def->size > vi->size) {
2496 pr_warn("map '%s': invalid def size.\n", map_name);
2497 return -EINVAL;
2498 }
2499
2500 map = bpf_object__add_map(obj);
2501 if (IS_ERR(map))
2502 return PTR_ERR(map);
2503 map->name = strdup(map_name);
2504 if (!map->name) {
2505 pr_warn("map '%s': failed to alloc map name.\n", map_name);
2506 return -ENOMEM;
2507 }
2508 map->libbpf_type = LIBBPF_MAP_UNSPEC;
2509 map->def.type = BPF_MAP_TYPE_UNSPEC;
2510 map->sec_idx = sec_idx;
2511 map->sec_offset = vi->offset;
2512 map->btf_var_idx = var_idx;
2513 pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
2514 map_name, map->sec_idx, map->sec_offset);
2515
2516 err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def);
2517 if (err)
2518 return err;
2519
2520 fill_map_from_def(map, &map_def);
2521
2522 if (map_def.pinning == LIBBPF_PIN_BY_NAME) {
2523 err = build_map_pin_path(map, pin_root_path);
2524 if (err) {
2525 pr_warn("map '%s': couldn't build pin path.\n", map->name);
2526 return err;
2527 }
2528 }
2529
2530 if (map_def.parts & MAP_DEF_INNER_MAP) {
2531 map->inner_map = calloc(1, sizeof(*map->inner_map));
2532 if (!map->inner_map)
2533 return -ENOMEM;
2534 map->inner_map->fd = -1;
2535 map->inner_map->sec_idx = sec_idx;
2536 map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1);
2537 if (!map->inner_map->name)
2538 return -ENOMEM;
2539 sprintf(map->inner_map->name, "%s.inner", map_name);
2540
2541 fill_map_from_def(map->inner_map, &inner_def);
2542 }
2543
2544 return 0;
2545 }
2546
bpf_object__init_user_btf_maps(struct bpf_object * obj,bool strict,const char * pin_root_path)2547 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2548 const char *pin_root_path)
2549 {
2550 const struct btf_type *sec = NULL;
2551 int nr_types, i, vlen, err;
2552 const struct btf_type *t;
2553 const char *name;
2554 Elf_Data *data;
2555 Elf_Scn *scn;
2556
2557 if (obj->efile.btf_maps_shndx < 0)
2558 return 0;
2559
2560 scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx);
2561 data = elf_sec_data(obj, scn);
2562 if (!scn || !data) {
2563 pr_warn("elf: failed to get %s map definitions for %s\n",
2564 MAPS_ELF_SEC, obj->path);
2565 return -EINVAL;
2566 }
2567
2568 nr_types = btf__type_cnt(obj->btf);
2569 for (i = 1; i < nr_types; i++) {
2570 t = btf__type_by_id(obj->btf, i);
2571 if (!btf_is_datasec(t))
2572 continue;
2573 name = btf__name_by_offset(obj->btf, t->name_off);
2574 if (strcmp(name, MAPS_ELF_SEC) == 0) {
2575 sec = t;
2576 obj->efile.btf_maps_sec_btf_id = i;
2577 break;
2578 }
2579 }
2580
2581 if (!sec) {
2582 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
2583 return -ENOENT;
2584 }
2585
2586 vlen = btf_vlen(sec);
2587 for (i = 0; i < vlen; i++) {
2588 err = bpf_object__init_user_btf_map(obj, sec, i,
2589 obj->efile.btf_maps_shndx,
2590 data, strict,
2591 pin_root_path);
2592 if (err)
2593 return err;
2594 }
2595
2596 return 0;
2597 }
2598
bpf_object__init_maps(struct bpf_object * obj,const struct bpf_object_open_opts * opts)2599 static int bpf_object__init_maps(struct bpf_object *obj,
2600 const struct bpf_object_open_opts *opts)
2601 {
2602 const char *pin_root_path;
2603 bool strict;
2604 int err;
2605
2606 strict = !OPTS_GET(opts, relaxed_maps, false);
2607 pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
2608
2609 err = bpf_object__init_user_maps(obj, strict);
2610 err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
2611 err = err ?: bpf_object__init_global_data_maps(obj);
2612 err = err ?: bpf_object__init_kconfig_map(obj);
2613 err = err ?: bpf_object__init_struct_ops_maps(obj);
2614
2615 return err;
2616 }
2617
section_have_execinstr(struct bpf_object * obj,int idx)2618 static bool section_have_execinstr(struct bpf_object *obj, int idx)
2619 {
2620 Elf64_Shdr *sh;
2621
2622 sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx));
2623 if (!sh)
2624 return false;
2625
2626 return sh->sh_flags & SHF_EXECINSTR;
2627 }
2628
btf_needs_sanitization(struct bpf_object * obj)2629 static bool btf_needs_sanitization(struct bpf_object *obj)
2630 {
2631 bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2632 bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2633 bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2634 bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2635 bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
2636 bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
2637
2638 return !has_func || !has_datasec || !has_func_global || !has_float ||
2639 !has_decl_tag || !has_type_tag;
2640 }
2641
bpf_object__sanitize_btf(struct bpf_object * obj,struct btf * btf)2642 static void bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf)
2643 {
2644 bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2645 bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2646 bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2647 bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2648 bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
2649 bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
2650 struct btf_type *t;
2651 int i, j, vlen;
2652
2653 for (i = 1; i < btf__type_cnt(btf); i++) {
2654 t = (struct btf_type *)btf__type_by_id(btf, i);
2655
2656 if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) {
2657 /* replace VAR/DECL_TAG with INT */
2658 t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
2659 /*
2660 * using size = 1 is the safest choice, 4 will be too
2661 * big and cause kernel BTF validation failure if
2662 * original variable took less than 4 bytes
2663 */
2664 t->size = 1;
2665 *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
2666 } else if (!has_datasec && btf_is_datasec(t)) {
2667 /* replace DATASEC with STRUCT */
2668 const struct btf_var_secinfo *v = btf_var_secinfos(t);
2669 struct btf_member *m = btf_members(t);
2670 struct btf_type *vt;
2671 char *name;
2672
2673 name = (char *)btf__name_by_offset(btf, t->name_off);
2674 while (*name) {
2675 if (*name == '.')
2676 *name = '_';
2677 name++;
2678 }
2679
2680 vlen = btf_vlen(t);
2681 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
2682 for (j = 0; j < vlen; j++, v++, m++) {
2683 /* order of field assignments is important */
2684 m->offset = v->offset * 8;
2685 m->type = v->type;
2686 /* preserve variable name as member name */
2687 vt = (void *)btf__type_by_id(btf, v->type);
2688 m->name_off = vt->name_off;
2689 }
2690 } else if (!has_func && btf_is_func_proto(t)) {
2691 /* replace FUNC_PROTO with ENUM */
2692 vlen = btf_vlen(t);
2693 t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
2694 t->size = sizeof(__u32); /* kernel enforced */
2695 } else if (!has_func && btf_is_func(t)) {
2696 /* replace FUNC with TYPEDEF */
2697 t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
2698 } else if (!has_func_global && btf_is_func(t)) {
2699 /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
2700 t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
2701 } else if (!has_float && btf_is_float(t)) {
2702 /* replace FLOAT with an equally-sized empty STRUCT;
2703 * since C compilers do not accept e.g. "float" as a
2704 * valid struct name, make it anonymous
2705 */
2706 t->name_off = 0;
2707 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0);
2708 } else if (!has_type_tag && btf_is_type_tag(t)) {
2709 /* replace TYPE_TAG with a CONST */
2710 t->name_off = 0;
2711 t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0);
2712 }
2713 }
2714 }
2715
libbpf_needs_btf(const struct bpf_object * obj)2716 static bool libbpf_needs_btf(const struct bpf_object *obj)
2717 {
2718 return obj->efile.btf_maps_shndx >= 0 ||
2719 obj->efile.st_ops_shndx >= 0 ||
2720 obj->nr_extern > 0;
2721 }
2722
kernel_needs_btf(const struct bpf_object * obj)2723 static bool kernel_needs_btf(const struct bpf_object *obj)
2724 {
2725 return obj->efile.st_ops_shndx >= 0;
2726 }
2727
bpf_object__init_btf(struct bpf_object * obj,Elf_Data * btf_data,Elf_Data * btf_ext_data)2728 static int bpf_object__init_btf(struct bpf_object *obj,
2729 Elf_Data *btf_data,
2730 Elf_Data *btf_ext_data)
2731 {
2732 int err = -ENOENT;
2733
2734 if (btf_data) {
2735 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
2736 err = libbpf_get_error(obj->btf);
2737 if (err) {
2738 obj->btf = NULL;
2739 pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err);
2740 goto out;
2741 }
2742 /* enforce 8-byte pointers for BPF-targeted BTFs */
2743 btf__set_pointer_size(obj->btf, 8);
2744 }
2745 if (btf_ext_data) {
2746 if (!obj->btf) {
2747 pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2748 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2749 goto out;
2750 }
2751 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
2752 err = libbpf_get_error(obj->btf_ext);
2753 if (err) {
2754 pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n",
2755 BTF_EXT_ELF_SEC, err);
2756 obj->btf_ext = NULL;
2757 goto out;
2758 }
2759 }
2760 out:
2761 if (err && libbpf_needs_btf(obj)) {
2762 pr_warn("BTF is required, but is missing or corrupted.\n");
2763 return err;
2764 }
2765 return 0;
2766 }
2767
compare_vsi_off(const void * _a,const void * _b)2768 static int compare_vsi_off(const void *_a, const void *_b)
2769 {
2770 const struct btf_var_secinfo *a = _a;
2771 const struct btf_var_secinfo *b = _b;
2772
2773 return a->offset - b->offset;
2774 }
2775
btf_fixup_datasec(struct bpf_object * obj,struct btf * btf,struct btf_type * t)2776 static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
2777 struct btf_type *t)
2778 {
2779 __u32 size = 0, off = 0, i, vars = btf_vlen(t);
2780 const char *name = btf__name_by_offset(btf, t->name_off);
2781 const struct btf_type *t_var;
2782 struct btf_var_secinfo *vsi;
2783 const struct btf_var *var;
2784 int ret;
2785
2786 if (!name) {
2787 pr_debug("No name found in string section for DATASEC kind.\n");
2788 return -ENOENT;
2789 }
2790
2791 /* .extern datasec size and var offsets were set correctly during
2792 * extern collection step, so just skip straight to sorting variables
2793 */
2794 if (t->size)
2795 goto sort_vars;
2796
2797 ret = find_elf_sec_sz(obj, name, &size);
2798 if (ret || !size || (t->size && t->size != size)) {
2799 pr_debug("Invalid size for section %s: %u bytes\n", name, size);
2800 return -ENOENT;
2801 }
2802
2803 t->size = size;
2804
2805 for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
2806 t_var = btf__type_by_id(btf, vsi->type);
2807 if (!t_var || !btf_is_var(t_var)) {
2808 pr_debug("Non-VAR type seen in section %s\n", name);
2809 return -EINVAL;
2810 }
2811
2812 var = btf_var(t_var);
2813 if (var->linkage == BTF_VAR_STATIC)
2814 continue;
2815
2816 name = btf__name_by_offset(btf, t_var->name_off);
2817 if (!name) {
2818 pr_debug("No name found in string section for VAR kind\n");
2819 return -ENOENT;
2820 }
2821
2822 ret = find_elf_var_offset(obj, name, &off);
2823 if (ret) {
2824 pr_debug("No offset found in symbol table for VAR %s\n",
2825 name);
2826 return -ENOENT;
2827 }
2828
2829 vsi->offset = off;
2830 }
2831
2832 sort_vars:
2833 qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
2834 return 0;
2835 }
2836
btf_finalize_data(struct bpf_object * obj,struct btf * btf)2837 static int btf_finalize_data(struct bpf_object *obj, struct btf *btf)
2838 {
2839 int err = 0;
2840 __u32 i, n = btf__type_cnt(btf);
2841
2842 for (i = 1; i < n; i++) {
2843 struct btf_type *t = btf_type_by_id(btf, i);
2844
2845 /* Loader needs to fix up some of the things compiler
2846 * couldn't get its hands on while emitting BTF. This
2847 * is section size and global variable offset. We use
2848 * the info from the ELF itself for this purpose.
2849 */
2850 if (btf_is_datasec(t)) {
2851 err = btf_fixup_datasec(obj, btf, t);
2852 if (err)
2853 break;
2854 }
2855 }
2856
2857 return libbpf_err(err);
2858 }
2859
btf__finalize_data(struct bpf_object * obj,struct btf * btf)2860 int btf__finalize_data(struct bpf_object *obj, struct btf *btf)
2861 {
2862 return btf_finalize_data(obj, btf);
2863 }
2864
bpf_object__finalize_btf(struct bpf_object * obj)2865 static int bpf_object__finalize_btf(struct bpf_object *obj)
2866 {
2867 int err;
2868
2869 if (!obj->btf)
2870 return 0;
2871
2872 err = btf_finalize_data(obj, obj->btf);
2873 if (err) {
2874 pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err);
2875 return err;
2876 }
2877
2878 return 0;
2879 }
2880
prog_needs_vmlinux_btf(struct bpf_program * prog)2881 static bool prog_needs_vmlinux_btf(struct bpf_program *prog)
2882 {
2883 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
2884 prog->type == BPF_PROG_TYPE_LSM)
2885 return true;
2886
2887 /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
2888 * also need vmlinux BTF
2889 */
2890 if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
2891 return true;
2892
2893 return false;
2894 }
2895
obj_needs_vmlinux_btf(const struct bpf_object * obj)2896 static bool obj_needs_vmlinux_btf(const struct bpf_object *obj)
2897 {
2898 struct bpf_program *prog;
2899 int i;
2900
2901 /* CO-RE relocations need kernel BTF, only when btf_custom_path
2902 * is not specified
2903 */
2904 if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path)
2905 return true;
2906
2907 /* Support for typed ksyms needs kernel BTF */
2908 for (i = 0; i < obj->nr_extern; i++) {
2909 const struct extern_desc *ext;
2910
2911 ext = &obj->externs[i];
2912 if (ext->type == EXT_KSYM && ext->ksym.type_id)
2913 return true;
2914 }
2915
2916 bpf_object__for_each_program(prog, obj) {
2917 if (!prog->load)
2918 continue;
2919 if (prog_needs_vmlinux_btf(prog))
2920 return true;
2921 }
2922
2923 return false;
2924 }
2925
bpf_object__load_vmlinux_btf(struct bpf_object * obj,bool force)2926 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force)
2927 {
2928 int err;
2929
2930 /* btf_vmlinux could be loaded earlier */
2931 if (obj->btf_vmlinux || obj->gen_loader)
2932 return 0;
2933
2934 if (!force && !obj_needs_vmlinux_btf(obj))
2935 return 0;
2936
2937 obj->btf_vmlinux = btf__load_vmlinux_btf();
2938 err = libbpf_get_error(obj->btf_vmlinux);
2939 if (err) {
2940 pr_warn("Error loading vmlinux BTF: %d\n", err);
2941 obj->btf_vmlinux = NULL;
2942 return err;
2943 }
2944 return 0;
2945 }
2946
bpf_object__sanitize_and_load_btf(struct bpf_object * obj)2947 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
2948 {
2949 struct btf *kern_btf = obj->btf;
2950 bool btf_mandatory, sanitize;
2951 int i, err = 0;
2952
2953 if (!obj->btf)
2954 return 0;
2955
2956 if (!kernel_supports(obj, FEAT_BTF)) {
2957 if (kernel_needs_btf(obj)) {
2958 err = -EOPNOTSUPP;
2959 goto report;
2960 }
2961 pr_debug("Kernel doesn't support BTF, skipping uploading it.\n");
2962 return 0;
2963 }
2964
2965 /* Even though some subprogs are global/weak, user might prefer more
2966 * permissive BPF verification process that BPF verifier performs for
2967 * static functions, taking into account more context from the caller
2968 * functions. In such case, they need to mark such subprogs with
2969 * __attribute__((visibility("hidden"))) and libbpf will adjust
2970 * corresponding FUNC BTF type to be marked as static and trigger more
2971 * involved BPF verification process.
2972 */
2973 for (i = 0; i < obj->nr_programs; i++) {
2974 struct bpf_program *prog = &obj->programs[i];
2975 struct btf_type *t;
2976 const char *name;
2977 int j, n;
2978
2979 if (!prog->mark_btf_static || !prog_is_subprog(obj, prog))
2980 continue;
2981
2982 n = btf__type_cnt(obj->btf);
2983 for (j = 1; j < n; j++) {
2984 t = btf_type_by_id(obj->btf, j);
2985 if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL)
2986 continue;
2987
2988 name = btf__str_by_offset(obj->btf, t->name_off);
2989 if (strcmp(name, prog->name) != 0)
2990 continue;
2991
2992 t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0);
2993 break;
2994 }
2995 }
2996
2997 sanitize = btf_needs_sanitization(obj);
2998 if (sanitize) {
2999 const void *raw_data;
3000 __u32 sz;
3001
3002 /* clone BTF to sanitize a copy and leave the original intact */
3003 raw_data = btf__raw_data(obj->btf, &sz);
3004 kern_btf = btf__new(raw_data, sz);
3005 err = libbpf_get_error(kern_btf);
3006 if (err)
3007 return err;
3008
3009 /* enforce 8-byte pointers for BPF-targeted BTFs */
3010 btf__set_pointer_size(obj->btf, 8);
3011 bpf_object__sanitize_btf(obj, kern_btf);
3012 }
3013
3014 if (obj->gen_loader) {
3015 __u32 raw_size = 0;
3016 const void *raw_data = btf__raw_data(kern_btf, &raw_size);
3017
3018 if (!raw_data)
3019 return -ENOMEM;
3020 bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size);
3021 /* Pretend to have valid FD to pass various fd >= 0 checks.
3022 * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
3023 */
3024 btf__set_fd(kern_btf, 0);
3025 } else {
3026 /* currently BPF_BTF_LOAD only supports log_level 1 */
3027 err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size,
3028 obj->log_level ? 1 : 0);
3029 }
3030 if (sanitize) {
3031 if (!err) {
3032 /* move fd to libbpf's BTF */
3033 btf__set_fd(obj->btf, btf__fd(kern_btf));
3034 btf__set_fd(kern_btf, -1);
3035 }
3036 btf__free(kern_btf);
3037 }
3038 report:
3039 if (err) {
3040 btf_mandatory = kernel_needs_btf(obj);
3041 pr_warn("Error loading .BTF into kernel: %d. %s\n", err,
3042 btf_mandatory ? "BTF is mandatory, can't proceed."
3043 : "BTF is optional, ignoring.");
3044 if (!btf_mandatory)
3045 err = 0;
3046 }
3047 return err;
3048 }
3049
elf_sym_str(const struct bpf_object * obj,size_t off)3050 static const char *elf_sym_str(const struct bpf_object *obj, size_t off)
3051 {
3052 const char *name;
3053
3054 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off);
3055 if (!name) {
3056 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3057 off, obj->path, elf_errmsg(-1));
3058 return NULL;
3059 }
3060
3061 return name;
3062 }
3063
elf_sec_str(const struct bpf_object * obj,size_t off)3064 static const char *elf_sec_str(const struct bpf_object *obj, size_t off)
3065 {
3066 const char *name;
3067
3068 name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off);
3069 if (!name) {
3070 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3071 off, obj->path, elf_errmsg(-1));
3072 return NULL;
3073 }
3074
3075 return name;
3076 }
3077
elf_sec_by_idx(const struct bpf_object * obj,size_t idx)3078 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx)
3079 {
3080 Elf_Scn *scn;
3081
3082 scn = elf_getscn(obj->efile.elf, idx);
3083 if (!scn) {
3084 pr_warn("elf: failed to get section(%zu) from %s: %s\n",
3085 idx, obj->path, elf_errmsg(-1));
3086 return NULL;
3087 }
3088 return scn;
3089 }
3090
elf_sec_by_name(const struct bpf_object * obj,const char * name)3091 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name)
3092 {
3093 Elf_Scn *scn = NULL;
3094 Elf *elf = obj->efile.elf;
3095 const char *sec_name;
3096
3097 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3098 sec_name = elf_sec_name(obj, scn);
3099 if (!sec_name)
3100 return NULL;
3101
3102 if (strcmp(sec_name, name) != 0)
3103 continue;
3104
3105 return scn;
3106 }
3107 return NULL;
3108 }
3109
elf_sec_hdr(const struct bpf_object * obj,Elf_Scn * scn)3110 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn)
3111 {
3112 Elf64_Shdr *shdr;
3113
3114 if (!scn)
3115 return NULL;
3116
3117 shdr = elf64_getshdr(scn);
3118 if (!shdr) {
3119 pr_warn("elf: failed to get section(%zu) header from %s: %s\n",
3120 elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3121 return NULL;
3122 }
3123
3124 return shdr;
3125 }
3126
elf_sec_name(const struct bpf_object * obj,Elf_Scn * scn)3127 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn)
3128 {
3129 const char *name;
3130 Elf64_Shdr *sh;
3131
3132 if (!scn)
3133 return NULL;
3134
3135 sh = elf_sec_hdr(obj, scn);
3136 if (!sh)
3137 return NULL;
3138
3139 name = elf_sec_str(obj, sh->sh_name);
3140 if (!name) {
3141 pr_warn("elf: failed to get section(%zu) name from %s: %s\n",
3142 elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3143 return NULL;
3144 }
3145
3146 return name;
3147 }
3148
elf_sec_data(const struct bpf_object * obj,Elf_Scn * scn)3149 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn)
3150 {
3151 Elf_Data *data;
3152
3153 if (!scn)
3154 return NULL;
3155
3156 data = elf_getdata(scn, 0);
3157 if (!data) {
3158 pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n",
3159 elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>",
3160 obj->path, elf_errmsg(-1));
3161 return NULL;
3162 }
3163
3164 return data;
3165 }
3166
elf_sym_by_idx(const struct bpf_object * obj,size_t idx)3167 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx)
3168 {
3169 if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym))
3170 return NULL;
3171
3172 return (Elf64_Sym *)obj->efile.symbols->d_buf + idx;
3173 }
3174
elf_rel_by_idx(Elf_Data * data,size_t idx)3175 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx)
3176 {
3177 if (idx >= data->d_size / sizeof(Elf64_Rel))
3178 return NULL;
3179
3180 return (Elf64_Rel *)data->d_buf + idx;
3181 }
3182
is_sec_name_dwarf(const char * name)3183 static bool is_sec_name_dwarf(const char *name)
3184 {
3185 /* approximation, but the actual list is too long */
3186 return str_has_pfx(name, ".debug_");
3187 }
3188
ignore_elf_section(Elf64_Shdr * hdr,const char * name)3189 static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name)
3190 {
3191 /* no special handling of .strtab */
3192 if (hdr->sh_type == SHT_STRTAB)
3193 return true;
3194
3195 /* ignore .llvm_addrsig section as well */
3196 if (hdr->sh_type == SHT_LLVM_ADDRSIG)
3197 return true;
3198
3199 /* no subprograms will lead to an empty .text section, ignore it */
3200 if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 &&
3201 strcmp(name, ".text") == 0)
3202 return true;
3203
3204 /* DWARF sections */
3205 if (is_sec_name_dwarf(name))
3206 return true;
3207
3208 if (str_has_pfx(name, ".rel")) {
3209 name += sizeof(".rel") - 1;
3210 /* DWARF section relocations */
3211 if (is_sec_name_dwarf(name))
3212 return true;
3213
3214 /* .BTF and .BTF.ext don't need relocations */
3215 if (strcmp(name, BTF_ELF_SEC) == 0 ||
3216 strcmp(name, BTF_EXT_ELF_SEC) == 0)
3217 return true;
3218 }
3219
3220 return false;
3221 }
3222
cmp_progs(const void * _a,const void * _b)3223 static int cmp_progs(const void *_a, const void *_b)
3224 {
3225 const struct bpf_program *a = _a;
3226 const struct bpf_program *b = _b;
3227
3228 if (a->sec_idx != b->sec_idx)
3229 return a->sec_idx < b->sec_idx ? -1 : 1;
3230
3231 /* sec_insn_off can't be the same within the section */
3232 return a->sec_insn_off < b->sec_insn_off ? -1 : 1;
3233 }
3234
bpf_object__elf_collect(struct bpf_object * obj)3235 static int bpf_object__elf_collect(struct bpf_object *obj)
3236 {
3237 struct elf_sec_desc *sec_desc;
3238 Elf *elf = obj->efile.elf;
3239 Elf_Data *btf_ext_data = NULL;
3240 Elf_Data *btf_data = NULL;
3241 int idx = 0, err = 0;
3242 const char *name;
3243 Elf_Data *data;
3244 Elf_Scn *scn;
3245 Elf64_Shdr *sh;
3246
3247 /* ELF section indices are 0-based, but sec #0 is special "invalid"
3248 * section. e_shnum does include sec #0, so e_shnum is the necessary
3249 * size of an array to keep all the sections.
3250 */
3251 obj->efile.sec_cnt = obj->efile.ehdr->e_shnum;
3252 obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs));
3253 if (!obj->efile.secs)
3254 return -ENOMEM;
3255
3256 /* a bunch of ELF parsing functionality depends on processing symbols,
3257 * so do the first pass and find the symbol table
3258 */
3259 scn = NULL;
3260 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3261 sh = elf_sec_hdr(obj, scn);
3262 if (!sh)
3263 return -LIBBPF_ERRNO__FORMAT;
3264
3265 if (sh->sh_type == SHT_SYMTAB) {
3266 if (obj->efile.symbols) {
3267 pr_warn("elf: multiple symbol tables in %s\n", obj->path);
3268 return -LIBBPF_ERRNO__FORMAT;
3269 }
3270
3271 data = elf_sec_data(obj, scn);
3272 if (!data)
3273 return -LIBBPF_ERRNO__FORMAT;
3274
3275 idx = elf_ndxscn(scn);
3276
3277 obj->efile.symbols = data;
3278 obj->efile.symbols_shndx = idx;
3279 obj->efile.strtabidx = sh->sh_link;
3280 }
3281 }
3282
3283 if (!obj->efile.symbols) {
3284 pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n",
3285 obj->path);
3286 return -ENOENT;
3287 }
3288
3289 scn = NULL;
3290 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3291 idx = elf_ndxscn(scn);
3292 sec_desc = &obj->efile.secs[idx];
3293
3294 sh = elf_sec_hdr(obj, scn);
3295 if (!sh)
3296 return -LIBBPF_ERRNO__FORMAT;
3297
3298 name = elf_sec_str(obj, sh->sh_name);
3299 if (!name)
3300 return -LIBBPF_ERRNO__FORMAT;
3301
3302 if (ignore_elf_section(sh, name))
3303 continue;
3304
3305 data = elf_sec_data(obj, scn);
3306 if (!data)
3307 return -LIBBPF_ERRNO__FORMAT;
3308
3309 pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
3310 idx, name, (unsigned long)data->d_size,
3311 (int)sh->sh_link, (unsigned long)sh->sh_flags,
3312 (int)sh->sh_type);
3313
3314 if (strcmp(name, "license") == 0) {
3315 err = bpf_object__init_license(obj, data->d_buf, data->d_size);
3316 if (err)
3317 return err;
3318 } else if (strcmp(name, "version") == 0) {
3319 err = bpf_object__init_kversion(obj, data->d_buf, data->d_size);
3320 if (err)
3321 return err;
3322 } else if (strcmp(name, "maps") == 0) {
3323 obj->efile.maps_shndx = idx;
3324 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
3325 obj->efile.btf_maps_shndx = idx;
3326 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
3327 if (sh->sh_type != SHT_PROGBITS)
3328 return -LIBBPF_ERRNO__FORMAT;
3329 btf_data = data;
3330 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
3331 if (sh->sh_type != SHT_PROGBITS)
3332 return -LIBBPF_ERRNO__FORMAT;
3333 btf_ext_data = data;
3334 } else if (sh->sh_type == SHT_SYMTAB) {
3335 /* already processed during the first pass above */
3336 } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) {
3337 if (sh->sh_flags & SHF_EXECINSTR) {
3338 if (strcmp(name, ".text") == 0)
3339 obj->efile.text_shndx = idx;
3340 err = bpf_object__add_programs(obj, data, name, idx);
3341 if (err)
3342 return err;
3343 } else if (strcmp(name, DATA_SEC) == 0 ||
3344 str_has_pfx(name, DATA_SEC ".")) {
3345 sec_desc->sec_type = SEC_DATA;
3346 sec_desc->shdr = sh;
3347 sec_desc->data = data;
3348 } else if (strcmp(name, RODATA_SEC) == 0 ||
3349 str_has_pfx(name, RODATA_SEC ".")) {
3350 sec_desc->sec_type = SEC_RODATA;
3351 sec_desc->shdr = sh;
3352 sec_desc->data = data;
3353 } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
3354 obj->efile.st_ops_data = data;
3355 obj->efile.st_ops_shndx = idx;
3356 } else {
3357 pr_info("elf: skipping unrecognized data section(%d) %s\n",
3358 idx, name);
3359 }
3360 } else if (sh->sh_type == SHT_REL) {
3361 int targ_sec_idx = sh->sh_info; /* points to other section */
3362
3363 if (sh->sh_entsize != sizeof(Elf64_Rel) ||
3364 targ_sec_idx >= obj->efile.sec_cnt)
3365 return -LIBBPF_ERRNO__FORMAT;
3366
3367 /* Only do relo for section with exec instructions */
3368 if (!section_have_execinstr(obj, targ_sec_idx) &&
3369 strcmp(name, ".rel" STRUCT_OPS_SEC) &&
3370 strcmp(name, ".rel" MAPS_ELF_SEC)) {
3371 pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n",
3372 idx, name, targ_sec_idx,
3373 elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>");
3374 continue;
3375 }
3376
3377 sec_desc->sec_type = SEC_RELO;
3378 sec_desc->shdr = sh;
3379 sec_desc->data = data;
3380 } else if (sh->sh_type == SHT_NOBITS && strcmp(name, BSS_SEC) == 0) {
3381 sec_desc->sec_type = SEC_BSS;
3382 sec_desc->shdr = sh;
3383 sec_desc->data = data;
3384 } else {
3385 pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name,
3386 (size_t)sh->sh_size);
3387 }
3388 }
3389
3390 if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
3391 pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path);
3392 return -LIBBPF_ERRNO__FORMAT;
3393 }
3394
3395 /* sort BPF programs by section name and in-section instruction offset
3396 * for faster search */
3397 if (obj->nr_programs)
3398 qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs);
3399
3400 return bpf_object__init_btf(obj, btf_data, btf_ext_data);
3401 }
3402
sym_is_extern(const Elf64_Sym * sym)3403 static bool sym_is_extern(const Elf64_Sym *sym)
3404 {
3405 int bind = ELF64_ST_BIND(sym->st_info);
3406 /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
3407 return sym->st_shndx == SHN_UNDEF &&
3408 (bind == STB_GLOBAL || bind == STB_WEAK) &&
3409 ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE;
3410 }
3411
sym_is_subprog(const Elf64_Sym * sym,int text_shndx)3412 static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx)
3413 {
3414 int bind = ELF64_ST_BIND(sym->st_info);
3415 int type = ELF64_ST_TYPE(sym->st_info);
3416
3417 /* in .text section */
3418 if (sym->st_shndx != text_shndx)
3419 return false;
3420
3421 /* local function */
3422 if (bind == STB_LOCAL && type == STT_SECTION)
3423 return true;
3424
3425 /* global function */
3426 return bind == STB_GLOBAL && type == STT_FUNC;
3427 }
3428
find_extern_btf_id(const struct btf * btf,const char * ext_name)3429 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
3430 {
3431 const struct btf_type *t;
3432 const char *tname;
3433 int i, n;
3434
3435 if (!btf)
3436 return -ESRCH;
3437
3438 n = btf__type_cnt(btf);
3439 for (i = 1; i < n; i++) {
3440 t = btf__type_by_id(btf, i);
3441
3442 if (!btf_is_var(t) && !btf_is_func(t))
3443 continue;
3444
3445 tname = btf__name_by_offset(btf, t->name_off);
3446 if (strcmp(tname, ext_name))
3447 continue;
3448
3449 if (btf_is_var(t) &&
3450 btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
3451 return -EINVAL;
3452
3453 if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN)
3454 return -EINVAL;
3455
3456 return i;
3457 }
3458
3459 return -ENOENT;
3460 }
3461
find_extern_sec_btf_id(struct btf * btf,int ext_btf_id)3462 static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
3463 const struct btf_var_secinfo *vs;
3464 const struct btf_type *t;
3465 int i, j, n;
3466
3467 if (!btf)
3468 return -ESRCH;
3469
3470 n = btf__type_cnt(btf);
3471 for (i = 1; i < n; i++) {
3472 t = btf__type_by_id(btf, i);
3473
3474 if (!btf_is_datasec(t))
3475 continue;
3476
3477 vs = btf_var_secinfos(t);
3478 for (j = 0; j < btf_vlen(t); j++, vs++) {
3479 if (vs->type == ext_btf_id)
3480 return i;
3481 }
3482 }
3483
3484 return -ENOENT;
3485 }
3486
find_kcfg_type(const struct btf * btf,int id,bool * is_signed)3487 static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
3488 bool *is_signed)
3489 {
3490 const struct btf_type *t;
3491 const char *name;
3492
3493 t = skip_mods_and_typedefs(btf, id, NULL);
3494 name = btf__name_by_offset(btf, t->name_off);
3495
3496 if (is_signed)
3497 *is_signed = false;
3498 switch (btf_kind(t)) {
3499 case BTF_KIND_INT: {
3500 int enc = btf_int_encoding(t);
3501
3502 if (enc & BTF_INT_BOOL)
3503 return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
3504 if (is_signed)
3505 *is_signed = enc & BTF_INT_SIGNED;
3506 if (t->size == 1)
3507 return KCFG_CHAR;
3508 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
3509 return KCFG_UNKNOWN;
3510 return KCFG_INT;
3511 }
3512 case BTF_KIND_ENUM:
3513 if (t->size != 4)
3514 return KCFG_UNKNOWN;
3515 if (strcmp(name, "libbpf_tristate"))
3516 return KCFG_UNKNOWN;
3517 return KCFG_TRISTATE;
3518 case BTF_KIND_ARRAY:
3519 if (btf_array(t)->nelems == 0)
3520 return KCFG_UNKNOWN;
3521 if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
3522 return KCFG_UNKNOWN;
3523 return KCFG_CHAR_ARR;
3524 default:
3525 return KCFG_UNKNOWN;
3526 }
3527 }
3528
cmp_externs(const void * _a,const void * _b)3529 static int cmp_externs(const void *_a, const void *_b)
3530 {
3531 const struct extern_desc *a = _a;
3532 const struct extern_desc *b = _b;
3533
3534 if (a->type != b->type)
3535 return a->type < b->type ? -1 : 1;
3536
3537 if (a->type == EXT_KCFG) {
3538 /* descending order by alignment requirements */
3539 if (a->kcfg.align != b->kcfg.align)
3540 return a->kcfg.align > b->kcfg.align ? -1 : 1;
3541 /* ascending order by size, within same alignment class */
3542 if (a->kcfg.sz != b->kcfg.sz)
3543 return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
3544 }
3545
3546 /* resolve ties by name */
3547 return strcmp(a->name, b->name);
3548 }
3549
find_int_btf_id(const struct btf * btf)3550 static int find_int_btf_id(const struct btf *btf)
3551 {
3552 const struct btf_type *t;
3553 int i, n;
3554
3555 n = btf__type_cnt(btf);
3556 for (i = 1; i < n; i++) {
3557 t = btf__type_by_id(btf, i);
3558
3559 if (btf_is_int(t) && btf_int_bits(t) == 32)
3560 return i;
3561 }
3562
3563 return 0;
3564 }
3565
add_dummy_ksym_var(struct btf * btf)3566 static int add_dummy_ksym_var(struct btf *btf)
3567 {
3568 int i, int_btf_id, sec_btf_id, dummy_var_btf_id;
3569 const struct btf_var_secinfo *vs;
3570 const struct btf_type *sec;
3571
3572 if (!btf)
3573 return 0;
3574
3575 sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC,
3576 BTF_KIND_DATASEC);
3577 if (sec_btf_id < 0)
3578 return 0;
3579
3580 sec = btf__type_by_id(btf, sec_btf_id);
3581 vs = btf_var_secinfos(sec);
3582 for (i = 0; i < btf_vlen(sec); i++, vs++) {
3583 const struct btf_type *vt;
3584
3585 vt = btf__type_by_id(btf, vs->type);
3586 if (btf_is_func(vt))
3587 break;
3588 }
3589
3590 /* No func in ksyms sec. No need to add dummy var. */
3591 if (i == btf_vlen(sec))
3592 return 0;
3593
3594 int_btf_id = find_int_btf_id(btf);
3595 dummy_var_btf_id = btf__add_var(btf,
3596 "dummy_ksym",
3597 BTF_VAR_GLOBAL_ALLOCATED,
3598 int_btf_id);
3599 if (dummy_var_btf_id < 0)
3600 pr_warn("cannot create a dummy_ksym var\n");
3601
3602 return dummy_var_btf_id;
3603 }
3604
bpf_object__collect_externs(struct bpf_object * obj)3605 static int bpf_object__collect_externs(struct bpf_object *obj)
3606 {
3607 struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
3608 const struct btf_type *t;
3609 struct extern_desc *ext;
3610 int i, n, off, dummy_var_btf_id;
3611 const char *ext_name, *sec_name;
3612 Elf_Scn *scn;
3613 Elf64_Shdr *sh;
3614
3615 if (!obj->efile.symbols)
3616 return 0;
3617
3618 scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
3619 sh = elf_sec_hdr(obj, scn);
3620 if (!sh || sh->sh_entsize != sizeof(Elf64_Sym))
3621 return -LIBBPF_ERRNO__FORMAT;
3622
3623 dummy_var_btf_id = add_dummy_ksym_var(obj->btf);
3624 if (dummy_var_btf_id < 0)
3625 return dummy_var_btf_id;
3626
3627 n = sh->sh_size / sh->sh_entsize;
3628 pr_debug("looking for externs among %d symbols...\n", n);
3629
3630 for (i = 0; i < n; i++) {
3631 Elf64_Sym *sym = elf_sym_by_idx(obj, i);
3632
3633 if (!sym)
3634 return -LIBBPF_ERRNO__FORMAT;
3635 if (!sym_is_extern(sym))
3636 continue;
3637 ext_name = elf_sym_str(obj, sym->st_name);
3638 if (!ext_name || !ext_name[0])
3639 continue;
3640
3641 ext = obj->externs;
3642 ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
3643 if (!ext)
3644 return -ENOMEM;
3645 obj->externs = ext;
3646 ext = &ext[obj->nr_extern];
3647 memset(ext, 0, sizeof(*ext));
3648 obj->nr_extern++;
3649
3650 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
3651 if (ext->btf_id <= 0) {
3652 pr_warn("failed to find BTF for extern '%s': %d\n",
3653 ext_name, ext->btf_id);
3654 return ext->btf_id;
3655 }
3656 t = btf__type_by_id(obj->btf, ext->btf_id);
3657 ext->name = btf__name_by_offset(obj->btf, t->name_off);
3658 ext->sym_idx = i;
3659 ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK;
3660
3661 ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
3662 if (ext->sec_btf_id <= 0) {
3663 pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
3664 ext_name, ext->btf_id, ext->sec_btf_id);
3665 return ext->sec_btf_id;
3666 }
3667 sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
3668 sec_name = btf__name_by_offset(obj->btf, sec->name_off);
3669
3670 if (strcmp(sec_name, KCONFIG_SEC) == 0) {
3671 if (btf_is_func(t)) {
3672 pr_warn("extern function %s is unsupported under %s section\n",
3673 ext->name, KCONFIG_SEC);
3674 return -ENOTSUP;
3675 }
3676 kcfg_sec = sec;
3677 ext->type = EXT_KCFG;
3678 ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
3679 if (ext->kcfg.sz <= 0) {
3680 pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
3681 ext_name, ext->kcfg.sz);
3682 return ext->kcfg.sz;
3683 }
3684 ext->kcfg.align = btf__align_of(obj->btf, t->type);
3685 if (ext->kcfg.align <= 0) {
3686 pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
3687 ext_name, ext->kcfg.align);
3688 return -EINVAL;
3689 }
3690 ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
3691 &ext->kcfg.is_signed);
3692 if (ext->kcfg.type == KCFG_UNKNOWN) {
3693 pr_warn("extern (kcfg) '%s' type is unsupported\n", ext_name);
3694 return -ENOTSUP;
3695 }
3696 } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
3697 ksym_sec = sec;
3698 ext->type = EXT_KSYM;
3699 skip_mods_and_typedefs(obj->btf, t->type,
3700 &ext->ksym.type_id);
3701 } else {
3702 pr_warn("unrecognized extern section '%s'\n", sec_name);
3703 return -ENOTSUP;
3704 }
3705 }
3706 pr_debug("collected %d externs total\n", obj->nr_extern);
3707
3708 if (!obj->nr_extern)
3709 return 0;
3710
3711 /* sort externs by type, for kcfg ones also by (align, size, name) */
3712 qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
3713
3714 /* for .ksyms section, we need to turn all externs into allocated
3715 * variables in BTF to pass kernel verification; we do this by
3716 * pretending that each extern is a 8-byte variable
3717 */
3718 if (ksym_sec) {
3719 /* find existing 4-byte integer type in BTF to use for fake
3720 * extern variables in DATASEC
3721 */
3722 int int_btf_id = find_int_btf_id(obj->btf);
3723 /* For extern function, a dummy_var added earlier
3724 * will be used to replace the vs->type and
3725 * its name string will be used to refill
3726 * the missing param's name.
3727 */
3728 const struct btf_type *dummy_var;
3729
3730 dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id);
3731 for (i = 0; i < obj->nr_extern; i++) {
3732 ext = &obj->externs[i];
3733 if (ext->type != EXT_KSYM)
3734 continue;
3735 pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
3736 i, ext->sym_idx, ext->name);
3737 }
3738
3739 sec = ksym_sec;
3740 n = btf_vlen(sec);
3741 for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
3742 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3743 struct btf_type *vt;
3744
3745 vt = (void *)btf__type_by_id(obj->btf, vs->type);
3746 ext_name = btf__name_by_offset(obj->btf, vt->name_off);
3747 ext = find_extern_by_name(obj, ext_name);
3748 if (!ext) {
3749 pr_warn("failed to find extern definition for BTF %s '%s'\n",
3750 btf_kind_str(vt), ext_name);
3751 return -ESRCH;
3752 }
3753 if (btf_is_func(vt)) {
3754 const struct btf_type *func_proto;
3755 struct btf_param *param;
3756 int j;
3757
3758 func_proto = btf__type_by_id(obj->btf,
3759 vt->type);
3760 param = btf_params(func_proto);
3761 /* Reuse the dummy_var string if the
3762 * func proto does not have param name.
3763 */
3764 for (j = 0; j < btf_vlen(func_proto); j++)
3765 if (param[j].type && !param[j].name_off)
3766 param[j].name_off =
3767 dummy_var->name_off;
3768 vs->type = dummy_var_btf_id;
3769 vt->info &= ~0xffff;
3770 vt->info |= BTF_FUNC_GLOBAL;
3771 } else {
3772 btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3773 vt->type = int_btf_id;
3774 }
3775 vs->offset = off;
3776 vs->size = sizeof(int);
3777 }
3778 sec->size = off;
3779 }
3780
3781 if (kcfg_sec) {
3782 sec = kcfg_sec;
3783 /* for kcfg externs calculate their offsets within a .kconfig map */
3784 off = 0;
3785 for (i = 0; i < obj->nr_extern; i++) {
3786 ext = &obj->externs[i];
3787 if (ext->type != EXT_KCFG)
3788 continue;
3789
3790 ext->kcfg.data_off = roundup(off, ext->kcfg.align);
3791 off = ext->kcfg.data_off + ext->kcfg.sz;
3792 pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
3793 i, ext->sym_idx, ext->kcfg.data_off, ext->name);
3794 }
3795 sec->size = off;
3796 n = btf_vlen(sec);
3797 for (i = 0; i < n; i++) {
3798 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3799
3800 t = btf__type_by_id(obj->btf, vs->type);
3801 ext_name = btf__name_by_offset(obj->btf, t->name_off);
3802 ext = find_extern_by_name(obj, ext_name);
3803 if (!ext) {
3804 pr_warn("failed to find extern definition for BTF var '%s'\n",
3805 ext_name);
3806 return -ESRCH;
3807 }
3808 btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3809 vs->offset = ext->kcfg.data_off;
3810 }
3811 }
3812 return 0;
3813 }
3814
3815 struct bpf_program *
bpf_object__find_program_by_title(const struct bpf_object * obj,const char * title)3816 bpf_object__find_program_by_title(const struct bpf_object *obj,
3817 const char *title)
3818 {
3819 struct bpf_program *pos;
3820
3821 bpf_object__for_each_program(pos, obj) {
3822 if (pos->sec_name && !strcmp(pos->sec_name, title))
3823 return pos;
3824 }
3825 return errno = ENOENT, NULL;
3826 }
3827
prog_is_subprog(const struct bpf_object * obj,const struct bpf_program * prog)3828 static bool prog_is_subprog(const struct bpf_object *obj,
3829 const struct bpf_program *prog)
3830 {
3831 /* For legacy reasons, libbpf supports an entry-point BPF programs
3832 * without SEC() attribute, i.e., those in the .text section. But if
3833 * there are 2 or more such programs in the .text section, they all
3834 * must be subprograms called from entry-point BPF programs in
3835 * designated SEC()'tions, otherwise there is no way to distinguish
3836 * which of those programs should be loaded vs which are a subprogram.
3837 * Similarly, if there is a function/program in .text and at least one
3838 * other BPF program with custom SEC() attribute, then we just assume
3839 * .text programs are subprograms (even if they are not called from
3840 * other programs), because libbpf never explicitly supported mixing
3841 * SEC()-designated BPF programs and .text entry-point BPF programs.
3842 */
3843 return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1;
3844 }
3845
3846 struct bpf_program *
bpf_object__find_program_by_name(const struct bpf_object * obj,const char * name)3847 bpf_object__find_program_by_name(const struct bpf_object *obj,
3848 const char *name)
3849 {
3850 struct bpf_program *prog;
3851
3852 bpf_object__for_each_program(prog, obj) {
3853 if (prog_is_subprog(obj, prog))
3854 continue;
3855 if (!strcmp(prog->name, name))
3856 return prog;
3857 }
3858 return errno = ENOENT, NULL;
3859 }
3860
bpf_object__shndx_is_data(const struct bpf_object * obj,int shndx)3861 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
3862 int shndx)
3863 {
3864 switch (obj->efile.secs[shndx].sec_type) {
3865 case SEC_BSS:
3866 case SEC_DATA:
3867 case SEC_RODATA:
3868 return true;
3869 default:
3870 return false;
3871 }
3872 }
3873
bpf_object__shndx_is_maps(const struct bpf_object * obj,int shndx)3874 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
3875 int shndx)
3876 {
3877 return shndx == obj->efile.maps_shndx ||
3878 shndx == obj->efile.btf_maps_shndx;
3879 }
3880
3881 static enum libbpf_map_type
bpf_object__section_to_libbpf_map_type(const struct bpf_object * obj,int shndx)3882 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
3883 {
3884 if (shndx == obj->efile.symbols_shndx)
3885 return LIBBPF_MAP_KCONFIG;
3886
3887 switch (obj->efile.secs[shndx].sec_type) {
3888 case SEC_BSS:
3889 return LIBBPF_MAP_BSS;
3890 case SEC_DATA:
3891 return LIBBPF_MAP_DATA;
3892 case SEC_RODATA:
3893 return LIBBPF_MAP_RODATA;
3894 default:
3895 return LIBBPF_MAP_UNSPEC;
3896 }
3897 }
3898
bpf_program__record_reloc(struct bpf_program * prog,struct reloc_desc * reloc_desc,__u32 insn_idx,const char * sym_name,const Elf64_Sym * sym,const Elf64_Rel * rel)3899 static int bpf_program__record_reloc(struct bpf_program *prog,
3900 struct reloc_desc *reloc_desc,
3901 __u32 insn_idx, const char *sym_name,
3902 const Elf64_Sym *sym, const Elf64_Rel *rel)
3903 {
3904 struct bpf_insn *insn = &prog->insns[insn_idx];
3905 size_t map_idx, nr_maps = prog->obj->nr_maps;
3906 struct bpf_object *obj = prog->obj;
3907 __u32 shdr_idx = sym->st_shndx;
3908 enum libbpf_map_type type;
3909 const char *sym_sec_name;
3910 struct bpf_map *map;
3911
3912 if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) {
3913 pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n",
3914 prog->name, sym_name, insn_idx, insn->code);
3915 return -LIBBPF_ERRNO__RELOC;
3916 }
3917
3918 if (sym_is_extern(sym)) {
3919 int sym_idx = ELF64_R_SYM(rel->r_info);
3920 int i, n = obj->nr_extern;
3921 struct extern_desc *ext;
3922
3923 for (i = 0; i < n; i++) {
3924 ext = &obj->externs[i];
3925 if (ext->sym_idx == sym_idx)
3926 break;
3927 }
3928 if (i >= n) {
3929 pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n",
3930 prog->name, sym_name, sym_idx);
3931 return -LIBBPF_ERRNO__RELOC;
3932 }
3933 pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n",
3934 prog->name, i, ext->name, ext->sym_idx, insn_idx);
3935 if (insn->code == (BPF_JMP | BPF_CALL))
3936 reloc_desc->type = RELO_EXTERN_FUNC;
3937 else
3938 reloc_desc->type = RELO_EXTERN_VAR;
3939 reloc_desc->insn_idx = insn_idx;
3940 reloc_desc->sym_off = i; /* sym_off stores extern index */
3941 return 0;
3942 }
3943
3944 /* sub-program call relocation */
3945 if (is_call_insn(insn)) {
3946 if (insn->src_reg != BPF_PSEUDO_CALL) {
3947 pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name);
3948 return -LIBBPF_ERRNO__RELOC;
3949 }
3950 /* text_shndx can be 0, if no default "main" program exists */
3951 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
3952 sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
3953 pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n",
3954 prog->name, sym_name, sym_sec_name);
3955 return -LIBBPF_ERRNO__RELOC;
3956 }
3957 if (sym->st_value % BPF_INSN_SZ) {
3958 pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n",
3959 prog->name, sym_name, (size_t)sym->st_value);
3960 return -LIBBPF_ERRNO__RELOC;
3961 }
3962 reloc_desc->type = RELO_CALL;
3963 reloc_desc->insn_idx = insn_idx;
3964 reloc_desc->sym_off = sym->st_value;
3965 return 0;
3966 }
3967
3968 if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
3969 pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n",
3970 prog->name, sym_name, shdr_idx);
3971 return -LIBBPF_ERRNO__RELOC;
3972 }
3973
3974 /* loading subprog addresses */
3975 if (sym_is_subprog(sym, obj->efile.text_shndx)) {
3976 /* global_func: sym->st_value = offset in the section, insn->imm = 0.
3977 * local_func: sym->st_value = 0, insn->imm = offset in the section.
3978 */
3979 if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) {
3980 pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n",
3981 prog->name, sym_name, (size_t)sym->st_value, insn->imm);
3982 return -LIBBPF_ERRNO__RELOC;
3983 }
3984
3985 reloc_desc->type = RELO_SUBPROG_ADDR;
3986 reloc_desc->insn_idx = insn_idx;
3987 reloc_desc->sym_off = sym->st_value;
3988 return 0;
3989 }
3990
3991 type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
3992 sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
3993
3994 /* generic map reference relocation */
3995 if (type == LIBBPF_MAP_UNSPEC) {
3996 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
3997 pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n",
3998 prog->name, sym_name, sym_sec_name);
3999 return -LIBBPF_ERRNO__RELOC;
4000 }
4001 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4002 map = &obj->maps[map_idx];
4003 if (map->libbpf_type != type ||
4004 map->sec_idx != sym->st_shndx ||
4005 map->sec_offset != sym->st_value)
4006 continue;
4007 pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n",
4008 prog->name, map_idx, map->name, map->sec_idx,
4009 map->sec_offset, insn_idx);
4010 break;
4011 }
4012 if (map_idx >= nr_maps) {
4013 pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n",
4014 prog->name, sym_sec_name, (size_t)sym->st_value);
4015 return -LIBBPF_ERRNO__RELOC;
4016 }
4017 reloc_desc->type = RELO_LD64;
4018 reloc_desc->insn_idx = insn_idx;
4019 reloc_desc->map_idx = map_idx;
4020 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
4021 return 0;
4022 }
4023
4024 /* global data map relocation */
4025 if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
4026 pr_warn("prog '%s': bad data relo against section '%s'\n",
4027 prog->name, sym_sec_name);
4028 return -LIBBPF_ERRNO__RELOC;
4029 }
4030 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4031 map = &obj->maps[map_idx];
4032 if (map->libbpf_type != type || map->sec_idx != sym->st_shndx)
4033 continue;
4034 pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n",
4035 prog->name, map_idx, map->name, map->sec_idx,
4036 map->sec_offset, insn_idx);
4037 break;
4038 }
4039 if (map_idx >= nr_maps) {
4040 pr_warn("prog '%s': data relo failed to find map for section '%s'\n",
4041 prog->name, sym_sec_name);
4042 return -LIBBPF_ERRNO__RELOC;
4043 }
4044
4045 reloc_desc->type = RELO_DATA;
4046 reloc_desc->insn_idx = insn_idx;
4047 reloc_desc->map_idx = map_idx;
4048 reloc_desc->sym_off = sym->st_value;
4049 return 0;
4050 }
4051
prog_contains_insn(const struct bpf_program * prog,size_t insn_idx)4052 static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx)
4053 {
4054 return insn_idx >= prog->sec_insn_off &&
4055 insn_idx < prog->sec_insn_off + prog->sec_insn_cnt;
4056 }
4057
find_prog_by_sec_insn(const struct bpf_object * obj,size_t sec_idx,size_t insn_idx)4058 static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj,
4059 size_t sec_idx, size_t insn_idx)
4060 {
4061 int l = 0, r = obj->nr_programs - 1, m;
4062 struct bpf_program *prog;
4063
4064 while (l < r) {
4065 m = l + (r - l + 1) / 2;
4066 prog = &obj->programs[m];
4067
4068 if (prog->sec_idx < sec_idx ||
4069 (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx))
4070 l = m;
4071 else
4072 r = m - 1;
4073 }
4074 /* matching program could be at index l, but it still might be the
4075 * wrong one, so we need to double check conditions for the last time
4076 */
4077 prog = &obj->programs[l];
4078 if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx))
4079 return prog;
4080 return NULL;
4081 }
4082
4083 static int
bpf_object__collect_prog_relos(struct bpf_object * obj,Elf64_Shdr * shdr,Elf_Data * data)4084 bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data)
4085 {
4086 const char *relo_sec_name, *sec_name;
4087 size_t sec_idx = shdr->sh_info, sym_idx;
4088 struct bpf_program *prog;
4089 struct reloc_desc *relos;
4090 int err, i, nrels;
4091 const char *sym_name;
4092 __u32 insn_idx;
4093 Elf_Scn *scn;
4094 Elf_Data *scn_data;
4095 Elf64_Sym *sym;
4096 Elf64_Rel *rel;
4097
4098 if (sec_idx >= obj->efile.sec_cnt)
4099 return -EINVAL;
4100
4101 scn = elf_sec_by_idx(obj, sec_idx);
4102 scn_data = elf_sec_data(obj, scn);
4103
4104 relo_sec_name = elf_sec_str(obj, shdr->sh_name);
4105 sec_name = elf_sec_name(obj, scn);
4106 if (!relo_sec_name || !sec_name)
4107 return -EINVAL;
4108
4109 pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n",
4110 relo_sec_name, sec_idx, sec_name);
4111 nrels = shdr->sh_size / shdr->sh_entsize;
4112
4113 for (i = 0; i < nrels; i++) {
4114 rel = elf_rel_by_idx(data, i);
4115 if (!rel) {
4116 pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i);
4117 return -LIBBPF_ERRNO__FORMAT;
4118 }
4119
4120 sym_idx = ELF64_R_SYM(rel->r_info);
4121 sym = elf_sym_by_idx(obj, sym_idx);
4122 if (!sym) {
4123 pr_warn("sec '%s': symbol #%zu not found for relo #%d\n",
4124 relo_sec_name, sym_idx, i);
4125 return -LIBBPF_ERRNO__FORMAT;
4126 }
4127
4128 if (sym->st_shndx >= obj->efile.sec_cnt) {
4129 pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n",
4130 relo_sec_name, sym_idx, (size_t)sym->st_shndx, i);
4131 return -LIBBPF_ERRNO__FORMAT;
4132 }
4133
4134 if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) {
4135 pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n",
4136 relo_sec_name, (size_t)rel->r_offset, i);
4137 return -LIBBPF_ERRNO__FORMAT;
4138 }
4139
4140 insn_idx = rel->r_offset / BPF_INSN_SZ;
4141 /* relocations against static functions are recorded as
4142 * relocations against the section that contains a function;
4143 * in such case, symbol will be STT_SECTION and sym.st_name
4144 * will point to empty string (0), so fetch section name
4145 * instead
4146 */
4147 if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0)
4148 sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx));
4149 else
4150 sym_name = elf_sym_str(obj, sym->st_name);
4151 sym_name = sym_name ?: "<?";
4152
4153 pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n",
4154 relo_sec_name, i, insn_idx, sym_name);
4155
4156 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
4157 if (!prog) {
4158 pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n",
4159 relo_sec_name, i, sec_name, insn_idx);
4160 continue;
4161 }
4162
4163 relos = libbpf_reallocarray(prog->reloc_desc,
4164 prog->nr_reloc + 1, sizeof(*relos));
4165 if (!relos)
4166 return -ENOMEM;
4167 prog->reloc_desc = relos;
4168
4169 /* adjust insn_idx to local BPF program frame of reference */
4170 insn_idx -= prog->sec_insn_off;
4171 err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc],
4172 insn_idx, sym_name, sym, rel);
4173 if (err)
4174 return err;
4175
4176 prog->nr_reloc++;
4177 }
4178 return 0;
4179 }
4180
bpf_map_find_btf_info(struct bpf_object * obj,struct bpf_map * map)4181 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
4182 {
4183 struct bpf_map_def *def = &map->def;
4184 __u32 key_type_id = 0, value_type_id = 0;
4185 int ret;
4186
4187 /* if it's BTF-defined map, we don't need to search for type IDs.
4188 * For struct_ops map, it does not need btf_key_type_id and
4189 * btf_value_type_id.
4190 */
4191 if (map->sec_idx == obj->efile.btf_maps_shndx ||
4192 bpf_map__is_struct_ops(map))
4193 return 0;
4194
4195 if (!bpf_map__is_internal(map)) {
4196 pr_warn("Use of BPF_ANNOTATE_KV_PAIR is deprecated, use BTF-defined maps in .maps section instead\n");
4197 #pragma GCC diagnostic push
4198 #pragma GCC diagnostic ignored "-Wdeprecated-declarations"
4199 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
4200 def->value_size, &key_type_id,
4201 &value_type_id);
4202 #pragma GCC diagnostic pop
4203 } else {
4204 /*
4205 * LLVM annotates global data differently in BTF, that is,
4206 * only as '.data', '.bss' or '.rodata'.
4207 */
4208 ret = btf__find_by_name(obj->btf, map->real_name);
4209 }
4210 if (ret < 0)
4211 return ret;
4212
4213 map->btf_key_type_id = key_type_id;
4214 map->btf_value_type_id = bpf_map__is_internal(map) ?
4215 ret : value_type_id;
4216 return 0;
4217 }
4218
bpf_get_map_info_from_fdinfo(int fd,struct bpf_map_info * info)4219 static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info)
4220 {
4221 char file[PATH_MAX], buff[4096];
4222 FILE *fp;
4223 __u32 val;
4224 int err;
4225
4226 snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
4227 memset(info, 0, sizeof(*info));
4228
4229 fp = fopen(file, "r");
4230 if (!fp) {
4231 err = -errno;
4232 pr_warn("failed to open %s: %d. No procfs support?\n", file,
4233 err);
4234 return err;
4235 }
4236
4237 while (fgets(buff, sizeof(buff), fp)) {
4238 if (sscanf(buff, "map_type:\t%u", &val) == 1)
4239 info->type = val;
4240 else if (sscanf(buff, "key_size:\t%u", &val) == 1)
4241 info->key_size = val;
4242 else if (sscanf(buff, "value_size:\t%u", &val) == 1)
4243 info->value_size = val;
4244 else if (sscanf(buff, "max_entries:\t%u", &val) == 1)
4245 info->max_entries = val;
4246 else if (sscanf(buff, "map_flags:\t%i", &val) == 1)
4247 info->map_flags = val;
4248 }
4249
4250 fclose(fp);
4251
4252 return 0;
4253 }
4254
bpf_map__reuse_fd(struct bpf_map * map,int fd)4255 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
4256 {
4257 struct bpf_map_info info = {};
4258 __u32 len = sizeof(info);
4259 int new_fd, err;
4260 char *new_name;
4261
4262 err = bpf_obj_get_info_by_fd(fd, &info, &len);
4263 if (err && errno == EINVAL)
4264 err = bpf_get_map_info_from_fdinfo(fd, &info);
4265 if (err)
4266 return libbpf_err(err);
4267
4268 new_name = strdup(info.name);
4269 if (!new_name)
4270 return libbpf_err(-errno);
4271
4272 new_fd = open("/", O_RDONLY | O_CLOEXEC);
4273 if (new_fd < 0) {
4274 err = -errno;
4275 goto err_free_new_name;
4276 }
4277
4278 new_fd = dup3(fd, new_fd, O_CLOEXEC);
4279 if (new_fd < 0) {
4280 err = -errno;
4281 goto err_close_new_fd;
4282 }
4283
4284 err = zclose(map->fd);
4285 if (err) {
4286 err = -errno;
4287 goto err_close_new_fd;
4288 }
4289 free(map->name);
4290
4291 map->fd = new_fd;
4292 map->name = new_name;
4293 map->def.type = info.type;
4294 map->def.key_size = info.key_size;
4295 map->def.value_size = info.value_size;
4296 map->def.max_entries = info.max_entries;
4297 map->def.map_flags = info.map_flags;
4298 map->btf_key_type_id = info.btf_key_type_id;
4299 map->btf_value_type_id = info.btf_value_type_id;
4300 map->reused = true;
4301 map->map_extra = info.map_extra;
4302
4303 return 0;
4304
4305 err_close_new_fd:
4306 close(new_fd);
4307 err_free_new_name:
4308 free(new_name);
4309 return libbpf_err(err);
4310 }
4311
bpf_map__max_entries(const struct bpf_map * map)4312 __u32 bpf_map__max_entries(const struct bpf_map *map)
4313 {
4314 return map->def.max_entries;
4315 }
4316
bpf_map__inner_map(struct bpf_map * map)4317 struct bpf_map *bpf_map__inner_map(struct bpf_map *map)
4318 {
4319 if (!bpf_map_type__is_map_in_map(map->def.type))
4320 return errno = EINVAL, NULL;
4321
4322 return map->inner_map;
4323 }
4324
bpf_map__set_max_entries(struct bpf_map * map,__u32 max_entries)4325 int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
4326 {
4327 if (map->fd >= 0)
4328 return libbpf_err(-EBUSY);
4329 map->def.max_entries = max_entries;
4330 return 0;
4331 }
4332
bpf_map__resize(struct bpf_map * map,__u32 max_entries)4333 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
4334 {
4335 if (!map || !max_entries)
4336 return libbpf_err(-EINVAL);
4337
4338 return bpf_map__set_max_entries(map, max_entries);
4339 }
4340
4341 static int
bpf_object__probe_loading(struct bpf_object * obj)4342 bpf_object__probe_loading(struct bpf_object *obj)
4343 {
4344 char *cp, errmsg[STRERR_BUFSIZE];
4345 struct bpf_insn insns[] = {
4346 BPF_MOV64_IMM(BPF_REG_0, 0),
4347 BPF_EXIT_INSN(),
4348 };
4349 int ret, insn_cnt = ARRAY_SIZE(insns);
4350
4351 if (obj->gen_loader)
4352 return 0;
4353
4354 ret = bump_rlimit_memlock();
4355 if (ret)
4356 pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret);
4357
4358 /* make sure basic loading works */
4359 ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4360 if (ret < 0)
4361 ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL);
4362 if (ret < 0) {
4363 ret = errno;
4364 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4365 pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF "
4366 "program. Make sure your kernel supports BPF "
4367 "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is "
4368 "set to big enough value.\n", __func__, cp, ret);
4369 return -ret;
4370 }
4371 close(ret);
4372
4373 return 0;
4374 }
4375
probe_fd(int fd)4376 static int probe_fd(int fd)
4377 {
4378 if (fd >= 0)
4379 close(fd);
4380 return fd >= 0;
4381 }
4382
probe_kern_prog_name(void)4383 static int probe_kern_prog_name(void)
4384 {
4385 struct bpf_insn insns[] = {
4386 BPF_MOV64_IMM(BPF_REG_0, 0),
4387 BPF_EXIT_INSN(),
4388 };
4389 int ret, insn_cnt = ARRAY_SIZE(insns);
4390
4391 /* make sure loading with name works */
4392 ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, "test", "GPL", insns, insn_cnt, NULL);
4393 return probe_fd(ret);
4394 }
4395
probe_kern_global_data(void)4396 static int probe_kern_global_data(void)
4397 {
4398 char *cp, errmsg[STRERR_BUFSIZE];
4399 struct bpf_insn insns[] = {
4400 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
4401 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
4402 BPF_MOV64_IMM(BPF_REG_0, 0),
4403 BPF_EXIT_INSN(),
4404 };
4405 int ret, map, insn_cnt = ARRAY_SIZE(insns);
4406
4407 map = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), 32, 1, NULL);
4408 if (map < 0) {
4409 ret = -errno;
4410 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4411 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4412 __func__, cp, -ret);
4413 return ret;
4414 }
4415
4416 insns[0].imm = map;
4417
4418 ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4419 close(map);
4420 return probe_fd(ret);
4421 }
4422
probe_kern_btf(void)4423 static int probe_kern_btf(void)
4424 {
4425 static const char strs[] = "\0int";
4426 __u32 types[] = {
4427 /* int */
4428 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4429 };
4430
4431 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4432 strs, sizeof(strs)));
4433 }
4434
probe_kern_btf_func(void)4435 static int probe_kern_btf_func(void)
4436 {
4437 static const char strs[] = "\0int\0x\0a";
4438 /* void x(int a) {} */
4439 __u32 types[] = {
4440 /* int */
4441 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4442 /* FUNC_PROTO */ /* [2] */
4443 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4444 BTF_PARAM_ENC(7, 1),
4445 /* FUNC x */ /* [3] */
4446 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
4447 };
4448
4449 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4450 strs, sizeof(strs)));
4451 }
4452
probe_kern_btf_func_global(void)4453 static int probe_kern_btf_func_global(void)
4454 {
4455 static const char strs[] = "\0int\0x\0a";
4456 /* static void x(int a) {} */
4457 __u32 types[] = {
4458 /* int */
4459 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4460 /* FUNC_PROTO */ /* [2] */
4461 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4462 BTF_PARAM_ENC(7, 1),
4463 /* FUNC x BTF_FUNC_GLOBAL */ /* [3] */
4464 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
4465 };
4466
4467 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4468 strs, sizeof(strs)));
4469 }
4470
probe_kern_btf_datasec(void)4471 static int probe_kern_btf_datasec(void)
4472 {
4473 static const char strs[] = "\0x\0.data";
4474 /* static int a; */
4475 __u32 types[] = {
4476 /* int */
4477 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4478 /* VAR x */ /* [2] */
4479 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4480 BTF_VAR_STATIC,
4481 /* DATASEC val */ /* [3] */
4482 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
4483 BTF_VAR_SECINFO_ENC(2, 0, 4),
4484 };
4485
4486 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4487 strs, sizeof(strs)));
4488 }
4489
probe_kern_btf_float(void)4490 static int probe_kern_btf_float(void)
4491 {
4492 static const char strs[] = "\0float";
4493 __u32 types[] = {
4494 /* float */
4495 BTF_TYPE_FLOAT_ENC(1, 4),
4496 };
4497
4498 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4499 strs, sizeof(strs)));
4500 }
4501
probe_kern_btf_decl_tag(void)4502 static int probe_kern_btf_decl_tag(void)
4503 {
4504 static const char strs[] = "\0tag";
4505 __u32 types[] = {
4506 /* int */
4507 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4508 /* VAR x */ /* [2] */
4509 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4510 BTF_VAR_STATIC,
4511 /* attr */
4512 BTF_TYPE_DECL_TAG_ENC(1, 2, -1),
4513 };
4514
4515 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4516 strs, sizeof(strs)));
4517 }
4518
probe_kern_btf_type_tag(void)4519 static int probe_kern_btf_type_tag(void)
4520 {
4521 static const char strs[] = "\0tag";
4522 __u32 types[] = {
4523 /* int */
4524 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
4525 /* attr */
4526 BTF_TYPE_TYPE_TAG_ENC(1, 1), /* [2] */
4527 /* ptr */
4528 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_PTR, 0, 0), 2), /* [3] */
4529 };
4530
4531 return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4532 strs, sizeof(strs)));
4533 }
4534
probe_kern_array_mmap(void)4535 static int probe_kern_array_mmap(void)
4536 {
4537 LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE);
4538 int fd;
4539
4540 fd = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), sizeof(int), 1, &opts);
4541 return probe_fd(fd);
4542 }
4543
probe_kern_exp_attach_type(void)4544 static int probe_kern_exp_attach_type(void)
4545 {
4546 LIBBPF_OPTS(bpf_prog_load_opts, opts, .expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE);
4547 struct bpf_insn insns[] = {
4548 BPF_MOV64_IMM(BPF_REG_0, 0),
4549 BPF_EXIT_INSN(),
4550 };
4551 int fd, insn_cnt = ARRAY_SIZE(insns);
4552
4553 /* use any valid combination of program type and (optional)
4554 * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS)
4555 * to see if kernel supports expected_attach_type field for
4556 * BPF_PROG_LOAD command
4557 */
4558 fd = bpf_prog_load(BPF_PROG_TYPE_CGROUP_SOCK, NULL, "GPL", insns, insn_cnt, &opts);
4559 return probe_fd(fd);
4560 }
4561
probe_kern_probe_read_kernel(void)4562 static int probe_kern_probe_read_kernel(void)
4563 {
4564 struct bpf_insn insns[] = {
4565 BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), /* r1 = r10 (fp) */
4566 BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8), /* r1 += -8 */
4567 BPF_MOV64_IMM(BPF_REG_2, 8), /* r2 = 8 */
4568 BPF_MOV64_IMM(BPF_REG_3, 0), /* r3 = 0 */
4569 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel),
4570 BPF_EXIT_INSN(),
4571 };
4572 int fd, insn_cnt = ARRAY_SIZE(insns);
4573
4574 fd = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL, "GPL", insns, insn_cnt, NULL);
4575 return probe_fd(fd);
4576 }
4577
probe_prog_bind_map(void)4578 static int probe_prog_bind_map(void)
4579 {
4580 char *cp, errmsg[STRERR_BUFSIZE];
4581 struct bpf_insn insns[] = {
4582 BPF_MOV64_IMM(BPF_REG_0, 0),
4583 BPF_EXIT_INSN(),
4584 };
4585 int ret, map, prog, insn_cnt = ARRAY_SIZE(insns);
4586
4587 map = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), 32, 1, NULL);
4588 if (map < 0) {
4589 ret = -errno;
4590 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4591 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4592 __func__, cp, -ret);
4593 return ret;
4594 }
4595
4596 prog = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4597 if (prog < 0) {
4598 close(map);
4599 return 0;
4600 }
4601
4602 ret = bpf_prog_bind_map(prog, map, NULL);
4603
4604 close(map);
4605 close(prog);
4606
4607 return ret >= 0;
4608 }
4609
probe_module_btf(void)4610 static int probe_module_btf(void)
4611 {
4612 static const char strs[] = "\0int";
4613 __u32 types[] = {
4614 /* int */
4615 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4616 };
4617 struct bpf_btf_info info;
4618 __u32 len = sizeof(info);
4619 char name[16];
4620 int fd, err;
4621
4622 fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs));
4623 if (fd < 0)
4624 return 0; /* BTF not supported at all */
4625
4626 memset(&info, 0, sizeof(info));
4627 info.name = ptr_to_u64(name);
4628 info.name_len = sizeof(name);
4629
4630 /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer;
4631 * kernel's module BTF support coincides with support for
4632 * name/name_len fields in struct bpf_btf_info.
4633 */
4634 err = bpf_obj_get_info_by_fd(fd, &info, &len);
4635 close(fd);
4636 return !err;
4637 }
4638
probe_perf_link(void)4639 static int probe_perf_link(void)
4640 {
4641 struct bpf_insn insns[] = {
4642 BPF_MOV64_IMM(BPF_REG_0, 0),
4643 BPF_EXIT_INSN(),
4644 };
4645 int prog_fd, link_fd, err;
4646
4647 prog_fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL",
4648 insns, ARRAY_SIZE(insns), NULL);
4649 if (prog_fd < 0)
4650 return -errno;
4651
4652 /* use invalid perf_event FD to get EBADF, if link is supported;
4653 * otherwise EINVAL should be returned
4654 */
4655 link_fd = bpf_link_create(prog_fd, -1, BPF_PERF_EVENT, NULL);
4656 err = -errno; /* close() can clobber errno */
4657
4658 if (link_fd >= 0)
4659 close(link_fd);
4660 close(prog_fd);
4661
4662 return link_fd < 0 && err == -EBADF;
4663 }
4664
4665 enum kern_feature_result {
4666 FEAT_UNKNOWN = 0,
4667 FEAT_SUPPORTED = 1,
4668 FEAT_MISSING = 2,
4669 };
4670
4671 typedef int (*feature_probe_fn)(void);
4672
4673 static struct kern_feature_desc {
4674 const char *desc;
4675 feature_probe_fn probe;
4676 enum kern_feature_result res;
4677 } feature_probes[__FEAT_CNT] = {
4678 [FEAT_PROG_NAME] = {
4679 "BPF program name", probe_kern_prog_name,
4680 },
4681 [FEAT_GLOBAL_DATA] = {
4682 "global variables", probe_kern_global_data,
4683 },
4684 [FEAT_BTF] = {
4685 "minimal BTF", probe_kern_btf,
4686 },
4687 [FEAT_BTF_FUNC] = {
4688 "BTF functions", probe_kern_btf_func,
4689 },
4690 [FEAT_BTF_GLOBAL_FUNC] = {
4691 "BTF global function", probe_kern_btf_func_global,
4692 },
4693 [FEAT_BTF_DATASEC] = {
4694 "BTF data section and variable", probe_kern_btf_datasec,
4695 },
4696 [FEAT_ARRAY_MMAP] = {
4697 "ARRAY map mmap()", probe_kern_array_mmap,
4698 },
4699 [FEAT_EXP_ATTACH_TYPE] = {
4700 "BPF_PROG_LOAD expected_attach_type attribute",
4701 probe_kern_exp_attach_type,
4702 },
4703 [FEAT_PROBE_READ_KERN] = {
4704 "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel,
4705 },
4706 [FEAT_PROG_BIND_MAP] = {
4707 "BPF_PROG_BIND_MAP support", probe_prog_bind_map,
4708 },
4709 [FEAT_MODULE_BTF] = {
4710 "module BTF support", probe_module_btf,
4711 },
4712 [FEAT_BTF_FLOAT] = {
4713 "BTF_KIND_FLOAT support", probe_kern_btf_float,
4714 },
4715 [FEAT_PERF_LINK] = {
4716 "BPF perf link support", probe_perf_link,
4717 },
4718 [FEAT_BTF_DECL_TAG] = {
4719 "BTF_KIND_DECL_TAG support", probe_kern_btf_decl_tag,
4720 },
4721 [FEAT_BTF_TYPE_TAG] = {
4722 "BTF_KIND_TYPE_TAG support", probe_kern_btf_type_tag,
4723 },
4724 [FEAT_MEMCG_ACCOUNT] = {
4725 "memcg-based memory accounting", probe_memcg_account,
4726 },
4727 };
4728
kernel_supports(const struct bpf_object * obj,enum kern_feature_id feat_id)4729 bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id)
4730 {
4731 struct kern_feature_desc *feat = &feature_probes[feat_id];
4732 int ret;
4733
4734 if (obj && obj->gen_loader)
4735 /* To generate loader program assume the latest kernel
4736 * to avoid doing extra prog_load, map_create syscalls.
4737 */
4738 return true;
4739
4740 if (READ_ONCE(feat->res) == FEAT_UNKNOWN) {
4741 ret = feat->probe();
4742 if (ret > 0) {
4743 WRITE_ONCE(feat->res, FEAT_SUPPORTED);
4744 } else if (ret == 0) {
4745 WRITE_ONCE(feat->res, FEAT_MISSING);
4746 } else {
4747 pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret);
4748 WRITE_ONCE(feat->res, FEAT_MISSING);
4749 }
4750 }
4751
4752 return READ_ONCE(feat->res) == FEAT_SUPPORTED;
4753 }
4754
map_is_reuse_compat(const struct bpf_map * map,int map_fd)4755 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
4756 {
4757 struct bpf_map_info map_info = {};
4758 char msg[STRERR_BUFSIZE];
4759 __u32 map_info_len;
4760 int err;
4761
4762 map_info_len = sizeof(map_info);
4763
4764 err = bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len);
4765 if (err && errno == EINVAL)
4766 err = bpf_get_map_info_from_fdinfo(map_fd, &map_info);
4767 if (err) {
4768 pr_warn("failed to get map info for map FD %d: %s\n", map_fd,
4769 libbpf_strerror_r(errno, msg, sizeof(msg)));
4770 return false;
4771 }
4772
4773 return (map_info.type == map->def.type &&
4774 map_info.key_size == map->def.key_size &&
4775 map_info.value_size == map->def.value_size &&
4776 map_info.max_entries == map->def.max_entries &&
4777 map_info.map_flags == map->def.map_flags &&
4778 map_info.map_extra == map->map_extra);
4779 }
4780
4781 static int
bpf_object__reuse_map(struct bpf_map * map)4782 bpf_object__reuse_map(struct bpf_map *map)
4783 {
4784 char *cp, errmsg[STRERR_BUFSIZE];
4785 int err, pin_fd;
4786
4787 pin_fd = bpf_obj_get(map->pin_path);
4788 if (pin_fd < 0) {
4789 err = -errno;
4790 if (err == -ENOENT) {
4791 pr_debug("found no pinned map to reuse at '%s'\n",
4792 map->pin_path);
4793 return 0;
4794 }
4795
4796 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
4797 pr_warn("couldn't retrieve pinned map '%s': %s\n",
4798 map->pin_path, cp);
4799 return err;
4800 }
4801
4802 if (!map_is_reuse_compat(map, pin_fd)) {
4803 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
4804 map->pin_path);
4805 close(pin_fd);
4806 return -EINVAL;
4807 }
4808
4809 err = bpf_map__reuse_fd(map, pin_fd);
4810 if (err) {
4811 close(pin_fd);
4812 return err;
4813 }
4814 map->pinned = true;
4815 pr_debug("reused pinned map at '%s'\n", map->pin_path);
4816
4817 return 0;
4818 }
4819
4820 static int
bpf_object__populate_internal_map(struct bpf_object * obj,struct bpf_map * map)4821 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
4822 {
4823 enum libbpf_map_type map_type = map->libbpf_type;
4824 char *cp, errmsg[STRERR_BUFSIZE];
4825 int err, zero = 0;
4826
4827 if (obj->gen_loader) {
4828 bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps,
4829 map->mmaped, map->def.value_size);
4830 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG)
4831 bpf_gen__map_freeze(obj->gen_loader, map - obj->maps);
4832 return 0;
4833 }
4834 err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
4835 if (err) {
4836 err = -errno;
4837 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4838 pr_warn("Error setting initial map(%s) contents: %s\n",
4839 map->name, cp);
4840 return err;
4841 }
4842
4843 /* Freeze .rodata and .kconfig map as read-only from syscall side. */
4844 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
4845 err = bpf_map_freeze(map->fd);
4846 if (err) {
4847 err = -errno;
4848 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4849 pr_warn("Error freezing map(%s) as read-only: %s\n",
4850 map->name, cp);
4851 return err;
4852 }
4853 }
4854 return 0;
4855 }
4856
4857 static void bpf_map__destroy(struct bpf_map *map);
4858
bpf_object__create_map(struct bpf_object * obj,struct bpf_map * map,bool is_inner)4859 static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner)
4860 {
4861 LIBBPF_OPTS(bpf_map_create_opts, create_attr);
4862 struct bpf_map_def *def = &map->def;
4863 const char *map_name = NULL;
4864 __u32 max_entries;
4865 int err = 0;
4866
4867 if (kernel_supports(obj, FEAT_PROG_NAME))
4868 map_name = map->name;
4869 create_attr.map_ifindex = map->map_ifindex;
4870 create_attr.map_flags = def->map_flags;
4871 create_attr.numa_node = map->numa_node;
4872 create_attr.map_extra = map->map_extra;
4873
4874 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !def->max_entries) {
4875 int nr_cpus;
4876
4877 nr_cpus = libbpf_num_possible_cpus();
4878 if (nr_cpus < 0) {
4879 pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
4880 map->name, nr_cpus);
4881 return nr_cpus;
4882 }
4883 pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
4884 max_entries = nr_cpus;
4885 } else {
4886 max_entries = def->max_entries;
4887 }
4888
4889 if (bpf_map__is_struct_ops(map))
4890 create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4891
4892 if (obj->btf && btf__fd(obj->btf) >= 0 && !bpf_map_find_btf_info(obj, map)) {
4893 create_attr.btf_fd = btf__fd(obj->btf);
4894 create_attr.btf_key_type_id = map->btf_key_type_id;
4895 create_attr.btf_value_type_id = map->btf_value_type_id;
4896 }
4897
4898 if (bpf_map_type__is_map_in_map(def->type)) {
4899 if (map->inner_map) {
4900 err = bpf_object__create_map(obj, map->inner_map, true);
4901 if (err) {
4902 pr_warn("map '%s': failed to create inner map: %d\n",
4903 map->name, err);
4904 return err;
4905 }
4906 map->inner_map_fd = bpf_map__fd(map->inner_map);
4907 }
4908 if (map->inner_map_fd >= 0)
4909 create_attr.inner_map_fd = map->inner_map_fd;
4910 }
4911
4912 switch (def->type) {
4913 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
4914 case BPF_MAP_TYPE_CGROUP_ARRAY:
4915 case BPF_MAP_TYPE_STACK_TRACE:
4916 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
4917 case BPF_MAP_TYPE_HASH_OF_MAPS:
4918 case BPF_MAP_TYPE_DEVMAP:
4919 case BPF_MAP_TYPE_DEVMAP_HASH:
4920 case BPF_MAP_TYPE_CPUMAP:
4921 case BPF_MAP_TYPE_XSKMAP:
4922 case BPF_MAP_TYPE_SOCKMAP:
4923 case BPF_MAP_TYPE_SOCKHASH:
4924 case BPF_MAP_TYPE_QUEUE:
4925 case BPF_MAP_TYPE_STACK:
4926 case BPF_MAP_TYPE_RINGBUF:
4927 create_attr.btf_fd = 0;
4928 create_attr.btf_key_type_id = 0;
4929 create_attr.btf_value_type_id = 0;
4930 map->btf_key_type_id = 0;
4931 map->btf_value_type_id = 0;
4932 default:
4933 break;
4934 }
4935
4936 if (obj->gen_loader) {
4937 bpf_gen__map_create(obj->gen_loader, def->type, map_name,
4938 def->key_size, def->value_size, max_entries,
4939 &create_attr, is_inner ? -1 : map - obj->maps);
4940 /* Pretend to have valid FD to pass various fd >= 0 checks.
4941 * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
4942 */
4943 map->fd = 0;
4944 } else {
4945 map->fd = bpf_map_create(def->type, map_name,
4946 def->key_size, def->value_size,
4947 max_entries, &create_attr);
4948 }
4949 if (map->fd < 0 && (create_attr.btf_key_type_id ||
4950 create_attr.btf_value_type_id)) {
4951 char *cp, errmsg[STRERR_BUFSIZE];
4952
4953 err = -errno;
4954 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4955 pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
4956 map->name, cp, err);
4957 create_attr.btf_fd = 0;
4958 create_attr.btf_key_type_id = 0;
4959 create_attr.btf_value_type_id = 0;
4960 map->btf_key_type_id = 0;
4961 map->btf_value_type_id = 0;
4962 map->fd = bpf_map_create(def->type, map_name,
4963 def->key_size, def->value_size,
4964 max_entries, &create_attr);
4965 }
4966
4967 err = map->fd < 0 ? -errno : 0;
4968
4969 if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
4970 if (obj->gen_loader)
4971 map->inner_map->fd = -1;
4972 bpf_map__destroy(map->inner_map);
4973 zfree(&map->inner_map);
4974 }
4975
4976 return err;
4977 }
4978
init_map_in_map_slots(struct bpf_object * obj,struct bpf_map * map)4979 static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map)
4980 {
4981 const struct bpf_map *targ_map;
4982 unsigned int i;
4983 int fd, err = 0;
4984
4985 for (i = 0; i < map->init_slots_sz; i++) {
4986 if (!map->init_slots[i])
4987 continue;
4988
4989 targ_map = map->init_slots[i];
4990 fd = bpf_map__fd(targ_map);
4991
4992 if (obj->gen_loader) {
4993 bpf_gen__populate_outer_map(obj->gen_loader,
4994 map - obj->maps, i,
4995 targ_map - obj->maps);
4996 } else {
4997 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
4998 }
4999 if (err) {
5000 err = -errno;
5001 pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n",
5002 map->name, i, targ_map->name, fd, err);
5003 return err;
5004 }
5005 pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
5006 map->name, i, targ_map->name, fd);
5007 }
5008
5009 zfree(&map->init_slots);
5010 map->init_slots_sz = 0;
5011
5012 return 0;
5013 }
5014
init_prog_array_slots(struct bpf_object * obj,struct bpf_map * map)5015 static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map)
5016 {
5017 const struct bpf_program *targ_prog;
5018 unsigned int i;
5019 int fd, err;
5020
5021 if (obj->gen_loader)
5022 return -ENOTSUP;
5023
5024 for (i = 0; i < map->init_slots_sz; i++) {
5025 if (!map->init_slots[i])
5026 continue;
5027
5028 targ_prog = map->init_slots[i];
5029 fd = bpf_program__fd(targ_prog);
5030
5031 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5032 if (err) {
5033 err = -errno;
5034 pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %d\n",
5035 map->name, i, targ_prog->name, fd, err);
5036 return err;
5037 }
5038 pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n",
5039 map->name, i, targ_prog->name, fd);
5040 }
5041
5042 zfree(&map->init_slots);
5043 map->init_slots_sz = 0;
5044
5045 return 0;
5046 }
5047
bpf_object_init_prog_arrays(struct bpf_object * obj)5048 static int bpf_object_init_prog_arrays(struct bpf_object *obj)
5049 {
5050 struct bpf_map *map;
5051 int i, err;
5052
5053 for (i = 0; i < obj->nr_maps; i++) {
5054 map = &obj->maps[i];
5055
5056 if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY)
5057 continue;
5058
5059 err = init_prog_array_slots(obj, map);
5060 if (err < 0) {
5061 zclose(map->fd);
5062 return err;
5063 }
5064 }
5065 return 0;
5066 }
5067
5068 static int
bpf_object__create_maps(struct bpf_object * obj)5069 bpf_object__create_maps(struct bpf_object *obj)
5070 {
5071 struct bpf_map *map;
5072 char *cp, errmsg[STRERR_BUFSIZE];
5073 unsigned int i, j;
5074 int err;
5075 bool retried;
5076
5077 for (i = 0; i < obj->nr_maps; i++) {
5078 map = &obj->maps[i];
5079
5080 /* To support old kernels, we skip creating global data maps
5081 * (.rodata, .data, .kconfig, etc); later on, during program
5082 * loading, if we detect that at least one of the to-be-loaded
5083 * programs is referencing any global data map, we'll error
5084 * out with program name and relocation index logged.
5085 * This approach allows to accommodate Clang emitting
5086 * unnecessary .rodata.str1.1 sections for string literals,
5087 * but also it allows to have CO-RE applications that use
5088 * global variables in some of BPF programs, but not others.
5089 * If those global variable-using programs are not loaded at
5090 * runtime due to bpf_program__set_autoload(prog, false),
5091 * bpf_object loading will succeed just fine even on old
5092 * kernels.
5093 */
5094 if (bpf_map__is_internal(map) &&
5095 !kernel_supports(obj, FEAT_GLOBAL_DATA)) {
5096 map->skipped = true;
5097 continue;
5098 }
5099
5100 retried = false;
5101 retry:
5102 if (map->pin_path) {
5103 err = bpf_object__reuse_map(map);
5104 if (err) {
5105 pr_warn("map '%s': error reusing pinned map\n",
5106 map->name);
5107 goto err_out;
5108 }
5109 if (retried && map->fd < 0) {
5110 pr_warn("map '%s': cannot find pinned map\n",
5111 map->name);
5112 err = -ENOENT;
5113 goto err_out;
5114 }
5115 }
5116
5117 if (map->fd >= 0) {
5118 pr_debug("map '%s': skipping creation (preset fd=%d)\n",
5119 map->name, map->fd);
5120 } else {
5121 err = bpf_object__create_map(obj, map, false);
5122 if (err)
5123 goto err_out;
5124
5125 pr_debug("map '%s': created successfully, fd=%d\n",
5126 map->name, map->fd);
5127
5128 if (bpf_map__is_internal(map)) {
5129 err = bpf_object__populate_internal_map(obj, map);
5130 if (err < 0) {
5131 zclose(map->fd);
5132 goto err_out;
5133 }
5134 }
5135
5136 if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) {
5137 err = init_map_in_map_slots(obj, map);
5138 if (err < 0) {
5139 zclose(map->fd);
5140 goto err_out;
5141 }
5142 }
5143 }
5144
5145 if (map->pin_path && !map->pinned) {
5146 err = bpf_map__pin(map, NULL);
5147 if (err) {
5148 zclose(map->fd);
5149 if (!retried && err == -EEXIST) {
5150 retried = true;
5151 goto retry;
5152 }
5153 pr_warn("map '%s': failed to auto-pin at '%s': %d\n",
5154 map->name, map->pin_path, err);
5155 goto err_out;
5156 }
5157 }
5158 }
5159
5160 return 0;
5161
5162 err_out:
5163 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5164 pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err);
5165 pr_perm_msg(err);
5166 for (j = 0; j < i; j++)
5167 zclose(obj->maps[j].fd);
5168 return err;
5169 }
5170
bpf_core_is_flavor_sep(const char * s)5171 static bool bpf_core_is_flavor_sep(const char *s)
5172 {
5173 /* check X___Y name pattern, where X and Y are not underscores */
5174 return s[0] != '_' && /* X */
5175 s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */
5176 s[4] != '_'; /* Y */
5177 }
5178
5179 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
5180 * before last triple underscore. Struct name part after last triple
5181 * underscore is ignored by BPF CO-RE relocation during relocation matching.
5182 */
bpf_core_essential_name_len(const char * name)5183 size_t bpf_core_essential_name_len(const char *name)
5184 {
5185 size_t n = strlen(name);
5186 int i;
5187
5188 for (i = n - 5; i >= 0; i--) {
5189 if (bpf_core_is_flavor_sep(name + i))
5190 return i + 1;
5191 }
5192 return n;
5193 }
5194
bpf_core_free_cands(struct bpf_core_cand_list * cands)5195 static void bpf_core_free_cands(struct bpf_core_cand_list *cands)
5196 {
5197 free(cands->cands);
5198 free(cands);
5199 }
5200
bpf_core_add_cands(struct bpf_core_cand * local_cand,size_t local_essent_len,const struct btf * targ_btf,const char * targ_btf_name,int targ_start_id,struct bpf_core_cand_list * cands)5201 static int bpf_core_add_cands(struct bpf_core_cand *local_cand,
5202 size_t local_essent_len,
5203 const struct btf *targ_btf,
5204 const char *targ_btf_name,
5205 int targ_start_id,
5206 struct bpf_core_cand_list *cands)
5207 {
5208 struct bpf_core_cand *new_cands, *cand;
5209 const struct btf_type *t, *local_t;
5210 const char *targ_name, *local_name;
5211 size_t targ_essent_len;
5212 int n, i;
5213
5214 local_t = btf__type_by_id(local_cand->btf, local_cand->id);
5215 local_name = btf__str_by_offset(local_cand->btf, local_t->name_off);
5216
5217 n = btf__type_cnt(targ_btf);
5218 for (i = targ_start_id; i < n; i++) {
5219 t = btf__type_by_id(targ_btf, i);
5220 if (btf_kind(t) != btf_kind(local_t))
5221 continue;
5222
5223 targ_name = btf__name_by_offset(targ_btf, t->name_off);
5224 if (str_is_empty(targ_name))
5225 continue;
5226
5227 targ_essent_len = bpf_core_essential_name_len(targ_name);
5228 if (targ_essent_len != local_essent_len)
5229 continue;
5230
5231 if (strncmp(local_name, targ_name, local_essent_len) != 0)
5232 continue;
5233
5234 pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n",
5235 local_cand->id, btf_kind_str(local_t),
5236 local_name, i, btf_kind_str(t), targ_name,
5237 targ_btf_name);
5238 new_cands = libbpf_reallocarray(cands->cands, cands->len + 1,
5239 sizeof(*cands->cands));
5240 if (!new_cands)
5241 return -ENOMEM;
5242
5243 cand = &new_cands[cands->len];
5244 cand->btf = targ_btf;
5245 cand->id = i;
5246
5247 cands->cands = new_cands;
5248 cands->len++;
5249 }
5250 return 0;
5251 }
5252
load_module_btfs(struct bpf_object * obj)5253 static int load_module_btfs(struct bpf_object *obj)
5254 {
5255 struct bpf_btf_info info;
5256 struct module_btf *mod_btf;
5257 struct btf *btf;
5258 char name[64];
5259 __u32 id = 0, len;
5260 int err, fd;
5261
5262 if (obj->btf_modules_loaded)
5263 return 0;
5264
5265 if (obj->gen_loader)
5266 return 0;
5267
5268 /* don't do this again, even if we find no module BTFs */
5269 obj->btf_modules_loaded = true;
5270
5271 /* kernel too old to support module BTFs */
5272 if (!kernel_supports(obj, FEAT_MODULE_BTF))
5273 return 0;
5274
5275 while (true) {
5276 err = bpf_btf_get_next_id(id, &id);
5277 if (err && errno == ENOENT)
5278 return 0;
5279 if (err) {
5280 err = -errno;
5281 pr_warn("failed to iterate BTF objects: %d\n", err);
5282 return err;
5283 }
5284
5285 fd = bpf_btf_get_fd_by_id(id);
5286 if (fd < 0) {
5287 if (errno == ENOENT)
5288 continue; /* expected race: BTF was unloaded */
5289 err = -errno;
5290 pr_warn("failed to get BTF object #%d FD: %d\n", id, err);
5291 return err;
5292 }
5293
5294 len = sizeof(info);
5295 memset(&info, 0, sizeof(info));
5296 info.name = ptr_to_u64(name);
5297 info.name_len = sizeof(name);
5298
5299 err = bpf_obj_get_info_by_fd(fd, &info, &len);
5300 if (err) {
5301 err = -errno;
5302 pr_warn("failed to get BTF object #%d info: %d\n", id, err);
5303 goto err_out;
5304 }
5305
5306 /* ignore non-module BTFs */
5307 if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) {
5308 close(fd);
5309 continue;
5310 }
5311
5312 btf = btf_get_from_fd(fd, obj->btf_vmlinux);
5313 err = libbpf_get_error(btf);
5314 if (err) {
5315 pr_warn("failed to load module [%s]'s BTF object #%d: %d\n",
5316 name, id, err);
5317 goto err_out;
5318 }
5319
5320 err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap,
5321 sizeof(*obj->btf_modules), obj->btf_module_cnt + 1);
5322 if (err)
5323 goto err_out;
5324
5325 mod_btf = &obj->btf_modules[obj->btf_module_cnt++];
5326
5327 mod_btf->btf = btf;
5328 mod_btf->id = id;
5329 mod_btf->fd = fd;
5330 mod_btf->name = strdup(name);
5331 if (!mod_btf->name) {
5332 err = -ENOMEM;
5333 goto err_out;
5334 }
5335 continue;
5336
5337 err_out:
5338 close(fd);
5339 return err;
5340 }
5341
5342 return 0;
5343 }
5344
5345 static struct bpf_core_cand_list *
bpf_core_find_cands(struct bpf_object * obj,const struct btf * local_btf,__u32 local_type_id)5346 bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id)
5347 {
5348 struct bpf_core_cand local_cand = {};
5349 struct bpf_core_cand_list *cands;
5350 const struct btf *main_btf;
5351 const struct btf_type *local_t;
5352 const char *local_name;
5353 size_t local_essent_len;
5354 int err, i;
5355
5356 local_cand.btf = local_btf;
5357 local_cand.id = local_type_id;
5358 local_t = btf__type_by_id(local_btf, local_type_id);
5359 if (!local_t)
5360 return ERR_PTR(-EINVAL);
5361
5362 local_name = btf__name_by_offset(local_btf, local_t->name_off);
5363 if (str_is_empty(local_name))
5364 return ERR_PTR(-EINVAL);
5365 local_essent_len = bpf_core_essential_name_len(local_name);
5366
5367 cands = calloc(1, sizeof(*cands));
5368 if (!cands)
5369 return ERR_PTR(-ENOMEM);
5370
5371 /* Attempt to find target candidates in vmlinux BTF first */
5372 main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux;
5373 err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands);
5374 if (err)
5375 goto err_out;
5376
5377 /* if vmlinux BTF has any candidate, don't got for module BTFs */
5378 if (cands->len)
5379 return cands;
5380
5381 /* if vmlinux BTF was overridden, don't attempt to load module BTFs */
5382 if (obj->btf_vmlinux_override)
5383 return cands;
5384
5385 /* now look through module BTFs, trying to still find candidates */
5386 err = load_module_btfs(obj);
5387 if (err)
5388 goto err_out;
5389
5390 for (i = 0; i < obj->btf_module_cnt; i++) {
5391 err = bpf_core_add_cands(&local_cand, local_essent_len,
5392 obj->btf_modules[i].btf,
5393 obj->btf_modules[i].name,
5394 btf__type_cnt(obj->btf_vmlinux),
5395 cands);
5396 if (err)
5397 goto err_out;
5398 }
5399
5400 return cands;
5401 err_out:
5402 bpf_core_free_cands(cands);
5403 return ERR_PTR(err);
5404 }
5405
5406 /* Check local and target types for compatibility. This check is used for
5407 * type-based CO-RE relocations and follow slightly different rules than
5408 * field-based relocations. This function assumes that root types were already
5409 * checked for name match. Beyond that initial root-level name check, names
5410 * are completely ignored. Compatibility rules are as follows:
5411 * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but
5412 * kind should match for local and target types (i.e., STRUCT is not
5413 * compatible with UNION);
5414 * - for ENUMs, the size is ignored;
5415 * - for INT, size and signedness are ignored;
5416 * - for ARRAY, dimensionality is ignored, element types are checked for
5417 * compatibility recursively;
5418 * - CONST/VOLATILE/RESTRICT modifiers are ignored;
5419 * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible;
5420 * - FUNC_PROTOs are compatible if they have compatible signature: same
5421 * number of input args and compatible return and argument types.
5422 * These rules are not set in stone and probably will be adjusted as we get
5423 * more experience with using BPF CO-RE relocations.
5424 */
bpf_core_types_are_compat(const struct btf * local_btf,__u32 local_id,const struct btf * targ_btf,__u32 targ_id)5425 int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id,
5426 const struct btf *targ_btf, __u32 targ_id)
5427 {
5428 const struct btf_type *local_type, *targ_type;
5429 int depth = 32; /* max recursion depth */
5430
5431 /* caller made sure that names match (ignoring flavor suffix) */
5432 local_type = btf__type_by_id(local_btf, local_id);
5433 targ_type = btf__type_by_id(targ_btf, targ_id);
5434 if (btf_kind(local_type) != btf_kind(targ_type))
5435 return 0;
5436
5437 recur:
5438 depth--;
5439 if (depth < 0)
5440 return -EINVAL;
5441
5442 local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
5443 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
5444 if (!local_type || !targ_type)
5445 return -EINVAL;
5446
5447 if (btf_kind(local_type) != btf_kind(targ_type))
5448 return 0;
5449
5450 switch (btf_kind(local_type)) {
5451 case BTF_KIND_UNKN:
5452 case BTF_KIND_STRUCT:
5453 case BTF_KIND_UNION:
5454 case BTF_KIND_ENUM:
5455 case BTF_KIND_FWD:
5456 return 1;
5457 case BTF_KIND_INT:
5458 /* just reject deprecated bitfield-like integers; all other
5459 * integers are by default compatible between each other
5460 */
5461 return btf_int_offset(local_type) == 0 && btf_int_offset(targ_type) == 0;
5462 case BTF_KIND_PTR:
5463 local_id = local_type->type;
5464 targ_id = targ_type->type;
5465 goto recur;
5466 case BTF_KIND_ARRAY:
5467 local_id = btf_array(local_type)->type;
5468 targ_id = btf_array(targ_type)->type;
5469 goto recur;
5470 case BTF_KIND_FUNC_PROTO: {
5471 struct btf_param *local_p = btf_params(local_type);
5472 struct btf_param *targ_p = btf_params(targ_type);
5473 __u16 local_vlen = btf_vlen(local_type);
5474 __u16 targ_vlen = btf_vlen(targ_type);
5475 int i, err;
5476
5477 if (local_vlen != targ_vlen)
5478 return 0;
5479
5480 for (i = 0; i < local_vlen; i++, local_p++, targ_p++) {
5481 skip_mods_and_typedefs(local_btf, local_p->type, &local_id);
5482 skip_mods_and_typedefs(targ_btf, targ_p->type, &targ_id);
5483 err = bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id);
5484 if (err <= 0)
5485 return err;
5486 }
5487
5488 /* tail recurse for return type check */
5489 skip_mods_and_typedefs(local_btf, local_type->type, &local_id);
5490 skip_mods_and_typedefs(targ_btf, targ_type->type, &targ_id);
5491 goto recur;
5492 }
5493 default:
5494 pr_warn("unexpected kind %s relocated, local [%d], target [%d]\n",
5495 btf_kind_str(local_type), local_id, targ_id);
5496 return 0;
5497 }
5498 }
5499
bpf_core_hash_fn(const void * key,void * ctx)5500 static size_t bpf_core_hash_fn(const void *key, void *ctx)
5501 {
5502 return (size_t)key;
5503 }
5504
bpf_core_equal_fn(const void * k1,const void * k2,void * ctx)5505 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
5506 {
5507 return k1 == k2;
5508 }
5509
u32_as_hash_key(__u32 x)5510 static void *u32_as_hash_key(__u32 x)
5511 {
5512 return (void *)(uintptr_t)x;
5513 }
5514
record_relo_core(struct bpf_program * prog,const struct bpf_core_relo * core_relo,int insn_idx)5515 static int record_relo_core(struct bpf_program *prog,
5516 const struct bpf_core_relo *core_relo, int insn_idx)
5517 {
5518 struct reloc_desc *relos, *relo;
5519
5520 relos = libbpf_reallocarray(prog->reloc_desc,
5521 prog->nr_reloc + 1, sizeof(*relos));
5522 if (!relos)
5523 return -ENOMEM;
5524 relo = &relos[prog->nr_reloc];
5525 relo->type = RELO_CORE;
5526 relo->insn_idx = insn_idx;
5527 relo->core_relo = core_relo;
5528 prog->reloc_desc = relos;
5529 prog->nr_reloc++;
5530 return 0;
5531 }
5532
bpf_core_apply_relo(struct bpf_program * prog,const struct bpf_core_relo * relo,int relo_idx,const struct btf * local_btf,struct hashmap * cand_cache)5533 static int bpf_core_apply_relo(struct bpf_program *prog,
5534 const struct bpf_core_relo *relo,
5535 int relo_idx,
5536 const struct btf *local_btf,
5537 struct hashmap *cand_cache)
5538 {
5539 struct bpf_core_spec specs_scratch[3] = {};
5540 const void *type_key = u32_as_hash_key(relo->type_id);
5541 struct bpf_core_cand_list *cands = NULL;
5542 const char *prog_name = prog->name;
5543 const struct btf_type *local_type;
5544 const char *local_name;
5545 __u32 local_id = relo->type_id;
5546 struct bpf_insn *insn;
5547 int insn_idx, err;
5548
5549 if (relo->insn_off % BPF_INSN_SZ)
5550 return -EINVAL;
5551 insn_idx = relo->insn_off / BPF_INSN_SZ;
5552 /* adjust insn_idx from section frame of reference to the local
5553 * program's frame of reference; (sub-)program code is not yet
5554 * relocated, so it's enough to just subtract in-section offset
5555 */
5556 insn_idx = insn_idx - prog->sec_insn_off;
5557 if (insn_idx >= prog->insns_cnt)
5558 return -EINVAL;
5559 insn = &prog->insns[insn_idx];
5560
5561 local_type = btf__type_by_id(local_btf, local_id);
5562 if (!local_type)
5563 return -EINVAL;
5564
5565 local_name = btf__name_by_offset(local_btf, local_type->name_off);
5566 if (!local_name)
5567 return -EINVAL;
5568
5569 if (prog->obj->gen_loader) {
5570 const char *spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
5571
5572 pr_debug("record_relo_core: prog %td insn[%d] %s %s %s final insn_idx %d\n",
5573 prog - prog->obj->programs, relo->insn_off / 8,
5574 btf_kind_str(local_type), local_name, spec_str, insn_idx);
5575 return record_relo_core(prog, relo, insn_idx);
5576 }
5577
5578 if (relo->kind != BPF_CORE_TYPE_ID_LOCAL &&
5579 !hashmap__find(cand_cache, type_key, (void **)&cands)) {
5580 cands = bpf_core_find_cands(prog->obj, local_btf, local_id);
5581 if (IS_ERR(cands)) {
5582 pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n",
5583 prog_name, relo_idx, local_id, btf_kind_str(local_type),
5584 local_name, PTR_ERR(cands));
5585 return PTR_ERR(cands);
5586 }
5587 err = hashmap__set(cand_cache, type_key, cands, NULL, NULL);
5588 if (err) {
5589 bpf_core_free_cands(cands);
5590 return err;
5591 }
5592 }
5593
5594 return bpf_core_apply_relo_insn(prog_name, insn, insn_idx, relo,
5595 relo_idx, local_btf, cands, specs_scratch);
5596 }
5597
5598 static int
bpf_object__relocate_core(struct bpf_object * obj,const char * targ_btf_path)5599 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
5600 {
5601 const struct btf_ext_info_sec *sec;
5602 const struct bpf_core_relo *rec;
5603 const struct btf_ext_info *seg;
5604 struct hashmap_entry *entry;
5605 struct hashmap *cand_cache = NULL;
5606 struct bpf_program *prog;
5607 const char *sec_name;
5608 int i, err = 0, insn_idx, sec_idx;
5609
5610 if (obj->btf_ext->core_relo_info.len == 0)
5611 return 0;
5612
5613 if (targ_btf_path) {
5614 obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL);
5615 err = libbpf_get_error(obj->btf_vmlinux_override);
5616 if (err) {
5617 pr_warn("failed to parse target BTF: %d\n", err);
5618 return err;
5619 }
5620 }
5621
5622 cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
5623 if (IS_ERR(cand_cache)) {
5624 err = PTR_ERR(cand_cache);
5625 goto out;
5626 }
5627
5628 seg = &obj->btf_ext->core_relo_info;
5629 for_each_btf_ext_sec(seg, sec) {
5630 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
5631 if (str_is_empty(sec_name)) {
5632 err = -EINVAL;
5633 goto out;
5634 }
5635 /* bpf_object's ELF is gone by now so it's not easy to find
5636 * section index by section name, but we can find *any*
5637 * bpf_program within desired section name and use it's
5638 * prog->sec_idx to do a proper search by section index and
5639 * instruction offset
5640 */
5641 prog = NULL;
5642 for (i = 0; i < obj->nr_programs; i++) {
5643 prog = &obj->programs[i];
5644 if (strcmp(prog->sec_name, sec_name) == 0)
5645 break;
5646 }
5647 if (!prog) {
5648 pr_warn("sec '%s': failed to find a BPF program\n", sec_name);
5649 return -ENOENT;
5650 }
5651 sec_idx = prog->sec_idx;
5652
5653 pr_debug("sec '%s': found %d CO-RE relocations\n",
5654 sec_name, sec->num_info);
5655
5656 for_each_btf_ext_rec(seg, sec, i, rec) {
5657 insn_idx = rec->insn_off / BPF_INSN_SZ;
5658 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
5659 if (!prog) {
5660 pr_warn("sec '%s': failed to find program at insn #%d for CO-RE offset relocation #%d\n",
5661 sec_name, insn_idx, i);
5662 err = -EINVAL;
5663 goto out;
5664 }
5665 /* no need to apply CO-RE relocation if the program is
5666 * not going to be loaded
5667 */
5668 if (!prog->load)
5669 continue;
5670
5671 err = bpf_core_apply_relo(prog, rec, i, obj->btf, cand_cache);
5672 if (err) {
5673 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
5674 prog->name, i, err);
5675 goto out;
5676 }
5677 }
5678 }
5679
5680 out:
5681 /* obj->btf_vmlinux and module BTFs are freed after object load */
5682 btf__free(obj->btf_vmlinux_override);
5683 obj->btf_vmlinux_override = NULL;
5684
5685 if (!IS_ERR_OR_NULL(cand_cache)) {
5686 hashmap__for_each_entry(cand_cache, entry, i) {
5687 bpf_core_free_cands(entry->value);
5688 }
5689 hashmap__free(cand_cache);
5690 }
5691 return err;
5692 }
5693
5694 /* Relocate data references within program code:
5695 * - map references;
5696 * - global variable references;
5697 * - extern references.
5698 */
5699 static int
bpf_object__relocate_data(struct bpf_object * obj,struct bpf_program * prog)5700 bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
5701 {
5702 int i;
5703
5704 for (i = 0; i < prog->nr_reloc; i++) {
5705 struct reloc_desc *relo = &prog->reloc_desc[i];
5706 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
5707 struct extern_desc *ext;
5708
5709 switch (relo->type) {
5710 case RELO_LD64:
5711 if (obj->gen_loader) {
5712 insn[0].src_reg = BPF_PSEUDO_MAP_IDX;
5713 insn[0].imm = relo->map_idx;
5714 } else {
5715 insn[0].src_reg = BPF_PSEUDO_MAP_FD;
5716 insn[0].imm = obj->maps[relo->map_idx].fd;
5717 }
5718 break;
5719 case RELO_DATA:
5720 insn[1].imm = insn[0].imm + relo->sym_off;
5721 if (obj->gen_loader) {
5722 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
5723 insn[0].imm = relo->map_idx;
5724 } else {
5725 const struct bpf_map *map = &obj->maps[relo->map_idx];
5726
5727 if (map->skipped) {
5728 pr_warn("prog '%s': relo #%d: kernel doesn't support global data\n",
5729 prog->name, i);
5730 return -ENOTSUP;
5731 }
5732 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5733 insn[0].imm = obj->maps[relo->map_idx].fd;
5734 }
5735 break;
5736 case RELO_EXTERN_VAR:
5737 ext = &obj->externs[relo->sym_off];
5738 if (ext->type == EXT_KCFG) {
5739 if (obj->gen_loader) {
5740 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
5741 insn[0].imm = obj->kconfig_map_idx;
5742 } else {
5743 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5744 insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
5745 }
5746 insn[1].imm = ext->kcfg.data_off;
5747 } else /* EXT_KSYM */ {
5748 if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */
5749 insn[0].src_reg = BPF_PSEUDO_BTF_ID;
5750 insn[0].imm = ext->ksym.kernel_btf_id;
5751 insn[1].imm = ext->ksym.kernel_btf_obj_fd;
5752 } else { /* typeless ksyms or unresolved typed ksyms */
5753 insn[0].imm = (__u32)ext->ksym.addr;
5754 insn[1].imm = ext->ksym.addr >> 32;
5755 }
5756 }
5757 break;
5758 case RELO_EXTERN_FUNC:
5759 ext = &obj->externs[relo->sym_off];
5760 insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL;
5761 if (ext->is_set) {
5762 insn[0].imm = ext->ksym.kernel_btf_id;
5763 insn[0].off = ext->ksym.btf_fd_idx;
5764 } else { /* unresolved weak kfunc */
5765 insn[0].imm = 0;
5766 insn[0].off = 0;
5767 }
5768 break;
5769 case RELO_SUBPROG_ADDR:
5770 if (insn[0].src_reg != BPF_PSEUDO_FUNC) {
5771 pr_warn("prog '%s': relo #%d: bad insn\n",
5772 prog->name, i);
5773 return -EINVAL;
5774 }
5775 /* handled already */
5776 break;
5777 case RELO_CALL:
5778 /* handled already */
5779 break;
5780 case RELO_CORE:
5781 /* will be handled by bpf_program_record_relos() */
5782 break;
5783 default:
5784 pr_warn("prog '%s': relo #%d: bad relo type %d\n",
5785 prog->name, i, relo->type);
5786 return -EINVAL;
5787 }
5788 }
5789
5790 return 0;
5791 }
5792
adjust_prog_btf_ext_info(const struct bpf_object * obj,const struct bpf_program * prog,const struct btf_ext_info * ext_info,void ** prog_info,__u32 * prog_rec_cnt,__u32 * prog_rec_sz)5793 static int adjust_prog_btf_ext_info(const struct bpf_object *obj,
5794 const struct bpf_program *prog,
5795 const struct btf_ext_info *ext_info,
5796 void **prog_info, __u32 *prog_rec_cnt,
5797 __u32 *prog_rec_sz)
5798 {
5799 void *copy_start = NULL, *copy_end = NULL;
5800 void *rec, *rec_end, *new_prog_info;
5801 const struct btf_ext_info_sec *sec;
5802 size_t old_sz, new_sz;
5803 const char *sec_name;
5804 int i, off_adj;
5805
5806 for_each_btf_ext_sec(ext_info, sec) {
5807 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
5808 if (!sec_name)
5809 return -EINVAL;
5810 if (strcmp(sec_name, prog->sec_name) != 0)
5811 continue;
5812
5813 for_each_btf_ext_rec(ext_info, sec, i, rec) {
5814 __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ;
5815
5816 if (insn_off < prog->sec_insn_off)
5817 continue;
5818 if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt)
5819 break;
5820
5821 if (!copy_start)
5822 copy_start = rec;
5823 copy_end = rec + ext_info->rec_size;
5824 }
5825
5826 if (!copy_start)
5827 return -ENOENT;
5828
5829 /* append func/line info of a given (sub-)program to the main
5830 * program func/line info
5831 */
5832 old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size;
5833 new_sz = old_sz + (copy_end - copy_start);
5834 new_prog_info = realloc(*prog_info, new_sz);
5835 if (!new_prog_info)
5836 return -ENOMEM;
5837 *prog_info = new_prog_info;
5838 *prog_rec_cnt = new_sz / ext_info->rec_size;
5839 memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start);
5840
5841 /* Kernel instruction offsets are in units of 8-byte
5842 * instructions, while .BTF.ext instruction offsets generated
5843 * by Clang are in units of bytes. So convert Clang offsets
5844 * into kernel offsets and adjust offset according to program
5845 * relocated position.
5846 */
5847 off_adj = prog->sub_insn_off - prog->sec_insn_off;
5848 rec = new_prog_info + old_sz;
5849 rec_end = new_prog_info + new_sz;
5850 for (; rec < rec_end; rec += ext_info->rec_size) {
5851 __u32 *insn_off = rec;
5852
5853 *insn_off = *insn_off / BPF_INSN_SZ + off_adj;
5854 }
5855 *prog_rec_sz = ext_info->rec_size;
5856 return 0;
5857 }
5858
5859 return -ENOENT;
5860 }
5861
5862 static int
reloc_prog_func_and_line_info(const struct bpf_object * obj,struct bpf_program * main_prog,const struct bpf_program * prog)5863 reloc_prog_func_and_line_info(const struct bpf_object *obj,
5864 struct bpf_program *main_prog,
5865 const struct bpf_program *prog)
5866 {
5867 int err;
5868
5869 /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't
5870 * supprot func/line info
5871 */
5872 if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC))
5873 return 0;
5874
5875 /* only attempt func info relocation if main program's func_info
5876 * relocation was successful
5877 */
5878 if (main_prog != prog && !main_prog->func_info)
5879 goto line_info;
5880
5881 err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info,
5882 &main_prog->func_info,
5883 &main_prog->func_info_cnt,
5884 &main_prog->func_info_rec_size);
5885 if (err) {
5886 if (err != -ENOENT) {
5887 pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n",
5888 prog->name, err);
5889 return err;
5890 }
5891 if (main_prog->func_info) {
5892 /*
5893 * Some info has already been found but has problem
5894 * in the last btf_ext reloc. Must have to error out.
5895 */
5896 pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name);
5897 return err;
5898 }
5899 /* Have problem loading the very first info. Ignore the rest. */
5900 pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n",
5901 prog->name);
5902 }
5903
5904 line_info:
5905 /* don't relocate line info if main program's relocation failed */
5906 if (main_prog != prog && !main_prog->line_info)
5907 return 0;
5908
5909 err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info,
5910 &main_prog->line_info,
5911 &main_prog->line_info_cnt,
5912 &main_prog->line_info_rec_size);
5913 if (err) {
5914 if (err != -ENOENT) {
5915 pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n",
5916 prog->name, err);
5917 return err;
5918 }
5919 if (main_prog->line_info) {
5920 /*
5921 * Some info has already been found but has problem
5922 * in the last btf_ext reloc. Must have to error out.
5923 */
5924 pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name);
5925 return err;
5926 }
5927 /* Have problem loading the very first info. Ignore the rest. */
5928 pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n",
5929 prog->name);
5930 }
5931 return 0;
5932 }
5933
cmp_relo_by_insn_idx(const void * key,const void * elem)5934 static int cmp_relo_by_insn_idx(const void *key, const void *elem)
5935 {
5936 size_t insn_idx = *(const size_t *)key;
5937 const struct reloc_desc *relo = elem;
5938
5939 if (insn_idx == relo->insn_idx)
5940 return 0;
5941 return insn_idx < relo->insn_idx ? -1 : 1;
5942 }
5943
find_prog_insn_relo(const struct bpf_program * prog,size_t insn_idx)5944 static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx)
5945 {
5946 if (!prog->nr_reloc)
5947 return NULL;
5948 return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc,
5949 sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx);
5950 }
5951
append_subprog_relos(struct bpf_program * main_prog,struct bpf_program * subprog)5952 static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog)
5953 {
5954 int new_cnt = main_prog->nr_reloc + subprog->nr_reloc;
5955 struct reloc_desc *relos;
5956 int i;
5957
5958 if (main_prog == subprog)
5959 return 0;
5960 relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos));
5961 if (!relos)
5962 return -ENOMEM;
5963 if (subprog->nr_reloc)
5964 memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc,
5965 sizeof(*relos) * subprog->nr_reloc);
5966
5967 for (i = main_prog->nr_reloc; i < new_cnt; i++)
5968 relos[i].insn_idx += subprog->sub_insn_off;
5969 /* After insn_idx adjustment the 'relos' array is still sorted
5970 * by insn_idx and doesn't break bsearch.
5971 */
5972 main_prog->reloc_desc = relos;
5973 main_prog->nr_reloc = new_cnt;
5974 return 0;
5975 }
5976
5977 static int
bpf_object__reloc_code(struct bpf_object * obj,struct bpf_program * main_prog,struct bpf_program * prog)5978 bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog,
5979 struct bpf_program *prog)
5980 {
5981 size_t sub_insn_idx, insn_idx, new_cnt;
5982 struct bpf_program *subprog;
5983 struct bpf_insn *insns, *insn;
5984 struct reloc_desc *relo;
5985 int err;
5986
5987 err = reloc_prog_func_and_line_info(obj, main_prog, prog);
5988 if (err)
5989 return err;
5990
5991 for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) {
5992 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
5993 if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn))
5994 continue;
5995
5996 relo = find_prog_insn_relo(prog, insn_idx);
5997 if (relo && relo->type == RELO_EXTERN_FUNC)
5998 /* kfunc relocations will be handled later
5999 * in bpf_object__relocate_data()
6000 */
6001 continue;
6002 if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) {
6003 pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n",
6004 prog->name, insn_idx, relo->type);
6005 return -LIBBPF_ERRNO__RELOC;
6006 }
6007 if (relo) {
6008 /* sub-program instruction index is a combination of
6009 * an offset of a symbol pointed to by relocation and
6010 * call instruction's imm field; for global functions,
6011 * call always has imm = -1, but for static functions
6012 * relocation is against STT_SECTION and insn->imm
6013 * points to a start of a static function
6014 *
6015 * for subprog addr relocation, the relo->sym_off + insn->imm is
6016 * the byte offset in the corresponding section.
6017 */
6018 if (relo->type == RELO_CALL)
6019 sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1;
6020 else
6021 sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ;
6022 } else if (insn_is_pseudo_func(insn)) {
6023 /*
6024 * RELO_SUBPROG_ADDR relo is always emitted even if both
6025 * functions are in the same section, so it shouldn't reach here.
6026 */
6027 pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n",
6028 prog->name, insn_idx);
6029 return -LIBBPF_ERRNO__RELOC;
6030 } else {
6031 /* if subprogram call is to a static function within
6032 * the same ELF section, there won't be any relocation
6033 * emitted, but it also means there is no additional
6034 * offset necessary, insns->imm is relative to
6035 * instruction's original position within the section
6036 */
6037 sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1;
6038 }
6039
6040 /* we enforce that sub-programs should be in .text section */
6041 subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx);
6042 if (!subprog) {
6043 pr_warn("prog '%s': no .text section found yet sub-program call exists\n",
6044 prog->name);
6045 return -LIBBPF_ERRNO__RELOC;
6046 }
6047
6048 /* if it's the first call instruction calling into this
6049 * subprogram (meaning this subprog hasn't been processed
6050 * yet) within the context of current main program:
6051 * - append it at the end of main program's instructions blog;
6052 * - process is recursively, while current program is put on hold;
6053 * - if that subprogram calls some other not yet processes
6054 * subprogram, same thing will happen recursively until
6055 * there are no more unprocesses subprograms left to append
6056 * and relocate.
6057 */
6058 if (subprog->sub_insn_off == 0) {
6059 subprog->sub_insn_off = main_prog->insns_cnt;
6060
6061 new_cnt = main_prog->insns_cnt + subprog->insns_cnt;
6062 insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns));
6063 if (!insns) {
6064 pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name);
6065 return -ENOMEM;
6066 }
6067 main_prog->insns = insns;
6068 main_prog->insns_cnt = new_cnt;
6069
6070 memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns,
6071 subprog->insns_cnt * sizeof(*insns));
6072
6073 pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n",
6074 main_prog->name, subprog->insns_cnt, subprog->name);
6075
6076 /* The subprog insns are now appended. Append its relos too. */
6077 err = append_subprog_relos(main_prog, subprog);
6078 if (err)
6079 return err;
6080 err = bpf_object__reloc_code(obj, main_prog, subprog);
6081 if (err)
6082 return err;
6083 }
6084
6085 /* main_prog->insns memory could have been re-allocated, so
6086 * calculate pointer again
6087 */
6088 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6089 /* calculate correct instruction position within current main
6090 * prog; each main prog can have a different set of
6091 * subprograms appended (potentially in different order as
6092 * well), so position of any subprog can be different for
6093 * different main programs */
6094 insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1;
6095
6096 pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n",
6097 prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off);
6098 }
6099
6100 return 0;
6101 }
6102
6103 /*
6104 * Relocate sub-program calls.
6105 *
6106 * Algorithm operates as follows. Each entry-point BPF program (referred to as
6107 * main prog) is processed separately. For each subprog (non-entry functions,
6108 * that can be called from either entry progs or other subprogs) gets their
6109 * sub_insn_off reset to zero. This serves as indicator that this subprogram
6110 * hasn't been yet appended and relocated within current main prog. Once its
6111 * relocated, sub_insn_off will point at the position within current main prog
6112 * where given subprog was appended. This will further be used to relocate all
6113 * the call instructions jumping into this subprog.
6114 *
6115 * We start with main program and process all call instructions. If the call
6116 * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off
6117 * is zero), subprog instructions are appended at the end of main program's
6118 * instruction array. Then main program is "put on hold" while we recursively
6119 * process newly appended subprogram. If that subprogram calls into another
6120 * subprogram that hasn't been appended, new subprogram is appended again to
6121 * the *main* prog's instructions (subprog's instructions are always left
6122 * untouched, as they need to be in unmodified state for subsequent main progs
6123 * and subprog instructions are always sent only as part of a main prog) and
6124 * the process continues recursively. Once all the subprogs called from a main
6125 * prog or any of its subprogs are appended (and relocated), all their
6126 * positions within finalized instructions array are known, so it's easy to
6127 * rewrite call instructions with correct relative offsets, corresponding to
6128 * desired target subprog.
6129 *
6130 * Its important to realize that some subprogs might not be called from some
6131 * main prog and any of its called/used subprogs. Those will keep their
6132 * subprog->sub_insn_off as zero at all times and won't be appended to current
6133 * main prog and won't be relocated within the context of current main prog.
6134 * They might still be used from other main progs later.
6135 *
6136 * Visually this process can be shown as below. Suppose we have two main
6137 * programs mainA and mainB and BPF object contains three subprogs: subA,
6138 * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and
6139 * subC both call subB:
6140 *
6141 * +--------+ +-------+
6142 * | v v |
6143 * +--+---+ +--+-+-+ +---+--+
6144 * | subA | | subB | | subC |
6145 * +--+---+ +------+ +---+--+
6146 * ^ ^
6147 * | |
6148 * +---+-------+ +------+----+
6149 * | mainA | | mainB |
6150 * +-----------+ +-----------+
6151 *
6152 * We'll start relocating mainA, will find subA, append it and start
6153 * processing sub A recursively:
6154 *
6155 * +-----------+------+
6156 * | mainA | subA |
6157 * +-----------+------+
6158 *
6159 * At this point we notice that subB is used from subA, so we append it and
6160 * relocate (there are no further subcalls from subB):
6161 *
6162 * +-----------+------+------+
6163 * | mainA | subA | subB |
6164 * +-----------+------+------+
6165 *
6166 * At this point, we relocate subA calls, then go one level up and finish with
6167 * relocatin mainA calls. mainA is done.
6168 *
6169 * For mainB process is similar but results in different order. We start with
6170 * mainB and skip subA and subB, as mainB never calls them (at least
6171 * directly), but we see subC is needed, so we append and start processing it:
6172 *
6173 * +-----------+------+
6174 * | mainB | subC |
6175 * +-----------+------+
6176 * Now we see subC needs subB, so we go back to it, append and relocate it:
6177 *
6178 * +-----------+------+------+
6179 * | mainB | subC | subB |
6180 * +-----------+------+------+
6181 *
6182 * At this point we unwind recursion, relocate calls in subC, then in mainB.
6183 */
6184 static int
bpf_object__relocate_calls(struct bpf_object * obj,struct bpf_program * prog)6185 bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog)
6186 {
6187 struct bpf_program *subprog;
6188 int i, err;
6189
6190 /* mark all subprogs as not relocated (yet) within the context of
6191 * current main program
6192 */
6193 for (i = 0; i < obj->nr_programs; i++) {
6194 subprog = &obj->programs[i];
6195 if (!prog_is_subprog(obj, subprog))
6196 continue;
6197
6198 subprog->sub_insn_off = 0;
6199 }
6200
6201 err = bpf_object__reloc_code(obj, prog, prog);
6202 if (err)
6203 return err;
6204
6205
6206 return 0;
6207 }
6208
6209 static void
bpf_object__free_relocs(struct bpf_object * obj)6210 bpf_object__free_relocs(struct bpf_object *obj)
6211 {
6212 struct bpf_program *prog;
6213 int i;
6214
6215 /* free up relocation descriptors */
6216 for (i = 0; i < obj->nr_programs; i++) {
6217 prog = &obj->programs[i];
6218 zfree(&prog->reloc_desc);
6219 prog->nr_reloc = 0;
6220 }
6221 }
6222
cmp_relocs(const void * _a,const void * _b)6223 static int cmp_relocs(const void *_a, const void *_b)
6224 {
6225 const struct reloc_desc *a = _a;
6226 const struct reloc_desc *b = _b;
6227
6228 if (a->insn_idx != b->insn_idx)
6229 return a->insn_idx < b->insn_idx ? -1 : 1;
6230
6231 /* no two relocations should have the same insn_idx, but ... */
6232 if (a->type != b->type)
6233 return a->type < b->type ? -1 : 1;
6234
6235 return 0;
6236 }
6237
bpf_object__sort_relos(struct bpf_object * obj)6238 static void bpf_object__sort_relos(struct bpf_object *obj)
6239 {
6240 int i;
6241
6242 for (i = 0; i < obj->nr_programs; i++) {
6243 struct bpf_program *p = &obj->programs[i];
6244
6245 if (!p->nr_reloc)
6246 continue;
6247
6248 qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs);
6249 }
6250 }
6251
6252 static int
bpf_object__relocate(struct bpf_object * obj,const char * targ_btf_path)6253 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
6254 {
6255 struct bpf_program *prog;
6256 size_t i, j;
6257 int err;
6258
6259 if (obj->btf_ext) {
6260 err = bpf_object__relocate_core(obj, targ_btf_path);
6261 if (err) {
6262 pr_warn("failed to perform CO-RE relocations: %d\n",
6263 err);
6264 return err;
6265 }
6266 if (obj->gen_loader)
6267 bpf_object__sort_relos(obj);
6268 }
6269
6270 /* Before relocating calls pre-process relocations and mark
6271 * few ld_imm64 instructions that points to subprogs.
6272 * Otherwise bpf_object__reloc_code() later would have to consider
6273 * all ld_imm64 insns as relocation candidates. That would
6274 * reduce relocation speed, since amount of find_prog_insn_relo()
6275 * would increase and most of them will fail to find a relo.
6276 */
6277 for (i = 0; i < obj->nr_programs; i++) {
6278 prog = &obj->programs[i];
6279 for (j = 0; j < prog->nr_reloc; j++) {
6280 struct reloc_desc *relo = &prog->reloc_desc[j];
6281 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
6282
6283 /* mark the insn, so it's recognized by insn_is_pseudo_func() */
6284 if (relo->type == RELO_SUBPROG_ADDR)
6285 insn[0].src_reg = BPF_PSEUDO_FUNC;
6286 }
6287 }
6288
6289 /* relocate subprogram calls and append used subprograms to main
6290 * programs; each copy of subprogram code needs to be relocated
6291 * differently for each main program, because its code location might
6292 * have changed.
6293 * Append subprog relos to main programs to allow data relos to be
6294 * processed after text is completely relocated.
6295 */
6296 for (i = 0; i < obj->nr_programs; i++) {
6297 prog = &obj->programs[i];
6298 /* sub-program's sub-calls are relocated within the context of
6299 * its main program only
6300 */
6301 if (prog_is_subprog(obj, prog))
6302 continue;
6303 if (!prog->load)
6304 continue;
6305
6306 err = bpf_object__relocate_calls(obj, prog);
6307 if (err) {
6308 pr_warn("prog '%s': failed to relocate calls: %d\n",
6309 prog->name, err);
6310 return err;
6311 }
6312 }
6313 /* Process data relos for main programs */
6314 for (i = 0; i < obj->nr_programs; i++) {
6315 prog = &obj->programs[i];
6316 if (prog_is_subprog(obj, prog))
6317 continue;
6318 if (!prog->load)
6319 continue;
6320 err = bpf_object__relocate_data(obj, prog);
6321 if (err) {
6322 pr_warn("prog '%s': failed to relocate data references: %d\n",
6323 prog->name, err);
6324 return err;
6325 }
6326 }
6327 if (!obj->gen_loader)
6328 bpf_object__free_relocs(obj);
6329 return 0;
6330 }
6331
6332 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
6333 Elf64_Shdr *shdr, Elf_Data *data);
6334
bpf_object__collect_map_relos(struct bpf_object * obj,Elf64_Shdr * shdr,Elf_Data * data)6335 static int bpf_object__collect_map_relos(struct bpf_object *obj,
6336 Elf64_Shdr *shdr, Elf_Data *data)
6337 {
6338 const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *);
6339 int i, j, nrels, new_sz;
6340 const struct btf_var_secinfo *vi = NULL;
6341 const struct btf_type *sec, *var, *def;
6342 struct bpf_map *map = NULL, *targ_map = NULL;
6343 struct bpf_program *targ_prog = NULL;
6344 bool is_prog_array, is_map_in_map;
6345 const struct btf_member *member;
6346 const char *name, *mname, *type;
6347 unsigned int moff;
6348 Elf64_Sym *sym;
6349 Elf64_Rel *rel;
6350 void *tmp;
6351
6352 if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
6353 return -EINVAL;
6354 sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
6355 if (!sec)
6356 return -EINVAL;
6357
6358 nrels = shdr->sh_size / shdr->sh_entsize;
6359 for (i = 0; i < nrels; i++) {
6360 rel = elf_rel_by_idx(data, i);
6361 if (!rel) {
6362 pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
6363 return -LIBBPF_ERRNO__FORMAT;
6364 }
6365
6366 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
6367 if (!sym) {
6368 pr_warn(".maps relo #%d: symbol %zx not found\n",
6369 i, (size_t)ELF64_R_SYM(rel->r_info));
6370 return -LIBBPF_ERRNO__FORMAT;
6371 }
6372 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
6373
6374 pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n",
6375 i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value,
6376 (size_t)rel->r_offset, sym->st_name, name);
6377
6378 for (j = 0; j < obj->nr_maps; j++) {
6379 map = &obj->maps[j];
6380 if (map->sec_idx != obj->efile.btf_maps_shndx)
6381 continue;
6382
6383 vi = btf_var_secinfos(sec) + map->btf_var_idx;
6384 if (vi->offset <= rel->r_offset &&
6385 rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size)
6386 break;
6387 }
6388 if (j == obj->nr_maps) {
6389 pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n",
6390 i, name, (size_t)rel->r_offset);
6391 return -EINVAL;
6392 }
6393
6394 is_map_in_map = bpf_map_type__is_map_in_map(map->def.type);
6395 is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY;
6396 type = is_map_in_map ? "map" : "prog";
6397 if (is_map_in_map) {
6398 if (sym->st_shndx != obj->efile.btf_maps_shndx) {
6399 pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
6400 i, name);
6401 return -LIBBPF_ERRNO__RELOC;
6402 }
6403 if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
6404 map->def.key_size != sizeof(int)) {
6405 pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
6406 i, map->name, sizeof(int));
6407 return -EINVAL;
6408 }
6409 targ_map = bpf_object__find_map_by_name(obj, name);
6410 if (!targ_map) {
6411 pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n",
6412 i, name);
6413 return -ESRCH;
6414 }
6415 } else if (is_prog_array) {
6416 targ_prog = bpf_object__find_program_by_name(obj, name);
6417 if (!targ_prog) {
6418 pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n",
6419 i, name);
6420 return -ESRCH;
6421 }
6422 if (targ_prog->sec_idx != sym->st_shndx ||
6423 targ_prog->sec_insn_off * 8 != sym->st_value ||
6424 prog_is_subprog(obj, targ_prog)) {
6425 pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n",
6426 i, name);
6427 return -LIBBPF_ERRNO__RELOC;
6428 }
6429 } else {
6430 return -EINVAL;
6431 }
6432
6433 var = btf__type_by_id(obj->btf, vi->type);
6434 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
6435 if (btf_vlen(def) == 0)
6436 return -EINVAL;
6437 member = btf_members(def) + btf_vlen(def) - 1;
6438 mname = btf__name_by_offset(obj->btf, member->name_off);
6439 if (strcmp(mname, "values"))
6440 return -EINVAL;
6441
6442 moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
6443 if (rel->r_offset - vi->offset < moff)
6444 return -EINVAL;
6445
6446 moff = rel->r_offset - vi->offset - moff;
6447 /* here we use BPF pointer size, which is always 64 bit, as we
6448 * are parsing ELF that was built for BPF target
6449 */
6450 if (moff % bpf_ptr_sz)
6451 return -EINVAL;
6452 moff /= bpf_ptr_sz;
6453 if (moff >= map->init_slots_sz) {
6454 new_sz = moff + 1;
6455 tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz);
6456 if (!tmp)
6457 return -ENOMEM;
6458 map->init_slots = tmp;
6459 memset(map->init_slots + map->init_slots_sz, 0,
6460 (new_sz - map->init_slots_sz) * host_ptr_sz);
6461 map->init_slots_sz = new_sz;
6462 }
6463 map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog;
6464
6465 pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n",
6466 i, map->name, moff, type, name);
6467 }
6468
6469 return 0;
6470 }
6471
bpf_object__collect_relos(struct bpf_object * obj)6472 static int bpf_object__collect_relos(struct bpf_object *obj)
6473 {
6474 int i, err;
6475
6476 for (i = 0; i < obj->efile.sec_cnt; i++) {
6477 struct elf_sec_desc *sec_desc = &obj->efile.secs[i];
6478 Elf64_Shdr *shdr;
6479 Elf_Data *data;
6480 int idx;
6481
6482 if (sec_desc->sec_type != SEC_RELO)
6483 continue;
6484
6485 shdr = sec_desc->shdr;
6486 data = sec_desc->data;
6487 idx = shdr->sh_info;
6488
6489 if (shdr->sh_type != SHT_REL) {
6490 pr_warn("internal error at %d\n", __LINE__);
6491 return -LIBBPF_ERRNO__INTERNAL;
6492 }
6493
6494 if (idx == obj->efile.st_ops_shndx)
6495 err = bpf_object__collect_st_ops_relos(obj, shdr, data);
6496 else if (idx == obj->efile.btf_maps_shndx)
6497 err = bpf_object__collect_map_relos(obj, shdr, data);
6498 else
6499 err = bpf_object__collect_prog_relos(obj, shdr, data);
6500 if (err)
6501 return err;
6502 }
6503
6504 bpf_object__sort_relos(obj);
6505 return 0;
6506 }
6507
insn_is_helper_call(struct bpf_insn * insn,enum bpf_func_id * func_id)6508 static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id)
6509 {
6510 if (BPF_CLASS(insn->code) == BPF_JMP &&
6511 BPF_OP(insn->code) == BPF_CALL &&
6512 BPF_SRC(insn->code) == BPF_K &&
6513 insn->src_reg == 0 &&
6514 insn->dst_reg == 0) {
6515 *func_id = insn->imm;
6516 return true;
6517 }
6518 return false;
6519 }
6520
bpf_object__sanitize_prog(struct bpf_object * obj,struct bpf_program * prog)6521 static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog)
6522 {
6523 struct bpf_insn *insn = prog->insns;
6524 enum bpf_func_id func_id;
6525 int i;
6526
6527 if (obj->gen_loader)
6528 return 0;
6529
6530 for (i = 0; i < prog->insns_cnt; i++, insn++) {
6531 if (!insn_is_helper_call(insn, &func_id))
6532 continue;
6533
6534 /* on kernels that don't yet support
6535 * bpf_probe_read_{kernel,user}[_str] helpers, fall back
6536 * to bpf_probe_read() which works well for old kernels
6537 */
6538 switch (func_id) {
6539 case BPF_FUNC_probe_read_kernel:
6540 case BPF_FUNC_probe_read_user:
6541 if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
6542 insn->imm = BPF_FUNC_probe_read;
6543 break;
6544 case BPF_FUNC_probe_read_kernel_str:
6545 case BPF_FUNC_probe_read_user_str:
6546 if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
6547 insn->imm = BPF_FUNC_probe_read_str;
6548 break;
6549 default:
6550 break;
6551 }
6552 }
6553 return 0;
6554 }
6555
6556 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
6557 int *btf_obj_fd, int *btf_type_id);
6558
6559 /* this is called as prog->sec_def->preload_fn for libbpf-supported sec_defs */
libbpf_preload_prog(struct bpf_program * prog,struct bpf_prog_load_opts * opts,long cookie)6560 static int libbpf_preload_prog(struct bpf_program *prog,
6561 struct bpf_prog_load_opts *opts, long cookie)
6562 {
6563 enum sec_def_flags def = cookie;
6564
6565 /* old kernels might not support specifying expected_attach_type */
6566 if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE))
6567 opts->expected_attach_type = 0;
6568
6569 if (def & SEC_SLEEPABLE)
6570 opts->prog_flags |= BPF_F_SLEEPABLE;
6571
6572 if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS))
6573 opts->prog_flags |= BPF_F_XDP_HAS_FRAGS;
6574
6575 if (def & SEC_DEPRECATED)
6576 pr_warn("SEC(\"%s\") is deprecated, please see https://github.com/libbpf/libbpf/wiki/Libbpf-1.0-migration-guide#bpf-program-sec-annotation-deprecations for details\n",
6577 prog->sec_name);
6578
6579 if ((prog->type == BPF_PROG_TYPE_TRACING ||
6580 prog->type == BPF_PROG_TYPE_LSM ||
6581 prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) {
6582 int btf_obj_fd = 0, btf_type_id = 0, err;
6583 const char *attach_name;
6584
6585 attach_name = strchr(prog->sec_name, '/') + 1;
6586 err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id);
6587 if (err)
6588 return err;
6589
6590 /* cache resolved BTF FD and BTF type ID in the prog */
6591 prog->attach_btf_obj_fd = btf_obj_fd;
6592 prog->attach_btf_id = btf_type_id;
6593
6594 /* but by now libbpf common logic is not utilizing
6595 * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because
6596 * this callback is called after opts were populated by
6597 * libbpf, so this callback has to update opts explicitly here
6598 */
6599 opts->attach_btf_obj_fd = btf_obj_fd;
6600 opts->attach_btf_id = btf_type_id;
6601 }
6602 return 0;
6603 }
6604
bpf_object_load_prog_instance(struct bpf_object * obj,struct bpf_program * prog,struct bpf_insn * insns,int insns_cnt,const char * license,__u32 kern_version,int * prog_fd)6605 static int bpf_object_load_prog_instance(struct bpf_object *obj, struct bpf_program *prog,
6606 struct bpf_insn *insns, int insns_cnt,
6607 const char *license, __u32 kern_version,
6608 int *prog_fd)
6609 {
6610 LIBBPF_OPTS(bpf_prog_load_opts, load_attr);
6611 const char *prog_name = NULL;
6612 char *cp, errmsg[STRERR_BUFSIZE];
6613 size_t log_buf_size = 0;
6614 char *log_buf = NULL, *tmp;
6615 int btf_fd, ret, err;
6616 bool own_log_buf = true;
6617 __u32 log_level = prog->log_level;
6618
6619 if (prog->type == BPF_PROG_TYPE_UNSPEC) {
6620 /*
6621 * The program type must be set. Most likely we couldn't find a proper
6622 * section definition at load time, and thus we didn't infer the type.
6623 */
6624 pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n",
6625 prog->name, prog->sec_name);
6626 return -EINVAL;
6627 }
6628
6629 if (!insns || !insns_cnt)
6630 return -EINVAL;
6631
6632 load_attr.expected_attach_type = prog->expected_attach_type;
6633 if (kernel_supports(obj, FEAT_PROG_NAME))
6634 prog_name = prog->name;
6635 load_attr.attach_prog_fd = prog->attach_prog_fd;
6636 load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd;
6637 load_attr.attach_btf_id = prog->attach_btf_id;
6638 load_attr.kern_version = kern_version;
6639 load_attr.prog_ifindex = prog->prog_ifindex;
6640
6641 /* specify func_info/line_info only if kernel supports them */
6642 btf_fd = bpf_object__btf_fd(obj);
6643 if (btf_fd >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) {
6644 load_attr.prog_btf_fd = btf_fd;
6645 load_attr.func_info = prog->func_info;
6646 load_attr.func_info_rec_size = prog->func_info_rec_size;
6647 load_attr.func_info_cnt = prog->func_info_cnt;
6648 load_attr.line_info = prog->line_info;
6649 load_attr.line_info_rec_size = prog->line_info_rec_size;
6650 load_attr.line_info_cnt = prog->line_info_cnt;
6651 }
6652 load_attr.log_level = log_level;
6653 load_attr.prog_flags = prog->prog_flags;
6654 load_attr.fd_array = obj->fd_array;
6655
6656 /* adjust load_attr if sec_def provides custom preload callback */
6657 if (prog->sec_def && prog->sec_def->preload_fn) {
6658 err = prog->sec_def->preload_fn(prog, &load_attr, prog->sec_def->cookie);
6659 if (err < 0) {
6660 pr_warn("prog '%s': failed to prepare load attributes: %d\n",
6661 prog->name, err);
6662 return err;
6663 }
6664 }
6665
6666 if (obj->gen_loader) {
6667 bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name,
6668 license, insns, insns_cnt, &load_attr,
6669 prog - obj->programs);
6670 *prog_fd = -1;
6671 return 0;
6672 }
6673
6674 retry_load:
6675 /* if log_level is zero, we don't request logs initiallly even if
6676 * custom log_buf is specified; if the program load fails, then we'll
6677 * bump log_level to 1 and use either custom log_buf or we'll allocate
6678 * our own and retry the load to get details on what failed
6679 */
6680 if (log_level) {
6681 if (prog->log_buf) {
6682 log_buf = prog->log_buf;
6683 log_buf_size = prog->log_size;
6684 own_log_buf = false;
6685 } else if (obj->log_buf) {
6686 log_buf = obj->log_buf;
6687 log_buf_size = obj->log_size;
6688 own_log_buf = false;
6689 } else {
6690 log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2);
6691 tmp = realloc(log_buf, log_buf_size);
6692 if (!tmp) {
6693 ret = -ENOMEM;
6694 goto out;
6695 }
6696 log_buf = tmp;
6697 log_buf[0] = '\0';
6698 own_log_buf = true;
6699 }
6700 }
6701
6702 load_attr.log_buf = log_buf;
6703 load_attr.log_size = log_buf_size;
6704 load_attr.log_level = log_level;
6705
6706 ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr);
6707 if (ret >= 0) {
6708 if (log_level && own_log_buf) {
6709 pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
6710 prog->name, log_buf);
6711 }
6712
6713 if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) {
6714 struct bpf_map *map;
6715 int i;
6716
6717 for (i = 0; i < obj->nr_maps; i++) {
6718 map = &prog->obj->maps[i];
6719 if (map->libbpf_type != LIBBPF_MAP_RODATA)
6720 continue;
6721
6722 if (bpf_prog_bind_map(ret, bpf_map__fd(map), NULL)) {
6723 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6724 pr_warn("prog '%s': failed to bind map '%s': %s\n",
6725 prog->name, map->real_name, cp);
6726 /* Don't fail hard if can't bind rodata. */
6727 }
6728 }
6729 }
6730
6731 *prog_fd = ret;
6732 ret = 0;
6733 goto out;
6734 }
6735
6736 if (log_level == 0) {
6737 log_level = 1;
6738 goto retry_load;
6739 }
6740 /* On ENOSPC, increase log buffer size and retry, unless custom
6741 * log_buf is specified.
6742 * Be careful to not overflow u32, though. Kernel's log buf size limit
6743 * isn't part of UAPI so it can always be bumped to full 4GB. So don't
6744 * multiply by 2 unless we are sure we'll fit within 32 bits.
6745 * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2).
6746 */
6747 if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2)
6748 goto retry_load;
6749
6750 ret = -errno;
6751 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6752 pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, cp);
6753 pr_perm_msg(ret);
6754
6755 if (own_log_buf && log_buf && log_buf[0] != '\0') {
6756 pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
6757 prog->name, log_buf);
6758 }
6759 if (insns_cnt >= BPF_MAXINSNS) {
6760 pr_warn("prog '%s': program too large (%d insns), at most %d insns\n",
6761 prog->name, insns_cnt, BPF_MAXINSNS);
6762 }
6763
6764 out:
6765 if (own_log_buf)
6766 free(log_buf);
6767 return ret;
6768 }
6769
bpf_program_record_relos(struct bpf_program * prog)6770 static int bpf_program_record_relos(struct bpf_program *prog)
6771 {
6772 struct bpf_object *obj = prog->obj;
6773 int i;
6774
6775 for (i = 0; i < prog->nr_reloc; i++) {
6776 struct reloc_desc *relo = &prog->reloc_desc[i];
6777 struct extern_desc *ext = &obj->externs[relo->sym_off];
6778
6779 switch (relo->type) {
6780 case RELO_EXTERN_VAR:
6781 if (ext->type != EXT_KSYM)
6782 continue;
6783 bpf_gen__record_extern(obj->gen_loader, ext->name,
6784 ext->is_weak, !ext->ksym.type_id,
6785 BTF_KIND_VAR, relo->insn_idx);
6786 break;
6787 case RELO_EXTERN_FUNC:
6788 bpf_gen__record_extern(obj->gen_loader, ext->name,
6789 ext->is_weak, false, BTF_KIND_FUNC,
6790 relo->insn_idx);
6791 break;
6792 case RELO_CORE: {
6793 struct bpf_core_relo cr = {
6794 .insn_off = relo->insn_idx * 8,
6795 .type_id = relo->core_relo->type_id,
6796 .access_str_off = relo->core_relo->access_str_off,
6797 .kind = relo->core_relo->kind,
6798 };
6799
6800 bpf_gen__record_relo_core(obj->gen_loader, &cr);
6801 break;
6802 }
6803 default:
6804 continue;
6805 }
6806 }
6807 return 0;
6808 }
6809
bpf_object_load_prog(struct bpf_object * obj,struct bpf_program * prog,const char * license,__u32 kern_ver)6810 static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog,
6811 const char *license, __u32 kern_ver)
6812 {
6813 int err = 0, fd, i;
6814
6815 if (obj->loaded) {
6816 pr_warn("prog '%s': can't load after object was loaded\n", prog->name);
6817 return libbpf_err(-EINVAL);
6818 }
6819
6820 if (prog->instances.nr < 0 || !prog->instances.fds) {
6821 if (prog->preprocessor) {
6822 pr_warn("Internal error: can't load program '%s'\n",
6823 prog->name);
6824 return libbpf_err(-LIBBPF_ERRNO__INTERNAL);
6825 }
6826
6827 prog->instances.fds = malloc(sizeof(int));
6828 if (!prog->instances.fds) {
6829 pr_warn("Not enough memory for BPF fds\n");
6830 return libbpf_err(-ENOMEM);
6831 }
6832 prog->instances.nr = 1;
6833 prog->instances.fds[0] = -1;
6834 }
6835
6836 if (!prog->preprocessor) {
6837 if (prog->instances.nr != 1) {
6838 pr_warn("prog '%s': inconsistent nr(%d) != 1\n",
6839 prog->name, prog->instances.nr);
6840 }
6841 if (obj->gen_loader)
6842 bpf_program_record_relos(prog);
6843 err = bpf_object_load_prog_instance(obj, prog,
6844 prog->insns, prog->insns_cnt,
6845 license, kern_ver, &fd);
6846 if (!err)
6847 prog->instances.fds[0] = fd;
6848 goto out;
6849 }
6850
6851 for (i = 0; i < prog->instances.nr; i++) {
6852 struct bpf_prog_prep_result result;
6853 bpf_program_prep_t preprocessor = prog->preprocessor;
6854
6855 memset(&result, 0, sizeof(result));
6856 err = preprocessor(prog, i, prog->insns,
6857 prog->insns_cnt, &result);
6858 if (err) {
6859 pr_warn("Preprocessing the %dth instance of program '%s' failed\n",
6860 i, prog->name);
6861 goto out;
6862 }
6863
6864 if (!result.new_insn_ptr || !result.new_insn_cnt) {
6865 pr_debug("Skip loading the %dth instance of program '%s'\n",
6866 i, prog->name);
6867 prog->instances.fds[i] = -1;
6868 if (result.pfd)
6869 *result.pfd = -1;
6870 continue;
6871 }
6872
6873 err = bpf_object_load_prog_instance(obj, prog,
6874 result.new_insn_ptr, result.new_insn_cnt,
6875 license, kern_ver, &fd);
6876 if (err) {
6877 pr_warn("Loading the %dth instance of program '%s' failed\n",
6878 i, prog->name);
6879 goto out;
6880 }
6881
6882 if (result.pfd)
6883 *result.pfd = fd;
6884 prog->instances.fds[i] = fd;
6885 }
6886 out:
6887 if (err)
6888 pr_warn("failed to load program '%s'\n", prog->name);
6889 return libbpf_err(err);
6890 }
6891
bpf_program__load(struct bpf_program * prog,const char * license,__u32 kern_ver)6892 int bpf_program__load(struct bpf_program *prog, const char *license, __u32 kern_ver)
6893 {
6894 return bpf_object_load_prog(prog->obj, prog, license, kern_ver);
6895 }
6896
6897 static int
bpf_object__load_progs(struct bpf_object * obj,int log_level)6898 bpf_object__load_progs(struct bpf_object *obj, int log_level)
6899 {
6900 struct bpf_program *prog;
6901 size_t i;
6902 int err;
6903
6904 for (i = 0; i < obj->nr_programs; i++) {
6905 prog = &obj->programs[i];
6906 err = bpf_object__sanitize_prog(obj, prog);
6907 if (err)
6908 return err;
6909 }
6910
6911 for (i = 0; i < obj->nr_programs; i++) {
6912 prog = &obj->programs[i];
6913 if (prog_is_subprog(obj, prog))
6914 continue;
6915 if (!prog->load) {
6916 pr_debug("prog '%s': skipped loading\n", prog->name);
6917 continue;
6918 }
6919 prog->log_level |= log_level;
6920 err = bpf_object_load_prog(obj, prog, obj->license, obj->kern_version);
6921 if (err)
6922 return err;
6923 }
6924 if (obj->gen_loader)
6925 bpf_object__free_relocs(obj);
6926 return 0;
6927 }
6928
6929 static const struct bpf_sec_def *find_sec_def(const char *sec_name);
6930
bpf_object_init_progs(struct bpf_object * obj,const struct bpf_object_open_opts * opts)6931 static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts)
6932 {
6933 struct bpf_program *prog;
6934 int err;
6935
6936 bpf_object__for_each_program(prog, obj) {
6937 prog->sec_def = find_sec_def(prog->sec_name);
6938 if (!prog->sec_def) {
6939 /* couldn't guess, but user might manually specify */
6940 pr_debug("prog '%s': unrecognized ELF section name '%s'\n",
6941 prog->name, prog->sec_name);
6942 continue;
6943 }
6944
6945 bpf_program__set_type(prog, prog->sec_def->prog_type);
6946 bpf_program__set_expected_attach_type(prog, prog->sec_def->expected_attach_type);
6947
6948 #pragma GCC diagnostic push
6949 #pragma GCC diagnostic ignored "-Wdeprecated-declarations"
6950 if (prog->sec_def->prog_type == BPF_PROG_TYPE_TRACING ||
6951 prog->sec_def->prog_type == BPF_PROG_TYPE_EXT)
6952 prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
6953 #pragma GCC diagnostic pop
6954
6955 /* sec_def can have custom callback which should be called
6956 * after bpf_program is initialized to adjust its properties
6957 */
6958 if (prog->sec_def->init_fn) {
6959 err = prog->sec_def->init_fn(prog, prog->sec_def->cookie);
6960 if (err < 0) {
6961 pr_warn("prog '%s': failed to initialize: %d\n",
6962 prog->name, err);
6963 return err;
6964 }
6965 }
6966 }
6967
6968 return 0;
6969 }
6970
bpf_object_open(const char * path,const void * obj_buf,size_t obj_buf_sz,const struct bpf_object_open_opts * opts)6971 static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz,
6972 const struct bpf_object_open_opts *opts)
6973 {
6974 const char *obj_name, *kconfig, *btf_tmp_path;
6975 struct bpf_object *obj;
6976 char tmp_name[64];
6977 int err;
6978 char *log_buf;
6979 size_t log_size;
6980 __u32 log_level;
6981
6982 if (elf_version(EV_CURRENT) == EV_NONE) {
6983 pr_warn("failed to init libelf for %s\n",
6984 path ? : "(mem buf)");
6985 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
6986 }
6987
6988 if (!OPTS_VALID(opts, bpf_object_open_opts))
6989 return ERR_PTR(-EINVAL);
6990
6991 obj_name = OPTS_GET(opts, object_name, NULL);
6992 if (obj_buf) {
6993 if (!obj_name) {
6994 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
6995 (unsigned long)obj_buf,
6996 (unsigned long)obj_buf_sz);
6997 obj_name = tmp_name;
6998 }
6999 path = obj_name;
7000 pr_debug("loading object '%s' from buffer\n", obj_name);
7001 }
7002
7003 log_buf = OPTS_GET(opts, kernel_log_buf, NULL);
7004 log_size = OPTS_GET(opts, kernel_log_size, 0);
7005 log_level = OPTS_GET(opts, kernel_log_level, 0);
7006 if (log_size > UINT_MAX)
7007 return ERR_PTR(-EINVAL);
7008 if (log_size && !log_buf)
7009 return ERR_PTR(-EINVAL);
7010
7011 obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
7012 if (IS_ERR(obj))
7013 return obj;
7014
7015 obj->log_buf = log_buf;
7016 obj->log_size = log_size;
7017 obj->log_level = log_level;
7018
7019 btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL);
7020 if (btf_tmp_path) {
7021 if (strlen(btf_tmp_path) >= PATH_MAX) {
7022 err = -ENAMETOOLONG;
7023 goto out;
7024 }
7025 obj->btf_custom_path = strdup(btf_tmp_path);
7026 if (!obj->btf_custom_path) {
7027 err = -ENOMEM;
7028 goto out;
7029 }
7030 }
7031
7032 kconfig = OPTS_GET(opts, kconfig, NULL);
7033 if (kconfig) {
7034 obj->kconfig = strdup(kconfig);
7035 if (!obj->kconfig) {
7036 err = -ENOMEM;
7037 goto out;
7038 }
7039 }
7040
7041 err = bpf_object__elf_init(obj);
7042 err = err ? : bpf_object__check_endianness(obj);
7043 err = err ? : bpf_object__elf_collect(obj);
7044 err = err ? : bpf_object__collect_externs(obj);
7045 err = err ? : bpf_object__finalize_btf(obj);
7046 err = err ? : bpf_object__init_maps(obj, opts);
7047 err = err ? : bpf_object_init_progs(obj, opts);
7048 err = err ? : bpf_object__collect_relos(obj);
7049 if (err)
7050 goto out;
7051
7052 bpf_object__elf_finish(obj);
7053
7054 return obj;
7055 out:
7056 bpf_object__close(obj);
7057 return ERR_PTR(err);
7058 }
7059
7060 static struct bpf_object *
__bpf_object__open_xattr(struct bpf_object_open_attr * attr,int flags)7061 __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
7062 {
7063 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
7064 .relaxed_maps = flags & MAPS_RELAX_COMPAT,
7065 );
7066
7067 /* param validation */
7068 if (!attr->file)
7069 return NULL;
7070
7071 pr_debug("loading %s\n", attr->file);
7072 return bpf_object_open(attr->file, NULL, 0, &opts);
7073 }
7074
bpf_object__open_xattr(struct bpf_object_open_attr * attr)7075 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
7076 {
7077 return libbpf_ptr(__bpf_object__open_xattr(attr, 0));
7078 }
7079
bpf_object__open(const char * path)7080 struct bpf_object *bpf_object__open(const char *path)
7081 {
7082 struct bpf_object_open_attr attr = {
7083 .file = path,
7084 .prog_type = BPF_PROG_TYPE_UNSPEC,
7085 };
7086
7087 return libbpf_ptr(__bpf_object__open_xattr(&attr, 0));
7088 }
7089
7090 struct bpf_object *
bpf_object__open_file(const char * path,const struct bpf_object_open_opts * opts)7091 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
7092 {
7093 if (!path)
7094 return libbpf_err_ptr(-EINVAL);
7095
7096 pr_debug("loading %s\n", path);
7097
7098 return libbpf_ptr(bpf_object_open(path, NULL, 0, opts));
7099 }
7100
7101 struct bpf_object *
bpf_object__open_mem(const void * obj_buf,size_t obj_buf_sz,const struct bpf_object_open_opts * opts)7102 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
7103 const struct bpf_object_open_opts *opts)
7104 {
7105 if (!obj_buf || obj_buf_sz == 0)
7106 return libbpf_err_ptr(-EINVAL);
7107
7108 return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, opts));
7109 }
7110
7111 struct bpf_object *
bpf_object__open_buffer(const void * obj_buf,size_t obj_buf_sz,const char * name)7112 bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
7113 const char *name)
7114 {
7115 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
7116 .object_name = name,
7117 /* wrong default, but backwards-compatible */
7118 .relaxed_maps = true,
7119 );
7120
7121 /* returning NULL is wrong, but backwards-compatible */
7122 if (!obj_buf || obj_buf_sz == 0)
7123 return errno = EINVAL, NULL;
7124
7125 return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, &opts));
7126 }
7127
bpf_object_unload(struct bpf_object * obj)7128 static int bpf_object_unload(struct bpf_object *obj)
7129 {
7130 size_t i;
7131
7132 if (!obj)
7133 return libbpf_err(-EINVAL);
7134
7135 for (i = 0; i < obj->nr_maps; i++) {
7136 zclose(obj->maps[i].fd);
7137 if (obj->maps[i].st_ops)
7138 zfree(&obj->maps[i].st_ops->kern_vdata);
7139 }
7140
7141 for (i = 0; i < obj->nr_programs; i++)
7142 bpf_program__unload(&obj->programs[i]);
7143
7144 return 0;
7145 }
7146
7147 int bpf_object__unload(struct bpf_object *obj) __attribute__((alias("bpf_object_unload")));
7148
bpf_object__sanitize_maps(struct bpf_object * obj)7149 static int bpf_object__sanitize_maps(struct bpf_object *obj)
7150 {
7151 struct bpf_map *m;
7152
7153 bpf_object__for_each_map(m, obj) {
7154 if (!bpf_map__is_internal(m))
7155 continue;
7156 if (!kernel_supports(obj, FEAT_ARRAY_MMAP))
7157 m->def.map_flags ^= BPF_F_MMAPABLE;
7158 }
7159
7160 return 0;
7161 }
7162
bpf_object__read_kallsyms_file(struct bpf_object * obj)7163 static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
7164 {
7165 char sym_type, sym_name[500];
7166 unsigned long long sym_addr;
7167 const struct btf_type *t;
7168 struct extern_desc *ext;
7169 int ret, err = 0;
7170 FILE *f;
7171
7172 f = fopen("/proc/kallsyms", "r");
7173 if (!f) {
7174 err = -errno;
7175 pr_warn("failed to open /proc/kallsyms: %d\n", err);
7176 return err;
7177 }
7178
7179 while (true) {
7180 ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
7181 &sym_addr, &sym_type, sym_name);
7182 if (ret == EOF && feof(f))
7183 break;
7184 if (ret != 3) {
7185 pr_warn("failed to read kallsyms entry: %d\n", ret);
7186 err = -EINVAL;
7187 goto out;
7188 }
7189
7190 ext = find_extern_by_name(obj, sym_name);
7191 if (!ext || ext->type != EXT_KSYM)
7192 continue;
7193
7194 t = btf__type_by_id(obj->btf, ext->btf_id);
7195 if (!btf_is_var(t))
7196 continue;
7197
7198 if (ext->is_set && ext->ksym.addr != sym_addr) {
7199 pr_warn("extern (ksym) '%s' resolution is ambiguous: 0x%llx or 0x%llx\n",
7200 sym_name, ext->ksym.addr, sym_addr);
7201 err = -EINVAL;
7202 goto out;
7203 }
7204 if (!ext->is_set) {
7205 ext->is_set = true;
7206 ext->ksym.addr = sym_addr;
7207 pr_debug("extern (ksym) %s=0x%llx\n", sym_name, sym_addr);
7208 }
7209 }
7210
7211 out:
7212 fclose(f);
7213 return err;
7214 }
7215
find_ksym_btf_id(struct bpf_object * obj,const char * ksym_name,__u16 kind,struct btf ** res_btf,struct module_btf ** res_mod_btf)7216 static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name,
7217 __u16 kind, struct btf **res_btf,
7218 struct module_btf **res_mod_btf)
7219 {
7220 struct module_btf *mod_btf;
7221 struct btf *btf;
7222 int i, id, err;
7223
7224 btf = obj->btf_vmlinux;
7225 mod_btf = NULL;
7226 id = btf__find_by_name_kind(btf, ksym_name, kind);
7227
7228 if (id == -ENOENT) {
7229 err = load_module_btfs(obj);
7230 if (err)
7231 return err;
7232
7233 for (i = 0; i < obj->btf_module_cnt; i++) {
7234 /* we assume module_btf's BTF FD is always >0 */
7235 mod_btf = &obj->btf_modules[i];
7236 btf = mod_btf->btf;
7237 id = btf__find_by_name_kind_own(btf, ksym_name, kind);
7238 if (id != -ENOENT)
7239 break;
7240 }
7241 }
7242 if (id <= 0)
7243 return -ESRCH;
7244
7245 *res_btf = btf;
7246 *res_mod_btf = mod_btf;
7247 return id;
7248 }
7249
bpf_object__resolve_ksym_var_btf_id(struct bpf_object * obj,struct extern_desc * ext)7250 static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj,
7251 struct extern_desc *ext)
7252 {
7253 const struct btf_type *targ_var, *targ_type;
7254 __u32 targ_type_id, local_type_id;
7255 struct module_btf *mod_btf = NULL;
7256 const char *targ_var_name;
7257 struct btf *btf = NULL;
7258 int id, err;
7259
7260 id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf);
7261 if (id < 0) {
7262 if (id == -ESRCH && ext->is_weak)
7263 return 0;
7264 pr_warn("extern (var ksym) '%s': not found in kernel BTF\n",
7265 ext->name);
7266 return id;
7267 }
7268
7269 /* find local type_id */
7270 local_type_id = ext->ksym.type_id;
7271
7272 /* find target type_id */
7273 targ_var = btf__type_by_id(btf, id);
7274 targ_var_name = btf__name_by_offset(btf, targ_var->name_off);
7275 targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id);
7276
7277 err = bpf_core_types_are_compat(obj->btf, local_type_id,
7278 btf, targ_type_id);
7279 if (err <= 0) {
7280 const struct btf_type *local_type;
7281 const char *targ_name, *local_name;
7282
7283 local_type = btf__type_by_id(obj->btf, local_type_id);
7284 local_name = btf__name_by_offset(obj->btf, local_type->name_off);
7285 targ_name = btf__name_by_offset(btf, targ_type->name_off);
7286
7287 pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n",
7288 ext->name, local_type_id,
7289 btf_kind_str(local_type), local_name, targ_type_id,
7290 btf_kind_str(targ_type), targ_name);
7291 return -EINVAL;
7292 }
7293
7294 ext->is_set = true;
7295 ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
7296 ext->ksym.kernel_btf_id = id;
7297 pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n",
7298 ext->name, id, btf_kind_str(targ_var), targ_var_name);
7299
7300 return 0;
7301 }
7302
bpf_object__resolve_ksym_func_btf_id(struct bpf_object * obj,struct extern_desc * ext)7303 static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj,
7304 struct extern_desc *ext)
7305 {
7306 int local_func_proto_id, kfunc_proto_id, kfunc_id;
7307 struct module_btf *mod_btf = NULL;
7308 const struct btf_type *kern_func;
7309 struct btf *kern_btf = NULL;
7310 int ret;
7311
7312 local_func_proto_id = ext->ksym.type_id;
7313
7314 kfunc_id = find_ksym_btf_id(obj, ext->name, BTF_KIND_FUNC, &kern_btf, &mod_btf);
7315 if (kfunc_id < 0) {
7316 if (kfunc_id == -ESRCH && ext->is_weak)
7317 return 0;
7318 pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n",
7319 ext->name);
7320 return kfunc_id;
7321 }
7322
7323 kern_func = btf__type_by_id(kern_btf, kfunc_id);
7324 kfunc_proto_id = kern_func->type;
7325
7326 ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id,
7327 kern_btf, kfunc_proto_id);
7328 if (ret <= 0) {
7329 pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with kernel [%d]\n",
7330 ext->name, local_func_proto_id, kfunc_proto_id);
7331 return -EINVAL;
7332 }
7333
7334 /* set index for module BTF fd in fd_array, if unset */
7335 if (mod_btf && !mod_btf->fd_array_idx) {
7336 /* insn->off is s16 */
7337 if (obj->fd_array_cnt == INT16_MAX) {
7338 pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n",
7339 ext->name, mod_btf->fd_array_idx);
7340 return -E2BIG;
7341 }
7342 /* Cannot use index 0 for module BTF fd */
7343 if (!obj->fd_array_cnt)
7344 obj->fd_array_cnt = 1;
7345
7346 ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int),
7347 obj->fd_array_cnt + 1);
7348 if (ret)
7349 return ret;
7350 mod_btf->fd_array_idx = obj->fd_array_cnt;
7351 /* we assume module BTF FD is always >0 */
7352 obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd;
7353 }
7354
7355 ext->is_set = true;
7356 ext->ksym.kernel_btf_id = kfunc_id;
7357 ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0;
7358 pr_debug("extern (func ksym) '%s': resolved to kernel [%d]\n",
7359 ext->name, kfunc_id);
7360
7361 return 0;
7362 }
7363
bpf_object__resolve_ksyms_btf_id(struct bpf_object * obj)7364 static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj)
7365 {
7366 const struct btf_type *t;
7367 struct extern_desc *ext;
7368 int i, err;
7369
7370 for (i = 0; i < obj->nr_extern; i++) {
7371 ext = &obj->externs[i];
7372 if (ext->type != EXT_KSYM || !ext->ksym.type_id)
7373 continue;
7374
7375 if (obj->gen_loader) {
7376 ext->is_set = true;
7377 ext->ksym.kernel_btf_obj_fd = 0;
7378 ext->ksym.kernel_btf_id = 0;
7379 continue;
7380 }
7381 t = btf__type_by_id(obj->btf, ext->btf_id);
7382 if (btf_is_var(t))
7383 err = bpf_object__resolve_ksym_var_btf_id(obj, ext);
7384 else
7385 err = bpf_object__resolve_ksym_func_btf_id(obj, ext);
7386 if (err)
7387 return err;
7388 }
7389 return 0;
7390 }
7391
bpf_object__resolve_externs(struct bpf_object * obj,const char * extra_kconfig)7392 static int bpf_object__resolve_externs(struct bpf_object *obj,
7393 const char *extra_kconfig)
7394 {
7395 bool need_config = false, need_kallsyms = false;
7396 bool need_vmlinux_btf = false;
7397 struct extern_desc *ext;
7398 void *kcfg_data = NULL;
7399 int err, i;
7400
7401 if (obj->nr_extern == 0)
7402 return 0;
7403
7404 if (obj->kconfig_map_idx >= 0)
7405 kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
7406
7407 for (i = 0; i < obj->nr_extern; i++) {
7408 ext = &obj->externs[i];
7409
7410 if (ext->type == EXT_KCFG &&
7411 strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
7412 void *ext_val = kcfg_data + ext->kcfg.data_off;
7413 __u32 kver = get_kernel_version();
7414
7415 if (!kver) {
7416 pr_warn("failed to get kernel version\n");
7417 return -EINVAL;
7418 }
7419 err = set_kcfg_value_num(ext, ext_val, kver);
7420 if (err)
7421 return err;
7422 pr_debug("extern (kcfg) %s=0x%x\n", ext->name, kver);
7423 } else if (ext->type == EXT_KCFG && str_has_pfx(ext->name, "CONFIG_")) {
7424 need_config = true;
7425 } else if (ext->type == EXT_KSYM) {
7426 if (ext->ksym.type_id)
7427 need_vmlinux_btf = true;
7428 else
7429 need_kallsyms = true;
7430 } else {
7431 pr_warn("unrecognized extern '%s'\n", ext->name);
7432 return -EINVAL;
7433 }
7434 }
7435 if (need_config && extra_kconfig) {
7436 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
7437 if (err)
7438 return -EINVAL;
7439 need_config = false;
7440 for (i = 0; i < obj->nr_extern; i++) {
7441 ext = &obj->externs[i];
7442 if (ext->type == EXT_KCFG && !ext->is_set) {
7443 need_config = true;
7444 break;
7445 }
7446 }
7447 }
7448 if (need_config) {
7449 err = bpf_object__read_kconfig_file(obj, kcfg_data);
7450 if (err)
7451 return -EINVAL;
7452 }
7453 if (need_kallsyms) {
7454 err = bpf_object__read_kallsyms_file(obj);
7455 if (err)
7456 return -EINVAL;
7457 }
7458 if (need_vmlinux_btf) {
7459 err = bpf_object__resolve_ksyms_btf_id(obj);
7460 if (err)
7461 return -EINVAL;
7462 }
7463 for (i = 0; i < obj->nr_extern; i++) {
7464 ext = &obj->externs[i];
7465
7466 if (!ext->is_set && !ext->is_weak) {
7467 pr_warn("extern %s (strong) not resolved\n", ext->name);
7468 return -ESRCH;
7469 } else if (!ext->is_set) {
7470 pr_debug("extern %s (weak) not resolved, defaulting to zero\n",
7471 ext->name);
7472 }
7473 }
7474
7475 return 0;
7476 }
7477
bpf_object_load(struct bpf_object * obj,int extra_log_level,const char * target_btf_path)7478 static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path)
7479 {
7480 int err, i;
7481
7482 if (!obj)
7483 return libbpf_err(-EINVAL);
7484
7485 if (obj->loaded) {
7486 pr_warn("object '%s': load can't be attempted twice\n", obj->name);
7487 return libbpf_err(-EINVAL);
7488 }
7489
7490 if (obj->gen_loader)
7491 bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps);
7492
7493 err = bpf_object__probe_loading(obj);
7494 err = err ? : bpf_object__load_vmlinux_btf(obj, false);
7495 err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
7496 err = err ? : bpf_object__sanitize_and_load_btf(obj);
7497 err = err ? : bpf_object__sanitize_maps(obj);
7498 err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
7499 err = err ? : bpf_object__create_maps(obj);
7500 err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path);
7501 err = err ? : bpf_object__load_progs(obj, extra_log_level);
7502 err = err ? : bpf_object_init_prog_arrays(obj);
7503
7504 if (obj->gen_loader) {
7505 /* reset FDs */
7506 if (obj->btf)
7507 btf__set_fd(obj->btf, -1);
7508 for (i = 0; i < obj->nr_maps; i++)
7509 obj->maps[i].fd = -1;
7510 if (!err)
7511 err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps);
7512 }
7513
7514 /* clean up fd_array */
7515 zfree(&obj->fd_array);
7516
7517 /* clean up module BTFs */
7518 for (i = 0; i < obj->btf_module_cnt; i++) {
7519 close(obj->btf_modules[i].fd);
7520 btf__free(obj->btf_modules[i].btf);
7521 free(obj->btf_modules[i].name);
7522 }
7523 free(obj->btf_modules);
7524
7525 /* clean up vmlinux BTF */
7526 btf__free(obj->btf_vmlinux);
7527 obj->btf_vmlinux = NULL;
7528
7529 obj->loaded = true; /* doesn't matter if successfully or not */
7530
7531 if (err)
7532 goto out;
7533
7534 return 0;
7535 out:
7536 /* unpin any maps that were auto-pinned during load */
7537 for (i = 0; i < obj->nr_maps; i++)
7538 if (obj->maps[i].pinned && !obj->maps[i].reused)
7539 bpf_map__unpin(&obj->maps[i], NULL);
7540
7541 bpf_object_unload(obj);
7542 pr_warn("failed to load object '%s'\n", obj->path);
7543 return libbpf_err(err);
7544 }
7545
bpf_object__load_xattr(struct bpf_object_load_attr * attr)7546 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
7547 {
7548 return bpf_object_load(attr->obj, attr->log_level, attr->target_btf_path);
7549 }
7550
bpf_object__load(struct bpf_object * obj)7551 int bpf_object__load(struct bpf_object *obj)
7552 {
7553 return bpf_object_load(obj, 0, NULL);
7554 }
7555
make_parent_dir(const char * path)7556 static int make_parent_dir(const char *path)
7557 {
7558 char *cp, errmsg[STRERR_BUFSIZE];
7559 char *dname, *dir;
7560 int err = 0;
7561
7562 dname = strdup(path);
7563 if (dname == NULL)
7564 return -ENOMEM;
7565
7566 dir = dirname(dname);
7567 if (mkdir(dir, 0700) && errno != EEXIST)
7568 err = -errno;
7569
7570 free(dname);
7571 if (err) {
7572 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
7573 pr_warn("failed to mkdir %s: %s\n", path, cp);
7574 }
7575 return err;
7576 }
7577
check_path(const char * path)7578 static int check_path(const char *path)
7579 {
7580 char *cp, errmsg[STRERR_BUFSIZE];
7581 struct statfs st_fs;
7582 char *dname, *dir;
7583 int err = 0;
7584
7585 if (path == NULL)
7586 return -EINVAL;
7587
7588 dname = strdup(path);
7589 if (dname == NULL)
7590 return -ENOMEM;
7591
7592 dir = dirname(dname);
7593 if (statfs(dir, &st_fs)) {
7594 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
7595 pr_warn("failed to statfs %s: %s\n", dir, cp);
7596 err = -errno;
7597 }
7598 free(dname);
7599
7600 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
7601 pr_warn("specified path %s is not on BPF FS\n", path);
7602 err = -EINVAL;
7603 }
7604
7605 return err;
7606 }
7607
bpf_program_pin_instance(struct bpf_program * prog,const char * path,int instance)7608 static int bpf_program_pin_instance(struct bpf_program *prog, const char *path, int instance)
7609 {
7610 char *cp, errmsg[STRERR_BUFSIZE];
7611 int err;
7612
7613 err = make_parent_dir(path);
7614 if (err)
7615 return libbpf_err(err);
7616
7617 err = check_path(path);
7618 if (err)
7619 return libbpf_err(err);
7620
7621 if (prog == NULL) {
7622 pr_warn("invalid program pointer\n");
7623 return libbpf_err(-EINVAL);
7624 }
7625
7626 if (instance < 0 || instance >= prog->instances.nr) {
7627 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
7628 instance, prog->name, prog->instances.nr);
7629 return libbpf_err(-EINVAL);
7630 }
7631
7632 if (bpf_obj_pin(prog->instances.fds[instance], path)) {
7633 err = -errno;
7634 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
7635 pr_warn("failed to pin program: %s\n", cp);
7636 return libbpf_err(err);
7637 }
7638 pr_debug("pinned program '%s'\n", path);
7639
7640 return 0;
7641 }
7642
bpf_program_unpin_instance(struct bpf_program * prog,const char * path,int instance)7643 static int bpf_program_unpin_instance(struct bpf_program *prog, const char *path, int instance)
7644 {
7645 int err;
7646
7647 err = check_path(path);
7648 if (err)
7649 return libbpf_err(err);
7650
7651 if (prog == NULL) {
7652 pr_warn("invalid program pointer\n");
7653 return libbpf_err(-EINVAL);
7654 }
7655
7656 if (instance < 0 || instance >= prog->instances.nr) {
7657 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
7658 instance, prog->name, prog->instances.nr);
7659 return libbpf_err(-EINVAL);
7660 }
7661
7662 err = unlink(path);
7663 if (err != 0)
7664 return libbpf_err(-errno);
7665
7666 pr_debug("unpinned program '%s'\n", path);
7667
7668 return 0;
7669 }
7670
7671 __attribute__((alias("bpf_program_pin_instance")))
7672 int bpf_object__pin_instance(struct bpf_program *prog, const char *path, int instance);
7673
7674 __attribute__((alias("bpf_program_unpin_instance")))
7675 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path, int instance);
7676
bpf_program__pin(struct bpf_program * prog,const char * path)7677 int bpf_program__pin(struct bpf_program *prog, const char *path)
7678 {
7679 int i, err;
7680
7681 err = make_parent_dir(path);
7682 if (err)
7683 return libbpf_err(err);
7684
7685 err = check_path(path);
7686 if (err)
7687 return libbpf_err(err);
7688
7689 if (prog == NULL) {
7690 pr_warn("invalid program pointer\n");
7691 return libbpf_err(-EINVAL);
7692 }
7693
7694 if (prog->instances.nr <= 0) {
7695 pr_warn("no instances of prog %s to pin\n", prog->name);
7696 return libbpf_err(-EINVAL);
7697 }
7698
7699 if (prog->instances.nr == 1) {
7700 /* don't create subdirs when pinning single instance */
7701 return bpf_program_pin_instance(prog, path, 0);
7702 }
7703
7704 for (i = 0; i < prog->instances.nr; i++) {
7705 char buf[PATH_MAX];
7706 int len;
7707
7708 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7709 if (len < 0) {
7710 err = -EINVAL;
7711 goto err_unpin;
7712 } else if (len >= PATH_MAX) {
7713 err = -ENAMETOOLONG;
7714 goto err_unpin;
7715 }
7716
7717 err = bpf_program_pin_instance(prog, buf, i);
7718 if (err)
7719 goto err_unpin;
7720 }
7721
7722 return 0;
7723
7724 err_unpin:
7725 for (i = i - 1; i >= 0; i--) {
7726 char buf[PATH_MAX];
7727 int len;
7728
7729 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7730 if (len < 0)
7731 continue;
7732 else if (len >= PATH_MAX)
7733 continue;
7734
7735 bpf_program_unpin_instance(prog, buf, i);
7736 }
7737
7738 rmdir(path);
7739
7740 return libbpf_err(err);
7741 }
7742
bpf_program__unpin(struct bpf_program * prog,const char * path)7743 int bpf_program__unpin(struct bpf_program *prog, const char *path)
7744 {
7745 int i, err;
7746
7747 err = check_path(path);
7748 if (err)
7749 return libbpf_err(err);
7750
7751 if (prog == NULL) {
7752 pr_warn("invalid program pointer\n");
7753 return libbpf_err(-EINVAL);
7754 }
7755
7756 if (prog->instances.nr <= 0) {
7757 pr_warn("no instances of prog %s to pin\n", prog->name);
7758 return libbpf_err(-EINVAL);
7759 }
7760
7761 if (prog->instances.nr == 1) {
7762 /* don't create subdirs when pinning single instance */
7763 return bpf_program_unpin_instance(prog, path, 0);
7764 }
7765
7766 for (i = 0; i < prog->instances.nr; i++) {
7767 char buf[PATH_MAX];
7768 int len;
7769
7770 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7771 if (len < 0)
7772 return libbpf_err(-EINVAL);
7773 else if (len >= PATH_MAX)
7774 return libbpf_err(-ENAMETOOLONG);
7775
7776 err = bpf_program_unpin_instance(prog, buf, i);
7777 if (err)
7778 return err;
7779 }
7780
7781 err = rmdir(path);
7782 if (err)
7783 return libbpf_err(-errno);
7784
7785 return 0;
7786 }
7787
bpf_map__pin(struct bpf_map * map,const char * path)7788 int bpf_map__pin(struct bpf_map *map, const char *path)
7789 {
7790 char *cp, errmsg[STRERR_BUFSIZE];
7791 int err;
7792
7793 if (map == NULL) {
7794 pr_warn("invalid map pointer\n");
7795 return libbpf_err(-EINVAL);
7796 }
7797
7798 if (map->pin_path) {
7799 if (path && strcmp(path, map->pin_path)) {
7800 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
7801 bpf_map__name(map), map->pin_path, path);
7802 return libbpf_err(-EINVAL);
7803 } else if (map->pinned) {
7804 pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
7805 bpf_map__name(map), map->pin_path);
7806 return 0;
7807 }
7808 } else {
7809 if (!path) {
7810 pr_warn("missing a path to pin map '%s' at\n",
7811 bpf_map__name(map));
7812 return libbpf_err(-EINVAL);
7813 } else if (map->pinned) {
7814 pr_warn("map '%s' already pinned\n", bpf_map__name(map));
7815 return libbpf_err(-EEXIST);
7816 }
7817
7818 map->pin_path = strdup(path);
7819 if (!map->pin_path) {
7820 err = -errno;
7821 goto out_err;
7822 }
7823 }
7824
7825 err = make_parent_dir(map->pin_path);
7826 if (err)
7827 return libbpf_err(err);
7828
7829 err = check_path(map->pin_path);
7830 if (err)
7831 return libbpf_err(err);
7832
7833 if (bpf_obj_pin(map->fd, map->pin_path)) {
7834 err = -errno;
7835 goto out_err;
7836 }
7837
7838 map->pinned = true;
7839 pr_debug("pinned map '%s'\n", map->pin_path);
7840
7841 return 0;
7842
7843 out_err:
7844 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
7845 pr_warn("failed to pin map: %s\n", cp);
7846 return libbpf_err(err);
7847 }
7848
bpf_map__unpin(struct bpf_map * map,const char * path)7849 int bpf_map__unpin(struct bpf_map *map, const char *path)
7850 {
7851 int err;
7852
7853 if (map == NULL) {
7854 pr_warn("invalid map pointer\n");
7855 return libbpf_err(-EINVAL);
7856 }
7857
7858 if (map->pin_path) {
7859 if (path && strcmp(path, map->pin_path)) {
7860 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
7861 bpf_map__name(map), map->pin_path, path);
7862 return libbpf_err(-EINVAL);
7863 }
7864 path = map->pin_path;
7865 } else if (!path) {
7866 pr_warn("no path to unpin map '%s' from\n",
7867 bpf_map__name(map));
7868 return libbpf_err(-EINVAL);
7869 }
7870
7871 err = check_path(path);
7872 if (err)
7873 return libbpf_err(err);
7874
7875 err = unlink(path);
7876 if (err != 0)
7877 return libbpf_err(-errno);
7878
7879 map->pinned = false;
7880 pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
7881
7882 return 0;
7883 }
7884
bpf_map__set_pin_path(struct bpf_map * map,const char * path)7885 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
7886 {
7887 char *new = NULL;
7888
7889 if (path) {
7890 new = strdup(path);
7891 if (!new)
7892 return libbpf_err(-errno);
7893 }
7894
7895 free(map->pin_path);
7896 map->pin_path = new;
7897 return 0;
7898 }
7899
7900 __alias(bpf_map__pin_path)
7901 const char *bpf_map__get_pin_path(const struct bpf_map *map);
7902
bpf_map__pin_path(const struct bpf_map * map)7903 const char *bpf_map__pin_path(const struct bpf_map *map)
7904 {
7905 return map->pin_path;
7906 }
7907
bpf_map__is_pinned(const struct bpf_map * map)7908 bool bpf_map__is_pinned(const struct bpf_map *map)
7909 {
7910 return map->pinned;
7911 }
7912
sanitize_pin_path(char * s)7913 static void sanitize_pin_path(char *s)
7914 {
7915 /* bpffs disallows periods in path names */
7916 while (*s) {
7917 if (*s == '.')
7918 *s = '_';
7919 s++;
7920 }
7921 }
7922
bpf_object__pin_maps(struct bpf_object * obj,const char * path)7923 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
7924 {
7925 struct bpf_map *map;
7926 int err;
7927
7928 if (!obj)
7929 return libbpf_err(-ENOENT);
7930
7931 if (!obj->loaded) {
7932 pr_warn("object not yet loaded; load it first\n");
7933 return libbpf_err(-ENOENT);
7934 }
7935
7936 bpf_object__for_each_map(map, obj) {
7937 char *pin_path = NULL;
7938 char buf[PATH_MAX];
7939
7940 if (map->skipped)
7941 continue;
7942
7943 if (path) {
7944 int len;
7945
7946 len = snprintf(buf, PATH_MAX, "%s/%s", path,
7947 bpf_map__name(map));
7948 if (len < 0) {
7949 err = -EINVAL;
7950 goto err_unpin_maps;
7951 } else if (len >= PATH_MAX) {
7952 err = -ENAMETOOLONG;
7953 goto err_unpin_maps;
7954 }
7955 sanitize_pin_path(buf);
7956 pin_path = buf;
7957 } else if (!map->pin_path) {
7958 continue;
7959 }
7960
7961 err = bpf_map__pin(map, pin_path);
7962 if (err)
7963 goto err_unpin_maps;
7964 }
7965
7966 return 0;
7967
7968 err_unpin_maps:
7969 while ((map = bpf_object__prev_map(obj, map))) {
7970 if (!map->pin_path)
7971 continue;
7972
7973 bpf_map__unpin(map, NULL);
7974 }
7975
7976 return libbpf_err(err);
7977 }
7978
bpf_object__unpin_maps(struct bpf_object * obj,const char * path)7979 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
7980 {
7981 struct bpf_map *map;
7982 int err;
7983
7984 if (!obj)
7985 return libbpf_err(-ENOENT);
7986
7987 bpf_object__for_each_map(map, obj) {
7988 char *pin_path = NULL;
7989 char buf[PATH_MAX];
7990
7991 if (path) {
7992 int len;
7993
7994 len = snprintf(buf, PATH_MAX, "%s/%s", path,
7995 bpf_map__name(map));
7996 if (len < 0)
7997 return libbpf_err(-EINVAL);
7998 else if (len >= PATH_MAX)
7999 return libbpf_err(-ENAMETOOLONG);
8000 sanitize_pin_path(buf);
8001 pin_path = buf;
8002 } else if (!map->pin_path) {
8003 continue;
8004 }
8005
8006 err = bpf_map__unpin(map, pin_path);
8007 if (err)
8008 return libbpf_err(err);
8009 }
8010
8011 return 0;
8012 }
8013
bpf_object__pin_programs(struct bpf_object * obj,const char * path)8014 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
8015 {
8016 struct bpf_program *prog;
8017 int err;
8018
8019 if (!obj)
8020 return libbpf_err(-ENOENT);
8021
8022 if (!obj->loaded) {
8023 pr_warn("object not yet loaded; load it first\n");
8024 return libbpf_err(-ENOENT);
8025 }
8026
8027 bpf_object__for_each_program(prog, obj) {
8028 char buf[PATH_MAX];
8029 int len;
8030
8031 len = snprintf(buf, PATH_MAX, "%s/%s", path,
8032 prog->pin_name);
8033 if (len < 0) {
8034 err = -EINVAL;
8035 goto err_unpin_programs;
8036 } else if (len >= PATH_MAX) {
8037 err = -ENAMETOOLONG;
8038 goto err_unpin_programs;
8039 }
8040
8041 err = bpf_program__pin(prog, buf);
8042 if (err)
8043 goto err_unpin_programs;
8044 }
8045
8046 return 0;
8047
8048 err_unpin_programs:
8049 while ((prog = bpf_object__prev_program(obj, prog))) {
8050 char buf[PATH_MAX];
8051 int len;
8052
8053 len = snprintf(buf, PATH_MAX, "%s/%s", path,
8054 prog->pin_name);
8055 if (len < 0)
8056 continue;
8057 else if (len >= PATH_MAX)
8058 continue;
8059
8060 bpf_program__unpin(prog, buf);
8061 }
8062
8063 return libbpf_err(err);
8064 }
8065
bpf_object__unpin_programs(struct bpf_object * obj,const char * path)8066 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
8067 {
8068 struct bpf_program *prog;
8069 int err;
8070
8071 if (!obj)
8072 return libbpf_err(-ENOENT);
8073
8074 bpf_object__for_each_program(prog, obj) {
8075 char buf[PATH_MAX];
8076 int len;
8077
8078 len = snprintf(buf, PATH_MAX, "%s/%s", path,
8079 prog->pin_name);
8080 if (len < 0)
8081 return libbpf_err(-EINVAL);
8082 else if (len >= PATH_MAX)
8083 return libbpf_err(-ENAMETOOLONG);
8084
8085 err = bpf_program__unpin(prog, buf);
8086 if (err)
8087 return libbpf_err(err);
8088 }
8089
8090 return 0;
8091 }
8092
bpf_object__pin(struct bpf_object * obj,const char * path)8093 int bpf_object__pin(struct bpf_object *obj, const char *path)
8094 {
8095 int err;
8096
8097 err = bpf_object__pin_maps(obj, path);
8098 if (err)
8099 return libbpf_err(err);
8100
8101 err = bpf_object__pin_programs(obj, path);
8102 if (err) {
8103 bpf_object__unpin_maps(obj, path);
8104 return libbpf_err(err);
8105 }
8106
8107 return 0;
8108 }
8109
bpf_map__destroy(struct bpf_map * map)8110 static void bpf_map__destroy(struct bpf_map *map)
8111 {
8112 if (map->clear_priv)
8113 map->clear_priv(map, map->priv);
8114 map->priv = NULL;
8115 map->clear_priv = NULL;
8116
8117 if (map->inner_map) {
8118 bpf_map__destroy(map->inner_map);
8119 zfree(&map->inner_map);
8120 }
8121
8122 zfree(&map->init_slots);
8123 map->init_slots_sz = 0;
8124
8125 if (map->mmaped) {
8126 munmap(map->mmaped, bpf_map_mmap_sz(map));
8127 map->mmaped = NULL;
8128 }
8129
8130 if (map->st_ops) {
8131 zfree(&map->st_ops->data);
8132 zfree(&map->st_ops->progs);
8133 zfree(&map->st_ops->kern_func_off);
8134 zfree(&map->st_ops);
8135 }
8136
8137 zfree(&map->name);
8138 zfree(&map->real_name);
8139 zfree(&map->pin_path);
8140
8141 if (map->fd >= 0)
8142 zclose(map->fd);
8143 }
8144
bpf_object__close(struct bpf_object * obj)8145 void bpf_object__close(struct bpf_object *obj)
8146 {
8147 size_t i;
8148
8149 if (IS_ERR_OR_NULL(obj))
8150 return;
8151
8152 if (obj->clear_priv)
8153 obj->clear_priv(obj, obj->priv);
8154
8155 bpf_gen__free(obj->gen_loader);
8156 bpf_object__elf_finish(obj);
8157 bpf_object_unload(obj);
8158 btf__free(obj->btf);
8159 btf_ext__free(obj->btf_ext);
8160
8161 for (i = 0; i < obj->nr_maps; i++)
8162 bpf_map__destroy(&obj->maps[i]);
8163
8164 zfree(&obj->btf_custom_path);
8165 zfree(&obj->kconfig);
8166 zfree(&obj->externs);
8167 obj->nr_extern = 0;
8168
8169 zfree(&obj->maps);
8170 obj->nr_maps = 0;
8171
8172 if (obj->programs && obj->nr_programs) {
8173 for (i = 0; i < obj->nr_programs; i++)
8174 bpf_program__exit(&obj->programs[i]);
8175 }
8176 zfree(&obj->programs);
8177
8178 list_del(&obj->list);
8179 free(obj);
8180 }
8181
8182 struct bpf_object *
bpf_object__next(struct bpf_object * prev)8183 bpf_object__next(struct bpf_object *prev)
8184 {
8185 struct bpf_object *next;
8186 bool strict = (libbpf_mode & LIBBPF_STRICT_NO_OBJECT_LIST);
8187
8188 if (strict)
8189 return NULL;
8190
8191 if (!prev)
8192 next = list_first_entry(&bpf_objects_list,
8193 struct bpf_object,
8194 list);
8195 else
8196 next = list_next_entry(prev, list);
8197
8198 /* Empty list is noticed here so don't need checking on entry. */
8199 if (&next->list == &bpf_objects_list)
8200 return NULL;
8201
8202 return next;
8203 }
8204
bpf_object__name(const struct bpf_object * obj)8205 const char *bpf_object__name(const struct bpf_object *obj)
8206 {
8207 return obj ? obj->name : libbpf_err_ptr(-EINVAL);
8208 }
8209
bpf_object__kversion(const struct bpf_object * obj)8210 unsigned int bpf_object__kversion(const struct bpf_object *obj)
8211 {
8212 return obj ? obj->kern_version : 0;
8213 }
8214
bpf_object__btf(const struct bpf_object * obj)8215 struct btf *bpf_object__btf(const struct bpf_object *obj)
8216 {
8217 return obj ? obj->btf : NULL;
8218 }
8219
bpf_object__btf_fd(const struct bpf_object * obj)8220 int bpf_object__btf_fd(const struct bpf_object *obj)
8221 {
8222 return obj->btf ? btf__fd(obj->btf) : -1;
8223 }
8224
bpf_object__set_kversion(struct bpf_object * obj,__u32 kern_version)8225 int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version)
8226 {
8227 if (obj->loaded)
8228 return libbpf_err(-EINVAL);
8229
8230 obj->kern_version = kern_version;
8231
8232 return 0;
8233 }
8234
bpf_object__set_priv(struct bpf_object * obj,void * priv,bpf_object_clear_priv_t clear_priv)8235 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
8236 bpf_object_clear_priv_t clear_priv)
8237 {
8238 if (obj->priv && obj->clear_priv)
8239 obj->clear_priv(obj, obj->priv);
8240
8241 obj->priv = priv;
8242 obj->clear_priv = clear_priv;
8243 return 0;
8244 }
8245
bpf_object__priv(const struct bpf_object * obj)8246 void *bpf_object__priv(const struct bpf_object *obj)
8247 {
8248 return obj ? obj->priv : libbpf_err_ptr(-EINVAL);
8249 }
8250
bpf_object__gen_loader(struct bpf_object * obj,struct gen_loader_opts * opts)8251 int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts)
8252 {
8253 struct bpf_gen *gen;
8254
8255 if (!opts)
8256 return -EFAULT;
8257 if (!OPTS_VALID(opts, gen_loader_opts))
8258 return -EINVAL;
8259 gen = calloc(sizeof(*gen), 1);
8260 if (!gen)
8261 return -ENOMEM;
8262 gen->opts = opts;
8263 obj->gen_loader = gen;
8264 return 0;
8265 }
8266
8267 static struct bpf_program *
__bpf_program__iter(const struct bpf_program * p,const struct bpf_object * obj,bool forward)8268 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
8269 bool forward)
8270 {
8271 size_t nr_programs = obj->nr_programs;
8272 ssize_t idx;
8273
8274 if (!nr_programs)
8275 return NULL;
8276
8277 if (!p)
8278 /* Iter from the beginning */
8279 return forward ? &obj->programs[0] :
8280 &obj->programs[nr_programs - 1];
8281
8282 if (p->obj != obj) {
8283 pr_warn("error: program handler doesn't match object\n");
8284 return errno = EINVAL, NULL;
8285 }
8286
8287 idx = (p - obj->programs) + (forward ? 1 : -1);
8288 if (idx >= obj->nr_programs || idx < 0)
8289 return NULL;
8290 return &obj->programs[idx];
8291 }
8292
8293 struct bpf_program *
bpf_program__next(struct bpf_program * prev,const struct bpf_object * obj)8294 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
8295 {
8296 return bpf_object__next_program(obj, prev);
8297 }
8298
8299 struct bpf_program *
bpf_object__next_program(const struct bpf_object * obj,struct bpf_program * prev)8300 bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev)
8301 {
8302 struct bpf_program *prog = prev;
8303
8304 do {
8305 prog = __bpf_program__iter(prog, obj, true);
8306 } while (prog && prog_is_subprog(obj, prog));
8307
8308 return prog;
8309 }
8310
8311 struct bpf_program *
bpf_program__prev(struct bpf_program * next,const struct bpf_object * obj)8312 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
8313 {
8314 return bpf_object__prev_program(obj, next);
8315 }
8316
8317 struct bpf_program *
bpf_object__prev_program(const struct bpf_object * obj,struct bpf_program * next)8318 bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next)
8319 {
8320 struct bpf_program *prog = next;
8321
8322 do {
8323 prog = __bpf_program__iter(prog, obj, false);
8324 } while (prog && prog_is_subprog(obj, prog));
8325
8326 return prog;
8327 }
8328
bpf_program__set_priv(struct bpf_program * prog,void * priv,bpf_program_clear_priv_t clear_priv)8329 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
8330 bpf_program_clear_priv_t clear_priv)
8331 {
8332 if (prog->priv && prog->clear_priv)
8333 prog->clear_priv(prog, prog->priv);
8334
8335 prog->priv = priv;
8336 prog->clear_priv = clear_priv;
8337 return 0;
8338 }
8339
bpf_program__priv(const struct bpf_program * prog)8340 void *bpf_program__priv(const struct bpf_program *prog)
8341 {
8342 return prog ? prog->priv : libbpf_err_ptr(-EINVAL);
8343 }
8344
bpf_program__set_ifindex(struct bpf_program * prog,__u32 ifindex)8345 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
8346 {
8347 prog->prog_ifindex = ifindex;
8348 }
8349
bpf_program__name(const struct bpf_program * prog)8350 const char *bpf_program__name(const struct bpf_program *prog)
8351 {
8352 return prog->name;
8353 }
8354
bpf_program__section_name(const struct bpf_program * prog)8355 const char *bpf_program__section_name(const struct bpf_program *prog)
8356 {
8357 return prog->sec_name;
8358 }
8359
bpf_program__title(const struct bpf_program * prog,bool needs_copy)8360 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
8361 {
8362 const char *title;
8363
8364 title = prog->sec_name;
8365 if (needs_copy) {
8366 title = strdup(title);
8367 if (!title) {
8368 pr_warn("failed to strdup program title\n");
8369 return libbpf_err_ptr(-ENOMEM);
8370 }
8371 }
8372
8373 return title;
8374 }
8375
bpf_program__autoload(const struct bpf_program * prog)8376 bool bpf_program__autoload(const struct bpf_program *prog)
8377 {
8378 return prog->load;
8379 }
8380
bpf_program__set_autoload(struct bpf_program * prog,bool autoload)8381 int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
8382 {
8383 if (prog->obj->loaded)
8384 return libbpf_err(-EINVAL);
8385
8386 prog->load = autoload;
8387 return 0;
8388 }
8389
8390 static int bpf_program_nth_fd(const struct bpf_program *prog, int n);
8391
bpf_program__fd(const struct bpf_program * prog)8392 int bpf_program__fd(const struct bpf_program *prog)
8393 {
8394 return bpf_program_nth_fd(prog, 0);
8395 }
8396
bpf_program__size(const struct bpf_program * prog)8397 size_t bpf_program__size(const struct bpf_program *prog)
8398 {
8399 return prog->insns_cnt * BPF_INSN_SZ;
8400 }
8401
bpf_program__insns(const struct bpf_program * prog)8402 const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog)
8403 {
8404 return prog->insns;
8405 }
8406
bpf_program__insn_cnt(const struct bpf_program * prog)8407 size_t bpf_program__insn_cnt(const struct bpf_program *prog)
8408 {
8409 return prog->insns_cnt;
8410 }
8411
bpf_program__set_prep(struct bpf_program * prog,int nr_instances,bpf_program_prep_t prep)8412 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
8413 bpf_program_prep_t prep)
8414 {
8415 int *instances_fds;
8416
8417 if (nr_instances <= 0 || !prep)
8418 return libbpf_err(-EINVAL);
8419
8420 if (prog->instances.nr > 0 || prog->instances.fds) {
8421 pr_warn("Can't set pre-processor after loading\n");
8422 return libbpf_err(-EINVAL);
8423 }
8424
8425 instances_fds = malloc(sizeof(int) * nr_instances);
8426 if (!instances_fds) {
8427 pr_warn("alloc memory failed for fds\n");
8428 return libbpf_err(-ENOMEM);
8429 }
8430
8431 /* fill all fd with -1 */
8432 memset(instances_fds, -1, sizeof(int) * nr_instances);
8433
8434 prog->instances.nr = nr_instances;
8435 prog->instances.fds = instances_fds;
8436 prog->preprocessor = prep;
8437 return 0;
8438 }
8439
8440 __attribute__((alias("bpf_program_nth_fd")))
8441 int bpf_program__nth_fd(const struct bpf_program *prog, int n);
8442
bpf_program_nth_fd(const struct bpf_program * prog,int n)8443 static int bpf_program_nth_fd(const struct bpf_program *prog, int n)
8444 {
8445 int fd;
8446
8447 if (!prog)
8448 return libbpf_err(-EINVAL);
8449
8450 if (n >= prog->instances.nr || n < 0) {
8451 pr_warn("Can't get the %dth fd from program %s: only %d instances\n",
8452 n, prog->name, prog->instances.nr);
8453 return libbpf_err(-EINVAL);
8454 }
8455
8456 fd = prog->instances.fds[n];
8457 if (fd < 0) {
8458 pr_warn("%dth instance of program '%s' is invalid\n",
8459 n, prog->name);
8460 return libbpf_err(-ENOENT);
8461 }
8462
8463 return fd;
8464 }
8465
8466 __alias(bpf_program__type)
8467 enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog);
8468
bpf_program__type(const struct bpf_program * prog)8469 enum bpf_prog_type bpf_program__type(const struct bpf_program *prog)
8470 {
8471 return prog->type;
8472 }
8473
bpf_program__set_type(struct bpf_program * prog,enum bpf_prog_type type)8474 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
8475 {
8476 prog->type = type;
8477 }
8478
bpf_program__is_type(const struct bpf_program * prog,enum bpf_prog_type type)8479 static bool bpf_program__is_type(const struct bpf_program *prog,
8480 enum bpf_prog_type type)
8481 {
8482 return prog ? (prog->type == type) : false;
8483 }
8484
8485 #define BPF_PROG_TYPE_FNS(NAME, TYPE) \
8486 int bpf_program__set_##NAME(struct bpf_program *prog) \
8487 { \
8488 if (!prog) \
8489 return libbpf_err(-EINVAL); \
8490 bpf_program__set_type(prog, TYPE); \
8491 return 0; \
8492 } \
8493 \
8494 bool bpf_program__is_##NAME(const struct bpf_program *prog) \
8495 { \
8496 return bpf_program__is_type(prog, TYPE); \
8497 } \
8498
8499 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
8500 BPF_PROG_TYPE_FNS(lsm, BPF_PROG_TYPE_LSM);
8501 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
8502 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
8503 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
8504 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
8505 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
8506 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
8507 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
8508 BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
8509 BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS);
8510 BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT);
8511 BPF_PROG_TYPE_FNS(sk_lookup, BPF_PROG_TYPE_SK_LOOKUP);
8512
8513 __alias(bpf_program__expected_attach_type)
8514 enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog);
8515
bpf_program__expected_attach_type(const struct bpf_program * prog)8516 enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog)
8517 {
8518 return prog->expected_attach_type;
8519 }
8520
bpf_program__set_expected_attach_type(struct bpf_program * prog,enum bpf_attach_type type)8521 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
8522 enum bpf_attach_type type)
8523 {
8524 prog->expected_attach_type = type;
8525 }
8526
bpf_program__flags(const struct bpf_program * prog)8527 __u32 bpf_program__flags(const struct bpf_program *prog)
8528 {
8529 return prog->prog_flags;
8530 }
8531
bpf_program__set_flags(struct bpf_program * prog,__u32 flags)8532 int bpf_program__set_flags(struct bpf_program *prog, __u32 flags)
8533 {
8534 if (prog->obj->loaded)
8535 return libbpf_err(-EBUSY);
8536
8537 prog->prog_flags = flags;
8538 return 0;
8539 }
8540
bpf_program__log_level(const struct bpf_program * prog)8541 __u32 bpf_program__log_level(const struct bpf_program *prog)
8542 {
8543 return prog->log_level;
8544 }
8545
bpf_program__set_log_level(struct bpf_program * prog,__u32 log_level)8546 int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level)
8547 {
8548 if (prog->obj->loaded)
8549 return libbpf_err(-EBUSY);
8550
8551 prog->log_level = log_level;
8552 return 0;
8553 }
8554
bpf_program__log_buf(const struct bpf_program * prog,size_t * log_size)8555 const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size)
8556 {
8557 *log_size = prog->log_size;
8558 return prog->log_buf;
8559 }
8560
bpf_program__set_log_buf(struct bpf_program * prog,char * log_buf,size_t log_size)8561 int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size)
8562 {
8563 if (log_size && !log_buf)
8564 return -EINVAL;
8565 if (prog->log_size > UINT_MAX)
8566 return -EINVAL;
8567 if (prog->obj->loaded)
8568 return -EBUSY;
8569
8570 prog->log_buf = log_buf;
8571 prog->log_size = log_size;
8572 return 0;
8573 }
8574
8575 #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \
8576 .sec = sec_pfx, \
8577 .prog_type = BPF_PROG_TYPE_##ptype, \
8578 .expected_attach_type = atype, \
8579 .cookie = (long)(flags), \
8580 .preload_fn = libbpf_preload_prog, \
8581 __VA_ARGS__ \
8582 }
8583
8584 static struct bpf_link *attach_kprobe(const struct bpf_program *prog, long cookie);
8585 static struct bpf_link *attach_tp(const struct bpf_program *prog, long cookie);
8586 static struct bpf_link *attach_raw_tp(const struct bpf_program *prog, long cookie);
8587 static struct bpf_link *attach_trace(const struct bpf_program *prog, long cookie);
8588 static struct bpf_link *attach_lsm(const struct bpf_program *prog, long cookie);
8589 static struct bpf_link *attach_iter(const struct bpf_program *prog, long cookie);
8590
8591 static const struct bpf_sec_def section_defs[] = {
8592 SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE | SEC_SLOPPY_PFX),
8593 SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8594 SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8595 SEC_DEF("kprobe/", KPROBE, 0, SEC_NONE, attach_kprobe),
8596 SEC_DEF("uprobe/", KPROBE, 0, SEC_NONE),
8597 SEC_DEF("kretprobe/", KPROBE, 0, SEC_NONE, attach_kprobe),
8598 SEC_DEF("uretprobe/", KPROBE, 0, SEC_NONE),
8599 SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE),
8600 SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE | SEC_SLOPPY_PFX | SEC_DEPRECATED),
8601 SEC_DEF("action", SCHED_ACT, 0, SEC_NONE | SEC_SLOPPY_PFX),
8602 SEC_DEF("tracepoint/", TRACEPOINT, 0, SEC_NONE, attach_tp),
8603 SEC_DEF("tp/", TRACEPOINT, 0, SEC_NONE, attach_tp),
8604 SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
8605 SEC_DEF("raw_tp/", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
8606 SEC_DEF("raw_tracepoint.w/", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
8607 SEC_DEF("raw_tp.w/", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
8608 SEC_DEF("tp_btf/", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace),
8609 SEC_DEF("fentry/", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace),
8610 SEC_DEF("fmod_ret/", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace),
8611 SEC_DEF("fexit/", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace),
8612 SEC_DEF("fentry.s/", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8613 SEC_DEF("fmod_ret.s/", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8614 SEC_DEF("fexit.s/", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8615 SEC_DEF("freplace/", EXT, 0, SEC_ATTACH_BTF, attach_trace),
8616 SEC_DEF("lsm/", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm),
8617 SEC_DEF("lsm.s/", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm),
8618 SEC_DEF("iter/", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter),
8619 SEC_DEF("iter.s/", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter),
8620 SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE),
8621 SEC_DEF("xdp.frags/devmap", XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS),
8622 SEC_DEF("xdp/devmap", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE),
8623 SEC_DEF("xdp_devmap/", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE | SEC_DEPRECATED),
8624 SEC_DEF("xdp.frags/cpumap", XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS),
8625 SEC_DEF("xdp/cpumap", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE),
8626 SEC_DEF("xdp_cpumap/", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE | SEC_DEPRECATED),
8627 SEC_DEF("xdp.frags", XDP, BPF_XDP, SEC_XDP_FRAGS),
8628 SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8629 SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE | SEC_SLOPPY_PFX),
8630 SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE | SEC_SLOPPY_PFX),
8631 SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE | SEC_SLOPPY_PFX),
8632 SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE | SEC_SLOPPY_PFX),
8633 SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE | SEC_SLOPPY_PFX),
8634 SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8635 SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8636 SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE | SEC_SLOPPY_PFX),
8637 SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8638 SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8639 SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8640 SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8641 SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8642 SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8643 SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8644 SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8645 SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8646 SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE | SEC_SLOPPY_PFX),
8647 SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8648 SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8649 SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8650 SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8651 SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8652 SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8653 SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8654 SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8655 SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8656 SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8657 SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8658 SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8659 SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8660 SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8661 SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8662 SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8663 SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8664 SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8665 SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE),
8666 SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8667 };
8668
8669 #define MAX_TYPE_NAME_SIZE 32
8670
find_sec_def(const char * sec_name)8671 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
8672 {
8673 const struct bpf_sec_def *sec_def;
8674 enum sec_def_flags sec_flags;
8675 int i, n = ARRAY_SIZE(section_defs), len;
8676 bool strict = libbpf_mode & LIBBPF_STRICT_SEC_NAME;
8677
8678 for (i = 0; i < n; i++) {
8679 sec_def = §ion_defs[i];
8680 sec_flags = sec_def->cookie;
8681 len = strlen(sec_def->sec);
8682
8683 /* "type/" always has to have proper SEC("type/extras") form */
8684 if (sec_def->sec[len - 1] == '/') {
8685 if (str_has_pfx(sec_name, sec_def->sec))
8686 return sec_def;
8687 continue;
8688 }
8689
8690 /* "type+" means it can be either exact SEC("type") or
8691 * well-formed SEC("type/extras") with proper '/' separator
8692 */
8693 if (sec_def->sec[len - 1] == '+') {
8694 len--;
8695 /* not even a prefix */
8696 if (strncmp(sec_name, sec_def->sec, len) != 0)
8697 continue;
8698 /* exact match or has '/' separator */
8699 if (sec_name[len] == '\0' || sec_name[len] == '/')
8700 return sec_def;
8701 continue;
8702 }
8703
8704 /* SEC_SLOPPY_PFX definitions are allowed to be just prefix
8705 * matches, unless strict section name mode
8706 * (LIBBPF_STRICT_SEC_NAME) is enabled, in which case the
8707 * match has to be exact.
8708 */
8709 if ((sec_flags & SEC_SLOPPY_PFX) && !strict) {
8710 if (str_has_pfx(sec_name, sec_def->sec))
8711 return sec_def;
8712 continue;
8713 }
8714
8715 /* Definitions not marked SEC_SLOPPY_PFX (e.g.,
8716 * SEC("syscall")) are exact matches in both modes.
8717 */
8718 if (strcmp(sec_name, sec_def->sec) == 0)
8719 return sec_def;
8720 }
8721 return NULL;
8722 }
8723
libbpf_get_type_names(bool attach_type)8724 static char *libbpf_get_type_names(bool attach_type)
8725 {
8726 int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
8727 char *buf;
8728
8729 buf = malloc(len);
8730 if (!buf)
8731 return NULL;
8732
8733 buf[0] = '\0';
8734 /* Forge string buf with all available names */
8735 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
8736 const struct bpf_sec_def *sec_def = §ion_defs[i];
8737
8738 if (attach_type) {
8739 if (sec_def->preload_fn != libbpf_preload_prog)
8740 continue;
8741
8742 if (!(sec_def->cookie & SEC_ATTACHABLE))
8743 continue;
8744 }
8745
8746 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
8747 free(buf);
8748 return NULL;
8749 }
8750 strcat(buf, " ");
8751 strcat(buf, section_defs[i].sec);
8752 }
8753
8754 return buf;
8755 }
8756
libbpf_prog_type_by_name(const char * name,enum bpf_prog_type * prog_type,enum bpf_attach_type * expected_attach_type)8757 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
8758 enum bpf_attach_type *expected_attach_type)
8759 {
8760 const struct bpf_sec_def *sec_def;
8761 char *type_names;
8762
8763 if (!name)
8764 return libbpf_err(-EINVAL);
8765
8766 sec_def = find_sec_def(name);
8767 if (sec_def) {
8768 *prog_type = sec_def->prog_type;
8769 *expected_attach_type = sec_def->expected_attach_type;
8770 return 0;
8771 }
8772
8773 pr_debug("failed to guess program type from ELF section '%s'\n", name);
8774 type_names = libbpf_get_type_names(false);
8775 if (type_names != NULL) {
8776 pr_debug("supported section(type) names are:%s\n", type_names);
8777 free(type_names);
8778 }
8779
8780 return libbpf_err(-ESRCH);
8781 }
8782
find_struct_ops_map_by_offset(struct bpf_object * obj,size_t offset)8783 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
8784 size_t offset)
8785 {
8786 struct bpf_map *map;
8787 size_t i;
8788
8789 for (i = 0; i < obj->nr_maps; i++) {
8790 map = &obj->maps[i];
8791 if (!bpf_map__is_struct_ops(map))
8792 continue;
8793 if (map->sec_offset <= offset &&
8794 offset - map->sec_offset < map->def.value_size)
8795 return map;
8796 }
8797
8798 return NULL;
8799 }
8800
8801 /* Collect the reloc from ELF and populate the st_ops->progs[] */
bpf_object__collect_st_ops_relos(struct bpf_object * obj,Elf64_Shdr * shdr,Elf_Data * data)8802 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
8803 Elf64_Shdr *shdr, Elf_Data *data)
8804 {
8805 const struct btf_member *member;
8806 struct bpf_struct_ops *st_ops;
8807 struct bpf_program *prog;
8808 unsigned int shdr_idx;
8809 const struct btf *btf;
8810 struct bpf_map *map;
8811 unsigned int moff, insn_idx;
8812 const char *name;
8813 __u32 member_idx;
8814 Elf64_Sym *sym;
8815 Elf64_Rel *rel;
8816 int i, nrels;
8817
8818 btf = obj->btf;
8819 nrels = shdr->sh_size / shdr->sh_entsize;
8820 for (i = 0; i < nrels; i++) {
8821 rel = elf_rel_by_idx(data, i);
8822 if (!rel) {
8823 pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
8824 return -LIBBPF_ERRNO__FORMAT;
8825 }
8826
8827 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
8828 if (!sym) {
8829 pr_warn("struct_ops reloc: symbol %zx not found\n",
8830 (size_t)ELF64_R_SYM(rel->r_info));
8831 return -LIBBPF_ERRNO__FORMAT;
8832 }
8833
8834 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
8835 map = find_struct_ops_map_by_offset(obj, rel->r_offset);
8836 if (!map) {
8837 pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n",
8838 (size_t)rel->r_offset);
8839 return -EINVAL;
8840 }
8841
8842 moff = rel->r_offset - map->sec_offset;
8843 shdr_idx = sym->st_shndx;
8844 st_ops = map->st_ops;
8845 pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel->r_offset %zu map->sec_offset %zu name %d (\'%s\')\n",
8846 map->name,
8847 (long long)(rel->r_info >> 32),
8848 (long long)sym->st_value,
8849 shdr_idx, (size_t)rel->r_offset,
8850 map->sec_offset, sym->st_name, name);
8851
8852 if (shdr_idx >= SHN_LORESERVE) {
8853 pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n",
8854 map->name, (size_t)rel->r_offset, shdr_idx);
8855 return -LIBBPF_ERRNO__RELOC;
8856 }
8857 if (sym->st_value % BPF_INSN_SZ) {
8858 pr_warn("struct_ops reloc %s: invalid target program offset %llu\n",
8859 map->name, (unsigned long long)sym->st_value);
8860 return -LIBBPF_ERRNO__FORMAT;
8861 }
8862 insn_idx = sym->st_value / BPF_INSN_SZ;
8863
8864 member = find_member_by_offset(st_ops->type, moff * 8);
8865 if (!member) {
8866 pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
8867 map->name, moff);
8868 return -EINVAL;
8869 }
8870 member_idx = member - btf_members(st_ops->type);
8871 name = btf__name_by_offset(btf, member->name_off);
8872
8873 if (!resolve_func_ptr(btf, member->type, NULL)) {
8874 pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
8875 map->name, name);
8876 return -EINVAL;
8877 }
8878
8879 prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx);
8880 if (!prog) {
8881 pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
8882 map->name, shdr_idx, name);
8883 return -EINVAL;
8884 }
8885
8886 /* prevent the use of BPF prog with invalid type */
8887 if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) {
8888 pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n",
8889 map->name, prog->name);
8890 return -EINVAL;
8891 }
8892
8893 /* if we haven't yet processed this BPF program, record proper
8894 * attach_btf_id and member_idx
8895 */
8896 if (!prog->attach_btf_id) {
8897 prog->attach_btf_id = st_ops->type_id;
8898 prog->expected_attach_type = member_idx;
8899 }
8900
8901 /* struct_ops BPF prog can be re-used between multiple
8902 * .struct_ops as long as it's the same struct_ops struct
8903 * definition and the same function pointer field
8904 */
8905 if (prog->attach_btf_id != st_ops->type_id ||
8906 prog->expected_attach_type != member_idx) {
8907 pr_warn("struct_ops reloc %s: cannot use prog %s in sec %s with type %u attach_btf_id %u expected_attach_type %u for func ptr %s\n",
8908 map->name, prog->name, prog->sec_name, prog->type,
8909 prog->attach_btf_id, prog->expected_attach_type, name);
8910 return -EINVAL;
8911 }
8912
8913 st_ops->progs[member_idx] = prog;
8914 }
8915
8916 return 0;
8917 }
8918
8919 #define BTF_TRACE_PREFIX "btf_trace_"
8920 #define BTF_LSM_PREFIX "bpf_lsm_"
8921 #define BTF_ITER_PREFIX "bpf_iter_"
8922 #define BTF_MAX_NAME_SIZE 128
8923
btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type,const char ** prefix,int * kind)8924 void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type,
8925 const char **prefix, int *kind)
8926 {
8927 switch (attach_type) {
8928 case BPF_TRACE_RAW_TP:
8929 *prefix = BTF_TRACE_PREFIX;
8930 *kind = BTF_KIND_TYPEDEF;
8931 break;
8932 case BPF_LSM_MAC:
8933 *prefix = BTF_LSM_PREFIX;
8934 *kind = BTF_KIND_FUNC;
8935 break;
8936 case BPF_TRACE_ITER:
8937 *prefix = BTF_ITER_PREFIX;
8938 *kind = BTF_KIND_FUNC;
8939 break;
8940 default:
8941 *prefix = "";
8942 *kind = BTF_KIND_FUNC;
8943 }
8944 }
8945
find_btf_by_prefix_kind(const struct btf * btf,const char * prefix,const char * name,__u32 kind)8946 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
8947 const char *name, __u32 kind)
8948 {
8949 char btf_type_name[BTF_MAX_NAME_SIZE];
8950 int ret;
8951
8952 ret = snprintf(btf_type_name, sizeof(btf_type_name),
8953 "%s%s", prefix, name);
8954 /* snprintf returns the number of characters written excluding the
8955 * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
8956 * indicates truncation.
8957 */
8958 if (ret < 0 || ret >= sizeof(btf_type_name))
8959 return -ENAMETOOLONG;
8960 return btf__find_by_name_kind(btf, btf_type_name, kind);
8961 }
8962
find_attach_btf_id(struct btf * btf,const char * name,enum bpf_attach_type attach_type)8963 static inline int find_attach_btf_id(struct btf *btf, const char *name,
8964 enum bpf_attach_type attach_type)
8965 {
8966 const char *prefix;
8967 int kind;
8968
8969 btf_get_kernel_prefix_kind(attach_type, &prefix, &kind);
8970 return find_btf_by_prefix_kind(btf, prefix, name, kind);
8971 }
8972
libbpf_find_vmlinux_btf_id(const char * name,enum bpf_attach_type attach_type)8973 int libbpf_find_vmlinux_btf_id(const char *name,
8974 enum bpf_attach_type attach_type)
8975 {
8976 struct btf *btf;
8977 int err;
8978
8979 btf = btf__load_vmlinux_btf();
8980 err = libbpf_get_error(btf);
8981 if (err) {
8982 pr_warn("vmlinux BTF is not found\n");
8983 return libbpf_err(err);
8984 }
8985
8986 err = find_attach_btf_id(btf, name, attach_type);
8987 if (err <= 0)
8988 pr_warn("%s is not found in vmlinux BTF\n", name);
8989
8990 btf__free(btf);
8991 return libbpf_err(err);
8992 }
8993
libbpf_find_prog_btf_id(const char * name,__u32 attach_prog_fd)8994 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
8995 {
8996 struct bpf_prog_info info = {};
8997 __u32 info_len = sizeof(info);
8998 struct btf *btf;
8999 int err;
9000
9001 err = bpf_obj_get_info_by_fd(attach_prog_fd, &info, &info_len);
9002 if (err) {
9003 pr_warn("failed bpf_obj_get_info_by_fd for FD %d: %d\n",
9004 attach_prog_fd, err);
9005 return err;
9006 }
9007
9008 err = -EINVAL;
9009 if (!info.btf_id) {
9010 pr_warn("The target program doesn't have BTF\n");
9011 goto out;
9012 }
9013 btf = btf__load_from_kernel_by_id(info.btf_id);
9014 err = libbpf_get_error(btf);
9015 if (err) {
9016 pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err);
9017 goto out;
9018 }
9019 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
9020 btf__free(btf);
9021 if (err <= 0) {
9022 pr_warn("%s is not found in prog's BTF\n", name);
9023 goto out;
9024 }
9025 out:
9026 return err;
9027 }
9028
find_kernel_btf_id(struct bpf_object * obj,const char * attach_name,enum bpf_attach_type attach_type,int * btf_obj_fd,int * btf_type_id)9029 static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name,
9030 enum bpf_attach_type attach_type,
9031 int *btf_obj_fd, int *btf_type_id)
9032 {
9033 int ret, i;
9034
9035 ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type);
9036 if (ret > 0) {
9037 *btf_obj_fd = 0; /* vmlinux BTF */
9038 *btf_type_id = ret;
9039 return 0;
9040 }
9041 if (ret != -ENOENT)
9042 return ret;
9043
9044 ret = load_module_btfs(obj);
9045 if (ret)
9046 return ret;
9047
9048 for (i = 0; i < obj->btf_module_cnt; i++) {
9049 const struct module_btf *mod = &obj->btf_modules[i];
9050
9051 ret = find_attach_btf_id(mod->btf, attach_name, attach_type);
9052 if (ret > 0) {
9053 *btf_obj_fd = mod->fd;
9054 *btf_type_id = ret;
9055 return 0;
9056 }
9057 if (ret == -ENOENT)
9058 continue;
9059
9060 return ret;
9061 }
9062
9063 return -ESRCH;
9064 }
9065
libbpf_find_attach_btf_id(struct bpf_program * prog,const char * attach_name,int * btf_obj_fd,int * btf_type_id)9066 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
9067 int *btf_obj_fd, int *btf_type_id)
9068 {
9069 enum bpf_attach_type attach_type = prog->expected_attach_type;
9070 __u32 attach_prog_fd = prog->attach_prog_fd;
9071 int err = 0;
9072
9073 /* BPF program's BTF ID */
9074 if (attach_prog_fd) {
9075 err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd);
9076 if (err < 0) {
9077 pr_warn("failed to find BPF program (FD %d) BTF ID for '%s': %d\n",
9078 attach_prog_fd, attach_name, err);
9079 return err;
9080 }
9081 *btf_obj_fd = 0;
9082 *btf_type_id = err;
9083 return 0;
9084 }
9085
9086 /* kernel/module BTF ID */
9087 if (prog->obj->gen_loader) {
9088 bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type);
9089 *btf_obj_fd = 0;
9090 *btf_type_id = 1;
9091 } else {
9092 err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id);
9093 }
9094 if (err) {
9095 pr_warn("failed to find kernel BTF type ID of '%s': %d\n", attach_name, err);
9096 return err;
9097 }
9098 return 0;
9099 }
9100
libbpf_attach_type_by_name(const char * name,enum bpf_attach_type * attach_type)9101 int libbpf_attach_type_by_name(const char *name,
9102 enum bpf_attach_type *attach_type)
9103 {
9104 char *type_names;
9105 const struct bpf_sec_def *sec_def;
9106
9107 if (!name)
9108 return libbpf_err(-EINVAL);
9109
9110 sec_def = find_sec_def(name);
9111 if (!sec_def) {
9112 pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
9113 type_names = libbpf_get_type_names(true);
9114 if (type_names != NULL) {
9115 pr_debug("attachable section(type) names are:%s\n", type_names);
9116 free(type_names);
9117 }
9118
9119 return libbpf_err(-EINVAL);
9120 }
9121
9122 if (sec_def->preload_fn != libbpf_preload_prog)
9123 return libbpf_err(-EINVAL);
9124 if (!(sec_def->cookie & SEC_ATTACHABLE))
9125 return libbpf_err(-EINVAL);
9126
9127 *attach_type = sec_def->expected_attach_type;
9128 return 0;
9129 }
9130
bpf_map__fd(const struct bpf_map * map)9131 int bpf_map__fd(const struct bpf_map *map)
9132 {
9133 return map ? map->fd : libbpf_err(-EINVAL);
9134 }
9135
bpf_map__def(const struct bpf_map * map)9136 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
9137 {
9138 return map ? &map->def : libbpf_err_ptr(-EINVAL);
9139 }
9140
map_uses_real_name(const struct bpf_map * map)9141 static bool map_uses_real_name(const struct bpf_map *map)
9142 {
9143 /* Since libbpf started to support custom .data.* and .rodata.* maps,
9144 * their user-visible name differs from kernel-visible name. Users see
9145 * such map's corresponding ELF section name as a map name.
9146 * This check distinguishes .data/.rodata from .data.* and .rodata.*
9147 * maps to know which name has to be returned to the user.
9148 */
9149 if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0)
9150 return true;
9151 if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0)
9152 return true;
9153 return false;
9154 }
9155
bpf_map__name(const struct bpf_map * map)9156 const char *bpf_map__name(const struct bpf_map *map)
9157 {
9158 if (!map)
9159 return NULL;
9160
9161 if (map_uses_real_name(map))
9162 return map->real_name;
9163
9164 return map->name;
9165 }
9166
bpf_map__type(const struct bpf_map * map)9167 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
9168 {
9169 return map->def.type;
9170 }
9171
bpf_map__set_type(struct bpf_map * map,enum bpf_map_type type)9172 int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
9173 {
9174 if (map->fd >= 0)
9175 return libbpf_err(-EBUSY);
9176 map->def.type = type;
9177 return 0;
9178 }
9179
bpf_map__map_flags(const struct bpf_map * map)9180 __u32 bpf_map__map_flags(const struct bpf_map *map)
9181 {
9182 return map->def.map_flags;
9183 }
9184
bpf_map__set_map_flags(struct bpf_map * map,__u32 flags)9185 int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
9186 {
9187 if (map->fd >= 0)
9188 return libbpf_err(-EBUSY);
9189 map->def.map_flags = flags;
9190 return 0;
9191 }
9192
bpf_map__map_extra(const struct bpf_map * map)9193 __u64 bpf_map__map_extra(const struct bpf_map *map)
9194 {
9195 return map->map_extra;
9196 }
9197
bpf_map__set_map_extra(struct bpf_map * map,__u64 map_extra)9198 int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra)
9199 {
9200 if (map->fd >= 0)
9201 return libbpf_err(-EBUSY);
9202 map->map_extra = map_extra;
9203 return 0;
9204 }
9205
bpf_map__numa_node(const struct bpf_map * map)9206 __u32 bpf_map__numa_node(const struct bpf_map *map)
9207 {
9208 return map->numa_node;
9209 }
9210
bpf_map__set_numa_node(struct bpf_map * map,__u32 numa_node)9211 int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
9212 {
9213 if (map->fd >= 0)
9214 return libbpf_err(-EBUSY);
9215 map->numa_node = numa_node;
9216 return 0;
9217 }
9218
bpf_map__key_size(const struct bpf_map * map)9219 __u32 bpf_map__key_size(const struct bpf_map *map)
9220 {
9221 return map->def.key_size;
9222 }
9223
bpf_map__set_key_size(struct bpf_map * map,__u32 size)9224 int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
9225 {
9226 if (map->fd >= 0)
9227 return libbpf_err(-EBUSY);
9228 map->def.key_size = size;
9229 return 0;
9230 }
9231
bpf_map__value_size(const struct bpf_map * map)9232 __u32 bpf_map__value_size(const struct bpf_map *map)
9233 {
9234 return map->def.value_size;
9235 }
9236
bpf_map__set_value_size(struct bpf_map * map,__u32 size)9237 int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
9238 {
9239 if (map->fd >= 0)
9240 return libbpf_err(-EBUSY);
9241 map->def.value_size = size;
9242 return 0;
9243 }
9244
bpf_map__btf_key_type_id(const struct bpf_map * map)9245 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
9246 {
9247 return map ? map->btf_key_type_id : 0;
9248 }
9249
bpf_map__btf_value_type_id(const struct bpf_map * map)9250 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
9251 {
9252 return map ? map->btf_value_type_id : 0;
9253 }
9254
bpf_map__set_priv(struct bpf_map * map,void * priv,bpf_map_clear_priv_t clear_priv)9255 int bpf_map__set_priv(struct bpf_map *map, void *priv,
9256 bpf_map_clear_priv_t clear_priv)
9257 {
9258 if (!map)
9259 return libbpf_err(-EINVAL);
9260
9261 if (map->priv) {
9262 if (map->clear_priv)
9263 map->clear_priv(map, map->priv);
9264 }
9265
9266 map->priv = priv;
9267 map->clear_priv = clear_priv;
9268 return 0;
9269 }
9270
bpf_map__priv(const struct bpf_map * map)9271 void *bpf_map__priv(const struct bpf_map *map)
9272 {
9273 return map ? map->priv : libbpf_err_ptr(-EINVAL);
9274 }
9275
bpf_map__set_initial_value(struct bpf_map * map,const void * data,size_t size)9276 int bpf_map__set_initial_value(struct bpf_map *map,
9277 const void *data, size_t size)
9278 {
9279 if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG ||
9280 size != map->def.value_size || map->fd >= 0)
9281 return libbpf_err(-EINVAL);
9282
9283 memcpy(map->mmaped, data, size);
9284 return 0;
9285 }
9286
bpf_map__initial_value(struct bpf_map * map,size_t * psize)9287 const void *bpf_map__initial_value(struct bpf_map *map, size_t *psize)
9288 {
9289 if (!map->mmaped)
9290 return NULL;
9291 *psize = map->def.value_size;
9292 return map->mmaped;
9293 }
9294
bpf_map__is_offload_neutral(const struct bpf_map * map)9295 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
9296 {
9297 return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
9298 }
9299
bpf_map__is_internal(const struct bpf_map * map)9300 bool bpf_map__is_internal(const struct bpf_map *map)
9301 {
9302 return map->libbpf_type != LIBBPF_MAP_UNSPEC;
9303 }
9304
bpf_map__ifindex(const struct bpf_map * map)9305 __u32 bpf_map__ifindex(const struct bpf_map *map)
9306 {
9307 return map->map_ifindex;
9308 }
9309
bpf_map__set_ifindex(struct bpf_map * map,__u32 ifindex)9310 int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
9311 {
9312 if (map->fd >= 0)
9313 return libbpf_err(-EBUSY);
9314 map->map_ifindex = ifindex;
9315 return 0;
9316 }
9317
bpf_map__set_inner_map_fd(struct bpf_map * map,int fd)9318 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
9319 {
9320 if (!bpf_map_type__is_map_in_map(map->def.type)) {
9321 pr_warn("error: unsupported map type\n");
9322 return libbpf_err(-EINVAL);
9323 }
9324 if (map->inner_map_fd != -1) {
9325 pr_warn("error: inner_map_fd already specified\n");
9326 return libbpf_err(-EINVAL);
9327 }
9328 if (map->inner_map) {
9329 bpf_map__destroy(map->inner_map);
9330 zfree(&map->inner_map);
9331 }
9332 map->inner_map_fd = fd;
9333 return 0;
9334 }
9335
9336 static struct bpf_map *
__bpf_map__iter(const struct bpf_map * m,const struct bpf_object * obj,int i)9337 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
9338 {
9339 ssize_t idx;
9340 struct bpf_map *s, *e;
9341
9342 if (!obj || !obj->maps)
9343 return errno = EINVAL, NULL;
9344
9345 s = obj->maps;
9346 e = obj->maps + obj->nr_maps;
9347
9348 if ((m < s) || (m >= e)) {
9349 pr_warn("error in %s: map handler doesn't belong to object\n",
9350 __func__);
9351 return errno = EINVAL, NULL;
9352 }
9353
9354 idx = (m - obj->maps) + i;
9355 if (idx >= obj->nr_maps || idx < 0)
9356 return NULL;
9357 return &obj->maps[idx];
9358 }
9359
9360 struct bpf_map *
bpf_map__next(const struct bpf_map * prev,const struct bpf_object * obj)9361 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
9362 {
9363 return bpf_object__next_map(obj, prev);
9364 }
9365
9366 struct bpf_map *
bpf_object__next_map(const struct bpf_object * obj,const struct bpf_map * prev)9367 bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev)
9368 {
9369 if (prev == NULL)
9370 return obj->maps;
9371
9372 return __bpf_map__iter(prev, obj, 1);
9373 }
9374
9375 struct bpf_map *
bpf_map__prev(const struct bpf_map * next,const struct bpf_object * obj)9376 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
9377 {
9378 return bpf_object__prev_map(obj, next);
9379 }
9380
9381 struct bpf_map *
bpf_object__prev_map(const struct bpf_object * obj,const struct bpf_map * next)9382 bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next)
9383 {
9384 if (next == NULL) {
9385 if (!obj->nr_maps)
9386 return NULL;
9387 return obj->maps + obj->nr_maps - 1;
9388 }
9389
9390 return __bpf_map__iter(next, obj, -1);
9391 }
9392
9393 struct bpf_map *
bpf_object__find_map_by_name(const struct bpf_object * obj,const char * name)9394 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
9395 {
9396 struct bpf_map *pos;
9397
9398 bpf_object__for_each_map(pos, obj) {
9399 /* if it's a special internal map name (which always starts
9400 * with dot) then check if that special name matches the
9401 * real map name (ELF section name)
9402 */
9403 if (name[0] == '.') {
9404 if (pos->real_name && strcmp(pos->real_name, name) == 0)
9405 return pos;
9406 continue;
9407 }
9408 /* otherwise map name has to be an exact match */
9409 if (map_uses_real_name(pos)) {
9410 if (strcmp(pos->real_name, name) == 0)
9411 return pos;
9412 continue;
9413 }
9414 if (strcmp(pos->name, name) == 0)
9415 return pos;
9416 }
9417 return errno = ENOENT, NULL;
9418 }
9419
9420 int
bpf_object__find_map_fd_by_name(const struct bpf_object * obj,const char * name)9421 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
9422 {
9423 return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
9424 }
9425
9426 struct bpf_map *
bpf_object__find_map_by_offset(struct bpf_object * obj,size_t offset)9427 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
9428 {
9429 return libbpf_err_ptr(-ENOTSUP);
9430 }
9431
libbpf_get_error(const void * ptr)9432 long libbpf_get_error(const void *ptr)
9433 {
9434 if (!IS_ERR_OR_NULL(ptr))
9435 return 0;
9436
9437 if (IS_ERR(ptr))
9438 errno = -PTR_ERR(ptr);
9439
9440 /* If ptr == NULL, then errno should be already set by the failing
9441 * API, because libbpf never returns NULL on success and it now always
9442 * sets errno on error. So no extra errno handling for ptr == NULL
9443 * case.
9444 */
9445 return -errno;
9446 }
9447
9448 __attribute__((alias("bpf_prog_load_xattr2")))
9449 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
9450 struct bpf_object **pobj, int *prog_fd);
9451
bpf_prog_load_xattr2(const struct bpf_prog_load_attr * attr,struct bpf_object ** pobj,int * prog_fd)9452 static int bpf_prog_load_xattr2(const struct bpf_prog_load_attr *attr,
9453 struct bpf_object **pobj, int *prog_fd)
9454 {
9455 struct bpf_object_open_attr open_attr = {};
9456 struct bpf_program *prog, *first_prog = NULL;
9457 struct bpf_object *obj;
9458 struct bpf_map *map;
9459 int err;
9460
9461 if (!attr)
9462 return libbpf_err(-EINVAL);
9463 if (!attr->file)
9464 return libbpf_err(-EINVAL);
9465
9466 open_attr.file = attr->file;
9467 open_attr.prog_type = attr->prog_type;
9468
9469 obj = __bpf_object__open_xattr(&open_attr, 0);
9470 err = libbpf_get_error(obj);
9471 if (err)
9472 return libbpf_err(-ENOENT);
9473
9474 bpf_object__for_each_program(prog, obj) {
9475 enum bpf_attach_type attach_type = attr->expected_attach_type;
9476 /*
9477 * to preserve backwards compatibility, bpf_prog_load treats
9478 * attr->prog_type, if specified, as an override to whatever
9479 * bpf_object__open guessed
9480 */
9481 if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) {
9482 bpf_program__set_type(prog, attr->prog_type);
9483 bpf_program__set_expected_attach_type(prog,
9484 attach_type);
9485 }
9486 if (bpf_program__type(prog) == BPF_PROG_TYPE_UNSPEC) {
9487 /*
9488 * we haven't guessed from section name and user
9489 * didn't provide a fallback type, too bad...
9490 */
9491 bpf_object__close(obj);
9492 return libbpf_err(-EINVAL);
9493 }
9494
9495 prog->prog_ifindex = attr->ifindex;
9496 prog->log_level = attr->log_level;
9497 prog->prog_flags |= attr->prog_flags;
9498 if (!first_prog)
9499 first_prog = prog;
9500 }
9501
9502 bpf_object__for_each_map(map, obj) {
9503 if (map->def.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
9504 map->map_ifindex = attr->ifindex;
9505 }
9506
9507 if (!first_prog) {
9508 pr_warn("object file doesn't contain bpf program\n");
9509 bpf_object__close(obj);
9510 return libbpf_err(-ENOENT);
9511 }
9512
9513 err = bpf_object__load(obj);
9514 if (err) {
9515 bpf_object__close(obj);
9516 return libbpf_err(err);
9517 }
9518
9519 *pobj = obj;
9520 *prog_fd = bpf_program__fd(first_prog);
9521 return 0;
9522 }
9523
9524 COMPAT_VERSION(bpf_prog_load_deprecated, bpf_prog_load, LIBBPF_0.0.1)
bpf_prog_load_deprecated(const char * file,enum bpf_prog_type type,struct bpf_object ** pobj,int * prog_fd)9525 int bpf_prog_load_deprecated(const char *file, enum bpf_prog_type type,
9526 struct bpf_object **pobj, int *prog_fd)
9527 {
9528 struct bpf_prog_load_attr attr;
9529
9530 memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
9531 attr.file = file;
9532 attr.prog_type = type;
9533 attr.expected_attach_type = 0;
9534
9535 return bpf_prog_load_xattr2(&attr, pobj, prog_fd);
9536 }
9537
9538 struct bpf_link {
9539 int (*detach)(struct bpf_link *link);
9540 void (*dealloc)(struct bpf_link *link);
9541 char *pin_path; /* NULL, if not pinned */
9542 int fd; /* hook FD, -1 if not applicable */
9543 bool disconnected;
9544 };
9545
9546 /* Replace link's underlying BPF program with the new one */
bpf_link__update_program(struct bpf_link * link,struct bpf_program * prog)9547 int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
9548 {
9549 int ret;
9550
9551 ret = bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL);
9552 return libbpf_err_errno(ret);
9553 }
9554
9555 /* Release "ownership" of underlying BPF resource (typically, BPF program
9556 * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
9557 * link, when destructed through bpf_link__destroy() call won't attempt to
9558 * detach/unregisted that BPF resource. This is useful in situations where,
9559 * say, attached BPF program has to outlive userspace program that attached it
9560 * in the system. Depending on type of BPF program, though, there might be
9561 * additional steps (like pinning BPF program in BPF FS) necessary to ensure
9562 * exit of userspace program doesn't trigger automatic detachment and clean up
9563 * inside the kernel.
9564 */
bpf_link__disconnect(struct bpf_link * link)9565 void bpf_link__disconnect(struct bpf_link *link)
9566 {
9567 link->disconnected = true;
9568 }
9569
bpf_link__destroy(struct bpf_link * link)9570 int bpf_link__destroy(struct bpf_link *link)
9571 {
9572 int err = 0;
9573
9574 if (IS_ERR_OR_NULL(link))
9575 return 0;
9576
9577 if (!link->disconnected && link->detach)
9578 err = link->detach(link);
9579 if (link->pin_path)
9580 free(link->pin_path);
9581 if (link->dealloc)
9582 link->dealloc(link);
9583 else
9584 free(link);
9585
9586 return libbpf_err(err);
9587 }
9588
bpf_link__fd(const struct bpf_link * link)9589 int bpf_link__fd(const struct bpf_link *link)
9590 {
9591 return link->fd;
9592 }
9593
bpf_link__pin_path(const struct bpf_link * link)9594 const char *bpf_link__pin_path(const struct bpf_link *link)
9595 {
9596 return link->pin_path;
9597 }
9598
bpf_link__detach_fd(struct bpf_link * link)9599 static int bpf_link__detach_fd(struct bpf_link *link)
9600 {
9601 return libbpf_err_errno(close(link->fd));
9602 }
9603
bpf_link__open(const char * path)9604 struct bpf_link *bpf_link__open(const char *path)
9605 {
9606 struct bpf_link *link;
9607 int fd;
9608
9609 fd = bpf_obj_get(path);
9610 if (fd < 0) {
9611 fd = -errno;
9612 pr_warn("failed to open link at %s: %d\n", path, fd);
9613 return libbpf_err_ptr(fd);
9614 }
9615
9616 link = calloc(1, sizeof(*link));
9617 if (!link) {
9618 close(fd);
9619 return libbpf_err_ptr(-ENOMEM);
9620 }
9621 link->detach = &bpf_link__detach_fd;
9622 link->fd = fd;
9623
9624 link->pin_path = strdup(path);
9625 if (!link->pin_path) {
9626 bpf_link__destroy(link);
9627 return libbpf_err_ptr(-ENOMEM);
9628 }
9629
9630 return link;
9631 }
9632
bpf_link__detach(struct bpf_link * link)9633 int bpf_link__detach(struct bpf_link *link)
9634 {
9635 return bpf_link_detach(link->fd) ? -errno : 0;
9636 }
9637
bpf_link__pin(struct bpf_link * link,const char * path)9638 int bpf_link__pin(struct bpf_link *link, const char *path)
9639 {
9640 int err;
9641
9642 if (link->pin_path)
9643 return libbpf_err(-EBUSY);
9644 err = make_parent_dir(path);
9645 if (err)
9646 return libbpf_err(err);
9647 err = check_path(path);
9648 if (err)
9649 return libbpf_err(err);
9650
9651 link->pin_path = strdup(path);
9652 if (!link->pin_path)
9653 return libbpf_err(-ENOMEM);
9654
9655 if (bpf_obj_pin(link->fd, link->pin_path)) {
9656 err = -errno;
9657 zfree(&link->pin_path);
9658 return libbpf_err(err);
9659 }
9660
9661 pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
9662 return 0;
9663 }
9664
bpf_link__unpin(struct bpf_link * link)9665 int bpf_link__unpin(struct bpf_link *link)
9666 {
9667 int err;
9668
9669 if (!link->pin_path)
9670 return libbpf_err(-EINVAL);
9671
9672 err = unlink(link->pin_path);
9673 if (err != 0)
9674 return -errno;
9675
9676 pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
9677 zfree(&link->pin_path);
9678 return 0;
9679 }
9680
9681 struct bpf_link_perf {
9682 struct bpf_link link;
9683 int perf_event_fd;
9684 /* legacy kprobe support: keep track of probe identifier and type */
9685 char *legacy_probe_name;
9686 bool legacy_is_kprobe;
9687 bool legacy_is_retprobe;
9688 };
9689
9690 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe);
9691 static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe);
9692
bpf_link_perf_detach(struct bpf_link * link)9693 static int bpf_link_perf_detach(struct bpf_link *link)
9694 {
9695 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
9696 int err = 0;
9697
9698 if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0)
9699 err = -errno;
9700
9701 if (perf_link->perf_event_fd != link->fd)
9702 close(perf_link->perf_event_fd);
9703 close(link->fd);
9704
9705 /* legacy uprobe/kprobe needs to be removed after perf event fd closure */
9706 if (perf_link->legacy_probe_name) {
9707 if (perf_link->legacy_is_kprobe) {
9708 err = remove_kprobe_event_legacy(perf_link->legacy_probe_name,
9709 perf_link->legacy_is_retprobe);
9710 } else {
9711 err = remove_uprobe_event_legacy(perf_link->legacy_probe_name,
9712 perf_link->legacy_is_retprobe);
9713 }
9714 }
9715
9716 return err;
9717 }
9718
bpf_link_perf_dealloc(struct bpf_link * link)9719 static void bpf_link_perf_dealloc(struct bpf_link *link)
9720 {
9721 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
9722
9723 free(perf_link->legacy_probe_name);
9724 free(perf_link);
9725 }
9726
bpf_program__attach_perf_event_opts(const struct bpf_program * prog,int pfd,const struct bpf_perf_event_opts * opts)9727 struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd,
9728 const struct bpf_perf_event_opts *opts)
9729 {
9730 char errmsg[STRERR_BUFSIZE];
9731 struct bpf_link_perf *link;
9732 int prog_fd, link_fd = -1, err;
9733
9734 if (!OPTS_VALID(opts, bpf_perf_event_opts))
9735 return libbpf_err_ptr(-EINVAL);
9736
9737 if (pfd < 0) {
9738 pr_warn("prog '%s': invalid perf event FD %d\n",
9739 prog->name, pfd);
9740 return libbpf_err_ptr(-EINVAL);
9741 }
9742 prog_fd = bpf_program__fd(prog);
9743 if (prog_fd < 0) {
9744 pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n",
9745 prog->name);
9746 return libbpf_err_ptr(-EINVAL);
9747 }
9748
9749 link = calloc(1, sizeof(*link));
9750 if (!link)
9751 return libbpf_err_ptr(-ENOMEM);
9752 link->link.detach = &bpf_link_perf_detach;
9753 link->link.dealloc = &bpf_link_perf_dealloc;
9754 link->perf_event_fd = pfd;
9755
9756 if (kernel_supports(prog->obj, FEAT_PERF_LINK)) {
9757 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts,
9758 .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0));
9759
9760 link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts);
9761 if (link_fd < 0) {
9762 err = -errno;
9763 pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n",
9764 prog->name, pfd,
9765 err, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9766 goto err_out;
9767 }
9768 link->link.fd = link_fd;
9769 } else {
9770 if (OPTS_GET(opts, bpf_cookie, 0)) {
9771 pr_warn("prog '%s': user context value is not supported\n", prog->name);
9772 err = -EOPNOTSUPP;
9773 goto err_out;
9774 }
9775
9776 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
9777 err = -errno;
9778 pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n",
9779 prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9780 if (err == -EPROTO)
9781 pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
9782 prog->name, pfd);
9783 goto err_out;
9784 }
9785 link->link.fd = pfd;
9786 }
9787 if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
9788 err = -errno;
9789 pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n",
9790 prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9791 goto err_out;
9792 }
9793
9794 return &link->link;
9795 err_out:
9796 if (link_fd >= 0)
9797 close(link_fd);
9798 free(link);
9799 return libbpf_err_ptr(err);
9800 }
9801
bpf_program__attach_perf_event(const struct bpf_program * prog,int pfd)9802 struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd)
9803 {
9804 return bpf_program__attach_perf_event_opts(prog, pfd, NULL);
9805 }
9806
9807 /*
9808 * this function is expected to parse integer in the range of [0, 2^31-1] from
9809 * given file using scanf format string fmt. If actual parsed value is
9810 * negative, the result might be indistinguishable from error
9811 */
parse_uint_from_file(const char * file,const char * fmt)9812 static int parse_uint_from_file(const char *file, const char *fmt)
9813 {
9814 char buf[STRERR_BUFSIZE];
9815 int err, ret;
9816 FILE *f;
9817
9818 f = fopen(file, "r");
9819 if (!f) {
9820 err = -errno;
9821 pr_debug("failed to open '%s': %s\n", file,
9822 libbpf_strerror_r(err, buf, sizeof(buf)));
9823 return err;
9824 }
9825 err = fscanf(f, fmt, &ret);
9826 if (err != 1) {
9827 err = err == EOF ? -EIO : -errno;
9828 pr_debug("failed to parse '%s': %s\n", file,
9829 libbpf_strerror_r(err, buf, sizeof(buf)));
9830 fclose(f);
9831 return err;
9832 }
9833 fclose(f);
9834 return ret;
9835 }
9836
determine_kprobe_perf_type(void)9837 static int determine_kprobe_perf_type(void)
9838 {
9839 const char *file = "/sys/bus/event_source/devices/kprobe/type";
9840
9841 return parse_uint_from_file(file, "%d\n");
9842 }
9843
determine_uprobe_perf_type(void)9844 static int determine_uprobe_perf_type(void)
9845 {
9846 const char *file = "/sys/bus/event_source/devices/uprobe/type";
9847
9848 return parse_uint_from_file(file, "%d\n");
9849 }
9850
determine_kprobe_retprobe_bit(void)9851 static int determine_kprobe_retprobe_bit(void)
9852 {
9853 const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
9854
9855 return parse_uint_from_file(file, "config:%d\n");
9856 }
9857
determine_uprobe_retprobe_bit(void)9858 static int determine_uprobe_retprobe_bit(void)
9859 {
9860 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
9861
9862 return parse_uint_from_file(file, "config:%d\n");
9863 }
9864
9865 #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32
9866 #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32
9867
perf_event_open_probe(bool uprobe,bool retprobe,const char * name,uint64_t offset,int pid,size_t ref_ctr_off)9868 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
9869 uint64_t offset, int pid, size_t ref_ctr_off)
9870 {
9871 struct perf_event_attr attr = {};
9872 char errmsg[STRERR_BUFSIZE];
9873 int type, pfd, err;
9874
9875 if (ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS))
9876 return -EINVAL;
9877
9878 type = uprobe ? determine_uprobe_perf_type()
9879 : determine_kprobe_perf_type();
9880 if (type < 0) {
9881 pr_warn("failed to determine %s perf type: %s\n",
9882 uprobe ? "uprobe" : "kprobe",
9883 libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
9884 return type;
9885 }
9886 if (retprobe) {
9887 int bit = uprobe ? determine_uprobe_retprobe_bit()
9888 : determine_kprobe_retprobe_bit();
9889
9890 if (bit < 0) {
9891 pr_warn("failed to determine %s retprobe bit: %s\n",
9892 uprobe ? "uprobe" : "kprobe",
9893 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
9894 return bit;
9895 }
9896 attr.config |= 1 << bit;
9897 }
9898 attr.size = sizeof(attr);
9899 attr.type = type;
9900 attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT;
9901 attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
9902 attr.config2 = offset; /* kprobe_addr or probe_offset */
9903
9904 /* pid filter is meaningful only for uprobes */
9905 pfd = syscall(__NR_perf_event_open, &attr,
9906 pid < 0 ? -1 : pid /* pid */,
9907 pid == -1 ? 0 : -1 /* cpu */,
9908 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
9909 if (pfd < 0) {
9910 err = -errno;
9911 pr_warn("%s perf_event_open() failed: %s\n",
9912 uprobe ? "uprobe" : "kprobe",
9913 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9914 return err;
9915 }
9916 return pfd;
9917 }
9918
append_to_file(const char * file,const char * fmt,...)9919 static int append_to_file(const char *file, const char *fmt, ...)
9920 {
9921 int fd, n, err = 0;
9922 va_list ap;
9923
9924 fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0);
9925 if (fd < 0)
9926 return -errno;
9927
9928 va_start(ap, fmt);
9929 n = vdprintf(fd, fmt, ap);
9930 va_end(ap);
9931
9932 if (n < 0)
9933 err = -errno;
9934
9935 close(fd);
9936 return err;
9937 }
9938
gen_kprobe_legacy_event_name(char * buf,size_t buf_sz,const char * kfunc_name,size_t offset)9939 static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz,
9940 const char *kfunc_name, size_t offset)
9941 {
9942 static int index = 0;
9943
9944 snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset,
9945 __sync_fetch_and_add(&index, 1));
9946 }
9947
add_kprobe_event_legacy(const char * probe_name,bool retprobe,const char * kfunc_name,size_t offset)9948 static int add_kprobe_event_legacy(const char *probe_name, bool retprobe,
9949 const char *kfunc_name, size_t offset)
9950 {
9951 const char *file = "/sys/kernel/debug/tracing/kprobe_events";
9952
9953 return append_to_file(file, "%c:%s/%s %s+0x%zx",
9954 retprobe ? 'r' : 'p',
9955 retprobe ? "kretprobes" : "kprobes",
9956 probe_name, kfunc_name, offset);
9957 }
9958
remove_kprobe_event_legacy(const char * probe_name,bool retprobe)9959 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe)
9960 {
9961 const char *file = "/sys/kernel/debug/tracing/kprobe_events";
9962
9963 return append_to_file(file, "-:%s/%s", retprobe ? "kretprobes" : "kprobes", probe_name);
9964 }
9965
determine_kprobe_perf_type_legacy(const char * probe_name,bool retprobe)9966 static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe)
9967 {
9968 char file[256];
9969
9970 snprintf(file, sizeof(file),
9971 "/sys/kernel/debug/tracing/events/%s/%s/id",
9972 retprobe ? "kretprobes" : "kprobes", probe_name);
9973
9974 return parse_uint_from_file(file, "%d\n");
9975 }
9976
perf_event_kprobe_open_legacy(const char * probe_name,bool retprobe,const char * kfunc_name,size_t offset,int pid)9977 static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe,
9978 const char *kfunc_name, size_t offset, int pid)
9979 {
9980 struct perf_event_attr attr = {};
9981 char errmsg[STRERR_BUFSIZE];
9982 int type, pfd, err;
9983
9984 err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset);
9985 if (err < 0) {
9986 pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n",
9987 kfunc_name, offset,
9988 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9989 return err;
9990 }
9991 type = determine_kprobe_perf_type_legacy(probe_name, retprobe);
9992 if (type < 0) {
9993 pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n",
9994 kfunc_name, offset,
9995 libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
9996 return type;
9997 }
9998 attr.size = sizeof(attr);
9999 attr.config = type;
10000 attr.type = PERF_TYPE_TRACEPOINT;
10001
10002 pfd = syscall(__NR_perf_event_open, &attr,
10003 pid < 0 ? -1 : pid, /* pid */
10004 pid == -1 ? 0 : -1, /* cpu */
10005 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10006 if (pfd < 0) {
10007 err = -errno;
10008 pr_warn("legacy kprobe perf_event_open() failed: %s\n",
10009 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10010 return err;
10011 }
10012 return pfd;
10013 }
10014
10015 struct bpf_link *
bpf_program__attach_kprobe_opts(const struct bpf_program * prog,const char * func_name,const struct bpf_kprobe_opts * opts)10016 bpf_program__attach_kprobe_opts(const struct bpf_program *prog,
10017 const char *func_name,
10018 const struct bpf_kprobe_opts *opts)
10019 {
10020 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
10021 char errmsg[STRERR_BUFSIZE];
10022 char *legacy_probe = NULL;
10023 struct bpf_link *link;
10024 size_t offset;
10025 bool retprobe, legacy;
10026 int pfd, err;
10027
10028 if (!OPTS_VALID(opts, bpf_kprobe_opts))
10029 return libbpf_err_ptr(-EINVAL);
10030
10031 retprobe = OPTS_GET(opts, retprobe, false);
10032 offset = OPTS_GET(opts, offset, 0);
10033 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10034
10035 legacy = determine_kprobe_perf_type() < 0;
10036 if (!legacy) {
10037 pfd = perf_event_open_probe(false /* uprobe */, retprobe,
10038 func_name, offset,
10039 -1 /* pid */, 0 /* ref_ctr_off */);
10040 } else {
10041 char probe_name[256];
10042
10043 gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name),
10044 func_name, offset);
10045
10046 legacy_probe = strdup(probe_name);
10047 if (!legacy_probe)
10048 return libbpf_err_ptr(-ENOMEM);
10049
10050 pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name,
10051 offset, -1 /* pid */);
10052 }
10053 if (pfd < 0) {
10054 err = -errno;
10055 pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n",
10056 prog->name, retprobe ? "kretprobe" : "kprobe",
10057 func_name, offset,
10058 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10059 goto err_out;
10060 }
10061 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
10062 err = libbpf_get_error(link);
10063 if (err) {
10064 close(pfd);
10065 pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n",
10066 prog->name, retprobe ? "kretprobe" : "kprobe",
10067 func_name, offset,
10068 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10069 goto err_out;
10070 }
10071 if (legacy) {
10072 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10073
10074 perf_link->legacy_probe_name = legacy_probe;
10075 perf_link->legacy_is_kprobe = true;
10076 perf_link->legacy_is_retprobe = retprobe;
10077 }
10078
10079 return link;
10080 err_out:
10081 free(legacy_probe);
10082 return libbpf_err_ptr(err);
10083 }
10084
bpf_program__attach_kprobe(const struct bpf_program * prog,bool retprobe,const char * func_name)10085 struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog,
10086 bool retprobe,
10087 const char *func_name)
10088 {
10089 DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts,
10090 .retprobe = retprobe,
10091 );
10092
10093 return bpf_program__attach_kprobe_opts(prog, func_name, &opts);
10094 }
10095
attach_kprobe(const struct bpf_program * prog,long cookie)10096 static struct bpf_link *attach_kprobe(const struct bpf_program *prog, long cookie)
10097 {
10098 DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts);
10099 unsigned long offset = 0;
10100 struct bpf_link *link;
10101 const char *func_name;
10102 char *func;
10103 int n, err;
10104
10105 opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/");
10106 if (opts.retprobe)
10107 func_name = prog->sec_name + sizeof("kretprobe/") - 1;
10108 else
10109 func_name = prog->sec_name + sizeof("kprobe/") - 1;
10110
10111 n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset);
10112 if (n < 1) {
10113 err = -EINVAL;
10114 pr_warn("kprobe name is invalid: %s\n", func_name);
10115 return libbpf_err_ptr(err);
10116 }
10117 if (opts.retprobe && offset != 0) {
10118 free(func);
10119 err = -EINVAL;
10120 pr_warn("kretprobes do not support offset specification\n");
10121 return libbpf_err_ptr(err);
10122 }
10123
10124 opts.offset = offset;
10125 link = bpf_program__attach_kprobe_opts(prog, func, &opts);
10126 free(func);
10127 return link;
10128 }
10129
gen_uprobe_legacy_event_name(char * buf,size_t buf_sz,const char * binary_path,uint64_t offset)10130 static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz,
10131 const char *binary_path, uint64_t offset)
10132 {
10133 int i;
10134
10135 snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset);
10136
10137 /* sanitize binary_path in the probe name */
10138 for (i = 0; buf[i]; i++) {
10139 if (!isalnum(buf[i]))
10140 buf[i] = '_';
10141 }
10142 }
10143
add_uprobe_event_legacy(const char * probe_name,bool retprobe,const char * binary_path,size_t offset)10144 static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe,
10145 const char *binary_path, size_t offset)
10146 {
10147 const char *file = "/sys/kernel/debug/tracing/uprobe_events";
10148
10149 return append_to_file(file, "%c:%s/%s %s:0x%zx",
10150 retprobe ? 'r' : 'p',
10151 retprobe ? "uretprobes" : "uprobes",
10152 probe_name, binary_path, offset);
10153 }
10154
remove_uprobe_event_legacy(const char * probe_name,bool retprobe)10155 static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe)
10156 {
10157 const char *file = "/sys/kernel/debug/tracing/uprobe_events";
10158
10159 return append_to_file(file, "-:%s/%s", retprobe ? "uretprobes" : "uprobes", probe_name);
10160 }
10161
determine_uprobe_perf_type_legacy(const char * probe_name,bool retprobe)10162 static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe)
10163 {
10164 char file[512];
10165
10166 snprintf(file, sizeof(file),
10167 "/sys/kernel/debug/tracing/events/%s/%s/id",
10168 retprobe ? "uretprobes" : "uprobes", probe_name);
10169
10170 return parse_uint_from_file(file, "%d\n");
10171 }
10172
perf_event_uprobe_open_legacy(const char * probe_name,bool retprobe,const char * binary_path,size_t offset,int pid)10173 static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe,
10174 const char *binary_path, size_t offset, int pid)
10175 {
10176 struct perf_event_attr attr;
10177 int type, pfd, err;
10178
10179 err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset);
10180 if (err < 0) {
10181 pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n",
10182 binary_path, (size_t)offset, err);
10183 return err;
10184 }
10185 type = determine_uprobe_perf_type_legacy(probe_name, retprobe);
10186 if (type < 0) {
10187 pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n",
10188 binary_path, offset, err);
10189 return type;
10190 }
10191
10192 memset(&attr, 0, sizeof(attr));
10193 attr.size = sizeof(attr);
10194 attr.config = type;
10195 attr.type = PERF_TYPE_TRACEPOINT;
10196
10197 pfd = syscall(__NR_perf_event_open, &attr,
10198 pid < 0 ? -1 : pid, /* pid */
10199 pid == -1 ? 0 : -1, /* cpu */
10200 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10201 if (pfd < 0) {
10202 err = -errno;
10203 pr_warn("legacy uprobe perf_event_open() failed: %d\n", err);
10204 return err;
10205 }
10206 return pfd;
10207 }
10208
10209 LIBBPF_API struct bpf_link *
bpf_program__attach_uprobe_opts(const struct bpf_program * prog,pid_t pid,const char * binary_path,size_t func_offset,const struct bpf_uprobe_opts * opts)10210 bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid,
10211 const char *binary_path, size_t func_offset,
10212 const struct bpf_uprobe_opts *opts)
10213 {
10214 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
10215 char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL;
10216 struct bpf_link *link;
10217 size_t ref_ctr_off;
10218 int pfd, err;
10219 bool retprobe, legacy;
10220
10221 if (!OPTS_VALID(opts, bpf_uprobe_opts))
10222 return libbpf_err_ptr(-EINVAL);
10223
10224 retprobe = OPTS_GET(opts, retprobe, false);
10225 ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0);
10226 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10227
10228 legacy = determine_uprobe_perf_type() < 0;
10229 if (!legacy) {
10230 pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path,
10231 func_offset, pid, ref_ctr_off);
10232 } else {
10233 char probe_name[512];
10234
10235 if (ref_ctr_off)
10236 return libbpf_err_ptr(-EINVAL);
10237
10238 gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name),
10239 binary_path, func_offset);
10240
10241 legacy_probe = strdup(probe_name);
10242 if (!legacy_probe)
10243 return libbpf_err_ptr(-ENOMEM);
10244
10245 pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe,
10246 binary_path, func_offset, pid);
10247 }
10248 if (pfd < 0) {
10249 err = -errno;
10250 pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
10251 prog->name, retprobe ? "uretprobe" : "uprobe",
10252 binary_path, func_offset,
10253 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10254 goto err_out;
10255 }
10256
10257 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
10258 err = libbpf_get_error(link);
10259 if (err) {
10260 close(pfd);
10261 pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n",
10262 prog->name, retprobe ? "uretprobe" : "uprobe",
10263 binary_path, func_offset,
10264 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10265 goto err_out;
10266 }
10267 if (legacy) {
10268 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10269
10270 perf_link->legacy_probe_name = legacy_probe;
10271 perf_link->legacy_is_kprobe = false;
10272 perf_link->legacy_is_retprobe = retprobe;
10273 }
10274 return link;
10275 err_out:
10276 free(legacy_probe);
10277 return libbpf_err_ptr(err);
10278
10279 }
10280
bpf_program__attach_uprobe(const struct bpf_program * prog,bool retprobe,pid_t pid,const char * binary_path,size_t func_offset)10281 struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog,
10282 bool retprobe, pid_t pid,
10283 const char *binary_path,
10284 size_t func_offset)
10285 {
10286 DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe);
10287
10288 return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts);
10289 }
10290
determine_tracepoint_id(const char * tp_category,const char * tp_name)10291 static int determine_tracepoint_id(const char *tp_category,
10292 const char *tp_name)
10293 {
10294 char file[PATH_MAX];
10295 int ret;
10296
10297 ret = snprintf(file, sizeof(file),
10298 "/sys/kernel/debug/tracing/events/%s/%s/id",
10299 tp_category, tp_name);
10300 if (ret < 0)
10301 return -errno;
10302 if (ret >= sizeof(file)) {
10303 pr_debug("tracepoint %s/%s path is too long\n",
10304 tp_category, tp_name);
10305 return -E2BIG;
10306 }
10307 return parse_uint_from_file(file, "%d\n");
10308 }
10309
perf_event_open_tracepoint(const char * tp_category,const char * tp_name)10310 static int perf_event_open_tracepoint(const char *tp_category,
10311 const char *tp_name)
10312 {
10313 struct perf_event_attr attr = {};
10314 char errmsg[STRERR_BUFSIZE];
10315 int tp_id, pfd, err;
10316
10317 tp_id = determine_tracepoint_id(tp_category, tp_name);
10318 if (tp_id < 0) {
10319 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
10320 tp_category, tp_name,
10321 libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
10322 return tp_id;
10323 }
10324
10325 attr.type = PERF_TYPE_TRACEPOINT;
10326 attr.size = sizeof(attr);
10327 attr.config = tp_id;
10328
10329 pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
10330 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10331 if (pfd < 0) {
10332 err = -errno;
10333 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
10334 tp_category, tp_name,
10335 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10336 return err;
10337 }
10338 return pfd;
10339 }
10340
bpf_program__attach_tracepoint_opts(const struct bpf_program * prog,const char * tp_category,const char * tp_name,const struct bpf_tracepoint_opts * opts)10341 struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog,
10342 const char *tp_category,
10343 const char *tp_name,
10344 const struct bpf_tracepoint_opts *opts)
10345 {
10346 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
10347 char errmsg[STRERR_BUFSIZE];
10348 struct bpf_link *link;
10349 int pfd, err;
10350
10351 if (!OPTS_VALID(opts, bpf_tracepoint_opts))
10352 return libbpf_err_ptr(-EINVAL);
10353
10354 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10355
10356 pfd = perf_event_open_tracepoint(tp_category, tp_name);
10357 if (pfd < 0) {
10358 pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
10359 prog->name, tp_category, tp_name,
10360 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
10361 return libbpf_err_ptr(pfd);
10362 }
10363 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
10364 err = libbpf_get_error(link);
10365 if (err) {
10366 close(pfd);
10367 pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n",
10368 prog->name, tp_category, tp_name,
10369 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10370 return libbpf_err_ptr(err);
10371 }
10372 return link;
10373 }
10374
bpf_program__attach_tracepoint(const struct bpf_program * prog,const char * tp_category,const char * tp_name)10375 struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog,
10376 const char *tp_category,
10377 const char *tp_name)
10378 {
10379 return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL);
10380 }
10381
attach_tp(const struct bpf_program * prog,long cookie)10382 static struct bpf_link *attach_tp(const struct bpf_program *prog, long cookie)
10383 {
10384 char *sec_name, *tp_cat, *tp_name;
10385 struct bpf_link *link;
10386
10387 sec_name = strdup(prog->sec_name);
10388 if (!sec_name)
10389 return libbpf_err_ptr(-ENOMEM);
10390
10391 /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */
10392 if (str_has_pfx(prog->sec_name, "tp/"))
10393 tp_cat = sec_name + sizeof("tp/") - 1;
10394 else
10395 tp_cat = sec_name + sizeof("tracepoint/") - 1;
10396 tp_name = strchr(tp_cat, '/');
10397 if (!tp_name) {
10398 free(sec_name);
10399 return libbpf_err_ptr(-EINVAL);
10400 }
10401 *tp_name = '\0';
10402 tp_name++;
10403
10404 link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
10405 free(sec_name);
10406 return link;
10407 }
10408
bpf_program__attach_raw_tracepoint(const struct bpf_program * prog,const char * tp_name)10409 struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog,
10410 const char *tp_name)
10411 {
10412 char errmsg[STRERR_BUFSIZE];
10413 struct bpf_link *link;
10414 int prog_fd, pfd;
10415
10416 prog_fd = bpf_program__fd(prog);
10417 if (prog_fd < 0) {
10418 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
10419 return libbpf_err_ptr(-EINVAL);
10420 }
10421
10422 link = calloc(1, sizeof(*link));
10423 if (!link)
10424 return libbpf_err_ptr(-ENOMEM);
10425 link->detach = &bpf_link__detach_fd;
10426
10427 pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
10428 if (pfd < 0) {
10429 pfd = -errno;
10430 free(link);
10431 pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n",
10432 prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
10433 return libbpf_err_ptr(pfd);
10434 }
10435 link->fd = pfd;
10436 return link;
10437 }
10438
attach_raw_tp(const struct bpf_program * prog,long cookie)10439 static struct bpf_link *attach_raw_tp(const struct bpf_program *prog, long cookie)
10440 {
10441 static const char *const prefixes[] = {
10442 "raw_tp/",
10443 "raw_tracepoint/",
10444 "raw_tp.w/",
10445 "raw_tracepoint.w/",
10446 };
10447 size_t i;
10448 const char *tp_name = NULL;
10449
10450 for (i = 0; i < ARRAY_SIZE(prefixes); i++) {
10451 if (str_has_pfx(prog->sec_name, prefixes[i])) {
10452 tp_name = prog->sec_name + strlen(prefixes[i]);
10453 break;
10454 }
10455 }
10456 if (!tp_name) {
10457 pr_warn("prog '%s': invalid section name '%s'\n",
10458 prog->name, prog->sec_name);
10459 return libbpf_err_ptr(-EINVAL);
10460 }
10461
10462 return bpf_program__attach_raw_tracepoint(prog, tp_name);
10463 }
10464
10465 /* Common logic for all BPF program types that attach to a btf_id */
bpf_program__attach_btf_id(const struct bpf_program * prog)10466 static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog)
10467 {
10468 char errmsg[STRERR_BUFSIZE];
10469 struct bpf_link *link;
10470 int prog_fd, pfd;
10471
10472 prog_fd = bpf_program__fd(prog);
10473 if (prog_fd < 0) {
10474 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
10475 return libbpf_err_ptr(-EINVAL);
10476 }
10477
10478 link = calloc(1, sizeof(*link));
10479 if (!link)
10480 return libbpf_err_ptr(-ENOMEM);
10481 link->detach = &bpf_link__detach_fd;
10482
10483 pfd = bpf_raw_tracepoint_open(NULL, prog_fd);
10484 if (pfd < 0) {
10485 pfd = -errno;
10486 free(link);
10487 pr_warn("prog '%s': failed to attach: %s\n",
10488 prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
10489 return libbpf_err_ptr(pfd);
10490 }
10491 link->fd = pfd;
10492 return (struct bpf_link *)link;
10493 }
10494
bpf_program__attach_trace(const struct bpf_program * prog)10495 struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog)
10496 {
10497 return bpf_program__attach_btf_id(prog);
10498 }
10499
bpf_program__attach_lsm(const struct bpf_program * prog)10500 struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog)
10501 {
10502 return bpf_program__attach_btf_id(prog);
10503 }
10504
attach_trace(const struct bpf_program * prog,long cookie)10505 static struct bpf_link *attach_trace(const struct bpf_program *prog, long cookie)
10506 {
10507 return bpf_program__attach_trace(prog);
10508 }
10509
attach_lsm(const struct bpf_program * prog,long cookie)10510 static struct bpf_link *attach_lsm(const struct bpf_program *prog, long cookie)
10511 {
10512 return bpf_program__attach_lsm(prog);
10513 }
10514
10515 static struct bpf_link *
bpf_program__attach_fd(const struct bpf_program * prog,int target_fd,int btf_id,const char * target_name)10516 bpf_program__attach_fd(const struct bpf_program *prog, int target_fd, int btf_id,
10517 const char *target_name)
10518 {
10519 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts,
10520 .target_btf_id = btf_id);
10521 enum bpf_attach_type attach_type;
10522 char errmsg[STRERR_BUFSIZE];
10523 struct bpf_link *link;
10524 int prog_fd, link_fd;
10525
10526 prog_fd = bpf_program__fd(prog);
10527 if (prog_fd < 0) {
10528 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
10529 return libbpf_err_ptr(-EINVAL);
10530 }
10531
10532 link = calloc(1, sizeof(*link));
10533 if (!link)
10534 return libbpf_err_ptr(-ENOMEM);
10535 link->detach = &bpf_link__detach_fd;
10536
10537 attach_type = bpf_program__expected_attach_type(prog);
10538 link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts);
10539 if (link_fd < 0) {
10540 link_fd = -errno;
10541 free(link);
10542 pr_warn("prog '%s': failed to attach to %s: %s\n",
10543 prog->name, target_name,
10544 libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
10545 return libbpf_err_ptr(link_fd);
10546 }
10547 link->fd = link_fd;
10548 return link;
10549 }
10550
10551 struct bpf_link *
bpf_program__attach_cgroup(const struct bpf_program * prog,int cgroup_fd)10552 bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd)
10553 {
10554 return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup");
10555 }
10556
10557 struct bpf_link *
bpf_program__attach_netns(const struct bpf_program * prog,int netns_fd)10558 bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd)
10559 {
10560 return bpf_program__attach_fd(prog, netns_fd, 0, "netns");
10561 }
10562
bpf_program__attach_xdp(const struct bpf_program * prog,int ifindex)10563 struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex)
10564 {
10565 /* target_fd/target_ifindex use the same field in LINK_CREATE */
10566 return bpf_program__attach_fd(prog, ifindex, 0, "xdp");
10567 }
10568
bpf_program__attach_freplace(const struct bpf_program * prog,int target_fd,const char * attach_func_name)10569 struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog,
10570 int target_fd,
10571 const char *attach_func_name)
10572 {
10573 int btf_id;
10574
10575 if (!!target_fd != !!attach_func_name) {
10576 pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n",
10577 prog->name);
10578 return libbpf_err_ptr(-EINVAL);
10579 }
10580
10581 if (prog->type != BPF_PROG_TYPE_EXT) {
10582 pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace",
10583 prog->name);
10584 return libbpf_err_ptr(-EINVAL);
10585 }
10586
10587 if (target_fd) {
10588 btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd);
10589 if (btf_id < 0)
10590 return libbpf_err_ptr(btf_id);
10591
10592 return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace");
10593 } else {
10594 /* no target, so use raw_tracepoint_open for compatibility
10595 * with old kernels
10596 */
10597 return bpf_program__attach_trace(prog);
10598 }
10599 }
10600
10601 struct bpf_link *
bpf_program__attach_iter(const struct bpf_program * prog,const struct bpf_iter_attach_opts * opts)10602 bpf_program__attach_iter(const struct bpf_program *prog,
10603 const struct bpf_iter_attach_opts *opts)
10604 {
10605 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
10606 char errmsg[STRERR_BUFSIZE];
10607 struct bpf_link *link;
10608 int prog_fd, link_fd;
10609 __u32 target_fd = 0;
10610
10611 if (!OPTS_VALID(opts, bpf_iter_attach_opts))
10612 return libbpf_err_ptr(-EINVAL);
10613
10614 link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0);
10615 link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0);
10616
10617 prog_fd = bpf_program__fd(prog);
10618 if (prog_fd < 0) {
10619 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
10620 return libbpf_err_ptr(-EINVAL);
10621 }
10622
10623 link = calloc(1, sizeof(*link));
10624 if (!link)
10625 return libbpf_err_ptr(-ENOMEM);
10626 link->detach = &bpf_link__detach_fd;
10627
10628 link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
10629 &link_create_opts);
10630 if (link_fd < 0) {
10631 link_fd = -errno;
10632 free(link);
10633 pr_warn("prog '%s': failed to attach to iterator: %s\n",
10634 prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
10635 return libbpf_err_ptr(link_fd);
10636 }
10637 link->fd = link_fd;
10638 return link;
10639 }
10640
attach_iter(const struct bpf_program * prog,long cookie)10641 static struct bpf_link *attach_iter(const struct bpf_program *prog, long cookie)
10642 {
10643 return bpf_program__attach_iter(prog, NULL);
10644 }
10645
bpf_program__attach(const struct bpf_program * prog)10646 struct bpf_link *bpf_program__attach(const struct bpf_program *prog)
10647 {
10648 if (!prog->sec_def || !prog->sec_def->attach_fn)
10649 return libbpf_err_ptr(-ESRCH);
10650
10651 return prog->sec_def->attach_fn(prog, prog->sec_def->cookie);
10652 }
10653
bpf_link__detach_struct_ops(struct bpf_link * link)10654 static int bpf_link__detach_struct_ops(struct bpf_link *link)
10655 {
10656 __u32 zero = 0;
10657
10658 if (bpf_map_delete_elem(link->fd, &zero))
10659 return -errno;
10660
10661 return 0;
10662 }
10663
bpf_map__attach_struct_ops(const struct bpf_map * map)10664 struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map)
10665 {
10666 struct bpf_struct_ops *st_ops;
10667 struct bpf_link *link;
10668 __u32 i, zero = 0;
10669 int err;
10670
10671 if (!bpf_map__is_struct_ops(map) || map->fd == -1)
10672 return libbpf_err_ptr(-EINVAL);
10673
10674 link = calloc(1, sizeof(*link));
10675 if (!link)
10676 return libbpf_err_ptr(-EINVAL);
10677
10678 st_ops = map->st_ops;
10679 for (i = 0; i < btf_vlen(st_ops->type); i++) {
10680 struct bpf_program *prog = st_ops->progs[i];
10681 void *kern_data;
10682 int prog_fd;
10683
10684 if (!prog)
10685 continue;
10686
10687 prog_fd = bpf_program__fd(prog);
10688 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
10689 *(unsigned long *)kern_data = prog_fd;
10690 }
10691
10692 err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0);
10693 if (err) {
10694 err = -errno;
10695 free(link);
10696 return libbpf_err_ptr(err);
10697 }
10698
10699 link->detach = bpf_link__detach_struct_ops;
10700 link->fd = map->fd;
10701
10702 return link;
10703 }
10704
10705 static enum bpf_perf_event_ret
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)10706 perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
10707 void **copy_mem, size_t *copy_size,
10708 bpf_perf_event_print_t fn, void *private_data)
10709 {
10710 struct perf_event_mmap_page *header = mmap_mem;
10711 __u64 data_head = ring_buffer_read_head(header);
10712 __u64 data_tail = header->data_tail;
10713 void *base = ((__u8 *)header) + page_size;
10714 int ret = LIBBPF_PERF_EVENT_CONT;
10715 struct perf_event_header *ehdr;
10716 size_t ehdr_size;
10717
10718 while (data_head != data_tail) {
10719 ehdr = base + (data_tail & (mmap_size - 1));
10720 ehdr_size = ehdr->size;
10721
10722 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
10723 void *copy_start = ehdr;
10724 size_t len_first = base + mmap_size - copy_start;
10725 size_t len_secnd = ehdr_size - len_first;
10726
10727 if (*copy_size < ehdr_size) {
10728 free(*copy_mem);
10729 *copy_mem = malloc(ehdr_size);
10730 if (!*copy_mem) {
10731 *copy_size = 0;
10732 ret = LIBBPF_PERF_EVENT_ERROR;
10733 break;
10734 }
10735 *copy_size = ehdr_size;
10736 }
10737
10738 memcpy(*copy_mem, copy_start, len_first);
10739 memcpy(*copy_mem + len_first, base, len_secnd);
10740 ehdr = *copy_mem;
10741 }
10742
10743 ret = fn(ehdr, private_data);
10744 data_tail += ehdr_size;
10745 if (ret != LIBBPF_PERF_EVENT_CONT)
10746 break;
10747 }
10748
10749 ring_buffer_write_tail(header, data_tail);
10750 return libbpf_err(ret);
10751 }
10752
10753 __attribute__((alias("perf_event_read_simple")))
10754 enum bpf_perf_event_ret
10755 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
10756 void **copy_mem, size_t *copy_size,
10757 bpf_perf_event_print_t fn, void *private_data);
10758
10759 struct perf_buffer;
10760
10761 struct perf_buffer_params {
10762 struct perf_event_attr *attr;
10763 /* if event_cb is specified, it takes precendence */
10764 perf_buffer_event_fn event_cb;
10765 /* sample_cb and lost_cb are higher-level common-case callbacks */
10766 perf_buffer_sample_fn sample_cb;
10767 perf_buffer_lost_fn lost_cb;
10768 void *ctx;
10769 int cpu_cnt;
10770 int *cpus;
10771 int *map_keys;
10772 };
10773
10774 struct perf_cpu_buf {
10775 struct perf_buffer *pb;
10776 void *base; /* mmap()'ed memory */
10777 void *buf; /* for reconstructing segmented data */
10778 size_t buf_size;
10779 int fd;
10780 int cpu;
10781 int map_key;
10782 };
10783
10784 struct perf_buffer {
10785 perf_buffer_event_fn event_cb;
10786 perf_buffer_sample_fn sample_cb;
10787 perf_buffer_lost_fn lost_cb;
10788 void *ctx; /* passed into callbacks */
10789
10790 size_t page_size;
10791 size_t mmap_size;
10792 struct perf_cpu_buf **cpu_bufs;
10793 struct epoll_event *events;
10794 int cpu_cnt; /* number of allocated CPU buffers */
10795 int epoll_fd; /* perf event FD */
10796 int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
10797 };
10798
perf_buffer__free_cpu_buf(struct perf_buffer * pb,struct perf_cpu_buf * cpu_buf)10799 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
10800 struct perf_cpu_buf *cpu_buf)
10801 {
10802 if (!cpu_buf)
10803 return;
10804 if (cpu_buf->base &&
10805 munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
10806 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
10807 if (cpu_buf->fd >= 0) {
10808 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
10809 close(cpu_buf->fd);
10810 }
10811 free(cpu_buf->buf);
10812 free(cpu_buf);
10813 }
10814
perf_buffer__free(struct perf_buffer * pb)10815 void perf_buffer__free(struct perf_buffer *pb)
10816 {
10817 int i;
10818
10819 if (IS_ERR_OR_NULL(pb))
10820 return;
10821 if (pb->cpu_bufs) {
10822 for (i = 0; i < pb->cpu_cnt; i++) {
10823 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
10824
10825 if (!cpu_buf)
10826 continue;
10827
10828 bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
10829 perf_buffer__free_cpu_buf(pb, cpu_buf);
10830 }
10831 free(pb->cpu_bufs);
10832 }
10833 if (pb->epoll_fd >= 0)
10834 close(pb->epoll_fd);
10835 free(pb->events);
10836 free(pb);
10837 }
10838
10839 static struct perf_cpu_buf *
perf_buffer__open_cpu_buf(struct perf_buffer * pb,struct perf_event_attr * attr,int cpu,int map_key)10840 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
10841 int cpu, int map_key)
10842 {
10843 struct perf_cpu_buf *cpu_buf;
10844 char msg[STRERR_BUFSIZE];
10845 int err;
10846
10847 cpu_buf = calloc(1, sizeof(*cpu_buf));
10848 if (!cpu_buf)
10849 return ERR_PTR(-ENOMEM);
10850
10851 cpu_buf->pb = pb;
10852 cpu_buf->cpu = cpu;
10853 cpu_buf->map_key = map_key;
10854
10855 cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
10856 -1, PERF_FLAG_FD_CLOEXEC);
10857 if (cpu_buf->fd < 0) {
10858 err = -errno;
10859 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
10860 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
10861 goto error;
10862 }
10863
10864 cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
10865 PROT_READ | PROT_WRITE, MAP_SHARED,
10866 cpu_buf->fd, 0);
10867 if (cpu_buf->base == MAP_FAILED) {
10868 cpu_buf->base = NULL;
10869 err = -errno;
10870 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
10871 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
10872 goto error;
10873 }
10874
10875 if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
10876 err = -errno;
10877 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
10878 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
10879 goto error;
10880 }
10881
10882 return cpu_buf;
10883
10884 error:
10885 perf_buffer__free_cpu_buf(pb, cpu_buf);
10886 return (struct perf_cpu_buf *)ERR_PTR(err);
10887 }
10888
10889 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
10890 struct perf_buffer_params *p);
10891
10892 DEFAULT_VERSION(perf_buffer__new_v0_6_0, perf_buffer__new, LIBBPF_0.6.0)
perf_buffer__new_v0_6_0(int map_fd,size_t page_cnt,perf_buffer_sample_fn sample_cb,perf_buffer_lost_fn lost_cb,void * ctx,const struct perf_buffer_opts * opts)10893 struct perf_buffer *perf_buffer__new_v0_6_0(int map_fd, size_t page_cnt,
10894 perf_buffer_sample_fn sample_cb,
10895 perf_buffer_lost_fn lost_cb,
10896 void *ctx,
10897 const struct perf_buffer_opts *opts)
10898 {
10899 struct perf_buffer_params p = {};
10900 struct perf_event_attr attr = {};
10901
10902 if (!OPTS_VALID(opts, perf_buffer_opts))
10903 return libbpf_err_ptr(-EINVAL);
10904
10905 attr.config = PERF_COUNT_SW_BPF_OUTPUT;
10906 attr.type = PERF_TYPE_SOFTWARE;
10907 attr.sample_type = PERF_SAMPLE_RAW;
10908 attr.sample_period = 1;
10909 attr.wakeup_events = 1;
10910
10911 p.attr = &attr;
10912 p.sample_cb = sample_cb;
10913 p.lost_cb = lost_cb;
10914 p.ctx = ctx;
10915
10916 return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
10917 }
10918
10919 COMPAT_VERSION(perf_buffer__new_deprecated, perf_buffer__new, LIBBPF_0.0.4)
perf_buffer__new_deprecated(int map_fd,size_t page_cnt,const struct perf_buffer_opts * opts)10920 struct perf_buffer *perf_buffer__new_deprecated(int map_fd, size_t page_cnt,
10921 const struct perf_buffer_opts *opts)
10922 {
10923 return perf_buffer__new_v0_6_0(map_fd, page_cnt,
10924 opts ? opts->sample_cb : NULL,
10925 opts ? opts->lost_cb : NULL,
10926 opts ? opts->ctx : NULL,
10927 NULL);
10928 }
10929
10930 DEFAULT_VERSION(perf_buffer__new_raw_v0_6_0, perf_buffer__new_raw, LIBBPF_0.6.0)
perf_buffer__new_raw_v0_6_0(int map_fd,size_t page_cnt,struct perf_event_attr * attr,perf_buffer_event_fn event_cb,void * ctx,const struct perf_buffer_raw_opts * opts)10931 struct perf_buffer *perf_buffer__new_raw_v0_6_0(int map_fd, size_t page_cnt,
10932 struct perf_event_attr *attr,
10933 perf_buffer_event_fn event_cb, void *ctx,
10934 const struct perf_buffer_raw_opts *opts)
10935 {
10936 struct perf_buffer_params p = {};
10937
10938 if (page_cnt == 0 || !attr)
10939 return libbpf_err_ptr(-EINVAL);
10940
10941 if (!OPTS_VALID(opts, perf_buffer_raw_opts))
10942 return libbpf_err_ptr(-EINVAL);
10943
10944 p.attr = attr;
10945 p.event_cb = event_cb;
10946 p.ctx = ctx;
10947 p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0);
10948 p.cpus = OPTS_GET(opts, cpus, NULL);
10949 p.map_keys = OPTS_GET(opts, map_keys, NULL);
10950
10951 return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
10952 }
10953
10954 COMPAT_VERSION(perf_buffer__new_raw_deprecated, perf_buffer__new_raw, LIBBPF_0.0.4)
perf_buffer__new_raw_deprecated(int map_fd,size_t page_cnt,const struct perf_buffer_raw_opts * opts)10955 struct perf_buffer *perf_buffer__new_raw_deprecated(int map_fd, size_t page_cnt,
10956 const struct perf_buffer_raw_opts *opts)
10957 {
10958 LIBBPF_OPTS(perf_buffer_raw_opts, inner_opts,
10959 .cpu_cnt = opts->cpu_cnt,
10960 .cpus = opts->cpus,
10961 .map_keys = opts->map_keys,
10962 );
10963
10964 return perf_buffer__new_raw_v0_6_0(map_fd, page_cnt, opts->attr,
10965 opts->event_cb, opts->ctx, &inner_opts);
10966 }
10967
__perf_buffer__new(int map_fd,size_t page_cnt,struct perf_buffer_params * p)10968 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
10969 struct perf_buffer_params *p)
10970 {
10971 const char *online_cpus_file = "/sys/devices/system/cpu/online";
10972 struct bpf_map_info map;
10973 char msg[STRERR_BUFSIZE];
10974 struct perf_buffer *pb;
10975 bool *online = NULL;
10976 __u32 map_info_len;
10977 int err, i, j, n;
10978
10979 if (page_cnt & (page_cnt - 1)) {
10980 pr_warn("page count should be power of two, but is %zu\n",
10981 page_cnt);
10982 return ERR_PTR(-EINVAL);
10983 }
10984
10985 /* best-effort sanity checks */
10986 memset(&map, 0, sizeof(map));
10987 map_info_len = sizeof(map);
10988 err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
10989 if (err) {
10990 err = -errno;
10991 /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
10992 * -EBADFD, -EFAULT, or -E2BIG on real error
10993 */
10994 if (err != -EINVAL) {
10995 pr_warn("failed to get map info for map FD %d: %s\n",
10996 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
10997 return ERR_PTR(err);
10998 }
10999 pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
11000 map_fd);
11001 } else {
11002 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
11003 pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
11004 map.name);
11005 return ERR_PTR(-EINVAL);
11006 }
11007 }
11008
11009 pb = calloc(1, sizeof(*pb));
11010 if (!pb)
11011 return ERR_PTR(-ENOMEM);
11012
11013 pb->event_cb = p->event_cb;
11014 pb->sample_cb = p->sample_cb;
11015 pb->lost_cb = p->lost_cb;
11016 pb->ctx = p->ctx;
11017
11018 pb->page_size = getpagesize();
11019 pb->mmap_size = pb->page_size * page_cnt;
11020 pb->map_fd = map_fd;
11021
11022 pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
11023 if (pb->epoll_fd < 0) {
11024 err = -errno;
11025 pr_warn("failed to create epoll instance: %s\n",
11026 libbpf_strerror_r(err, msg, sizeof(msg)));
11027 goto error;
11028 }
11029
11030 if (p->cpu_cnt > 0) {
11031 pb->cpu_cnt = p->cpu_cnt;
11032 } else {
11033 pb->cpu_cnt = libbpf_num_possible_cpus();
11034 if (pb->cpu_cnt < 0) {
11035 err = pb->cpu_cnt;
11036 goto error;
11037 }
11038 if (map.max_entries && map.max_entries < pb->cpu_cnt)
11039 pb->cpu_cnt = map.max_entries;
11040 }
11041
11042 pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
11043 if (!pb->events) {
11044 err = -ENOMEM;
11045 pr_warn("failed to allocate events: out of memory\n");
11046 goto error;
11047 }
11048 pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
11049 if (!pb->cpu_bufs) {
11050 err = -ENOMEM;
11051 pr_warn("failed to allocate buffers: out of memory\n");
11052 goto error;
11053 }
11054
11055 err = parse_cpu_mask_file(online_cpus_file, &online, &n);
11056 if (err) {
11057 pr_warn("failed to get online CPU mask: %d\n", err);
11058 goto error;
11059 }
11060
11061 for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
11062 struct perf_cpu_buf *cpu_buf;
11063 int cpu, map_key;
11064
11065 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
11066 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
11067
11068 /* in case user didn't explicitly requested particular CPUs to
11069 * be attached to, skip offline/not present CPUs
11070 */
11071 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
11072 continue;
11073
11074 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
11075 if (IS_ERR(cpu_buf)) {
11076 err = PTR_ERR(cpu_buf);
11077 goto error;
11078 }
11079
11080 pb->cpu_bufs[j] = cpu_buf;
11081
11082 err = bpf_map_update_elem(pb->map_fd, &map_key,
11083 &cpu_buf->fd, 0);
11084 if (err) {
11085 err = -errno;
11086 pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
11087 cpu, map_key, cpu_buf->fd,
11088 libbpf_strerror_r(err, msg, sizeof(msg)));
11089 goto error;
11090 }
11091
11092 pb->events[j].events = EPOLLIN;
11093 pb->events[j].data.ptr = cpu_buf;
11094 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
11095 &pb->events[j]) < 0) {
11096 err = -errno;
11097 pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
11098 cpu, cpu_buf->fd,
11099 libbpf_strerror_r(err, msg, sizeof(msg)));
11100 goto error;
11101 }
11102 j++;
11103 }
11104 pb->cpu_cnt = j;
11105 free(online);
11106
11107 return pb;
11108
11109 error:
11110 free(online);
11111 if (pb)
11112 perf_buffer__free(pb);
11113 return ERR_PTR(err);
11114 }
11115
11116 struct perf_sample_raw {
11117 struct perf_event_header header;
11118 uint32_t size;
11119 char data[];
11120 };
11121
11122 struct perf_sample_lost {
11123 struct perf_event_header header;
11124 uint64_t id;
11125 uint64_t lost;
11126 uint64_t sample_id;
11127 };
11128
11129 static enum bpf_perf_event_ret
perf_buffer__process_record(struct perf_event_header * e,void * ctx)11130 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
11131 {
11132 struct perf_cpu_buf *cpu_buf = ctx;
11133 struct perf_buffer *pb = cpu_buf->pb;
11134 void *data = e;
11135
11136 /* user wants full control over parsing perf event */
11137 if (pb->event_cb)
11138 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
11139
11140 switch (e->type) {
11141 case PERF_RECORD_SAMPLE: {
11142 struct perf_sample_raw *s = data;
11143
11144 if (pb->sample_cb)
11145 pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
11146 break;
11147 }
11148 case PERF_RECORD_LOST: {
11149 struct perf_sample_lost *s = data;
11150
11151 if (pb->lost_cb)
11152 pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
11153 break;
11154 }
11155 default:
11156 pr_warn("unknown perf sample type %d\n", e->type);
11157 return LIBBPF_PERF_EVENT_ERROR;
11158 }
11159 return LIBBPF_PERF_EVENT_CONT;
11160 }
11161
perf_buffer__process_records(struct perf_buffer * pb,struct perf_cpu_buf * cpu_buf)11162 static int perf_buffer__process_records(struct perf_buffer *pb,
11163 struct perf_cpu_buf *cpu_buf)
11164 {
11165 enum bpf_perf_event_ret ret;
11166
11167 ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size,
11168 pb->page_size, &cpu_buf->buf,
11169 &cpu_buf->buf_size,
11170 perf_buffer__process_record, cpu_buf);
11171 if (ret != LIBBPF_PERF_EVENT_CONT)
11172 return ret;
11173 return 0;
11174 }
11175
perf_buffer__epoll_fd(const struct perf_buffer * pb)11176 int perf_buffer__epoll_fd(const struct perf_buffer *pb)
11177 {
11178 return pb->epoll_fd;
11179 }
11180
perf_buffer__poll(struct perf_buffer * pb,int timeout_ms)11181 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
11182 {
11183 int i, cnt, err;
11184
11185 cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
11186 if (cnt < 0)
11187 return -errno;
11188
11189 for (i = 0; i < cnt; i++) {
11190 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
11191
11192 err = perf_buffer__process_records(pb, cpu_buf);
11193 if (err) {
11194 pr_warn("error while processing records: %d\n", err);
11195 return libbpf_err(err);
11196 }
11197 }
11198 return cnt;
11199 }
11200
11201 /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer
11202 * manager.
11203 */
perf_buffer__buffer_cnt(const struct perf_buffer * pb)11204 size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb)
11205 {
11206 return pb->cpu_cnt;
11207 }
11208
11209 /*
11210 * Return perf_event FD of a ring buffer in *buf_idx* slot of
11211 * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using
11212 * select()/poll()/epoll() Linux syscalls.
11213 */
perf_buffer__buffer_fd(const struct perf_buffer * pb,size_t buf_idx)11214 int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx)
11215 {
11216 struct perf_cpu_buf *cpu_buf;
11217
11218 if (buf_idx >= pb->cpu_cnt)
11219 return libbpf_err(-EINVAL);
11220
11221 cpu_buf = pb->cpu_bufs[buf_idx];
11222 if (!cpu_buf)
11223 return libbpf_err(-ENOENT);
11224
11225 return cpu_buf->fd;
11226 }
11227
11228 /*
11229 * Consume data from perf ring buffer corresponding to slot *buf_idx* in
11230 * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to
11231 * consume, do nothing and return success.
11232 * Returns:
11233 * - 0 on success;
11234 * - <0 on failure.
11235 */
perf_buffer__consume_buffer(struct perf_buffer * pb,size_t buf_idx)11236 int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx)
11237 {
11238 struct perf_cpu_buf *cpu_buf;
11239
11240 if (buf_idx >= pb->cpu_cnt)
11241 return libbpf_err(-EINVAL);
11242
11243 cpu_buf = pb->cpu_bufs[buf_idx];
11244 if (!cpu_buf)
11245 return libbpf_err(-ENOENT);
11246
11247 return perf_buffer__process_records(pb, cpu_buf);
11248 }
11249
perf_buffer__consume(struct perf_buffer * pb)11250 int perf_buffer__consume(struct perf_buffer *pb)
11251 {
11252 int i, err;
11253
11254 for (i = 0; i < pb->cpu_cnt; i++) {
11255 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
11256
11257 if (!cpu_buf)
11258 continue;
11259
11260 err = perf_buffer__process_records(pb, cpu_buf);
11261 if (err) {
11262 pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err);
11263 return libbpf_err(err);
11264 }
11265 }
11266 return 0;
11267 }
11268
11269 struct bpf_prog_info_array_desc {
11270 int array_offset; /* e.g. offset of jited_prog_insns */
11271 int count_offset; /* e.g. offset of jited_prog_len */
11272 int size_offset; /* > 0: offset of rec size,
11273 * < 0: fix size of -size_offset
11274 */
11275 };
11276
11277 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
11278 [BPF_PROG_INFO_JITED_INSNS] = {
11279 offsetof(struct bpf_prog_info, jited_prog_insns),
11280 offsetof(struct bpf_prog_info, jited_prog_len),
11281 -1,
11282 },
11283 [BPF_PROG_INFO_XLATED_INSNS] = {
11284 offsetof(struct bpf_prog_info, xlated_prog_insns),
11285 offsetof(struct bpf_prog_info, xlated_prog_len),
11286 -1,
11287 },
11288 [BPF_PROG_INFO_MAP_IDS] = {
11289 offsetof(struct bpf_prog_info, map_ids),
11290 offsetof(struct bpf_prog_info, nr_map_ids),
11291 -(int)sizeof(__u32),
11292 },
11293 [BPF_PROG_INFO_JITED_KSYMS] = {
11294 offsetof(struct bpf_prog_info, jited_ksyms),
11295 offsetof(struct bpf_prog_info, nr_jited_ksyms),
11296 -(int)sizeof(__u64),
11297 },
11298 [BPF_PROG_INFO_JITED_FUNC_LENS] = {
11299 offsetof(struct bpf_prog_info, jited_func_lens),
11300 offsetof(struct bpf_prog_info, nr_jited_func_lens),
11301 -(int)sizeof(__u32),
11302 },
11303 [BPF_PROG_INFO_FUNC_INFO] = {
11304 offsetof(struct bpf_prog_info, func_info),
11305 offsetof(struct bpf_prog_info, nr_func_info),
11306 offsetof(struct bpf_prog_info, func_info_rec_size),
11307 },
11308 [BPF_PROG_INFO_LINE_INFO] = {
11309 offsetof(struct bpf_prog_info, line_info),
11310 offsetof(struct bpf_prog_info, nr_line_info),
11311 offsetof(struct bpf_prog_info, line_info_rec_size),
11312 },
11313 [BPF_PROG_INFO_JITED_LINE_INFO] = {
11314 offsetof(struct bpf_prog_info, jited_line_info),
11315 offsetof(struct bpf_prog_info, nr_jited_line_info),
11316 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
11317 },
11318 [BPF_PROG_INFO_PROG_TAGS] = {
11319 offsetof(struct bpf_prog_info, prog_tags),
11320 offsetof(struct bpf_prog_info, nr_prog_tags),
11321 -(int)sizeof(__u8) * BPF_TAG_SIZE,
11322 },
11323
11324 };
11325
bpf_prog_info_read_offset_u32(struct bpf_prog_info * info,int offset)11326 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info,
11327 int offset)
11328 {
11329 __u32 *array = (__u32 *)info;
11330
11331 if (offset >= 0)
11332 return array[offset / sizeof(__u32)];
11333 return -(int)offset;
11334 }
11335
bpf_prog_info_read_offset_u64(struct bpf_prog_info * info,int offset)11336 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info,
11337 int offset)
11338 {
11339 __u64 *array = (__u64 *)info;
11340
11341 if (offset >= 0)
11342 return array[offset / sizeof(__u64)];
11343 return -(int)offset;
11344 }
11345
bpf_prog_info_set_offset_u32(struct bpf_prog_info * info,int offset,__u32 val)11346 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
11347 __u32 val)
11348 {
11349 __u32 *array = (__u32 *)info;
11350
11351 if (offset >= 0)
11352 array[offset / sizeof(__u32)] = val;
11353 }
11354
bpf_prog_info_set_offset_u64(struct bpf_prog_info * info,int offset,__u64 val)11355 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
11356 __u64 val)
11357 {
11358 __u64 *array = (__u64 *)info;
11359
11360 if (offset >= 0)
11361 array[offset / sizeof(__u64)] = val;
11362 }
11363
11364 struct bpf_prog_info_linear *
bpf_program__get_prog_info_linear(int fd,__u64 arrays)11365 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
11366 {
11367 struct bpf_prog_info_linear *info_linear;
11368 struct bpf_prog_info info = {};
11369 __u32 info_len = sizeof(info);
11370 __u32 data_len = 0;
11371 int i, err;
11372 void *ptr;
11373
11374 if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
11375 return libbpf_err_ptr(-EINVAL);
11376
11377 /* step 1: get array dimensions */
11378 err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
11379 if (err) {
11380 pr_debug("can't get prog info: %s", strerror(errno));
11381 return libbpf_err_ptr(-EFAULT);
11382 }
11383
11384 /* step 2: calculate total size of all arrays */
11385 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
11386 bool include_array = (arrays & (1UL << i)) > 0;
11387 struct bpf_prog_info_array_desc *desc;
11388 __u32 count, size;
11389
11390 desc = bpf_prog_info_array_desc + i;
11391
11392 /* kernel is too old to support this field */
11393 if (info_len < desc->array_offset + sizeof(__u32) ||
11394 info_len < desc->count_offset + sizeof(__u32) ||
11395 (desc->size_offset > 0 && info_len < desc->size_offset))
11396 include_array = false;
11397
11398 if (!include_array) {
11399 arrays &= ~(1UL << i); /* clear the bit */
11400 continue;
11401 }
11402
11403 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
11404 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
11405
11406 data_len += count * size;
11407 }
11408
11409 /* step 3: allocate continuous memory */
11410 data_len = roundup(data_len, sizeof(__u64));
11411 info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
11412 if (!info_linear)
11413 return libbpf_err_ptr(-ENOMEM);
11414
11415 /* step 4: fill data to info_linear->info */
11416 info_linear->arrays = arrays;
11417 memset(&info_linear->info, 0, sizeof(info));
11418 ptr = info_linear->data;
11419
11420 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
11421 struct bpf_prog_info_array_desc *desc;
11422 __u32 count, size;
11423
11424 if ((arrays & (1UL << i)) == 0)
11425 continue;
11426
11427 desc = bpf_prog_info_array_desc + i;
11428 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
11429 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
11430 bpf_prog_info_set_offset_u32(&info_linear->info,
11431 desc->count_offset, count);
11432 bpf_prog_info_set_offset_u32(&info_linear->info,
11433 desc->size_offset, size);
11434 bpf_prog_info_set_offset_u64(&info_linear->info,
11435 desc->array_offset,
11436 ptr_to_u64(ptr));
11437 ptr += count * size;
11438 }
11439
11440 /* step 5: call syscall again to get required arrays */
11441 err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
11442 if (err) {
11443 pr_debug("can't get prog info: %s", strerror(errno));
11444 free(info_linear);
11445 return libbpf_err_ptr(-EFAULT);
11446 }
11447
11448 /* step 6: verify the data */
11449 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
11450 struct bpf_prog_info_array_desc *desc;
11451 __u32 v1, v2;
11452
11453 if ((arrays & (1UL << i)) == 0)
11454 continue;
11455
11456 desc = bpf_prog_info_array_desc + i;
11457 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
11458 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
11459 desc->count_offset);
11460 if (v1 != v2)
11461 pr_warn("%s: mismatch in element count\n", __func__);
11462
11463 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
11464 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
11465 desc->size_offset);
11466 if (v1 != v2)
11467 pr_warn("%s: mismatch in rec size\n", __func__);
11468 }
11469
11470 /* step 7: update info_len and data_len */
11471 info_linear->info_len = sizeof(struct bpf_prog_info);
11472 info_linear->data_len = data_len;
11473
11474 return info_linear;
11475 }
11476
bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear * info_linear)11477 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
11478 {
11479 int i;
11480
11481 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
11482 struct bpf_prog_info_array_desc *desc;
11483 __u64 addr, offs;
11484
11485 if ((info_linear->arrays & (1UL << i)) == 0)
11486 continue;
11487
11488 desc = bpf_prog_info_array_desc + i;
11489 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
11490 desc->array_offset);
11491 offs = addr - ptr_to_u64(info_linear->data);
11492 bpf_prog_info_set_offset_u64(&info_linear->info,
11493 desc->array_offset, offs);
11494 }
11495 }
11496
bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear * info_linear)11497 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
11498 {
11499 int i;
11500
11501 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
11502 struct bpf_prog_info_array_desc *desc;
11503 __u64 addr, offs;
11504
11505 if ((info_linear->arrays & (1UL << i)) == 0)
11506 continue;
11507
11508 desc = bpf_prog_info_array_desc + i;
11509 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
11510 desc->array_offset);
11511 addr = offs + ptr_to_u64(info_linear->data);
11512 bpf_prog_info_set_offset_u64(&info_linear->info,
11513 desc->array_offset, addr);
11514 }
11515 }
11516
bpf_program__set_attach_target(struct bpf_program * prog,int attach_prog_fd,const char * attach_func_name)11517 int bpf_program__set_attach_target(struct bpf_program *prog,
11518 int attach_prog_fd,
11519 const char *attach_func_name)
11520 {
11521 int btf_obj_fd = 0, btf_id = 0, err;
11522
11523 if (!prog || attach_prog_fd < 0)
11524 return libbpf_err(-EINVAL);
11525
11526 if (prog->obj->loaded)
11527 return libbpf_err(-EINVAL);
11528
11529 if (attach_prog_fd && !attach_func_name) {
11530 /* remember attach_prog_fd and let bpf_program__load() find
11531 * BTF ID during the program load
11532 */
11533 prog->attach_prog_fd = attach_prog_fd;
11534 return 0;
11535 }
11536
11537 if (attach_prog_fd) {
11538 btf_id = libbpf_find_prog_btf_id(attach_func_name,
11539 attach_prog_fd);
11540 if (btf_id < 0)
11541 return libbpf_err(btf_id);
11542 } else {
11543 if (!attach_func_name)
11544 return libbpf_err(-EINVAL);
11545
11546 /* load btf_vmlinux, if not yet */
11547 err = bpf_object__load_vmlinux_btf(prog->obj, true);
11548 if (err)
11549 return libbpf_err(err);
11550 err = find_kernel_btf_id(prog->obj, attach_func_name,
11551 prog->expected_attach_type,
11552 &btf_obj_fd, &btf_id);
11553 if (err)
11554 return libbpf_err(err);
11555 }
11556
11557 prog->attach_btf_id = btf_id;
11558 prog->attach_btf_obj_fd = btf_obj_fd;
11559 prog->attach_prog_fd = attach_prog_fd;
11560 return 0;
11561 }
11562
parse_cpu_mask_str(const char * s,bool ** mask,int * mask_sz)11563 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
11564 {
11565 int err = 0, n, len, start, end = -1;
11566 bool *tmp;
11567
11568 *mask = NULL;
11569 *mask_sz = 0;
11570
11571 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
11572 while (*s) {
11573 if (*s == ',' || *s == '\n') {
11574 s++;
11575 continue;
11576 }
11577 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
11578 if (n <= 0 || n > 2) {
11579 pr_warn("Failed to get CPU range %s: %d\n", s, n);
11580 err = -EINVAL;
11581 goto cleanup;
11582 } else if (n == 1) {
11583 end = start;
11584 }
11585 if (start < 0 || start > end) {
11586 pr_warn("Invalid CPU range [%d,%d] in %s\n",
11587 start, end, s);
11588 err = -EINVAL;
11589 goto cleanup;
11590 }
11591 tmp = realloc(*mask, end + 1);
11592 if (!tmp) {
11593 err = -ENOMEM;
11594 goto cleanup;
11595 }
11596 *mask = tmp;
11597 memset(tmp + *mask_sz, 0, start - *mask_sz);
11598 memset(tmp + start, 1, end - start + 1);
11599 *mask_sz = end + 1;
11600 s += len;
11601 }
11602 if (!*mask_sz) {
11603 pr_warn("Empty CPU range\n");
11604 return -EINVAL;
11605 }
11606 return 0;
11607 cleanup:
11608 free(*mask);
11609 *mask = NULL;
11610 return err;
11611 }
11612
parse_cpu_mask_file(const char * fcpu,bool ** mask,int * mask_sz)11613 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
11614 {
11615 int fd, err = 0, len;
11616 char buf[128];
11617
11618 fd = open(fcpu, O_RDONLY | O_CLOEXEC);
11619 if (fd < 0) {
11620 err = -errno;
11621 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
11622 return err;
11623 }
11624 len = read(fd, buf, sizeof(buf));
11625 close(fd);
11626 if (len <= 0) {
11627 err = len ? -errno : -EINVAL;
11628 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
11629 return err;
11630 }
11631 if (len >= sizeof(buf)) {
11632 pr_warn("CPU mask is too big in file %s\n", fcpu);
11633 return -E2BIG;
11634 }
11635 buf[len] = '\0';
11636
11637 return parse_cpu_mask_str(buf, mask, mask_sz);
11638 }
11639
libbpf_num_possible_cpus(void)11640 int libbpf_num_possible_cpus(void)
11641 {
11642 static const char *fcpu = "/sys/devices/system/cpu/possible";
11643 static int cpus;
11644 int err, n, i, tmp_cpus;
11645 bool *mask;
11646
11647 tmp_cpus = READ_ONCE(cpus);
11648 if (tmp_cpus > 0)
11649 return tmp_cpus;
11650
11651 err = parse_cpu_mask_file(fcpu, &mask, &n);
11652 if (err)
11653 return libbpf_err(err);
11654
11655 tmp_cpus = 0;
11656 for (i = 0; i < n; i++) {
11657 if (mask[i])
11658 tmp_cpus++;
11659 }
11660 free(mask);
11661
11662 WRITE_ONCE(cpus, tmp_cpus);
11663 return tmp_cpus;
11664 }
11665
bpf_object__open_skeleton(struct bpf_object_skeleton * s,const struct bpf_object_open_opts * opts)11666 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
11667 const struct bpf_object_open_opts *opts)
11668 {
11669 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
11670 .object_name = s->name,
11671 );
11672 struct bpf_object *obj;
11673 int i, err;
11674
11675 /* Attempt to preserve opts->object_name, unless overriden by user
11676 * explicitly. Overwriting object name for skeletons is discouraged,
11677 * as it breaks global data maps, because they contain object name
11678 * prefix as their own map name prefix. When skeleton is generated,
11679 * bpftool is making an assumption that this name will stay the same.
11680 */
11681 if (opts) {
11682 memcpy(&skel_opts, opts, sizeof(*opts));
11683 if (!opts->object_name)
11684 skel_opts.object_name = s->name;
11685 }
11686
11687 obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
11688 err = libbpf_get_error(obj);
11689 if (err) {
11690 pr_warn("failed to initialize skeleton BPF object '%s': %d\n",
11691 s->name, err);
11692 return libbpf_err(err);
11693 }
11694
11695 *s->obj = obj;
11696
11697 for (i = 0; i < s->map_cnt; i++) {
11698 struct bpf_map **map = s->maps[i].map;
11699 const char *name = s->maps[i].name;
11700 void **mmaped = s->maps[i].mmaped;
11701
11702 *map = bpf_object__find_map_by_name(obj, name);
11703 if (!*map) {
11704 pr_warn("failed to find skeleton map '%s'\n", name);
11705 return libbpf_err(-ESRCH);
11706 }
11707
11708 /* externs shouldn't be pre-setup from user code */
11709 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
11710 *mmaped = (*map)->mmaped;
11711 }
11712
11713 for (i = 0; i < s->prog_cnt; i++) {
11714 struct bpf_program **prog = s->progs[i].prog;
11715 const char *name = s->progs[i].name;
11716
11717 *prog = bpf_object__find_program_by_name(obj, name);
11718 if (!*prog) {
11719 pr_warn("failed to find skeleton program '%s'\n", name);
11720 return libbpf_err(-ESRCH);
11721 }
11722 }
11723
11724 return 0;
11725 }
11726
bpf_object__load_skeleton(struct bpf_object_skeleton * s)11727 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
11728 {
11729 int i, err;
11730
11731 err = bpf_object__load(*s->obj);
11732 if (err) {
11733 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
11734 return libbpf_err(err);
11735 }
11736
11737 for (i = 0; i < s->map_cnt; i++) {
11738 struct bpf_map *map = *s->maps[i].map;
11739 size_t mmap_sz = bpf_map_mmap_sz(map);
11740 int prot, map_fd = bpf_map__fd(map);
11741 void **mmaped = s->maps[i].mmaped;
11742
11743 if (!mmaped)
11744 continue;
11745
11746 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
11747 *mmaped = NULL;
11748 continue;
11749 }
11750
11751 if (map->def.map_flags & BPF_F_RDONLY_PROG)
11752 prot = PROT_READ;
11753 else
11754 prot = PROT_READ | PROT_WRITE;
11755
11756 /* Remap anonymous mmap()-ed "map initialization image" as
11757 * a BPF map-backed mmap()-ed memory, but preserving the same
11758 * memory address. This will cause kernel to change process'
11759 * page table to point to a different piece of kernel memory,
11760 * but from userspace point of view memory address (and its
11761 * contents, being identical at this point) will stay the
11762 * same. This mapping will be released by bpf_object__close()
11763 * as per normal clean up procedure, so we don't need to worry
11764 * about it from skeleton's clean up perspective.
11765 */
11766 *mmaped = mmap(map->mmaped, mmap_sz, prot,
11767 MAP_SHARED | MAP_FIXED, map_fd, 0);
11768 if (*mmaped == MAP_FAILED) {
11769 err = -errno;
11770 *mmaped = NULL;
11771 pr_warn("failed to re-mmap() map '%s': %d\n",
11772 bpf_map__name(map), err);
11773 return libbpf_err(err);
11774 }
11775 }
11776
11777 return 0;
11778 }
11779
bpf_object__attach_skeleton(struct bpf_object_skeleton * s)11780 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
11781 {
11782 int i, err;
11783
11784 for (i = 0; i < s->prog_cnt; i++) {
11785 struct bpf_program *prog = *s->progs[i].prog;
11786 struct bpf_link **link = s->progs[i].link;
11787
11788 if (!prog->load)
11789 continue;
11790
11791 /* auto-attaching not supported for this program */
11792 if (!prog->sec_def || !prog->sec_def->attach_fn)
11793 continue;
11794
11795 *link = bpf_program__attach(prog);
11796 err = libbpf_get_error(*link);
11797 if (err) {
11798 pr_warn("failed to auto-attach program '%s': %d\n",
11799 bpf_program__name(prog), err);
11800 return libbpf_err(err);
11801 }
11802 }
11803
11804 return 0;
11805 }
11806
bpf_object__detach_skeleton(struct bpf_object_skeleton * s)11807 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
11808 {
11809 int i;
11810
11811 for (i = 0; i < s->prog_cnt; i++) {
11812 struct bpf_link **link = s->progs[i].link;
11813
11814 bpf_link__destroy(*link);
11815 *link = NULL;
11816 }
11817 }
11818
bpf_object__destroy_skeleton(struct bpf_object_skeleton * s)11819 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
11820 {
11821 if (!s)
11822 return;
11823
11824 if (s->progs)
11825 bpf_object__detach_skeleton(s);
11826 if (s->obj)
11827 bpf_object__close(*s->obj);
11828 free(s->maps);
11829 free(s->progs);
11830 free(s);
11831 }
11832