Commit | Line | Data |
---|---|---|
1bc38b8f | 1 | // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) |
6061a3d6 | 2 | |
1b76c13e WN |
3 | /* |
4 | * Common eBPF ELF object loading operations. | |
5 | * | |
6 | * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org> | |
7 | * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com> | |
8 | * Copyright (C) 2015 Huawei Inc. | |
f367540c | 9 | * Copyright (C) 2017 Nicira, Inc. |
d859900c | 10 | * Copyright (C) 2019 Isovalent, Inc. |
1b76c13e WN |
11 | */ |
12 | ||
b4269954 | 13 | #ifndef _GNU_SOURCE |
531b014e | 14 | #define _GNU_SOURCE |
b4269954 | 15 | #endif |
1b76c13e | 16 | #include <stdlib.h> |
b3f59d66 WN |
17 | #include <stdio.h> |
18 | #include <stdarg.h> | |
f367540c | 19 | #include <libgen.h> |
34090915 | 20 | #include <inttypes.h> |
8ab9da57 | 21 | #include <limits.h> |
b3f59d66 | 22 | #include <string.h> |
1b76c13e | 23 | #include <unistd.h> |
cdb2f920 | 24 | #include <endian.h> |
1a5e3fb1 WN |
25 | #include <fcntl.h> |
26 | #include <errno.h> | |
113e6b7e | 27 | #include <ctype.h> |
1b76c13e | 28 | #include <asm/unistd.h> |
e28ff1a8 | 29 | #include <linux/err.h> |
cb1e5e96 | 30 | #include <linux/kernel.h> |
1b76c13e | 31 | #include <linux/bpf.h> |
38d5d3b3 | 32 | #include <linux/btf.h> |
47eff617 | 33 | #include <linux/filter.h> |
9a208eff | 34 | #include <linux/list.h> |
f367540c | 35 | #include <linux/limits.h> |
438363c0 | 36 | #include <linux/perf_event.h> |
a64af0ef | 37 | #include <linux/ring_buffer.h> |
5e61f270 | 38 | #include <linux/version.h> |
fb84b822 | 39 | #include <sys/epoll.h> |
63f2f5ee | 40 | #include <sys/ioctl.h> |
fb84b822 | 41 | #include <sys/mman.h> |
f367540c JS |
42 | #include <sys/stat.h> |
43 | #include <sys/types.h> | |
44 | #include <sys/vfs.h> | |
ddc7c304 | 45 | #include <sys/utsname.h> |
dc3a2d25 | 46 | #include <sys/resource.h> |
1a5e3fb1 WN |
47 | #include <libelf.h> |
48 | #include <gelf.h> | |
166750bc | 49 | #include <zlib.h> |
1b76c13e WN |
50 | |
51 | #include "libbpf.h" | |
52d3352e | 52 | #include "bpf.h" |
8a138aed | 53 | #include "btf.h" |
6d41907c | 54 | #include "str_error.h" |
d7c4b398 | 55 | #include "libbpf_internal.h" |
ddc7c304 | 56 | #include "hashmap.h" |
67234743 | 57 | #include "bpf_gen_internal.h" |
b3f59d66 | 58 | |
f367540c JS |
59 | #ifndef BPF_FS_MAGIC |
60 | #define BPF_FS_MAGIC 0xcafe4a11 | |
61 | #endif | |
62 | ||
9c0f8cbd AN |
63 | #define BPF_INSN_SZ (sizeof(struct bpf_insn)) |
64 | ||
ff466b58 AI |
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 | ||
b3f59d66 WN |
70 | #define __printf(a, b) __attribute__((format(printf, a, b))) |
71 | ||
590a0088 | 72 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj); |
aea28a60 | 73 | static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog); |
590a0088 | 74 | |
a8a1f7d0 SF |
75 | static int __base_pr(enum libbpf_print_level level, const char *format, |
76 | va_list args) | |
b3f59d66 | 77 | { |
6f1ae8b6 YS |
78 | if (level == LIBBPF_DEBUG) |
79 | return 0; | |
80 | ||
a8a1f7d0 | 81 | return vfprintf(stderr, format, args); |
b3f59d66 WN |
82 | } |
83 | ||
a8a1f7d0 | 84 | static libbpf_print_fn_t __libbpf_pr = __base_pr; |
b3f59d66 | 85 | |
e87fd8ba | 86 | libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn) |
b3f59d66 | 87 | { |
e87fd8ba AN |
88 | libbpf_print_fn_t old_print_fn = __libbpf_pr; |
89 | ||
6f1ae8b6 | 90 | __libbpf_pr = fn; |
e87fd8ba | 91 | return old_print_fn; |
b3f59d66 | 92 | } |
1a5e3fb1 | 93 | |
8461ef8b YS |
94 | __printf(2, 3) |
95 | void libbpf_print(enum libbpf_print_level level, const char *format, ...) | |
96 | { | |
97 | va_list args; | |
98 | ||
6f1ae8b6 YS |
99 | if (!__libbpf_pr) |
100 | return; | |
101 | ||
8461ef8b | 102 | va_start(args, format); |
6f1ae8b6 | 103 | __libbpf_pr(level, format, args); |
8461ef8b YS |
104 | va_end(args); |
105 | } | |
106 | ||
dc3a2d25 THJ |
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) | |
b5c7d0d0 | 123 | snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur); |
dc3a2d25 THJ |
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 | ||
6371ca3b WN |
133 | #define STRERR_BUFSIZE 128 |
134 | ||
1a5e3fb1 WN |
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 | ||
34be1646 SL |
149 | static inline __u64 ptr_to_u64(const void *ptr) |
150 | { | |
151 | return (__u64) (unsigned long) ptr; | |
152 | } | |
153 | ||
5981881d AN |
154 | /* this goes away in libbpf 1.0 */ |
155 | enum libbpf_strict_mode libbpf_mode = LIBBPF_STRICT_NONE; | |
156 | ||
157 | int libbpf_set_strict_mode(enum libbpf_strict_mode mode) | |
158 | { | |
5981881d AN |
159 | libbpf_mode = mode; |
160 | return 0; | |
161 | } | |
162 | ||
7615209f AN |
163 | __u32 libbpf_major_version(void) |
164 | { | |
165 | return LIBBPF_MAJOR_VERSION; | |
166 | } | |
167 | ||
168 | __u32 libbpf_minor_version(void) | |
169 | { | |
170 | return LIBBPF_MINOR_VERSION; | |
171 | } | |
172 | ||
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 | ||
166750bc AN |
182 | enum reloc_type { |
183 | RELO_LD64, | |
184 | RELO_CALL, | |
185 | RELO_DATA, | |
0c091e5c | 186 | RELO_EXTERN_VAR, |
5bd022ec | 187 | RELO_EXTERN_FUNC, |
53eddb5e | 188 | RELO_SUBPROG_ADDR, |
d0e92887 | 189 | RELO_CORE, |
166750bc AN |
190 | }; |
191 | ||
192 | struct reloc_desc { | |
193 | enum reloc_type type; | |
194 | int insn_idx; | |
d0e92887 AS |
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 | }; | |
166750bc AN |
202 | }; |
203 | ||
15ea31fa AN |
204 | /* stored as sec_def->cookie for all libbpf-supported SEC()s */ |
205 | enum sec_def_flags { | |
206 | SEC_NONE = 0, | |
207 | /* expected_attach_type is optional, if kernel doesn't support that */ | |
208 | SEC_EXP_ATTACH_OPT = 1, | |
209 | /* legacy, only used by libbpf_get_type_names() and | |
210 | * libbpf_attach_type_by_name(), not used by libbpf itself at all. | |
211 | * This used to be associated with cgroup (and few other) BPF programs | |
212 | * that were attachable through BPF_PROG_ATTACH command. Pretty | |
213 | * meaningless nowadays, though. | |
214 | */ | |
215 | SEC_ATTACHABLE = 2, | |
216 | SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT, | |
217 | /* attachment target is specified through BTF ID in either kernel or | |
218 | * other BPF program's BTF object */ | |
219 | SEC_ATTACH_BTF = 4, | |
220 | /* BPF program type allows sleeping/blocking in kernel */ | |
221 | SEC_SLEEPABLE = 8, | |
dd94d45c AN |
222 | /* allow non-strict prefix matching */ |
223 | SEC_SLOPPY_PFX = 16, | |
082c4bfb LB |
224 | /* BPF program support non-linear XDP buffer */ |
225 | SEC_XDP_FRAGS = 32, | |
4a4d4cee LB |
226 | /* deprecated sec definitions not supposed to be used */ |
227 | SEC_DEPRECATED = 64, | |
15ea31fa AN |
228 | }; |
229 | ||
25498a19 | 230 | struct bpf_sec_def { |
697f104d | 231 | char *sec; |
25498a19 AN |
232 | enum bpf_prog_type prog_type; |
233 | enum bpf_attach_type expected_attach_type; | |
15ea31fa | 234 | long cookie; |
697f104d | 235 | int handler_id; |
12d9466d | 236 | |
4fa5bcfe AN |
237 | libbpf_prog_setup_fn_t prog_setup_fn; |
238 | libbpf_prog_prepare_load_fn_t prog_prepare_load_fn; | |
239 | libbpf_prog_attach_fn_t prog_attach_fn; | |
25498a19 AN |
240 | }; |
241 | ||
a5b8bd47 WN |
242 | /* |
243 | * bpf_prog should be a better name but it has been used in | |
244 | * linux/filter.h. | |
245 | */ | |
246 | struct bpf_program { | |
25498a19 | 247 | const struct bpf_sec_def *sec_def; |
52109584 | 248 | char *sec_name; |
c1122392 AN |
249 | size_t sec_idx; |
250 | /* this program's instruction offset (in number of instructions) | |
251 | * within its containing ELF section | |
252 | */ | |
253 | size_t sec_insn_off; | |
254 | /* number of original instructions in ELF section belonging to this | |
255 | * program, not taking into account subprogram instructions possible | |
256 | * appended later during relocation | |
257 | */ | |
258 | size_t sec_insn_cnt; | |
259 | /* Offset (in number of instructions) of the start of instruction | |
260 | * belonging to this BPF program within its containing main BPF | |
261 | * program. For the entry-point (main) BPF program, this is always | |
262 | * zero. For a sub-program, this gets reset before each of main BPF | |
263 | * programs are processed and relocated and is used to determined | |
264 | * whether sub-program was already appended to the main program, and | |
265 | * if yes, at which instruction offset. | |
266 | */ | |
267 | size_t sub_insn_off; | |
268 | ||
269 | char *name; | |
a77f879b | 270 | /* name with / replaced by _; makes recursive pinning |
33a2c75c SF |
271 | * in bpf_object__pin_programs easier |
272 | */ | |
273 | char *pin_name; | |
c1122392 AN |
274 | |
275 | /* instructions that belong to BPF program; insns[0] is located at | |
276 | * sec_insn_off instruction within its ELF section in ELF file, so | |
277 | * when mapping ELF file instruction index to the local instruction, | |
278 | * one needs to subtract sec_insn_off; and vice versa. | |
279 | */ | |
a5b8bd47 | 280 | struct bpf_insn *insns; |
c1122392 AN |
281 | /* actual number of instruction in this BPF program's image; for |
282 | * entry-point BPF programs this includes the size of main program | |
283 | * itself plus all the used sub-programs, appended at the end | |
284 | */ | |
c3c55696 | 285 | size_t insns_cnt; |
34090915 | 286 | |
166750bc | 287 | struct reloc_desc *reloc_desc; |
34090915 | 288 | int nr_reloc; |
b3ce9079 AN |
289 | |
290 | /* BPF verifier log settings */ | |
291 | char *log_buf; | |
292 | size_t log_size; | |
293 | __u32 log_level; | |
55cffde2 | 294 | |
b580563e WN |
295 | struct { |
296 | int nr; | |
297 | int *fds; | |
298 | } instances; | |
299 | bpf_program_prep_t preprocessor; | |
aa9b1ac3 WN |
300 | |
301 | struct bpf_object *obj; | |
302 | void *priv; | |
303 | bpf_program_clear_priv_t clear_priv; | |
d7be143b | 304 | |
a3820c48 | 305 | bool autoload; |
aea28a60 | 306 | bool mark_btf_static; |
c1122392 | 307 | enum bpf_prog_type type; |
d7be143b | 308 | enum bpf_attach_type expected_attach_type; |
c1122392 | 309 | int prog_ifindex; |
91abb4a6 | 310 | __u32 attach_btf_obj_fd; |
12a8654b | 311 | __u32 attach_btf_id; |
e7bf94db | 312 | __u32 attach_prog_fd; |
2993e051 YS |
313 | void *func_info; |
314 | __u32 func_info_rec_size; | |
f0187f0b | 315 | __u32 func_info_cnt; |
47eff617 | 316 | |
3d650141 MKL |
317 | void *line_info; |
318 | __u32 line_info_rec_size; | |
319 | __u32 line_info_cnt; | |
04656198 | 320 | __u32 prog_flags; |
a5b8bd47 WN |
321 | }; |
322 | ||
590a0088 MKL |
323 | struct bpf_struct_ops { |
324 | const char *tname; | |
325 | const struct btf_type *type; | |
326 | struct bpf_program **progs; | |
327 | __u32 *kern_func_off; | |
328 | /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */ | |
329 | void *data; | |
330 | /* e.g. struct bpf_struct_ops_tcp_congestion_ops in | |
331 | * btf_vmlinux's format. | |
332 | * struct bpf_struct_ops_tcp_congestion_ops { | |
333 | * [... some other kernel fields ...] | |
334 | * struct tcp_congestion_ops data; | |
335 | * } | |
336 | * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops) | |
337 | * bpf_map__init_kern_struct_ops() will populate the "kern_vdata" | |
338 | * from "data". | |
339 | */ | |
340 | void *kern_vdata; | |
341 | __u32 type_id; | |
342 | }; | |
343 | ||
ac9d1389 AN |
344 | #define DATA_SEC ".data" |
345 | #define BSS_SEC ".bss" | |
346 | #define RODATA_SEC ".rodata" | |
81bfdd08 | 347 | #define KCONFIG_SEC ".kconfig" |
1c0c7074 | 348 | #define KSYMS_SEC ".ksyms" |
590a0088 | 349 | #define STRUCT_OPS_SEC ".struct_ops" |
ac9d1389 | 350 | |
d859900c DB |
351 | enum libbpf_map_type { |
352 | LIBBPF_MAP_UNSPEC, | |
353 | LIBBPF_MAP_DATA, | |
354 | LIBBPF_MAP_BSS, | |
355 | LIBBPF_MAP_RODATA, | |
81bfdd08 | 356 | LIBBPF_MAP_KCONFIG, |
d859900c DB |
357 | }; |
358 | ||
9d759a9b | 359 | struct bpf_map { |
561bbcca | 360 | char *name; |
aed65917 AN |
361 | /* real_name is defined for special internal maps (.rodata*, |
362 | * .data*, .bss, .kconfig) and preserves their original ELF section | |
363 | * name. This is important to be be able to find corresponding BTF | |
364 | * DATASEC information. | |
365 | */ | |
366 | char *real_name; | |
01af3bf0 | 367 | int fd; |
db48814b AN |
368 | int sec_idx; |
369 | size_t sec_offset; | |
f0307a7e | 370 | int map_ifindex; |
addb9fc9 | 371 | int inner_map_fd; |
9d759a9b | 372 | struct bpf_map_def def; |
1bdb6c9a | 373 | __u32 numa_node; |
646f02ff | 374 | __u32 btf_var_idx; |
5b891af7 MKL |
375 | __u32 btf_key_type_id; |
376 | __u32 btf_value_type_id; | |
590a0088 | 377 | __u32 btf_vmlinux_value_type_id; |
9d759a9b WN |
378 | void *priv; |
379 | bpf_map_clear_priv_t clear_priv; | |
d859900c | 380 | enum libbpf_map_type libbpf_type; |
eba9c5f4 | 381 | void *mmaped; |
590a0088 | 382 | struct bpf_struct_ops *st_ops; |
646f02ff AN |
383 | struct bpf_map *inner_map; |
384 | void **init_slots; | |
385 | int init_slots_sz; | |
4580b25f THJ |
386 | char *pin_path; |
387 | bool pinned; | |
ec6d5f47 | 388 | bool reused; |
229fae38 | 389 | bool skipped; |
47512102 | 390 | __u64 map_extra; |
d859900c DB |
391 | }; |
392 | ||
166750bc AN |
393 | enum extern_type { |
394 | EXT_UNKNOWN, | |
2e33efe3 | 395 | EXT_KCFG, |
1c0c7074 | 396 | EXT_KSYM, |
2e33efe3 AN |
397 | }; |
398 | ||
399 | enum kcfg_type { | |
400 | KCFG_UNKNOWN, | |
401 | KCFG_CHAR, | |
402 | KCFG_BOOL, | |
403 | KCFG_INT, | |
404 | KCFG_TRISTATE, | |
405 | KCFG_CHAR_ARR, | |
166750bc AN |
406 | }; |
407 | ||
408 | struct extern_desc { | |
2e33efe3 | 409 | enum extern_type type; |
166750bc AN |
410 | int sym_idx; |
411 | int btf_id; | |
2e33efe3 AN |
412 | int sec_btf_id; |
413 | const char *name; | |
166750bc | 414 | bool is_set; |
2e33efe3 AN |
415 | bool is_weak; |
416 | union { | |
417 | struct { | |
418 | enum kcfg_type type; | |
419 | int sz; | |
420 | int align; | |
421 | int data_off; | |
422 | bool is_signed; | |
423 | } kcfg; | |
1c0c7074 AN |
424 | struct { |
425 | unsigned long long addr; | |
d370bbe1 HL |
426 | |
427 | /* target btf_id of the corresponding kernel var. */ | |
284d2587 AN |
428 | int kernel_btf_obj_fd; |
429 | int kernel_btf_id; | |
d370bbe1 HL |
430 | |
431 | /* local btf_id of the ksym extern's type. */ | |
432 | __u32 type_id; | |
9dbe6015 KKD |
433 | /* BTF fd index to be patched in for insn->off, this is |
434 | * 0 for vmlinux BTF, index in obj->fd_array for module | |
435 | * BTF | |
436 | */ | |
437 | __s16 btf_fd_idx; | |
1c0c7074 | 438 | } ksym; |
2e33efe3 | 439 | }; |
166750bc AN |
440 | }; |
441 | ||
9a208eff WN |
442 | static LIST_HEAD(bpf_objects_list); |
443 | ||
4f33a53d AN |
444 | struct module_btf { |
445 | struct btf *btf; | |
446 | char *name; | |
447 | __u32 id; | |
91abb4a6 | 448 | int fd; |
9dbe6015 | 449 | int fd_array_idx; |
4f33a53d AN |
450 | }; |
451 | ||
25bbbd7a AN |
452 | enum sec_type { |
453 | SEC_UNUSED = 0, | |
454 | SEC_RELO, | |
455 | SEC_BSS, | |
456 | SEC_DATA, | |
457 | SEC_RODATA, | |
458 | }; | |
459 | ||
460 | struct elf_sec_desc { | |
461 | enum sec_type sec_type; | |
462 | Elf64_Shdr *shdr; | |
463 | Elf_Data *data; | |
464 | }; | |
465 | ||
29a30ff5 AN |
466 | struct elf_state { |
467 | int fd; | |
468 | const void *obj_buf; | |
469 | size_t obj_buf_sz; | |
470 | Elf *elf; | |
ad23b723 | 471 | Elf64_Ehdr *ehdr; |
29a30ff5 | 472 | Elf_Data *symbols; |
29a30ff5 AN |
473 | Elf_Data *st_ops_data; |
474 | size_t shstrndx; /* section index for section name strings */ | |
475 | size_t strtabidx; | |
25bbbd7a AN |
476 | struct elf_sec_desc *secs; |
477 | int sec_cnt; | |
29a30ff5 AN |
478 | int maps_shndx; |
479 | int btf_maps_shndx; | |
480 | __u32 btf_maps_sec_btf_id; | |
481 | int text_shndx; | |
482 | int symbols_shndx; | |
29a30ff5 | 483 | int st_ops_shndx; |
4f33a53d AN |
484 | }; |
485 | ||
2e4913e0 AN |
486 | struct usdt_manager; |
487 | ||
1a5e3fb1 | 488 | struct bpf_object { |
d859900c | 489 | char name[BPF_OBJ_NAME_LEN]; |
cb1e5e96 | 490 | char license[64]; |
438363c0 | 491 | __u32 kern_version; |
0b3d1efa | 492 | |
a5b8bd47 WN |
493 | struct bpf_program *programs; |
494 | size_t nr_programs; | |
9d759a9b WN |
495 | struct bpf_map *maps; |
496 | size_t nr_maps; | |
bf829271 | 497 | size_t maps_cap; |
9d759a9b | 498 | |
8601fd42 | 499 | char *kconfig; |
166750bc AN |
500 | struct extern_desc *externs; |
501 | int nr_extern; | |
81bfdd08 | 502 | int kconfig_map_idx; |
166750bc | 503 | |
52d3352e | 504 | bool loaded; |
c3c55696 | 505 | bool has_subcalls; |
25bbbd7a | 506 | bool has_rodata; |
a5b8bd47 | 507 | |
e2fa0156 AS |
508 | struct bpf_gen *gen_loader; |
509 | ||
29a30ff5 AN |
510 | /* Information when doing ELF related work. Only valid if efile.elf is not NULL */ |
511 | struct elf_state efile; | |
1a5e3fb1 | 512 | /* |
29a30ff5 | 513 | * All loaded bpf_object are linked in a list, which is |
9a208eff WN |
514 | * hidden to caller. bpf_objects__<func> handlers deal with |
515 | * all objects. | |
516 | */ | |
517 | struct list_head list; | |
10931d24 | 518 | |
8a138aed | 519 | struct btf *btf; |
0f7515ca AN |
520 | struct btf_ext *btf_ext; |
521 | ||
a6ed02ca KS |
522 | /* Parse and load BTF vmlinux if any of the programs in the object need |
523 | * it at load time. | |
524 | */ | |
525 | struct btf *btf_vmlinux; | |
1373ff59 SC |
526 | /* Path to the custom BTF to be used for BPF CO-RE relocations as an |
527 | * override for vmlinux BTF. | |
528 | */ | |
529 | char *btf_custom_path; | |
0f7515ca AN |
530 | /* vmlinux BTF override for CO-RE relocations */ |
531 | struct btf *btf_vmlinux_override; | |
4f33a53d AN |
532 | /* Lazily initialized kernel module BTFs */ |
533 | struct module_btf *btf_modules; | |
534 | bool btf_modules_loaded; | |
535 | size_t btf_module_cnt; | |
536 | size_t btf_module_cap; | |
8a138aed | 537 | |
e0e3ea88 AN |
538 | /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */ |
539 | char *log_buf; | |
540 | size_t log_size; | |
541 | __u32 log_level; | |
542 | ||
10931d24 WN |
543 | void *priv; |
544 | bpf_object_clear_priv_t clear_priv; | |
545 | ||
9dbe6015 KKD |
546 | int *fd_array; |
547 | size_t fd_array_cap; | |
548 | size_t fd_array_cnt; | |
549 | ||
2e4913e0 AN |
550 | struct usdt_manager *usdt_man; |
551 | ||
1a5e3fb1 WN |
552 | char path[]; |
553 | }; | |
1a5e3fb1 | 554 | |
88a82120 AN |
555 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off); |
556 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off); | |
557 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx); | |
558 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name); | |
ad23b723 | 559 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn); |
88a82120 AN |
560 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn); |
561 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn); | |
ad23b723 AN |
562 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx); |
563 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx); | |
88a82120 | 564 | |
29cd77f4 | 565 | void bpf_program__unload(struct bpf_program *prog) |
55cffde2 | 566 | { |
b580563e WN |
567 | int i; |
568 | ||
55cffde2 WN |
569 | if (!prog) |
570 | return; | |
571 | ||
b580563e WN |
572 | /* |
573 | * If the object is opened but the program was never loaded, | |
574 | * it is possible that prog->instances.nr == -1. | |
575 | */ | |
576 | if (prog->instances.nr > 0) { | |
577 | for (i = 0; i < prog->instances.nr; i++) | |
578 | zclose(prog->instances.fds[i]); | |
579 | } else if (prog->instances.nr != -1) { | |
be18010e KW |
580 | pr_warn("Internal error: instances.nr is %d\n", |
581 | prog->instances.nr); | |
b580563e WN |
582 | } |
583 | ||
584 | prog->instances.nr = -1; | |
585 | zfree(&prog->instances.fds); | |
2993e051 | 586 | |
2993e051 | 587 | zfree(&prog->func_info); |
07a09d1b | 588 | zfree(&prog->line_info); |
55cffde2 WN |
589 | } |
590 | ||
a5b8bd47 WN |
591 | static void bpf_program__exit(struct bpf_program *prog) |
592 | { | |
593 | if (!prog) | |
594 | return; | |
595 | ||
aa9b1ac3 WN |
596 | if (prog->clear_priv) |
597 | prog->clear_priv(prog, prog->priv); | |
598 | ||
599 | prog->priv = NULL; | |
600 | prog->clear_priv = NULL; | |
601 | ||
55cffde2 | 602 | bpf_program__unload(prog); |
88cda1c9 | 603 | zfree(&prog->name); |
52109584 | 604 | zfree(&prog->sec_name); |
33a2c75c | 605 | zfree(&prog->pin_name); |
a5b8bd47 | 606 | zfree(&prog->insns); |
34090915 WN |
607 | zfree(&prog->reloc_desc); |
608 | ||
609 | prog->nr_reloc = 0; | |
a5b8bd47 | 610 | prog->insns_cnt = 0; |
c1122392 | 611 | prog->sec_idx = -1; |
a5b8bd47 WN |
612 | } |
613 | ||
33a2c75c SF |
614 | static char *__bpf_program__pin_name(struct bpf_program *prog) |
615 | { | |
616 | char *name, *p; | |
617 | ||
a77f879b SF |
618 | if (libbpf_mode & LIBBPF_STRICT_SEC_NAME) |
619 | name = strdup(prog->name); | |
620 | else | |
621 | name = strdup(prog->sec_name); | |
622 | ||
623 | if (!name) | |
624 | return NULL; | |
625 | ||
626 | p = name; | |
627 | ||
33a2c75c SF |
628 | while ((p = strchr(p, '/'))) |
629 | *p = '_'; | |
630 | ||
631 | return name; | |
632 | } | |
633 | ||
c3c55696 AN |
634 | static bool insn_is_subprog_call(const struct bpf_insn *insn) |
635 | { | |
636 | return BPF_CLASS(insn->code) == BPF_JMP && | |
637 | BPF_OP(insn->code) == BPF_CALL && | |
638 | BPF_SRC(insn->code) == BPF_K && | |
639 | insn->src_reg == BPF_PSEUDO_CALL && | |
640 | insn->dst_reg == 0 && | |
641 | insn->off == 0; | |
642 | } | |
643 | ||
aa0b8d43 MKL |
644 | static bool is_call_insn(const struct bpf_insn *insn) |
645 | { | |
646 | return insn->code == (BPF_JMP | BPF_CALL); | |
647 | } | |
648 | ||
53eddb5e YS |
649 | static bool insn_is_pseudo_func(struct bpf_insn *insn) |
650 | { | |
aa0b8d43 | 651 | return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC; |
53eddb5e YS |
652 | } |
653 | ||
a5b8bd47 | 654 | static int |
c3c55696 AN |
655 | bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog, |
656 | const char *name, size_t sec_idx, const char *sec_name, | |
657 | size_t sec_off, void *insn_data, size_t insn_data_sz) | |
a5b8bd47 | 658 | { |
c1122392 AN |
659 | if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) { |
660 | pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n", | |
661 | sec_name, name, sec_off, insn_data_sz); | |
a5b8bd47 WN |
662 | return -EINVAL; |
663 | } | |
664 | ||
1ad9cbb8 | 665 | memset(prog, 0, sizeof(*prog)); |
c3c55696 AN |
666 | prog->obj = obj; |
667 | ||
c1122392 AN |
668 | prog->sec_idx = sec_idx; |
669 | prog->sec_insn_off = sec_off / BPF_INSN_SZ; | |
670 | prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ; | |
671 | /* insns_cnt can later be increased by appending used subprograms */ | |
672 | prog->insns_cnt = prog->sec_insn_cnt; | |
a5b8bd47 | 673 | |
c1122392 | 674 | prog->type = BPF_PROG_TYPE_UNSPEC; |
a3820c48 AN |
675 | |
676 | /* libbpf's convention for SEC("?abc...") is that it's just like | |
677 | * SEC("abc...") but the corresponding bpf_program starts out with | |
678 | * autoload set to false. | |
679 | */ | |
680 | if (sec_name[0] == '?') { | |
681 | prog->autoload = false; | |
682 | /* from now on forget there was ? in section name */ | |
683 | sec_name++; | |
684 | } else { | |
685 | prog->autoload = true; | |
686 | } | |
a5b8bd47 | 687 | |
c1122392 AN |
688 | prog->instances.fds = NULL; |
689 | prog->instances.nr = -1; | |
690 | ||
b3ce9079 AN |
691 | /* inherit object's log_level */ |
692 | prog->log_level = obj->log_level; | |
693 | ||
52109584 AN |
694 | prog->sec_name = strdup(sec_name); |
695 | if (!prog->sec_name) | |
c1122392 AN |
696 | goto errout; |
697 | ||
698 | prog->name = strdup(name); | |
699 | if (!prog->name) | |
a5b8bd47 | 700 | goto errout; |
a5b8bd47 | 701 | |
33a2c75c | 702 | prog->pin_name = __bpf_program__pin_name(prog); |
c1122392 | 703 | if (!prog->pin_name) |
33a2c75c | 704 | goto errout; |
33a2c75c | 705 | |
c1122392 AN |
706 | prog->insns = malloc(insn_data_sz); |
707 | if (!prog->insns) | |
a5b8bd47 | 708 | goto errout; |
c1122392 | 709 | memcpy(prog->insns, insn_data, insn_data_sz); |
a5b8bd47 WN |
710 | |
711 | return 0; | |
712 | errout: | |
c1122392 | 713 | pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name); |
a5b8bd47 WN |
714 | bpf_program__exit(prog); |
715 | return -ENOMEM; | |
716 | } | |
717 | ||
718 | static int | |
c1122392 AN |
719 | bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data, |
720 | const char *sec_name, int sec_idx) | |
a5b8bd47 | 721 | { |
6245947c | 722 | Elf_Data *symbols = obj->efile.symbols; |
c1122392 AN |
723 | struct bpf_program *prog, *progs; |
724 | void *data = sec_data->d_buf; | |
6245947c AN |
725 | size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms; |
726 | int nr_progs, err, i; | |
c1122392 | 727 | const char *name; |
ad23b723 | 728 | Elf64_Sym *sym; |
a5b8bd47 WN |
729 | |
730 | progs = obj->programs; | |
731 | nr_progs = obj->nr_programs; | |
ad23b723 | 732 | nr_syms = symbols->d_size / sizeof(Elf64_Sym); |
c1122392 | 733 | sec_off = 0; |
a5b8bd47 | 734 | |
6245947c | 735 | for (i = 0; i < nr_syms; i++) { |
ad23b723 AN |
736 | sym = elf_sym_by_idx(obj, i); |
737 | ||
738 | if (sym->st_shndx != sec_idx) | |
6245947c | 739 | continue; |
ad23b723 | 740 | if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC) |
6245947c | 741 | continue; |
88cda1c9 | 742 | |
ad23b723 AN |
743 | prog_sz = sym->st_size; |
744 | sec_off = sym->st_value; | |
88cda1c9 | 745 | |
ad23b723 | 746 | name = elf_sym_str(obj, sym->st_name); |
c1122392 AN |
747 | if (!name) { |
748 | pr_warn("sec '%s': failed to get symbol name for offset %zu\n", | |
749 | sec_name, sec_off); | |
750 | return -LIBBPF_ERRNO__FORMAT; | |
751 | } | |
88cda1c9 | 752 | |
c1122392 AN |
753 | if (sec_off + prog_sz > sec_sz) { |
754 | pr_warn("sec '%s': program at offset %zu crosses section boundary\n", | |
755 | sec_name, sec_off); | |
756 | return -LIBBPF_ERRNO__FORMAT; | |
757 | } | |
88cda1c9 | 758 | |
ad23b723 | 759 | if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) { |
513f485c AN |
760 | pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name); |
761 | return -ENOTSUP; | |
762 | } | |
763 | ||
c3c55696 AN |
764 | pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n", |
765 | sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz); | |
88cda1c9 | 766 | |
c3c55696 | 767 | progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs)); |
c1122392 AN |
768 | if (!progs) { |
769 | /* | |
770 | * In this case the original obj->programs | |
771 | * is still valid, so don't need special treat for | |
772 | * bpf_close_object(). | |
773 | */ | |
774 | pr_warn("sec '%s': failed to alloc memory for new program '%s'\n", | |
775 | sec_name, name); | |
776 | return -ENOMEM; | |
88cda1c9 | 777 | } |
c1122392 | 778 | obj->programs = progs; |
88cda1c9 | 779 | |
c1122392 | 780 | prog = &progs[nr_progs]; |
9a94f277 | 781 | |
c3c55696 AN |
782 | err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name, |
783 | sec_off, data + sec_off, prog_sz); | |
c1122392 AN |
784 | if (err) |
785 | return err; | |
9a94f277 | 786 | |
e5670fa0 AN |
787 | /* if function is a global/weak symbol, but has restricted |
788 | * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC | |
789 | * as static to enable more permissive BPF verification mode | |
790 | * with more outside context available to BPF verifier | |
aea28a60 | 791 | */ |
ad23b723 AN |
792 | if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL |
793 | && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN | |
794 | || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL)) | |
aea28a60 AN |
795 | prog->mark_btf_static = true; |
796 | ||
c1122392 AN |
797 | nr_progs++; |
798 | obj->nr_programs = nr_progs; | |
88cda1c9 MKL |
799 | } |
800 | ||
801 | return 0; | |
802 | } | |
803 | ||
5b3d7298 | 804 | __u32 get_kernel_version(void) |
5e61f270 | 805 | { |
5b3d7298 AN |
806 | /* On Ubuntu LINUX_VERSION_CODE doesn't correspond to info.release, |
807 | * but Ubuntu provides /proc/version_signature file, as described at | |
808 | * https://ubuntu.com/kernel, with an example contents below, which we | |
809 | * can use to get a proper LINUX_VERSION_CODE. | |
810 | * | |
811 | * Ubuntu 5.4.0-12.15-generic 5.4.8 | |
812 | * | |
813 | * In the above, 5.4.8 is what kernel is actually expecting, while | |
814 | * uname() call will return 5.4.0 in info.release. | |
815 | */ | |
816 | const char *ubuntu_kver_file = "/proc/version_signature"; | |
5e61f270 AN |
817 | __u32 major, minor, patch; |
818 | struct utsname info; | |
819 | ||
5b3d7298 AN |
820 | if (access(ubuntu_kver_file, R_OK) == 0) { |
821 | FILE *f; | |
822 | ||
823 | f = fopen(ubuntu_kver_file, "r"); | |
824 | if (f) { | |
825 | if (fscanf(f, "%*s %*s %d.%d.%d\n", &major, &minor, &patch) == 3) { | |
826 | fclose(f); | |
827 | return KERNEL_VERSION(major, minor, patch); | |
828 | } | |
829 | fclose(f); | |
830 | } | |
831 | /* something went wrong, fall back to uname() approach */ | |
832 | } | |
833 | ||
5e61f270 AN |
834 | uname(&info); |
835 | if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3) | |
836 | return 0; | |
837 | return KERNEL_VERSION(major, minor, patch); | |
838 | } | |
839 | ||
590a0088 MKL |
840 | static const struct btf_member * |
841 | find_member_by_offset(const struct btf_type *t, __u32 bit_offset) | |
842 | { | |
843 | struct btf_member *m; | |
844 | int i; | |
845 | ||
846 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
847 | if (btf_member_bit_offset(t, i) == bit_offset) | |
848 | return m; | |
849 | } | |
850 | ||
851 | return NULL; | |
852 | } | |
853 | ||
854 | static const struct btf_member * | |
855 | find_member_by_name(const struct btf *btf, const struct btf_type *t, | |
856 | const char *name) | |
857 | { | |
858 | struct btf_member *m; | |
859 | int i; | |
860 | ||
861 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
862 | if (!strcmp(btf__name_by_offset(btf, m->name_off), name)) | |
863 | return m; | |
864 | } | |
865 | ||
866 | return NULL; | |
867 | } | |
868 | ||
869 | #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_" | |
a6ed02ca KS |
870 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
871 | const char *name, __u32 kind); | |
590a0088 MKL |
872 | |
873 | static int | |
874 | find_struct_ops_kern_types(const struct btf *btf, const char *tname, | |
875 | const struct btf_type **type, __u32 *type_id, | |
876 | const struct btf_type **vtype, __u32 *vtype_id, | |
877 | const struct btf_member **data_member) | |
878 | { | |
879 | const struct btf_type *kern_type, *kern_vtype; | |
880 | const struct btf_member *kern_data_member; | |
881 | __s32 kern_vtype_id, kern_type_id; | |
590a0088 MKL |
882 | __u32 i; |
883 | ||
884 | kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT); | |
885 | if (kern_type_id < 0) { | |
886 | pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n", | |
887 | tname); | |
888 | return kern_type_id; | |
889 | } | |
890 | kern_type = btf__type_by_id(btf, kern_type_id); | |
891 | ||
892 | /* Find the corresponding "map_value" type that will be used | |
893 | * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example, | |
894 | * find "struct bpf_struct_ops_tcp_congestion_ops" from the | |
895 | * btf_vmlinux. | |
896 | */ | |
a6ed02ca KS |
897 | kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX, |
898 | tname, BTF_KIND_STRUCT); | |
590a0088 | 899 | if (kern_vtype_id < 0) { |
a6ed02ca KS |
900 | pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n", |
901 | STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
902 | return kern_vtype_id; |
903 | } | |
904 | kern_vtype = btf__type_by_id(btf, kern_vtype_id); | |
905 | ||
906 | /* Find "struct tcp_congestion_ops" from | |
907 | * struct bpf_struct_ops_tcp_congestion_ops { | |
908 | * [ ... ] | |
909 | * struct tcp_congestion_ops data; | |
910 | * } | |
911 | */ | |
912 | kern_data_member = btf_members(kern_vtype); | |
913 | for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) { | |
914 | if (kern_data_member->type == kern_type_id) | |
915 | break; | |
916 | } | |
917 | if (i == btf_vlen(kern_vtype)) { | |
a6ed02ca KS |
918 | pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n", |
919 | tname, STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
920 | return -EINVAL; |
921 | } | |
922 | ||
923 | *type = kern_type; | |
924 | *type_id = kern_type_id; | |
925 | *vtype = kern_vtype; | |
926 | *vtype_id = kern_vtype_id; | |
927 | *data_member = kern_data_member; | |
928 | ||
929 | return 0; | |
930 | } | |
931 | ||
932 | static bool bpf_map__is_struct_ops(const struct bpf_map *map) | |
933 | { | |
934 | return map->def.type == BPF_MAP_TYPE_STRUCT_OPS; | |
935 | } | |
936 | ||
937 | /* Init the map's fields that depend on kern_btf */ | |
938 | static int bpf_map__init_kern_struct_ops(struct bpf_map *map, | |
939 | const struct btf *btf, | |
940 | const struct btf *kern_btf) | |
941 | { | |
942 | const struct btf_member *member, *kern_member, *kern_data_member; | |
943 | const struct btf_type *type, *kern_type, *kern_vtype; | |
944 | __u32 i, kern_type_id, kern_vtype_id, kern_data_off; | |
945 | struct bpf_struct_ops *st_ops; | |
946 | void *data, *kern_data; | |
947 | const char *tname; | |
948 | int err; | |
949 | ||
950 | st_ops = map->st_ops; | |
951 | type = st_ops->type; | |
952 | tname = st_ops->tname; | |
953 | err = find_struct_ops_kern_types(kern_btf, tname, | |
954 | &kern_type, &kern_type_id, | |
955 | &kern_vtype, &kern_vtype_id, | |
956 | &kern_data_member); | |
957 | if (err) | |
958 | return err; | |
959 | ||
960 | pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n", | |
961 | map->name, st_ops->type_id, kern_type_id, kern_vtype_id); | |
962 | ||
963 | map->def.value_size = kern_vtype->size; | |
964 | map->btf_vmlinux_value_type_id = kern_vtype_id; | |
965 | ||
966 | st_ops->kern_vdata = calloc(1, kern_vtype->size); | |
967 | if (!st_ops->kern_vdata) | |
968 | return -ENOMEM; | |
969 | ||
970 | data = st_ops->data; | |
971 | kern_data_off = kern_data_member->offset / 8; | |
972 | kern_data = st_ops->kern_vdata + kern_data_off; | |
973 | ||
974 | member = btf_members(type); | |
975 | for (i = 0; i < btf_vlen(type); i++, member++) { | |
976 | const struct btf_type *mtype, *kern_mtype; | |
977 | __u32 mtype_id, kern_mtype_id; | |
978 | void *mdata, *kern_mdata; | |
979 | __s64 msize, kern_msize; | |
980 | __u32 moff, kern_moff; | |
981 | __u32 kern_member_idx; | |
982 | const char *mname; | |
983 | ||
984 | mname = btf__name_by_offset(btf, member->name_off); | |
985 | kern_member = find_member_by_name(kern_btf, kern_type, mname); | |
986 | if (!kern_member) { | |
987 | pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n", | |
988 | map->name, mname); | |
989 | return -ENOTSUP; | |
990 | } | |
991 | ||
992 | kern_member_idx = kern_member - btf_members(kern_type); | |
993 | if (btf_member_bitfield_size(type, i) || | |
994 | btf_member_bitfield_size(kern_type, kern_member_idx)) { | |
995 | pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n", | |
996 | map->name, mname); | |
997 | return -ENOTSUP; | |
998 | } | |
999 | ||
1000 | moff = member->offset / 8; | |
1001 | kern_moff = kern_member->offset / 8; | |
1002 | ||
1003 | mdata = data + moff; | |
1004 | kern_mdata = kern_data + kern_moff; | |
1005 | ||
1006 | mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id); | |
1007 | kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type, | |
1008 | &kern_mtype_id); | |
1009 | if (BTF_INFO_KIND(mtype->info) != | |
1010 | BTF_INFO_KIND(kern_mtype->info)) { | |
1011 | pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n", | |
1012 | map->name, mname, BTF_INFO_KIND(mtype->info), | |
1013 | BTF_INFO_KIND(kern_mtype->info)); | |
1014 | return -ENOTSUP; | |
1015 | } | |
1016 | ||
1017 | if (btf_is_ptr(mtype)) { | |
1018 | struct bpf_program *prog; | |
1019 | ||
d2836ddd MKL |
1020 | prog = st_ops->progs[i]; |
1021 | if (!prog) | |
1022 | continue; | |
1023 | ||
590a0088 MKL |
1024 | kern_mtype = skip_mods_and_typedefs(kern_btf, |
1025 | kern_mtype->type, | |
1026 | &kern_mtype_id); | |
d2836ddd MKL |
1027 | |
1028 | /* mtype->type must be a func_proto which was | |
1029 | * guaranteed in bpf_object__collect_st_ops_relos(), | |
1030 | * so only check kern_mtype for func_proto here. | |
1031 | */ | |
1032 | if (!btf_is_func_proto(kern_mtype)) { | |
1033 | pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n", | |
590a0088 MKL |
1034 | map->name, mname); |
1035 | return -ENOTSUP; | |
1036 | } | |
1037 | ||
590a0088 MKL |
1038 | prog->attach_btf_id = kern_type_id; |
1039 | prog->expected_attach_type = kern_member_idx; | |
1040 | ||
1041 | st_ops->kern_func_off[i] = kern_data_off + kern_moff; | |
1042 | ||
1043 | pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n", | |
1044 | map->name, mname, prog->name, moff, | |
1045 | kern_moff); | |
1046 | ||
1047 | continue; | |
1048 | } | |
1049 | ||
1050 | msize = btf__resolve_size(btf, mtype_id); | |
1051 | kern_msize = btf__resolve_size(kern_btf, kern_mtype_id); | |
1052 | if (msize < 0 || kern_msize < 0 || msize != kern_msize) { | |
1053 | pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n", | |
1054 | map->name, mname, (ssize_t)msize, | |
1055 | (ssize_t)kern_msize); | |
1056 | return -ENOTSUP; | |
1057 | } | |
1058 | ||
1059 | pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n", | |
1060 | map->name, mname, (unsigned int)msize, | |
1061 | moff, kern_moff); | |
1062 | memcpy(kern_mdata, mdata, msize); | |
1063 | } | |
1064 | ||
1065 | return 0; | |
1066 | } | |
1067 | ||
1068 | static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj) | |
1069 | { | |
590a0088 MKL |
1070 | struct bpf_map *map; |
1071 | size_t i; | |
1072 | int err; | |
1073 | ||
1074 | for (i = 0; i < obj->nr_maps; i++) { | |
1075 | map = &obj->maps[i]; | |
1076 | ||
1077 | if (!bpf_map__is_struct_ops(map)) | |
1078 | continue; | |
1079 | ||
a6ed02ca KS |
1080 | err = bpf_map__init_kern_struct_ops(map, obj->btf, |
1081 | obj->btf_vmlinux); | |
1082 | if (err) | |
590a0088 | 1083 | return err; |
590a0088 MKL |
1084 | } |
1085 | ||
590a0088 MKL |
1086 | return 0; |
1087 | } | |
1088 | ||
1089 | static int bpf_object__init_struct_ops_maps(struct bpf_object *obj) | |
1090 | { | |
1091 | const struct btf_type *type, *datasec; | |
1092 | const struct btf_var_secinfo *vsi; | |
1093 | struct bpf_struct_ops *st_ops; | |
1094 | const char *tname, *var_name; | |
1095 | __s32 type_id, datasec_id; | |
1096 | const struct btf *btf; | |
1097 | struct bpf_map *map; | |
1098 | __u32 i; | |
1099 | ||
1100 | if (obj->efile.st_ops_shndx == -1) | |
1101 | return 0; | |
1102 | ||
1103 | btf = obj->btf; | |
1104 | datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC, | |
1105 | BTF_KIND_DATASEC); | |
1106 | if (datasec_id < 0) { | |
1107 | pr_warn("struct_ops init: DATASEC %s not found\n", | |
1108 | STRUCT_OPS_SEC); | |
1109 | return -EINVAL; | |
1110 | } | |
1111 | ||
1112 | datasec = btf__type_by_id(btf, datasec_id); | |
1113 | vsi = btf_var_secinfos(datasec); | |
1114 | for (i = 0; i < btf_vlen(datasec); i++, vsi++) { | |
1115 | type = btf__type_by_id(obj->btf, vsi->type); | |
1116 | var_name = btf__name_by_offset(obj->btf, type->name_off); | |
1117 | ||
1118 | type_id = btf__resolve_type(obj->btf, vsi->type); | |
1119 | if (type_id < 0) { | |
1120 | pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n", | |
1121 | vsi->type, STRUCT_OPS_SEC); | |
1122 | return -EINVAL; | |
1123 | } | |
1124 | ||
1125 | type = btf__type_by_id(obj->btf, type_id); | |
1126 | tname = btf__name_by_offset(obj->btf, type->name_off); | |
1127 | if (!tname[0]) { | |
1128 | pr_warn("struct_ops init: anonymous type is not supported\n"); | |
1129 | return -ENOTSUP; | |
1130 | } | |
1131 | if (!btf_is_struct(type)) { | |
1132 | pr_warn("struct_ops init: %s is not a struct\n", tname); | |
1133 | return -EINVAL; | |
1134 | } | |
1135 | ||
1136 | map = bpf_object__add_map(obj); | |
1137 | if (IS_ERR(map)) | |
1138 | return PTR_ERR(map); | |
1139 | ||
1140 | map->sec_idx = obj->efile.st_ops_shndx; | |
1141 | map->sec_offset = vsi->offset; | |
1142 | map->name = strdup(var_name); | |
1143 | if (!map->name) | |
1144 | return -ENOMEM; | |
1145 | ||
1146 | map->def.type = BPF_MAP_TYPE_STRUCT_OPS; | |
1147 | map->def.key_size = sizeof(int); | |
1148 | map->def.value_size = type->size; | |
1149 | map->def.max_entries = 1; | |
1150 | ||
1151 | map->st_ops = calloc(1, sizeof(*map->st_ops)); | |
1152 | if (!map->st_ops) | |
1153 | return -ENOMEM; | |
1154 | st_ops = map->st_ops; | |
1155 | st_ops->data = malloc(type->size); | |
1156 | st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs)); | |
1157 | st_ops->kern_func_off = malloc(btf_vlen(type) * | |
1158 | sizeof(*st_ops->kern_func_off)); | |
1159 | if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off) | |
1160 | return -ENOMEM; | |
1161 | ||
1162 | if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) { | |
1163 | pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n", | |
1164 | var_name, STRUCT_OPS_SEC); | |
1165 | return -EINVAL; | |
1166 | } | |
1167 | ||
1168 | memcpy(st_ops->data, | |
1169 | obj->efile.st_ops_data->d_buf + vsi->offset, | |
1170 | type->size); | |
1171 | st_ops->tname = tname; | |
1172 | st_ops->type = type; | |
1173 | st_ops->type_id = type_id; | |
1174 | ||
1175 | pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n", | |
1176 | tname, type_id, var_name, vsi->offset); | |
1177 | } | |
1178 | ||
1179 | return 0; | |
1180 | } | |
1181 | ||
6c956392 | 1182 | static struct bpf_object *bpf_object__new(const char *path, |
5e61f270 | 1183 | const void *obj_buf, |
2ce8450e AN |
1184 | size_t obj_buf_sz, |
1185 | const char *obj_name) | |
1a5e3fb1 | 1186 | { |
689624f0 | 1187 | bool strict = (libbpf_mode & LIBBPF_STRICT_NO_OBJECT_LIST); |
1a5e3fb1 | 1188 | struct bpf_object *obj; |
d859900c | 1189 | char *end; |
1a5e3fb1 WN |
1190 | |
1191 | obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1); | |
1192 | if (!obj) { | |
be18010e | 1193 | pr_warn("alloc memory failed for %s\n", path); |
6371ca3b | 1194 | return ERR_PTR(-ENOMEM); |
1a5e3fb1 WN |
1195 | } |
1196 | ||
1197 | strcpy(obj->path, path); | |
2ce8450e | 1198 | if (obj_name) { |
9fc205b4 | 1199 | libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name)); |
2ce8450e AN |
1200 | } else { |
1201 | /* Using basename() GNU version which doesn't modify arg. */ | |
9fc205b4 | 1202 | libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name)); |
2ce8450e AN |
1203 | end = strchr(obj->name, '.'); |
1204 | if (end) | |
1205 | *end = 0; | |
1206 | } | |
6c956392 | 1207 | |
d859900c | 1208 | obj->efile.fd = -1; |
6c956392 | 1209 | /* |
76e1022b | 1210 | * Caller of this function should also call |
6c956392 WN |
1211 | * bpf_object__elf_finish() after data collection to return |
1212 | * obj_buf to user. If not, we should duplicate the buffer to | |
1213 | * avoid user freeing them before elf finish. | |
1214 | */ | |
1215 | obj->efile.obj_buf = obj_buf; | |
1216 | obj->efile.obj_buf_sz = obj_buf_sz; | |
666810e8 | 1217 | obj->efile.maps_shndx = -1; |
abd29c93 | 1218 | obj->efile.btf_maps_shndx = -1; |
590a0088 | 1219 | obj->efile.st_ops_shndx = -1; |
81bfdd08 | 1220 | obj->kconfig_map_idx = -1; |
6c956392 | 1221 | |
5e61f270 | 1222 | obj->kern_version = get_kernel_version(); |
52d3352e | 1223 | obj->loaded = false; |
9a208eff WN |
1224 | |
1225 | INIT_LIST_HEAD(&obj->list); | |
689624f0 JB |
1226 | if (!strict) |
1227 | list_add(&obj->list, &bpf_objects_list); | |
1a5e3fb1 WN |
1228 | return obj; |
1229 | } | |
1230 | ||
1231 | static void bpf_object__elf_finish(struct bpf_object *obj) | |
1232 | { | |
29a30ff5 | 1233 | if (!obj->efile.elf) |
1a5e3fb1 WN |
1234 | return; |
1235 | ||
b71a2ebf GC |
1236 | elf_end(obj->efile.elf); |
1237 | obj->efile.elf = NULL; | |
bec7d68c | 1238 | obj->efile.symbols = NULL; |
590a0088 | 1239 | obj->efile.st_ops_data = NULL; |
b62f06e8 | 1240 | |
25bbbd7a AN |
1241 | zfree(&obj->efile.secs); |
1242 | obj->efile.sec_cnt = 0; | |
1a5e3fb1 | 1243 | zclose(obj->efile.fd); |
6c956392 WN |
1244 | obj->efile.obj_buf = NULL; |
1245 | obj->efile.obj_buf_sz = 0; | |
1a5e3fb1 WN |
1246 | } |
1247 | ||
1248 | static int bpf_object__elf_init(struct bpf_object *obj) | |
1249 | { | |
ad23b723 | 1250 | Elf64_Ehdr *ehdr; |
1a5e3fb1 | 1251 | int err = 0; |
ad23b723 | 1252 | Elf *elf; |
1a5e3fb1 | 1253 | |
29a30ff5 | 1254 | if (obj->efile.elf) { |
88a82120 | 1255 | pr_warn("elf: init internal error\n"); |
6371ca3b | 1256 | return -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1257 | } |
1258 | ||
6c956392 WN |
1259 | if (obj->efile.obj_buf_sz > 0) { |
1260 | /* | |
1261 | * obj_buf should have been validated by | |
1262 | * bpf_object__open_buffer(). | |
1263 | */ | |
ad23b723 | 1264 | elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz); |
6c956392 | 1265 | } else { |
92274e24 | 1266 | obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC); |
6c956392 | 1267 | if (obj->efile.fd < 0) { |
be5c5d4e | 1268 | char errmsg[STRERR_BUFSIZE], *cp; |
1ce6a9fc | 1269 | |
be5c5d4e AN |
1270 | err = -errno; |
1271 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
88a82120 | 1272 | pr_warn("elf: failed to open %s: %s\n", obj->path, cp); |
be5c5d4e | 1273 | return err; |
6c956392 WN |
1274 | } |
1275 | ||
ad23b723 | 1276 | elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL); |
1a5e3fb1 WN |
1277 | } |
1278 | ||
ad23b723 | 1279 | if (!elf) { |
88a82120 | 1280 | pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1281 | err = -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1282 | goto errout; |
1283 | } | |
1284 | ||
ad23b723 AN |
1285 | obj->efile.elf = elf; |
1286 | ||
1287 | if (elf_kind(elf) != ELF_K_ELF) { | |
1288 | err = -LIBBPF_ERRNO__FORMAT; | |
1289 | pr_warn("elf: '%s' is not a proper ELF object\n", obj->path); | |
1290 | goto errout; | |
1291 | } | |
1292 | ||
1293 | if (gelf_getclass(elf) != ELFCLASS64) { | |
1294 | err = -LIBBPF_ERRNO__FORMAT; | |
1295 | pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path); | |
1296 | goto errout; | |
1297 | } | |
1298 | ||
1299 | obj->efile.ehdr = ehdr = elf64_getehdr(elf); | |
1300 | if (!obj->efile.ehdr) { | |
88a82120 | 1301 | pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1302 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1303 | goto errout; |
1304 | } | |
1a5e3fb1 | 1305 | |
ad23b723 | 1306 | if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) { |
88a82120 AN |
1307 | pr_warn("elf: failed to get section names section index for %s: %s\n", |
1308 | obj->path, elf_errmsg(-1)); | |
1309 | err = -LIBBPF_ERRNO__FORMAT; | |
1310 | goto errout; | |
1311 | } | |
1312 | ||
1313 | /* Elf is corrupted/truncated, avoid calling elf_strptr. */ | |
ad23b723 | 1314 | if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) { |
88a82120 AN |
1315 | pr_warn("elf: failed to get section names strings from %s: %s\n", |
1316 | obj->path, elf_errmsg(-1)); | |
8f3f5792 NK |
1317 | err = -LIBBPF_ERRNO__FORMAT; |
1318 | goto errout; | |
88a82120 AN |
1319 | } |
1320 | ||
9b16137a | 1321 | /* Old LLVM set e_machine to EM_NONE */ |
ad23b723 | 1322 | if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) { |
88a82120 | 1323 | pr_warn("elf: %s is not a valid eBPF object file\n", obj->path); |
6371ca3b | 1324 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1325 | goto errout; |
1326 | } | |
1327 | ||
1328 | return 0; | |
1329 | errout: | |
1330 | bpf_object__elf_finish(obj); | |
1331 | return err; | |
1332 | } | |
1333 | ||
12ef5634 | 1334 | static int bpf_object__check_endianness(struct bpf_object *obj) |
cc4228d5 | 1335 | { |
3930198d | 1336 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
ad23b723 | 1337 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB) |
12ef5634 | 1338 | return 0; |
3930198d | 1339 | #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
ad23b723 | 1340 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB) |
12ef5634 AN |
1341 | return 0; |
1342 | #else | |
1343 | # error "Unrecognized __BYTE_ORDER__" | |
1344 | #endif | |
88a82120 | 1345 | pr_warn("elf: endianness mismatch in %s.\n", obj->path); |
6371ca3b | 1346 | return -LIBBPF_ERRNO__ENDIAN; |
cc4228d5 WN |
1347 | } |
1348 | ||
cb1e5e96 | 1349 | static int |
399dc65e | 1350 | bpf_object__init_license(struct bpf_object *obj, void *data, size_t size) |
cb1e5e96 | 1351 | { |
f9798239 AN |
1352 | /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't |
1353 | * go over allowed ELF data section buffer | |
1354 | */ | |
1355 | libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license))); | |
cb1e5e96 WN |
1356 | pr_debug("license of %s is %s\n", obj->path, obj->license); |
1357 | return 0; | |
1358 | } | |
1359 | ||
54b8625c JF |
1360 | static int |
1361 | bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size) | |
1362 | { | |
1363 | __u32 kver; | |
1364 | ||
1365 | if (size != sizeof(kver)) { | |
be18010e | 1366 | pr_warn("invalid kver section in %s\n", obj->path); |
54b8625c JF |
1367 | return -LIBBPF_ERRNO__FORMAT; |
1368 | } | |
1369 | memcpy(&kver, data, sizeof(kver)); | |
1370 | obj->kern_version = kver; | |
1371 | pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version); | |
1372 | return 0; | |
1373 | } | |
1374 | ||
addb9fc9 NS |
1375 | static bool bpf_map_type__is_map_in_map(enum bpf_map_type type) |
1376 | { | |
1377 | if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1378 | type == BPF_MAP_TYPE_HASH_OF_MAPS) | |
1379 | return true; | |
1380 | return false; | |
1381 | } | |
1382 | ||
b96c07f3 | 1383 | static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size) |
1713d68b | 1384 | { |
25bbbd7a AN |
1385 | Elf_Data *data; |
1386 | Elf_Scn *scn; | |
1713d68b | 1387 | |
25bbbd7a | 1388 | if (!name) |
1713d68b | 1389 | return -EINVAL; |
88a82120 | 1390 | |
25bbbd7a AN |
1391 | scn = elf_sec_by_name(obj, name); |
1392 | data = elf_sec_data(obj, scn); | |
1393 | if (data) { | |
25bbbd7a | 1394 | *size = data->d_size; |
08894d9c | 1395 | return 0; /* found it */ |
1713d68b DB |
1396 | } |
1397 | ||
08894d9c | 1398 | return -ENOENT; |
1713d68b DB |
1399 | } |
1400 | ||
b96c07f3 | 1401 | static int find_elf_var_offset(const struct bpf_object *obj, const char *name, __u32 *off) |
1713d68b DB |
1402 | { |
1403 | Elf_Data *symbols = obj->efile.symbols; | |
1404 | const char *sname; | |
1405 | size_t si; | |
1406 | ||
1407 | if (!name || !off) | |
1408 | return -EINVAL; | |
1409 | ||
ad23b723 AN |
1410 | for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) { |
1411 | Elf64_Sym *sym = elf_sym_by_idx(obj, si); | |
1713d68b | 1412 | |
3a06ec0a AN |
1413 | if (ELF64_ST_TYPE(sym->st_info) != STT_OBJECT) |
1414 | continue; | |
1415 | ||
1416 | if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL && | |
1417 | ELF64_ST_BIND(sym->st_info) != STB_WEAK) | |
1713d68b DB |
1418 | continue; |
1419 | ||
ad23b723 | 1420 | sname = elf_sym_str(obj, sym->st_name); |
1713d68b | 1421 | if (!sname) { |
ad23b723 | 1422 | pr_warn("failed to get sym name string for var %s\n", name); |
1713d68b DB |
1423 | return -EIO; |
1424 | } | |
1425 | if (strcmp(name, sname) == 0) { | |
ad23b723 | 1426 | *off = sym->st_value; |
1713d68b DB |
1427 | return 0; |
1428 | } | |
1429 | } | |
1430 | ||
1431 | return -ENOENT; | |
1432 | } | |
1433 | ||
bf829271 | 1434 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj) |
d859900c | 1435 | { |
bf829271 AN |
1436 | struct bpf_map *new_maps; |
1437 | size_t new_cap; | |
1438 | int i; | |
1439 | ||
1440 | if (obj->nr_maps < obj->maps_cap) | |
1441 | return &obj->maps[obj->nr_maps++]; | |
1442 | ||
95064979 | 1443 | new_cap = max((size_t)4, obj->maps_cap * 3 / 2); |
029258d7 | 1444 | new_maps = libbpf_reallocarray(obj->maps, new_cap, sizeof(*obj->maps)); |
bf829271 | 1445 | if (!new_maps) { |
be18010e | 1446 | pr_warn("alloc maps for object failed\n"); |
bf829271 AN |
1447 | return ERR_PTR(-ENOMEM); |
1448 | } | |
1449 | ||
1450 | obj->maps_cap = new_cap; | |
1451 | obj->maps = new_maps; | |
1452 | ||
1453 | /* zero out new maps */ | |
1454 | memset(obj->maps + obj->nr_maps, 0, | |
1455 | (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps)); | |
1456 | /* | |
1457 | * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin) | |
1458 | * when failure (zclose won't close negative fd)). | |
1459 | */ | |
1460 | for (i = obj->nr_maps; i < obj->maps_cap; i++) { | |
1461 | obj->maps[i].fd = -1; | |
1462 | obj->maps[i].inner_map_fd = -1; | |
1463 | } | |
1464 | ||
1465 | return &obj->maps[obj->nr_maps++]; | |
d859900c DB |
1466 | } |
1467 | ||
eba9c5f4 AN |
1468 | static size_t bpf_map_mmap_sz(const struct bpf_map *map) |
1469 | { | |
1470 | long page_sz = sysconf(_SC_PAGE_SIZE); | |
1471 | size_t map_sz; | |
1472 | ||
c701917e | 1473 | map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries; |
eba9c5f4 AN |
1474 | map_sz = roundup(map_sz, page_sz); |
1475 | return map_sz; | |
1476 | } | |
1477 | ||
aed65917 | 1478 | static char *internal_map_name(struct bpf_object *obj, const char *real_name) |
81bfdd08 | 1479 | { |
113e6b7e | 1480 | char map_name[BPF_OBJ_NAME_LEN], *p; |
aed65917 AN |
1481 | int pfx_len, sfx_len = max((size_t)7, strlen(real_name)); |
1482 | ||
1483 | /* This is one of the more confusing parts of libbpf for various | |
1484 | * reasons, some of which are historical. The original idea for naming | |
1485 | * internal names was to include as much of BPF object name prefix as | |
1486 | * possible, so that it can be distinguished from similar internal | |
1487 | * maps of a different BPF object. | |
1488 | * As an example, let's say we have bpf_object named 'my_object_name' | |
1489 | * and internal map corresponding to '.rodata' ELF section. The final | |
1490 | * map name advertised to user and to the kernel will be | |
1491 | * 'my_objec.rodata', taking first 8 characters of object name and | |
1492 | * entire 7 characters of '.rodata'. | |
1493 | * Somewhat confusingly, if internal map ELF section name is shorter | |
1494 | * than 7 characters, e.g., '.bss', we still reserve 7 characters | |
1495 | * for the suffix, even though we only have 4 actual characters, and | |
1496 | * resulting map will be called 'my_objec.bss', not even using all 15 | |
1497 | * characters allowed by the kernel. Oh well, at least the truncated | |
1498 | * object name is somewhat consistent in this case. But if the map | |
1499 | * name is '.kconfig', we'll still have entirety of '.kconfig' added | |
1500 | * (8 chars) and thus will be left with only first 7 characters of the | |
1501 | * object name ('my_obje'). Happy guessing, user, that the final map | |
1502 | * name will be "my_obje.kconfig". | |
1503 | * Now, with libbpf starting to support arbitrarily named .rodata.* | |
1504 | * and .data.* data sections, it's possible that ELF section name is | |
1505 | * longer than allowed 15 chars, so we now need to be careful to take | |
1506 | * only up to 15 first characters of ELF name, taking no BPF object | |
1507 | * name characters at all. So '.rodata.abracadabra' will result in | |
1508 | * '.rodata.abracad' kernel and user-visible name. | |
1509 | * We need to keep this convoluted logic intact for .data, .bss and | |
1510 | * .rodata maps, but for new custom .data.custom and .rodata.custom | |
1511 | * maps we use their ELF names as is, not prepending bpf_object name | |
1512 | * in front. We still need to truncate them to 15 characters for the | |
1513 | * kernel. Full name can be recovered for such maps by using DATASEC | |
1514 | * BTF type associated with such map's value type, though. | |
1515 | */ | |
1516 | if (sfx_len >= BPF_OBJ_NAME_LEN) | |
1517 | sfx_len = BPF_OBJ_NAME_LEN - 1; | |
1518 | ||
1519 | /* if there are two or more dots in map name, it's a custom dot map */ | |
1520 | if (strchr(real_name + 1, '.') != NULL) | |
1521 | pfx_len = 0; | |
1522 | else | |
1523 | pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name)); | |
81bfdd08 AN |
1524 | |
1525 | snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name, | |
aed65917 | 1526 | sfx_len, real_name); |
81bfdd08 | 1527 | |
113e6b7e THJ |
1528 | /* sanitise map name to characters allowed by kernel */ |
1529 | for (p = map_name; *p && p < map_name + sizeof(map_name); p++) | |
1530 | if (!isalnum(*p) && *p != '_' && *p != '.') | |
1531 | *p = '_'; | |
1532 | ||
81bfdd08 AN |
1533 | return strdup(map_name); |
1534 | } | |
1535 | ||
262cfb74 DK |
1536 | static int |
1537 | bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map); | |
1538 | ||
d859900c | 1539 | static int |
bf829271 | 1540 | bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type, |
aed65917 | 1541 | const char *real_name, int sec_idx, void *data, size_t data_sz) |
d859900c | 1542 | { |
bf829271 AN |
1543 | struct bpf_map_def *def; |
1544 | struct bpf_map *map; | |
eba9c5f4 | 1545 | int err; |
bf829271 AN |
1546 | |
1547 | map = bpf_object__add_map(obj); | |
1548 | if (IS_ERR(map)) | |
1549 | return PTR_ERR(map); | |
d859900c DB |
1550 | |
1551 | map->libbpf_type = type; | |
db48814b AN |
1552 | map->sec_idx = sec_idx; |
1553 | map->sec_offset = 0; | |
aed65917 AN |
1554 | map->real_name = strdup(real_name); |
1555 | map->name = internal_map_name(obj, real_name); | |
1556 | if (!map->real_name || !map->name) { | |
1557 | zfree(&map->real_name); | |
1558 | zfree(&map->name); | |
d859900c DB |
1559 | return -ENOMEM; |
1560 | } | |
1561 | ||
bf829271 | 1562 | def = &map->def; |
d859900c DB |
1563 | def->type = BPF_MAP_TYPE_ARRAY; |
1564 | def->key_size = sizeof(int); | |
eba9c5f4 | 1565 | def->value_size = data_sz; |
d859900c | 1566 | def->max_entries = 1; |
81bfdd08 | 1567 | def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG |
166750bc | 1568 | ? BPF_F_RDONLY_PROG : 0; |
0d13bfce | 1569 | def->map_flags |= BPF_F_MMAPABLE; |
7fe74b43 AN |
1570 | |
1571 | pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n", | |
81bfdd08 | 1572 | map->name, map->sec_idx, map->sec_offset, def->map_flags); |
7fe74b43 | 1573 | |
eba9c5f4 AN |
1574 | map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE, |
1575 | MAP_SHARED | MAP_ANONYMOUS, -1, 0); | |
1576 | if (map->mmaped == MAP_FAILED) { | |
1577 | err = -errno; | |
1578 | map->mmaped = NULL; | |
1579 | pr_warn("failed to alloc map '%s' content buffer: %d\n", | |
1580 | map->name, err); | |
aed65917 | 1581 | zfree(&map->real_name); |
eba9c5f4 AN |
1582 | zfree(&map->name); |
1583 | return err; | |
d859900c DB |
1584 | } |
1585 | ||
262cfb74 DK |
1586 | /* failures are fine because of maps like .rodata.str1.1 */ |
1587 | (void) bpf_map_find_btf_info(obj, map); | |
1588 | ||
166750bc | 1589 | if (data) |
eba9c5f4 AN |
1590 | memcpy(map->mmaped, data, data_sz); |
1591 | ||
e1d1dc46 | 1592 | pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name); |
d859900c DB |
1593 | return 0; |
1594 | } | |
1595 | ||
bf829271 AN |
1596 | static int bpf_object__init_global_data_maps(struct bpf_object *obj) |
1597 | { | |
25bbbd7a | 1598 | struct elf_sec_desc *sec_desc; |
aed65917 | 1599 | const char *sec_name; |
25bbbd7a | 1600 | int err = 0, sec_idx; |
bf829271 | 1601 | |
bf829271 AN |
1602 | /* |
1603 | * Populate obj->maps with libbpf internal maps. | |
1604 | */ | |
25bbbd7a AN |
1605 | for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) { |
1606 | sec_desc = &obj->efile.secs[sec_idx]; | |
1607 | ||
1608 | switch (sec_desc->sec_type) { | |
1609 | case SEC_DATA: | |
aed65917 | 1610 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1611 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA, |
aed65917 | 1612 | sec_name, sec_idx, |
25bbbd7a AN |
1613 | sec_desc->data->d_buf, |
1614 | sec_desc->data->d_size); | |
1615 | break; | |
1616 | case SEC_RODATA: | |
1617 | obj->has_rodata = true; | |
aed65917 | 1618 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1619 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA, |
aed65917 | 1620 | sec_name, sec_idx, |
25bbbd7a AN |
1621 | sec_desc->data->d_buf, |
1622 | sec_desc->data->d_size); | |
1623 | break; | |
1624 | case SEC_BSS: | |
aed65917 | 1625 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1626 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS, |
aed65917 | 1627 | sec_name, sec_idx, |
25bbbd7a AN |
1628 | NULL, |
1629 | sec_desc->data->d_size); | |
1630 | break; | |
1631 | default: | |
1632 | /* skip */ | |
1633 | break; | |
1634 | } | |
bf829271 AN |
1635 | if (err) |
1636 | return err; | |
1637 | } | |
1638 | return 0; | |
1639 | } | |
1640 | ||
166750bc AN |
1641 | |
1642 | static struct extern_desc *find_extern_by_name(const struct bpf_object *obj, | |
1643 | const void *name) | |
1644 | { | |
1645 | int i; | |
1646 | ||
1647 | for (i = 0; i < obj->nr_extern; i++) { | |
1648 | if (strcmp(obj->externs[i].name, name) == 0) | |
1649 | return &obj->externs[i]; | |
1650 | } | |
1651 | return NULL; | |
1652 | } | |
1653 | ||
2e33efe3 AN |
1654 | static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val, |
1655 | char value) | |
166750bc | 1656 | { |
2e33efe3 AN |
1657 | switch (ext->kcfg.type) { |
1658 | case KCFG_BOOL: | |
166750bc | 1659 | if (value == 'm') { |
2e33efe3 | 1660 | pr_warn("extern (kcfg) %s=%c should be tristate or char\n", |
166750bc AN |
1661 | ext->name, value); |
1662 | return -EINVAL; | |
1663 | } | |
1664 | *(bool *)ext_val = value == 'y' ? true : false; | |
1665 | break; | |
2e33efe3 | 1666 | case KCFG_TRISTATE: |
166750bc AN |
1667 | if (value == 'y') |
1668 | *(enum libbpf_tristate *)ext_val = TRI_YES; | |
1669 | else if (value == 'm') | |
1670 | *(enum libbpf_tristate *)ext_val = TRI_MODULE; | |
1671 | else /* value == 'n' */ | |
1672 | *(enum libbpf_tristate *)ext_val = TRI_NO; | |
1673 | break; | |
2e33efe3 | 1674 | case KCFG_CHAR: |
166750bc AN |
1675 | *(char *)ext_val = value; |
1676 | break; | |
2e33efe3 AN |
1677 | case KCFG_UNKNOWN: |
1678 | case KCFG_INT: | |
1679 | case KCFG_CHAR_ARR: | |
166750bc | 1680 | default: |
2e33efe3 | 1681 | pr_warn("extern (kcfg) %s=%c should be bool, tristate, or char\n", |
166750bc AN |
1682 | ext->name, value); |
1683 | return -EINVAL; | |
1684 | } | |
1685 | ext->is_set = true; | |
1686 | return 0; | |
1687 | } | |
1688 | ||
2e33efe3 AN |
1689 | static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val, |
1690 | const char *value) | |
166750bc AN |
1691 | { |
1692 | size_t len; | |
1693 | ||
2e33efe3 AN |
1694 | if (ext->kcfg.type != KCFG_CHAR_ARR) { |
1695 | pr_warn("extern (kcfg) %s=%s should be char array\n", ext->name, value); | |
166750bc AN |
1696 | return -EINVAL; |
1697 | } | |
1698 | ||
1699 | len = strlen(value); | |
1700 | if (value[len - 1] != '"') { | |
2e33efe3 | 1701 | pr_warn("extern (kcfg) '%s': invalid string config '%s'\n", |
166750bc AN |
1702 | ext->name, value); |
1703 | return -EINVAL; | |
1704 | } | |
1705 | ||
1706 | /* strip quotes */ | |
1707 | len -= 2; | |
2e33efe3 AN |
1708 | if (len >= ext->kcfg.sz) { |
1709 | pr_warn("extern (kcfg) '%s': long string config %s of (%zu bytes) truncated to %d bytes\n", | |
1710 | ext->name, value, len, ext->kcfg.sz - 1); | |
1711 | len = ext->kcfg.sz - 1; | |
166750bc AN |
1712 | } |
1713 | memcpy(ext_val, value + 1, len); | |
1714 | ext_val[len] = '\0'; | |
1715 | ext->is_set = true; | |
1716 | return 0; | |
1717 | } | |
1718 | ||
1719 | static int parse_u64(const char *value, __u64 *res) | |
1720 | { | |
1721 | char *value_end; | |
1722 | int err; | |
1723 | ||
1724 | errno = 0; | |
1725 | *res = strtoull(value, &value_end, 0); | |
1726 | if (errno) { | |
1727 | err = -errno; | |
1728 | pr_warn("failed to parse '%s' as integer: %d\n", value, err); | |
1729 | return err; | |
1730 | } | |
1731 | if (*value_end) { | |
1732 | pr_warn("failed to parse '%s' as integer completely\n", value); | |
1733 | return -EINVAL; | |
1734 | } | |
1735 | return 0; | |
1736 | } | |
1737 | ||
2e33efe3 | 1738 | static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v) |
166750bc | 1739 | { |
2e33efe3 | 1740 | int bit_sz = ext->kcfg.sz * 8; |
166750bc | 1741 | |
2e33efe3 | 1742 | if (ext->kcfg.sz == 8) |
166750bc AN |
1743 | return true; |
1744 | ||
1745 | /* Validate that value stored in u64 fits in integer of `ext->sz` | |
1746 | * bytes size without any loss of information. If the target integer | |
1747 | * is signed, we rely on the following limits of integer type of | |
1748 | * Y bits and subsequent transformation: | |
1749 | * | |
1750 | * -2^(Y-1) <= X <= 2^(Y-1) - 1 | |
1751 | * 0 <= X + 2^(Y-1) <= 2^Y - 1 | |
1752 | * 0 <= X + 2^(Y-1) < 2^Y | |
1753 | * | |
1754 | * For unsigned target integer, check that all the (64 - Y) bits are | |
1755 | * zero. | |
1756 | */ | |
2e33efe3 | 1757 | if (ext->kcfg.is_signed) |
166750bc AN |
1758 | return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz); |
1759 | else | |
1760 | return (v >> bit_sz) == 0; | |
1761 | } | |
1762 | ||
2e33efe3 AN |
1763 | static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val, |
1764 | __u64 value) | |
166750bc | 1765 | { |
2e33efe3 AN |
1766 | if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) { |
1767 | pr_warn("extern (kcfg) %s=%llu should be integer\n", | |
7745ff98 | 1768 | ext->name, (unsigned long long)value); |
166750bc AN |
1769 | return -EINVAL; |
1770 | } | |
2e33efe3 AN |
1771 | if (!is_kcfg_value_in_range(ext, value)) { |
1772 | pr_warn("extern (kcfg) %s=%llu value doesn't fit in %d bytes\n", | |
1773 | ext->name, (unsigned long long)value, ext->kcfg.sz); | |
166750bc AN |
1774 | return -ERANGE; |
1775 | } | |
2e33efe3 | 1776 | switch (ext->kcfg.sz) { |
166750bc AN |
1777 | case 1: *(__u8 *)ext_val = value; break; |
1778 | case 2: *(__u16 *)ext_val = value; break; | |
1779 | case 4: *(__u32 *)ext_val = value; break; | |
1780 | case 8: *(__u64 *)ext_val = value; break; | |
1781 | default: | |
1782 | return -EINVAL; | |
1783 | } | |
1784 | ext->is_set = true; | |
1785 | return 0; | |
1786 | } | |
1787 | ||
8601fd42 AN |
1788 | static int bpf_object__process_kconfig_line(struct bpf_object *obj, |
1789 | char *buf, void *data) | |
166750bc | 1790 | { |
166750bc | 1791 | struct extern_desc *ext; |
8601fd42 | 1792 | char *sep, *value; |
166750bc AN |
1793 | int len, err = 0; |
1794 | void *ext_val; | |
1795 | __u64 num; | |
166750bc | 1796 | |
13d35a0c | 1797 | if (!str_has_pfx(buf, "CONFIG_")) |
8601fd42 | 1798 | return 0; |
166750bc | 1799 | |
8601fd42 AN |
1800 | sep = strchr(buf, '='); |
1801 | if (!sep) { | |
1802 | pr_warn("failed to parse '%s': no separator\n", buf); | |
1803 | return -EINVAL; | |
1804 | } | |
1805 | ||
1806 | /* Trim ending '\n' */ | |
1807 | len = strlen(buf); | |
1808 | if (buf[len - 1] == '\n') | |
1809 | buf[len - 1] = '\0'; | |
1810 | /* Split on '=' and ensure that a value is present. */ | |
1811 | *sep = '\0'; | |
1812 | if (!sep[1]) { | |
1813 | *sep = '='; | |
1814 | pr_warn("failed to parse '%s': no value\n", buf); | |
1815 | return -EINVAL; | |
1816 | } | |
1817 | ||
1818 | ext = find_extern_by_name(obj, buf); | |
1819 | if (!ext || ext->is_set) | |
1820 | return 0; | |
1821 | ||
2e33efe3 | 1822 | ext_val = data + ext->kcfg.data_off; |
8601fd42 AN |
1823 | value = sep + 1; |
1824 | ||
1825 | switch (*value) { | |
1826 | case 'y': case 'n': case 'm': | |
2e33efe3 | 1827 | err = set_kcfg_value_tri(ext, ext_val, *value); |
8601fd42 AN |
1828 | break; |
1829 | case '"': | |
2e33efe3 | 1830 | err = set_kcfg_value_str(ext, ext_val, value); |
8601fd42 AN |
1831 | break; |
1832 | default: | |
1833 | /* assume integer */ | |
1834 | err = parse_u64(value, &num); | |
1835 | if (err) { | |
2e33efe3 | 1836 | pr_warn("extern (kcfg) %s=%s should be integer\n", |
8601fd42 AN |
1837 | ext->name, value); |
1838 | return err; | |
1839 | } | |
2e33efe3 | 1840 | err = set_kcfg_value_num(ext, ext_val, num); |
8601fd42 | 1841 | break; |
166750bc | 1842 | } |
8601fd42 AN |
1843 | if (err) |
1844 | return err; | |
2e33efe3 | 1845 | pr_debug("extern (kcfg) %s=%s\n", ext->name, value); |
8601fd42 AN |
1846 | return 0; |
1847 | } | |
1848 | ||
1849 | static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data) | |
1850 | { | |
1851 | char buf[PATH_MAX]; | |
1852 | struct utsname uts; | |
1853 | int len, err = 0; | |
1854 | gzFile file; | |
1855 | ||
1856 | uname(&uts); | |
1857 | len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release); | |
1858 | if (len < 0) | |
1859 | return -EINVAL; | |
1860 | else if (len >= PATH_MAX) | |
1861 | return -ENAMETOOLONG; | |
1862 | ||
1863 | /* gzopen also accepts uncompressed files. */ | |
1864 | file = gzopen(buf, "r"); | |
1865 | if (!file) | |
1866 | file = gzopen("/proc/config.gz", "r"); | |
1867 | ||
166750bc | 1868 | if (!file) { |
8601fd42 | 1869 | pr_warn("failed to open system Kconfig\n"); |
166750bc AN |
1870 | return -ENOENT; |
1871 | } | |
1872 | ||
1873 | while (gzgets(file, buf, sizeof(buf))) { | |
8601fd42 AN |
1874 | err = bpf_object__process_kconfig_line(obj, buf, data); |
1875 | if (err) { | |
1876 | pr_warn("error parsing system Kconfig line '%s': %d\n", | |
1877 | buf, err); | |
166750bc AN |
1878 | goto out; |
1879 | } | |
8601fd42 | 1880 | } |
166750bc | 1881 | |
8601fd42 AN |
1882 | out: |
1883 | gzclose(file); | |
1884 | return err; | |
1885 | } | |
166750bc | 1886 | |
8601fd42 AN |
1887 | static int bpf_object__read_kconfig_mem(struct bpf_object *obj, |
1888 | const char *config, void *data) | |
1889 | { | |
1890 | char buf[PATH_MAX]; | |
1891 | int err = 0; | |
1892 | FILE *file; | |
166750bc | 1893 | |
8601fd42 AN |
1894 | file = fmemopen((void *)config, strlen(config), "r"); |
1895 | if (!file) { | |
1896 | err = -errno; | |
1897 | pr_warn("failed to open in-memory Kconfig: %d\n", err); | |
1898 | return err; | |
1899 | } | |
1900 | ||
1901 | while (fgets(buf, sizeof(buf), file)) { | |
1902 | err = bpf_object__process_kconfig_line(obj, buf, data); | |
1903 | if (err) { | |
1904 | pr_warn("error parsing in-memory Kconfig line '%s': %d\n", | |
1905 | buf, err); | |
166750bc AN |
1906 | break; |
1907 | } | |
166750bc AN |
1908 | } |
1909 | ||
8601fd42 | 1910 | fclose(file); |
166750bc AN |
1911 | return err; |
1912 | } | |
1913 | ||
81bfdd08 | 1914 | static int bpf_object__init_kconfig_map(struct bpf_object *obj) |
166750bc | 1915 | { |
2e33efe3 | 1916 | struct extern_desc *last_ext = NULL, *ext; |
166750bc | 1917 | size_t map_sz; |
2e33efe3 | 1918 | int i, err; |
166750bc | 1919 | |
2e33efe3 AN |
1920 | for (i = 0; i < obj->nr_extern; i++) { |
1921 | ext = &obj->externs[i]; | |
1922 | if (ext->type == EXT_KCFG) | |
1923 | last_ext = ext; | |
1924 | } | |
166750bc | 1925 | |
2e33efe3 AN |
1926 | if (!last_ext) |
1927 | return 0; | |
166750bc | 1928 | |
2e33efe3 | 1929 | map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz; |
81bfdd08 | 1930 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG, |
aed65917 | 1931 | ".kconfig", obj->efile.symbols_shndx, |
166750bc AN |
1932 | NULL, map_sz); |
1933 | if (err) | |
1934 | return err; | |
1935 | ||
81bfdd08 | 1936 | obj->kconfig_map_idx = obj->nr_maps - 1; |
166750bc AN |
1937 | |
1938 | return 0; | |
1939 | } | |
1940 | ||
bf829271 | 1941 | static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict) |
561bbcca | 1942 | { |
561bbcca | 1943 | Elf_Data *symbols = obj->efile.symbols; |
bf829271 | 1944 | int i, map_def_sz = 0, nr_maps = 0, nr_syms; |
d859900c | 1945 | Elf_Data *data = NULL; |
bf829271 AN |
1946 | Elf_Scn *scn; |
1947 | ||
1948 | if (obj->efile.maps_shndx < 0) | |
1949 | return 0; | |
561bbcca | 1950 | |
93b8952d AN |
1951 | if (libbpf_mode & LIBBPF_STRICT_MAP_DEFINITIONS) { |
1952 | pr_warn("legacy map definitions in SEC(\"maps\") are not supported\n"); | |
1953 | return -EOPNOTSUPP; | |
1954 | } | |
1955 | ||
4708bbda EL |
1956 | if (!symbols) |
1957 | return -EINVAL; | |
1958 | ||
88a82120 AN |
1959 | scn = elf_sec_by_idx(obj, obj->efile.maps_shndx); |
1960 | data = elf_sec_data(obj, scn); | |
bf829271 | 1961 | if (!scn || !data) { |
88a82120 AN |
1962 | pr_warn("elf: failed to get legacy map definitions for %s\n", |
1963 | obj->path); | |
bf829271 | 1964 | return -EINVAL; |
4708bbda | 1965 | } |
561bbcca | 1966 | |
4708bbda EL |
1967 | /* |
1968 | * Count number of maps. Each map has a name. | |
1969 | * Array of maps is not supported: only the first element is | |
1970 | * considered. | |
1971 | * | |
1972 | * TODO: Detect array of map and report error. | |
1973 | */ | |
ad23b723 | 1974 | nr_syms = symbols->d_size / sizeof(Elf64_Sym); |
bf829271 | 1975 | for (i = 0; i < nr_syms; i++) { |
ad23b723 | 1976 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
4708bbda | 1977 | |
ad23b723 | 1978 | if (sym->st_shndx != obj->efile.maps_shndx) |
4708bbda | 1979 | continue; |
ad23b723 | 1980 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION) |
161ecd53 | 1981 | continue; |
4708bbda EL |
1982 | nr_maps++; |
1983 | } | |
b13c5c14 | 1984 | /* Assume equally sized map definitions */ |
88a82120 AN |
1985 | pr_debug("elf: found %d legacy map definitions (%zd bytes) in %s\n", |
1986 | nr_maps, data->d_size, obj->path); | |
bf829271 | 1987 | |
98e527af | 1988 | if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) { |
88a82120 AN |
1989 | pr_warn("elf: unable to determine legacy map definition size in %s\n", |
1990 | obj->path); | |
bf829271 | 1991 | return -EINVAL; |
addb9fc9 | 1992 | } |
98e527af | 1993 | map_def_sz = data->d_size / nr_maps; |
4708bbda | 1994 | |
bf829271 AN |
1995 | /* Fill obj->maps using data in "maps" section. */ |
1996 | for (i = 0; i < nr_syms; i++) { | |
ad23b723 | 1997 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
561bbcca | 1998 | const char *map_name; |
4708bbda | 1999 | struct bpf_map_def *def; |
bf829271 | 2000 | struct bpf_map *map; |
561bbcca | 2001 | |
ad23b723 | 2002 | if (sym->st_shndx != obj->efile.maps_shndx) |
561bbcca | 2003 | continue; |
ad23b723 | 2004 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION) |
c3e8c44a | 2005 | continue; |
561bbcca | 2006 | |
bf829271 AN |
2007 | map = bpf_object__add_map(obj); |
2008 | if (IS_ERR(map)) | |
2009 | return PTR_ERR(map); | |
2010 | ||
ad23b723 | 2011 | map_name = elf_sym_str(obj, sym->st_name); |
c51829bb | 2012 | if (!map_name) { |
be18010e KW |
2013 | pr_warn("failed to get map #%d name sym string for obj %s\n", |
2014 | i, obj->path); | |
c51829bb AN |
2015 | return -LIBBPF_ERRNO__FORMAT; |
2016 | } | |
d859900c | 2017 | |
93b8952d AN |
2018 | pr_warn("map '%s' (legacy): legacy map definitions are deprecated, use BTF-defined maps instead\n", map_name); |
2019 | ||
ad23b723 | 2020 | if (ELF64_ST_BIND(sym->st_info) == STB_LOCAL) { |
c1cccec9 AN |
2021 | pr_warn("map '%s' (legacy): static maps are not supported\n", map_name); |
2022 | return -ENOTSUP; | |
2023 | } | |
2024 | ||
bf829271 | 2025 | map->libbpf_type = LIBBPF_MAP_UNSPEC; |
ad23b723 AN |
2026 | map->sec_idx = sym->st_shndx; |
2027 | map->sec_offset = sym->st_value; | |
db48814b AN |
2028 | pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n", |
2029 | map_name, map->sec_idx, map->sec_offset); | |
ad23b723 | 2030 | if (sym->st_value + map_def_sz > data->d_size) { |
be18010e KW |
2031 | pr_warn("corrupted maps section in %s: last map \"%s\" too small\n", |
2032 | obj->path, map_name); | |
4708bbda | 2033 | return -EINVAL; |
561bbcca | 2034 | } |
4708bbda | 2035 | |
bf829271 AN |
2036 | map->name = strdup(map_name); |
2037 | if (!map->name) { | |
aed65917 | 2038 | pr_warn("map '%s': failed to alloc map name\n", map_name); |
973170e6 WN |
2039 | return -ENOMEM; |
2040 | } | |
bf829271 | 2041 | pr_debug("map %d is \"%s\"\n", i, map->name); |
ad23b723 | 2042 | def = (struct bpf_map_def *)(data->d_buf + sym->st_value); |
b13c5c14 CG |
2043 | /* |
2044 | * If the definition of the map in the object file fits in | |
2045 | * bpf_map_def, copy it. Any extra fields in our version | |
2046 | * of bpf_map_def will default to zero as a result of the | |
2047 | * calloc above. | |
2048 | */ | |
2049 | if (map_def_sz <= sizeof(struct bpf_map_def)) { | |
bf829271 | 2050 | memcpy(&map->def, def, map_def_sz); |
b13c5c14 CG |
2051 | } else { |
2052 | /* | |
2053 | * Here the map structure being read is bigger than what | |
2054 | * we expect, truncate if the excess bits are all zero. | |
2055 | * If they are not zero, reject this map as | |
2056 | * incompatible. | |
2057 | */ | |
2058 | char *b; | |
8983b731 | 2059 | |
b13c5c14 CG |
2060 | for (b = ((char *)def) + sizeof(struct bpf_map_def); |
2061 | b < ((char *)def) + map_def_sz; b++) { | |
2062 | if (*b != 0) { | |
8983b731 | 2063 | pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n", |
be18010e | 2064 | obj->path, map_name); |
c034a177 JF |
2065 | if (strict) |
2066 | return -EINVAL; | |
b13c5c14 CG |
2067 | } |
2068 | } | |
bf829271 | 2069 | memcpy(&map->def, def, sizeof(struct bpf_map_def)); |
b13c5c14 | 2070 | } |
262cfb74 DK |
2071 | |
2072 | /* btf info may not exist but fill it in if it does exist */ | |
2073 | (void) bpf_map_find_btf_info(obj, map); | |
561bbcca | 2074 | } |
bf829271 AN |
2075 | return 0; |
2076 | } | |
4708bbda | 2077 | |
42869d28 | 2078 | const struct btf_type * |
ddc7c304 | 2079 | skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id) |
abd29c93 AN |
2080 | { |
2081 | const struct btf_type *t = btf__type_by_id(btf, id); | |
8837fe5d | 2082 | |
ddc7c304 AN |
2083 | if (res_id) |
2084 | *res_id = id; | |
2085 | ||
2086 | while (btf_is_mod(t) || btf_is_typedef(t)) { | |
2087 | if (res_id) | |
2088 | *res_id = t->type; | |
2089 | t = btf__type_by_id(btf, t->type); | |
abd29c93 | 2090 | } |
ddc7c304 AN |
2091 | |
2092 | return t; | |
abd29c93 AN |
2093 | } |
2094 | ||
590a0088 MKL |
2095 | static const struct btf_type * |
2096 | resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id) | |
2097 | { | |
2098 | const struct btf_type *t; | |
2099 | ||
2100 | t = skip_mods_and_typedefs(btf, id, NULL); | |
2101 | if (!btf_is_ptr(t)) | |
2102 | return NULL; | |
2103 | ||
2104 | t = skip_mods_and_typedefs(btf, t->type, res_id); | |
2105 | ||
2106 | return btf_is_func_proto(t) ? t : NULL; | |
2107 | } | |
2108 | ||
774e132e | 2109 | static const char *__btf_kind_str(__u16 kind) |
81ba0889 | 2110 | { |
774e132e | 2111 | switch (kind) { |
81ba0889 AN |
2112 | case BTF_KIND_UNKN: return "void"; |
2113 | case BTF_KIND_INT: return "int"; | |
2114 | case BTF_KIND_PTR: return "ptr"; | |
2115 | case BTF_KIND_ARRAY: return "array"; | |
2116 | case BTF_KIND_STRUCT: return "struct"; | |
2117 | case BTF_KIND_UNION: return "union"; | |
2118 | case BTF_KIND_ENUM: return "enum"; | |
2119 | case BTF_KIND_FWD: return "fwd"; | |
2120 | case BTF_KIND_TYPEDEF: return "typedef"; | |
2121 | case BTF_KIND_VOLATILE: return "volatile"; | |
2122 | case BTF_KIND_CONST: return "const"; | |
2123 | case BTF_KIND_RESTRICT: return "restrict"; | |
2124 | case BTF_KIND_FUNC: return "func"; | |
2125 | case BTF_KIND_FUNC_PROTO: return "func_proto"; | |
2126 | case BTF_KIND_VAR: return "var"; | |
2127 | case BTF_KIND_DATASEC: return "datasec"; | |
22541a9e | 2128 | case BTF_KIND_FLOAT: return "float"; |
223f903e | 2129 | case BTF_KIND_DECL_TAG: return "decl_tag"; |
2dc1e488 | 2130 | case BTF_KIND_TYPE_TAG: return "type_tag"; |
81ba0889 AN |
2131 | default: return "unknown"; |
2132 | } | |
2133 | } | |
2134 | ||
42869d28 | 2135 | const char *btf_kind_str(const struct btf_type *t) |
774e132e MKL |
2136 | { |
2137 | return __btf_kind_str(btf_kind(t)); | |
2138 | } | |
2139 | ||
ef99b02b AN |
2140 | /* |
2141 | * Fetch integer attribute of BTF map definition. Such attributes are | |
2142 | * represented using a pointer to an array, in which dimensionality of array | |
2143 | * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY]; | |
2144 | * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF | |
2145 | * type definition, while using only sizeof(void *) space in ELF data section. | |
2146 | */ | |
2147 | static bool get_map_field_int(const char *map_name, const struct btf *btf, | |
8983b731 AN |
2148 | const struct btf_member *m, __u32 *res) |
2149 | { | |
ddc7c304 | 2150 | const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL); |
abd29c93 | 2151 | const char *name = btf__name_by_offset(btf, m->name_off); |
ef99b02b AN |
2152 | const struct btf_array *arr_info; |
2153 | const struct btf_type *arr_t; | |
abd29c93 | 2154 | |
b03bc685 | 2155 | if (!btf_is_ptr(t)) { |
81ba0889 AN |
2156 | pr_warn("map '%s': attr '%s': expected PTR, got %s.\n", |
2157 | map_name, name, btf_kind_str(t)); | |
abd29c93 AN |
2158 | return false; |
2159 | } | |
ef99b02b AN |
2160 | |
2161 | arr_t = btf__type_by_id(btf, t->type); | |
2162 | if (!arr_t) { | |
be18010e KW |
2163 | pr_warn("map '%s': attr '%s': type [%u] not found.\n", |
2164 | map_name, name, t->type); | |
abd29c93 AN |
2165 | return false; |
2166 | } | |
b03bc685 | 2167 | if (!btf_is_array(arr_t)) { |
81ba0889 AN |
2168 | pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n", |
2169 | map_name, name, btf_kind_str(arr_t)); | |
abd29c93 AN |
2170 | return false; |
2171 | } | |
b03bc685 | 2172 | arr_info = btf_array(arr_t); |
ef99b02b | 2173 | *res = arr_info->nelems; |
abd29c93 AN |
2174 | return true; |
2175 | } | |
2176 | ||
57a00f41 THJ |
2177 | static int build_map_pin_path(struct bpf_map *map, const char *path) |
2178 | { | |
2179 | char buf[PATH_MAX]; | |
6e9cab2e | 2180 | int len; |
57a00f41 THJ |
2181 | |
2182 | if (!path) | |
2183 | path = "/sys/fs/bpf"; | |
2184 | ||
2185 | len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map)); | |
2186 | if (len < 0) | |
2187 | return -EINVAL; | |
2188 | else if (len >= PATH_MAX) | |
2189 | return -ENAMETOOLONG; | |
2190 | ||
6e9cab2e | 2191 | return bpf_map__set_pin_path(map, buf); |
57a00f41 THJ |
2192 | } |
2193 | ||
c7ef5ec9 AN |
2194 | int parse_btf_map_def(const char *map_name, struct btf *btf, |
2195 | const struct btf_type *def_t, bool strict, | |
2196 | struct btf_map_def *map_def, struct btf_map_def *inner_def) | |
abd29c93 | 2197 | { |
41017e56 | 2198 | const struct btf_type *t; |
abd29c93 | 2199 | const struct btf_member *m; |
c7ef5ec9 | 2200 | bool is_inner = inner_def == NULL; |
abd29c93 AN |
2201 | int vlen, i; |
2202 | ||
c7ef5ec9 AN |
2203 | vlen = btf_vlen(def_t); |
2204 | m = btf_members(def_t); | |
abd29c93 | 2205 | for (i = 0; i < vlen; i++, m++) { |
c7ef5ec9 | 2206 | const char *name = btf__name_by_offset(btf, m->name_off); |
abd29c93 AN |
2207 | |
2208 | if (!name) { | |
c7ef5ec9 | 2209 | pr_warn("map '%s': invalid field #%d.\n", map_name, i); |
abd29c93 AN |
2210 | return -EINVAL; |
2211 | } | |
2212 | if (strcmp(name, "type") == 0) { | |
c7ef5ec9 | 2213 | if (!get_map_field_int(map_name, btf, m, &map_def->map_type)) |
abd29c93 | 2214 | return -EINVAL; |
c7ef5ec9 | 2215 | map_def->parts |= MAP_DEF_MAP_TYPE; |
abd29c93 | 2216 | } else if (strcmp(name, "max_entries") == 0) { |
c7ef5ec9 | 2217 | if (!get_map_field_int(map_name, btf, m, &map_def->max_entries)) |
abd29c93 | 2218 | return -EINVAL; |
c7ef5ec9 | 2219 | map_def->parts |= MAP_DEF_MAX_ENTRIES; |
abd29c93 | 2220 | } else if (strcmp(name, "map_flags") == 0) { |
c7ef5ec9 | 2221 | if (!get_map_field_int(map_name, btf, m, &map_def->map_flags)) |
abd29c93 | 2222 | return -EINVAL; |
c7ef5ec9 | 2223 | map_def->parts |= MAP_DEF_MAP_FLAGS; |
1bdb6c9a | 2224 | } else if (strcmp(name, "numa_node") == 0) { |
c7ef5ec9 | 2225 | if (!get_map_field_int(map_name, btf, m, &map_def->numa_node)) |
1bdb6c9a | 2226 | return -EINVAL; |
c7ef5ec9 | 2227 | map_def->parts |= MAP_DEF_NUMA_NODE; |
abd29c93 AN |
2228 | } else if (strcmp(name, "key_size") == 0) { |
2229 | __u32 sz; | |
2230 | ||
c7ef5ec9 | 2231 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2232 | return -EINVAL; |
c7ef5ec9 | 2233 | if (map_def->key_size && map_def->key_size != sz) { |
be18010e | 2234 | pr_warn("map '%s': conflicting key size %u != %u.\n", |
c7ef5ec9 | 2235 | map_name, map_def->key_size, sz); |
abd29c93 AN |
2236 | return -EINVAL; |
2237 | } | |
c7ef5ec9 AN |
2238 | map_def->key_size = sz; |
2239 | map_def->parts |= MAP_DEF_KEY_SIZE; | |
abd29c93 AN |
2240 | } else if (strcmp(name, "key") == 0) { |
2241 | __s64 sz; | |
2242 | ||
c7ef5ec9 | 2243 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2244 | if (!t) { |
be18010e | 2245 | pr_warn("map '%s': key type [%d] not found.\n", |
c7ef5ec9 | 2246 | map_name, m->type); |
abd29c93 AN |
2247 | return -EINVAL; |
2248 | } | |
b03bc685 | 2249 | if (!btf_is_ptr(t)) { |
81ba0889 | 2250 | pr_warn("map '%s': key spec is not PTR: %s.\n", |
c7ef5ec9 | 2251 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2252 | return -EINVAL; |
2253 | } | |
c7ef5ec9 | 2254 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2255 | if (sz < 0) { |
679152d3 | 2256 | pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n", |
c7ef5ec9 | 2257 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2258 | return sz; |
2259 | } | |
c7ef5ec9 | 2260 | if (map_def->key_size && map_def->key_size != sz) { |
679152d3 | 2261 | pr_warn("map '%s': conflicting key size %u != %zd.\n", |
c7ef5ec9 | 2262 | map_name, map_def->key_size, (ssize_t)sz); |
abd29c93 AN |
2263 | return -EINVAL; |
2264 | } | |
c7ef5ec9 AN |
2265 | map_def->key_size = sz; |
2266 | map_def->key_type_id = t->type; | |
2267 | map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE; | |
abd29c93 AN |
2268 | } else if (strcmp(name, "value_size") == 0) { |
2269 | __u32 sz; | |
2270 | ||
c7ef5ec9 | 2271 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2272 | return -EINVAL; |
c7ef5ec9 | 2273 | if (map_def->value_size && map_def->value_size != sz) { |
be18010e | 2274 | pr_warn("map '%s': conflicting value size %u != %u.\n", |
c7ef5ec9 | 2275 | map_name, map_def->value_size, sz); |
abd29c93 AN |
2276 | return -EINVAL; |
2277 | } | |
c7ef5ec9 AN |
2278 | map_def->value_size = sz; |
2279 | map_def->parts |= MAP_DEF_VALUE_SIZE; | |
abd29c93 AN |
2280 | } else if (strcmp(name, "value") == 0) { |
2281 | __s64 sz; | |
2282 | ||
c7ef5ec9 | 2283 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2284 | if (!t) { |
be18010e | 2285 | pr_warn("map '%s': value type [%d] not found.\n", |
c7ef5ec9 | 2286 | map_name, m->type); |
abd29c93 AN |
2287 | return -EINVAL; |
2288 | } | |
b03bc685 | 2289 | if (!btf_is_ptr(t)) { |
81ba0889 | 2290 | pr_warn("map '%s': value spec is not PTR: %s.\n", |
c7ef5ec9 | 2291 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2292 | return -EINVAL; |
2293 | } | |
c7ef5ec9 | 2294 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2295 | if (sz < 0) { |
679152d3 | 2296 | pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n", |
c7ef5ec9 | 2297 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2298 | return sz; |
2299 | } | |
c7ef5ec9 | 2300 | if (map_def->value_size && map_def->value_size != sz) { |
679152d3 | 2301 | pr_warn("map '%s': conflicting value size %u != %zd.\n", |
c7ef5ec9 | 2302 | map_name, map_def->value_size, (ssize_t)sz); |
abd29c93 AN |
2303 | return -EINVAL; |
2304 | } | |
c7ef5ec9 AN |
2305 | map_def->value_size = sz; |
2306 | map_def->value_type_id = t->type; | |
2307 | map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE; | |
646f02ff AN |
2308 | } |
2309 | else if (strcmp(name, "values") == 0) { | |
341ac5ff HC |
2310 | bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type); |
2311 | bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY; | |
2312 | const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value"; | |
c7ef5ec9 | 2313 | char inner_map_name[128]; |
646f02ff AN |
2314 | int err; |
2315 | ||
2316 | if (is_inner) { | |
2317 | pr_warn("map '%s': multi-level inner maps not supported.\n", | |
c7ef5ec9 | 2318 | map_name); |
646f02ff AN |
2319 | return -ENOTSUP; |
2320 | } | |
2321 | if (i != vlen - 1) { | |
2322 | pr_warn("map '%s': '%s' member should be last.\n", | |
c7ef5ec9 | 2323 | map_name, name); |
646f02ff AN |
2324 | return -EINVAL; |
2325 | } | |
341ac5ff HC |
2326 | if (!is_map_in_map && !is_prog_array) { |
2327 | pr_warn("map '%s': should be map-in-map or prog-array.\n", | |
c7ef5ec9 | 2328 | map_name); |
646f02ff AN |
2329 | return -ENOTSUP; |
2330 | } | |
c7ef5ec9 | 2331 | if (map_def->value_size && map_def->value_size != 4) { |
646f02ff | 2332 | pr_warn("map '%s': conflicting value size %u != 4.\n", |
c7ef5ec9 | 2333 | map_name, map_def->value_size); |
646f02ff AN |
2334 | return -EINVAL; |
2335 | } | |
c7ef5ec9 AN |
2336 | map_def->value_size = 4; |
2337 | t = btf__type_by_id(btf, m->type); | |
646f02ff | 2338 | if (!t) { |
341ac5ff HC |
2339 | pr_warn("map '%s': %s type [%d] not found.\n", |
2340 | map_name, desc, m->type); | |
646f02ff AN |
2341 | return -EINVAL; |
2342 | } | |
2343 | if (!btf_is_array(t) || btf_array(t)->nelems) { | |
341ac5ff HC |
2344 | pr_warn("map '%s': %s spec is not a zero-sized array.\n", |
2345 | map_name, desc); | |
646f02ff AN |
2346 | return -EINVAL; |
2347 | } | |
c7ef5ec9 | 2348 | t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL); |
646f02ff | 2349 | if (!btf_is_ptr(t)) { |
341ac5ff HC |
2350 | pr_warn("map '%s': %s def is of unexpected kind %s.\n", |
2351 | map_name, desc, btf_kind_str(t)); | |
646f02ff AN |
2352 | return -EINVAL; |
2353 | } | |
c7ef5ec9 | 2354 | t = skip_mods_and_typedefs(btf, t->type, NULL); |
341ac5ff HC |
2355 | if (is_prog_array) { |
2356 | if (!btf_is_func_proto(t)) { | |
2357 | pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n", | |
2358 | map_name, btf_kind_str(t)); | |
2359 | return -EINVAL; | |
2360 | } | |
2361 | continue; | |
2362 | } | |
646f02ff | 2363 | if (!btf_is_struct(t)) { |
81ba0889 | 2364 | pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n", |
c7ef5ec9 | 2365 | map_name, btf_kind_str(t)); |
646f02ff AN |
2366 | return -EINVAL; |
2367 | } | |
2368 | ||
c7ef5ec9 AN |
2369 | snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name); |
2370 | err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL); | |
646f02ff AN |
2371 | if (err) |
2372 | return err; | |
c7ef5ec9 AN |
2373 | |
2374 | map_def->parts |= MAP_DEF_INNER_MAP; | |
57a00f41 THJ |
2375 | } else if (strcmp(name, "pinning") == 0) { |
2376 | __u32 val; | |
57a00f41 | 2377 | |
646f02ff | 2378 | if (is_inner) { |
c7ef5ec9 | 2379 | pr_warn("map '%s': inner def can't be pinned.\n", map_name); |
646f02ff AN |
2380 | return -EINVAL; |
2381 | } | |
c7ef5ec9 | 2382 | if (!get_map_field_int(map_name, btf, m, &val)) |
57a00f41 | 2383 | return -EINVAL; |
c7ef5ec9 | 2384 | if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) { |
57a00f41 | 2385 | pr_warn("map '%s': invalid pinning value %u.\n", |
c7ef5ec9 | 2386 | map_name, val); |
57a00f41 THJ |
2387 | return -EINVAL; |
2388 | } | |
c7ef5ec9 AN |
2389 | map_def->pinning = val; |
2390 | map_def->parts |= MAP_DEF_PINNING; | |
47512102 JK |
2391 | } else if (strcmp(name, "map_extra") == 0) { |
2392 | __u32 map_extra; | |
2393 | ||
2394 | if (!get_map_field_int(map_name, btf, m, &map_extra)) | |
2395 | return -EINVAL; | |
2396 | map_def->map_extra = map_extra; | |
2397 | map_def->parts |= MAP_DEF_MAP_EXTRA; | |
abd29c93 AN |
2398 | } else { |
2399 | if (strict) { | |
c7ef5ec9 | 2400 | pr_warn("map '%s': unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2401 | return -ENOTSUP; |
2402 | } | |
c7ef5ec9 | 2403 | pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2404 | } |
2405 | } | |
2406 | ||
c7ef5ec9 AN |
2407 | if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) { |
2408 | pr_warn("map '%s': map type isn't specified.\n", map_name); | |
abd29c93 AN |
2409 | return -EINVAL; |
2410 | } | |
2411 | ||
2412 | return 0; | |
2413 | } | |
2414 | ||
c7ef5ec9 AN |
2415 | static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def) |
2416 | { | |
2417 | map->def.type = def->map_type; | |
2418 | map->def.key_size = def->key_size; | |
2419 | map->def.value_size = def->value_size; | |
2420 | map->def.max_entries = def->max_entries; | |
2421 | map->def.map_flags = def->map_flags; | |
47512102 | 2422 | map->map_extra = def->map_extra; |
c7ef5ec9 AN |
2423 | |
2424 | map->numa_node = def->numa_node; | |
2425 | map->btf_key_type_id = def->key_type_id; | |
2426 | map->btf_value_type_id = def->value_type_id; | |
2427 | ||
2428 | if (def->parts & MAP_DEF_MAP_TYPE) | |
2429 | pr_debug("map '%s': found type = %u.\n", map->name, def->map_type); | |
2430 | ||
2431 | if (def->parts & MAP_DEF_KEY_TYPE) | |
2432 | pr_debug("map '%s': found key [%u], sz = %u.\n", | |
2433 | map->name, def->key_type_id, def->key_size); | |
2434 | else if (def->parts & MAP_DEF_KEY_SIZE) | |
2435 | pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size); | |
2436 | ||
2437 | if (def->parts & MAP_DEF_VALUE_TYPE) | |
2438 | pr_debug("map '%s': found value [%u], sz = %u.\n", | |
2439 | map->name, def->value_type_id, def->value_size); | |
2440 | else if (def->parts & MAP_DEF_VALUE_SIZE) | |
2441 | pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size); | |
2442 | ||
2443 | if (def->parts & MAP_DEF_MAX_ENTRIES) | |
2444 | pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries); | |
2445 | if (def->parts & MAP_DEF_MAP_FLAGS) | |
47512102 JK |
2446 | pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags); |
2447 | if (def->parts & MAP_DEF_MAP_EXTRA) | |
2448 | pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name, | |
2449 | (unsigned long long)def->map_extra); | |
c7ef5ec9 AN |
2450 | if (def->parts & MAP_DEF_PINNING) |
2451 | pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning); | |
2452 | if (def->parts & MAP_DEF_NUMA_NODE) | |
2453 | pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node); | |
2454 | ||
2455 | if (def->parts & MAP_DEF_INNER_MAP) | |
2456 | pr_debug("map '%s': found inner map definition.\n", map->name); | |
2457 | } | |
2458 | ||
c1cccec9 AN |
2459 | static const char *btf_var_linkage_str(__u32 linkage) |
2460 | { | |
2461 | switch (linkage) { | |
2462 | case BTF_VAR_STATIC: return "static"; | |
2463 | case BTF_VAR_GLOBAL_ALLOCATED: return "global"; | |
2464 | case BTF_VAR_GLOBAL_EXTERN: return "extern"; | |
2465 | default: return "unknown"; | |
2466 | } | |
2467 | } | |
2468 | ||
41017e56 AN |
2469 | static int bpf_object__init_user_btf_map(struct bpf_object *obj, |
2470 | const struct btf_type *sec, | |
2471 | int var_idx, int sec_idx, | |
2472 | const Elf_Data *data, bool strict, | |
2473 | const char *pin_root_path) | |
2474 | { | |
c7ef5ec9 | 2475 | struct btf_map_def map_def = {}, inner_def = {}; |
41017e56 AN |
2476 | const struct btf_type *var, *def; |
2477 | const struct btf_var_secinfo *vi; | |
2478 | const struct btf_var *var_extra; | |
2479 | const char *map_name; | |
2480 | struct bpf_map *map; | |
c7ef5ec9 | 2481 | int err; |
41017e56 AN |
2482 | |
2483 | vi = btf_var_secinfos(sec) + var_idx; | |
2484 | var = btf__type_by_id(obj->btf, vi->type); | |
2485 | var_extra = btf_var(var); | |
2486 | map_name = btf__name_by_offset(obj->btf, var->name_off); | |
2487 | ||
2488 | if (map_name == NULL || map_name[0] == '\0') { | |
2489 | pr_warn("map #%d: empty name.\n", var_idx); | |
2490 | return -EINVAL; | |
2491 | } | |
2492 | if ((__u64)vi->offset + vi->size > data->d_size) { | |
2493 | pr_warn("map '%s' BTF data is corrupted.\n", map_name); | |
2494 | return -EINVAL; | |
2495 | } | |
2496 | if (!btf_is_var(var)) { | |
81ba0889 AN |
2497 | pr_warn("map '%s': unexpected var kind %s.\n", |
2498 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2499 | return -EINVAL; |
2500 | } | |
c1cccec9 AN |
2501 | if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) { |
2502 | pr_warn("map '%s': unsupported map linkage %s.\n", | |
2503 | map_name, btf_var_linkage_str(var_extra->linkage)); | |
41017e56 AN |
2504 | return -EOPNOTSUPP; |
2505 | } | |
2506 | ||
2507 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
2508 | if (!btf_is_struct(def)) { | |
81ba0889 AN |
2509 | pr_warn("map '%s': unexpected def kind %s.\n", |
2510 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2511 | return -EINVAL; |
2512 | } | |
2513 | if (def->size > vi->size) { | |
2514 | pr_warn("map '%s': invalid def size.\n", map_name); | |
2515 | return -EINVAL; | |
2516 | } | |
2517 | ||
2518 | map = bpf_object__add_map(obj); | |
2519 | if (IS_ERR(map)) | |
2520 | return PTR_ERR(map); | |
2521 | map->name = strdup(map_name); | |
2522 | if (!map->name) { | |
2523 | pr_warn("map '%s': failed to alloc map name.\n", map_name); | |
2524 | return -ENOMEM; | |
2525 | } | |
2526 | map->libbpf_type = LIBBPF_MAP_UNSPEC; | |
2527 | map->def.type = BPF_MAP_TYPE_UNSPEC; | |
2528 | map->sec_idx = sec_idx; | |
2529 | map->sec_offset = vi->offset; | |
646f02ff | 2530 | map->btf_var_idx = var_idx; |
41017e56 AN |
2531 | pr_debug("map '%s': at sec_idx %d, offset %zu.\n", |
2532 | map_name, map->sec_idx, map->sec_offset); | |
2533 | ||
c7ef5ec9 AN |
2534 | err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def); |
2535 | if (err) | |
2536 | return err; | |
2537 | ||
2538 | fill_map_from_def(map, &map_def); | |
2539 | ||
2540 | if (map_def.pinning == LIBBPF_PIN_BY_NAME) { | |
2541 | err = build_map_pin_path(map, pin_root_path); | |
2542 | if (err) { | |
2543 | pr_warn("map '%s': couldn't build pin path.\n", map->name); | |
2544 | return err; | |
2545 | } | |
2546 | } | |
2547 | ||
2548 | if (map_def.parts & MAP_DEF_INNER_MAP) { | |
2549 | map->inner_map = calloc(1, sizeof(*map->inner_map)); | |
2550 | if (!map->inner_map) | |
2551 | return -ENOMEM; | |
2552 | map->inner_map->fd = -1; | |
2553 | map->inner_map->sec_idx = sec_idx; | |
2554 | map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1); | |
2555 | if (!map->inner_map->name) | |
2556 | return -ENOMEM; | |
2557 | sprintf(map->inner_map->name, "%s.inner", map_name); | |
2558 | ||
2559 | fill_map_from_def(map->inner_map, &inner_def); | |
2560 | } | |
2561 | ||
262cfb74 DK |
2562 | err = bpf_map_find_btf_info(obj, map); |
2563 | if (err) | |
2564 | return err; | |
2565 | ||
c7ef5ec9 | 2566 | return 0; |
41017e56 AN |
2567 | } |
2568 | ||
57a00f41 THJ |
2569 | static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict, |
2570 | const char *pin_root_path) | |
abd29c93 AN |
2571 | { |
2572 | const struct btf_type *sec = NULL; | |
2573 | int nr_types, i, vlen, err; | |
2574 | const struct btf_type *t; | |
2575 | const char *name; | |
2576 | Elf_Data *data; | |
2577 | Elf_Scn *scn; | |
2578 | ||
2579 | if (obj->efile.btf_maps_shndx < 0) | |
2580 | return 0; | |
2581 | ||
88a82120 AN |
2582 | scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx); |
2583 | data = elf_sec_data(obj, scn); | |
abd29c93 | 2584 | if (!scn || !data) { |
88a82120 AN |
2585 | pr_warn("elf: failed to get %s map definitions for %s\n", |
2586 | MAPS_ELF_SEC, obj->path); | |
abd29c93 AN |
2587 | return -EINVAL; |
2588 | } | |
2589 | ||
6a886de0 HC |
2590 | nr_types = btf__type_cnt(obj->btf); |
2591 | for (i = 1; i < nr_types; i++) { | |
abd29c93 | 2592 | t = btf__type_by_id(obj->btf, i); |
b03bc685 | 2593 | if (!btf_is_datasec(t)) |
abd29c93 AN |
2594 | continue; |
2595 | name = btf__name_by_offset(obj->btf, t->name_off); | |
2596 | if (strcmp(name, MAPS_ELF_SEC) == 0) { | |
2597 | sec = t; | |
646f02ff | 2598 | obj->efile.btf_maps_sec_btf_id = i; |
abd29c93 AN |
2599 | break; |
2600 | } | |
2601 | } | |
2602 | ||
2603 | if (!sec) { | |
be18010e | 2604 | pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC); |
abd29c93 AN |
2605 | return -ENOENT; |
2606 | } | |
2607 | ||
b03bc685 | 2608 | vlen = btf_vlen(sec); |
abd29c93 AN |
2609 | for (i = 0; i < vlen; i++) { |
2610 | err = bpf_object__init_user_btf_map(obj, sec, i, | |
2611 | obj->efile.btf_maps_shndx, | |
8983b731 AN |
2612 | data, strict, |
2613 | pin_root_path); | |
abd29c93 AN |
2614 | if (err) |
2615 | return err; | |
2616 | } | |
2617 | ||
2618 | return 0; | |
2619 | } | |
2620 | ||
0d13bfce | 2621 | static int bpf_object__init_maps(struct bpf_object *obj, |
01af3bf0 | 2622 | const struct bpf_object_open_opts *opts) |
bf829271 | 2623 | { |
166750bc AN |
2624 | const char *pin_root_path; |
2625 | bool strict; | |
bf829271 | 2626 | int err; |
8837fe5d | 2627 | |
166750bc AN |
2628 | strict = !OPTS_GET(opts, relaxed_maps, false); |
2629 | pin_root_path = OPTS_GET(opts, pin_root_path, NULL); | |
bf829271 | 2630 | |
166750bc AN |
2631 | err = bpf_object__init_user_maps(obj, strict); |
2632 | err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path); | |
2633 | err = err ?: bpf_object__init_global_data_maps(obj); | |
81bfdd08 | 2634 | err = err ?: bpf_object__init_kconfig_map(obj); |
590a0088 | 2635 | err = err ?: bpf_object__init_struct_ops_maps(obj); |
bf829271 | 2636 | |
3b3af91c | 2637 | return err; |
561bbcca WN |
2638 | } |
2639 | ||
e3d91b0c JDB |
2640 | static bool section_have_execinstr(struct bpf_object *obj, int idx) |
2641 | { | |
ad23b723 | 2642 | Elf64_Shdr *sh; |
e3d91b0c | 2643 | |
ad23b723 AN |
2644 | sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx)); |
2645 | if (!sh) | |
e3d91b0c JDB |
2646 | return false; |
2647 | ||
ad23b723 | 2648 | return sh->sh_flags & SHF_EXECINSTR; |
e3d91b0c JDB |
2649 | } |
2650 | ||
0f0e55d8 AN |
2651 | static bool btf_needs_sanitization(struct bpf_object *obj) |
2652 | { | |
9ca1f56a AS |
2653 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
2654 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
2655 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
2656 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 2657 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
2dc1e488 | 2658 | bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG); |
0f0e55d8 | 2659 | |
2dc1e488 YS |
2660 | return !has_func || !has_datasec || !has_func_global || !has_float || |
2661 | !has_decl_tag || !has_type_tag; | |
0f0e55d8 AN |
2662 | } |
2663 | ||
2664 | static void bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf) | |
d7c4b398 | 2665 | { |
9ca1f56a AS |
2666 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
2667 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
2668 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
2669 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 2670 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
2dc1e488 | 2671 | bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG); |
d7c4b398 AN |
2672 | struct btf_type *t; |
2673 | int i, j, vlen; | |
d7c4b398 | 2674 | |
6a886de0 | 2675 | for (i = 1; i < btf__type_cnt(btf); i++) { |
d7c4b398 | 2676 | t = (struct btf_type *)btf__type_by_id(btf, i); |
d7c4b398 | 2677 | |
223f903e YS |
2678 | if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) { |
2679 | /* replace VAR/DECL_TAG with INT */ | |
d7c4b398 | 2680 | t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0); |
1d4126c4 AN |
2681 | /* |
2682 | * using size = 1 is the safest choice, 4 will be too | |
2683 | * big and cause kernel BTF validation failure if | |
2684 | * original variable took less than 4 bytes | |
2685 | */ | |
2686 | t->size = 1; | |
708852dc | 2687 | *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8); |
b03bc685 | 2688 | } else if (!has_datasec && btf_is_datasec(t)) { |
d7c4b398 | 2689 | /* replace DATASEC with STRUCT */ |
b03bc685 AN |
2690 | const struct btf_var_secinfo *v = btf_var_secinfos(t); |
2691 | struct btf_member *m = btf_members(t); | |
d7c4b398 AN |
2692 | struct btf_type *vt; |
2693 | char *name; | |
2694 | ||
2695 | name = (char *)btf__name_by_offset(btf, t->name_off); | |
2696 | while (*name) { | |
2697 | if (*name == '.') | |
2698 | *name = '_'; | |
2699 | name++; | |
2700 | } | |
2701 | ||
b03bc685 | 2702 | vlen = btf_vlen(t); |
d7c4b398 AN |
2703 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen); |
2704 | for (j = 0; j < vlen; j++, v++, m++) { | |
2705 | /* order of field assignments is important */ | |
2706 | m->offset = v->offset * 8; | |
2707 | m->type = v->type; | |
2708 | /* preserve variable name as member name */ | |
2709 | vt = (void *)btf__type_by_id(btf, v->type); | |
2710 | m->name_off = vt->name_off; | |
2711 | } | |
b03bc685 | 2712 | } else if (!has_func && btf_is_func_proto(t)) { |
d7c4b398 | 2713 | /* replace FUNC_PROTO with ENUM */ |
b03bc685 | 2714 | vlen = btf_vlen(t); |
d7c4b398 AN |
2715 | t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen); |
2716 | t->size = sizeof(__u32); /* kernel enforced */ | |
b03bc685 | 2717 | } else if (!has_func && btf_is_func(t)) { |
d7c4b398 AN |
2718 | /* replace FUNC with TYPEDEF */ |
2719 | t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0); | |
2d3eb67f AS |
2720 | } else if (!has_func_global && btf_is_func(t)) { |
2721 | /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */ | |
2722 | t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0); | |
22541a9e IL |
2723 | } else if (!has_float && btf_is_float(t)) { |
2724 | /* replace FLOAT with an equally-sized empty STRUCT; | |
2725 | * since C compilers do not accept e.g. "float" as a | |
2726 | * valid struct name, make it anonymous | |
2727 | */ | |
2728 | t->name_off = 0; | |
2729 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0); | |
2dc1e488 YS |
2730 | } else if (!has_type_tag && btf_is_type_tag(t)) { |
2731 | /* replace TYPE_TAG with a CONST */ | |
2732 | t->name_off = 0; | |
2733 | t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0); | |
d7c4b398 AN |
2734 | } |
2735 | } | |
2736 | } | |
2737 | ||
b35f14f4 | 2738 | static bool libbpf_needs_btf(const struct bpf_object *obj) |
abd29c93 | 2739 | { |
b35f14f4 AN |
2740 | return obj->efile.btf_maps_shndx >= 0 || |
2741 | obj->efile.st_ops_shndx >= 0 || | |
2742 | obj->nr_extern > 0; | |
2743 | } | |
2744 | ||
2745 | static bool kernel_needs_btf(const struct bpf_object *obj) | |
2746 | { | |
2747 | return obj->efile.st_ops_shndx >= 0; | |
abd29c93 AN |
2748 | } |
2749 | ||
063183bf | 2750 | static int bpf_object__init_btf(struct bpf_object *obj, |
9c6660d0 AN |
2751 | Elf_Data *btf_data, |
2752 | Elf_Data *btf_ext_data) | |
2753 | { | |
b7d7f3e1 | 2754 | int err = -ENOENT; |
9c6660d0 AN |
2755 | |
2756 | if (btf_data) { | |
2757 | obj->btf = btf__new(btf_data->d_buf, btf_data->d_size); | |
e9fc3ce9 AN |
2758 | err = libbpf_get_error(obj->btf); |
2759 | if (err) { | |
b7d7f3e1 | 2760 | obj->btf = NULL; |
e9fc3ce9 | 2761 | pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err); |
9c6660d0 AN |
2762 | goto out; |
2763 | } | |
4c01925f AN |
2764 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
2765 | btf__set_pointer_size(obj->btf, 8); | |
9c6660d0 AN |
2766 | } |
2767 | if (btf_ext_data) { | |
11d5daa8 AN |
2768 | struct btf_ext_info *ext_segs[3]; |
2769 | int seg_num, sec_num; | |
2770 | ||
9c6660d0 AN |
2771 | if (!obj->btf) { |
2772 | pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n", | |
2773 | BTF_EXT_ELF_SEC, BTF_ELF_SEC); | |
2774 | goto out; | |
2775 | } | |
e9fc3ce9 AN |
2776 | obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size); |
2777 | err = libbpf_get_error(obj->btf_ext); | |
2778 | if (err) { | |
2779 | pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n", | |
2780 | BTF_EXT_ELF_SEC, err); | |
9c6660d0 AN |
2781 | obj->btf_ext = NULL; |
2782 | goto out; | |
2783 | } | |
11d5daa8 AN |
2784 | |
2785 | /* setup .BTF.ext to ELF section mapping */ | |
2786 | ext_segs[0] = &obj->btf_ext->func_info; | |
2787 | ext_segs[1] = &obj->btf_ext->line_info; | |
2788 | ext_segs[2] = &obj->btf_ext->core_relo_info; | |
2789 | for (seg_num = 0; seg_num < ARRAY_SIZE(ext_segs); seg_num++) { | |
2790 | struct btf_ext_info *seg = ext_segs[seg_num]; | |
2791 | const struct btf_ext_info_sec *sec; | |
2792 | const char *sec_name; | |
2793 | Elf_Scn *scn; | |
2794 | ||
2795 | if (seg->sec_cnt == 0) | |
2796 | continue; | |
2797 | ||
2798 | seg->sec_idxs = calloc(seg->sec_cnt, sizeof(*seg->sec_idxs)); | |
2799 | if (!seg->sec_idxs) { | |
2800 | err = -ENOMEM; | |
2801 | goto out; | |
2802 | } | |
2803 | ||
2804 | sec_num = 0; | |
2805 | for_each_btf_ext_sec(seg, sec) { | |
2806 | /* preventively increment index to avoid doing | |
2807 | * this before every continue below | |
2808 | */ | |
2809 | sec_num++; | |
2810 | ||
2811 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); | |
2812 | if (str_is_empty(sec_name)) | |
2813 | continue; | |
2814 | scn = elf_sec_by_name(obj, sec_name); | |
2815 | if (!scn) | |
2816 | continue; | |
2817 | ||
2818 | seg->sec_idxs[sec_num - 1] = elf_ndxscn(scn); | |
2819 | } | |
2820 | } | |
9c6660d0 AN |
2821 | } |
2822 | out: | |
b35f14f4 | 2823 | if (err && libbpf_needs_btf(obj)) { |
be18010e | 2824 | pr_warn("BTF is required, but is missing or corrupted.\n"); |
b7d7f3e1 | 2825 | return err; |
abd29c93 | 2826 | } |
9c6660d0 AN |
2827 | return 0; |
2828 | } | |
2829 | ||
b96c07f3 AN |
2830 | static int compare_vsi_off(const void *_a, const void *_b) |
2831 | { | |
2832 | const struct btf_var_secinfo *a = _a; | |
2833 | const struct btf_var_secinfo *b = _b; | |
2834 | ||
2835 | return a->offset - b->offset; | |
2836 | } | |
2837 | ||
2838 | static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf, | |
2839 | struct btf_type *t) | |
2840 | { | |
2841 | __u32 size = 0, off = 0, i, vars = btf_vlen(t); | |
2842 | const char *name = btf__name_by_offset(btf, t->name_off); | |
2843 | const struct btf_type *t_var; | |
2844 | struct btf_var_secinfo *vsi; | |
2845 | const struct btf_var *var; | |
2846 | int ret; | |
2847 | ||
2848 | if (!name) { | |
2849 | pr_debug("No name found in string section for DATASEC kind.\n"); | |
2850 | return -ENOENT; | |
2851 | } | |
2852 | ||
2853 | /* .extern datasec size and var offsets were set correctly during | |
2854 | * extern collection step, so just skip straight to sorting variables | |
2855 | */ | |
2856 | if (t->size) | |
2857 | goto sort_vars; | |
2858 | ||
2859 | ret = find_elf_sec_sz(obj, name, &size); | |
6966d4c4 | 2860 | if (ret || !size) { |
b96c07f3 AN |
2861 | pr_debug("Invalid size for section %s: %u bytes\n", name, size); |
2862 | return -ENOENT; | |
2863 | } | |
2864 | ||
2865 | t->size = size; | |
2866 | ||
2867 | for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) { | |
2868 | t_var = btf__type_by_id(btf, vsi->type); | |
88918dc1 | 2869 | if (!t_var || !btf_is_var(t_var)) { |
b96c07f3 AN |
2870 | pr_debug("Non-VAR type seen in section %s\n", name); |
2871 | return -EINVAL; | |
2872 | } | |
2873 | ||
88918dc1 | 2874 | var = btf_var(t_var); |
b96c07f3 AN |
2875 | if (var->linkage == BTF_VAR_STATIC) |
2876 | continue; | |
2877 | ||
2878 | name = btf__name_by_offset(btf, t_var->name_off); | |
2879 | if (!name) { | |
2880 | pr_debug("No name found in string section for VAR kind\n"); | |
2881 | return -ENOENT; | |
2882 | } | |
2883 | ||
2884 | ret = find_elf_var_offset(obj, name, &off); | |
2885 | if (ret) { | |
2886 | pr_debug("No offset found in symbol table for VAR %s\n", | |
2887 | name); | |
2888 | return -ENOENT; | |
2889 | } | |
2890 | ||
2891 | vsi->offset = off; | |
2892 | } | |
2893 | ||
2894 | sort_vars: | |
2895 | qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off); | |
2896 | return 0; | |
2897 | } | |
2898 | ||
2899 | static int btf_finalize_data(struct bpf_object *obj, struct btf *btf) | |
2900 | { | |
2901 | int err = 0; | |
6a886de0 | 2902 | __u32 i, n = btf__type_cnt(btf); |
b96c07f3 | 2903 | |
6a886de0 | 2904 | for (i = 1; i < n; i++) { |
b96c07f3 AN |
2905 | struct btf_type *t = btf_type_by_id(btf, i); |
2906 | ||
2907 | /* Loader needs to fix up some of the things compiler | |
2908 | * couldn't get its hands on while emitting BTF. This | |
2909 | * is section size and global variable offset. We use | |
2910 | * the info from the ELF itself for this purpose. | |
2911 | */ | |
2912 | if (btf_is_datasec(t)) { | |
2913 | err = btf_fixup_datasec(obj, btf, t); | |
2914 | if (err) | |
2915 | break; | |
2916 | } | |
2917 | } | |
2918 | ||
2919 | return libbpf_err(err); | |
2920 | } | |
2921 | ||
2922 | int btf__finalize_data(struct bpf_object *obj, struct btf *btf) | |
2923 | { | |
2924 | return btf_finalize_data(obj, btf); | |
2925 | } | |
2926 | ||
166750bc AN |
2927 | static int bpf_object__finalize_btf(struct bpf_object *obj) |
2928 | { | |
2929 | int err; | |
2930 | ||
2931 | if (!obj->btf) | |
2932 | return 0; | |
2933 | ||
b96c07f3 | 2934 | err = btf_finalize_data(obj, obj->btf); |
bfc96656 AN |
2935 | if (err) { |
2936 | pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err); | |
2937 | return err; | |
166750bc | 2938 | } |
bfc96656 | 2939 | |
166750bc AN |
2940 | return 0; |
2941 | } | |
2942 | ||
fe62de31 | 2943 | static bool prog_needs_vmlinux_btf(struct bpf_program *prog) |
a6ed02ca | 2944 | { |
1e092a03 KS |
2945 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS || |
2946 | prog->type == BPF_PROG_TYPE_LSM) | |
a6ed02ca KS |
2947 | return true; |
2948 | ||
2949 | /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs | |
2950 | * also need vmlinux BTF | |
2951 | */ | |
2952 | if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd) | |
2953 | return true; | |
2954 | ||
2955 | return false; | |
2956 | } | |
2957 | ||
fe62de31 | 2958 | static bool obj_needs_vmlinux_btf(const struct bpf_object *obj) |
a6ed02ca KS |
2959 | { |
2960 | struct bpf_program *prog; | |
fe62de31 | 2961 | int i; |
a6ed02ca | 2962 | |
1373ff59 SC |
2963 | /* CO-RE relocations need kernel BTF, only when btf_custom_path |
2964 | * is not specified | |
2965 | */ | |
2966 | if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path) | |
fe62de31 | 2967 | return true; |
192b6638 | 2968 | |
d370bbe1 HL |
2969 | /* Support for typed ksyms needs kernel BTF */ |
2970 | for (i = 0; i < obj->nr_extern; i++) { | |
2971 | const struct extern_desc *ext; | |
2972 | ||
2973 | ext = &obj->externs[i]; | |
fe62de31 AN |
2974 | if (ext->type == EXT_KSYM && ext->ksym.type_id) |
2975 | return true; | |
d370bbe1 HL |
2976 | } |
2977 | ||
a6ed02ca | 2978 | bpf_object__for_each_program(prog, obj) { |
a3820c48 | 2979 | if (!prog->autoload) |
d9297581 | 2980 | continue; |
fe62de31 AN |
2981 | if (prog_needs_vmlinux_btf(prog)) |
2982 | return true; | |
a6ed02ca KS |
2983 | } |
2984 | ||
fe62de31 AN |
2985 | return false; |
2986 | } | |
2987 | ||
2988 | static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force) | |
2989 | { | |
2990 | int err; | |
2991 | ||
2992 | /* btf_vmlinux could be loaded earlier */ | |
67234743 | 2993 | if (obj->btf_vmlinux || obj->gen_loader) |
fe62de31 AN |
2994 | return 0; |
2995 | ||
2996 | if (!force && !obj_needs_vmlinux_btf(obj)) | |
192b6638 AN |
2997 | return 0; |
2998 | ||
a710eed3 | 2999 | obj->btf_vmlinux = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
3000 | err = libbpf_get_error(obj->btf_vmlinux); |
3001 | if (err) { | |
192b6638 AN |
3002 | pr_warn("Error loading vmlinux BTF: %d\n", err); |
3003 | obj->btf_vmlinux = NULL; | |
3004 | return err; | |
3005 | } | |
a6ed02ca KS |
3006 | return 0; |
3007 | } | |
3008 | ||
063183bf AN |
3009 | static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj) |
3010 | { | |
0f0e55d8 AN |
3011 | struct btf *kern_btf = obj->btf; |
3012 | bool btf_mandatory, sanitize; | |
aea28a60 | 3013 | int i, err = 0; |
063183bf AN |
3014 | |
3015 | if (!obj->btf) | |
3016 | return 0; | |
3017 | ||
9ca1f56a | 3018 | if (!kernel_supports(obj, FEAT_BTF)) { |
68b08647 AN |
3019 | if (kernel_needs_btf(obj)) { |
3020 | err = -EOPNOTSUPP; | |
3021 | goto report; | |
3022 | } | |
3023 | pr_debug("Kernel doesn't support BTF, skipping uploading it.\n"); | |
3024 | return 0; | |
3025 | } | |
3026 | ||
aea28a60 AN |
3027 | /* Even though some subprogs are global/weak, user might prefer more |
3028 | * permissive BPF verification process that BPF verifier performs for | |
3029 | * static functions, taking into account more context from the caller | |
3030 | * functions. In such case, they need to mark such subprogs with | |
3031 | * __attribute__((visibility("hidden"))) and libbpf will adjust | |
3032 | * corresponding FUNC BTF type to be marked as static and trigger more | |
3033 | * involved BPF verification process. | |
3034 | */ | |
3035 | for (i = 0; i < obj->nr_programs; i++) { | |
3036 | struct bpf_program *prog = &obj->programs[i]; | |
3037 | struct btf_type *t; | |
3038 | const char *name; | |
3039 | int j, n; | |
3040 | ||
3041 | if (!prog->mark_btf_static || !prog_is_subprog(obj, prog)) | |
3042 | continue; | |
3043 | ||
6a886de0 HC |
3044 | n = btf__type_cnt(obj->btf); |
3045 | for (j = 1; j < n; j++) { | |
aea28a60 AN |
3046 | t = btf_type_by_id(obj->btf, j); |
3047 | if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) | |
3048 | continue; | |
3049 | ||
3050 | name = btf__str_by_offset(obj->btf, t->name_off); | |
3051 | if (strcmp(name, prog->name) != 0) | |
3052 | continue; | |
3053 | ||
3054 | t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0); | |
3055 | break; | |
3056 | } | |
3057 | } | |
3058 | ||
0f0e55d8 AN |
3059 | sanitize = btf_needs_sanitization(obj); |
3060 | if (sanitize) { | |
5c3320d7 | 3061 | const void *raw_data; |
0f0e55d8 | 3062 | __u32 sz; |
063183bf | 3063 | |
0f0e55d8 | 3064 | /* clone BTF to sanitize a copy and leave the original intact */ |
6a886de0 | 3065 | raw_data = btf__raw_data(obj->btf, &sz); |
5c3320d7 | 3066 | kern_btf = btf__new(raw_data, sz); |
e9fc3ce9 AN |
3067 | err = libbpf_get_error(kern_btf); |
3068 | if (err) | |
3069 | return err; | |
04efe591 | 3070 | |
4c01925f AN |
3071 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
3072 | btf__set_pointer_size(obj->btf, 8); | |
0f0e55d8 | 3073 | bpf_object__sanitize_btf(obj, kern_btf); |
063183bf | 3074 | } |
0f0e55d8 | 3075 | |
67234743 AS |
3076 | if (obj->gen_loader) { |
3077 | __u32 raw_size = 0; | |
6a886de0 | 3078 | const void *raw_data = btf__raw_data(kern_btf, &raw_size); |
67234743 AS |
3079 | |
3080 | if (!raw_data) | |
3081 | return -ENOMEM; | |
3082 | bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size); | |
3083 | /* Pretend to have valid FD to pass various fd >= 0 checks. | |
3084 | * This fd == 0 will not be used with any syscall and will be reset to -1 eventually. | |
3085 | */ | |
3086 | btf__set_fd(kern_btf, 0); | |
3087 | } else { | |
e0e3ea88 AN |
3088 | /* currently BPF_BTF_LOAD only supports log_level 1 */ |
3089 | err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size, | |
3090 | obj->log_level ? 1 : 0); | |
67234743 | 3091 | } |
0f0e55d8 AN |
3092 | if (sanitize) { |
3093 | if (!err) { | |
3094 | /* move fd to libbpf's BTF */ | |
3095 | btf__set_fd(obj->btf, btf__fd(kern_btf)); | |
3096 | btf__set_fd(kern_btf, -1); | |
3097 | } | |
3098 | btf__free(kern_btf); | |
3099 | } | |
68b08647 | 3100 | report: |
0f0e55d8 AN |
3101 | if (err) { |
3102 | btf_mandatory = kernel_needs_btf(obj); | |
3103 | pr_warn("Error loading .BTF into kernel: %d. %s\n", err, | |
3104 | btf_mandatory ? "BTF is mandatory, can't proceed." | |
3105 | : "BTF is optional, ignoring."); | |
3106 | if (!btf_mandatory) | |
3107 | err = 0; | |
3108 | } | |
3109 | return err; | |
063183bf AN |
3110 | } |
3111 | ||
88a82120 AN |
3112 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off) |
3113 | { | |
3114 | const char *name; | |
3115 | ||
3116 | name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off); | |
3117 | if (!name) { | |
3118 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3119 | off, obj->path, elf_errmsg(-1)); | |
3120 | return NULL; | |
3121 | } | |
3122 | ||
3123 | return name; | |
3124 | } | |
3125 | ||
3126 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off) | |
3127 | { | |
3128 | const char *name; | |
3129 | ||
3130 | name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off); | |
3131 | if (!name) { | |
3132 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3133 | off, obj->path, elf_errmsg(-1)); | |
3134 | return NULL; | |
3135 | } | |
3136 | ||
3137 | return name; | |
3138 | } | |
3139 | ||
3140 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx) | |
3141 | { | |
3142 | Elf_Scn *scn; | |
3143 | ||
3144 | scn = elf_getscn(obj->efile.elf, idx); | |
3145 | if (!scn) { | |
3146 | pr_warn("elf: failed to get section(%zu) from %s: %s\n", | |
3147 | idx, obj->path, elf_errmsg(-1)); | |
3148 | return NULL; | |
3149 | } | |
3150 | return scn; | |
3151 | } | |
3152 | ||
3153 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name) | |
3154 | { | |
3155 | Elf_Scn *scn = NULL; | |
3156 | Elf *elf = obj->efile.elf; | |
3157 | const char *sec_name; | |
3158 | ||
3159 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
3160 | sec_name = elf_sec_name(obj, scn); | |
3161 | if (!sec_name) | |
3162 | return NULL; | |
3163 | ||
3164 | if (strcmp(sec_name, name) != 0) | |
3165 | continue; | |
3166 | ||
3167 | return scn; | |
3168 | } | |
3169 | return NULL; | |
3170 | } | |
3171 | ||
ad23b723 | 3172 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn) |
88a82120 | 3173 | { |
ad23b723 AN |
3174 | Elf64_Shdr *shdr; |
3175 | ||
88a82120 | 3176 | if (!scn) |
ad23b723 | 3177 | return NULL; |
88a82120 | 3178 | |
ad23b723 AN |
3179 | shdr = elf64_getshdr(scn); |
3180 | if (!shdr) { | |
88a82120 AN |
3181 | pr_warn("elf: failed to get section(%zu) header from %s: %s\n", |
3182 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
ad23b723 | 3183 | return NULL; |
88a82120 AN |
3184 | } |
3185 | ||
ad23b723 | 3186 | return shdr; |
88a82120 AN |
3187 | } |
3188 | ||
3189 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn) | |
3190 | { | |
3191 | const char *name; | |
ad23b723 | 3192 | Elf64_Shdr *sh; |
88a82120 AN |
3193 | |
3194 | if (!scn) | |
3195 | return NULL; | |
3196 | ||
ad23b723 AN |
3197 | sh = elf_sec_hdr(obj, scn); |
3198 | if (!sh) | |
88a82120 AN |
3199 | return NULL; |
3200 | ||
ad23b723 | 3201 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 AN |
3202 | if (!name) { |
3203 | pr_warn("elf: failed to get section(%zu) name from %s: %s\n", | |
3204 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
3205 | return NULL; | |
3206 | } | |
3207 | ||
3208 | return name; | |
3209 | } | |
3210 | ||
3211 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn) | |
3212 | { | |
3213 | Elf_Data *data; | |
3214 | ||
3215 | if (!scn) | |
3216 | return NULL; | |
3217 | ||
3218 | data = elf_getdata(scn, 0); | |
3219 | if (!data) { | |
3220 | pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n", | |
3221 | elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>", | |
3222 | obj->path, elf_errmsg(-1)); | |
3223 | return NULL; | |
3224 | } | |
3225 | ||
3226 | return data; | |
3227 | } | |
3228 | ||
ad23b723 AN |
3229 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx) |
3230 | { | |
3231 | if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym)) | |
3232 | return NULL; | |
3233 | ||
3234 | return (Elf64_Sym *)obj->efile.symbols->d_buf + idx; | |
3235 | } | |
3236 | ||
3237 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx) | |
3238 | { | |
3239 | if (idx >= data->d_size / sizeof(Elf64_Rel)) | |
3240 | return NULL; | |
3241 | ||
3242 | return (Elf64_Rel *)data->d_buf + idx; | |
3243 | } | |
3244 | ||
50e09460 AN |
3245 | static bool is_sec_name_dwarf(const char *name) |
3246 | { | |
3247 | /* approximation, but the actual list is too long */ | |
13d35a0c | 3248 | return str_has_pfx(name, ".debug_"); |
50e09460 AN |
3249 | } |
3250 | ||
ad23b723 | 3251 | static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name) |
50e09460 AN |
3252 | { |
3253 | /* no special handling of .strtab */ | |
3254 | if (hdr->sh_type == SHT_STRTAB) | |
3255 | return true; | |
3256 | ||
3257 | /* ignore .llvm_addrsig section as well */ | |
faf6ed32 | 3258 | if (hdr->sh_type == SHT_LLVM_ADDRSIG) |
50e09460 AN |
3259 | return true; |
3260 | ||
3261 | /* no subprograms will lead to an empty .text section, ignore it */ | |
3262 | if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 && | |
3263 | strcmp(name, ".text") == 0) | |
3264 | return true; | |
3265 | ||
3266 | /* DWARF sections */ | |
3267 | if (is_sec_name_dwarf(name)) | |
3268 | return true; | |
3269 | ||
13d35a0c | 3270 | if (str_has_pfx(name, ".rel")) { |
50e09460 AN |
3271 | name += sizeof(".rel") - 1; |
3272 | /* DWARF section relocations */ | |
3273 | if (is_sec_name_dwarf(name)) | |
3274 | return true; | |
3275 | ||
3276 | /* .BTF and .BTF.ext don't need relocations */ | |
3277 | if (strcmp(name, BTF_ELF_SEC) == 0 || | |
3278 | strcmp(name, BTF_EXT_ELF_SEC) == 0) | |
3279 | return true; | |
3280 | } | |
3281 | ||
3282 | return false; | |
3283 | } | |
3284 | ||
db2b8b06 AN |
3285 | static int cmp_progs(const void *_a, const void *_b) |
3286 | { | |
3287 | const struct bpf_program *a = _a; | |
3288 | const struct bpf_program *b = _b; | |
3289 | ||
3290 | if (a->sec_idx != b->sec_idx) | |
3291 | return a->sec_idx < b->sec_idx ? -1 : 1; | |
3292 | ||
3293 | /* sec_insn_off can't be the same within the section */ | |
3294 | return a->sec_insn_off < b->sec_insn_off ? -1 : 1; | |
3295 | } | |
3296 | ||
0d13bfce | 3297 | static int bpf_object__elf_collect(struct bpf_object *obj) |
29603665 | 3298 | { |
25bbbd7a | 3299 | struct elf_sec_desc *sec_desc; |
29603665 | 3300 | Elf *elf = obj->efile.elf; |
f0187f0b | 3301 | Elf_Data *btf_ext_data = NULL; |
1713d68b | 3302 | Elf_Data *btf_data = NULL; |
666810e8 | 3303 | int idx = 0, err = 0; |
0201c575 AN |
3304 | const char *name; |
3305 | Elf_Data *data; | |
3306 | Elf_Scn *scn; | |
ad23b723 | 3307 | Elf64_Shdr *sh; |
29603665 | 3308 | |
0d6988e1 AN |
3309 | /* ELF section indices are 0-based, but sec #0 is special "invalid" |
3310 | * section. e_shnum does include sec #0, so e_shnum is the necessary | |
3311 | * size of an array to keep all the sections. | |
25bbbd7a | 3312 | */ |
0d6988e1 | 3313 | obj->efile.sec_cnt = obj->efile.ehdr->e_shnum; |
25bbbd7a AN |
3314 | obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs)); |
3315 | if (!obj->efile.secs) | |
3316 | return -ENOMEM; | |
29603665 | 3317 | |
0201c575 AN |
3318 | /* a bunch of ELF parsing functionality depends on processing symbols, |
3319 | * so do the first pass and find the symbol table | |
3320 | */ | |
3321 | scn = NULL; | |
29603665 | 3322 | while ((scn = elf_nextscn(elf, scn)) != NULL) { |
ad23b723 AN |
3323 | sh = elf_sec_hdr(obj, scn); |
3324 | if (!sh) | |
0201c575 AN |
3325 | return -LIBBPF_ERRNO__FORMAT; |
3326 | ||
ad23b723 | 3327 | if (sh->sh_type == SHT_SYMTAB) { |
0201c575 AN |
3328 | if (obj->efile.symbols) { |
3329 | pr_warn("elf: multiple symbol tables in %s\n", obj->path); | |
3330 | return -LIBBPF_ERRNO__FORMAT; | |
3331 | } | |
29603665 | 3332 | |
0201c575 AN |
3333 | data = elf_sec_data(obj, scn); |
3334 | if (!data) | |
3335 | return -LIBBPF_ERRNO__FORMAT; | |
3336 | ||
25bbbd7a AN |
3337 | idx = elf_ndxscn(scn); |
3338 | ||
0201c575 | 3339 | obj->efile.symbols = data; |
25bbbd7a | 3340 | obj->efile.symbols_shndx = idx; |
ad23b723 | 3341 | obj->efile.strtabidx = sh->sh_link; |
0201c575 AN |
3342 | } |
3343 | } | |
3344 | ||
03e601f4 THJ |
3345 | if (!obj->efile.symbols) { |
3346 | pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n", | |
3347 | obj->path); | |
3348 | return -ENOENT; | |
3349 | } | |
3350 | ||
0201c575 AN |
3351 | scn = NULL; |
3352 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
25bbbd7a AN |
3353 | idx = elf_ndxscn(scn); |
3354 | sec_desc = &obj->efile.secs[idx]; | |
88a82120 | 3355 | |
ad23b723 AN |
3356 | sh = elf_sec_hdr(obj, scn); |
3357 | if (!sh) | |
01b29d1d | 3358 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3359 | |
ad23b723 | 3360 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 | 3361 | if (!name) |
01b29d1d | 3362 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3363 | |
ad23b723 | 3364 | if (ignore_elf_section(sh, name)) |
50e09460 AN |
3365 | continue; |
3366 | ||
88a82120 AN |
3367 | data = elf_sec_data(obj, scn); |
3368 | if (!data) | |
01b29d1d | 3369 | return -LIBBPF_ERRNO__FORMAT; |
88a82120 AN |
3370 | |
3371 | pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n", | |
077c066a | 3372 | idx, name, (unsigned long)data->d_size, |
ad23b723 AN |
3373 | (int)sh->sh_link, (unsigned long)sh->sh_flags, |
3374 | (int)sh->sh_type); | |
cb1e5e96 | 3375 | |
1713d68b | 3376 | if (strcmp(name, "license") == 0) { |
88a82120 | 3377 | err = bpf_object__init_license(obj, data->d_buf, data->d_size); |
01b29d1d AN |
3378 | if (err) |
3379 | return err; | |
1713d68b | 3380 | } else if (strcmp(name, "version") == 0) { |
88a82120 | 3381 | err = bpf_object__init_kversion(obj, data->d_buf, data->d_size); |
54b8625c JF |
3382 | if (err) |
3383 | return err; | |
1713d68b | 3384 | } else if (strcmp(name, "maps") == 0) { |
666810e8 | 3385 | obj->efile.maps_shndx = idx; |
abd29c93 AN |
3386 | } else if (strcmp(name, MAPS_ELF_SEC) == 0) { |
3387 | obj->efile.btf_maps_shndx = idx; | |
1713d68b | 3388 | } else if (strcmp(name, BTF_ELF_SEC) == 0) { |
62554d52 AN |
3389 | if (sh->sh_type != SHT_PROGBITS) |
3390 | return -LIBBPF_ERRNO__FORMAT; | |
1713d68b | 3391 | btf_data = data; |
2993e051 | 3392 | } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) { |
62554d52 AN |
3393 | if (sh->sh_type != SHT_PROGBITS) |
3394 | return -LIBBPF_ERRNO__FORMAT; | |
f0187f0b | 3395 | btf_ext_data = data; |
ad23b723 | 3396 | } else if (sh->sh_type == SHT_SYMTAB) { |
0201c575 | 3397 | /* already processed during the first pass above */ |
ad23b723 AN |
3398 | } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) { |
3399 | if (sh->sh_flags & SHF_EXECINSTR) { | |
f8c7a4d4 JS |
3400 | if (strcmp(name, ".text") == 0) |
3401 | obj->efile.text_shndx = idx; | |
c1122392 | 3402 | err = bpf_object__add_programs(obj, data, name, idx); |
88a82120 | 3403 | if (err) |
01b29d1d | 3404 | return err; |
aed65917 AN |
3405 | } else if (strcmp(name, DATA_SEC) == 0 || |
3406 | str_has_pfx(name, DATA_SEC ".")) { | |
25bbbd7a AN |
3407 | sec_desc->sec_type = SEC_DATA; |
3408 | sec_desc->shdr = sh; | |
3409 | sec_desc->data = data; | |
aed65917 AN |
3410 | } else if (strcmp(name, RODATA_SEC) == 0 || |
3411 | str_has_pfx(name, RODATA_SEC ".")) { | |
25bbbd7a AN |
3412 | sec_desc->sec_type = SEC_RODATA; |
3413 | sec_desc->shdr = sh; | |
3414 | sec_desc->data = data; | |
590a0088 MKL |
3415 | } else if (strcmp(name, STRUCT_OPS_SEC) == 0) { |
3416 | obj->efile.st_ops_data = data; | |
3417 | obj->efile.st_ops_shndx = idx; | |
d859900c | 3418 | } else { |
50e09460 AN |
3419 | pr_info("elf: skipping unrecognized data section(%d) %s\n", |
3420 | idx, name); | |
a5b8bd47 | 3421 | } |
ad23b723 | 3422 | } else if (sh->sh_type == SHT_REL) { |
25bbbd7a | 3423 | int targ_sec_idx = sh->sh_info; /* points to other section */ |
e3d91b0c | 3424 | |
b7332d28 AN |
3425 | if (sh->sh_entsize != sizeof(Elf64_Rel) || |
3426 | targ_sec_idx >= obj->efile.sec_cnt) | |
3427 | return -LIBBPF_ERRNO__FORMAT; | |
3428 | ||
e3d91b0c | 3429 | /* Only do relo for section with exec instructions */ |
25bbbd7a | 3430 | if (!section_have_execinstr(obj, targ_sec_idx) && |
646f02ff AN |
3431 | strcmp(name, ".rel" STRUCT_OPS_SEC) && |
3432 | strcmp(name, ".rel" MAPS_ELF_SEC)) { | |
50e09460 | 3433 | pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n", |
25bbbd7a AN |
3434 | idx, name, targ_sec_idx, |
3435 | elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>"); | |
e3d91b0c JDB |
3436 | continue; |
3437 | } | |
b62f06e8 | 3438 | |
25bbbd7a AN |
3439 | sec_desc->sec_type = SEC_RELO; |
3440 | sec_desc->shdr = sh; | |
3441 | sec_desc->data = data; | |
ad23b723 | 3442 | } else if (sh->sh_type == SHT_NOBITS && strcmp(name, BSS_SEC) == 0) { |
25bbbd7a AN |
3443 | sec_desc->sec_type = SEC_BSS; |
3444 | sec_desc->shdr = sh; | |
3445 | sec_desc->data = data; | |
077c066a | 3446 | } else { |
2e80be60 | 3447 | pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name, |
ad23b723 | 3448 | (size_t)sh->sh_size); |
bec7d68c | 3449 | } |
29603665 | 3450 | } |
561bbcca | 3451 | |
d3a3aa0c | 3452 | if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) { |
88a82120 | 3453 | pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path); |
f102154d | 3454 | return -LIBBPF_ERRNO__FORMAT; |
77ba9a5b | 3455 | } |
db2b8b06 AN |
3456 | |
3457 | /* sort BPF programs by section name and in-section instruction offset | |
3458 | * for faster search */ | |
2a6a9bf2 AN |
3459 | if (obj->nr_programs) |
3460 | qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs); | |
db2b8b06 | 3461 | |
0d13bfce | 3462 | return bpf_object__init_btf(obj, btf_data, btf_ext_data); |
29603665 WN |
3463 | } |
3464 | ||
ad23b723 | 3465 | static bool sym_is_extern(const Elf64_Sym *sym) |
166750bc | 3466 | { |
ad23b723 | 3467 | int bind = ELF64_ST_BIND(sym->st_info); |
166750bc AN |
3468 | /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */ |
3469 | return sym->st_shndx == SHN_UNDEF && | |
3470 | (bind == STB_GLOBAL || bind == STB_WEAK) && | |
ad23b723 | 3471 | ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE; |
166750bc AN |
3472 | } |
3473 | ||
ad23b723 | 3474 | static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx) |
53eddb5e | 3475 | { |
ad23b723 AN |
3476 | int bind = ELF64_ST_BIND(sym->st_info); |
3477 | int type = ELF64_ST_TYPE(sym->st_info); | |
53eddb5e YS |
3478 | |
3479 | /* in .text section */ | |
3480 | if (sym->st_shndx != text_shndx) | |
3481 | return false; | |
3482 | ||
3483 | /* local function */ | |
3484 | if (bind == STB_LOCAL && type == STT_SECTION) | |
3485 | return true; | |
3486 | ||
3487 | /* global function */ | |
3488 | return bind == STB_GLOBAL && type == STT_FUNC; | |
3489 | } | |
3490 | ||
166750bc AN |
3491 | static int find_extern_btf_id(const struct btf *btf, const char *ext_name) |
3492 | { | |
3493 | const struct btf_type *t; | |
5bd022ec | 3494 | const char *tname; |
166750bc AN |
3495 | int i, n; |
3496 | ||
3497 | if (!btf) | |
3498 | return -ESRCH; | |
3499 | ||
6a886de0 HC |
3500 | n = btf__type_cnt(btf); |
3501 | for (i = 1; i < n; i++) { | |
166750bc AN |
3502 | t = btf__type_by_id(btf, i); |
3503 | ||
5bd022ec | 3504 | if (!btf_is_var(t) && !btf_is_func(t)) |
166750bc AN |
3505 | continue; |
3506 | ||
5bd022ec MKL |
3507 | tname = btf__name_by_offset(btf, t->name_off); |
3508 | if (strcmp(tname, ext_name)) | |
166750bc AN |
3509 | continue; |
3510 | ||
5bd022ec MKL |
3511 | if (btf_is_var(t) && |
3512 | btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN) | |
3513 | return -EINVAL; | |
3514 | ||
3515 | if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN) | |
166750bc AN |
3516 | return -EINVAL; |
3517 | ||
3518 | return i; | |
3519 | } | |
3520 | ||
3521 | return -ENOENT; | |
3522 | } | |
3523 | ||
2e33efe3 AN |
3524 | static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) { |
3525 | const struct btf_var_secinfo *vs; | |
3526 | const struct btf_type *t; | |
3527 | int i, j, n; | |
3528 | ||
3529 | if (!btf) | |
3530 | return -ESRCH; | |
3531 | ||
6a886de0 HC |
3532 | n = btf__type_cnt(btf); |
3533 | for (i = 1; i < n; i++) { | |
2e33efe3 AN |
3534 | t = btf__type_by_id(btf, i); |
3535 | ||
3536 | if (!btf_is_datasec(t)) | |
3537 | continue; | |
3538 | ||
3539 | vs = btf_var_secinfos(t); | |
3540 | for (j = 0; j < btf_vlen(t); j++, vs++) { | |
3541 | if (vs->type == ext_btf_id) | |
3542 | return i; | |
3543 | } | |
3544 | } | |
3545 | ||
3546 | return -ENOENT; | |
3547 | } | |
3548 | ||
3549 | static enum kcfg_type find_kcfg_type(const struct btf *btf, int id, | |
3550 | bool *is_signed) | |
166750bc AN |
3551 | { |
3552 | const struct btf_type *t; | |
3553 | const char *name; | |
3554 | ||
3555 | t = skip_mods_and_typedefs(btf, id, NULL); | |
3556 | name = btf__name_by_offset(btf, t->name_off); | |
3557 | ||
3558 | if (is_signed) | |
3559 | *is_signed = false; | |
3560 | switch (btf_kind(t)) { | |
3561 | case BTF_KIND_INT: { | |
3562 | int enc = btf_int_encoding(t); | |
3563 | ||
3564 | if (enc & BTF_INT_BOOL) | |
2e33efe3 | 3565 | return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN; |
166750bc AN |
3566 | if (is_signed) |
3567 | *is_signed = enc & BTF_INT_SIGNED; | |
3568 | if (t->size == 1) | |
2e33efe3 | 3569 | return KCFG_CHAR; |
166750bc | 3570 | if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1))) |
2e33efe3 AN |
3571 | return KCFG_UNKNOWN; |
3572 | return KCFG_INT; | |
166750bc AN |
3573 | } |
3574 | case BTF_KIND_ENUM: | |
3575 | if (t->size != 4) | |
2e33efe3 | 3576 | return KCFG_UNKNOWN; |
166750bc | 3577 | if (strcmp(name, "libbpf_tristate")) |
2e33efe3 AN |
3578 | return KCFG_UNKNOWN; |
3579 | return KCFG_TRISTATE; | |
166750bc AN |
3580 | case BTF_KIND_ARRAY: |
3581 | if (btf_array(t)->nelems == 0) | |
2e33efe3 AN |
3582 | return KCFG_UNKNOWN; |
3583 | if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR) | |
3584 | return KCFG_UNKNOWN; | |
3585 | return KCFG_CHAR_ARR; | |
166750bc | 3586 | default: |
2e33efe3 | 3587 | return KCFG_UNKNOWN; |
166750bc AN |
3588 | } |
3589 | } | |
3590 | ||
3591 | static int cmp_externs(const void *_a, const void *_b) | |
3592 | { | |
3593 | const struct extern_desc *a = _a; | |
3594 | const struct extern_desc *b = _b; | |
3595 | ||
2e33efe3 AN |
3596 | if (a->type != b->type) |
3597 | return a->type < b->type ? -1 : 1; | |
3598 | ||
3599 | if (a->type == EXT_KCFG) { | |
3600 | /* descending order by alignment requirements */ | |
3601 | if (a->kcfg.align != b->kcfg.align) | |
3602 | return a->kcfg.align > b->kcfg.align ? -1 : 1; | |
3603 | /* ascending order by size, within same alignment class */ | |
3604 | if (a->kcfg.sz != b->kcfg.sz) | |
3605 | return a->kcfg.sz < b->kcfg.sz ? -1 : 1; | |
3606 | } | |
3607 | ||
166750bc AN |
3608 | /* resolve ties by name */ |
3609 | return strcmp(a->name, b->name); | |
3610 | } | |
3611 | ||
1c0c7074 AN |
3612 | static int find_int_btf_id(const struct btf *btf) |
3613 | { | |
3614 | const struct btf_type *t; | |
3615 | int i, n; | |
3616 | ||
6a886de0 HC |
3617 | n = btf__type_cnt(btf); |
3618 | for (i = 1; i < n; i++) { | |
1c0c7074 AN |
3619 | t = btf__type_by_id(btf, i); |
3620 | ||
3621 | if (btf_is_int(t) && btf_int_bits(t) == 32) | |
3622 | return i; | |
3623 | } | |
3624 | ||
3625 | return 0; | |
3626 | } | |
3627 | ||
5bd022ec MKL |
3628 | static int add_dummy_ksym_var(struct btf *btf) |
3629 | { | |
3630 | int i, int_btf_id, sec_btf_id, dummy_var_btf_id; | |
3631 | const struct btf_var_secinfo *vs; | |
3632 | const struct btf_type *sec; | |
3633 | ||
9683e577 IR |
3634 | if (!btf) |
3635 | return 0; | |
3636 | ||
5bd022ec MKL |
3637 | sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC, |
3638 | BTF_KIND_DATASEC); | |
3639 | if (sec_btf_id < 0) | |
3640 | return 0; | |
3641 | ||
3642 | sec = btf__type_by_id(btf, sec_btf_id); | |
3643 | vs = btf_var_secinfos(sec); | |
3644 | for (i = 0; i < btf_vlen(sec); i++, vs++) { | |
3645 | const struct btf_type *vt; | |
3646 | ||
3647 | vt = btf__type_by_id(btf, vs->type); | |
3648 | if (btf_is_func(vt)) | |
3649 | break; | |
3650 | } | |
3651 | ||
3652 | /* No func in ksyms sec. No need to add dummy var. */ | |
3653 | if (i == btf_vlen(sec)) | |
3654 | return 0; | |
3655 | ||
3656 | int_btf_id = find_int_btf_id(btf); | |
3657 | dummy_var_btf_id = btf__add_var(btf, | |
3658 | "dummy_ksym", | |
3659 | BTF_VAR_GLOBAL_ALLOCATED, | |
3660 | int_btf_id); | |
3661 | if (dummy_var_btf_id < 0) | |
3662 | pr_warn("cannot create a dummy_ksym var\n"); | |
3663 | ||
3664 | return dummy_var_btf_id; | |
3665 | } | |
3666 | ||
166750bc AN |
3667 | static int bpf_object__collect_externs(struct bpf_object *obj) |
3668 | { | |
1c0c7074 | 3669 | struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL; |
166750bc AN |
3670 | const struct btf_type *t; |
3671 | struct extern_desc *ext; | |
5bd022ec | 3672 | int i, n, off, dummy_var_btf_id; |
2e33efe3 | 3673 | const char *ext_name, *sec_name; |
166750bc | 3674 | Elf_Scn *scn; |
ad23b723 | 3675 | Elf64_Shdr *sh; |
166750bc AN |
3676 | |
3677 | if (!obj->efile.symbols) | |
3678 | return 0; | |
3679 | ||
88a82120 | 3680 | scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx); |
ad23b723 | 3681 | sh = elf_sec_hdr(obj, scn); |
83390787 | 3682 | if (!sh || sh->sh_entsize != sizeof(Elf64_Sym)) |
166750bc | 3683 | return -LIBBPF_ERRNO__FORMAT; |
166750bc | 3684 | |
5bd022ec MKL |
3685 | dummy_var_btf_id = add_dummy_ksym_var(obj->btf); |
3686 | if (dummy_var_btf_id < 0) | |
3687 | return dummy_var_btf_id; | |
3688 | ||
ad23b723 | 3689 | n = sh->sh_size / sh->sh_entsize; |
166750bc | 3690 | pr_debug("looking for externs among %d symbols...\n", n); |
88a82120 | 3691 | |
166750bc | 3692 | for (i = 0; i < n; i++) { |
ad23b723 | 3693 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
166750bc | 3694 | |
ad23b723 | 3695 | if (!sym) |
166750bc | 3696 | return -LIBBPF_ERRNO__FORMAT; |
ad23b723 | 3697 | if (!sym_is_extern(sym)) |
166750bc | 3698 | continue; |
ad23b723 | 3699 | ext_name = elf_sym_str(obj, sym->st_name); |
166750bc AN |
3700 | if (!ext_name || !ext_name[0]) |
3701 | continue; | |
3702 | ||
3703 | ext = obj->externs; | |
029258d7 | 3704 | ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext)); |
166750bc AN |
3705 | if (!ext) |
3706 | return -ENOMEM; | |
3707 | obj->externs = ext; | |
3708 | ext = &ext[obj->nr_extern]; | |
3709 | memset(ext, 0, sizeof(*ext)); | |
3710 | obj->nr_extern++; | |
3711 | ||
3712 | ext->btf_id = find_extern_btf_id(obj->btf, ext_name); | |
3713 | if (ext->btf_id <= 0) { | |
3714 | pr_warn("failed to find BTF for extern '%s': %d\n", | |
3715 | ext_name, ext->btf_id); | |
3716 | return ext->btf_id; | |
3717 | } | |
3718 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
3719 | ext->name = btf__name_by_offset(obj->btf, t->name_off); | |
3720 | ext->sym_idx = i; | |
ad23b723 | 3721 | ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK; |
2e33efe3 AN |
3722 | |
3723 | ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id); | |
3724 | if (ext->sec_btf_id <= 0) { | |
3725 | pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n", | |
3726 | ext_name, ext->btf_id, ext->sec_btf_id); | |
3727 | return ext->sec_btf_id; | |
166750bc | 3728 | } |
2e33efe3 AN |
3729 | sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id); |
3730 | sec_name = btf__name_by_offset(obj->btf, sec->name_off); | |
3731 | ||
3732 | if (strcmp(sec_name, KCONFIG_SEC) == 0) { | |
5bd022ec MKL |
3733 | if (btf_is_func(t)) { |
3734 | pr_warn("extern function %s is unsupported under %s section\n", | |
3735 | ext->name, KCONFIG_SEC); | |
3736 | return -ENOTSUP; | |
3737 | } | |
2e33efe3 AN |
3738 | kcfg_sec = sec; |
3739 | ext->type = EXT_KCFG; | |
3740 | ext->kcfg.sz = btf__resolve_size(obj->btf, t->type); | |
3741 | if (ext->kcfg.sz <= 0) { | |
3742 | pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n", | |
3743 | ext_name, ext->kcfg.sz); | |
3744 | return ext->kcfg.sz; | |
3745 | } | |
3746 | ext->kcfg.align = btf__align_of(obj->btf, t->type); | |
3747 | if (ext->kcfg.align <= 0) { | |
3748 | pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n", | |
3749 | ext_name, ext->kcfg.align); | |
3750 | return -EINVAL; | |
3751 | } | |
3752 | ext->kcfg.type = find_kcfg_type(obj->btf, t->type, | |
3753 | &ext->kcfg.is_signed); | |
3754 | if (ext->kcfg.type == KCFG_UNKNOWN) { | |
3755 | pr_warn("extern (kcfg) '%s' type is unsupported\n", ext_name); | |
3756 | return -ENOTSUP; | |
3757 | } | |
1c0c7074 | 3758 | } else if (strcmp(sec_name, KSYMS_SEC) == 0) { |
1c0c7074 AN |
3759 | ksym_sec = sec; |
3760 | ext->type = EXT_KSYM; | |
d370bbe1 HL |
3761 | skip_mods_and_typedefs(obj->btf, t->type, |
3762 | &ext->ksym.type_id); | |
2e33efe3 AN |
3763 | } else { |
3764 | pr_warn("unrecognized extern section '%s'\n", sec_name); | |
166750bc AN |
3765 | return -ENOTSUP; |
3766 | } | |
3767 | } | |
3768 | pr_debug("collected %d externs total\n", obj->nr_extern); | |
3769 | ||
3770 | if (!obj->nr_extern) | |
3771 | return 0; | |
3772 | ||
2e33efe3 | 3773 | /* sort externs by type, for kcfg ones also by (align, size, name) */ |
166750bc | 3774 | qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs); |
166750bc | 3775 | |
1c0c7074 AN |
3776 | /* for .ksyms section, we need to turn all externs into allocated |
3777 | * variables in BTF to pass kernel verification; we do this by | |
3778 | * pretending that each extern is a 8-byte variable | |
3779 | */ | |
3780 | if (ksym_sec) { | |
3781 | /* find existing 4-byte integer type in BTF to use for fake | |
3782 | * extern variables in DATASEC | |
3783 | */ | |
3784 | int int_btf_id = find_int_btf_id(obj->btf); | |
5bd022ec MKL |
3785 | /* For extern function, a dummy_var added earlier |
3786 | * will be used to replace the vs->type and | |
3787 | * its name string will be used to refill | |
3788 | * the missing param's name. | |
3789 | */ | |
3790 | const struct btf_type *dummy_var; | |
1c0c7074 | 3791 | |
5bd022ec | 3792 | dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id); |
1c0c7074 AN |
3793 | for (i = 0; i < obj->nr_extern; i++) { |
3794 | ext = &obj->externs[i]; | |
3795 | if (ext->type != EXT_KSYM) | |
3796 | continue; | |
3797 | pr_debug("extern (ksym) #%d: symbol %d, name %s\n", | |
3798 | i, ext->sym_idx, ext->name); | |
3799 | } | |
3800 | ||
3801 | sec = ksym_sec; | |
3802 | n = btf_vlen(sec); | |
3803 | for (i = 0, off = 0; i < n; i++, off += sizeof(int)) { | |
3804 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
3805 | struct btf_type *vt; | |
3806 | ||
3807 | vt = (void *)btf__type_by_id(obj->btf, vs->type); | |
3808 | ext_name = btf__name_by_offset(obj->btf, vt->name_off); | |
3809 | ext = find_extern_by_name(obj, ext_name); | |
3810 | if (!ext) { | |
5bd022ec MKL |
3811 | pr_warn("failed to find extern definition for BTF %s '%s'\n", |
3812 | btf_kind_str(vt), ext_name); | |
1c0c7074 AN |
3813 | return -ESRCH; |
3814 | } | |
5bd022ec MKL |
3815 | if (btf_is_func(vt)) { |
3816 | const struct btf_type *func_proto; | |
3817 | struct btf_param *param; | |
3818 | int j; | |
3819 | ||
3820 | func_proto = btf__type_by_id(obj->btf, | |
3821 | vt->type); | |
3822 | param = btf_params(func_proto); | |
3823 | /* Reuse the dummy_var string if the | |
3824 | * func proto does not have param name. | |
3825 | */ | |
3826 | for (j = 0; j < btf_vlen(func_proto); j++) | |
3827 | if (param[j].type && !param[j].name_off) | |
3828 | param[j].name_off = | |
3829 | dummy_var->name_off; | |
3830 | vs->type = dummy_var_btf_id; | |
3831 | vt->info &= ~0xffff; | |
3832 | vt->info |= BTF_FUNC_GLOBAL; | |
3833 | } else { | |
3834 | btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
3835 | vt->type = int_btf_id; | |
3836 | } | |
1c0c7074 AN |
3837 | vs->offset = off; |
3838 | vs->size = sizeof(int); | |
3839 | } | |
3840 | sec->size = off; | |
3841 | } | |
3842 | ||
2e33efe3 AN |
3843 | if (kcfg_sec) { |
3844 | sec = kcfg_sec; | |
3845 | /* for kcfg externs calculate their offsets within a .kconfig map */ | |
3846 | off = 0; | |
3847 | for (i = 0; i < obj->nr_extern; i++) { | |
3848 | ext = &obj->externs[i]; | |
3849 | if (ext->type != EXT_KCFG) | |
3850 | continue; | |
166750bc | 3851 | |
2e33efe3 AN |
3852 | ext->kcfg.data_off = roundup(off, ext->kcfg.align); |
3853 | off = ext->kcfg.data_off + ext->kcfg.sz; | |
1c0c7074 | 3854 | pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n", |
2e33efe3 AN |
3855 | i, ext->sym_idx, ext->kcfg.data_off, ext->name); |
3856 | } | |
3857 | sec->size = off; | |
3858 | n = btf_vlen(sec); | |
3859 | for (i = 0; i < n; i++) { | |
3860 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
3861 | ||
3862 | t = btf__type_by_id(obj->btf, vs->type); | |
3863 | ext_name = btf__name_by_offset(obj->btf, t->name_off); | |
3864 | ext = find_extern_by_name(obj, ext_name); | |
3865 | if (!ext) { | |
3866 | pr_warn("failed to find extern definition for BTF var '%s'\n", | |
3867 | ext_name); | |
3868 | return -ESRCH; | |
3869 | } | |
3870 | btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
3871 | vs->offset = ext->kcfg.data_off; | |
166750bc | 3872 | } |
166750bc | 3873 | } |
166750bc AN |
3874 | return 0; |
3875 | } | |
3876 | ||
6d4b198b | 3877 | struct bpf_program * |
a324aae3 AN |
3878 | bpf_object__find_program_by_title(const struct bpf_object *obj, |
3879 | const char *title) | |
6d4b198b JK |
3880 | { |
3881 | struct bpf_program *pos; | |
3882 | ||
3883 | bpf_object__for_each_program(pos, obj) { | |
52109584 | 3884 | if (pos->sec_name && !strcmp(pos->sec_name, title)) |
6d4b198b JK |
3885 | return pos; |
3886 | } | |
e9fc3ce9 | 3887 | return errno = ENOENT, NULL; |
6d4b198b JK |
3888 | } |
3889 | ||
c3c55696 AN |
3890 | static bool prog_is_subprog(const struct bpf_object *obj, |
3891 | const struct bpf_program *prog) | |
3892 | { | |
197afc63 AN |
3893 | /* For legacy reasons, libbpf supports an entry-point BPF programs |
3894 | * without SEC() attribute, i.e., those in the .text section. But if | |
3895 | * there are 2 or more such programs in the .text section, they all | |
3896 | * must be subprograms called from entry-point BPF programs in | |
3897 | * designated SEC()'tions, otherwise there is no way to distinguish | |
3898 | * which of those programs should be loaded vs which are a subprogram. | |
3899 | * Similarly, if there is a function/program in .text and at least one | |
3900 | * other BPF program with custom SEC() attribute, then we just assume | |
3901 | * .text programs are subprograms (even if they are not called from | |
3902 | * other programs), because libbpf never explicitly supported mixing | |
3903 | * SEC()-designated BPF programs and .text entry-point BPF programs. | |
bc380eb9 DK |
3904 | * |
3905 | * In libbpf 1.0 strict mode, we always consider .text | |
3906 | * programs to be subprograms. | |
197afc63 | 3907 | */ |
bc380eb9 DK |
3908 | |
3909 | if (libbpf_mode & LIBBPF_STRICT_SEC_NAME) | |
3910 | return prog->sec_idx == obj->efile.text_shndx; | |
3911 | ||
197afc63 | 3912 | return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1; |
c3c55696 AN |
3913 | } |
3914 | ||
01af3bf0 AN |
3915 | struct bpf_program * |
3916 | bpf_object__find_program_by_name(const struct bpf_object *obj, | |
3917 | const char *name) | |
3918 | { | |
3919 | struct bpf_program *prog; | |
3920 | ||
3921 | bpf_object__for_each_program(prog, obj) { | |
c3c55696 AN |
3922 | if (prog_is_subprog(obj, prog)) |
3923 | continue; | |
01af3bf0 AN |
3924 | if (!strcmp(prog->name, name)) |
3925 | return prog; | |
3926 | } | |
e9fc3ce9 | 3927 | return errno = ENOENT, NULL; |
01af3bf0 AN |
3928 | } |
3929 | ||
d859900c DB |
3930 | static bool bpf_object__shndx_is_data(const struct bpf_object *obj, |
3931 | int shndx) | |
3932 | { | |
25bbbd7a AN |
3933 | switch (obj->efile.secs[shndx].sec_type) { |
3934 | case SEC_BSS: | |
3935 | case SEC_DATA: | |
3936 | case SEC_RODATA: | |
3937 | return true; | |
3938 | default: | |
3939 | return false; | |
3940 | } | |
d859900c DB |
3941 | } |
3942 | ||
3943 | static bool bpf_object__shndx_is_maps(const struct bpf_object *obj, | |
3944 | int shndx) | |
3945 | { | |
abd29c93 AN |
3946 | return shndx == obj->efile.maps_shndx || |
3947 | shndx == obj->efile.btf_maps_shndx; | |
d859900c DB |
3948 | } |
3949 | ||
d859900c DB |
3950 | static enum libbpf_map_type |
3951 | bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx) | |
3952 | { | |
25bbbd7a AN |
3953 | if (shndx == obj->efile.symbols_shndx) |
3954 | return LIBBPF_MAP_KCONFIG; | |
3955 | ||
3956 | switch (obj->efile.secs[shndx].sec_type) { | |
3957 | case SEC_BSS: | |
d859900c | 3958 | return LIBBPF_MAP_BSS; |
25bbbd7a AN |
3959 | case SEC_DATA: |
3960 | return LIBBPF_MAP_DATA; | |
3961 | case SEC_RODATA: | |
d859900c | 3962 | return LIBBPF_MAP_RODATA; |
25bbbd7a | 3963 | default: |
d859900c | 3964 | return LIBBPF_MAP_UNSPEC; |
25bbbd7a | 3965 | } |
d859900c DB |
3966 | } |
3967 | ||
1f8e2bcb AN |
3968 | static int bpf_program__record_reloc(struct bpf_program *prog, |
3969 | struct reloc_desc *reloc_desc, | |
9c0f8cbd | 3970 | __u32 insn_idx, const char *sym_name, |
ad23b723 | 3971 | const Elf64_Sym *sym, const Elf64_Rel *rel) |
1f8e2bcb AN |
3972 | { |
3973 | struct bpf_insn *insn = &prog->insns[insn_idx]; | |
3974 | size_t map_idx, nr_maps = prog->obj->nr_maps; | |
3975 | struct bpf_object *obj = prog->obj; | |
3976 | __u32 shdr_idx = sym->st_shndx; | |
3977 | enum libbpf_map_type type; | |
9c0f8cbd | 3978 | const char *sym_sec_name; |
1f8e2bcb AN |
3979 | struct bpf_map *map; |
3980 | ||
aa0b8d43 | 3981 | if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) { |
9c0f8cbd AN |
3982 | pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n", |
3983 | prog->name, sym_name, insn_idx, insn->code); | |
1f8e2bcb AN |
3984 | return -LIBBPF_ERRNO__RELOC; |
3985 | } | |
166750bc AN |
3986 | |
3987 | if (sym_is_extern(sym)) { | |
ad23b723 | 3988 | int sym_idx = ELF64_R_SYM(rel->r_info); |
166750bc AN |
3989 | int i, n = obj->nr_extern; |
3990 | struct extern_desc *ext; | |
3991 | ||
3992 | for (i = 0; i < n; i++) { | |
3993 | ext = &obj->externs[i]; | |
3994 | if (ext->sym_idx == sym_idx) | |
3995 | break; | |
3996 | } | |
3997 | if (i >= n) { | |
9c0f8cbd AN |
3998 | pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n", |
3999 | prog->name, sym_name, sym_idx); | |
166750bc AN |
4000 | return -LIBBPF_ERRNO__RELOC; |
4001 | } | |
9c0f8cbd AN |
4002 | pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n", |
4003 | prog->name, i, ext->name, ext->sym_idx, insn_idx); | |
5bd022ec MKL |
4004 | if (insn->code == (BPF_JMP | BPF_CALL)) |
4005 | reloc_desc->type = RELO_EXTERN_FUNC; | |
4006 | else | |
4007 | reloc_desc->type = RELO_EXTERN_VAR; | |
166750bc | 4008 | reloc_desc->insn_idx = insn_idx; |
2e33efe3 | 4009 | reloc_desc->sym_off = i; /* sym_off stores extern index */ |
166750bc AN |
4010 | return 0; |
4011 | } | |
4012 | ||
aa0b8d43 MKL |
4013 | /* sub-program call relocation */ |
4014 | if (is_call_insn(insn)) { | |
4015 | if (insn->src_reg != BPF_PSEUDO_CALL) { | |
4016 | pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name); | |
4017 | return -LIBBPF_ERRNO__RELOC; | |
4018 | } | |
4019 | /* text_shndx can be 0, if no default "main" program exists */ | |
4020 | if (!shdr_idx || shdr_idx != obj->efile.text_shndx) { | |
4021 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); | |
4022 | pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n", | |
4023 | prog->name, sym_name, sym_sec_name); | |
4024 | return -LIBBPF_ERRNO__RELOC; | |
4025 | } | |
4026 | if (sym->st_value % BPF_INSN_SZ) { | |
4027 | pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n", | |
4028 | prog->name, sym_name, (size_t)sym->st_value); | |
4029 | return -LIBBPF_ERRNO__RELOC; | |
4030 | } | |
4031 | reloc_desc->type = RELO_CALL; | |
4032 | reloc_desc->insn_idx = insn_idx; | |
4033 | reloc_desc->sym_off = sym->st_value; | |
4034 | return 0; | |
4035 | } | |
4036 | ||
1f8e2bcb | 4037 | if (!shdr_idx || shdr_idx >= SHN_LORESERVE) { |
9c0f8cbd AN |
4038 | pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n", |
4039 | prog->name, sym_name, shdr_idx); | |
1f8e2bcb AN |
4040 | return -LIBBPF_ERRNO__RELOC; |
4041 | } | |
4042 | ||
53eddb5e YS |
4043 | /* loading subprog addresses */ |
4044 | if (sym_is_subprog(sym, obj->efile.text_shndx)) { | |
4045 | /* global_func: sym->st_value = offset in the section, insn->imm = 0. | |
4046 | * local_func: sym->st_value = 0, insn->imm = offset in the section. | |
4047 | */ | |
4048 | if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) { | |
4049 | pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n", | |
4050 | prog->name, sym_name, (size_t)sym->st_value, insn->imm); | |
4051 | return -LIBBPF_ERRNO__RELOC; | |
4052 | } | |
4053 | ||
4054 | reloc_desc->type = RELO_SUBPROG_ADDR; | |
4055 | reloc_desc->insn_idx = insn_idx; | |
4056 | reloc_desc->sym_off = sym->st_value; | |
4057 | return 0; | |
4058 | } | |
4059 | ||
1f8e2bcb | 4060 | type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx); |
9c0f8cbd | 4061 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); |
1f8e2bcb AN |
4062 | |
4063 | /* generic map reference relocation */ | |
4064 | if (type == LIBBPF_MAP_UNSPEC) { | |
4065 | if (!bpf_object__shndx_is_maps(obj, shdr_idx)) { | |
9c0f8cbd AN |
4066 | pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n", |
4067 | prog->name, sym_name, sym_sec_name); | |
1f8e2bcb AN |
4068 | return -LIBBPF_ERRNO__RELOC; |
4069 | } | |
4070 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { | |
4071 | map = &obj->maps[map_idx]; | |
4072 | if (map->libbpf_type != type || | |
4073 | map->sec_idx != sym->st_shndx || | |
4074 | map->sec_offset != sym->st_value) | |
4075 | continue; | |
9c0f8cbd AN |
4076 | pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n", |
4077 | prog->name, map_idx, map->name, map->sec_idx, | |
1f8e2bcb AN |
4078 | map->sec_offset, insn_idx); |
4079 | break; | |
4080 | } | |
4081 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
4082 | pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n", |
4083 | prog->name, sym_sec_name, (size_t)sym->st_value); | |
1f8e2bcb AN |
4084 | return -LIBBPF_ERRNO__RELOC; |
4085 | } | |
4086 | reloc_desc->type = RELO_LD64; | |
4087 | reloc_desc->insn_idx = insn_idx; | |
4088 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 4089 | reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */ |
1f8e2bcb AN |
4090 | return 0; |
4091 | } | |
4092 | ||
4093 | /* global data map relocation */ | |
4094 | if (!bpf_object__shndx_is_data(obj, shdr_idx)) { | |
9c0f8cbd AN |
4095 | pr_warn("prog '%s': bad data relo against section '%s'\n", |
4096 | prog->name, sym_sec_name); | |
1f8e2bcb | 4097 | return -LIBBPF_ERRNO__RELOC; |
1f8e2bcb | 4098 | } |
1f8e2bcb AN |
4099 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { |
4100 | map = &obj->maps[map_idx]; | |
25bbbd7a | 4101 | if (map->libbpf_type != type || map->sec_idx != sym->st_shndx) |
1f8e2bcb | 4102 | continue; |
9c0f8cbd AN |
4103 | pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n", |
4104 | prog->name, map_idx, map->name, map->sec_idx, | |
4105 | map->sec_offset, insn_idx); | |
1f8e2bcb AN |
4106 | break; |
4107 | } | |
4108 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
4109 | pr_warn("prog '%s': data relo failed to find map for section '%s'\n", |
4110 | prog->name, sym_sec_name); | |
1f8e2bcb AN |
4111 | return -LIBBPF_ERRNO__RELOC; |
4112 | } | |
4113 | ||
4114 | reloc_desc->type = RELO_DATA; | |
4115 | reloc_desc->insn_idx = insn_idx; | |
4116 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 4117 | reloc_desc->sym_off = sym->st_value; |
1f8e2bcb AN |
4118 | return 0; |
4119 | } | |
4120 | ||
db2b8b06 AN |
4121 | static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx) |
4122 | { | |
4123 | return insn_idx >= prog->sec_insn_off && | |
4124 | insn_idx < prog->sec_insn_off + prog->sec_insn_cnt; | |
4125 | } | |
4126 | ||
4127 | static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj, | |
4128 | size_t sec_idx, size_t insn_idx) | |
4129 | { | |
4130 | int l = 0, r = obj->nr_programs - 1, m; | |
4131 | struct bpf_program *prog; | |
4132 | ||
4133 | while (l < r) { | |
4134 | m = l + (r - l + 1) / 2; | |
4135 | prog = &obj->programs[m]; | |
4136 | ||
4137 | if (prog->sec_idx < sec_idx || | |
4138 | (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx)) | |
4139 | l = m; | |
4140 | else | |
4141 | r = m - 1; | |
4142 | } | |
4143 | /* matching program could be at index l, but it still might be the | |
4144 | * wrong one, so we need to double check conditions for the last time | |
4145 | */ | |
4146 | prog = &obj->programs[l]; | |
4147 | if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx)) | |
4148 | return prog; | |
4149 | return NULL; | |
4150 | } | |
4151 | ||
34090915 | 4152 | static int |
ad23b723 | 4153 | bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data) |
34090915 | 4154 | { |
9c0f8cbd | 4155 | const char *relo_sec_name, *sec_name; |
b7332d28 | 4156 | size_t sec_idx = shdr->sh_info, sym_idx; |
c3c55696 AN |
4157 | struct bpf_program *prog; |
4158 | struct reloc_desc *relos; | |
1f8e2bcb | 4159 | int err, i, nrels; |
c3c55696 AN |
4160 | const char *sym_name; |
4161 | __u32 insn_idx; | |
6245947c AN |
4162 | Elf_Scn *scn; |
4163 | Elf_Data *scn_data; | |
ad23b723 AN |
4164 | Elf64_Sym *sym; |
4165 | Elf64_Rel *rel; | |
34090915 | 4166 | |
b7332d28 AN |
4167 | if (sec_idx >= obj->efile.sec_cnt) |
4168 | return -EINVAL; | |
4169 | ||
6245947c AN |
4170 | scn = elf_sec_by_idx(obj, sec_idx); |
4171 | scn_data = elf_sec_data(obj, scn); | |
4172 | ||
9c0f8cbd | 4173 | relo_sec_name = elf_sec_str(obj, shdr->sh_name); |
6245947c | 4174 | sec_name = elf_sec_name(obj, scn); |
9c0f8cbd AN |
4175 | if (!relo_sec_name || !sec_name) |
4176 | return -EINVAL; | |
4177 | ||
4178 | pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n", | |
4179 | relo_sec_name, sec_idx, sec_name); | |
34090915 WN |
4180 | nrels = shdr->sh_size / shdr->sh_entsize; |
4181 | ||
34090915 | 4182 | for (i = 0; i < nrels; i++) { |
ad23b723 AN |
4183 | rel = elf_rel_by_idx(data, i); |
4184 | if (!rel) { | |
9c0f8cbd | 4185 | pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i); |
6371ca3b | 4186 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4187 | } |
ad23b723 | 4188 | |
b7332d28 AN |
4189 | sym_idx = ELF64_R_SYM(rel->r_info); |
4190 | sym = elf_sym_by_idx(obj, sym_idx); | |
ad23b723 | 4191 | if (!sym) { |
b7332d28 AN |
4192 | pr_warn("sec '%s': symbol #%zu not found for relo #%d\n", |
4193 | relo_sec_name, sym_idx, i); | |
4194 | return -LIBBPF_ERRNO__FORMAT; | |
4195 | } | |
4196 | ||
4197 | if (sym->st_shndx >= obj->efile.sec_cnt) { | |
4198 | pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n", | |
4199 | relo_sec_name, sym_idx, (size_t)sym->st_shndx, i); | |
6371ca3b | 4200 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4201 | } |
6245947c | 4202 | |
ad23b723 | 4203 | if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) { |
9c0f8cbd | 4204 | pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n", |
b7332d28 | 4205 | relo_sec_name, (size_t)rel->r_offset, i); |
1f8e2bcb | 4206 | return -LIBBPF_ERRNO__FORMAT; |
9c0f8cbd | 4207 | } |
d859900c | 4208 | |
ad23b723 | 4209 | insn_idx = rel->r_offset / BPF_INSN_SZ; |
c3c55696 AN |
4210 | /* relocations against static functions are recorded as |
4211 | * relocations against the section that contains a function; | |
4212 | * in such case, symbol will be STT_SECTION and sym.st_name | |
4213 | * will point to empty string (0), so fetch section name | |
4214 | * instead | |
4215 | */ | |
ad23b723 AN |
4216 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0) |
4217 | sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx)); | |
c3c55696 | 4218 | else |
ad23b723 | 4219 | sym_name = elf_sym_str(obj, sym->st_name); |
c3c55696 | 4220 | sym_name = sym_name ?: "<?"; |
d859900c | 4221 | |
9c0f8cbd AN |
4222 | pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n", |
4223 | relo_sec_name, i, insn_idx, sym_name); | |
666810e8 | 4224 | |
c3c55696 AN |
4225 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); |
4226 | if (!prog) { | |
6245947c | 4227 | pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n", |
c3c55696 | 4228 | relo_sec_name, i, sec_name, insn_idx); |
6245947c | 4229 | continue; |
c3c55696 AN |
4230 | } |
4231 | ||
4232 | relos = libbpf_reallocarray(prog->reloc_desc, | |
4233 | prog->nr_reloc + 1, sizeof(*relos)); | |
4234 | if (!relos) | |
4235 | return -ENOMEM; | |
4236 | prog->reloc_desc = relos; | |
4237 | ||
4238 | /* adjust insn_idx to local BPF program frame of reference */ | |
4239 | insn_idx -= prog->sec_insn_off; | |
4240 | err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc], | |
ad23b723 | 4241 | insn_idx, sym_name, sym, rel); |
1f8e2bcb AN |
4242 | if (err) |
4243 | return err; | |
c3c55696 AN |
4244 | |
4245 | prog->nr_reloc++; | |
34090915 WN |
4246 | } |
4247 | return 0; | |
4248 | } | |
4249 | ||
abd29c93 | 4250 | static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map) |
8a138aed MKL |
4251 | { |
4252 | struct bpf_map_def *def = &map->def; | |
d859900c | 4253 | __u32 key_type_id = 0, value_type_id = 0; |
96408c43 | 4254 | int ret; |
8a138aed | 4255 | |
a8fee962 AN |
4256 | if (!obj->btf) |
4257 | return -ENOENT; | |
4258 | ||
590a0088 MKL |
4259 | /* if it's BTF-defined map, we don't need to search for type IDs. |
4260 | * For struct_ops map, it does not need btf_key_type_id and | |
4261 | * btf_value_type_id. | |
4262 | */ | |
4263 | if (map->sec_idx == obj->efile.btf_maps_shndx || | |
4264 | bpf_map__is_struct_ops(map)) | |
abd29c93 AN |
4265 | return 0; |
4266 | ||
d859900c | 4267 | if (!bpf_map__is_internal(map)) { |
93b8952d | 4268 | pr_warn("Use of BPF_ANNOTATE_KV_PAIR is deprecated, use BTF-defined maps in .maps section instead\n"); |
227a0713 AN |
4269 | #pragma GCC diagnostic push |
4270 | #pragma GCC diagnostic ignored "-Wdeprecated-declarations" | |
abd29c93 | 4271 | ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size, |
d859900c DB |
4272 | def->value_size, &key_type_id, |
4273 | &value_type_id); | |
227a0713 | 4274 | #pragma GCC diagnostic pop |
d859900c DB |
4275 | } else { |
4276 | /* | |
4277 | * LLVM annotates global data differently in BTF, that is, | |
4278 | * only as '.data', '.bss' or '.rodata'. | |
4279 | */ | |
aed65917 | 4280 | ret = btf__find_by_name(obj->btf, map->real_name); |
d859900c DB |
4281 | } |
4282 | if (ret < 0) | |
96408c43 | 4283 | return ret; |
8a138aed | 4284 | |
96408c43 | 4285 | map->btf_key_type_id = key_type_id; |
d859900c DB |
4286 | map->btf_value_type_id = bpf_map__is_internal(map) ? |
4287 | ret : value_type_id; | |
8a138aed MKL |
4288 | return 0; |
4289 | } | |
4290 | ||
97eb3138 MP |
4291 | static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info) |
4292 | { | |
4293 | char file[PATH_MAX], buff[4096]; | |
4294 | FILE *fp; | |
4295 | __u32 val; | |
4296 | int err; | |
4297 | ||
4298 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
4299 | memset(info, 0, sizeof(*info)); | |
4300 | ||
4301 | fp = fopen(file, "r"); | |
4302 | if (!fp) { | |
4303 | err = -errno; | |
4304 | pr_warn("failed to open %s: %d. No procfs support?\n", file, | |
4305 | err); | |
4306 | return err; | |
4307 | } | |
4308 | ||
4309 | while (fgets(buff, sizeof(buff), fp)) { | |
4310 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
4311 | info->type = val; | |
4312 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) | |
4313 | info->key_size = val; | |
4314 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) | |
4315 | info->value_size = val; | |
4316 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) | |
4317 | info->max_entries = val; | |
4318 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) | |
4319 | info->map_flags = val; | |
4320 | } | |
4321 | ||
4322 | fclose(fp); | |
4323 | ||
4324 | return 0; | |
4325 | } | |
4326 | ||
26736eb9 JK |
4327 | int bpf_map__reuse_fd(struct bpf_map *map, int fd) |
4328 | { | |
4329 | struct bpf_map_info info = {}; | |
4330 | __u32 len = sizeof(info); | |
4331 | int new_fd, err; | |
4332 | char *new_name; | |
4333 | ||
4334 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
97eb3138 MP |
4335 | if (err && errno == EINVAL) |
4336 | err = bpf_get_map_info_from_fdinfo(fd, &info); | |
26736eb9 | 4337 | if (err) |
e9fc3ce9 | 4338 | return libbpf_err(err); |
26736eb9 JK |
4339 | |
4340 | new_name = strdup(info.name); | |
4341 | if (!new_name) | |
e9fc3ce9 | 4342 | return libbpf_err(-errno); |
26736eb9 JK |
4343 | |
4344 | new_fd = open("/", O_RDONLY | O_CLOEXEC); | |
d1b4574a THJ |
4345 | if (new_fd < 0) { |
4346 | err = -errno; | |
26736eb9 | 4347 | goto err_free_new_name; |
d1b4574a | 4348 | } |
26736eb9 JK |
4349 | |
4350 | new_fd = dup3(fd, new_fd, O_CLOEXEC); | |
d1b4574a THJ |
4351 | if (new_fd < 0) { |
4352 | err = -errno; | |
26736eb9 | 4353 | goto err_close_new_fd; |
d1b4574a | 4354 | } |
26736eb9 JK |
4355 | |
4356 | err = zclose(map->fd); | |
d1b4574a THJ |
4357 | if (err) { |
4358 | err = -errno; | |
26736eb9 | 4359 | goto err_close_new_fd; |
d1b4574a | 4360 | } |
26736eb9 JK |
4361 | free(map->name); |
4362 | ||
4363 | map->fd = new_fd; | |
4364 | map->name = new_name; | |
4365 | map->def.type = info.type; | |
4366 | map->def.key_size = info.key_size; | |
4367 | map->def.value_size = info.value_size; | |
4368 | map->def.max_entries = info.max_entries; | |
4369 | map->def.map_flags = info.map_flags; | |
4370 | map->btf_key_type_id = info.btf_key_type_id; | |
4371 | map->btf_value_type_id = info.btf_value_type_id; | |
ec6d5f47 | 4372 | map->reused = true; |
47512102 | 4373 | map->map_extra = info.map_extra; |
26736eb9 JK |
4374 | |
4375 | return 0; | |
4376 | ||
4377 | err_close_new_fd: | |
4378 | close(new_fd); | |
4379 | err_free_new_name: | |
4380 | free(new_name); | |
e9fc3ce9 | 4381 | return libbpf_err(err); |
26736eb9 JK |
4382 | } |
4383 | ||
1bdb6c9a | 4384 | __u32 bpf_map__max_entries(const struct bpf_map *map) |
1a11a4c7 | 4385 | { |
1bdb6c9a AN |
4386 | return map->def.max_entries; |
4387 | } | |
1a11a4c7 | 4388 | |
b3278099 AN |
4389 | struct bpf_map *bpf_map__inner_map(struct bpf_map *map) |
4390 | { | |
4391 | if (!bpf_map_type__is_map_in_map(map->def.type)) | |
e9fc3ce9 | 4392 | return errno = EINVAL, NULL; |
b3278099 AN |
4393 | |
4394 | return map->inner_map; | |
4395 | } | |
4396 | ||
1bdb6c9a AN |
4397 | int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries) |
4398 | { | |
1a11a4c7 | 4399 | if (map->fd >= 0) |
e9fc3ce9 | 4400 | return libbpf_err(-EBUSY); |
1a11a4c7 | 4401 | map->def.max_entries = max_entries; |
1a11a4c7 AI |
4402 | return 0; |
4403 | } | |
4404 | ||
1bdb6c9a AN |
4405 | int bpf_map__resize(struct bpf_map *map, __u32 max_entries) |
4406 | { | |
4407 | if (!map || !max_entries) | |
e9fc3ce9 | 4408 | return libbpf_err(-EINVAL); |
1bdb6c9a AN |
4409 | |
4410 | return bpf_map__set_max_entries(map, max_entries); | |
4411 | } | |
4412 | ||
47eff617 | 4413 | static int |
fd9eef1a | 4414 | bpf_object__probe_loading(struct bpf_object *obj) |
47eff617 | 4415 | { |
47eff617 SF |
4416 | char *cp, errmsg[STRERR_BUFSIZE]; |
4417 | struct bpf_insn insns[] = { | |
4418 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4419 | BPF_EXIT_INSN(), | |
4420 | }; | |
e32660ac | 4421 | int ret, insn_cnt = ARRAY_SIZE(insns); |
47eff617 | 4422 | |
f9bceaa5 SF |
4423 | if (obj->gen_loader) |
4424 | return 0; | |
4425 | ||
e542f2c4 AN |
4426 | ret = bump_rlimit_memlock(); |
4427 | if (ret) | |
4428 | pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret); | |
4429 | ||
47eff617 | 4430 | /* make sure basic loading works */ |
e32660ac AN |
4431 | ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL); |
4432 | if (ret < 0) | |
4433 | ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL); | |
47eff617 | 4434 | if (ret < 0) { |
fd9eef1a EC |
4435 | ret = errno; |
4436 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4437 | pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF " | |
4438 | "program. Make sure your kernel supports BPF " | |
4439 | "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is " | |
4440 | "set to big enough value.\n", __func__, cp, ret); | |
4441 | return -ret; | |
47eff617 SF |
4442 | } |
4443 | close(ret); | |
4444 | ||
fd9eef1a EC |
4445 | return 0; |
4446 | } | |
4447 | ||
bb180fb2 AN |
4448 | static int probe_fd(int fd) |
4449 | { | |
4450 | if (fd >= 0) | |
4451 | close(fd); | |
4452 | return fd >= 0; | |
4453 | } | |
4454 | ||
47b6cb4d | 4455 | static int probe_kern_prog_name(void) |
fd9eef1a | 4456 | { |
fd9eef1a EC |
4457 | struct bpf_insn insns[] = { |
4458 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4459 | BPF_EXIT_INSN(), | |
4460 | }; | |
e32660ac | 4461 | int ret, insn_cnt = ARRAY_SIZE(insns); |
fd9eef1a EC |
4462 | |
4463 | /* make sure loading with name works */ | |
e32660ac | 4464 | ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, "test", "GPL", insns, insn_cnt, NULL); |
bb180fb2 | 4465 | return probe_fd(ret); |
47eff617 SF |
4466 | } |
4467 | ||
47b6cb4d | 4468 | static int probe_kern_global_data(void) |
8837fe5d | 4469 | { |
8837fe5d DB |
4470 | char *cp, errmsg[STRERR_BUFSIZE]; |
4471 | struct bpf_insn insns[] = { | |
4472 | BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16), | |
4473 | BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42), | |
4474 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4475 | BPF_EXIT_INSN(), | |
4476 | }; | |
e32660ac | 4477 | int ret, map, insn_cnt = ARRAY_SIZE(insns); |
8837fe5d | 4478 | |
a9606f40 | 4479 | map = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), 32, 1, NULL); |
8837fe5d | 4480 | if (map < 0) { |
23ab656b THJ |
4481 | ret = -errno; |
4482 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
be18010e | 4483 | pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n", |
23ab656b THJ |
4484 | __func__, cp, -ret); |
4485 | return ret; | |
8837fe5d DB |
4486 | } |
4487 | ||
4488 | insns[0].imm = map; | |
4489 | ||
e32660ac | 4490 | ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL); |
47b6cb4d | 4491 | close(map); |
bb180fb2 | 4492 | return probe_fd(ret); |
8837fe5d DB |
4493 | } |
4494 | ||
68b08647 AN |
4495 | static int probe_kern_btf(void) |
4496 | { | |
4497 | static const char strs[] = "\0int"; | |
4498 | __u32 types[] = { | |
4499 | /* int */ | |
4500 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), | |
4501 | }; | |
4502 | ||
4503 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4504 | strs, sizeof(strs))); | |
4505 | } | |
4506 | ||
47b6cb4d | 4507 | static int probe_kern_btf_func(void) |
d7c4b398 | 4508 | { |
8983b731 | 4509 | static const char strs[] = "\0int\0x\0a"; |
d7c4b398 AN |
4510 | /* void x(int a) {} */ |
4511 | __u32 types[] = { | |
4512 | /* int */ | |
4513 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4514 | /* FUNC_PROTO */ /* [2] */ | |
4515 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0), | |
4516 | BTF_PARAM_ENC(7, 1), | |
4517 | /* FUNC x */ /* [3] */ | |
4518 | BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2), | |
4519 | }; | |
d7c4b398 | 4520 | |
bb180fb2 AN |
4521 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4522 | strs, sizeof(strs))); | |
d7c4b398 AN |
4523 | } |
4524 | ||
47b6cb4d | 4525 | static int probe_kern_btf_func_global(void) |
2d3eb67f AS |
4526 | { |
4527 | static const char strs[] = "\0int\0x\0a"; | |
4528 | /* static void x(int a) {} */ | |
4529 | __u32 types[] = { | |
4530 | /* int */ | |
4531 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4532 | /* FUNC_PROTO */ /* [2] */ | |
4533 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0), | |
4534 | BTF_PARAM_ENC(7, 1), | |
4535 | /* FUNC x BTF_FUNC_GLOBAL */ /* [3] */ | |
4536 | BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2), | |
4537 | }; | |
2d3eb67f | 4538 | |
bb180fb2 AN |
4539 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4540 | strs, sizeof(strs))); | |
2d3eb67f AS |
4541 | } |
4542 | ||
47b6cb4d | 4543 | static int probe_kern_btf_datasec(void) |
d7c4b398 | 4544 | { |
8983b731 | 4545 | static const char strs[] = "\0x\0.data"; |
d7c4b398 AN |
4546 | /* static int a; */ |
4547 | __u32 types[] = { | |
4548 | /* int */ | |
4549 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4550 | /* VAR x */ /* [2] */ | |
4551 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1), | |
4552 | BTF_VAR_STATIC, | |
4553 | /* DATASEC val */ /* [3] */ | |
4554 | BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4), | |
4555 | BTF_VAR_SECINFO_ENC(2, 0, 4), | |
4556 | }; | |
cfd49210 | 4557 | |
bb180fb2 AN |
4558 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4559 | strs, sizeof(strs))); | |
d7c4b398 AN |
4560 | } |
4561 | ||
22541a9e IL |
4562 | static int probe_kern_btf_float(void) |
4563 | { | |
4564 | static const char strs[] = "\0float"; | |
4565 | __u32 types[] = { | |
4566 | /* float */ | |
4567 | BTF_TYPE_FLOAT_ENC(1, 4), | |
4568 | }; | |
4569 | ||
4570 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4571 | strs, sizeof(strs))); | |
4572 | } | |
4573 | ||
223f903e | 4574 | static int probe_kern_btf_decl_tag(void) |
5b84bd10 YS |
4575 | { |
4576 | static const char strs[] = "\0tag"; | |
4577 | __u32 types[] = { | |
4578 | /* int */ | |
4579 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4580 | /* VAR x */ /* [2] */ | |
4581 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1), | |
4582 | BTF_VAR_STATIC, | |
4583 | /* attr */ | |
223f903e | 4584 | BTF_TYPE_DECL_TAG_ENC(1, 2, -1), |
5b84bd10 YS |
4585 | }; |
4586 | ||
4587 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4588 | strs, sizeof(strs))); | |
4589 | } | |
4590 | ||
2dc1e488 YS |
4591 | static int probe_kern_btf_type_tag(void) |
4592 | { | |
4593 | static const char strs[] = "\0tag"; | |
4594 | __u32 types[] = { | |
4595 | /* int */ | |
4596 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4597 | /* attr */ | |
4598 | BTF_TYPE_TYPE_TAG_ENC(1, 1), /* [2] */ | |
4599 | /* ptr */ | |
4600 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_PTR, 0, 0), 2), /* [3] */ | |
4601 | }; | |
4602 | ||
4603 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4604 | strs, sizeof(strs))); | |
4605 | } | |
4606 | ||
47b6cb4d | 4607 | static int probe_kern_array_mmap(void) |
7fe74b43 | 4608 | { |
a9606f40 AN |
4609 | LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); |
4610 | int fd; | |
7fe74b43 | 4611 | |
a9606f40 AN |
4612 | fd = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), sizeof(int), 1, &opts); |
4613 | return probe_fd(fd); | |
7fe74b43 AN |
4614 | } |
4615 | ||
47b6cb4d | 4616 | static int probe_kern_exp_attach_type(void) |
25498a19 | 4617 | { |
e32660ac | 4618 | LIBBPF_OPTS(bpf_prog_load_opts, opts, .expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE); |
25498a19 AN |
4619 | struct bpf_insn insns[] = { |
4620 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4621 | BPF_EXIT_INSN(), | |
4622 | }; | |
e32660ac | 4623 | int fd, insn_cnt = ARRAY_SIZE(insns); |
25498a19 | 4624 | |
25498a19 AN |
4625 | /* use any valid combination of program type and (optional) |
4626 | * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS) | |
4627 | * to see if kernel supports expected_attach_type field for | |
4628 | * BPF_PROG_LOAD command | |
4629 | */ | |
e32660ac AN |
4630 | fd = bpf_prog_load(BPF_PROG_TYPE_CGROUP_SOCK, NULL, "GPL", insns, insn_cnt, &opts); |
4631 | return probe_fd(fd); | |
25498a19 AN |
4632 | } |
4633 | ||
109cea5a AN |
4634 | static int probe_kern_probe_read_kernel(void) |
4635 | { | |
109cea5a AN |
4636 | struct bpf_insn insns[] = { |
4637 | BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), /* r1 = r10 (fp) */ | |
4638 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8), /* r1 += -8 */ | |
4639 | BPF_MOV64_IMM(BPF_REG_2, 8), /* r2 = 8 */ | |
4640 | BPF_MOV64_IMM(BPF_REG_3, 0), /* r3 = 0 */ | |
4641 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel), | |
4642 | BPF_EXIT_INSN(), | |
4643 | }; | |
e32660ac | 4644 | int fd, insn_cnt = ARRAY_SIZE(insns); |
109cea5a | 4645 | |
d252a4a4 | 4646 | fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL); |
e32660ac | 4647 | return probe_fd(fd); |
109cea5a AN |
4648 | } |
4649 | ||
5d23328d YZ |
4650 | static int probe_prog_bind_map(void) |
4651 | { | |
5d23328d YZ |
4652 | char *cp, errmsg[STRERR_BUFSIZE]; |
4653 | struct bpf_insn insns[] = { | |
4654 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4655 | BPF_EXIT_INSN(), | |
4656 | }; | |
e32660ac | 4657 | int ret, map, prog, insn_cnt = ARRAY_SIZE(insns); |
5d23328d | 4658 | |
a9606f40 | 4659 | map = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), 32, 1, NULL); |
5d23328d YZ |
4660 | if (map < 0) { |
4661 | ret = -errno; | |
4662 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4663 | pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n", | |
4664 | __func__, cp, -ret); | |
4665 | return ret; | |
4666 | } | |
4667 | ||
e32660ac | 4668 | prog = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL); |
5d23328d YZ |
4669 | if (prog < 0) { |
4670 | close(map); | |
4671 | return 0; | |
4672 | } | |
4673 | ||
4674 | ret = bpf_prog_bind_map(prog, map, NULL); | |
4675 | ||
4676 | close(map); | |
4677 | close(prog); | |
4678 | ||
4679 | return ret >= 0; | |
4680 | } | |
4681 | ||
4f33a53d AN |
4682 | static int probe_module_btf(void) |
4683 | { | |
4684 | static const char strs[] = "\0int"; | |
4685 | __u32 types[] = { | |
4686 | /* int */ | |
4687 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), | |
4688 | }; | |
4689 | struct bpf_btf_info info; | |
4690 | __u32 len = sizeof(info); | |
4691 | char name[16]; | |
4692 | int fd, err; | |
4693 | ||
4694 | fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs)); | |
4695 | if (fd < 0) | |
4696 | return 0; /* BTF not supported at all */ | |
4697 | ||
4698 | memset(&info, 0, sizeof(info)); | |
4699 | info.name = ptr_to_u64(name); | |
4700 | info.name_len = sizeof(name); | |
4701 | ||
4702 | /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer; | |
4703 | * kernel's module BTF support coincides with support for | |
4704 | * name/name_len fields in struct bpf_btf_info. | |
4705 | */ | |
4706 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
4707 | close(fd); | |
4708 | return !err; | |
4709 | } | |
4710 | ||
668ace0e AN |
4711 | static int probe_perf_link(void) |
4712 | { | |
668ace0e AN |
4713 | struct bpf_insn insns[] = { |
4714 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4715 | BPF_EXIT_INSN(), | |
4716 | }; | |
4717 | int prog_fd, link_fd, err; | |
4718 | ||
e32660ac AN |
4719 | prog_fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", |
4720 | insns, ARRAY_SIZE(insns), NULL); | |
668ace0e AN |
4721 | if (prog_fd < 0) |
4722 | return -errno; | |
4723 | ||
4724 | /* use invalid perf_event FD to get EBADF, if link is supported; | |
4725 | * otherwise EINVAL should be returned | |
4726 | */ | |
4727 | link_fd = bpf_link_create(prog_fd, -1, BPF_PERF_EVENT, NULL); | |
4728 | err = -errno; /* close() can clobber errno */ | |
4729 | ||
4730 | if (link_fd >= 0) | |
4731 | close(link_fd); | |
4732 | close(prog_fd); | |
4733 | ||
4734 | return link_fd < 0 && err == -EBADF; | |
4735 | } | |
4736 | ||
2e4913e0 AN |
4737 | static int probe_kern_bpf_cookie(void) |
4738 | { | |
4739 | struct bpf_insn insns[] = { | |
4740 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_get_attach_cookie), | |
4741 | BPF_EXIT_INSN(), | |
4742 | }; | |
4743 | int ret, insn_cnt = ARRAY_SIZE(insns); | |
4744 | ||
4745 | ret = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL, "GPL", insns, insn_cnt, NULL); | |
4746 | return probe_fd(ret); | |
4747 | } | |
4748 | ||
47b6cb4d AN |
4749 | enum kern_feature_result { |
4750 | FEAT_UNKNOWN = 0, | |
4751 | FEAT_SUPPORTED = 1, | |
4752 | FEAT_MISSING = 2, | |
4753 | }; | |
4754 | ||
4755 | typedef int (*feature_probe_fn)(void); | |
4756 | ||
4757 | static struct kern_feature_desc { | |
4758 | const char *desc; | |
4759 | feature_probe_fn probe; | |
4760 | enum kern_feature_result res; | |
4761 | } feature_probes[__FEAT_CNT] = { | |
4762 | [FEAT_PROG_NAME] = { | |
4763 | "BPF program name", probe_kern_prog_name, | |
4764 | }, | |
4765 | [FEAT_GLOBAL_DATA] = { | |
4766 | "global variables", probe_kern_global_data, | |
4767 | }, | |
68b08647 AN |
4768 | [FEAT_BTF] = { |
4769 | "minimal BTF", probe_kern_btf, | |
4770 | }, | |
47b6cb4d AN |
4771 | [FEAT_BTF_FUNC] = { |
4772 | "BTF functions", probe_kern_btf_func, | |
4773 | }, | |
4774 | [FEAT_BTF_GLOBAL_FUNC] = { | |
4775 | "BTF global function", probe_kern_btf_func_global, | |
4776 | }, | |
4777 | [FEAT_BTF_DATASEC] = { | |
4778 | "BTF data section and variable", probe_kern_btf_datasec, | |
4779 | }, | |
4780 | [FEAT_ARRAY_MMAP] = { | |
4781 | "ARRAY map mmap()", probe_kern_array_mmap, | |
4782 | }, | |
4783 | [FEAT_EXP_ATTACH_TYPE] = { | |
4784 | "BPF_PROG_LOAD expected_attach_type attribute", | |
4785 | probe_kern_exp_attach_type, | |
4786 | }, | |
109cea5a AN |
4787 | [FEAT_PROBE_READ_KERN] = { |
4788 | "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel, | |
5d23328d YZ |
4789 | }, |
4790 | [FEAT_PROG_BIND_MAP] = { | |
4791 | "BPF_PROG_BIND_MAP support", probe_prog_bind_map, | |
4f33a53d AN |
4792 | }, |
4793 | [FEAT_MODULE_BTF] = { | |
4794 | "module BTF support", probe_module_btf, | |
4795 | }, | |
22541a9e IL |
4796 | [FEAT_BTF_FLOAT] = { |
4797 | "BTF_KIND_FLOAT support", probe_kern_btf_float, | |
4798 | }, | |
668ace0e AN |
4799 | [FEAT_PERF_LINK] = { |
4800 | "BPF perf link support", probe_perf_link, | |
4801 | }, | |
223f903e YS |
4802 | [FEAT_BTF_DECL_TAG] = { |
4803 | "BTF_KIND_DECL_TAG support", probe_kern_btf_decl_tag, | |
5b84bd10 | 4804 | }, |
2dc1e488 YS |
4805 | [FEAT_BTF_TYPE_TAG] = { |
4806 | "BTF_KIND_TYPE_TAG support", probe_kern_btf_type_tag, | |
4807 | }, | |
e542f2c4 AN |
4808 | [FEAT_MEMCG_ACCOUNT] = { |
4809 | "memcg-based memory accounting", probe_memcg_account, | |
4810 | }, | |
2e4913e0 AN |
4811 | [FEAT_BPF_COOKIE] = { |
4812 | "BPF cookie support", probe_kern_bpf_cookie, | |
4813 | }, | |
47b6cb4d | 4814 | }; |
8837fe5d | 4815 | |
e542f2c4 | 4816 | bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id) |
47b6cb4d AN |
4817 | { |
4818 | struct kern_feature_desc *feat = &feature_probes[feat_id]; | |
4819 | int ret; | |
4820 | ||
e542f2c4 | 4821 | if (obj && obj->gen_loader) |
67234743 AS |
4822 | /* To generate loader program assume the latest kernel |
4823 | * to avoid doing extra prog_load, map_create syscalls. | |
4824 | */ | |
4825 | return true; | |
4826 | ||
47b6cb4d AN |
4827 | if (READ_ONCE(feat->res) == FEAT_UNKNOWN) { |
4828 | ret = feat->probe(); | |
4829 | if (ret > 0) { | |
4830 | WRITE_ONCE(feat->res, FEAT_SUPPORTED); | |
4831 | } else if (ret == 0) { | |
4832 | WRITE_ONCE(feat->res, FEAT_MISSING); | |
4833 | } else { | |
4834 | pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret); | |
4835 | WRITE_ONCE(feat->res, FEAT_MISSING); | |
4836 | } | |
8837fe5d DB |
4837 | } |
4838 | ||
47b6cb4d | 4839 | return READ_ONCE(feat->res) == FEAT_SUPPORTED; |
47eff617 SF |
4840 | } |
4841 | ||
57a00f41 THJ |
4842 | static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd) |
4843 | { | |
4844 | struct bpf_map_info map_info = {}; | |
4845 | char msg[STRERR_BUFSIZE]; | |
4846 | __u32 map_info_len; | |
97eb3138 | 4847 | int err; |
57a00f41 THJ |
4848 | |
4849 | map_info_len = sizeof(map_info); | |
4850 | ||
97eb3138 MP |
4851 | err = bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len); |
4852 | if (err && errno == EINVAL) | |
4853 | err = bpf_get_map_info_from_fdinfo(map_fd, &map_info); | |
4854 | if (err) { | |
4855 | pr_warn("failed to get map info for map FD %d: %s\n", map_fd, | |
4856 | libbpf_strerror_r(errno, msg, sizeof(msg))); | |
57a00f41 THJ |
4857 | return false; |
4858 | } | |
4859 | ||
4860 | return (map_info.type == map->def.type && | |
4861 | map_info.key_size == map->def.key_size && | |
4862 | map_info.value_size == map->def.value_size && | |
4863 | map_info.max_entries == map->def.max_entries && | |
47512102 JK |
4864 | map_info.map_flags == map->def.map_flags && |
4865 | map_info.map_extra == map->map_extra); | |
57a00f41 THJ |
4866 | } |
4867 | ||
4868 | static int | |
4869 | bpf_object__reuse_map(struct bpf_map *map) | |
4870 | { | |
4871 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4872 | int err, pin_fd; | |
4873 | ||
4874 | pin_fd = bpf_obj_get(map->pin_path); | |
4875 | if (pin_fd < 0) { | |
4876 | err = -errno; | |
4877 | if (err == -ENOENT) { | |
4878 | pr_debug("found no pinned map to reuse at '%s'\n", | |
4879 | map->pin_path); | |
4880 | return 0; | |
4881 | } | |
4882 | ||
4883 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
4884 | pr_warn("couldn't retrieve pinned map '%s': %s\n", | |
4885 | map->pin_path, cp); | |
4886 | return err; | |
4887 | } | |
4888 | ||
4889 | if (!map_is_reuse_compat(map, pin_fd)) { | |
4890 | pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n", | |
4891 | map->pin_path); | |
4892 | close(pin_fd); | |
4893 | return -EINVAL; | |
4894 | } | |
4895 | ||
4896 | err = bpf_map__reuse_fd(map, pin_fd); | |
d0f325c3 | 4897 | close(pin_fd); |
57a00f41 | 4898 | if (err) { |
57a00f41 THJ |
4899 | return err; |
4900 | } | |
4901 | map->pinned = true; | |
4902 | pr_debug("reused pinned map at '%s'\n", map->pin_path); | |
4903 | ||
4904 | return 0; | |
4905 | } | |
4906 | ||
d859900c DB |
4907 | static int |
4908 | bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map) | |
4909 | { | |
166750bc | 4910 | enum libbpf_map_type map_type = map->libbpf_type; |
d859900c DB |
4911 | char *cp, errmsg[STRERR_BUFSIZE]; |
4912 | int err, zero = 0; | |
d859900c | 4913 | |
67234743 AS |
4914 | if (obj->gen_loader) { |
4915 | bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps, | |
4916 | map->mmaped, map->def.value_size); | |
4917 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) | |
4918 | bpf_gen__map_freeze(obj->gen_loader, map - obj->maps); | |
4919 | return 0; | |
4920 | } | |
eba9c5f4 AN |
4921 | err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0); |
4922 | if (err) { | |
4923 | err = -errno; | |
4924 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
4925 | pr_warn("Error setting initial map(%s) contents: %s\n", | |
4926 | map->name, cp); | |
4927 | return err; | |
4928 | } | |
d859900c | 4929 | |
81bfdd08 AN |
4930 | /* Freeze .rodata and .kconfig map as read-only from syscall side. */ |
4931 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) { | |
d859900c DB |
4932 | err = bpf_map_freeze(map->fd); |
4933 | if (err) { | |
eba9c5f4 AN |
4934 | err = -errno; |
4935 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e KW |
4936 | pr_warn("Error freezing map(%s) as read-only: %s\n", |
4937 | map->name, cp); | |
eba9c5f4 | 4938 | return err; |
d859900c DB |
4939 | } |
4940 | } | |
eba9c5f4 | 4941 | return 0; |
d859900c DB |
4942 | } |
4943 | ||
2d39d7c5 AN |
4944 | static void bpf_map__destroy(struct bpf_map *map); |
4945 | ||
67234743 | 4946 | static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner) |
2d39d7c5 | 4947 | { |
992c4225 | 4948 | LIBBPF_OPTS(bpf_map_create_opts, create_attr); |
2d39d7c5 | 4949 | struct bpf_map_def *def = &map->def; |
992c4225 | 4950 | const char *map_name = NULL; |
a21ab4c5 | 4951 | int err = 0; |
2d39d7c5 | 4952 | |
9ca1f56a | 4953 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
992c4225 | 4954 | map_name = map->name; |
2d39d7c5 | 4955 | create_attr.map_ifindex = map->map_ifindex; |
2d39d7c5 | 4956 | create_attr.map_flags = def->map_flags; |
1bdb6c9a | 4957 | create_attr.numa_node = map->numa_node; |
47512102 | 4958 | create_attr.map_extra = map->map_extra; |
2d39d7c5 | 4959 | |
2d39d7c5 | 4960 | if (bpf_map__is_struct_ops(map)) |
992c4225 | 4961 | create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id; |
2d39d7c5 | 4962 | |
262cfb74 | 4963 | if (obj->btf && btf__fd(obj->btf) >= 0) { |
2d39d7c5 AN |
4964 | create_attr.btf_fd = btf__fd(obj->btf); |
4965 | create_attr.btf_key_type_id = map->btf_key_type_id; | |
4966 | create_attr.btf_value_type_id = map->btf_value_type_id; | |
4967 | } | |
4968 | ||
646f02ff AN |
4969 | if (bpf_map_type__is_map_in_map(def->type)) { |
4970 | if (map->inner_map) { | |
67234743 | 4971 | err = bpf_object__create_map(obj, map->inner_map, true); |
646f02ff AN |
4972 | if (err) { |
4973 | pr_warn("map '%s': failed to create inner map: %d\n", | |
4974 | map->name, err); | |
4975 | return err; | |
4976 | } | |
4977 | map->inner_map_fd = bpf_map__fd(map->inner_map); | |
4978 | } | |
4979 | if (map->inner_map_fd >= 0) | |
4980 | create_attr.inner_map_fd = map->inner_map_fd; | |
4981 | } | |
4982 | ||
f7310523 HC |
4983 | switch (def->type) { |
4984 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
4985 | case BPF_MAP_TYPE_CGROUP_ARRAY: | |
4986 | case BPF_MAP_TYPE_STACK_TRACE: | |
4987 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
4988 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
4989 | case BPF_MAP_TYPE_DEVMAP: | |
4990 | case BPF_MAP_TYPE_DEVMAP_HASH: | |
4991 | case BPF_MAP_TYPE_CPUMAP: | |
4992 | case BPF_MAP_TYPE_XSKMAP: | |
4993 | case BPF_MAP_TYPE_SOCKMAP: | |
4994 | case BPF_MAP_TYPE_SOCKHASH: | |
4995 | case BPF_MAP_TYPE_QUEUE: | |
4996 | case BPF_MAP_TYPE_STACK: | |
4997 | case BPF_MAP_TYPE_RINGBUF: | |
4998 | create_attr.btf_fd = 0; | |
4999 | create_attr.btf_key_type_id = 0; | |
5000 | create_attr.btf_value_type_id = 0; | |
5001 | map->btf_key_type_id = 0; | |
5002 | map->btf_value_type_id = 0; | |
5003 | default: | |
5004 | break; | |
5005 | } | |
5006 | ||
67234743 | 5007 | if (obj->gen_loader) { |
992c4225 | 5008 | bpf_gen__map_create(obj->gen_loader, def->type, map_name, |
a4fbfdd7 | 5009 | def->key_size, def->value_size, def->max_entries, |
992c4225 | 5010 | &create_attr, is_inner ? -1 : map - obj->maps); |
67234743 AS |
5011 | /* Pretend to have valid FD to pass various fd >= 0 checks. |
5012 | * This fd == 0 will not be used with any syscall and will be reset to -1 eventually. | |
5013 | */ | |
5014 | map->fd = 0; | |
5015 | } else { | |
992c4225 AN |
5016 | map->fd = bpf_map_create(def->type, map_name, |
5017 | def->key_size, def->value_size, | |
a4fbfdd7 | 5018 | def->max_entries, &create_attr); |
67234743 | 5019 | } |
2d39d7c5 AN |
5020 | if (map->fd < 0 && (create_attr.btf_key_type_id || |
5021 | create_attr.btf_value_type_id)) { | |
5022 | char *cp, errmsg[STRERR_BUFSIZE]; | |
2d39d7c5 | 5023 | |
a21ab4c5 | 5024 | err = -errno; |
2d39d7c5 AN |
5025 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); |
5026 | pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n", | |
5027 | map->name, cp, err); | |
5028 | create_attr.btf_fd = 0; | |
5029 | create_attr.btf_key_type_id = 0; | |
5030 | create_attr.btf_value_type_id = 0; | |
5031 | map->btf_key_type_id = 0; | |
5032 | map->btf_value_type_id = 0; | |
992c4225 AN |
5033 | map->fd = bpf_map_create(def->type, map_name, |
5034 | def->key_size, def->value_size, | |
a4fbfdd7 | 5035 | def->max_entries, &create_attr); |
2d39d7c5 AN |
5036 | } |
5037 | ||
a21ab4c5 | 5038 | err = map->fd < 0 ? -errno : 0; |
2d39d7c5 | 5039 | |
646f02ff | 5040 | if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) { |
67234743 AS |
5041 | if (obj->gen_loader) |
5042 | map->inner_map->fd = -1; | |
646f02ff AN |
5043 | bpf_map__destroy(map->inner_map); |
5044 | zfree(&map->inner_map); | |
5045 | } | |
5046 | ||
a21ab4c5 | 5047 | return err; |
2d39d7c5 AN |
5048 | } |
5049 | ||
341ac5ff | 5050 | static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map) |
a0f2b7ac HL |
5051 | { |
5052 | const struct bpf_map *targ_map; | |
5053 | unsigned int i; | |
67234743 | 5054 | int fd, err = 0; |
a0f2b7ac HL |
5055 | |
5056 | for (i = 0; i < map->init_slots_sz; i++) { | |
5057 | if (!map->init_slots[i]) | |
5058 | continue; | |
5059 | ||
5060 | targ_map = map->init_slots[i]; | |
5061 | fd = bpf_map__fd(targ_map); | |
341ac5ff | 5062 | |
67234743 | 5063 | if (obj->gen_loader) { |
be05c944 AS |
5064 | bpf_gen__populate_outer_map(obj->gen_loader, |
5065 | map - obj->maps, i, | |
5066 | targ_map - obj->maps); | |
67234743 AS |
5067 | } else { |
5068 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
5069 | } | |
a0f2b7ac HL |
5070 | if (err) { |
5071 | err = -errno; | |
5072 | pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n", | |
341ac5ff | 5073 | map->name, i, targ_map->name, fd, err); |
a0f2b7ac HL |
5074 | return err; |
5075 | } | |
5076 | pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n", | |
5077 | map->name, i, targ_map->name, fd); | |
5078 | } | |
5079 | ||
5080 | zfree(&map->init_slots); | |
5081 | map->init_slots_sz = 0; | |
5082 | ||
5083 | return 0; | |
5084 | } | |
5085 | ||
341ac5ff HC |
5086 | static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map) |
5087 | { | |
5088 | const struct bpf_program *targ_prog; | |
5089 | unsigned int i; | |
5090 | int fd, err; | |
5091 | ||
5092 | if (obj->gen_loader) | |
5093 | return -ENOTSUP; | |
5094 | ||
5095 | for (i = 0; i < map->init_slots_sz; i++) { | |
5096 | if (!map->init_slots[i]) | |
5097 | continue; | |
5098 | ||
5099 | targ_prog = map->init_slots[i]; | |
5100 | fd = bpf_program__fd(targ_prog); | |
5101 | ||
5102 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
5103 | if (err) { | |
5104 | err = -errno; | |
5105 | pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %d\n", | |
5106 | map->name, i, targ_prog->name, fd, err); | |
5107 | return err; | |
5108 | } | |
5109 | pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n", | |
5110 | map->name, i, targ_prog->name, fd); | |
5111 | } | |
5112 | ||
5113 | zfree(&map->init_slots); | |
5114 | map->init_slots_sz = 0; | |
5115 | ||
5116 | return 0; | |
5117 | } | |
5118 | ||
5119 | static int bpf_object_init_prog_arrays(struct bpf_object *obj) | |
5120 | { | |
5121 | struct bpf_map *map; | |
5122 | int i, err; | |
5123 | ||
5124 | for (i = 0; i < obj->nr_maps; i++) { | |
5125 | map = &obj->maps[i]; | |
5126 | ||
5127 | if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY) | |
5128 | continue; | |
5129 | ||
5130 | err = init_prog_array_slots(obj, map); | |
5131 | if (err < 0) { | |
5132 | zclose(map->fd); | |
5133 | return err; | |
5134 | } | |
5135 | } | |
5136 | return 0; | |
5137 | } | |
5138 | ||
a4fbfdd7 ST |
5139 | static int map_set_def_max_entries(struct bpf_map *map) |
5140 | { | |
5141 | if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) { | |
5142 | int nr_cpus; | |
5143 | ||
5144 | nr_cpus = libbpf_num_possible_cpus(); | |
5145 | if (nr_cpus < 0) { | |
5146 | pr_warn("map '%s': failed to determine number of system CPUs: %d\n", | |
5147 | map->name, nr_cpus); | |
5148 | return nr_cpus; | |
5149 | } | |
5150 | pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus); | |
5151 | map->def.max_entries = nr_cpus; | |
5152 | } | |
5153 | ||
5154 | return 0; | |
5155 | } | |
5156 | ||
52d3352e WN |
5157 | static int |
5158 | bpf_object__create_maps(struct bpf_object *obj) | |
5159 | { | |
2d39d7c5 AN |
5160 | struct bpf_map *map; |
5161 | char *cp, errmsg[STRERR_BUFSIZE]; | |
5162 | unsigned int i, j; | |
8a138aed | 5163 | int err; |
043c5bb3 | 5164 | bool retried; |
52d3352e | 5165 | |
9d759a9b | 5166 | for (i = 0; i < obj->nr_maps; i++) { |
2d39d7c5 | 5167 | map = &obj->maps[i]; |
8a138aed | 5168 | |
16e0c35c AN |
5169 | /* To support old kernels, we skip creating global data maps |
5170 | * (.rodata, .data, .kconfig, etc); later on, during program | |
5171 | * loading, if we detect that at least one of the to-be-loaded | |
5172 | * programs is referencing any global data map, we'll error | |
5173 | * out with program name and relocation index logged. | |
5174 | * This approach allows to accommodate Clang emitting | |
5175 | * unnecessary .rodata.str1.1 sections for string literals, | |
5176 | * but also it allows to have CO-RE applications that use | |
5177 | * global variables in some of BPF programs, but not others. | |
5178 | * If those global variable-using programs are not loaded at | |
5179 | * runtime due to bpf_program__set_autoload(prog, false), | |
5180 | * bpf_object loading will succeed just fine even on old | |
5181 | * kernels. | |
5182 | */ | |
5183 | if (bpf_map__is_internal(map) && | |
229fae38 SC |
5184 | !kernel_supports(obj, FEAT_GLOBAL_DATA)) { |
5185 | map->skipped = true; | |
16e0c35c | 5186 | continue; |
229fae38 | 5187 | } |
16e0c35c | 5188 | |
a4fbfdd7 ST |
5189 | err = map_set_def_max_entries(map); |
5190 | if (err) | |
5191 | goto err_out; | |
5192 | ||
043c5bb3 MP |
5193 | retried = false; |
5194 | retry: | |
57a00f41 THJ |
5195 | if (map->pin_path) { |
5196 | err = bpf_object__reuse_map(map); | |
5197 | if (err) { | |
2d39d7c5 | 5198 | pr_warn("map '%s': error reusing pinned map\n", |
57a00f41 | 5199 | map->name); |
2d39d7c5 | 5200 | goto err_out; |
57a00f41 | 5201 | } |
043c5bb3 MP |
5202 | if (retried && map->fd < 0) { |
5203 | pr_warn("map '%s': cannot find pinned map\n", | |
5204 | map->name); | |
5205 | err = -ENOENT; | |
5206 | goto err_out; | |
5207 | } | |
57a00f41 THJ |
5208 | } |
5209 | ||
26736eb9 | 5210 | if (map->fd >= 0) { |
2d39d7c5 | 5211 | pr_debug("map '%s': skipping creation (preset fd=%d)\n", |
26736eb9 | 5212 | map->name, map->fd); |
2c193d32 | 5213 | } else { |
67234743 | 5214 | err = bpf_object__create_map(obj, map, false); |
2c193d32 | 5215 | if (err) |
d859900c | 5216 | goto err_out; |
d859900c | 5217 | |
2c193d32 HL |
5218 | pr_debug("map '%s': created successfully, fd=%d\n", |
5219 | map->name, map->fd); | |
646f02ff | 5220 | |
2c193d32 HL |
5221 | if (bpf_map__is_internal(map)) { |
5222 | err = bpf_object__populate_internal_map(obj, map); | |
5223 | if (err < 0) { | |
5224 | zclose(map->fd); | |
5225 | goto err_out; | |
5226 | } | |
d859900c | 5227 | } |
646f02ff | 5228 | |
341ac5ff HC |
5229 | if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) { |
5230 | err = init_map_in_map_slots(obj, map); | |
2c193d32 HL |
5231 | if (err < 0) { |
5232 | zclose(map->fd); | |
646f02ff AN |
5233 | goto err_out; |
5234 | } | |
646f02ff | 5235 | } |
646f02ff AN |
5236 | } |
5237 | ||
57a00f41 THJ |
5238 | if (map->pin_path && !map->pinned) { |
5239 | err = bpf_map__pin(map, NULL); | |
5240 | if (err) { | |
043c5bb3 MP |
5241 | zclose(map->fd); |
5242 | if (!retried && err == -EEXIST) { | |
5243 | retried = true; | |
5244 | goto retry; | |
5245 | } | |
2d39d7c5 AN |
5246 | pr_warn("map '%s': failed to auto-pin at '%s': %d\n", |
5247 | map->name, map->pin_path, err); | |
2d39d7c5 | 5248 | goto err_out; |
57a00f41 THJ |
5249 | } |
5250 | } | |
52d3352e WN |
5251 | } |
5252 | ||
52d3352e | 5253 | return 0; |
2d39d7c5 AN |
5254 | |
5255 | err_out: | |
5256 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
5257 | pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err); | |
5258 | pr_perm_msg(err); | |
5259 | for (j = 0; j < i; j++) | |
5260 | zclose(obj->maps[j].fd); | |
5261 | return err; | |
52d3352e WN |
5262 | } |
5263 | ||
ddc7c304 AN |
5264 | static bool bpf_core_is_flavor_sep(const char *s) |
5265 | { | |
5266 | /* check X___Y name pattern, where X and Y are not underscores */ | |
5267 | return s[0] != '_' && /* X */ | |
5268 | s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */ | |
5269 | s[4] != '_'; /* Y */ | |
5270 | } | |
5271 | ||
5272 | /* Given 'some_struct_name___with_flavor' return the length of a name prefix | |
5273 | * before last triple underscore. Struct name part after last triple | |
5274 | * underscore is ignored by BPF CO-RE relocation during relocation matching. | |
5275 | */ | |
b0588390 | 5276 | size_t bpf_core_essential_name_len(const char *name) |
ddc7c304 AN |
5277 | { |
5278 | size_t n = strlen(name); | |
5279 | int i; | |
5280 | ||
5281 | for (i = n - 5; i >= 0; i--) { | |
5282 | if (bpf_core_is_flavor_sep(name + i)) | |
5283 | return i + 1; | |
5284 | } | |
5285 | return n; | |
5286 | } | |
5287 | ||
8de6cae4 | 5288 | void bpf_core_free_cands(struct bpf_core_cand_list *cands) |
ddc7c304 | 5289 | { |
8de6cae4 MV |
5290 | if (!cands) |
5291 | return; | |
5292 | ||
0f7515ca AN |
5293 | free(cands->cands); |
5294 | free(cands); | |
ddc7c304 AN |
5295 | } |
5296 | ||
8de6cae4 MV |
5297 | int bpf_core_add_cands(struct bpf_core_cand *local_cand, |
5298 | size_t local_essent_len, | |
5299 | const struct btf *targ_btf, | |
5300 | const char *targ_btf_name, | |
5301 | int targ_start_id, | |
5302 | struct bpf_core_cand_list *cands) | |
ddc7c304 | 5303 | { |
301ba4d7 | 5304 | struct bpf_core_cand *new_cands, *cand; |
03d5b991 AN |
5305 | const struct btf_type *t, *local_t; |
5306 | const char *targ_name, *local_name; | |
0f7515ca AN |
5307 | size_t targ_essent_len; |
5308 | int n, i; | |
ddc7c304 | 5309 | |
03d5b991 AN |
5310 | local_t = btf__type_by_id(local_cand->btf, local_cand->id); |
5311 | local_name = btf__str_by_offset(local_cand->btf, local_t->name_off); | |
5312 | ||
6a886de0 HC |
5313 | n = btf__type_cnt(targ_btf); |
5314 | for (i = targ_start_id; i < n; i++) { | |
ddc7c304 | 5315 | t = btf__type_by_id(targ_btf, i); |
03d5b991 | 5316 | if (btf_kind(t) != btf_kind(local_t)) |
ddc7c304 AN |
5317 | continue; |
5318 | ||
3fc32f40 AN |
5319 | targ_name = btf__name_by_offset(targ_btf, t->name_off); |
5320 | if (str_is_empty(targ_name)) | |
d121e1d3 AN |
5321 | continue; |
5322 | ||
ddc7c304 AN |
5323 | targ_essent_len = bpf_core_essential_name_len(targ_name); |
5324 | if (targ_essent_len != local_essent_len) | |
5325 | continue; | |
5326 | ||
03d5b991 | 5327 | if (strncmp(local_name, targ_name, local_essent_len) != 0) |
0f7515ca AN |
5328 | continue; |
5329 | ||
5330 | pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n", | |
03d5b991 AN |
5331 | local_cand->id, btf_kind_str(local_t), |
5332 | local_name, i, btf_kind_str(t), targ_name, | |
0f7515ca AN |
5333 | targ_btf_name); |
5334 | new_cands = libbpf_reallocarray(cands->cands, cands->len + 1, | |
5335 | sizeof(*cands->cands)); | |
5336 | if (!new_cands) | |
5337 | return -ENOMEM; | |
5338 | ||
5339 | cand = &new_cands[cands->len]; | |
5340 | cand->btf = targ_btf; | |
0f7515ca AN |
5341 | cand->id = i; |
5342 | ||
5343 | cands->cands = new_cands; | |
5344 | cands->len++; | |
ddc7c304 | 5345 | } |
0f7515ca AN |
5346 | return 0; |
5347 | } | |
5348 | ||
4f33a53d AN |
5349 | static int load_module_btfs(struct bpf_object *obj) |
5350 | { | |
5351 | struct bpf_btf_info info; | |
5352 | struct module_btf *mod_btf; | |
5353 | struct btf *btf; | |
5354 | char name[64]; | |
5355 | __u32 id = 0, len; | |
5356 | int err, fd; | |
5357 | ||
5358 | if (obj->btf_modules_loaded) | |
5359 | return 0; | |
5360 | ||
67234743 AS |
5361 | if (obj->gen_loader) |
5362 | return 0; | |
5363 | ||
4f33a53d AN |
5364 | /* don't do this again, even if we find no module BTFs */ |
5365 | obj->btf_modules_loaded = true; | |
5366 | ||
5367 | /* kernel too old to support module BTFs */ | |
9ca1f56a | 5368 | if (!kernel_supports(obj, FEAT_MODULE_BTF)) |
4f33a53d AN |
5369 | return 0; |
5370 | ||
5371 | while (true) { | |
5372 | err = bpf_btf_get_next_id(id, &id); | |
5373 | if (err && errno == ENOENT) | |
5374 | return 0; | |
5375 | if (err) { | |
5376 | err = -errno; | |
5377 | pr_warn("failed to iterate BTF objects: %d\n", err); | |
5378 | return err; | |
ddc7c304 | 5379 | } |
4f33a53d AN |
5380 | |
5381 | fd = bpf_btf_get_fd_by_id(id); | |
5382 | if (fd < 0) { | |
5383 | if (errno == ENOENT) | |
5384 | continue; /* expected race: BTF was unloaded */ | |
5385 | err = -errno; | |
5386 | pr_warn("failed to get BTF object #%d FD: %d\n", id, err); | |
5387 | return err; | |
5388 | } | |
5389 | ||
5390 | len = sizeof(info); | |
5391 | memset(&info, 0, sizeof(info)); | |
5392 | info.name = ptr_to_u64(name); | |
5393 | info.name_len = sizeof(name); | |
5394 | ||
5395 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
5396 | if (err) { | |
5397 | err = -errno; | |
5398 | pr_warn("failed to get BTF object #%d info: %d\n", id, err); | |
91abb4a6 | 5399 | goto err_out; |
4f33a53d AN |
5400 | } |
5401 | ||
5402 | /* ignore non-module BTFs */ | |
5403 | if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) { | |
5404 | close(fd); | |
5405 | continue; | |
5406 | } | |
5407 | ||
5408 | btf = btf_get_from_fd(fd, obj->btf_vmlinux); | |
e9fc3ce9 AN |
5409 | err = libbpf_get_error(btf); |
5410 | if (err) { | |
5411 | pr_warn("failed to load module [%s]'s BTF object #%d: %d\n", | |
5412 | name, id, err); | |
91abb4a6 | 5413 | goto err_out; |
4f33a53d AN |
5414 | } |
5415 | ||
3b029e06 AN |
5416 | err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap, |
5417 | sizeof(*obj->btf_modules), obj->btf_module_cnt + 1); | |
4f33a53d | 5418 | if (err) |
91abb4a6 | 5419 | goto err_out; |
4f33a53d AN |
5420 | |
5421 | mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; | |
5422 | ||
5423 | mod_btf->btf = btf; | |
5424 | mod_btf->id = id; | |
91abb4a6 | 5425 | mod_btf->fd = fd; |
4f33a53d | 5426 | mod_btf->name = strdup(name); |
91abb4a6 AN |
5427 | if (!mod_btf->name) { |
5428 | err = -ENOMEM; | |
5429 | goto err_out; | |
5430 | } | |
5431 | continue; | |
5432 | ||
5433 | err_out: | |
5434 | close(fd); | |
5435 | return err; | |
ddc7c304 | 5436 | } |
4f33a53d AN |
5437 | |
5438 | return 0; | |
5439 | } | |
5440 | ||
301ba4d7 | 5441 | static struct bpf_core_cand_list * |
0f7515ca AN |
5442 | bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id) |
5443 | { | |
301ba4d7 AS |
5444 | struct bpf_core_cand local_cand = {}; |
5445 | struct bpf_core_cand_list *cands; | |
4f33a53d | 5446 | const struct btf *main_btf; |
03d5b991 AN |
5447 | const struct btf_type *local_t; |
5448 | const char *local_name; | |
0f7515ca | 5449 | size_t local_essent_len; |
4f33a53d | 5450 | int err, i; |
0f7515ca AN |
5451 | |
5452 | local_cand.btf = local_btf; | |
03d5b991 AN |
5453 | local_cand.id = local_type_id; |
5454 | local_t = btf__type_by_id(local_btf, local_type_id); | |
5455 | if (!local_t) | |
0f7515ca AN |
5456 | return ERR_PTR(-EINVAL); |
5457 | ||
03d5b991 AN |
5458 | local_name = btf__name_by_offset(local_btf, local_t->name_off); |
5459 | if (str_is_empty(local_name)) | |
0f7515ca | 5460 | return ERR_PTR(-EINVAL); |
03d5b991 | 5461 | local_essent_len = bpf_core_essential_name_len(local_name); |
0f7515ca AN |
5462 | |
5463 | cands = calloc(1, sizeof(*cands)); | |
5464 | if (!cands) | |
5465 | return ERR_PTR(-ENOMEM); | |
5466 | ||
5467 | /* Attempt to find target candidates in vmlinux BTF first */ | |
4f33a53d AN |
5468 | main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux; |
5469 | err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands); | |
5470 | if (err) | |
5471 | goto err_out; | |
5472 | ||
5473 | /* if vmlinux BTF has any candidate, don't got for module BTFs */ | |
5474 | if (cands->len) | |
5475 | return cands; | |
5476 | ||
5477 | /* if vmlinux BTF was overridden, don't attempt to load module BTFs */ | |
5478 | if (obj->btf_vmlinux_override) | |
5479 | return cands; | |
5480 | ||
5481 | /* now look through module BTFs, trying to still find candidates */ | |
5482 | err = load_module_btfs(obj); | |
5483 | if (err) | |
5484 | goto err_out; | |
5485 | ||
5486 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
5487 | err = bpf_core_add_cands(&local_cand, local_essent_len, | |
5488 | obj->btf_modules[i].btf, | |
5489 | obj->btf_modules[i].name, | |
6a886de0 | 5490 | btf__type_cnt(obj->btf_vmlinux), |
4f33a53d AN |
5491 | cands); |
5492 | if (err) | |
5493 | goto err_out; | |
0f7515ca AN |
5494 | } |
5495 | ||
5496 | return cands; | |
ddc7c304 | 5497 | err_out: |
4f33a53d | 5498 | bpf_core_free_cands(cands); |
ddc7c304 AN |
5499 | return ERR_PTR(err); |
5500 | } | |
5501 | ||
3fc32f40 AN |
5502 | /* Check local and target types for compatibility. This check is used for |
5503 | * type-based CO-RE relocations and follow slightly different rules than | |
5504 | * field-based relocations. This function assumes that root types were already | |
5505 | * checked for name match. Beyond that initial root-level name check, names | |
5506 | * are completely ignored. Compatibility rules are as follows: | |
5507 | * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but | |
5508 | * kind should match for local and target types (i.e., STRUCT is not | |
5509 | * compatible with UNION); | |
5510 | * - for ENUMs, the size is ignored; | |
5511 | * - for INT, size and signedness are ignored; | |
5512 | * - for ARRAY, dimensionality is ignored, element types are checked for | |
5513 | * compatibility recursively; | |
5514 | * - CONST/VOLATILE/RESTRICT modifiers are ignored; | |
5515 | * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible; | |
5516 | * - FUNC_PROTOs are compatible if they have compatible signature: same | |
5517 | * number of input args and compatible return and argument types. | |
5518 | * These rules are not set in stone and probably will be adjusted as we get | |
5519 | * more experience with using BPF CO-RE relocations. | |
5520 | */ | |
b0588390 AS |
5521 | int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id, |
5522 | const struct btf *targ_btf, __u32 targ_id) | |
3fc32f40 AN |
5523 | { |
5524 | const struct btf_type *local_type, *targ_type; | |
5525 | int depth = 32; /* max recursion depth */ | |
5526 | ||
5527 | /* caller made sure that names match (ignoring flavor suffix) */ | |
5528 | local_type = btf__type_by_id(local_btf, local_id); | |
f872e4bc | 5529 | targ_type = btf__type_by_id(targ_btf, targ_id); |
3fc32f40 AN |
5530 | if (btf_kind(local_type) != btf_kind(targ_type)) |
5531 | return 0; | |
5532 | ||
5533 | recur: | |
5534 | depth--; | |
5535 | if (depth < 0) | |
5536 | return -EINVAL; | |
5537 | ||
5538 | local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id); | |
5539 | targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id); | |
5540 | if (!local_type || !targ_type) | |
5541 | return -EINVAL; | |
5542 | ||
5543 | if (btf_kind(local_type) != btf_kind(targ_type)) | |
5544 | return 0; | |
5545 | ||
5546 | switch (btf_kind(local_type)) { | |
5547 | case BTF_KIND_UNKN: | |
5548 | case BTF_KIND_STRUCT: | |
5549 | case BTF_KIND_UNION: | |
5550 | case BTF_KIND_ENUM: | |
5551 | case BTF_KIND_FWD: | |
5552 | return 1; | |
5553 | case BTF_KIND_INT: | |
5554 | /* just reject deprecated bitfield-like integers; all other | |
5555 | * integers are by default compatible between each other | |
5556 | */ | |
5557 | return btf_int_offset(local_type) == 0 && btf_int_offset(targ_type) == 0; | |
5558 | case BTF_KIND_PTR: | |
5559 | local_id = local_type->type; | |
5560 | targ_id = targ_type->type; | |
5561 | goto recur; | |
5562 | case BTF_KIND_ARRAY: | |
5563 | local_id = btf_array(local_type)->type; | |
5564 | targ_id = btf_array(targ_type)->type; | |
5565 | goto recur; | |
5566 | case BTF_KIND_FUNC_PROTO: { | |
5567 | struct btf_param *local_p = btf_params(local_type); | |
5568 | struct btf_param *targ_p = btf_params(targ_type); | |
5569 | __u16 local_vlen = btf_vlen(local_type); | |
5570 | __u16 targ_vlen = btf_vlen(targ_type); | |
5571 | int i, err; | |
5572 | ||
5573 | if (local_vlen != targ_vlen) | |
5574 | return 0; | |
5575 | ||
5576 | for (i = 0; i < local_vlen; i++, local_p++, targ_p++) { | |
5577 | skip_mods_and_typedefs(local_btf, local_p->type, &local_id); | |
5578 | skip_mods_and_typedefs(targ_btf, targ_p->type, &targ_id); | |
5579 | err = bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id); | |
5580 | if (err <= 0) | |
5581 | return err; | |
5582 | } | |
5583 | ||
5584 | /* tail recurse for return type check */ | |
5585 | skip_mods_and_typedefs(local_btf, local_type->type, &local_id); | |
5586 | skip_mods_and_typedefs(targ_btf, targ_type->type, &targ_id); | |
5587 | goto recur; | |
5588 | } | |
5589 | default: | |
5590 | pr_warn("unexpected kind %s relocated, local [%d], target [%d]\n", | |
5591 | btf_kind_str(local_type), local_id, targ_id); | |
5592 | return 0; | |
5593 | } | |
5594 | } | |
5595 | ||
ddc7c304 AN |
5596 | static size_t bpf_core_hash_fn(const void *key, void *ctx) |
5597 | { | |
5598 | return (size_t)key; | |
5599 | } | |
5600 | ||
5601 | static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx) | |
5602 | { | |
5603 | return k1 == k2; | |
5604 | } | |
5605 | ||
5606 | static void *u32_as_hash_key(__u32 x) | |
5607 | { | |
5608 | return (void *)(uintptr_t)x; | |
5609 | } | |
5610 | ||
d0e92887 AS |
5611 | static int record_relo_core(struct bpf_program *prog, |
5612 | const struct bpf_core_relo *core_relo, int insn_idx) | |
5613 | { | |
5614 | struct reloc_desc *relos, *relo; | |
5615 | ||
5616 | relos = libbpf_reallocarray(prog->reloc_desc, | |
5617 | prog->nr_reloc + 1, sizeof(*relos)); | |
5618 | if (!relos) | |
5619 | return -ENOMEM; | |
5620 | relo = &relos[prog->nr_reloc]; | |
5621 | relo->type = RELO_CORE; | |
5622 | relo->insn_idx = insn_idx; | |
5623 | relo->core_relo = core_relo; | |
5624 | prog->reloc_desc = relos; | |
5625 | prog->nr_reloc++; | |
5626 | return 0; | |
5627 | } | |
5628 | ||
adb8fa19 MV |
5629 | static int bpf_core_resolve_relo(struct bpf_program *prog, |
5630 | const struct bpf_core_relo *relo, | |
5631 | int relo_idx, | |
5632 | const struct btf *local_btf, | |
5633 | struct hashmap *cand_cache, | |
5634 | struct bpf_core_relo_res *targ_res) | |
3ee4f533 | 5635 | { |
78c1f8d0 | 5636 | struct bpf_core_spec specs_scratch[3] = {}; |
3ee4f533 | 5637 | const void *type_key = u32_as_hash_key(relo->type_id); |
301ba4d7 | 5638 | struct bpf_core_cand_list *cands = NULL; |
3ee4f533 AS |
5639 | const char *prog_name = prog->name; |
5640 | const struct btf_type *local_type; | |
5641 | const char *local_name; | |
5642 | __u32 local_id = relo->type_id; | |
adb8fa19 | 5643 | int err; |
3ee4f533 AS |
5644 | |
5645 | local_type = btf__type_by_id(local_btf, local_id); | |
5646 | if (!local_type) | |
5647 | return -EINVAL; | |
5648 | ||
5649 | local_name = btf__name_by_offset(local_btf, local_type->name_off); | |
5650 | if (!local_name) | |
5651 | return -EINVAL; | |
5652 | ||
46334a0c | 5653 | if (relo->kind != BPF_CORE_TYPE_ID_LOCAL && |
3ee4f533 AS |
5654 | !hashmap__find(cand_cache, type_key, (void **)&cands)) { |
5655 | cands = bpf_core_find_cands(prog->obj, local_btf, local_id); | |
5656 | if (IS_ERR(cands)) { | |
5657 | pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n", | |
5658 | prog_name, relo_idx, local_id, btf_kind_str(local_type), | |
5659 | local_name, PTR_ERR(cands)); | |
5660 | return PTR_ERR(cands); | |
5661 | } | |
5662 | err = hashmap__set(cand_cache, type_key, cands, NULL, NULL); | |
5663 | if (err) { | |
5664 | bpf_core_free_cands(cands); | |
5665 | return err; | |
5666 | } | |
5667 | } | |
5668 | ||
adb8fa19 MV |
5669 | return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch, |
5670 | targ_res); | |
3ee4f533 AS |
5671 | } |
5672 | ||
ddc7c304 | 5673 | static int |
28b93c64 | 5674 | bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) |
ddc7c304 AN |
5675 | { |
5676 | const struct btf_ext_info_sec *sec; | |
adb8fa19 | 5677 | struct bpf_core_relo_res targ_res; |
28b93c64 | 5678 | const struct bpf_core_relo *rec; |
ddc7c304 AN |
5679 | const struct btf_ext_info *seg; |
5680 | struct hashmap_entry *entry; | |
5681 | struct hashmap *cand_cache = NULL; | |
5682 | struct bpf_program *prog; | |
adb8fa19 | 5683 | struct bpf_insn *insn; |
ddc7c304 | 5684 | const char *sec_name; |
11d5daa8 | 5685 | int i, err = 0, insn_idx, sec_idx, sec_num; |
ddc7c304 | 5686 | |
28b93c64 AN |
5687 | if (obj->btf_ext->core_relo_info.len == 0) |
5688 | return 0; | |
5689 | ||
0f7515ca AN |
5690 | if (targ_btf_path) { |
5691 | obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL); | |
e9fc3ce9 AN |
5692 | err = libbpf_get_error(obj->btf_vmlinux_override); |
5693 | if (err) { | |
0f7515ca AN |
5694 | pr_warn("failed to parse target BTF: %d\n", err); |
5695 | return err; | |
5696 | } | |
ddc7c304 AN |
5697 | } |
5698 | ||
5699 | cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL); | |
5700 | if (IS_ERR(cand_cache)) { | |
5701 | err = PTR_ERR(cand_cache); | |
5702 | goto out; | |
5703 | } | |
5704 | ||
28b93c64 | 5705 | seg = &obj->btf_ext->core_relo_info; |
11d5daa8 | 5706 | sec_num = 0; |
ddc7c304 | 5707 | for_each_btf_ext_sec(seg, sec) { |
11d5daa8 AN |
5708 | sec_idx = seg->sec_idxs[sec_num]; |
5709 | sec_num++; | |
5710 | ||
ddc7c304 AN |
5711 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); |
5712 | if (str_is_empty(sec_name)) { | |
5713 | err = -EINVAL; | |
5714 | goto out; | |
5715 | } | |
ddc7c304 | 5716 | |
11d5daa8 | 5717 | pr_debug("sec '%s': found %d CO-RE relocations\n", sec_name, sec->num_info); |
ddc7c304 AN |
5718 | |
5719 | for_each_btf_ext_rec(seg, sec, i, rec) { | |
adb8fa19 MV |
5720 | if (rec->insn_off % BPF_INSN_SZ) |
5721 | return -EINVAL; | |
db2b8b06 AN |
5722 | insn_idx = rec->insn_off / BPF_INSN_SZ; |
5723 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); | |
5724 | if (!prog) { | |
e89d57d9 AN |
5725 | /* When __weak subprog is "overridden" by another instance |
5726 | * of the subprog from a different object file, linker still | |
5727 | * appends all the .BTF.ext info that used to belong to that | |
5728 | * eliminated subprogram. | |
5729 | * This is similar to what x86-64 linker does for relocations. | |
5730 | * So just ignore such relocations just like we ignore | |
5731 | * subprog instructions when discovering subprograms. | |
5732 | */ | |
5733 | pr_debug("sec '%s': skipping CO-RE relocation #%d for insn #%d belonging to eliminated weak subprogram\n", | |
5734 | sec_name, i, insn_idx); | |
5735 | continue; | |
db2b8b06 | 5736 | } |
47f7cf63 AN |
5737 | /* no need to apply CO-RE relocation if the program is |
5738 | * not going to be loaded | |
5739 | */ | |
a3820c48 | 5740 | if (!prog->autoload) |
47f7cf63 | 5741 | continue; |
db2b8b06 | 5742 | |
adb8fa19 MV |
5743 | /* adjust insn_idx from section frame of reference to the local |
5744 | * program's frame of reference; (sub-)program code is not yet | |
5745 | * relocated, so it's enough to just subtract in-section offset | |
5746 | */ | |
5747 | insn_idx = insn_idx - prog->sec_insn_off; | |
5748 | if (insn_idx >= prog->insns_cnt) | |
5749 | return -EINVAL; | |
5750 | insn = &prog->insns[insn_idx]; | |
5751 | ||
5752 | if (prog->obj->gen_loader) { | |
5753 | err = record_relo_core(prog, rec, insn_idx); | |
5754 | if (err) { | |
5755 | pr_warn("prog '%s': relo #%d: failed to record relocation: %d\n", | |
5756 | prog->name, i, err); | |
5757 | goto out; | |
5758 | } | |
5759 | continue; | |
5760 | } | |
5761 | ||
5762 | err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res); | |
ddc7c304 | 5763 | if (err) { |
be18010e | 5764 | pr_warn("prog '%s': relo #%d: failed to relocate: %d\n", |
9c0f8cbd | 5765 | prog->name, i, err); |
ddc7c304 AN |
5766 | goto out; |
5767 | } | |
adb8fa19 MV |
5768 | |
5769 | err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res); | |
5770 | if (err) { | |
5771 | pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n", | |
5772 | prog->name, i, insn_idx, err); | |
5773 | goto out; | |
5774 | } | |
ddc7c304 AN |
5775 | } |
5776 | } | |
5777 | ||
5778 | out: | |
4f33a53d | 5779 | /* obj->btf_vmlinux and module BTFs are freed after object load */ |
0f7515ca AN |
5780 | btf__free(obj->btf_vmlinux_override); |
5781 | obj->btf_vmlinux_override = NULL; | |
5782 | ||
ddc7c304 AN |
5783 | if (!IS_ERR_OR_NULL(cand_cache)) { |
5784 | hashmap__for_each_entry(cand_cache, entry, i) { | |
5785 | bpf_core_free_cands(entry->value); | |
5786 | } | |
5787 | hashmap__free(cand_cache); | |
5788 | } | |
5789 | return err; | |
5790 | } | |
5791 | ||
c3c55696 AN |
5792 | /* Relocate data references within program code: |
5793 | * - map references; | |
5794 | * - global variable references; | |
5795 | * - extern references. | |
5796 | */ | |
48cca7e4 | 5797 | static int |
c3c55696 | 5798 | bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog) |
8a47a6c5 | 5799 | { |
c3c55696 | 5800 | int i; |
8a47a6c5 WN |
5801 | |
5802 | for (i = 0; i < prog->nr_reloc; i++) { | |
53f8dd43 | 5803 | struct reloc_desc *relo = &prog->reloc_desc[i]; |
166750bc | 5804 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; |
2e33efe3 | 5805 | struct extern_desc *ext; |
8a47a6c5 | 5806 | |
166750bc AN |
5807 | switch (relo->type) { |
5808 | case RELO_LD64: | |
e2fa0156 AS |
5809 | if (obj->gen_loader) { |
5810 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX; | |
5811 | insn[0].imm = relo->map_idx; | |
5812 | } else { | |
5813 | insn[0].src_reg = BPF_PSEUDO_MAP_FD; | |
5814 | insn[0].imm = obj->maps[relo->map_idx].fd; | |
5815 | } | |
166750bc AN |
5816 | break; |
5817 | case RELO_DATA: | |
166750bc | 5818 | insn[1].imm = insn[0].imm + relo->sym_off; |
e2fa0156 AS |
5819 | if (obj->gen_loader) { |
5820 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
5821 | insn[0].imm = relo->map_idx; | |
5822 | } else { | |
16e0c35c AN |
5823 | const struct bpf_map *map = &obj->maps[relo->map_idx]; |
5824 | ||
229fae38 | 5825 | if (map->skipped) { |
16e0c35c AN |
5826 | pr_warn("prog '%s': relo #%d: kernel doesn't support global data\n", |
5827 | prog->name, i); | |
5828 | return -ENOTSUP; | |
5829 | } | |
e2fa0156 AS |
5830 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; |
5831 | insn[0].imm = obj->maps[relo->map_idx].fd; | |
5832 | } | |
166750bc | 5833 | break; |
0c091e5c | 5834 | case RELO_EXTERN_VAR: |
2e33efe3 | 5835 | ext = &obj->externs[relo->sym_off]; |
1c0c7074 | 5836 | if (ext->type == EXT_KCFG) { |
e2fa0156 AS |
5837 | if (obj->gen_loader) { |
5838 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
5839 | insn[0].imm = obj->kconfig_map_idx; | |
5840 | } else { | |
5841 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; | |
5842 | insn[0].imm = obj->maps[obj->kconfig_map_idx].fd; | |
5843 | } | |
1c0c7074 AN |
5844 | insn[1].imm = ext->kcfg.data_off; |
5845 | } else /* EXT_KSYM */ { | |
2211c825 | 5846 | if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */ |
d370bbe1 | 5847 | insn[0].src_reg = BPF_PSEUDO_BTF_ID; |
284d2587 AN |
5848 | insn[0].imm = ext->ksym.kernel_btf_id; |
5849 | insn[1].imm = ext->ksym.kernel_btf_obj_fd; | |
2211c825 | 5850 | } else { /* typeless ksyms or unresolved typed ksyms */ |
d370bbe1 HL |
5851 | insn[0].imm = (__u32)ext->ksym.addr; |
5852 | insn[1].imm = ext->ksym.addr >> 32; | |
5853 | } | |
1c0c7074 | 5854 | } |
166750bc | 5855 | break; |
5bd022ec MKL |
5856 | case RELO_EXTERN_FUNC: |
5857 | ext = &obj->externs[relo->sym_off]; | |
5858 | insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL; | |
466b2e13 KKD |
5859 | if (ext->is_set) { |
5860 | insn[0].imm = ext->ksym.kernel_btf_id; | |
5861 | insn[0].off = ext->ksym.btf_fd_idx; | |
5862 | } else { /* unresolved weak kfunc */ | |
5863 | insn[0].imm = 0; | |
5864 | insn[0].off = 0; | |
5865 | } | |
5bd022ec | 5866 | break; |
53eddb5e | 5867 | case RELO_SUBPROG_ADDR: |
b1268826 AS |
5868 | if (insn[0].src_reg != BPF_PSEUDO_FUNC) { |
5869 | pr_warn("prog '%s': relo #%d: bad insn\n", | |
5870 | prog->name, i); | |
5871 | return -EINVAL; | |
5872 | } | |
5873 | /* handled already */ | |
53eddb5e | 5874 | break; |
166750bc | 5875 | case RELO_CALL: |
b1268826 | 5876 | /* handled already */ |
166750bc | 5877 | break; |
d0e92887 AS |
5878 | case RELO_CORE: |
5879 | /* will be handled by bpf_program_record_relos() */ | |
5880 | break; | |
166750bc | 5881 | default: |
9c0f8cbd AN |
5882 | pr_warn("prog '%s': relo #%d: bad relo type %d\n", |
5883 | prog->name, i, relo->type); | |
166750bc | 5884 | return -EINVAL; |
8a47a6c5 | 5885 | } |
8a47a6c5 WN |
5886 | } |
5887 | ||
c3c55696 AN |
5888 | return 0; |
5889 | } | |
5890 | ||
8505e870 AN |
5891 | static int adjust_prog_btf_ext_info(const struct bpf_object *obj, |
5892 | const struct bpf_program *prog, | |
5893 | const struct btf_ext_info *ext_info, | |
5894 | void **prog_info, __u32 *prog_rec_cnt, | |
5895 | __u32 *prog_rec_sz) | |
5896 | { | |
5897 | void *copy_start = NULL, *copy_end = NULL; | |
5898 | void *rec, *rec_end, *new_prog_info; | |
5899 | const struct btf_ext_info_sec *sec; | |
5900 | size_t old_sz, new_sz; | |
11d5daa8 | 5901 | int i, sec_num, sec_idx, off_adj; |
8505e870 | 5902 | |
11d5daa8 | 5903 | sec_num = 0; |
8505e870 | 5904 | for_each_btf_ext_sec(ext_info, sec) { |
11d5daa8 AN |
5905 | sec_idx = ext_info->sec_idxs[sec_num]; |
5906 | sec_num++; | |
5907 | if (prog->sec_idx != sec_idx) | |
8505e870 AN |
5908 | continue; |
5909 | ||
5910 | for_each_btf_ext_rec(ext_info, sec, i, rec) { | |
5911 | __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ; | |
5912 | ||
5913 | if (insn_off < prog->sec_insn_off) | |
5914 | continue; | |
5915 | if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt) | |
5916 | break; | |
5917 | ||
5918 | if (!copy_start) | |
5919 | copy_start = rec; | |
5920 | copy_end = rec + ext_info->rec_size; | |
5921 | } | |
5922 | ||
5923 | if (!copy_start) | |
5924 | return -ENOENT; | |
5925 | ||
5926 | /* append func/line info of a given (sub-)program to the main | |
5927 | * program func/line info | |
5928 | */ | |
8eb62958 | 5929 | old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size; |
8505e870 AN |
5930 | new_sz = old_sz + (copy_end - copy_start); |
5931 | new_prog_info = realloc(*prog_info, new_sz); | |
5932 | if (!new_prog_info) | |
5933 | return -ENOMEM; | |
5934 | *prog_info = new_prog_info; | |
5935 | *prog_rec_cnt = new_sz / ext_info->rec_size; | |
5936 | memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start); | |
5937 | ||
5938 | /* Kernel instruction offsets are in units of 8-byte | |
5939 | * instructions, while .BTF.ext instruction offsets generated | |
5940 | * by Clang are in units of bytes. So convert Clang offsets | |
5941 | * into kernel offsets and adjust offset according to program | |
5942 | * relocated position. | |
5943 | */ | |
5944 | off_adj = prog->sub_insn_off - prog->sec_insn_off; | |
5945 | rec = new_prog_info + old_sz; | |
5946 | rec_end = new_prog_info + new_sz; | |
5947 | for (; rec < rec_end; rec += ext_info->rec_size) { | |
5948 | __u32 *insn_off = rec; | |
5949 | ||
5950 | *insn_off = *insn_off / BPF_INSN_SZ + off_adj; | |
5951 | } | |
5952 | *prog_rec_sz = ext_info->rec_size; | |
5953 | return 0; | |
5954 | } | |
5955 | ||
5956 | return -ENOENT; | |
5957 | } | |
5958 | ||
5959 | static int | |
5960 | reloc_prog_func_and_line_info(const struct bpf_object *obj, | |
5961 | struct bpf_program *main_prog, | |
5962 | const struct bpf_program *prog) | |
5963 | { | |
5964 | int err; | |
5965 | ||
5966 | /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't | |
5967 | * supprot func/line info | |
5968 | */ | |
9ca1f56a | 5969 | if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC)) |
8505e870 AN |
5970 | return 0; |
5971 | ||
5972 | /* only attempt func info relocation if main program's func_info | |
5973 | * relocation was successful | |
5974 | */ | |
5975 | if (main_prog != prog && !main_prog->func_info) | |
5976 | goto line_info; | |
5977 | ||
5978 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info, | |
5979 | &main_prog->func_info, | |
5980 | &main_prog->func_info_cnt, | |
5981 | &main_prog->func_info_rec_size); | |
5982 | if (err) { | |
5983 | if (err != -ENOENT) { | |
5984 | pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n", | |
5985 | prog->name, err); | |
5986 | return err; | |
5987 | } | |
5988 | if (main_prog->func_info) { | |
5989 | /* | |
5990 | * Some info has already been found but has problem | |
5991 | * in the last btf_ext reloc. Must have to error out. | |
5992 | */ | |
5993 | pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name); | |
5994 | return err; | |
5995 | } | |
5996 | /* Have problem loading the very first info. Ignore the rest. */ | |
5997 | pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n", | |
5998 | prog->name); | |
5999 | } | |
6000 | ||
6001 | line_info: | |
6002 | /* don't relocate line info if main program's relocation failed */ | |
6003 | if (main_prog != prog && !main_prog->line_info) | |
6004 | return 0; | |
6005 | ||
6006 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info, | |
6007 | &main_prog->line_info, | |
6008 | &main_prog->line_info_cnt, | |
6009 | &main_prog->line_info_rec_size); | |
6010 | if (err) { | |
6011 | if (err != -ENOENT) { | |
6012 | pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n", | |
6013 | prog->name, err); | |
6014 | return err; | |
6015 | } | |
6016 | if (main_prog->line_info) { | |
6017 | /* | |
6018 | * Some info has already been found but has problem | |
6019 | * in the last btf_ext reloc. Must have to error out. | |
6020 | */ | |
6021 | pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name); | |
6022 | return err; | |
6023 | } | |
6024 | /* Have problem loading the very first info. Ignore the rest. */ | |
6025 | pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n", | |
6026 | prog->name); | |
6027 | } | |
6028 | return 0; | |
6029 | } | |
6030 | ||
c3c55696 AN |
6031 | static int cmp_relo_by_insn_idx(const void *key, const void *elem) |
6032 | { | |
6033 | size_t insn_idx = *(const size_t *)key; | |
6034 | const struct reloc_desc *relo = elem; | |
6035 | ||
6036 | if (insn_idx == relo->insn_idx) | |
6037 | return 0; | |
6038 | return insn_idx < relo->insn_idx ? -1 : 1; | |
6039 | } | |
6040 | ||
6041 | static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx) | |
6042 | { | |
2a6a9bf2 AN |
6043 | if (!prog->nr_reloc) |
6044 | return NULL; | |
c3c55696 AN |
6045 | return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc, |
6046 | sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx); | |
6047 | } | |
6048 | ||
b1268826 AS |
6049 | static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog) |
6050 | { | |
6051 | int new_cnt = main_prog->nr_reloc + subprog->nr_reloc; | |
6052 | struct reloc_desc *relos; | |
6053 | int i; | |
6054 | ||
6055 | if (main_prog == subprog) | |
6056 | return 0; | |
6057 | relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos)); | |
6058 | if (!relos) | |
6059 | return -ENOMEM; | |
2a6a9bf2 AN |
6060 | if (subprog->nr_reloc) |
6061 | memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc, | |
6062 | sizeof(*relos) * subprog->nr_reloc); | |
b1268826 AS |
6063 | |
6064 | for (i = main_prog->nr_reloc; i < new_cnt; i++) | |
6065 | relos[i].insn_idx += subprog->sub_insn_off; | |
6066 | /* After insn_idx adjustment the 'relos' array is still sorted | |
6067 | * by insn_idx and doesn't break bsearch. | |
6068 | */ | |
6069 | main_prog->reloc_desc = relos; | |
6070 | main_prog->nr_reloc = new_cnt; | |
6071 | return 0; | |
6072 | } | |
6073 | ||
c3c55696 AN |
6074 | static int |
6075 | bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog, | |
6076 | struct bpf_program *prog) | |
6077 | { | |
6078 | size_t sub_insn_idx, insn_idx, new_cnt; | |
6079 | struct bpf_program *subprog; | |
6080 | struct bpf_insn *insns, *insn; | |
6081 | struct reloc_desc *relo; | |
6082 | int err; | |
6083 | ||
6084 | err = reloc_prog_func_and_line_info(obj, main_prog, prog); | |
6085 | if (err) | |
6086 | return err; | |
6087 | ||
6088 | for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) { | |
6089 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
53eddb5e | 6090 | if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn)) |
c3c55696 AN |
6091 | continue; |
6092 | ||
6093 | relo = find_prog_insn_relo(prog, insn_idx); | |
b1268826 AS |
6094 | if (relo && relo->type == RELO_EXTERN_FUNC) |
6095 | /* kfunc relocations will be handled later | |
6096 | * in bpf_object__relocate_data() | |
6097 | */ | |
6098 | continue; | |
53eddb5e | 6099 | if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) { |
c3c55696 AN |
6100 | pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n", |
6101 | prog->name, insn_idx, relo->type); | |
6102 | return -LIBBPF_ERRNO__RELOC; | |
6103 | } | |
6104 | if (relo) { | |
6105 | /* sub-program instruction index is a combination of | |
6106 | * an offset of a symbol pointed to by relocation and | |
6107 | * call instruction's imm field; for global functions, | |
6108 | * call always has imm = -1, but for static functions | |
6109 | * relocation is against STT_SECTION and insn->imm | |
6110 | * points to a start of a static function | |
53eddb5e YS |
6111 | * |
6112 | * for subprog addr relocation, the relo->sym_off + insn->imm is | |
6113 | * the byte offset in the corresponding section. | |
c3c55696 | 6114 | */ |
53eddb5e YS |
6115 | if (relo->type == RELO_CALL) |
6116 | sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1; | |
6117 | else | |
6118 | sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ; | |
6119 | } else if (insn_is_pseudo_func(insn)) { | |
6120 | /* | |
6121 | * RELO_SUBPROG_ADDR relo is always emitted even if both | |
6122 | * functions are in the same section, so it shouldn't reach here. | |
6123 | */ | |
6124 | pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n", | |
6125 | prog->name, insn_idx); | |
6126 | return -LIBBPF_ERRNO__RELOC; | |
c3c55696 AN |
6127 | } else { |
6128 | /* if subprogram call is to a static function within | |
6129 | * the same ELF section, there won't be any relocation | |
6130 | * emitted, but it also means there is no additional | |
6131 | * offset necessary, insns->imm is relative to | |
6132 | * instruction's original position within the section | |
6133 | */ | |
6134 | sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1; | |
6135 | } | |
6136 | ||
6137 | /* we enforce that sub-programs should be in .text section */ | |
6138 | subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx); | |
6139 | if (!subprog) { | |
6140 | pr_warn("prog '%s': no .text section found yet sub-program call exists\n", | |
6141 | prog->name); | |
6142 | return -LIBBPF_ERRNO__RELOC; | |
6143 | } | |
6144 | ||
6145 | /* if it's the first call instruction calling into this | |
6146 | * subprogram (meaning this subprog hasn't been processed | |
6147 | * yet) within the context of current main program: | |
6148 | * - append it at the end of main program's instructions blog; | |
6149 | * - process is recursively, while current program is put on hold; | |
6150 | * - if that subprogram calls some other not yet processes | |
6151 | * subprogram, same thing will happen recursively until | |
6152 | * there are no more unprocesses subprograms left to append | |
6153 | * and relocate. | |
6154 | */ | |
6155 | if (subprog->sub_insn_off == 0) { | |
6156 | subprog->sub_insn_off = main_prog->insns_cnt; | |
6157 | ||
6158 | new_cnt = main_prog->insns_cnt + subprog->insns_cnt; | |
6159 | insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns)); | |
6160 | if (!insns) { | |
6161 | pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name); | |
6162 | return -ENOMEM; | |
6163 | } | |
6164 | main_prog->insns = insns; | |
6165 | main_prog->insns_cnt = new_cnt; | |
6166 | ||
6167 | memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns, | |
6168 | subprog->insns_cnt * sizeof(*insns)); | |
6169 | ||
6170 | pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n", | |
6171 | main_prog->name, subprog->insns_cnt, subprog->name); | |
6172 | ||
b1268826 AS |
6173 | /* The subprog insns are now appended. Append its relos too. */ |
6174 | err = append_subprog_relos(main_prog, subprog); | |
6175 | if (err) | |
6176 | return err; | |
c3c55696 AN |
6177 | err = bpf_object__reloc_code(obj, main_prog, subprog); |
6178 | if (err) | |
6179 | return err; | |
6180 | } | |
6181 | ||
6182 | /* main_prog->insns memory could have been re-allocated, so | |
6183 | * calculate pointer again | |
6184 | */ | |
6185 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
6186 | /* calculate correct instruction position within current main | |
6187 | * prog; each main prog can have a different set of | |
6188 | * subprograms appended (potentially in different order as | |
6189 | * well), so position of any subprog can be different for | |
6190 | * different main programs */ | |
6191 | insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1; | |
6192 | ||
c3c55696 AN |
6193 | pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n", |
6194 | prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off); | |
6195 | } | |
6196 | ||
6197 | return 0; | |
6198 | } | |
6199 | ||
6200 | /* | |
6201 | * Relocate sub-program calls. | |
6202 | * | |
6203 | * Algorithm operates as follows. Each entry-point BPF program (referred to as | |
6204 | * main prog) is processed separately. For each subprog (non-entry functions, | |
6205 | * that can be called from either entry progs or other subprogs) gets their | |
6206 | * sub_insn_off reset to zero. This serves as indicator that this subprogram | |
6207 | * hasn't been yet appended and relocated within current main prog. Once its | |
6208 | * relocated, sub_insn_off will point at the position within current main prog | |
6209 | * where given subprog was appended. This will further be used to relocate all | |
6210 | * the call instructions jumping into this subprog. | |
6211 | * | |
6212 | * We start with main program and process all call instructions. If the call | |
6213 | * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off | |
6214 | * is zero), subprog instructions are appended at the end of main program's | |
6215 | * instruction array. Then main program is "put on hold" while we recursively | |
6216 | * process newly appended subprogram. If that subprogram calls into another | |
6217 | * subprogram that hasn't been appended, new subprogram is appended again to | |
6218 | * the *main* prog's instructions (subprog's instructions are always left | |
6219 | * untouched, as they need to be in unmodified state for subsequent main progs | |
6220 | * and subprog instructions are always sent only as part of a main prog) and | |
6221 | * the process continues recursively. Once all the subprogs called from a main | |
6222 | * prog or any of its subprogs are appended (and relocated), all their | |
6223 | * positions within finalized instructions array are known, so it's easy to | |
6224 | * rewrite call instructions with correct relative offsets, corresponding to | |
6225 | * desired target subprog. | |
6226 | * | |
6227 | * Its important to realize that some subprogs might not be called from some | |
6228 | * main prog and any of its called/used subprogs. Those will keep their | |
6229 | * subprog->sub_insn_off as zero at all times and won't be appended to current | |
6230 | * main prog and won't be relocated within the context of current main prog. | |
6231 | * They might still be used from other main progs later. | |
6232 | * | |
6233 | * Visually this process can be shown as below. Suppose we have two main | |
6234 | * programs mainA and mainB and BPF object contains three subprogs: subA, | |
6235 | * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and | |
6236 | * subC both call subB: | |
6237 | * | |
6238 | * +--------+ +-------+ | |
6239 | * | v v | | |
6240 | * +--+---+ +--+-+-+ +---+--+ | |
6241 | * | subA | | subB | | subC | | |
6242 | * +--+---+ +------+ +---+--+ | |
6243 | * ^ ^ | |
6244 | * | | | |
6245 | * +---+-------+ +------+----+ | |
6246 | * | mainA | | mainB | | |
6247 | * +-----------+ +-----------+ | |
6248 | * | |
6249 | * We'll start relocating mainA, will find subA, append it and start | |
6250 | * processing sub A recursively: | |
6251 | * | |
6252 | * +-----------+------+ | |
6253 | * | mainA | subA | | |
6254 | * +-----------+------+ | |
6255 | * | |
6256 | * At this point we notice that subB is used from subA, so we append it and | |
6257 | * relocate (there are no further subcalls from subB): | |
6258 | * | |
6259 | * +-----------+------+------+ | |
6260 | * | mainA | subA | subB | | |
6261 | * +-----------+------+------+ | |
6262 | * | |
6263 | * At this point, we relocate subA calls, then go one level up and finish with | |
6264 | * relocatin mainA calls. mainA is done. | |
6265 | * | |
6266 | * For mainB process is similar but results in different order. We start with | |
6267 | * mainB and skip subA and subB, as mainB never calls them (at least | |
6268 | * directly), but we see subC is needed, so we append and start processing it: | |
6269 | * | |
6270 | * +-----------+------+ | |
6271 | * | mainB | subC | | |
6272 | * +-----------+------+ | |
6273 | * Now we see subC needs subB, so we go back to it, append and relocate it: | |
6274 | * | |
6275 | * +-----------+------+------+ | |
6276 | * | mainB | subC | subB | | |
6277 | * +-----------+------+------+ | |
6278 | * | |
6279 | * At this point we unwind recursion, relocate calls in subC, then in mainB. | |
6280 | */ | |
6281 | static int | |
6282 | bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog) | |
6283 | { | |
6284 | struct bpf_program *subprog; | |
d3d93e34 | 6285 | int i, err; |
c3c55696 | 6286 | |
c3c55696 AN |
6287 | /* mark all subprogs as not relocated (yet) within the context of |
6288 | * current main program | |
6289 | */ | |
6290 | for (i = 0; i < obj->nr_programs; i++) { | |
6291 | subprog = &obj->programs[i]; | |
6292 | if (!prog_is_subprog(obj, subprog)) | |
6293 | continue; | |
6294 | ||
6295 | subprog->sub_insn_off = 0; | |
c3c55696 AN |
6296 | } |
6297 | ||
6298 | err = bpf_object__reloc_code(obj, prog, prog); | |
6299 | if (err) | |
6300 | return err; | |
6301 | ||
6302 | ||
8a47a6c5 WN |
6303 | return 0; |
6304 | } | |
6305 | ||
67234743 AS |
6306 | static void |
6307 | bpf_object__free_relocs(struct bpf_object *obj) | |
6308 | { | |
6309 | struct bpf_program *prog; | |
6310 | int i; | |
6311 | ||
6312 | /* free up relocation descriptors */ | |
6313 | for (i = 0; i < obj->nr_programs; i++) { | |
6314 | prog = &obj->programs[i]; | |
6315 | zfree(&prog->reloc_desc); | |
6316 | prog->nr_reloc = 0; | |
6317 | } | |
6318 | } | |
6319 | ||
d0e92887 AS |
6320 | static int cmp_relocs(const void *_a, const void *_b) |
6321 | { | |
6322 | const struct reloc_desc *a = _a; | |
6323 | const struct reloc_desc *b = _b; | |
6324 | ||
6325 | if (a->insn_idx != b->insn_idx) | |
6326 | return a->insn_idx < b->insn_idx ? -1 : 1; | |
6327 | ||
6328 | /* no two relocations should have the same insn_idx, but ... */ | |
6329 | if (a->type != b->type) | |
6330 | return a->type < b->type ? -1 : 1; | |
6331 | ||
6332 | return 0; | |
6333 | } | |
6334 | ||
6335 | static void bpf_object__sort_relos(struct bpf_object *obj) | |
6336 | { | |
6337 | int i; | |
6338 | ||
6339 | for (i = 0; i < obj->nr_programs; i++) { | |
6340 | struct bpf_program *p = &obj->programs[i]; | |
6341 | ||
6342 | if (!p->nr_reloc) | |
6343 | continue; | |
6344 | ||
6345 | qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs); | |
6346 | } | |
6347 | } | |
6348 | ||
8a47a6c5 | 6349 | static int |
ddc7c304 | 6350 | bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path) |
8a47a6c5 WN |
6351 | { |
6352 | struct bpf_program *prog; | |
b1268826 | 6353 | size_t i, j; |
8a47a6c5 WN |
6354 | int err; |
6355 | ||
ddc7c304 AN |
6356 | if (obj->btf_ext) { |
6357 | err = bpf_object__relocate_core(obj, targ_btf_path); | |
6358 | if (err) { | |
be18010e KW |
6359 | pr_warn("failed to perform CO-RE relocations: %d\n", |
6360 | err); | |
ddc7c304 AN |
6361 | return err; |
6362 | } | |
d0e92887 AS |
6363 | if (obj->gen_loader) |
6364 | bpf_object__sort_relos(obj); | |
ddc7c304 | 6365 | } |
b1268826 AS |
6366 | |
6367 | /* Before relocating calls pre-process relocations and mark | |
6368 | * few ld_imm64 instructions that points to subprogs. | |
6369 | * Otherwise bpf_object__reloc_code() later would have to consider | |
6370 | * all ld_imm64 insns as relocation candidates. That would | |
6371 | * reduce relocation speed, since amount of find_prog_insn_relo() | |
6372 | * would increase and most of them will fail to find a relo. | |
9173cac3 AN |
6373 | */ |
6374 | for (i = 0; i < obj->nr_programs; i++) { | |
6375 | prog = &obj->programs[i]; | |
b1268826 AS |
6376 | for (j = 0; j < prog->nr_reloc; j++) { |
6377 | struct reloc_desc *relo = &prog->reloc_desc[j]; | |
6378 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; | |
6379 | ||
6380 | /* mark the insn, so it's recognized by insn_is_pseudo_func() */ | |
6381 | if (relo->type == RELO_SUBPROG_ADDR) | |
6382 | insn[0].src_reg = BPF_PSEUDO_FUNC; | |
9173cac3 | 6383 | } |
9173cac3 | 6384 | } |
b1268826 AS |
6385 | |
6386 | /* relocate subprogram calls and append used subprograms to main | |
c3c55696 AN |
6387 | * programs; each copy of subprogram code needs to be relocated |
6388 | * differently for each main program, because its code location might | |
b1268826 AS |
6389 | * have changed. |
6390 | * Append subprog relos to main programs to allow data relos to be | |
6391 | * processed after text is completely relocated. | |
9173cac3 | 6392 | */ |
8a47a6c5 WN |
6393 | for (i = 0; i < obj->nr_programs; i++) { |
6394 | prog = &obj->programs[i]; | |
c3c55696 AN |
6395 | /* sub-program's sub-calls are relocated within the context of |
6396 | * its main program only | |
6397 | */ | |
6398 | if (prog_is_subprog(obj, prog)) | |
9173cac3 | 6399 | continue; |
a3820c48 | 6400 | if (!prog->autoload) |
16e0c35c | 6401 | continue; |
8a47a6c5 | 6402 | |
c3c55696 | 6403 | err = bpf_object__relocate_calls(obj, prog); |
8a47a6c5 | 6404 | if (err) { |
9c0f8cbd AN |
6405 | pr_warn("prog '%s': failed to relocate calls: %d\n", |
6406 | prog->name, err); | |
8a47a6c5 WN |
6407 | return err; |
6408 | } | |
6409 | } | |
b1268826 AS |
6410 | /* Process data relos for main programs */ |
6411 | for (i = 0; i < obj->nr_programs; i++) { | |
6412 | prog = &obj->programs[i]; | |
6413 | if (prog_is_subprog(obj, prog)) | |
6414 | continue; | |
a3820c48 | 6415 | if (!prog->autoload) |
16e0c35c | 6416 | continue; |
b1268826 AS |
6417 | err = bpf_object__relocate_data(obj, prog); |
6418 | if (err) { | |
6419 | pr_warn("prog '%s': failed to relocate data references: %d\n", | |
6420 | prog->name, err); | |
6421 | return err; | |
6422 | } | |
6423 | } | |
67234743 AS |
6424 | if (!obj->gen_loader) |
6425 | bpf_object__free_relocs(obj); | |
8a47a6c5 WN |
6426 | return 0; |
6427 | } | |
6428 | ||
646f02ff | 6429 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 6430 | Elf64_Shdr *shdr, Elf_Data *data); |
646f02ff AN |
6431 | |
6432 | static int bpf_object__collect_map_relos(struct bpf_object *obj, | |
ad23b723 | 6433 | Elf64_Shdr *shdr, Elf_Data *data) |
646f02ff | 6434 | { |
15728ad3 AN |
6435 | const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *); |
6436 | int i, j, nrels, new_sz; | |
063e6881 | 6437 | const struct btf_var_secinfo *vi = NULL; |
646f02ff | 6438 | const struct btf_type *sec, *var, *def; |
341ac5ff HC |
6439 | struct bpf_map *map = NULL, *targ_map = NULL; |
6440 | struct bpf_program *targ_prog = NULL; | |
6441 | bool is_prog_array, is_map_in_map; | |
646f02ff | 6442 | const struct btf_member *member; |
341ac5ff | 6443 | const char *name, *mname, *type; |
646f02ff | 6444 | unsigned int moff; |
ad23b723 AN |
6445 | Elf64_Sym *sym; |
6446 | Elf64_Rel *rel; | |
646f02ff AN |
6447 | void *tmp; |
6448 | ||
6449 | if (!obj->efile.btf_maps_sec_btf_id || !obj->btf) | |
6450 | return -EINVAL; | |
6451 | sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id); | |
6452 | if (!sec) | |
6453 | return -EINVAL; | |
6454 | ||
646f02ff AN |
6455 | nrels = shdr->sh_size / shdr->sh_entsize; |
6456 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
6457 | rel = elf_rel_by_idx(data, i); |
6458 | if (!rel) { | |
646f02ff AN |
6459 | pr_warn(".maps relo #%d: failed to get ELF relo\n", i); |
6460 | return -LIBBPF_ERRNO__FORMAT; | |
6461 | } | |
ad23b723 AN |
6462 | |
6463 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); | |
6464 | if (!sym) { | |
646f02ff | 6465 | pr_warn(".maps relo #%d: symbol %zx not found\n", |
ad23b723 | 6466 | i, (size_t)ELF64_R_SYM(rel->r_info)); |
646f02ff AN |
6467 | return -LIBBPF_ERRNO__FORMAT; |
6468 | } | |
ad23b723 | 6469 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
646f02ff | 6470 | |
ad23b723 AN |
6471 | pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n", |
6472 | i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value, | |
6473 | (size_t)rel->r_offset, sym->st_name, name); | |
646f02ff AN |
6474 | |
6475 | for (j = 0; j < obj->nr_maps; j++) { | |
6476 | map = &obj->maps[j]; | |
6477 | if (map->sec_idx != obj->efile.btf_maps_shndx) | |
6478 | continue; | |
6479 | ||
6480 | vi = btf_var_secinfos(sec) + map->btf_var_idx; | |
ad23b723 AN |
6481 | if (vi->offset <= rel->r_offset && |
6482 | rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size) | |
646f02ff AN |
6483 | break; |
6484 | } | |
6485 | if (j == obj->nr_maps) { | |
ad23b723 AN |
6486 | pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n", |
6487 | i, name, (size_t)rel->r_offset); | |
646f02ff AN |
6488 | return -EINVAL; |
6489 | } | |
6490 | ||
341ac5ff HC |
6491 | is_map_in_map = bpf_map_type__is_map_in_map(map->def.type); |
6492 | is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY; | |
6493 | type = is_map_in_map ? "map" : "prog"; | |
6494 | if (is_map_in_map) { | |
6495 | if (sym->st_shndx != obj->efile.btf_maps_shndx) { | |
6496 | pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n", | |
6497 | i, name); | |
6498 | return -LIBBPF_ERRNO__RELOC; | |
6499 | } | |
6500 | if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS && | |
6501 | map->def.key_size != sizeof(int)) { | |
6502 | pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n", | |
6503 | i, map->name, sizeof(int)); | |
6504 | return -EINVAL; | |
6505 | } | |
6506 | targ_map = bpf_object__find_map_by_name(obj, name); | |
6507 | if (!targ_map) { | |
6508 | pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n", | |
6509 | i, name); | |
6510 | return -ESRCH; | |
6511 | } | |
6512 | } else if (is_prog_array) { | |
6513 | targ_prog = bpf_object__find_program_by_name(obj, name); | |
6514 | if (!targ_prog) { | |
6515 | pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n", | |
6516 | i, name); | |
6517 | return -ESRCH; | |
6518 | } | |
6519 | if (targ_prog->sec_idx != sym->st_shndx || | |
6520 | targ_prog->sec_insn_off * 8 != sym->st_value || | |
6521 | prog_is_subprog(obj, targ_prog)) { | |
6522 | pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n", | |
6523 | i, name); | |
6524 | return -LIBBPF_ERRNO__RELOC; | |
6525 | } | |
6526 | } else { | |
646f02ff AN |
6527 | return -EINVAL; |
6528 | } | |
6529 | ||
646f02ff AN |
6530 | var = btf__type_by_id(obj->btf, vi->type); |
6531 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
6532 | if (btf_vlen(def) == 0) | |
6533 | return -EINVAL; | |
6534 | member = btf_members(def) + btf_vlen(def) - 1; | |
6535 | mname = btf__name_by_offset(obj->btf, member->name_off); | |
6536 | if (strcmp(mname, "values")) | |
6537 | return -EINVAL; | |
6538 | ||
6539 | moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8; | |
ad23b723 | 6540 | if (rel->r_offset - vi->offset < moff) |
646f02ff AN |
6541 | return -EINVAL; |
6542 | ||
ad23b723 | 6543 | moff = rel->r_offset - vi->offset - moff; |
15728ad3 AN |
6544 | /* here we use BPF pointer size, which is always 64 bit, as we |
6545 | * are parsing ELF that was built for BPF target | |
6546 | */ | |
6547 | if (moff % bpf_ptr_sz) | |
646f02ff | 6548 | return -EINVAL; |
15728ad3 | 6549 | moff /= bpf_ptr_sz; |
646f02ff AN |
6550 | if (moff >= map->init_slots_sz) { |
6551 | new_sz = moff + 1; | |
029258d7 | 6552 | tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz); |
646f02ff AN |
6553 | if (!tmp) |
6554 | return -ENOMEM; | |
6555 | map->init_slots = tmp; | |
6556 | memset(map->init_slots + map->init_slots_sz, 0, | |
15728ad3 | 6557 | (new_sz - map->init_slots_sz) * host_ptr_sz); |
646f02ff AN |
6558 | map->init_slots_sz = new_sz; |
6559 | } | |
341ac5ff | 6560 | map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog; |
646f02ff | 6561 | |
341ac5ff HC |
6562 | pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n", |
6563 | i, map->name, moff, type, name); | |
646f02ff AN |
6564 | } |
6565 | ||
6566 | return 0; | |
6567 | } | |
590a0088 | 6568 | |
c3c55696 AN |
6569 | static int bpf_object__collect_relos(struct bpf_object *obj) |
6570 | { | |
6571 | int i, err; | |
34090915 | 6572 | |
25bbbd7a AN |
6573 | for (i = 0; i < obj->efile.sec_cnt; i++) { |
6574 | struct elf_sec_desc *sec_desc = &obj->efile.secs[i]; | |
6575 | Elf64_Shdr *shdr; | |
6576 | Elf_Data *data; | |
6577 | int idx; | |
6578 | ||
6579 | if (sec_desc->sec_type != SEC_RELO) | |
6580 | continue; | |
6581 | ||
6582 | shdr = sec_desc->shdr; | |
6583 | data = sec_desc->data; | |
6584 | idx = shdr->sh_info; | |
34090915 WN |
6585 | |
6586 | if (shdr->sh_type != SHT_REL) { | |
be18010e | 6587 | pr_warn("internal error at %d\n", __LINE__); |
6371ca3b | 6588 | return -LIBBPF_ERRNO__INTERNAL; |
34090915 WN |
6589 | } |
6590 | ||
c3c55696 | 6591 | if (idx == obj->efile.st_ops_shndx) |
646f02ff | 6592 | err = bpf_object__collect_st_ops_relos(obj, shdr, data); |
c3c55696 | 6593 | else if (idx == obj->efile.btf_maps_shndx) |
646f02ff | 6594 | err = bpf_object__collect_map_relos(obj, shdr, data); |
c3c55696 AN |
6595 | else |
6596 | err = bpf_object__collect_prog_relos(obj, shdr, data); | |
34090915 | 6597 | if (err) |
6371ca3b | 6598 | return err; |
34090915 | 6599 | } |
c3c55696 | 6600 | |
d0e92887 | 6601 | bpf_object__sort_relos(obj); |
34090915 WN |
6602 | return 0; |
6603 | } | |
6604 | ||
109cea5a AN |
6605 | static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id) |
6606 | { | |
9b2f6fec | 6607 | if (BPF_CLASS(insn->code) == BPF_JMP && |
109cea5a AN |
6608 | BPF_OP(insn->code) == BPF_CALL && |
6609 | BPF_SRC(insn->code) == BPF_K && | |
9b2f6fec AN |
6610 | insn->src_reg == 0 && |
6611 | insn->dst_reg == 0) { | |
6612 | *func_id = insn->imm; | |
109cea5a AN |
6613 | return true; |
6614 | } | |
6615 | return false; | |
6616 | } | |
6617 | ||
42869d28 | 6618 | static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog) |
109cea5a AN |
6619 | { |
6620 | struct bpf_insn *insn = prog->insns; | |
6621 | enum bpf_func_id func_id; | |
6622 | int i; | |
6623 | ||
67234743 AS |
6624 | if (obj->gen_loader) |
6625 | return 0; | |
6626 | ||
109cea5a AN |
6627 | for (i = 0; i < prog->insns_cnt; i++, insn++) { |
6628 | if (!insn_is_helper_call(insn, &func_id)) | |
6629 | continue; | |
6630 | ||
6631 | /* on kernels that don't yet support | |
6632 | * bpf_probe_read_{kernel,user}[_str] helpers, fall back | |
6633 | * to bpf_probe_read() which works well for old kernels | |
6634 | */ | |
6635 | switch (func_id) { | |
6636 | case BPF_FUNC_probe_read_kernel: | |
6637 | case BPF_FUNC_probe_read_user: | |
9ca1f56a | 6638 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
6639 | insn->imm = BPF_FUNC_probe_read; |
6640 | break; | |
6641 | case BPF_FUNC_probe_read_kernel_str: | |
6642 | case BPF_FUNC_probe_read_user_str: | |
9ca1f56a | 6643 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
6644 | insn->imm = BPF_FUNC_probe_read_str; |
6645 | break; | |
6646 | default: | |
6647 | break; | |
6648 | } | |
6649 | } | |
6650 | return 0; | |
6651 | } | |
6652 | ||
15ea31fa AN |
6653 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
6654 | int *btf_obj_fd, int *btf_type_id); | |
12d9466d | 6655 | |
4fa5bcfe AN |
6656 | /* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */ |
6657 | static int libbpf_prepare_prog_load(struct bpf_program *prog, | |
6658 | struct bpf_prog_load_opts *opts, long cookie) | |
12d9466d | 6659 | { |
15ea31fa AN |
6660 | enum sec_def_flags def = cookie; |
6661 | ||
12d9466d | 6662 | /* old kernels might not support specifying expected_attach_type */ |
15ea31fa | 6663 | if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE)) |
d10ef2b8 | 6664 | opts->expected_attach_type = 0; |
12d9466d | 6665 | |
15ea31fa | 6666 | if (def & SEC_SLEEPABLE) |
d10ef2b8 | 6667 | opts->prog_flags |= BPF_F_SLEEPABLE; |
12d9466d | 6668 | |
082c4bfb LB |
6669 | if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS)) |
6670 | opts->prog_flags |= BPF_F_XDP_HAS_FRAGS; | |
6671 | ||
4a4d4cee LB |
6672 | if (def & SEC_DEPRECATED) |
6673 | 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", | |
6674 | prog->sec_name); | |
6675 | ||
12d9466d AN |
6676 | if ((prog->type == BPF_PROG_TYPE_TRACING || |
6677 | prog->type == BPF_PROG_TYPE_LSM || | |
6678 | prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) { | |
6679 | int btf_obj_fd = 0, btf_type_id = 0, err; | |
15ea31fa | 6680 | const char *attach_name; |
12d9466d | 6681 | |
15ea31fa AN |
6682 | attach_name = strchr(prog->sec_name, '/') + 1; |
6683 | err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id); | |
12d9466d AN |
6684 | if (err) |
6685 | return err; | |
6686 | ||
6687 | /* cache resolved BTF FD and BTF type ID in the prog */ | |
6688 | prog->attach_btf_obj_fd = btf_obj_fd; | |
6689 | prog->attach_btf_id = btf_type_id; | |
6690 | ||
6691 | /* but by now libbpf common logic is not utilizing | |
6692 | * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because | |
d10ef2b8 AN |
6693 | * this callback is called after opts were populated by |
6694 | * libbpf, so this callback has to update opts explicitly here | |
12d9466d | 6695 | */ |
d10ef2b8 AN |
6696 | opts->attach_btf_obj_fd = btf_obj_fd; |
6697 | opts->attach_btf_id = btf_type_id; | |
12d9466d AN |
6698 | } |
6699 | return 0; | |
6700 | } | |
6701 | ||
be2f2d16 AN |
6702 | static int bpf_object_load_prog_instance(struct bpf_object *obj, struct bpf_program *prog, |
6703 | struct bpf_insn *insns, int insns_cnt, | |
6704 | const char *license, __u32 kern_version, | |
6705 | int *prog_fd) | |
55cffde2 | 6706 | { |
d10ef2b8 AN |
6707 | LIBBPF_OPTS(bpf_prog_load_opts, load_attr); |
6708 | const char *prog_name = NULL; | |
1ce6a9fc | 6709 | char *cp, errmsg[STRERR_BUFSIZE]; |
8395f320 | 6710 | size_t log_buf_size = 0; |
b3ce9079 | 6711 | char *log_buf = NULL, *tmp; |
12d9466d | 6712 | int btf_fd, ret, err; |
b3ce9079 AN |
6713 | bool own_log_buf = true; |
6714 | __u32 log_level = prog->log_level; | |
55cffde2 | 6715 | |
80b2b5c3 AM |
6716 | if (prog->type == BPF_PROG_TYPE_UNSPEC) { |
6717 | /* | |
6718 | * The program type must be set. Most likely we couldn't find a proper | |
6719 | * section definition at load time, and thus we didn't infer the type. | |
6720 | */ | |
6721 | pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n", | |
6722 | prog->name, prog->sec_name); | |
6723 | return -EINVAL; | |
6724 | } | |
6725 | ||
fba01a06 AN |
6726 | if (!insns || !insns_cnt) |
6727 | return -EINVAL; | |
6728 | ||
12d9466d | 6729 | load_attr.expected_attach_type = prog->expected_attach_type; |
25bbbd7a | 6730 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
d10ef2b8 | 6731 | prog_name = prog->name; |
12d9466d AN |
6732 | load_attr.attach_prog_fd = prog->attach_prog_fd; |
6733 | load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd; | |
6aef10a4 AN |
6734 | load_attr.attach_btf_id = prog->attach_btf_id; |
6735 | load_attr.kern_version = kern_version; | |
6736 | load_attr.prog_ifindex = prog->prog_ifindex; | |
6737 | ||
0f0e55d8 | 6738 | /* specify func_info/line_info only if kernel supports them */ |
25bbbd7a AN |
6739 | btf_fd = bpf_object__btf_fd(obj); |
6740 | if (btf_fd >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) { | |
0f0e55d8 AN |
6741 | load_attr.prog_btf_fd = btf_fd; |
6742 | load_attr.func_info = prog->func_info; | |
6743 | load_attr.func_info_rec_size = prog->func_info_rec_size; | |
6744 | load_attr.func_info_cnt = prog->func_info_cnt; | |
6745 | load_attr.line_info = prog->line_info; | |
6746 | load_attr.line_info_rec_size = prog->line_info_rec_size; | |
6747 | load_attr.line_info_cnt = prog->line_info_cnt; | |
6748 | } | |
b3ce9079 | 6749 | load_attr.log_level = log_level; |
04656198 | 6750 | load_attr.prog_flags = prog->prog_flags; |
25bbbd7a | 6751 | load_attr.fd_array = obj->fd_array; |
55cffde2 | 6752 | |
12d9466d | 6753 | /* adjust load_attr if sec_def provides custom preload callback */ |
4fa5bcfe AN |
6754 | if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) { |
6755 | err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie); | |
12d9466d AN |
6756 | if (err < 0) { |
6757 | pr_warn("prog '%s': failed to prepare load attributes: %d\n", | |
6758 | prog->name, err); | |
6759 | return err; | |
6760 | } | |
6761 | } | |
6762 | ||
25bbbd7a | 6763 | if (obj->gen_loader) { |
d10ef2b8 AN |
6764 | bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name, |
6765 | license, insns, insns_cnt, &load_attr, | |
25bbbd7a | 6766 | prog - obj->programs); |
be2f2d16 | 6767 | *prog_fd = -1; |
67234743 AS |
6768 | return 0; |
6769 | } | |
8395f320 | 6770 | |
b3ce9079 AN |
6771 | retry_load: |
6772 | /* if log_level is zero, we don't request logs initiallly even if | |
6773 | * custom log_buf is specified; if the program load fails, then we'll | |
6774 | * bump log_level to 1 and use either custom log_buf or we'll allocate | |
6775 | * our own and retry the load to get details on what failed | |
6776 | */ | |
6777 | if (log_level) { | |
6778 | if (prog->log_buf) { | |
6779 | log_buf = prog->log_buf; | |
6780 | log_buf_size = prog->log_size; | |
6781 | own_log_buf = false; | |
6782 | } else if (obj->log_buf) { | |
6783 | log_buf = obj->log_buf; | |
6784 | log_buf_size = obj->log_size; | |
6785 | own_log_buf = false; | |
6786 | } else { | |
6787 | log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2); | |
6788 | tmp = realloc(log_buf, log_buf_size); | |
6789 | if (!tmp) { | |
6790 | ret = -ENOMEM; | |
6791 | goto out; | |
6792 | } | |
6793 | log_buf = tmp; | |
6794 | log_buf[0] = '\0'; | |
6795 | own_log_buf = true; | |
6796 | } | |
8395f320 | 6797 | } |
55cffde2 | 6798 | |
6aef10a4 | 6799 | load_attr.log_buf = log_buf; |
d10ef2b8 | 6800 | load_attr.log_size = log_buf_size; |
b3ce9079 | 6801 | load_attr.log_level = log_level; |
55cffde2 | 6802 | |
b3ce9079 | 6803 | ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr); |
55cffde2 | 6804 | if (ret >= 0) { |
b3ce9079 | 6805 | if (log_level && own_log_buf) { |
ad9a7f96 AN |
6806 | pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n", |
6807 | prog->name, log_buf); | |
6808 | } | |
5d23328d | 6809 | |
25bbbd7a AN |
6810 | if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) { |
6811 | struct bpf_map *map; | |
6812 | int i; | |
6813 | ||
6814 | for (i = 0; i < obj->nr_maps; i++) { | |
6815 | map = &prog->obj->maps[i]; | |
6816 | if (map->libbpf_type != LIBBPF_MAP_RODATA) | |
6817 | continue; | |
5d23328d | 6818 | |
25bbbd7a AN |
6819 | if (bpf_prog_bind_map(ret, bpf_map__fd(map), NULL)) { |
6820 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); | |
ad9a7f96 AN |
6821 | pr_warn("prog '%s': failed to bind map '%s': %s\n", |
6822 | prog->name, map->real_name, cp); | |
25bbbd7a AN |
6823 | /* Don't fail hard if can't bind rodata. */ |
6824 | } | |
5d23328d YZ |
6825 | } |
6826 | } | |
6827 | ||
be2f2d16 | 6828 | *prog_fd = ret; |
55cffde2 WN |
6829 | ret = 0; |
6830 | goto out; | |
6831 | } | |
6832 | ||
b3ce9079 AN |
6833 | if (log_level == 0) { |
6834 | log_level = 1; | |
da11b417 AS |
6835 | goto retry_load; |
6836 | } | |
b3ce9079 AN |
6837 | /* On ENOSPC, increase log buffer size and retry, unless custom |
6838 | * log_buf is specified. | |
6839 | * Be careful to not overflow u32, though. Kernel's log buf size limit | |
6840 | * isn't part of UAPI so it can always be bumped to full 4GB. So don't | |
6841 | * multiply by 2 unless we are sure we'll fit within 32 bits. | |
6842 | * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2). | |
6843 | */ | |
6844 | if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2) | |
6845 | goto retry_load; | |
2eda2145 AN |
6846 | |
6847 | ret = -errno; | |
24d6a808 | 6848 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
ad9a7f96 | 6849 | pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, cp); |
dc3a2d25 | 6850 | pr_perm_msg(ret); |
55cffde2 | 6851 | |
b3ce9079 | 6852 | if (own_log_buf && log_buf && log_buf[0] != '\0') { |
ad9a7f96 AN |
6853 | pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n", |
6854 | prog->name, log_buf); | |
6855 | } | |
55cffde2 WN |
6856 | |
6857 | out: | |
b3ce9079 AN |
6858 | if (own_log_buf) |
6859 | free(log_buf); | |
55cffde2 WN |
6860 | return ret; |
6861 | } | |
6862 | ||
d0e92887 | 6863 | static int bpf_program_record_relos(struct bpf_program *prog) |
67234743 AS |
6864 | { |
6865 | struct bpf_object *obj = prog->obj; | |
6866 | int i; | |
6867 | ||
6868 | for (i = 0; i < prog->nr_reloc; i++) { | |
6869 | struct reloc_desc *relo = &prog->reloc_desc[i]; | |
6870 | struct extern_desc *ext = &obj->externs[relo->sym_off]; | |
6871 | ||
6872 | switch (relo->type) { | |
6873 | case RELO_EXTERN_VAR: | |
6874 | if (ext->type != EXT_KSYM) | |
6875 | continue; | |
c24941cd KKD |
6876 | bpf_gen__record_extern(obj->gen_loader, ext->name, |
6877 | ext->is_weak, !ext->ksym.type_id, | |
18f4fccb | 6878 | BTF_KIND_VAR, relo->insn_idx); |
67234743 AS |
6879 | break; |
6880 | case RELO_EXTERN_FUNC: | |
c24941cd KKD |
6881 | bpf_gen__record_extern(obj->gen_loader, ext->name, |
6882 | ext->is_weak, false, BTF_KIND_FUNC, | |
67234743 AS |
6883 | relo->insn_idx); |
6884 | break; | |
d0e92887 AS |
6885 | case RELO_CORE: { |
6886 | struct bpf_core_relo cr = { | |
6887 | .insn_off = relo->insn_idx * 8, | |
6888 | .type_id = relo->core_relo->type_id, | |
6889 | .access_str_off = relo->core_relo->access_str_off, | |
6890 | .kind = relo->core_relo->kind, | |
6891 | }; | |
6892 | ||
6893 | bpf_gen__record_relo_core(obj->gen_loader, &cr); | |
6894 | break; | |
6895 | } | |
67234743 AS |
6896 | default: |
6897 | continue; | |
6898 | } | |
6899 | } | |
6900 | return 0; | |
6901 | } | |
6902 | ||
be2f2d16 AN |
6903 | static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog, |
6904 | const char *license, __u32 kern_ver) | |
55cffde2 | 6905 | { |
91abb4a6 | 6906 | int err = 0, fd, i; |
13acb508 | 6907 | |
be2f2d16 | 6908 | if (obj->loaded) { |
52109584 | 6909 | pr_warn("prog '%s': can't load after object was loaded\n", prog->name); |
e9fc3ce9 | 6910 | return libbpf_err(-EINVAL); |
d9297581 AN |
6911 | } |
6912 | ||
b580563e WN |
6913 | if (prog->instances.nr < 0 || !prog->instances.fds) { |
6914 | if (prog->preprocessor) { | |
be18010e | 6915 | pr_warn("Internal error: can't load program '%s'\n", |
52109584 | 6916 | prog->name); |
e9fc3ce9 | 6917 | return libbpf_err(-LIBBPF_ERRNO__INTERNAL); |
b580563e | 6918 | } |
55cffde2 | 6919 | |
b580563e WN |
6920 | prog->instances.fds = malloc(sizeof(int)); |
6921 | if (!prog->instances.fds) { | |
be18010e | 6922 | pr_warn("Not enough memory for BPF fds\n"); |
e9fc3ce9 | 6923 | return libbpf_err(-ENOMEM); |
b580563e WN |
6924 | } |
6925 | prog->instances.nr = 1; | |
6926 | prog->instances.fds[0] = -1; | |
6927 | } | |
6928 | ||
6929 | if (!prog->preprocessor) { | |
6930 | if (prog->instances.nr != 1) { | |
52109584 AN |
6931 | pr_warn("prog '%s': inconsistent nr(%d) != 1\n", |
6932 | prog->name, prog->instances.nr); | |
b580563e | 6933 | } |
be2f2d16 | 6934 | if (obj->gen_loader) |
d0e92887 | 6935 | bpf_program_record_relos(prog); |
be2f2d16 AN |
6936 | err = bpf_object_load_prog_instance(obj, prog, |
6937 | prog->insns, prog->insns_cnt, | |
6938 | license, kern_ver, &fd); | |
b580563e WN |
6939 | if (!err) |
6940 | prog->instances.fds[0] = fd; | |
6941 | goto out; | |
6942 | } | |
6943 | ||
6944 | for (i = 0; i < prog->instances.nr; i++) { | |
6945 | struct bpf_prog_prep_result result; | |
6946 | bpf_program_prep_t preprocessor = prog->preprocessor; | |
6947 | ||
1ad9cbb8 | 6948 | memset(&result, 0, sizeof(result)); |
b580563e WN |
6949 | err = preprocessor(prog, i, prog->insns, |
6950 | prog->insns_cnt, &result); | |
6951 | if (err) { | |
be18010e | 6952 | pr_warn("Preprocessing the %dth instance of program '%s' failed\n", |
52109584 | 6953 | i, prog->name); |
b580563e WN |
6954 | goto out; |
6955 | } | |
6956 | ||
6957 | if (!result.new_insn_ptr || !result.new_insn_cnt) { | |
6958 | pr_debug("Skip loading the %dth instance of program '%s'\n", | |
52109584 | 6959 | i, prog->name); |
b580563e WN |
6960 | prog->instances.fds[i] = -1; |
6961 | if (result.pfd) | |
6962 | *result.pfd = -1; | |
6963 | continue; | |
6964 | } | |
6965 | ||
be2f2d16 AN |
6966 | err = bpf_object_load_prog_instance(obj, prog, |
6967 | result.new_insn_ptr, result.new_insn_cnt, | |
6968 | license, kern_ver, &fd); | |
b580563e | 6969 | if (err) { |
be18010e | 6970 | pr_warn("Loading the %dth instance of program '%s' failed\n", |
52109584 | 6971 | i, prog->name); |
b580563e WN |
6972 | goto out; |
6973 | } | |
6974 | ||
6975 | if (result.pfd) | |
6976 | *result.pfd = fd; | |
6977 | prog->instances.fds[i] = fd; | |
6978 | } | |
6979 | out: | |
55cffde2 | 6980 | if (err) |
52109584 | 6981 | pr_warn("failed to load program '%s'\n", prog->name); |
e9fc3ce9 | 6982 | return libbpf_err(err); |
55cffde2 WN |
6983 | } |
6984 | ||
be2f2d16 AN |
6985 | int bpf_program__load(struct bpf_program *prog, const char *license, __u32 kern_ver) |
6986 | { | |
6987 | return bpf_object_load_prog(prog->obj, prog, license, kern_ver); | |
6988 | } | |
6989 | ||
55cffde2 | 6990 | static int |
60276f98 | 6991 | bpf_object__load_progs(struct bpf_object *obj, int log_level) |
55cffde2 | 6992 | { |
d9297581 | 6993 | struct bpf_program *prog; |
55cffde2 WN |
6994 | size_t i; |
6995 | int err; | |
6996 | ||
109cea5a AN |
6997 | for (i = 0; i < obj->nr_programs; i++) { |
6998 | prog = &obj->programs[i]; | |
6999 | err = bpf_object__sanitize_prog(obj, prog); | |
7000 | if (err) | |
7001 | return err; | |
7002 | } | |
7003 | ||
55cffde2 | 7004 | for (i = 0; i < obj->nr_programs; i++) { |
d9297581 | 7005 | prog = &obj->programs[i]; |
c3c55696 | 7006 | if (prog_is_subprog(obj, prog)) |
48cca7e4 | 7007 | continue; |
a3820c48 | 7008 | if (!prog->autoload) { |
9c0f8cbd | 7009 | pr_debug("prog '%s': skipped loading\n", prog->name); |
d9297581 AN |
7010 | continue; |
7011 | } | |
7012 | prog->log_level |= log_level; | |
be2f2d16 | 7013 | err = bpf_object_load_prog(obj, prog, obj->license, obj->kern_version); |
55cffde2 WN |
7014 | if (err) |
7015 | return err; | |
7016 | } | |
67234743 AS |
7017 | if (obj->gen_loader) |
7018 | bpf_object__free_relocs(obj); | |
55cffde2 WN |
7019 | return 0; |
7020 | } | |
7021 | ||
25498a19 AN |
7022 | static const struct bpf_sec_def *find_sec_def(const char *sec_name); |
7023 | ||
91b4d1d1 AN |
7024 | static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts) |
7025 | { | |
7026 | struct bpf_program *prog; | |
12d9466d | 7027 | int err; |
91b4d1d1 AN |
7028 | |
7029 | bpf_object__for_each_program(prog, obj) { | |
7030 | prog->sec_def = find_sec_def(prog->sec_name); | |
7031 | if (!prog->sec_def) { | |
7032 | /* couldn't guess, but user might manually specify */ | |
7033 | pr_debug("prog '%s': unrecognized ELF section name '%s'\n", | |
7034 | prog->name, prog->sec_name); | |
7035 | continue; | |
7036 | } | |
7037 | ||
df286716 GS |
7038 | prog->type = prog->sec_def->prog_type; |
7039 | prog->expected_attach_type = prog->sec_def->expected_attach_type; | |
91b4d1d1 | 7040 | |
91b555d7 AN |
7041 | #pragma GCC diagnostic push |
7042 | #pragma GCC diagnostic ignored "-Wdeprecated-declarations" | |
91b4d1d1 AN |
7043 | if (prog->sec_def->prog_type == BPF_PROG_TYPE_TRACING || |
7044 | prog->sec_def->prog_type == BPF_PROG_TYPE_EXT) | |
7045 | prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0); | |
91b555d7 | 7046 | #pragma GCC diagnostic pop |
12d9466d AN |
7047 | |
7048 | /* sec_def can have custom callback which should be called | |
7049 | * after bpf_program is initialized to adjust its properties | |
7050 | */ | |
4fa5bcfe AN |
7051 | if (prog->sec_def->prog_setup_fn) { |
7052 | err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie); | |
12d9466d AN |
7053 | if (err < 0) { |
7054 | pr_warn("prog '%s': failed to initialize: %d\n", | |
7055 | prog->name, err); | |
7056 | return err; | |
7057 | } | |
7058 | } | |
91b4d1d1 AN |
7059 | } |
7060 | ||
7061 | return 0; | |
7062 | } | |
7063 | ||
ad9a7f96 AN |
7064 | static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz, |
7065 | const struct bpf_object_open_opts *opts) | |
1a5e3fb1 | 7066 | { |
1373ff59 | 7067 | const char *obj_name, *kconfig, *btf_tmp_path; |
1a5e3fb1 | 7068 | struct bpf_object *obj; |
291ee02b | 7069 | char tmp_name[64]; |
6371ca3b | 7070 | int err; |
e0e3ea88 AN |
7071 | char *log_buf; |
7072 | size_t log_size; | |
7073 | __u32 log_level; | |
1a5e3fb1 WN |
7074 | |
7075 | if (elf_version(EV_CURRENT) == EV_NONE) { | |
be18010e KW |
7076 | pr_warn("failed to init libelf for %s\n", |
7077 | path ? : "(mem buf)"); | |
6371ca3b | 7078 | return ERR_PTR(-LIBBPF_ERRNO__LIBELF); |
1a5e3fb1 WN |
7079 | } |
7080 | ||
291ee02b AN |
7081 | if (!OPTS_VALID(opts, bpf_object_open_opts)) |
7082 | return ERR_PTR(-EINVAL); | |
7083 | ||
1aace10f | 7084 | obj_name = OPTS_GET(opts, object_name, NULL); |
291ee02b AN |
7085 | if (obj_buf) { |
7086 | if (!obj_name) { | |
7087 | snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx", | |
7088 | (unsigned long)obj_buf, | |
7089 | (unsigned long)obj_buf_sz); | |
7090 | obj_name = tmp_name; | |
7091 | } | |
7092 | path = obj_name; | |
7093 | pr_debug("loading object '%s' from buffer\n", obj_name); | |
7094 | } | |
7095 | ||
e0e3ea88 AN |
7096 | log_buf = OPTS_GET(opts, kernel_log_buf, NULL); |
7097 | log_size = OPTS_GET(opts, kernel_log_size, 0); | |
7098 | log_level = OPTS_GET(opts, kernel_log_level, 0); | |
7099 | if (log_size > UINT_MAX) | |
7100 | return ERR_PTR(-EINVAL); | |
7101 | if (log_size && !log_buf) | |
7102 | return ERR_PTR(-EINVAL); | |
7103 | ||
2ce8450e | 7104 | obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name); |
6371ca3b WN |
7105 | if (IS_ERR(obj)) |
7106 | return obj; | |
1a5e3fb1 | 7107 | |
e0e3ea88 AN |
7108 | obj->log_buf = log_buf; |
7109 | obj->log_size = log_size; | |
7110 | obj->log_level = log_level; | |
7111 | ||
1373ff59 SC |
7112 | btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL); |
7113 | if (btf_tmp_path) { | |
7114 | if (strlen(btf_tmp_path) >= PATH_MAX) { | |
7115 | err = -ENAMETOOLONG; | |
7116 | goto out; | |
7117 | } | |
7118 | obj->btf_custom_path = strdup(btf_tmp_path); | |
7119 | if (!obj->btf_custom_path) { | |
7120 | err = -ENOMEM; | |
7121 | goto out; | |
7122 | } | |
7123 | } | |
7124 | ||
8601fd42 AN |
7125 | kconfig = OPTS_GET(opts, kconfig, NULL); |
7126 | if (kconfig) { | |
7127 | obj->kconfig = strdup(kconfig); | |
18353c87 SC |
7128 | if (!obj->kconfig) { |
7129 | err = -ENOMEM; | |
7130 | goto out; | |
7131 | } | |
166750bc | 7132 | } |
291ee02b | 7133 | |
0d13bfce AN |
7134 | err = bpf_object__elf_init(obj); |
7135 | err = err ? : bpf_object__check_endianness(obj); | |
7136 | err = err ? : bpf_object__elf_collect(obj); | |
166750bc AN |
7137 | err = err ? : bpf_object__collect_externs(obj); |
7138 | err = err ? : bpf_object__finalize_btf(obj); | |
0d13bfce | 7139 | err = err ? : bpf_object__init_maps(obj, opts); |
91b4d1d1 | 7140 | err = err ? : bpf_object_init_progs(obj, opts); |
c3c55696 | 7141 | err = err ? : bpf_object__collect_relos(obj); |
0d13bfce AN |
7142 | if (err) |
7143 | goto out; | |
dd4436bb | 7144 | |
91b4d1d1 | 7145 | bpf_object__elf_finish(obj); |
dd4436bb | 7146 | |
1a5e3fb1 WN |
7147 | return obj; |
7148 | out: | |
7149 | bpf_object__close(obj); | |
6371ca3b | 7150 | return ERR_PTR(err); |
1a5e3fb1 WN |
7151 | } |
7152 | ||
5e61f270 AN |
7153 | static struct bpf_object * |
7154 | __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags) | |
1a5e3fb1 | 7155 | { |
e00aca65 | 7156 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts, |
291ee02b AN |
7157 | .relaxed_maps = flags & MAPS_RELAX_COMPAT, |
7158 | ); | |
7159 | ||
1a5e3fb1 | 7160 | /* param validation */ |
07f2d4ea | 7161 | if (!attr->file) |
1a5e3fb1 WN |
7162 | return NULL; |
7163 | ||
07f2d4ea | 7164 | pr_debug("loading %s\n", attr->file); |
ad9a7f96 | 7165 | return bpf_object_open(attr->file, NULL, 0, &opts); |
c034a177 JF |
7166 | } |
7167 | ||
7168 | struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr) | |
7169 | { | |
e9fc3ce9 | 7170 | return libbpf_ptr(__bpf_object__open_xattr(attr, 0)); |
07f2d4ea JK |
7171 | } |
7172 | ||
7173 | struct bpf_object *bpf_object__open(const char *path) | |
7174 | { | |
7175 | struct bpf_object_open_attr attr = { | |
7176 | .file = path, | |
7177 | .prog_type = BPF_PROG_TYPE_UNSPEC, | |
7178 | }; | |
1a5e3fb1 | 7179 | |
e9fc3ce9 | 7180 | return libbpf_ptr(__bpf_object__open_xattr(&attr, 0)); |
6c956392 WN |
7181 | } |
7182 | ||
2ce8450e | 7183 | struct bpf_object * |
01af3bf0 | 7184 | bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts) |
2ce8450e | 7185 | { |
2ce8450e | 7186 | if (!path) |
e9fc3ce9 | 7187 | return libbpf_err_ptr(-EINVAL); |
2ce8450e AN |
7188 | |
7189 | pr_debug("loading %s\n", path); | |
7190 | ||
ad9a7f96 | 7191 | return libbpf_ptr(bpf_object_open(path, NULL, 0, opts)); |
2ce8450e AN |
7192 | } |
7193 | ||
7194 | struct bpf_object * | |
7195 | bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz, | |
01af3bf0 | 7196 | const struct bpf_object_open_opts *opts) |
6c956392 | 7197 | { |
2ce8450e | 7198 | if (!obj_buf || obj_buf_sz == 0) |
e9fc3ce9 | 7199 | return libbpf_err_ptr(-EINVAL); |
6c956392 | 7200 | |
ad9a7f96 | 7201 | return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, opts)); |
2ce8450e AN |
7202 | } |
7203 | ||
7204 | struct bpf_object * | |
7205 | bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz, | |
7206 | const char *name) | |
7207 | { | |
e00aca65 | 7208 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts, |
2ce8450e AN |
7209 | .object_name = name, |
7210 | /* wrong default, but backwards-compatible */ | |
7211 | .relaxed_maps = true, | |
7212 | ); | |
7213 | ||
7214 | /* returning NULL is wrong, but backwards-compatible */ | |
7215 | if (!obj_buf || obj_buf_sz == 0) | |
e9fc3ce9 | 7216 | return errno = EINVAL, NULL; |
6c956392 | 7217 | |
ad9a7f96 | 7218 | return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, &opts)); |
1a5e3fb1 WN |
7219 | } |
7220 | ||
4a404a7e | 7221 | static int bpf_object_unload(struct bpf_object *obj) |
52d3352e WN |
7222 | { |
7223 | size_t i; | |
7224 | ||
7225 | if (!obj) | |
e9fc3ce9 | 7226 | return libbpf_err(-EINVAL); |
52d3352e | 7227 | |
590a0088 | 7228 | for (i = 0; i < obj->nr_maps; i++) { |
9d759a9b | 7229 | zclose(obj->maps[i].fd); |
590a0088 MKL |
7230 | if (obj->maps[i].st_ops) |
7231 | zfree(&obj->maps[i].st_ops->kern_vdata); | |
7232 | } | |
52d3352e | 7233 | |
55cffde2 WN |
7234 | for (i = 0; i < obj->nr_programs; i++) |
7235 | bpf_program__unload(&obj->programs[i]); | |
7236 | ||
52d3352e WN |
7237 | return 0; |
7238 | } | |
7239 | ||
4a404a7e HC |
7240 | int bpf_object__unload(struct bpf_object *obj) __attribute__((alias("bpf_object_unload"))); |
7241 | ||
0d13bfce AN |
7242 | static int bpf_object__sanitize_maps(struct bpf_object *obj) |
7243 | { | |
7244 | struct bpf_map *m; | |
7245 | ||
7246 | bpf_object__for_each_map(m, obj) { | |
7247 | if (!bpf_map__is_internal(m)) | |
7248 | continue; | |
9ca1f56a | 7249 | if (!kernel_supports(obj, FEAT_ARRAY_MMAP)) |
0d13bfce AN |
7250 | m->def.map_flags ^= BPF_F_MMAPABLE; |
7251 | } | |
7252 | ||
7253 | return 0; | |
7254 | } | |
7255 | ||
85153ac0 | 7256 | int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *ctx) |
1c0c7074 AN |
7257 | { |
7258 | char sym_type, sym_name[500]; | |
7259 | unsigned long long sym_addr; | |
1c0c7074 AN |
7260 | int ret, err = 0; |
7261 | FILE *f; | |
7262 | ||
7263 | f = fopen("/proc/kallsyms", "r"); | |
7264 | if (!f) { | |
7265 | err = -errno; | |
7266 | pr_warn("failed to open /proc/kallsyms: %d\n", err); | |
7267 | return err; | |
7268 | } | |
7269 | ||
7270 | while (true) { | |
7271 | ret = fscanf(f, "%llx %c %499s%*[^\n]\n", | |
7272 | &sym_addr, &sym_type, sym_name); | |
7273 | if (ret == EOF && feof(f)) | |
7274 | break; | |
7275 | if (ret != 3) { | |
135c783f | 7276 | pr_warn("failed to read kallsyms entry: %d\n", ret); |
1c0c7074 | 7277 | err = -EINVAL; |
85153ac0 | 7278 | break; |
1c0c7074 AN |
7279 | } |
7280 | ||
85153ac0 JO |
7281 | err = cb(sym_addr, sym_type, sym_name, ctx); |
7282 | if (err) | |
7283 | break; | |
1c0c7074 AN |
7284 | } |
7285 | ||
1c0c7074 AN |
7286 | fclose(f); |
7287 | return err; | |
7288 | } | |
7289 | ||
85153ac0 JO |
7290 | static int kallsyms_cb(unsigned long long sym_addr, char sym_type, |
7291 | const char *sym_name, void *ctx) | |
7292 | { | |
7293 | struct bpf_object *obj = ctx; | |
7294 | const struct btf_type *t; | |
7295 | struct extern_desc *ext; | |
7296 | ||
7297 | ext = find_extern_by_name(obj, sym_name); | |
7298 | if (!ext || ext->type != EXT_KSYM) | |
7299 | return 0; | |
7300 | ||
7301 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
7302 | if (!btf_is_var(t)) | |
7303 | return 0; | |
7304 | ||
7305 | if (ext->is_set && ext->ksym.addr != sym_addr) { | |
7306 | pr_warn("extern (ksym) '%s' resolution is ambiguous: 0x%llx or 0x%llx\n", | |
7307 | sym_name, ext->ksym.addr, sym_addr); | |
7308 | return -EINVAL; | |
7309 | } | |
7310 | if (!ext->is_set) { | |
7311 | ext->is_set = true; | |
7312 | ext->ksym.addr = sym_addr; | |
7313 | pr_debug("extern (ksym) %s=0x%llx\n", sym_name, sym_addr); | |
7314 | } | |
7315 | return 0; | |
7316 | } | |
7317 | ||
7318 | static int bpf_object__read_kallsyms_file(struct bpf_object *obj) | |
7319 | { | |
7320 | return libbpf_kallsyms_parse(kallsyms_cb, obj); | |
7321 | } | |
7322 | ||
774e132e MKL |
7323 | static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name, |
7324 | __u16 kind, struct btf **res_btf, | |
9dbe6015 | 7325 | struct module_btf **res_mod_btf) |
d370bbe1 | 7326 | { |
9dbe6015 | 7327 | struct module_btf *mod_btf; |
284d2587 | 7328 | struct btf *btf; |
9dbe6015 | 7329 | int i, id, err; |
d370bbe1 | 7330 | |
933d1aa3 | 7331 | btf = obj->btf_vmlinux; |
9dbe6015 | 7332 | mod_btf = NULL; |
774e132e MKL |
7333 | id = btf__find_by_name_kind(btf, ksym_name, kind); |
7334 | ||
933d1aa3 MKL |
7335 | if (id == -ENOENT) { |
7336 | err = load_module_btfs(obj); | |
7337 | if (err) | |
7338 | return err; | |
d370bbe1 | 7339 | |
933d1aa3 | 7340 | for (i = 0; i < obj->btf_module_cnt; i++) { |
9dbe6015 KKD |
7341 | /* we assume module_btf's BTF FD is always >0 */ |
7342 | mod_btf = &obj->btf_modules[i]; | |
7343 | btf = mod_btf->btf; | |
7344 | id = btf__find_by_name_kind_own(btf, ksym_name, kind); | |
933d1aa3 MKL |
7345 | if (id != -ENOENT) |
7346 | break; | |
7347 | } | |
7348 | } | |
2211c825 | 7349 | if (id <= 0) |
933d1aa3 | 7350 | return -ESRCH; |
d370bbe1 | 7351 | |
774e132e | 7352 | *res_btf = btf; |
9dbe6015 | 7353 | *res_mod_btf = mod_btf; |
774e132e MKL |
7354 | return id; |
7355 | } | |
7356 | ||
7357 | static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj, | |
7358 | struct extern_desc *ext) | |
7359 | { | |
7360 | const struct btf_type *targ_var, *targ_type; | |
7361 | __u32 targ_type_id, local_type_id; | |
9dbe6015 | 7362 | struct module_btf *mod_btf = NULL; |
774e132e | 7363 | const char *targ_var_name; |
774e132e | 7364 | struct btf *btf = NULL; |
9dbe6015 | 7365 | int id, err; |
774e132e | 7366 | |
9dbe6015 | 7367 | id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf); |
466b2e13 KKD |
7368 | if (id < 0) { |
7369 | if (id == -ESRCH && ext->is_weak) | |
7370 | return 0; | |
2211c825 HL |
7371 | pr_warn("extern (var ksym) '%s': not found in kernel BTF\n", |
7372 | ext->name); | |
774e132e | 7373 | return id; |
2211c825 | 7374 | } |
774e132e | 7375 | |
933d1aa3 MKL |
7376 | /* find local type_id */ |
7377 | local_type_id = ext->ksym.type_id; | |
284d2587 | 7378 | |
933d1aa3 MKL |
7379 | /* find target type_id */ |
7380 | targ_var = btf__type_by_id(btf, id); | |
7381 | targ_var_name = btf__name_by_offset(btf, targ_var->name_off); | |
7382 | targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id); | |
d370bbe1 | 7383 | |
933d1aa3 MKL |
7384 | err = bpf_core_types_are_compat(obj->btf, local_type_id, |
7385 | btf, targ_type_id); | |
7386 | if (err <= 0) { | |
7387 | const struct btf_type *local_type; | |
7388 | const char *targ_name, *local_name; | |
d370bbe1 | 7389 | |
933d1aa3 MKL |
7390 | local_type = btf__type_by_id(obj->btf, local_type_id); |
7391 | local_name = btf__name_by_offset(obj->btf, local_type->name_off); | |
7392 | targ_name = btf__name_by_offset(btf, targ_type->name_off); | |
d370bbe1 | 7393 | |
933d1aa3 MKL |
7394 | pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n", |
7395 | ext->name, local_type_id, | |
7396 | btf_kind_str(local_type), local_name, targ_type_id, | |
7397 | btf_kind_str(targ_type), targ_name); | |
7398 | return -EINVAL; | |
7399 | } | |
d370bbe1 | 7400 | |
933d1aa3 | 7401 | ext->is_set = true; |
9dbe6015 | 7402 | ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0; |
933d1aa3 MKL |
7403 | ext->ksym.kernel_btf_id = id; |
7404 | pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n", | |
7405 | ext->name, id, btf_kind_str(targ_var), targ_var_name); | |
d370bbe1 | 7406 | |
933d1aa3 MKL |
7407 | return 0; |
7408 | } | |
d370bbe1 | 7409 | |
5bd022ec MKL |
7410 | static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj, |
7411 | struct extern_desc *ext) | |
7412 | { | |
7413 | int local_func_proto_id, kfunc_proto_id, kfunc_id; | |
9dbe6015 | 7414 | struct module_btf *mod_btf = NULL; |
5bd022ec MKL |
7415 | const struct btf_type *kern_func; |
7416 | struct btf *kern_btf = NULL; | |
9dbe6015 | 7417 | int ret; |
5bd022ec MKL |
7418 | |
7419 | local_func_proto_id = ext->ksym.type_id; | |
7420 | ||
9dbe6015 | 7421 | kfunc_id = find_ksym_btf_id(obj, ext->name, BTF_KIND_FUNC, &kern_btf, &mod_btf); |
5bd022ec | 7422 | if (kfunc_id < 0) { |
466b2e13 KKD |
7423 | if (kfunc_id == -ESRCH && ext->is_weak) |
7424 | return 0; | |
7425 | pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n", | |
5bd022ec MKL |
7426 | ext->name); |
7427 | return kfunc_id; | |
7428 | } | |
7429 | ||
5bd022ec MKL |
7430 | kern_func = btf__type_by_id(kern_btf, kfunc_id); |
7431 | kfunc_proto_id = kern_func->type; | |
7432 | ||
7433 | ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id, | |
7434 | kern_btf, kfunc_proto_id); | |
7435 | if (ret <= 0) { | |
7436 | pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with kernel [%d]\n", | |
7437 | ext->name, local_func_proto_id, kfunc_proto_id); | |
7438 | return -EINVAL; | |
7439 | } | |
7440 | ||
9dbe6015 KKD |
7441 | /* set index for module BTF fd in fd_array, if unset */ |
7442 | if (mod_btf && !mod_btf->fd_array_idx) { | |
7443 | /* insn->off is s16 */ | |
7444 | if (obj->fd_array_cnt == INT16_MAX) { | |
7445 | pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n", | |
7446 | ext->name, mod_btf->fd_array_idx); | |
7447 | return -E2BIG; | |
7448 | } | |
7449 | /* Cannot use index 0 for module BTF fd */ | |
7450 | if (!obj->fd_array_cnt) | |
7451 | obj->fd_array_cnt = 1; | |
7452 | ||
7453 | ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int), | |
7454 | obj->fd_array_cnt + 1); | |
7455 | if (ret) | |
7456 | return ret; | |
7457 | mod_btf->fd_array_idx = obj->fd_array_cnt; | |
7458 | /* we assume module BTF FD is always >0 */ | |
7459 | obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd; | |
7460 | } | |
7461 | ||
5bd022ec | 7462 | ext->is_set = true; |
5bd022ec | 7463 | ext->ksym.kernel_btf_id = kfunc_id; |
9dbe6015 | 7464 | ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0; |
5bd022ec MKL |
7465 | pr_debug("extern (func ksym) '%s': resolved to kernel [%d]\n", |
7466 | ext->name, kfunc_id); | |
7467 | ||
7468 | return 0; | |
7469 | } | |
7470 | ||
933d1aa3 MKL |
7471 | static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj) |
7472 | { | |
5bd022ec | 7473 | const struct btf_type *t; |
933d1aa3 MKL |
7474 | struct extern_desc *ext; |
7475 | int i, err; | |
7476 | ||
7477 | for (i = 0; i < obj->nr_extern; i++) { | |
7478 | ext = &obj->externs[i]; | |
7479 | if (ext->type != EXT_KSYM || !ext->ksym.type_id) | |
7480 | continue; | |
7481 | ||
67234743 AS |
7482 | if (obj->gen_loader) { |
7483 | ext->is_set = true; | |
7484 | ext->ksym.kernel_btf_obj_fd = 0; | |
7485 | ext->ksym.kernel_btf_id = 0; | |
7486 | continue; | |
7487 | } | |
5bd022ec MKL |
7488 | t = btf__type_by_id(obj->btf, ext->btf_id); |
7489 | if (btf_is_var(t)) | |
7490 | err = bpf_object__resolve_ksym_var_btf_id(obj, ext); | |
7491 | else | |
7492 | err = bpf_object__resolve_ksym_func_btf_id(obj, ext); | |
933d1aa3 MKL |
7493 | if (err) |
7494 | return err; | |
d370bbe1 HL |
7495 | } |
7496 | return 0; | |
7497 | } | |
7498 | ||
166750bc | 7499 | static int bpf_object__resolve_externs(struct bpf_object *obj, |
8601fd42 | 7500 | const char *extra_kconfig) |
166750bc | 7501 | { |
1c0c7074 | 7502 | bool need_config = false, need_kallsyms = false; |
d370bbe1 | 7503 | bool need_vmlinux_btf = false; |
166750bc | 7504 | struct extern_desc *ext; |
2e33efe3 | 7505 | void *kcfg_data = NULL; |
166750bc | 7506 | int err, i; |
166750bc AN |
7507 | |
7508 | if (obj->nr_extern == 0) | |
7509 | return 0; | |
7510 | ||
2e33efe3 AN |
7511 | if (obj->kconfig_map_idx >= 0) |
7512 | kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped; | |
166750bc AN |
7513 | |
7514 | for (i = 0; i < obj->nr_extern; i++) { | |
7515 | ext = &obj->externs[i]; | |
7516 | ||
2e33efe3 AN |
7517 | if (ext->type == EXT_KCFG && |
7518 | strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) { | |
7519 | void *ext_val = kcfg_data + ext->kcfg.data_off; | |
166750bc AN |
7520 | __u32 kver = get_kernel_version(); |
7521 | ||
7522 | if (!kver) { | |
7523 | pr_warn("failed to get kernel version\n"); | |
7524 | return -EINVAL; | |
7525 | } | |
2e33efe3 | 7526 | err = set_kcfg_value_num(ext, ext_val, kver); |
166750bc AN |
7527 | if (err) |
7528 | return err; | |
2e33efe3 | 7529 | pr_debug("extern (kcfg) %s=0x%x\n", ext->name, kver); |
13d35a0c | 7530 | } else if (ext->type == EXT_KCFG && str_has_pfx(ext->name, "CONFIG_")) { |
166750bc | 7531 | need_config = true; |
1c0c7074 | 7532 | } else if (ext->type == EXT_KSYM) { |
d370bbe1 HL |
7533 | if (ext->ksym.type_id) |
7534 | need_vmlinux_btf = true; | |
7535 | else | |
7536 | need_kallsyms = true; | |
166750bc AN |
7537 | } else { |
7538 | pr_warn("unrecognized extern '%s'\n", ext->name); | |
7539 | return -EINVAL; | |
7540 | } | |
7541 | } | |
8601fd42 | 7542 | if (need_config && extra_kconfig) { |
2e33efe3 | 7543 | err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data); |
8601fd42 AN |
7544 | if (err) |
7545 | return -EINVAL; | |
7546 | need_config = false; | |
7547 | for (i = 0; i < obj->nr_extern; i++) { | |
7548 | ext = &obj->externs[i]; | |
2e33efe3 | 7549 | if (ext->type == EXT_KCFG && !ext->is_set) { |
8601fd42 AN |
7550 | need_config = true; |
7551 | break; | |
7552 | } | |
7553 | } | |
7554 | } | |
166750bc | 7555 | if (need_config) { |
2e33efe3 | 7556 | err = bpf_object__read_kconfig_file(obj, kcfg_data); |
166750bc AN |
7557 | if (err) |
7558 | return -EINVAL; | |
7559 | } | |
1c0c7074 AN |
7560 | if (need_kallsyms) { |
7561 | err = bpf_object__read_kallsyms_file(obj); | |
7562 | if (err) | |
7563 | return -EINVAL; | |
7564 | } | |
d370bbe1 HL |
7565 | if (need_vmlinux_btf) { |
7566 | err = bpf_object__resolve_ksyms_btf_id(obj); | |
7567 | if (err) | |
7568 | return -EINVAL; | |
7569 | } | |
166750bc AN |
7570 | for (i = 0; i < obj->nr_extern; i++) { |
7571 | ext = &obj->externs[i]; | |
7572 | ||
7573 | if (!ext->is_set && !ext->is_weak) { | |
7574 | pr_warn("extern %s (strong) not resolved\n", ext->name); | |
7575 | return -ESRCH; | |
7576 | } else if (!ext->is_set) { | |
7577 | pr_debug("extern %s (weak) not resolved, defaulting to zero\n", | |
7578 | ext->name); | |
7579 | } | |
7580 | } | |
7581 | ||
7582 | return 0; | |
7583 | } | |
7584 | ||
e7b924ca | 7585 | static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path) |
52d3352e | 7586 | { |
ec6d5f47 | 7587 | int err, i; |
6371ca3b | 7588 | |
52d3352e | 7589 | if (!obj) |
e9fc3ce9 | 7590 | return libbpf_err(-EINVAL); |
52d3352e WN |
7591 | |
7592 | if (obj->loaded) { | |
d9297581 | 7593 | pr_warn("object '%s': load can't be attempted twice\n", obj->name); |
e9fc3ce9 | 7594 | return libbpf_err(-EINVAL); |
52d3352e WN |
7595 | } |
7596 | ||
67234743 | 7597 | if (obj->gen_loader) |
be315829 | 7598 | bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps); |
67234743 | 7599 | |
fd9eef1a | 7600 | err = bpf_object__probe_loading(obj); |
fe62de31 | 7601 | err = err ? : bpf_object__load_vmlinux_btf(obj, false); |
8601fd42 | 7602 | err = err ? : bpf_object__resolve_externs(obj, obj->kconfig); |
0d13bfce AN |
7603 | err = err ? : bpf_object__sanitize_and_load_btf(obj); |
7604 | err = err ? : bpf_object__sanitize_maps(obj); | |
590a0088 | 7605 | err = err ? : bpf_object__init_kern_struct_ops_maps(obj); |
0d13bfce | 7606 | err = err ? : bpf_object__create_maps(obj); |
e7b924ca AN |
7607 | err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path); |
7608 | err = err ? : bpf_object__load_progs(obj, extra_log_level); | |
341ac5ff | 7609 | err = err ? : bpf_object_init_prog_arrays(obj); |
a6ed02ca | 7610 | |
67234743 AS |
7611 | if (obj->gen_loader) { |
7612 | /* reset FDs */ | |
4729445b KKD |
7613 | if (obj->btf) |
7614 | btf__set_fd(obj->btf, -1); | |
67234743 AS |
7615 | for (i = 0; i < obj->nr_maps; i++) |
7616 | obj->maps[i].fd = -1; | |
7617 | if (!err) | |
ba05fd36 | 7618 | err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps); |
67234743 AS |
7619 | } |
7620 | ||
9dbe6015 KKD |
7621 | /* clean up fd_array */ |
7622 | zfree(&obj->fd_array); | |
7623 | ||
4f33a53d AN |
7624 | /* clean up module BTFs */ |
7625 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
91abb4a6 | 7626 | close(obj->btf_modules[i].fd); |
4f33a53d AN |
7627 | btf__free(obj->btf_modules[i].btf); |
7628 | free(obj->btf_modules[i].name); | |
7629 | } | |
7630 | free(obj->btf_modules); | |
7631 | ||
7632 | /* clean up vmlinux BTF */ | |
a6ed02ca KS |
7633 | btf__free(obj->btf_vmlinux); |
7634 | obj->btf_vmlinux = NULL; | |
7635 | ||
d9297581 AN |
7636 | obj->loaded = true; /* doesn't matter if successfully or not */ |
7637 | ||
0d13bfce AN |
7638 | if (err) |
7639 | goto out; | |
52d3352e WN |
7640 | |
7641 | return 0; | |
7642 | out: | |
ec6d5f47 THJ |
7643 | /* unpin any maps that were auto-pinned during load */ |
7644 | for (i = 0; i < obj->nr_maps; i++) | |
7645 | if (obj->maps[i].pinned && !obj->maps[i].reused) | |
7646 | bpf_map__unpin(&obj->maps[i], NULL); | |
7647 | ||
4a404a7e | 7648 | bpf_object_unload(obj); |
be18010e | 7649 | pr_warn("failed to load object '%s'\n", obj->path); |
e9fc3ce9 | 7650 | return libbpf_err(err); |
52d3352e WN |
7651 | } |
7652 | ||
e7b924ca | 7653 | int bpf_object__load_xattr(struct bpf_object_load_attr *attr) |
60276f98 | 7654 | { |
e7b924ca AN |
7655 | return bpf_object_load(attr->obj, attr->log_level, attr->target_btf_path); |
7656 | } | |
60276f98 | 7657 | |
e7b924ca AN |
7658 | int bpf_object__load(struct bpf_object *obj) |
7659 | { | |
7660 | return bpf_object_load(obj, 0, NULL); | |
60276f98 QM |
7661 | } |
7662 | ||
196f8487 THJ |
7663 | static int make_parent_dir(const char *path) |
7664 | { | |
7665 | char *cp, errmsg[STRERR_BUFSIZE]; | |
7666 | char *dname, *dir; | |
7667 | int err = 0; | |
7668 | ||
7669 | dname = strdup(path); | |
7670 | if (dname == NULL) | |
7671 | return -ENOMEM; | |
7672 | ||
7673 | dir = dirname(dname); | |
7674 | if (mkdir(dir, 0700) && errno != EEXIST) | |
7675 | err = -errno; | |
7676 | ||
7677 | free(dname); | |
7678 | if (err) { | |
7679 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
7680 | pr_warn("failed to mkdir %s: %s\n", path, cp); | |
7681 | } | |
7682 | return err; | |
7683 | } | |
7684 | ||
f367540c JS |
7685 | static int check_path(const char *path) |
7686 | { | |
1ce6a9fc | 7687 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
7688 | struct statfs st_fs; |
7689 | char *dname, *dir; | |
7690 | int err = 0; | |
7691 | ||
7692 | if (path == NULL) | |
7693 | return -EINVAL; | |
7694 | ||
7695 | dname = strdup(path); | |
7696 | if (dname == NULL) | |
7697 | return -ENOMEM; | |
7698 | ||
7699 | dir = dirname(dname); | |
7700 | if (statfs(dir, &st_fs)) { | |
24d6a808 | 7701 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
be18010e | 7702 | pr_warn("failed to statfs %s: %s\n", dir, cp); |
f367540c JS |
7703 | err = -errno; |
7704 | } | |
7705 | free(dname); | |
7706 | ||
7707 | if (!err && st_fs.f_type != BPF_FS_MAGIC) { | |
be18010e | 7708 | pr_warn("specified path %s is not on BPF FS\n", path); |
f367540c JS |
7709 | err = -EINVAL; |
7710 | } | |
7711 | ||
7712 | return err; | |
7713 | } | |
7714 | ||
e21d585c | 7715 | static int bpf_program_pin_instance(struct bpf_program *prog, const char *path, int instance) |
f367540c | 7716 | { |
1ce6a9fc | 7717 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
7718 | int err; |
7719 | ||
196f8487 THJ |
7720 | err = make_parent_dir(path); |
7721 | if (err) | |
e9fc3ce9 | 7722 | return libbpf_err(err); |
196f8487 | 7723 | |
f367540c JS |
7724 | err = check_path(path); |
7725 | if (err) | |
e9fc3ce9 | 7726 | return libbpf_err(err); |
f367540c JS |
7727 | |
7728 | if (prog == NULL) { | |
be18010e | 7729 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7730 | return libbpf_err(-EINVAL); |
f367540c JS |
7731 | } |
7732 | ||
7733 | if (instance < 0 || instance >= prog->instances.nr) { | |
be18010e | 7734 | pr_warn("invalid prog instance %d of prog %s (max %d)\n", |
52109584 | 7735 | instance, prog->name, prog->instances.nr); |
e9fc3ce9 | 7736 | return libbpf_err(-EINVAL); |
f367540c JS |
7737 | } |
7738 | ||
7739 | if (bpf_obj_pin(prog->instances.fds[instance], path)) { | |
23ab656b THJ |
7740 | err = -errno; |
7741 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e | 7742 | pr_warn("failed to pin program: %s\n", cp); |
e9fc3ce9 | 7743 | return libbpf_err(err); |
f367540c JS |
7744 | } |
7745 | pr_debug("pinned program '%s'\n", path); | |
7746 | ||
7747 | return 0; | |
7748 | } | |
7749 | ||
e21d585c | 7750 | static int bpf_program_unpin_instance(struct bpf_program *prog, const char *path, int instance) |
0c19a9fb SF |
7751 | { |
7752 | int err; | |
7753 | ||
7754 | err = check_path(path); | |
7755 | if (err) | |
e9fc3ce9 | 7756 | return libbpf_err(err); |
0c19a9fb SF |
7757 | |
7758 | if (prog == NULL) { | |
be18010e | 7759 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7760 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7761 | } |
7762 | ||
7763 | if (instance < 0 || instance >= prog->instances.nr) { | |
be18010e | 7764 | pr_warn("invalid prog instance %d of prog %s (max %d)\n", |
52109584 | 7765 | instance, prog->name, prog->instances.nr); |
e9fc3ce9 | 7766 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7767 | } |
7768 | ||
7769 | err = unlink(path); | |
7770 | if (err != 0) | |
e9fc3ce9 AN |
7771 | return libbpf_err(-errno); |
7772 | ||
0c19a9fb SF |
7773 | pr_debug("unpinned program '%s'\n", path); |
7774 | ||
7775 | return 0; | |
7776 | } | |
7777 | ||
e21d585c AN |
7778 | __attribute__((alias("bpf_program_pin_instance"))) |
7779 | int bpf_object__pin_instance(struct bpf_program *prog, const char *path, int instance); | |
7780 | ||
7781 | __attribute__((alias("bpf_program_unpin_instance"))) | |
7782 | int bpf_program__unpin_instance(struct bpf_program *prog, const char *path, int instance); | |
7783 | ||
f367540c JS |
7784 | int bpf_program__pin(struct bpf_program *prog, const char *path) |
7785 | { | |
7786 | int i, err; | |
7787 | ||
196f8487 THJ |
7788 | err = make_parent_dir(path); |
7789 | if (err) | |
e9fc3ce9 | 7790 | return libbpf_err(err); |
196f8487 | 7791 | |
f367540c JS |
7792 | err = check_path(path); |
7793 | if (err) | |
e9fc3ce9 | 7794 | return libbpf_err(err); |
f367540c JS |
7795 | |
7796 | if (prog == NULL) { | |
be18010e | 7797 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7798 | return libbpf_err(-EINVAL); |
f367540c JS |
7799 | } |
7800 | ||
7801 | if (prog->instances.nr <= 0) { | |
52109584 | 7802 | pr_warn("no instances of prog %s to pin\n", prog->name); |
e9fc3ce9 | 7803 | return libbpf_err(-EINVAL); |
f367540c JS |
7804 | } |
7805 | ||
fd734c5c SF |
7806 | if (prog->instances.nr == 1) { |
7807 | /* don't create subdirs when pinning single instance */ | |
e21d585c | 7808 | return bpf_program_pin_instance(prog, path, 0); |
fd734c5c SF |
7809 | } |
7810 | ||
0c19a9fb SF |
7811 | for (i = 0; i < prog->instances.nr; i++) { |
7812 | char buf[PATH_MAX]; | |
7813 | int len; | |
7814 | ||
7815 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7816 | if (len < 0) { | |
7817 | err = -EINVAL; | |
7818 | goto err_unpin; | |
7819 | } else if (len >= PATH_MAX) { | |
7820 | err = -ENAMETOOLONG; | |
7821 | goto err_unpin; | |
7822 | } | |
7823 | ||
e21d585c | 7824 | err = bpf_program_pin_instance(prog, buf, i); |
0c19a9fb SF |
7825 | if (err) |
7826 | goto err_unpin; | |
7827 | } | |
7828 | ||
7829 | return 0; | |
7830 | ||
7831 | err_unpin: | |
7832 | for (i = i - 1; i >= 0; i--) { | |
7833 | char buf[PATH_MAX]; | |
7834 | int len; | |
7835 | ||
7836 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7837 | if (len < 0) | |
7838 | continue; | |
7839 | else if (len >= PATH_MAX) | |
7840 | continue; | |
7841 | ||
e21d585c | 7842 | bpf_program_unpin_instance(prog, buf, i); |
0c19a9fb SF |
7843 | } |
7844 | ||
7845 | rmdir(path); | |
7846 | ||
e9fc3ce9 | 7847 | return libbpf_err(err); |
0c19a9fb SF |
7848 | } |
7849 | ||
7850 | int bpf_program__unpin(struct bpf_program *prog, const char *path) | |
7851 | { | |
7852 | int i, err; | |
7853 | ||
7854 | err = check_path(path); | |
7855 | if (err) | |
e9fc3ce9 | 7856 | return libbpf_err(err); |
0c19a9fb SF |
7857 | |
7858 | if (prog == NULL) { | |
be18010e | 7859 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7860 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7861 | } |
7862 | ||
7863 | if (prog->instances.nr <= 0) { | |
52109584 | 7864 | pr_warn("no instances of prog %s to pin\n", prog->name); |
e9fc3ce9 | 7865 | return libbpf_err(-EINVAL); |
fd734c5c SF |
7866 | } |
7867 | ||
7868 | if (prog->instances.nr == 1) { | |
7869 | /* don't create subdirs when pinning single instance */ | |
e21d585c | 7870 | return bpf_program_unpin_instance(prog, path, 0); |
0c19a9fb SF |
7871 | } |
7872 | ||
f367540c JS |
7873 | for (i = 0; i < prog->instances.nr; i++) { |
7874 | char buf[PATH_MAX]; | |
7875 | int len; | |
7876 | ||
7877 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7878 | if (len < 0) | |
e9fc3ce9 | 7879 | return libbpf_err(-EINVAL); |
f367540c | 7880 | else if (len >= PATH_MAX) |
e9fc3ce9 | 7881 | return libbpf_err(-ENAMETOOLONG); |
f367540c | 7882 | |
e21d585c | 7883 | err = bpf_program_unpin_instance(prog, buf, i); |
f367540c JS |
7884 | if (err) |
7885 | return err; | |
7886 | } | |
7887 | ||
0c19a9fb SF |
7888 | err = rmdir(path); |
7889 | if (err) | |
e9fc3ce9 | 7890 | return libbpf_err(-errno); |
0c19a9fb | 7891 | |
f367540c JS |
7892 | return 0; |
7893 | } | |
7894 | ||
b6989f35 JS |
7895 | int bpf_map__pin(struct bpf_map *map, const char *path) |
7896 | { | |
1ce6a9fc | 7897 | char *cp, errmsg[STRERR_BUFSIZE]; |
b6989f35 JS |
7898 | int err; |
7899 | ||
b6989f35 | 7900 | if (map == NULL) { |
be18010e | 7901 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 7902 | return libbpf_err(-EINVAL); |
b6989f35 JS |
7903 | } |
7904 | ||
4580b25f THJ |
7905 | if (map->pin_path) { |
7906 | if (path && strcmp(path, map->pin_path)) { | |
7907 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
7908 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 7909 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7910 | } else if (map->pinned) { |
7911 | pr_debug("map '%s' already pinned at '%s'; not re-pinning\n", | |
7912 | bpf_map__name(map), map->pin_path); | |
7913 | return 0; | |
7914 | } | |
7915 | } else { | |
7916 | if (!path) { | |
7917 | pr_warn("missing a path to pin map '%s' at\n", | |
7918 | bpf_map__name(map)); | |
e9fc3ce9 | 7919 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7920 | } else if (map->pinned) { |
7921 | pr_warn("map '%s' already pinned\n", bpf_map__name(map)); | |
e9fc3ce9 | 7922 | return libbpf_err(-EEXIST); |
4580b25f THJ |
7923 | } |
7924 | ||
7925 | map->pin_path = strdup(path); | |
7926 | if (!map->pin_path) { | |
7927 | err = -errno; | |
7928 | goto out_err; | |
7929 | } | |
b6989f35 JS |
7930 | } |
7931 | ||
196f8487 THJ |
7932 | err = make_parent_dir(map->pin_path); |
7933 | if (err) | |
e9fc3ce9 | 7934 | return libbpf_err(err); |
196f8487 | 7935 | |
4580b25f THJ |
7936 | err = check_path(map->pin_path); |
7937 | if (err) | |
e9fc3ce9 | 7938 | return libbpf_err(err); |
4580b25f THJ |
7939 | |
7940 | if (bpf_obj_pin(map->fd, map->pin_path)) { | |
7941 | err = -errno; | |
7942 | goto out_err; | |
7943 | } | |
7944 | ||
7945 | map->pinned = true; | |
7946 | pr_debug("pinned map '%s'\n", map->pin_path); | |
0c19a9fb | 7947 | |
b6989f35 | 7948 | return 0; |
4580b25f THJ |
7949 | |
7950 | out_err: | |
7951 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
7952 | pr_warn("failed to pin map: %s\n", cp); | |
e9fc3ce9 | 7953 | return libbpf_err(err); |
b6989f35 JS |
7954 | } |
7955 | ||
0c19a9fb SF |
7956 | int bpf_map__unpin(struct bpf_map *map, const char *path) |
7957 | { | |
7958 | int err; | |
7959 | ||
0c19a9fb | 7960 | if (map == NULL) { |
be18010e | 7961 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 7962 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7963 | } |
7964 | ||
4580b25f THJ |
7965 | if (map->pin_path) { |
7966 | if (path && strcmp(path, map->pin_path)) { | |
7967 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
7968 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 7969 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7970 | } |
7971 | path = map->pin_path; | |
7972 | } else if (!path) { | |
7973 | pr_warn("no path to unpin map '%s' from\n", | |
7974 | bpf_map__name(map)); | |
e9fc3ce9 | 7975 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7976 | } |
7977 | ||
7978 | err = check_path(path); | |
7979 | if (err) | |
e9fc3ce9 | 7980 | return libbpf_err(err); |
4580b25f | 7981 | |
0c19a9fb SF |
7982 | err = unlink(path); |
7983 | if (err != 0) | |
e9fc3ce9 | 7984 | return libbpf_err(-errno); |
4580b25f THJ |
7985 | |
7986 | map->pinned = false; | |
7987 | pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path); | |
0c19a9fb SF |
7988 | |
7989 | return 0; | |
7990 | } | |
7991 | ||
4580b25f THJ |
7992 | int bpf_map__set_pin_path(struct bpf_map *map, const char *path) |
7993 | { | |
7994 | char *new = NULL; | |
7995 | ||
7996 | if (path) { | |
7997 | new = strdup(path); | |
7998 | if (!new) | |
e9fc3ce9 | 7999 | return libbpf_err(-errno); |
4580b25f THJ |
8000 | } |
8001 | ||
8002 | free(map->pin_path); | |
8003 | map->pin_path = new; | |
8004 | return 0; | |
8005 | } | |
8006 | ||
20eccf29 AN |
8007 | __alias(bpf_map__pin_path) |
8008 | const char *bpf_map__get_pin_path(const struct bpf_map *map); | |
4580b25f | 8009 | |
e244d34d EL |
8010 | const char *bpf_map__pin_path(const struct bpf_map *map) |
8011 | { | |
8012 | return map->pin_path; | |
8013 | } | |
8014 | ||
4580b25f THJ |
8015 | bool bpf_map__is_pinned(const struct bpf_map *map) |
8016 | { | |
8017 | return map->pinned; | |
8018 | } | |
8019 | ||
9cf309c5 THJ |
8020 | static void sanitize_pin_path(char *s) |
8021 | { | |
8022 | /* bpffs disallows periods in path names */ | |
8023 | while (*s) { | |
8024 | if (*s == '.') | |
8025 | *s = '_'; | |
8026 | s++; | |
8027 | } | |
8028 | } | |
8029 | ||
0c19a9fb | 8030 | int bpf_object__pin_maps(struct bpf_object *obj, const char *path) |
d5148d85 | 8031 | { |
d5148d85 JS |
8032 | struct bpf_map *map; |
8033 | int err; | |
8034 | ||
8035 | if (!obj) | |
e9fc3ce9 | 8036 | return libbpf_err(-ENOENT); |
d5148d85 JS |
8037 | |
8038 | if (!obj->loaded) { | |
be18010e | 8039 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 8040 | return libbpf_err(-ENOENT); |
d5148d85 JS |
8041 | } |
8042 | ||
f74a53d9 | 8043 | bpf_object__for_each_map(map, obj) { |
4580b25f | 8044 | char *pin_path = NULL; |
0c19a9fb | 8045 | char buf[PATH_MAX]; |
0c19a9fb | 8046 | |
229fae38 SC |
8047 | if (map->skipped) |
8048 | continue; | |
8049 | ||
4580b25f THJ |
8050 | if (path) { |
8051 | int len; | |
8052 | ||
8053 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
8054 | bpf_map__name(map)); | |
8055 | if (len < 0) { | |
8056 | err = -EINVAL; | |
8057 | goto err_unpin_maps; | |
8058 | } else if (len >= PATH_MAX) { | |
8059 | err = -ENAMETOOLONG; | |
8060 | goto err_unpin_maps; | |
8061 | } | |
9cf309c5 | 8062 | sanitize_pin_path(buf); |
4580b25f THJ |
8063 | pin_path = buf; |
8064 | } else if (!map->pin_path) { | |
8065 | continue; | |
0c19a9fb SF |
8066 | } |
8067 | ||
4580b25f | 8068 | err = bpf_map__pin(map, pin_path); |
0c19a9fb SF |
8069 | if (err) |
8070 | goto err_unpin_maps; | |
8071 | } | |
8072 | ||
8073 | return 0; | |
8074 | ||
8075 | err_unpin_maps: | |
bcc40fc0 | 8076 | while ((map = bpf_object__prev_map(obj, map))) { |
4580b25f | 8077 | if (!map->pin_path) |
0c19a9fb SF |
8078 | continue; |
8079 | ||
4580b25f | 8080 | bpf_map__unpin(map, NULL); |
0c19a9fb SF |
8081 | } |
8082 | ||
e9fc3ce9 | 8083 | return libbpf_err(err); |
0c19a9fb SF |
8084 | } |
8085 | ||
8086 | int bpf_object__unpin_maps(struct bpf_object *obj, const char *path) | |
8087 | { | |
8088 | struct bpf_map *map; | |
8089 | int err; | |
8090 | ||
8091 | if (!obj) | |
e9fc3ce9 | 8092 | return libbpf_err(-ENOENT); |
0c19a9fb | 8093 | |
f74a53d9 | 8094 | bpf_object__for_each_map(map, obj) { |
4580b25f | 8095 | char *pin_path = NULL; |
d5148d85 | 8096 | char buf[PATH_MAX]; |
d5148d85 | 8097 | |
4580b25f THJ |
8098 | if (path) { |
8099 | int len; | |
8100 | ||
8101 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
8102 | bpf_map__name(map)); | |
8103 | if (len < 0) | |
e9fc3ce9 | 8104 | return libbpf_err(-EINVAL); |
4580b25f | 8105 | else if (len >= PATH_MAX) |
e9fc3ce9 | 8106 | return libbpf_err(-ENAMETOOLONG); |
9cf309c5 | 8107 | sanitize_pin_path(buf); |
4580b25f THJ |
8108 | pin_path = buf; |
8109 | } else if (!map->pin_path) { | |
8110 | continue; | |
8111 | } | |
d5148d85 | 8112 | |
4580b25f | 8113 | err = bpf_map__unpin(map, pin_path); |
d5148d85 | 8114 | if (err) |
e9fc3ce9 | 8115 | return libbpf_err(err); |
d5148d85 JS |
8116 | } |
8117 | ||
0c19a9fb SF |
8118 | return 0; |
8119 | } | |
8120 | ||
8121 | int bpf_object__pin_programs(struct bpf_object *obj, const char *path) | |
8122 | { | |
8123 | struct bpf_program *prog; | |
8124 | int err; | |
8125 | ||
8126 | if (!obj) | |
e9fc3ce9 | 8127 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
8128 | |
8129 | if (!obj->loaded) { | |
be18010e | 8130 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 8131 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
8132 | } |
8133 | ||
0c19a9fb SF |
8134 | bpf_object__for_each_program(prog, obj) { |
8135 | char buf[PATH_MAX]; | |
8136 | int len; | |
8137 | ||
8138 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 8139 | prog->pin_name); |
0c19a9fb SF |
8140 | if (len < 0) { |
8141 | err = -EINVAL; | |
8142 | goto err_unpin_programs; | |
8143 | } else if (len >= PATH_MAX) { | |
8144 | err = -ENAMETOOLONG; | |
8145 | goto err_unpin_programs; | |
8146 | } | |
8147 | ||
8148 | err = bpf_program__pin(prog, buf); | |
8149 | if (err) | |
8150 | goto err_unpin_programs; | |
8151 | } | |
8152 | ||
8153 | return 0; | |
8154 | ||
8155 | err_unpin_programs: | |
bcc40fc0 | 8156 | while ((prog = bpf_object__prev_program(obj, prog))) { |
0c19a9fb SF |
8157 | char buf[PATH_MAX]; |
8158 | int len; | |
8159 | ||
8160 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 8161 | prog->pin_name); |
0c19a9fb SF |
8162 | if (len < 0) |
8163 | continue; | |
8164 | else if (len >= PATH_MAX) | |
8165 | continue; | |
8166 | ||
8167 | bpf_program__unpin(prog, buf); | |
8168 | } | |
8169 | ||
e9fc3ce9 | 8170 | return libbpf_err(err); |
0c19a9fb SF |
8171 | } |
8172 | ||
8173 | int bpf_object__unpin_programs(struct bpf_object *obj, const char *path) | |
8174 | { | |
8175 | struct bpf_program *prog; | |
8176 | int err; | |
8177 | ||
8178 | if (!obj) | |
e9fc3ce9 | 8179 | return libbpf_err(-ENOENT); |
0c19a9fb | 8180 | |
d5148d85 JS |
8181 | bpf_object__for_each_program(prog, obj) { |
8182 | char buf[PATH_MAX]; | |
8183 | int len; | |
8184 | ||
8185 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 8186 | prog->pin_name); |
d5148d85 | 8187 | if (len < 0) |
e9fc3ce9 | 8188 | return libbpf_err(-EINVAL); |
d5148d85 | 8189 | else if (len >= PATH_MAX) |
e9fc3ce9 | 8190 | return libbpf_err(-ENAMETOOLONG); |
d5148d85 | 8191 | |
0c19a9fb | 8192 | err = bpf_program__unpin(prog, buf); |
d5148d85 | 8193 | if (err) |
e9fc3ce9 | 8194 | return libbpf_err(err); |
d5148d85 JS |
8195 | } |
8196 | ||
8197 | return 0; | |
8198 | } | |
8199 | ||
0c19a9fb SF |
8200 | int bpf_object__pin(struct bpf_object *obj, const char *path) |
8201 | { | |
8202 | int err; | |
8203 | ||
8204 | err = bpf_object__pin_maps(obj, path); | |
8205 | if (err) | |
e9fc3ce9 | 8206 | return libbpf_err(err); |
0c19a9fb SF |
8207 | |
8208 | err = bpf_object__pin_programs(obj, path); | |
8209 | if (err) { | |
8210 | bpf_object__unpin_maps(obj, path); | |
e9fc3ce9 | 8211 | return libbpf_err(err); |
0c19a9fb SF |
8212 | } |
8213 | ||
8214 | return 0; | |
8215 | } | |
8216 | ||
2d39d7c5 AN |
8217 | static void bpf_map__destroy(struct bpf_map *map) |
8218 | { | |
8219 | if (map->clear_priv) | |
8220 | map->clear_priv(map, map->priv); | |
8221 | map->priv = NULL; | |
8222 | map->clear_priv = NULL; | |
8223 | ||
646f02ff AN |
8224 | if (map->inner_map) { |
8225 | bpf_map__destroy(map->inner_map); | |
8226 | zfree(&map->inner_map); | |
8227 | } | |
8228 | ||
8229 | zfree(&map->init_slots); | |
8230 | map->init_slots_sz = 0; | |
8231 | ||
2d39d7c5 AN |
8232 | if (map->mmaped) { |
8233 | munmap(map->mmaped, bpf_map_mmap_sz(map)); | |
8234 | map->mmaped = NULL; | |
8235 | } | |
8236 | ||
8237 | if (map->st_ops) { | |
8238 | zfree(&map->st_ops->data); | |
8239 | zfree(&map->st_ops->progs); | |
8240 | zfree(&map->st_ops->kern_func_off); | |
8241 | zfree(&map->st_ops); | |
8242 | } | |
8243 | ||
8244 | zfree(&map->name); | |
aed65917 | 8245 | zfree(&map->real_name); |
2d39d7c5 AN |
8246 | zfree(&map->pin_path); |
8247 | ||
8248 | if (map->fd >= 0) | |
8249 | zclose(map->fd); | |
8250 | } | |
8251 | ||
1a5e3fb1 WN |
8252 | void bpf_object__close(struct bpf_object *obj) |
8253 | { | |
a5b8bd47 WN |
8254 | size_t i; |
8255 | ||
50450fc7 | 8256 | if (IS_ERR_OR_NULL(obj)) |
1a5e3fb1 WN |
8257 | return; |
8258 | ||
10931d24 WN |
8259 | if (obj->clear_priv) |
8260 | obj->clear_priv(obj, obj->priv); | |
8261 | ||
2e4913e0 AN |
8262 | usdt_manager_free(obj->usdt_man); |
8263 | obj->usdt_man = NULL; | |
8264 | ||
67234743 | 8265 | bpf_gen__free(obj->gen_loader); |
1a5e3fb1 | 8266 | bpf_object__elf_finish(obj); |
4a404a7e | 8267 | bpf_object_unload(obj); |
8a138aed | 8268 | btf__free(obj->btf); |
2993e051 | 8269 | btf_ext__free(obj->btf_ext); |
1a5e3fb1 | 8270 | |
2d39d7c5 AN |
8271 | for (i = 0; i < obj->nr_maps; i++) |
8272 | bpf_map__destroy(&obj->maps[i]); | |
d859900c | 8273 | |
1373ff59 | 8274 | zfree(&obj->btf_custom_path); |
8601fd42 | 8275 | zfree(&obj->kconfig); |
166750bc AN |
8276 | zfree(&obj->externs); |
8277 | obj->nr_extern = 0; | |
8278 | ||
9d759a9b WN |
8279 | zfree(&obj->maps); |
8280 | obj->nr_maps = 0; | |
a5b8bd47 WN |
8281 | |
8282 | if (obj->programs && obj->nr_programs) { | |
8283 | for (i = 0; i < obj->nr_programs; i++) | |
8284 | bpf_program__exit(&obj->programs[i]); | |
8285 | } | |
8286 | zfree(&obj->programs); | |
8287 | ||
9a208eff | 8288 | list_del(&obj->list); |
1a5e3fb1 WN |
8289 | free(obj); |
8290 | } | |
aa9b1ac3 | 8291 | |
9a208eff WN |
8292 | struct bpf_object * |
8293 | bpf_object__next(struct bpf_object *prev) | |
8294 | { | |
8295 | struct bpf_object *next; | |
689624f0 JB |
8296 | bool strict = (libbpf_mode & LIBBPF_STRICT_NO_OBJECT_LIST); |
8297 | ||
8298 | if (strict) | |
8299 | return NULL; | |
9a208eff WN |
8300 | |
8301 | if (!prev) | |
8302 | next = list_first_entry(&bpf_objects_list, | |
8303 | struct bpf_object, | |
8304 | list); | |
8305 | else | |
8306 | next = list_next_entry(prev, list); | |
8307 | ||
8308 | /* Empty list is noticed here so don't need checking on entry. */ | |
8309 | if (&next->list == &bpf_objects_list) | |
8310 | return NULL; | |
8311 | ||
8312 | return next; | |
8313 | } | |
8314 | ||
a324aae3 | 8315 | const char *bpf_object__name(const struct bpf_object *obj) |
acf860ae | 8316 | { |
e9fc3ce9 | 8317 | return obj ? obj->name : libbpf_err_ptr(-EINVAL); |
acf860ae WN |
8318 | } |
8319 | ||
a324aae3 | 8320 | unsigned int bpf_object__kversion(const struct bpf_object *obj) |
45825d8a | 8321 | { |
a7fe0450 | 8322 | return obj ? obj->kern_version : 0; |
45825d8a WN |
8323 | } |
8324 | ||
a324aae3 | 8325 | struct btf *bpf_object__btf(const struct bpf_object *obj) |
789f6bab AI |
8326 | { |
8327 | return obj ? obj->btf : NULL; | |
8328 | } | |
8329 | ||
8a138aed MKL |
8330 | int bpf_object__btf_fd(const struct bpf_object *obj) |
8331 | { | |
8332 | return obj->btf ? btf__fd(obj->btf) : -1; | |
8333 | } | |
8334 | ||
155f556d RDT |
8335 | int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version) |
8336 | { | |
8337 | if (obj->loaded) | |
e9fc3ce9 | 8338 | return libbpf_err(-EINVAL); |
155f556d RDT |
8339 | |
8340 | obj->kern_version = kern_version; | |
8341 | ||
8342 | return 0; | |
8343 | } | |
8344 | ||
10931d24 WN |
8345 | int bpf_object__set_priv(struct bpf_object *obj, void *priv, |
8346 | bpf_object_clear_priv_t clear_priv) | |
8347 | { | |
8348 | if (obj->priv && obj->clear_priv) | |
8349 | obj->clear_priv(obj, obj->priv); | |
8350 | ||
8351 | obj->priv = priv; | |
8352 | obj->clear_priv = clear_priv; | |
8353 | return 0; | |
8354 | } | |
8355 | ||
a324aae3 | 8356 | void *bpf_object__priv(const struct bpf_object *obj) |
10931d24 | 8357 | { |
e9fc3ce9 | 8358 | return obj ? obj->priv : libbpf_err_ptr(-EINVAL); |
10931d24 WN |
8359 | } |
8360 | ||
67234743 AS |
8361 | int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts) |
8362 | { | |
8363 | struct bpf_gen *gen; | |
8364 | ||
8365 | if (!opts) | |
8366 | return -EFAULT; | |
8367 | if (!OPTS_VALID(opts, gen_loader_opts)) | |
8368 | return -EINVAL; | |
8369 | gen = calloc(sizeof(*gen), 1); | |
8370 | if (!gen) | |
8371 | return -ENOMEM; | |
8372 | gen->opts = opts; | |
8373 | obj->gen_loader = gen; | |
8374 | return 0; | |
8375 | } | |
8376 | ||
eac7d845 | 8377 | static struct bpf_program * |
a324aae3 AN |
8378 | __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj, |
8379 | bool forward) | |
aa9b1ac3 | 8380 | { |
a83d6e76 | 8381 | size_t nr_programs = obj->nr_programs; |
0c19a9fb | 8382 | ssize_t idx; |
aa9b1ac3 | 8383 | |
a83d6e76 | 8384 | if (!nr_programs) |
aa9b1ac3 | 8385 | return NULL; |
aa9b1ac3 | 8386 | |
a83d6e76 MKL |
8387 | if (!p) |
8388 | /* Iter from the beginning */ | |
8389 | return forward ? &obj->programs[0] : | |
8390 | &obj->programs[nr_programs - 1]; | |
8391 | ||
0c19a9fb | 8392 | if (p->obj != obj) { |
be18010e | 8393 | pr_warn("error: program handler doesn't match object\n"); |
e9fc3ce9 | 8394 | return errno = EINVAL, NULL; |
aa9b1ac3 WN |
8395 | } |
8396 | ||
a83d6e76 | 8397 | idx = (p - obj->programs) + (forward ? 1 : -1); |
0c19a9fb | 8398 | if (idx >= obj->nr_programs || idx < 0) |
aa9b1ac3 WN |
8399 | return NULL; |
8400 | return &obj->programs[idx]; | |
8401 | } | |
8402 | ||
eac7d845 | 8403 | struct bpf_program * |
a324aae3 | 8404 | bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj) |
2088a3a7 HC |
8405 | { |
8406 | return bpf_object__next_program(obj, prev); | |
8407 | } | |
8408 | ||
8409 | struct bpf_program * | |
8410 | bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev) | |
eac7d845 JK |
8411 | { |
8412 | struct bpf_program *prog = prev; | |
8413 | ||
8414 | do { | |
a83d6e76 | 8415 | prog = __bpf_program__iter(prog, obj, true); |
c3c55696 | 8416 | } while (prog && prog_is_subprog(obj, prog)); |
0c19a9fb SF |
8417 | |
8418 | return prog; | |
8419 | } | |
8420 | ||
8421 | struct bpf_program * | |
a324aae3 | 8422 | bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj) |
2088a3a7 HC |
8423 | { |
8424 | return bpf_object__prev_program(obj, next); | |
8425 | } | |
8426 | ||
8427 | struct bpf_program * | |
8428 | bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next) | |
0c19a9fb SF |
8429 | { |
8430 | struct bpf_program *prog = next; | |
8431 | ||
0c19a9fb | 8432 | do { |
a83d6e76 | 8433 | prog = __bpf_program__iter(prog, obj, false); |
c3c55696 | 8434 | } while (prog && prog_is_subprog(obj, prog)); |
eac7d845 JK |
8435 | |
8436 | return prog; | |
8437 | } | |
8438 | ||
edb13ed4 ACM |
8439 | int bpf_program__set_priv(struct bpf_program *prog, void *priv, |
8440 | bpf_program_clear_priv_t clear_priv) | |
aa9b1ac3 WN |
8441 | { |
8442 | if (prog->priv && prog->clear_priv) | |
8443 | prog->clear_priv(prog, prog->priv); | |
8444 | ||
8445 | prog->priv = priv; | |
8446 | prog->clear_priv = clear_priv; | |
8447 | return 0; | |
8448 | } | |
8449 | ||
a324aae3 | 8450 | void *bpf_program__priv(const struct bpf_program *prog) |
aa9b1ac3 | 8451 | { |
e9fc3ce9 | 8452 | return prog ? prog->priv : libbpf_err_ptr(-EINVAL); |
aa9b1ac3 WN |
8453 | } |
8454 | ||
9aba3613 JK |
8455 | void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex) |
8456 | { | |
8457 | prog->prog_ifindex = ifindex; | |
8458 | } | |
8459 | ||
01af3bf0 AN |
8460 | const char *bpf_program__name(const struct bpf_program *prog) |
8461 | { | |
8462 | return prog->name; | |
8463 | } | |
8464 | ||
52109584 AN |
8465 | const char *bpf_program__section_name(const struct bpf_program *prog) |
8466 | { | |
8467 | return prog->sec_name; | |
8468 | } | |
8469 | ||
a324aae3 | 8470 | const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy) |
aa9b1ac3 WN |
8471 | { |
8472 | const char *title; | |
8473 | ||
52109584 | 8474 | title = prog->sec_name; |
715f8db9 | 8475 | if (needs_copy) { |
aa9b1ac3 WN |
8476 | title = strdup(title); |
8477 | if (!title) { | |
be18010e | 8478 | pr_warn("failed to strdup program title\n"); |
e9fc3ce9 | 8479 | return libbpf_err_ptr(-ENOMEM); |
aa9b1ac3 WN |
8480 | } |
8481 | } | |
8482 | ||
8483 | return title; | |
8484 | } | |
8485 | ||
d9297581 AN |
8486 | bool bpf_program__autoload(const struct bpf_program *prog) |
8487 | { | |
a3820c48 | 8488 | return prog->autoload; |
d9297581 AN |
8489 | } |
8490 | ||
8491 | int bpf_program__set_autoload(struct bpf_program *prog, bool autoload) | |
8492 | { | |
8493 | if (prog->obj->loaded) | |
e9fc3ce9 | 8494 | return libbpf_err(-EINVAL); |
d9297581 | 8495 | |
a3820c48 | 8496 | prog->autoload = autoload; |
d9297581 AN |
8497 | return 0; |
8498 | } | |
8499 | ||
bcc40fc0 AN |
8500 | static int bpf_program_nth_fd(const struct bpf_program *prog, int n); |
8501 | ||
a324aae3 | 8502 | int bpf_program__fd(const struct bpf_program *prog) |
aa9b1ac3 | 8503 | { |
bcc40fc0 | 8504 | return bpf_program_nth_fd(prog, 0); |
b580563e WN |
8505 | } |
8506 | ||
1a734efe THJ |
8507 | size_t bpf_program__size(const struct bpf_program *prog) |
8508 | { | |
9c0f8cbd | 8509 | return prog->insns_cnt * BPF_INSN_SZ; |
1a734efe THJ |
8510 | } |
8511 | ||
65a7fa2e AN |
8512 | const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog) |
8513 | { | |
8514 | return prog->insns; | |
8515 | } | |
8516 | ||
8517 | size_t bpf_program__insn_cnt(const struct bpf_program *prog) | |
8518 | { | |
8519 | return prog->insns_cnt; | |
8520 | } | |
8521 | ||
b580563e WN |
8522 | int bpf_program__set_prep(struct bpf_program *prog, int nr_instances, |
8523 | bpf_program_prep_t prep) | |
8524 | { | |
8525 | int *instances_fds; | |
8526 | ||
8527 | if (nr_instances <= 0 || !prep) | |
e9fc3ce9 | 8528 | return libbpf_err(-EINVAL); |
b580563e WN |
8529 | |
8530 | if (prog->instances.nr > 0 || prog->instances.fds) { | |
be18010e | 8531 | pr_warn("Can't set pre-processor after loading\n"); |
e9fc3ce9 | 8532 | return libbpf_err(-EINVAL); |
b580563e WN |
8533 | } |
8534 | ||
8535 | instances_fds = malloc(sizeof(int) * nr_instances); | |
8536 | if (!instances_fds) { | |
be18010e | 8537 | pr_warn("alloc memory failed for fds\n"); |
e9fc3ce9 | 8538 | return libbpf_err(-ENOMEM); |
b580563e WN |
8539 | } |
8540 | ||
8541 | /* fill all fd with -1 */ | |
8542 | memset(instances_fds, -1, sizeof(int) * nr_instances); | |
8543 | ||
8544 | prog->instances.nr = nr_instances; | |
8545 | prog->instances.fds = instances_fds; | |
8546 | prog->preprocessor = prep; | |
8547 | return 0; | |
8548 | } | |
8549 | ||
bcc40fc0 AN |
8550 | __attribute__((alias("bpf_program_nth_fd"))) |
8551 | int bpf_program__nth_fd(const struct bpf_program *prog, int n); | |
8552 | ||
8553 | static int bpf_program_nth_fd(const struct bpf_program *prog, int n) | |
b580563e WN |
8554 | { |
8555 | int fd; | |
8556 | ||
1e960043 | 8557 | if (!prog) |
e9fc3ce9 | 8558 | return libbpf_err(-EINVAL); |
1e960043 | 8559 | |
b580563e | 8560 | if (n >= prog->instances.nr || n < 0) { |
be18010e | 8561 | pr_warn("Can't get the %dth fd from program %s: only %d instances\n", |
52109584 | 8562 | n, prog->name, prog->instances.nr); |
e9fc3ce9 | 8563 | return libbpf_err(-EINVAL); |
b580563e WN |
8564 | } |
8565 | ||
8566 | fd = prog->instances.fds[n]; | |
8567 | if (fd < 0) { | |
be18010e | 8568 | pr_warn("%dth instance of program '%s' is invalid\n", |
52109584 | 8569 | n, prog->name); |
e9fc3ce9 | 8570 | return libbpf_err(-ENOENT); |
b580563e WN |
8571 | } |
8572 | ||
8573 | return fd; | |
aa9b1ac3 | 8574 | } |
9d759a9b | 8575 | |
20eccf29 AN |
8576 | __alias(bpf_program__type) |
8577 | enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog); | |
8578 | ||
8579 | enum bpf_prog_type bpf_program__type(const struct bpf_program *prog) | |
f1eead9e AN |
8580 | { |
8581 | return prog->type; | |
8582 | } | |
8583 | ||
93442f13 | 8584 | int bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type) |
5f44e4c8 | 8585 | { |
93442f13 GS |
8586 | if (prog->obj->loaded) |
8587 | return libbpf_err(-EBUSY); | |
8588 | ||
5f44e4c8 | 8589 | prog->type = type; |
93442f13 | 8590 | return 0; |
5f44e4c8 WN |
8591 | } |
8592 | ||
a324aae3 | 8593 | static bool bpf_program__is_type(const struct bpf_program *prog, |
5f44e4c8 WN |
8594 | enum bpf_prog_type type) |
8595 | { | |
8596 | return prog ? (prog->type == type) : false; | |
8597 | } | |
8598 | ||
a324aae3 AN |
8599 | #define BPF_PROG_TYPE_FNS(NAME, TYPE) \ |
8600 | int bpf_program__set_##NAME(struct bpf_program *prog) \ | |
8601 | { \ | |
8602 | if (!prog) \ | |
e9fc3ce9 | 8603 | return libbpf_err(-EINVAL); \ |
df286716 | 8604 | return bpf_program__set_type(prog, TYPE); \ |
a324aae3 AN |
8605 | } \ |
8606 | \ | |
8607 | bool bpf_program__is_##NAME(const struct bpf_program *prog) \ | |
8608 | { \ | |
8609 | return bpf_program__is_type(prog, TYPE); \ | |
8610 | } \ | |
ed794073 | 8611 | |
7803ba73 | 8612 | BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER); |
1e092a03 | 8613 | BPF_PROG_TYPE_FNS(lsm, BPF_PROG_TYPE_LSM); |
ed794073 | 8614 | BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE); |
7803ba73 JS |
8615 | BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS); |
8616 | BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT); | |
ed794073 | 8617 | BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT); |
e14c93fd | 8618 | BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT); |
7803ba73 JS |
8619 | BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP); |
8620 | BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT); | |
12a8654b | 8621 | BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING); |
590a0088 | 8622 | BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS); |
2db6eab1 | 8623 | BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT); |
499dd29d | 8624 | BPF_PROG_TYPE_FNS(sk_lookup, BPF_PROG_TYPE_SK_LOOKUP); |
5f44e4c8 | 8625 | |
20eccf29 AN |
8626 | __alias(bpf_program__expected_attach_type) |
8627 | enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog); | |
8628 | ||
8629 | enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog) | |
f1eead9e AN |
8630 | { |
8631 | return prog->expected_attach_type; | |
8632 | } | |
8633 | ||
93442f13 | 8634 | int bpf_program__set_expected_attach_type(struct bpf_program *prog, |
16962b24 | 8635 | enum bpf_attach_type type) |
d7be143b | 8636 | { |
93442f13 GS |
8637 | if (prog->obj->loaded) |
8638 | return libbpf_err(-EBUSY); | |
8639 | ||
d7be143b | 8640 | prog->expected_attach_type = type; |
93442f13 | 8641 | return 0; |
d7be143b AI |
8642 | } |
8643 | ||
a6ca7158 AN |
8644 | __u32 bpf_program__flags(const struct bpf_program *prog) |
8645 | { | |
8646 | return prog->prog_flags; | |
8647 | } | |
8648 | ||
8cccee9e | 8649 | int bpf_program__set_flags(struct bpf_program *prog, __u32 flags) |
a6ca7158 AN |
8650 | { |
8651 | if (prog->obj->loaded) | |
8652 | return libbpf_err(-EBUSY); | |
8653 | ||
8cccee9e | 8654 | prog->prog_flags = flags; |
a6ca7158 AN |
8655 | return 0; |
8656 | } | |
8657 | ||
dbdd2c7f AN |
8658 | __u32 bpf_program__log_level(const struct bpf_program *prog) |
8659 | { | |
8660 | return prog->log_level; | |
8661 | } | |
8662 | ||
8663 | int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level) | |
8664 | { | |
8665 | if (prog->obj->loaded) | |
8666 | return libbpf_err(-EBUSY); | |
8667 | ||
8668 | prog->log_level = log_level; | |
8669 | return 0; | |
8670 | } | |
8671 | ||
b3ce9079 AN |
8672 | const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size) |
8673 | { | |
8674 | *log_size = prog->log_size; | |
8675 | return prog->log_buf; | |
8676 | } | |
8677 | ||
8678 | int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size) | |
8679 | { | |
8680 | if (log_size && !log_buf) | |
8681 | return -EINVAL; | |
8682 | if (prog->log_size > UINT_MAX) | |
8683 | return -EINVAL; | |
8684 | if (prog->obj->loaded) | |
8685 | return -EBUSY; | |
8686 | ||
8687 | prog->log_buf = log_buf; | |
8688 | prog->log_size = log_size; | |
a6ca7158 AN |
8689 | return 0; |
8690 | } | |
8691 | ||
15ea31fa | 8692 | #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \ |
697f104d | 8693 | .sec = (char *)sec_pfx, \ |
d7a18ea7 | 8694 | .prog_type = BPF_PROG_TYPE_##ptype, \ |
15ea31fa AN |
8695 | .expected_attach_type = atype, \ |
8696 | .cookie = (long)(flags), \ | |
4fa5bcfe | 8697 | .prog_prepare_load_fn = libbpf_prepare_prog_load, \ |
d7a18ea7 AN |
8698 | __VA_ARGS__ \ |
8699 | } | |
8700 | ||
4fa5bcfe | 8701 | static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
39f8dc43 | 8702 | static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
2e4913e0 | 8703 | static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
4fa5bcfe AN |
8704 | static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
8705 | static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
8706 | static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
ddc6b049 | 8707 | static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
4fa5bcfe AN |
8708 | static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
8709 | static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
d7a18ea7 | 8710 | |
d7a18ea7 | 8711 | static const struct bpf_sec_def section_defs[] = { |
dd94d45c AN |
8712 | SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE | SEC_SLOPPY_PFX), |
8713 | SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8714 | SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
15ea31fa | 8715 | SEC_DEF("kprobe/", KPROBE, 0, SEC_NONE, attach_kprobe), |
39f8dc43 | 8716 | SEC_DEF("uprobe+", KPROBE, 0, SEC_NONE, attach_uprobe), |
15ea31fa | 8717 | SEC_DEF("kretprobe/", KPROBE, 0, SEC_NONE, attach_kprobe), |
39f8dc43 | 8718 | SEC_DEF("uretprobe+", KPROBE, 0, SEC_NONE, attach_uprobe), |
ddc6b049 JO |
8719 | SEC_DEF("kprobe.multi/", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi), |
8720 | SEC_DEF("kretprobe.multi/", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi), | |
2e4913e0 | 8721 | SEC_DEF("usdt+", KPROBE, 0, SEC_NONE, attach_usdt), |
15ea31fa | 8722 | SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE), |
4a4d4cee | 8723 | SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE | SEC_SLOPPY_PFX | SEC_DEPRECATED), |
dd94d45c | 8724 | SEC_DEF("action", SCHED_ACT, 0, SEC_NONE | SEC_SLOPPY_PFX), |
15ea31fa AN |
8725 | SEC_DEF("tracepoint/", TRACEPOINT, 0, SEC_NONE, attach_tp), |
8726 | SEC_DEF("tp/", TRACEPOINT, 0, SEC_NONE, attach_tp), | |
8727 | SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
8728 | SEC_DEF("raw_tp/", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
ccaf12d6 HT |
8729 | SEC_DEF("raw_tracepoint.w/", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), |
8730 | SEC_DEF("raw_tp.w/", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), | |
15ea31fa AN |
8731 | SEC_DEF("tp_btf/", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace), |
8732 | SEC_DEF("fentry/", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace), | |
8733 | SEC_DEF("fmod_ret/", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace), | |
8734 | SEC_DEF("fexit/", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace), | |
8735 | SEC_DEF("fentry.s/", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8736 | SEC_DEF("fmod_ret.s/", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8737 | SEC_DEF("fexit.s/", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8738 | SEC_DEF("freplace/", EXT, 0, SEC_ATTACH_BTF, attach_trace), | |
8739 | SEC_DEF("lsm/", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm), | |
8740 | SEC_DEF("lsm.s/", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm), | |
8741 | SEC_DEF("iter/", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter), | |
a8b77f74 | 8742 | SEC_DEF("iter.s/", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter), |
15ea31fa | 8743 | SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE), |
082c4bfb | 8744 | SEC_DEF("xdp.frags/devmap", XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS), |
4a4d4cee LB |
8745 | SEC_DEF("xdp/devmap", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE), |
8746 | SEC_DEF("xdp_devmap/", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE | SEC_DEPRECATED), | |
082c4bfb | 8747 | SEC_DEF("xdp.frags/cpumap", XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS), |
4a4d4cee LB |
8748 | SEC_DEF("xdp/cpumap", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE), |
8749 | SEC_DEF("xdp_cpumap/", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE | SEC_DEPRECATED), | |
082c4bfb | 8750 | SEC_DEF("xdp.frags", XDP, BPF_XDP, SEC_XDP_FRAGS), |
dd94d45c AN |
8751 | SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), |
8752 | SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8753 | SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8754 | SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8755 | SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8756 | SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8757 | SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8758 | SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8759 | SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8760 | SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8761 | SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8762 | SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8763 | SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8764 | SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8765 | SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8766 | SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8767 | SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8768 | SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8769 | SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8770 | SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8771 | SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8772 | SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8773 | SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8774 | SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8775 | SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8776 | SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8777 | SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8778 | SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8779 | SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8780 | SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8781 | SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8782 | SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8783 | SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8784 | SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8785 | SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8786 | SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8787 | SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8788 | SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), | |
7c80c87a | 8789 | SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX), |
583c9009 | 8790 | }; |
d7be143b | 8791 | |
697f104d AN |
8792 | static size_t custom_sec_def_cnt; |
8793 | static struct bpf_sec_def *custom_sec_defs; | |
8794 | static struct bpf_sec_def custom_fallback_def; | |
8795 | static bool has_custom_fallback_def; | |
c76e4c22 | 8796 | |
697f104d AN |
8797 | static int last_custom_sec_def_handler_id; |
8798 | ||
8799 | int libbpf_register_prog_handler(const char *sec, | |
8800 | enum bpf_prog_type prog_type, | |
8801 | enum bpf_attach_type exp_attach_type, | |
8802 | const struct libbpf_prog_handler_opts *opts) | |
d7a18ea7 | 8803 | { |
697f104d | 8804 | struct bpf_sec_def *sec_def; |
d7a18ea7 | 8805 | |
697f104d AN |
8806 | if (!OPTS_VALID(opts, libbpf_prog_handler_opts)) |
8807 | return libbpf_err(-EINVAL); | |
dd94d45c | 8808 | |
697f104d AN |
8809 | if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */ |
8810 | return libbpf_err(-E2BIG); | |
dd94d45c | 8811 | |
697f104d AN |
8812 | if (sec) { |
8813 | sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1, | |
8814 | sizeof(*sec_def)); | |
8815 | if (!sec_def) | |
8816 | return libbpf_err(-ENOMEM); | |
dd94d45c | 8817 | |
697f104d AN |
8818 | custom_sec_defs = sec_def; |
8819 | sec_def = &custom_sec_defs[custom_sec_def_cnt]; | |
8820 | } else { | |
8821 | if (has_custom_fallback_def) | |
8822 | return libbpf_err(-EBUSY); | |
dd94d45c | 8823 | |
697f104d AN |
8824 | sec_def = &custom_fallback_def; |
8825 | } | |
8826 | ||
8827 | sec_def->sec = sec ? strdup(sec) : NULL; | |
8828 | if (sec && !sec_def->sec) | |
8829 | return libbpf_err(-ENOMEM); | |
8830 | ||
8831 | sec_def->prog_type = prog_type; | |
8832 | sec_def->expected_attach_type = exp_attach_type; | |
8833 | sec_def->cookie = OPTS_GET(opts, cookie, 0); | |
8834 | ||
8835 | sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL); | |
8836 | sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL); | |
8837 | sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL); | |
8838 | ||
8839 | sec_def->handler_id = ++last_custom_sec_def_handler_id; | |
8840 | ||
8841 | if (sec) | |
8842 | custom_sec_def_cnt++; | |
8843 | else | |
8844 | has_custom_fallback_def = true; | |
8845 | ||
8846 | return sec_def->handler_id; | |
8847 | } | |
8848 | ||
8849 | int libbpf_unregister_prog_handler(int handler_id) | |
8850 | { | |
8851 | struct bpf_sec_def *sec_defs; | |
8852 | int i; | |
8853 | ||
8854 | if (handler_id <= 0) | |
8855 | return libbpf_err(-EINVAL); | |
8856 | ||
8857 | if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) { | |
8858 | memset(&custom_fallback_def, 0, sizeof(custom_fallback_def)); | |
8859 | has_custom_fallback_def = false; | |
8860 | return 0; | |
8861 | } | |
8862 | ||
8863 | for (i = 0; i < custom_sec_def_cnt; i++) { | |
8864 | if (custom_sec_defs[i].handler_id == handler_id) | |
8865 | break; | |
8866 | } | |
8867 | ||
8868 | if (i == custom_sec_def_cnt) | |
8869 | return libbpf_err(-ENOENT); | |
8870 | ||
8871 | free(custom_sec_defs[i].sec); | |
8872 | for (i = i + 1; i < custom_sec_def_cnt; i++) | |
8873 | custom_sec_defs[i - 1] = custom_sec_defs[i]; | |
8874 | custom_sec_def_cnt--; | |
8875 | ||
8876 | /* try to shrink the array, but it's ok if we couldn't */ | |
8877 | sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs)); | |
8878 | if (sec_defs) | |
8879 | custom_sec_defs = sec_defs; | |
8880 | ||
8881 | return 0; | |
8882 | } | |
8883 | ||
8884 | static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name, | |
8885 | bool allow_sloppy) | |
8886 | { | |
8887 | size_t len = strlen(sec_def->sec); | |
8888 | ||
8889 | /* "type/" always has to have proper SEC("type/extras") form */ | |
8890 | if (sec_def->sec[len - 1] == '/') { | |
8891 | if (str_has_pfx(sec_name, sec_def->sec)) | |
8892 | return true; | |
8893 | return false; | |
8894 | } | |
8895 | ||
8896 | /* "type+" means it can be either exact SEC("type") or | |
8897 | * well-formed SEC("type/extras") with proper '/' separator | |
8898 | */ | |
8899 | if (sec_def->sec[len - 1] == '+') { | |
8900 | len--; | |
8901 | /* not even a prefix */ | |
8902 | if (strncmp(sec_name, sec_def->sec, len) != 0) | |
8903 | return false; | |
8904 | /* exact match or has '/' separator */ | |
8905 | if (sec_name[len] == '\0' || sec_name[len] == '/') | |
8906 | return true; | |
8907 | return false; | |
8908 | } | |
8909 | ||
8910 | /* SEC_SLOPPY_PFX definitions are allowed to be just prefix | |
8911 | * matches, unless strict section name mode | |
8912 | * (LIBBPF_STRICT_SEC_NAME) is enabled, in which case the | |
8913 | * match has to be exact. | |
8914 | */ | |
8915 | if (allow_sloppy && str_has_pfx(sec_name, sec_def->sec)) | |
8916 | return true; | |
8917 | ||
8918 | /* Definitions not marked SEC_SLOPPY_PFX (e.g., | |
8919 | * SEC("syscall")) are exact matches in both modes. | |
8920 | */ | |
8921 | return strcmp(sec_name, sec_def->sec) == 0; | |
8922 | } | |
8923 | ||
8924 | static const struct bpf_sec_def *find_sec_def(const char *sec_name) | |
8925 | { | |
8926 | const struct bpf_sec_def *sec_def; | |
8927 | int i, n; | |
8928 | bool strict = libbpf_mode & LIBBPF_STRICT_SEC_NAME, allow_sloppy; | |
8929 | ||
8930 | n = custom_sec_def_cnt; | |
8931 | for (i = 0; i < n; i++) { | |
8932 | sec_def = &custom_sec_defs[i]; | |
8933 | if (sec_def_matches(sec_def, sec_name, false)) | |
8934 | return sec_def; | |
8935 | } | |
8936 | ||
8937 | n = ARRAY_SIZE(section_defs); | |
8938 | for (i = 0; i < n; i++) { | |
8939 | sec_def = §ion_defs[i]; | |
8940 | allow_sloppy = (sec_def->cookie & SEC_SLOPPY_PFX) && !strict; | |
8941 | if (sec_def_matches(sec_def, sec_name, allow_sloppy)) | |
dd94d45c | 8942 | return sec_def; |
d7a18ea7 | 8943 | } |
697f104d AN |
8944 | |
8945 | if (has_custom_fallback_def) | |
8946 | return &custom_fallback_def; | |
8947 | ||
d7a18ea7 AN |
8948 | return NULL; |
8949 | } | |
8950 | ||
697f104d AN |
8951 | #define MAX_TYPE_NAME_SIZE 32 |
8952 | ||
c76e4c22 TS |
8953 | static char *libbpf_get_type_names(bool attach_type) |
8954 | { | |
d7a18ea7 | 8955 | int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE; |
c76e4c22 TS |
8956 | char *buf; |
8957 | ||
8958 | buf = malloc(len); | |
8959 | if (!buf) | |
8960 | return NULL; | |
8961 | ||
8962 | buf[0] = '\0'; | |
8963 | /* Forge string buf with all available names */ | |
d7a18ea7 | 8964 | for (i = 0; i < ARRAY_SIZE(section_defs); i++) { |
15ea31fa AN |
8965 | const struct bpf_sec_def *sec_def = §ion_defs[i]; |
8966 | ||
8967 | if (attach_type) { | |
4fa5bcfe | 8968 | if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load) |
15ea31fa AN |
8969 | continue; |
8970 | ||
8971 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
8972 | continue; | |
8973 | } | |
c76e4c22 | 8974 | |
d7a18ea7 | 8975 | if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) { |
c76e4c22 TS |
8976 | free(buf); |
8977 | return NULL; | |
8978 | } | |
8979 | strcat(buf, " "); | |
d7a18ea7 | 8980 | strcat(buf, section_defs[i].sec); |
c76e4c22 TS |
8981 | } |
8982 | ||
8983 | return buf; | |
8984 | } | |
8985 | ||
b60df2a0 JK |
8986 | int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type, |
8987 | enum bpf_attach_type *expected_attach_type) | |
583c9009 | 8988 | { |
d7a18ea7 | 8989 | const struct bpf_sec_def *sec_def; |
c76e4c22 | 8990 | char *type_names; |
583c9009 | 8991 | |
b60df2a0 | 8992 | if (!name) |
e9fc3ce9 | 8993 | return libbpf_err(-EINVAL); |
583c9009 | 8994 | |
d7a18ea7 AN |
8995 | sec_def = find_sec_def(name); |
8996 | if (sec_def) { | |
8997 | *prog_type = sec_def->prog_type; | |
8998 | *expected_attach_type = sec_def->expected_attach_type; | |
b60df2a0 JK |
8999 | return 0; |
9000 | } | |
d7a18ea7 | 9001 | |
4a3d6c6a | 9002 | pr_debug("failed to guess program type from ELF section '%s'\n", name); |
c76e4c22 TS |
9003 | type_names = libbpf_get_type_names(false); |
9004 | if (type_names != NULL) { | |
3f519353 | 9005 | pr_debug("supported section(type) names are:%s\n", type_names); |
c76e4c22 TS |
9006 | free(type_names); |
9007 | } | |
9008 | ||
e9fc3ce9 | 9009 | return libbpf_err(-ESRCH); |
b60df2a0 | 9010 | } |
583c9009 | 9011 | |
590a0088 MKL |
9012 | static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj, |
9013 | size_t offset) | |
9014 | { | |
9015 | struct bpf_map *map; | |
9016 | size_t i; | |
9017 | ||
9018 | for (i = 0; i < obj->nr_maps; i++) { | |
9019 | map = &obj->maps[i]; | |
9020 | if (!bpf_map__is_struct_ops(map)) | |
9021 | continue; | |
9022 | if (map->sec_offset <= offset && | |
9023 | offset - map->sec_offset < map->def.value_size) | |
9024 | return map; | |
9025 | } | |
9026 | ||
9027 | return NULL; | |
9028 | } | |
9029 | ||
9030 | /* Collect the reloc from ELF and populate the st_ops->progs[] */ | |
646f02ff | 9031 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 9032 | Elf64_Shdr *shdr, Elf_Data *data) |
590a0088 MKL |
9033 | { |
9034 | const struct btf_member *member; | |
9035 | struct bpf_struct_ops *st_ops; | |
9036 | struct bpf_program *prog; | |
9037 | unsigned int shdr_idx; | |
9038 | const struct btf *btf; | |
9039 | struct bpf_map *map; | |
7e06aad5 | 9040 | unsigned int moff, insn_idx; |
590a0088 | 9041 | const char *name; |
1d1a3bcf | 9042 | __u32 member_idx; |
ad23b723 AN |
9043 | Elf64_Sym *sym; |
9044 | Elf64_Rel *rel; | |
590a0088 MKL |
9045 | int i, nrels; |
9046 | ||
590a0088 MKL |
9047 | btf = obj->btf; |
9048 | nrels = shdr->sh_size / shdr->sh_entsize; | |
9049 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
9050 | rel = elf_rel_by_idx(data, i); |
9051 | if (!rel) { | |
590a0088 MKL |
9052 | pr_warn("struct_ops reloc: failed to get %d reloc\n", i); |
9053 | return -LIBBPF_ERRNO__FORMAT; | |
9054 | } | |
9055 | ||
ad23b723 AN |
9056 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); |
9057 | if (!sym) { | |
590a0088 | 9058 | pr_warn("struct_ops reloc: symbol %zx not found\n", |
ad23b723 | 9059 | (size_t)ELF64_R_SYM(rel->r_info)); |
590a0088 MKL |
9060 | return -LIBBPF_ERRNO__FORMAT; |
9061 | } | |
9062 | ||
ad23b723 AN |
9063 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
9064 | map = find_struct_ops_map_by_offset(obj, rel->r_offset); | |
590a0088 | 9065 | if (!map) { |
ad23b723 AN |
9066 | pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n", |
9067 | (size_t)rel->r_offset); | |
590a0088 MKL |
9068 | return -EINVAL; |
9069 | } | |
9070 | ||
ad23b723 AN |
9071 | moff = rel->r_offset - map->sec_offset; |
9072 | shdr_idx = sym->st_shndx; | |
590a0088 | 9073 | st_ops = map->st_ops; |
ad23b723 | 9074 | 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", |
590a0088 | 9075 | map->name, |
ad23b723 AN |
9076 | (long long)(rel->r_info >> 32), |
9077 | (long long)sym->st_value, | |
9078 | shdr_idx, (size_t)rel->r_offset, | |
9079 | map->sec_offset, sym->st_name, name); | |
590a0088 MKL |
9080 | |
9081 | if (shdr_idx >= SHN_LORESERVE) { | |
ad23b723 AN |
9082 | pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n", |
9083 | map->name, (size_t)rel->r_offset, shdr_idx); | |
590a0088 MKL |
9084 | return -LIBBPF_ERRNO__RELOC; |
9085 | } | |
ad23b723 | 9086 | if (sym->st_value % BPF_INSN_SZ) { |
7e06aad5 | 9087 | pr_warn("struct_ops reloc %s: invalid target program offset %llu\n", |
ad23b723 | 9088 | map->name, (unsigned long long)sym->st_value); |
7e06aad5 AN |
9089 | return -LIBBPF_ERRNO__FORMAT; |
9090 | } | |
ad23b723 | 9091 | insn_idx = sym->st_value / BPF_INSN_SZ; |
590a0088 MKL |
9092 | |
9093 | member = find_member_by_offset(st_ops->type, moff * 8); | |
9094 | if (!member) { | |
9095 | pr_warn("struct_ops reloc %s: cannot find member at moff %u\n", | |
9096 | map->name, moff); | |
9097 | return -EINVAL; | |
9098 | } | |
9099 | member_idx = member - btf_members(st_ops->type); | |
9100 | name = btf__name_by_offset(btf, member->name_off); | |
9101 | ||
9102 | if (!resolve_func_ptr(btf, member->type, NULL)) { | |
9103 | pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n", | |
9104 | map->name, name); | |
9105 | return -EINVAL; | |
9106 | } | |
9107 | ||
7e06aad5 | 9108 | prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx); |
590a0088 MKL |
9109 | if (!prog) { |
9110 | pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n", | |
9111 | map->name, shdr_idx, name); | |
9112 | return -EINVAL; | |
9113 | } | |
9114 | ||
91b4d1d1 AN |
9115 | /* prevent the use of BPF prog with invalid type */ |
9116 | if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) { | |
9117 | pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n", | |
9118 | map->name, prog->name); | |
9119 | return -EINVAL; | |
9120 | } | |
590a0088 | 9121 | |
91b4d1d1 AN |
9122 | /* if we haven't yet processed this BPF program, record proper |
9123 | * attach_btf_id and member_idx | |
9124 | */ | |
9125 | if (!prog->attach_btf_id) { | |
590a0088 MKL |
9126 | prog->attach_btf_id = st_ops->type_id; |
9127 | prog->expected_attach_type = member_idx; | |
590a0088 | 9128 | } |
91b4d1d1 AN |
9129 | |
9130 | /* struct_ops BPF prog can be re-used between multiple | |
9131 | * .struct_ops as long as it's the same struct_ops struct | |
9132 | * definition and the same function pointer field | |
9133 | */ | |
9134 | if (prog->attach_btf_id != st_ops->type_id || | |
9135 | prog->expected_attach_type != member_idx) { | |
9136 | 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", | |
9137 | map->name, prog->name, prog->sec_name, prog->type, | |
9138 | prog->attach_btf_id, prog->expected_attach_type, name); | |
9139 | return -EINVAL; | |
9140 | } | |
9141 | ||
590a0088 MKL |
9142 | st_ops->progs[member_idx] = prog; |
9143 | } | |
9144 | ||
9145 | return 0; | |
590a0088 MKL |
9146 | } |
9147 | ||
a6ed02ca | 9148 | #define BTF_TRACE_PREFIX "btf_trace_" |
1e092a03 | 9149 | #define BTF_LSM_PREFIX "bpf_lsm_" |
21aef70e | 9150 | #define BTF_ITER_PREFIX "bpf_iter_" |
a6ed02ca KS |
9151 | #define BTF_MAX_NAME_SIZE 128 |
9152 | ||
67234743 AS |
9153 | void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type, |
9154 | const char **prefix, int *kind) | |
9155 | { | |
9156 | switch (attach_type) { | |
9157 | case BPF_TRACE_RAW_TP: | |
9158 | *prefix = BTF_TRACE_PREFIX; | |
9159 | *kind = BTF_KIND_TYPEDEF; | |
9160 | break; | |
9161 | case BPF_LSM_MAC: | |
9162 | *prefix = BTF_LSM_PREFIX; | |
9163 | *kind = BTF_KIND_FUNC; | |
9164 | break; | |
9165 | case BPF_TRACE_ITER: | |
9166 | *prefix = BTF_ITER_PREFIX; | |
9167 | *kind = BTF_KIND_FUNC; | |
9168 | break; | |
9169 | default: | |
9170 | *prefix = ""; | |
9171 | *kind = BTF_KIND_FUNC; | |
9172 | } | |
9173 | } | |
9174 | ||
a6ed02ca KS |
9175 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
9176 | const char *name, __u32 kind) | |
9177 | { | |
9178 | char btf_type_name[BTF_MAX_NAME_SIZE]; | |
9179 | int ret; | |
9180 | ||
9181 | ret = snprintf(btf_type_name, sizeof(btf_type_name), | |
9182 | "%s%s", prefix, name); | |
9183 | /* snprintf returns the number of characters written excluding the | |
c139e40a | 9184 | * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it |
a6ed02ca KS |
9185 | * indicates truncation. |
9186 | */ | |
9187 | if (ret < 0 || ret >= sizeof(btf_type_name)) | |
9188 | return -ENAMETOOLONG; | |
9189 | return btf__find_by_name_kind(btf, btf_type_name, kind); | |
9190 | } | |
9191 | ||
91abb4a6 AN |
9192 | static inline int find_attach_btf_id(struct btf *btf, const char *name, |
9193 | enum bpf_attach_type attach_type) | |
a6ed02ca | 9194 | { |
67234743 AS |
9195 | const char *prefix; |
9196 | int kind; | |
a6ed02ca | 9197 | |
67234743 AS |
9198 | btf_get_kernel_prefix_kind(attach_type, &prefix, &kind); |
9199 | return find_btf_by_prefix_kind(btf, prefix, name, kind); | |
a6ed02ca KS |
9200 | } |
9201 | ||
b8c54ea4 AS |
9202 | int libbpf_find_vmlinux_btf_id(const char *name, |
9203 | enum bpf_attach_type attach_type) | |
12a8654b | 9204 | { |
a6ed02ca | 9205 | struct btf *btf; |
3521ffa2 | 9206 | int err; |
12a8654b | 9207 | |
a710eed3 | 9208 | btf = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
9209 | err = libbpf_get_error(btf); |
9210 | if (err) { | |
12a8654b | 9211 | pr_warn("vmlinux BTF is not found\n"); |
e9fc3ce9 | 9212 | return libbpf_err(err); |
12a8654b AS |
9213 | } |
9214 | ||
91abb4a6 AN |
9215 | err = find_attach_btf_id(btf, name, attach_type); |
9216 | if (err <= 0) | |
9217 | pr_warn("%s is not found in vmlinux BTF\n", name); | |
9218 | ||
3521ffa2 | 9219 | btf__free(btf); |
e9fc3ce9 | 9220 | return libbpf_err(err); |
b8c54ea4 AS |
9221 | } |
9222 | ||
e7bf94db AS |
9223 | static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd) |
9224 | { | |
ebc7b50a DM |
9225 | struct bpf_prog_info info = {}; |
9226 | __u32 info_len = sizeof(info); | |
6cc93e2f | 9227 | struct btf *btf; |
6d2d73cd | 9228 | int err; |
e7bf94db | 9229 | |
ebc7b50a | 9230 | err = bpf_obj_get_info_by_fd(attach_prog_fd, &info, &info_len); |
e9fc3ce9 | 9231 | if (err) { |
ebc7b50a DM |
9232 | pr_warn("failed bpf_obj_get_info_by_fd for FD %d: %d\n", |
9233 | attach_prog_fd, err); | |
e9fc3ce9 | 9234 | return err; |
e7bf94db | 9235 | } |
6d2d73cd QM |
9236 | |
9237 | err = -EINVAL; | |
ebc7b50a | 9238 | if (!info.btf_id) { |
e7bf94db AS |
9239 | pr_warn("The target program doesn't have BTF\n"); |
9240 | goto out; | |
9241 | } | |
ebc7b50a DM |
9242 | btf = btf__load_from_kernel_by_id(info.btf_id); |
9243 | err = libbpf_get_error(btf); | |
9244 | if (err) { | |
9245 | pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err); | |
e7bf94db AS |
9246 | goto out; |
9247 | } | |
9248 | err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC); | |
9249 | btf__free(btf); | |
9250 | if (err <= 0) { | |
9251 | pr_warn("%s is not found in prog's BTF\n", name); | |
9252 | goto out; | |
9253 | } | |
9254 | out: | |
e7bf94db AS |
9255 | return err; |
9256 | } | |
9257 | ||
91abb4a6 AN |
9258 | static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name, |
9259 | enum bpf_attach_type attach_type, | |
9260 | int *btf_obj_fd, int *btf_type_id) | |
9261 | { | |
9262 | int ret, i; | |
9263 | ||
9264 | ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type); | |
9265 | if (ret > 0) { | |
9266 | *btf_obj_fd = 0; /* vmlinux BTF */ | |
9267 | *btf_type_id = ret; | |
9268 | return 0; | |
9269 | } | |
9270 | if (ret != -ENOENT) | |
9271 | return ret; | |
9272 | ||
9273 | ret = load_module_btfs(obj); | |
9274 | if (ret) | |
9275 | return ret; | |
9276 | ||
9277 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
9278 | const struct module_btf *mod = &obj->btf_modules[i]; | |
9279 | ||
9280 | ret = find_attach_btf_id(mod->btf, attach_name, attach_type); | |
9281 | if (ret > 0) { | |
9282 | *btf_obj_fd = mod->fd; | |
9283 | *btf_type_id = ret; | |
9284 | return 0; | |
9285 | } | |
9286 | if (ret == -ENOENT) | |
9287 | continue; | |
9288 | ||
9289 | return ret; | |
9290 | } | |
9291 | ||
9292 | return -ESRCH; | |
9293 | } | |
9294 | ||
15ea31fa AN |
9295 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
9296 | int *btf_obj_fd, int *btf_type_id) | |
b8c54ea4 | 9297 | { |
a6ed02ca KS |
9298 | enum bpf_attach_type attach_type = prog->expected_attach_type; |
9299 | __u32 attach_prog_fd = prog->attach_prog_fd; | |
b6291a6f | 9300 | int err = 0; |
b8c54ea4 | 9301 | |
91abb4a6 AN |
9302 | /* BPF program's BTF ID */ |
9303 | if (attach_prog_fd) { | |
9304 | err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd); | |
9305 | if (err < 0) { | |
9306 | pr_warn("failed to find BPF program (FD %d) BTF ID for '%s': %d\n", | |
9307 | attach_prog_fd, attach_name, err); | |
9308 | return err; | |
9309 | } | |
9310 | *btf_obj_fd = 0; | |
9311 | *btf_type_id = err; | |
9312 | return 0; | |
9313 | } | |
9314 | ||
9315 | /* kernel/module BTF ID */ | |
67234743 AS |
9316 | if (prog->obj->gen_loader) { |
9317 | bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type); | |
9318 | *btf_obj_fd = 0; | |
9319 | *btf_type_id = 1; | |
9320 | } else { | |
9321 | err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id); | |
9322 | } | |
91abb4a6 AN |
9323 | if (err) { |
9324 | pr_warn("failed to find kernel BTF type ID of '%s': %d\n", attach_name, err); | |
b8c54ea4 | 9325 | return err; |
12a8654b | 9326 | } |
91abb4a6 | 9327 | return 0; |
12a8654b AS |
9328 | } |
9329 | ||
956b620f AI |
9330 | int libbpf_attach_type_by_name(const char *name, |
9331 | enum bpf_attach_type *attach_type) | |
9332 | { | |
c76e4c22 | 9333 | char *type_names; |
b6291a6f | 9334 | const struct bpf_sec_def *sec_def; |
956b620f AI |
9335 | |
9336 | if (!name) | |
e9fc3ce9 | 9337 | return libbpf_err(-EINVAL); |
956b620f | 9338 | |
b6291a6f AN |
9339 | sec_def = find_sec_def(name); |
9340 | if (!sec_def) { | |
9341 | pr_debug("failed to guess attach type based on ELF section name '%s'\n", name); | |
9342 | type_names = libbpf_get_type_names(true); | |
9343 | if (type_names != NULL) { | |
9344 | pr_debug("attachable section(type) names are:%s\n", type_names); | |
9345 | free(type_names); | |
9346 | } | |
9347 | ||
9348 | return libbpf_err(-EINVAL); | |
c76e4c22 TS |
9349 | } |
9350 | ||
4fa5bcfe | 9351 | if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load) |
15ea31fa AN |
9352 | return libbpf_err(-EINVAL); |
9353 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
b6291a6f AN |
9354 | return libbpf_err(-EINVAL); |
9355 | ||
9356 | *attach_type = sec_def->expected_attach_type; | |
9357 | return 0; | |
956b620f AI |
9358 | } |
9359 | ||
a324aae3 | 9360 | int bpf_map__fd(const struct bpf_map *map) |
9d759a9b | 9361 | { |
e9fc3ce9 | 9362 | return map ? map->fd : libbpf_err(-EINVAL); |
9d759a9b WN |
9363 | } |
9364 | ||
a324aae3 | 9365 | const struct bpf_map_def *bpf_map__def(const struct bpf_map *map) |
9d759a9b | 9366 | { |
e9fc3ce9 | 9367 | return map ? &map->def : libbpf_err_ptr(-EINVAL); |
9d759a9b WN |
9368 | } |
9369 | ||
aed65917 AN |
9370 | static bool map_uses_real_name(const struct bpf_map *map) |
9371 | { | |
9372 | /* Since libbpf started to support custom .data.* and .rodata.* maps, | |
9373 | * their user-visible name differs from kernel-visible name. Users see | |
9374 | * such map's corresponding ELF section name as a map name. | |
9375 | * This check distinguishes .data/.rodata from .data.* and .rodata.* | |
9376 | * maps to know which name has to be returned to the user. | |
9377 | */ | |
9378 | if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0) | |
9379 | return true; | |
9380 | if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0) | |
9381 | return true; | |
9382 | return false; | |
9383 | } | |
9384 | ||
a324aae3 | 9385 | const char *bpf_map__name(const struct bpf_map *map) |
561bbcca | 9386 | { |
aed65917 AN |
9387 | if (!map) |
9388 | return NULL; | |
9389 | ||
9390 | if (map_uses_real_name(map)) | |
9391 | return map->real_name; | |
9392 | ||
9393 | return map->name; | |
561bbcca WN |
9394 | } |
9395 | ||
1bdb6c9a AN |
9396 | enum bpf_map_type bpf_map__type(const struct bpf_map *map) |
9397 | { | |
9398 | return map->def.type; | |
9399 | } | |
9400 | ||
9401 | int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type) | |
9402 | { | |
9403 | if (map->fd >= 0) | |
e9fc3ce9 | 9404 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9405 | map->def.type = type; |
9406 | return 0; | |
9407 | } | |
9408 | ||
9409 | __u32 bpf_map__map_flags(const struct bpf_map *map) | |
9410 | { | |
9411 | return map->def.map_flags; | |
9412 | } | |
9413 | ||
9414 | int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags) | |
9415 | { | |
9416 | if (map->fd >= 0) | |
e9fc3ce9 | 9417 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9418 | map->def.map_flags = flags; |
9419 | return 0; | |
9420 | } | |
9421 | ||
47512102 JK |
9422 | __u64 bpf_map__map_extra(const struct bpf_map *map) |
9423 | { | |
9424 | return map->map_extra; | |
9425 | } | |
9426 | ||
9427 | int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra) | |
9428 | { | |
9429 | if (map->fd >= 0) | |
9430 | return libbpf_err(-EBUSY); | |
9431 | map->map_extra = map_extra; | |
9432 | return 0; | |
9433 | } | |
9434 | ||
1bdb6c9a AN |
9435 | __u32 bpf_map__numa_node(const struct bpf_map *map) |
9436 | { | |
9437 | return map->numa_node; | |
9438 | } | |
9439 | ||
9440 | int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node) | |
9441 | { | |
9442 | if (map->fd >= 0) | |
e9fc3ce9 | 9443 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9444 | map->numa_node = numa_node; |
9445 | return 0; | |
9446 | } | |
9447 | ||
9448 | __u32 bpf_map__key_size(const struct bpf_map *map) | |
9449 | { | |
9450 | return map->def.key_size; | |
9451 | } | |
9452 | ||
9453 | int bpf_map__set_key_size(struct bpf_map *map, __u32 size) | |
9454 | { | |
9455 | if (map->fd >= 0) | |
e9fc3ce9 | 9456 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9457 | map->def.key_size = size; |
9458 | return 0; | |
9459 | } | |
9460 | ||
9461 | __u32 bpf_map__value_size(const struct bpf_map *map) | |
9462 | { | |
9463 | return map->def.value_size; | |
9464 | } | |
9465 | ||
9466 | int bpf_map__set_value_size(struct bpf_map *map, __u32 size) | |
9467 | { | |
9468 | if (map->fd >= 0) | |
e9fc3ce9 | 9469 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9470 | map->def.value_size = size; |
9471 | return 0; | |
9472 | } | |
9473 | ||
5b891af7 | 9474 | __u32 bpf_map__btf_key_type_id(const struct bpf_map *map) |
8a138aed | 9475 | { |
61746dbe | 9476 | return map ? map->btf_key_type_id : 0; |
8a138aed MKL |
9477 | } |
9478 | ||
5b891af7 | 9479 | __u32 bpf_map__btf_value_type_id(const struct bpf_map *map) |
8a138aed | 9480 | { |
61746dbe | 9481 | return map ? map->btf_value_type_id : 0; |
8a138aed MKL |
9482 | } |
9483 | ||
edb13ed4 ACM |
9484 | int bpf_map__set_priv(struct bpf_map *map, void *priv, |
9485 | bpf_map_clear_priv_t clear_priv) | |
9d759a9b WN |
9486 | { |
9487 | if (!map) | |
e9fc3ce9 | 9488 | return libbpf_err(-EINVAL); |
9d759a9b WN |
9489 | |
9490 | if (map->priv) { | |
9491 | if (map->clear_priv) | |
9492 | map->clear_priv(map, map->priv); | |
9493 | } | |
9494 | ||
9495 | map->priv = priv; | |
9496 | map->clear_priv = clear_priv; | |
9497 | return 0; | |
9498 | } | |
9499 | ||
a324aae3 | 9500 | void *bpf_map__priv(const struct bpf_map *map) |
9d759a9b | 9501 | { |
e9fc3ce9 | 9502 | return map ? map->priv : libbpf_err_ptr(-EINVAL); |
9d759a9b WN |
9503 | } |
9504 | ||
e2842be5 THJ |
9505 | int bpf_map__set_initial_value(struct bpf_map *map, |
9506 | const void *data, size_t size) | |
9507 | { | |
9508 | if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG || | |
9509 | size != map->def.value_size || map->fd >= 0) | |
e9fc3ce9 | 9510 | return libbpf_err(-EINVAL); |
e2842be5 THJ |
9511 | |
9512 | memcpy(map->mmaped, data, size); | |
9513 | return 0; | |
9514 | } | |
9515 | ||
7723256b AS |
9516 | const void *bpf_map__initial_value(struct bpf_map *map, size_t *psize) |
9517 | { | |
9518 | if (!map->mmaped) | |
9519 | return NULL; | |
9520 | *psize = map->def.value_size; | |
9521 | return map->mmaped; | |
9522 | } | |
9523 | ||
a324aae3 | 9524 | bool bpf_map__is_offload_neutral(const struct bpf_map *map) |
f83fb22c JK |
9525 | { |
9526 | return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; | |
9527 | } | |
9528 | ||
a324aae3 | 9529 | bool bpf_map__is_internal(const struct bpf_map *map) |
d859900c DB |
9530 | { |
9531 | return map->libbpf_type != LIBBPF_MAP_UNSPEC; | |
9532 | } | |
9533 | ||
1bdb6c9a AN |
9534 | __u32 bpf_map__ifindex(const struct bpf_map *map) |
9535 | { | |
9536 | return map->map_ifindex; | |
9537 | } | |
9538 | ||
9539 | int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex) | |
9aba3613 | 9540 | { |
1bdb6c9a | 9541 | if (map->fd >= 0) |
e9fc3ce9 | 9542 | return libbpf_err(-EBUSY); |
9aba3613 | 9543 | map->map_ifindex = ifindex; |
1bdb6c9a | 9544 | return 0; |
9aba3613 JK |
9545 | } |
9546 | ||
addb9fc9 NS |
9547 | int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd) |
9548 | { | |
9549 | if (!bpf_map_type__is_map_in_map(map->def.type)) { | |
be18010e | 9550 | pr_warn("error: unsupported map type\n"); |
e9fc3ce9 | 9551 | return libbpf_err(-EINVAL); |
addb9fc9 NS |
9552 | } |
9553 | if (map->inner_map_fd != -1) { | |
be18010e | 9554 | pr_warn("error: inner_map_fd already specified\n"); |
e9fc3ce9 | 9555 | return libbpf_err(-EINVAL); |
addb9fc9 | 9556 | } |
8f7b239e AN |
9557 | if (map->inner_map) { |
9558 | bpf_map__destroy(map->inner_map); | |
9559 | zfree(&map->inner_map); | |
9560 | } | |
addb9fc9 NS |
9561 | map->inner_map_fd = fd; |
9562 | return 0; | |
9563 | } | |
9564 | ||
0c19a9fb | 9565 | static struct bpf_map * |
a324aae3 | 9566 | __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i) |
9d759a9b | 9567 | { |
0c19a9fb | 9568 | ssize_t idx; |
9d759a9b WN |
9569 | struct bpf_map *s, *e; |
9570 | ||
9571 | if (!obj || !obj->maps) | |
e9fc3ce9 | 9572 | return errno = EINVAL, NULL; |
9d759a9b WN |
9573 | |
9574 | s = obj->maps; | |
9575 | e = obj->maps + obj->nr_maps; | |
9576 | ||
0c19a9fb | 9577 | if ((m < s) || (m >= e)) { |
be18010e KW |
9578 | pr_warn("error in %s: map handler doesn't belong to object\n", |
9579 | __func__); | |
e9fc3ce9 | 9580 | return errno = EINVAL, NULL; |
9d759a9b WN |
9581 | } |
9582 | ||
0c19a9fb SF |
9583 | idx = (m - obj->maps) + i; |
9584 | if (idx >= obj->nr_maps || idx < 0) | |
9d759a9b WN |
9585 | return NULL; |
9586 | return &obj->maps[idx]; | |
9587 | } | |
561bbcca | 9588 | |
0c19a9fb | 9589 | struct bpf_map * |
a324aae3 | 9590 | bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj) |
2088a3a7 HC |
9591 | { |
9592 | return bpf_object__next_map(obj, prev); | |
9593 | } | |
9594 | ||
9595 | struct bpf_map * | |
9596 | bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev) | |
0c19a9fb SF |
9597 | { |
9598 | if (prev == NULL) | |
9599 | return obj->maps; | |
9600 | ||
9601 | return __bpf_map__iter(prev, obj, 1); | |
9602 | } | |
9603 | ||
9604 | struct bpf_map * | |
a324aae3 | 9605 | bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj) |
2088a3a7 HC |
9606 | { |
9607 | return bpf_object__prev_map(obj, next); | |
9608 | } | |
9609 | ||
9610 | struct bpf_map * | |
9611 | bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next) | |
0c19a9fb SF |
9612 | { |
9613 | if (next == NULL) { | |
9614 | if (!obj->nr_maps) | |
9615 | return NULL; | |
9616 | return obj->maps + obj->nr_maps - 1; | |
9617 | } | |
9618 | ||
9619 | return __bpf_map__iter(next, obj, -1); | |
9620 | } | |
9621 | ||
561bbcca | 9622 | struct bpf_map * |
a324aae3 | 9623 | bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name) |
561bbcca WN |
9624 | { |
9625 | struct bpf_map *pos; | |
9626 | ||
f74a53d9 | 9627 | bpf_object__for_each_map(pos, obj) { |
26071635 AN |
9628 | /* if it's a special internal map name (which always starts |
9629 | * with dot) then check if that special name matches the | |
9630 | * real map name (ELF section name) | |
9631 | */ | |
9632 | if (name[0] == '.') { | |
9633 | if (pos->real_name && strcmp(pos->real_name, name) == 0) | |
9634 | return pos; | |
9635 | continue; | |
9636 | } | |
9637 | /* otherwise map name has to be an exact match */ | |
aed65917 AN |
9638 | if (map_uses_real_name(pos)) { |
9639 | if (strcmp(pos->real_name, name) == 0) | |
9640 | return pos; | |
9641 | continue; | |
9642 | } | |
9643 | if (strcmp(pos->name, name) == 0) | |
561bbcca WN |
9644 | return pos; |
9645 | } | |
e9fc3ce9 | 9646 | return errno = ENOENT, NULL; |
561bbcca | 9647 | } |
5a6acad1 | 9648 | |
f3cea32d | 9649 | int |
a324aae3 | 9650 | bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name) |
f3cea32d MF |
9651 | { |
9652 | return bpf_map__fd(bpf_object__find_map_by_name(obj, name)); | |
9653 | } | |
9654 | ||
5a6acad1 WN |
9655 | struct bpf_map * |
9656 | bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset) | |
9657 | { | |
e9fc3ce9 | 9658 | return libbpf_err_ptr(-ENOTSUP); |
5a6acad1 | 9659 | } |
e28ff1a8 JS |
9660 | |
9661 | long libbpf_get_error(const void *ptr) | |
9662 | { | |
e9fc3ce9 AN |
9663 | if (!IS_ERR_OR_NULL(ptr)) |
9664 | return 0; | |
9665 | ||
9666 | if (IS_ERR(ptr)) | |
9667 | errno = -PTR_ERR(ptr); | |
9668 | ||
9669 | /* If ptr == NULL, then errno should be already set by the failing | |
9670 | * API, because libbpf never returns NULL on success and it now always | |
9671 | * sets errno on error. So no extra errno handling for ptr == NULL | |
9672 | * case. | |
9673 | */ | |
9674 | return -errno; | |
e28ff1a8 | 9675 | } |
6f6d33f3 | 9676 | |
e32660ac | 9677 | __attribute__((alias("bpf_prog_load_xattr2"))) |
d7be143b | 9678 | int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr, |
e32660ac AN |
9679 | struct bpf_object **pobj, int *prog_fd); |
9680 | ||
9681 | static int bpf_prog_load_xattr2(const struct bpf_prog_load_attr *attr, | |
9682 | struct bpf_object **pobj, int *prog_fd) | |
6f6d33f3 | 9683 | { |
33bae185 | 9684 | struct bpf_object_open_attr open_attr = {}; |
48cca7e4 | 9685 | struct bpf_program *prog, *first_prog = NULL; |
6f6d33f3 | 9686 | struct bpf_object *obj; |
f0307a7e | 9687 | struct bpf_map *map; |
6f6d33f3 JF |
9688 | int err; |
9689 | ||
d7be143b | 9690 | if (!attr) |
e9fc3ce9 | 9691 | return libbpf_err(-EINVAL); |
17387dd5 | 9692 | if (!attr->file) |
e9fc3ce9 | 9693 | return libbpf_err(-EINVAL); |
d7be143b | 9694 | |
33bae185 LY |
9695 | open_attr.file = attr->file; |
9696 | open_attr.prog_type = attr->prog_type; | |
9697 | ||
fc763870 | 9698 | obj = __bpf_object__open_xattr(&open_attr, 0); |
e9fc3ce9 AN |
9699 | err = libbpf_get_error(obj); |
9700 | if (err) | |
9701 | return libbpf_err(-ENOENT); | |
6f6d33f3 | 9702 | |
48cca7e4 | 9703 | bpf_object__for_each_program(prog, obj) { |
dd4436bb | 9704 | enum bpf_attach_type attach_type = attr->expected_attach_type; |
48cca7e4 | 9705 | /* |
dd4436bb AN |
9706 | * to preserve backwards compatibility, bpf_prog_load treats |
9707 | * attr->prog_type, if specified, as an override to whatever | |
9708 | * bpf_object__open guessed | |
48cca7e4 | 9709 | */ |
dd4436bb | 9710 | if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) { |
df286716 GS |
9711 | prog->type = attr->prog_type; |
9712 | prog->expected_attach_type = attach_type; | |
dd4436bb | 9713 | } |
20eccf29 | 9714 | if (bpf_program__type(prog) == BPF_PROG_TYPE_UNSPEC) { |
dd4436bb AN |
9715 | /* |
9716 | * we haven't guessed from section name and user | |
9717 | * didn't provide a fallback type, too bad... | |
9718 | */ | |
9719 | bpf_object__close(obj); | |
e9fc3ce9 | 9720 | return libbpf_err(-EINVAL); |
583c9009 | 9721 | } |
48cca7e4 | 9722 | |
dd4436bb | 9723 | prog->prog_ifindex = attr->ifindex; |
da11b417 | 9724 | prog->log_level = attr->log_level; |
2b288740 | 9725 | prog->prog_flags |= attr->prog_flags; |
69495d2a | 9726 | if (!first_prog) |
48cca7e4 AS |
9727 | first_prog = prog; |
9728 | } | |
9729 | ||
f74a53d9 | 9730 | bpf_object__for_each_map(map, obj) { |
a5dd9589 | 9731 | if (map->def.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
f83fb22c | 9732 | map->map_ifindex = attr->ifindex; |
f0307a7e DB |
9733 | } |
9734 | ||
48cca7e4 | 9735 | if (!first_prog) { |
be18010e | 9736 | pr_warn("object file doesn't contain bpf program\n"); |
48cca7e4 | 9737 | bpf_object__close(obj); |
e9fc3ce9 | 9738 | return libbpf_err(-ENOENT); |
583c9009 RG |
9739 | } |
9740 | ||
6f6d33f3 JF |
9741 | err = bpf_object__load(obj); |
9742 | if (err) { | |
9743 | bpf_object__close(obj); | |
e9fc3ce9 | 9744 | return libbpf_err(err); |
6f6d33f3 JF |
9745 | } |
9746 | ||
9747 | *pobj = obj; | |
48cca7e4 | 9748 | *prog_fd = bpf_program__fd(first_prog); |
6f6d33f3 JF |
9749 | return 0; |
9750 | } | |
d0cabbb0 | 9751 | |
e32660ac AN |
9752 | COMPAT_VERSION(bpf_prog_load_deprecated, bpf_prog_load, LIBBPF_0.0.1) |
9753 | int bpf_prog_load_deprecated(const char *file, enum bpf_prog_type type, | |
9754 | struct bpf_object **pobj, int *prog_fd) | |
9755 | { | |
9756 | struct bpf_prog_load_attr attr; | |
9757 | ||
9758 | memset(&attr, 0, sizeof(struct bpf_prog_load_attr)); | |
9759 | attr.file = file; | |
9760 | attr.prog_type = type; | |
9761 | attr.expected_attach_type = 0; | |
9762 | ||
9763 | return bpf_prog_load_xattr2(&attr, pobj, prog_fd); | |
9764 | } | |
9765 | ||
cc4f864b AN |
9766 | /* Replace link's underlying BPF program with the new one */ |
9767 | int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog) | |
9768 | { | |
e9fc3ce9 | 9769 | int ret; |
c139e40a | 9770 | |
e9fc3ce9 AN |
9771 | ret = bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL); |
9772 | return libbpf_err_errno(ret); | |
cc4f864b AN |
9773 | } |
9774 | ||
d6958706 AN |
9775 | /* Release "ownership" of underlying BPF resource (typically, BPF program |
9776 | * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected | |
9777 | * link, when destructed through bpf_link__destroy() call won't attempt to | |
9778 | * detach/unregisted that BPF resource. This is useful in situations where, | |
9779 | * say, attached BPF program has to outlive userspace program that attached it | |
9780 | * in the system. Depending on type of BPF program, though, there might be | |
9781 | * additional steps (like pinning BPF program in BPF FS) necessary to ensure | |
9782 | * exit of userspace program doesn't trigger automatic detachment and clean up | |
9783 | * inside the kernel. | |
9784 | */ | |
9785 | void bpf_link__disconnect(struct bpf_link *link) | |
9786 | { | |
9787 | link->disconnected = true; | |
9788 | } | |
9789 | ||
1c2e9efc AN |
9790 | int bpf_link__destroy(struct bpf_link *link) |
9791 | { | |
d6958706 | 9792 | int err = 0; |
1c2e9efc | 9793 | |
50450fc7 | 9794 | if (IS_ERR_OR_NULL(link)) |
1c2e9efc AN |
9795 | return 0; |
9796 | ||
d6958706 AN |
9797 | if (!link->disconnected && link->detach) |
9798 | err = link->detach(link); | |
c016b68e AN |
9799 | if (link->pin_path) |
9800 | free(link->pin_path); | |
d88b71d4 AN |
9801 | if (link->dealloc) |
9802 | link->dealloc(link); | |
9803 | else | |
9804 | free(link); | |
1c2e9efc | 9805 | |
e9fc3ce9 | 9806 | return libbpf_err(err); |
1c2e9efc AN |
9807 | } |
9808 | ||
c016b68e AN |
9809 | int bpf_link__fd(const struct bpf_link *link) |
9810 | { | |
9811 | return link->fd; | |
9812 | } | |
9813 | ||
9814 | const char *bpf_link__pin_path(const struct bpf_link *link) | |
9815 | { | |
9816 | return link->pin_path; | |
9817 | } | |
9818 | ||
9819 | static int bpf_link__detach_fd(struct bpf_link *link) | |
9820 | { | |
e9fc3ce9 | 9821 | return libbpf_err_errno(close(link->fd)); |
c016b68e AN |
9822 | } |
9823 | ||
9824 | struct bpf_link *bpf_link__open(const char *path) | |
9825 | { | |
9826 | struct bpf_link *link; | |
9827 | int fd; | |
9828 | ||
9829 | fd = bpf_obj_get(path); | |
9830 | if (fd < 0) { | |
9831 | fd = -errno; | |
9832 | pr_warn("failed to open link at %s: %d\n", path, fd); | |
e9fc3ce9 | 9833 | return libbpf_err_ptr(fd); |
c016b68e AN |
9834 | } |
9835 | ||
9836 | link = calloc(1, sizeof(*link)); | |
9837 | if (!link) { | |
9838 | close(fd); | |
e9fc3ce9 | 9839 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
9840 | } |
9841 | link->detach = &bpf_link__detach_fd; | |
9842 | link->fd = fd; | |
9843 | ||
9844 | link->pin_path = strdup(path); | |
9845 | if (!link->pin_path) { | |
9846 | bpf_link__destroy(link); | |
e9fc3ce9 | 9847 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
9848 | } |
9849 | ||
9850 | return link; | |
9851 | } | |
9852 | ||
2e49527e AN |
9853 | int bpf_link__detach(struct bpf_link *link) |
9854 | { | |
9855 | return bpf_link_detach(link->fd) ? -errno : 0; | |
9856 | } | |
9857 | ||
c016b68e AN |
9858 | int bpf_link__pin(struct bpf_link *link, const char *path) |
9859 | { | |
9860 | int err; | |
9861 | ||
9862 | if (link->pin_path) | |
e9fc3ce9 | 9863 | return libbpf_err(-EBUSY); |
c016b68e AN |
9864 | err = make_parent_dir(path); |
9865 | if (err) | |
e9fc3ce9 | 9866 | return libbpf_err(err); |
c016b68e AN |
9867 | err = check_path(path); |
9868 | if (err) | |
e9fc3ce9 | 9869 | return libbpf_err(err); |
c016b68e AN |
9870 | |
9871 | link->pin_path = strdup(path); | |
9872 | if (!link->pin_path) | |
e9fc3ce9 | 9873 | return libbpf_err(-ENOMEM); |
c016b68e AN |
9874 | |
9875 | if (bpf_obj_pin(link->fd, link->pin_path)) { | |
9876 | err = -errno; | |
9877 | zfree(&link->pin_path); | |
e9fc3ce9 | 9878 | return libbpf_err(err); |
c016b68e AN |
9879 | } |
9880 | ||
9881 | pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path); | |
9882 | return 0; | |
9883 | } | |
9884 | ||
9885 | int bpf_link__unpin(struct bpf_link *link) | |
9886 | { | |
9887 | int err; | |
9888 | ||
9889 | if (!link->pin_path) | |
e9fc3ce9 | 9890 | return libbpf_err(-EINVAL); |
c016b68e AN |
9891 | |
9892 | err = unlink(link->pin_path); | |
9893 | if (err != 0) | |
af0efa05 | 9894 | return -errno; |
c016b68e AN |
9895 | |
9896 | pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path); | |
9897 | zfree(&link->pin_path); | |
9898 | return 0; | |
9899 | } | |
63f2f5ee | 9900 | |
668ace0e AN |
9901 | struct bpf_link_perf { |
9902 | struct bpf_link link; | |
9903 | int perf_event_fd; | |
ca304b40 RDT |
9904 | /* legacy kprobe support: keep track of probe identifier and type */ |
9905 | char *legacy_probe_name; | |
46ed5fc3 | 9906 | bool legacy_is_kprobe; |
ca304b40 | 9907 | bool legacy_is_retprobe; |
668ace0e AN |
9908 | }; |
9909 | ||
46ed5fc3 | 9910 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe); |
cc10623c | 9911 | static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe); |
46ed5fc3 | 9912 | |
668ace0e | 9913 | static int bpf_link_perf_detach(struct bpf_link *link) |
63f2f5ee | 9914 | { |
668ace0e AN |
9915 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); |
9916 | int err = 0; | |
63f2f5ee | 9917 | |
668ace0e | 9918 | if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0) |
63f2f5ee AN |
9919 | err = -errno; |
9920 | ||
668ace0e AN |
9921 | if (perf_link->perf_event_fd != link->fd) |
9922 | close(perf_link->perf_event_fd); | |
c016b68e | 9923 | close(link->fd); |
668ace0e | 9924 | |
cc10623c | 9925 | /* legacy uprobe/kprobe needs to be removed after perf event fd closure */ |
46ed5fc3 AN |
9926 | if (perf_link->legacy_probe_name) { |
9927 | if (perf_link->legacy_is_kprobe) { | |
9928 | err = remove_kprobe_event_legacy(perf_link->legacy_probe_name, | |
9929 | perf_link->legacy_is_retprobe); | |
cc10623c AN |
9930 | } else { |
9931 | err = remove_uprobe_event_legacy(perf_link->legacy_probe_name, | |
9932 | perf_link->legacy_is_retprobe); | |
46ed5fc3 AN |
9933 | } |
9934 | } | |
ca304b40 RDT |
9935 | |
9936 | return err; | |
63f2f5ee AN |
9937 | } |
9938 | ||
668ace0e AN |
9939 | static void bpf_link_perf_dealloc(struct bpf_link *link) |
9940 | { | |
9941 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
9942 | ||
ca304b40 | 9943 | free(perf_link->legacy_probe_name); |
668ace0e AN |
9944 | free(perf_link); |
9945 | } | |
9946 | ||
942025c9 | 9947 | struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd, |
47faff37 | 9948 | const struct bpf_perf_event_opts *opts) |
63f2f5ee AN |
9949 | { |
9950 | char errmsg[STRERR_BUFSIZE]; | |
668ace0e AN |
9951 | struct bpf_link_perf *link; |
9952 | int prog_fd, link_fd = -1, err; | |
63f2f5ee | 9953 | |
47faff37 AN |
9954 | if (!OPTS_VALID(opts, bpf_perf_event_opts)) |
9955 | return libbpf_err_ptr(-EINVAL); | |
9956 | ||
63f2f5ee | 9957 | if (pfd < 0) { |
52109584 AN |
9958 | pr_warn("prog '%s': invalid perf event FD %d\n", |
9959 | prog->name, pfd); | |
e9fc3ce9 | 9960 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
9961 | } |
9962 | prog_fd = bpf_program__fd(prog); | |
9963 | if (prog_fd < 0) { | |
52109584 AN |
9964 | pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n", |
9965 | prog->name); | |
e9fc3ce9 | 9966 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
9967 | } |
9968 | ||
d6958706 | 9969 | link = calloc(1, sizeof(*link)); |
63f2f5ee | 9970 | if (!link) |
e9fc3ce9 | 9971 | return libbpf_err_ptr(-ENOMEM); |
668ace0e AN |
9972 | link->link.detach = &bpf_link_perf_detach; |
9973 | link->link.dealloc = &bpf_link_perf_dealloc; | |
9974 | link->perf_event_fd = pfd; | |
63f2f5ee | 9975 | |
668ace0e | 9976 | if (kernel_supports(prog->obj, FEAT_PERF_LINK)) { |
47faff37 AN |
9977 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts, |
9978 | .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0)); | |
9979 | ||
9980 | link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts); | |
668ace0e AN |
9981 | if (link_fd < 0) { |
9982 | err = -errno; | |
9983 | pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n", | |
9984 | prog->name, pfd, | |
9985 | err, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
9986 | goto err_out; | |
9987 | } | |
9988 | link->link.fd = link_fd; | |
9989 | } else { | |
47faff37 AN |
9990 | if (OPTS_GET(opts, bpf_cookie, 0)) { |
9991 | pr_warn("prog '%s': user context value is not supported\n", prog->name); | |
9992 | err = -EOPNOTSUPP; | |
9993 | goto err_out; | |
9994 | } | |
9995 | ||
668ace0e AN |
9996 | if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) { |
9997 | err = -errno; | |
9998 | pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n", | |
9999 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10000 | if (err == -EPROTO) | |
10001 | pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n", | |
10002 | prog->name, pfd); | |
10003 | goto err_out; | |
10004 | } | |
10005 | link->link.fd = pfd; | |
63f2f5ee AN |
10006 | } |
10007 | if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
10008 | err = -errno; | |
668ace0e | 10009 | pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n", |
52109584 | 10010 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
668ace0e | 10011 | goto err_out; |
63f2f5ee | 10012 | } |
668ace0e AN |
10013 | |
10014 | return &link->link; | |
10015 | err_out: | |
10016 | if (link_fd >= 0) | |
10017 | close(link_fd); | |
10018 | free(link); | |
10019 | return libbpf_err_ptr(err); | |
63f2f5ee AN |
10020 | } |
10021 | ||
942025c9 | 10022 | struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd) |
47faff37 AN |
10023 | { |
10024 | return bpf_program__attach_perf_event_opts(prog, pfd, NULL); | |
10025 | } | |
10026 | ||
b2650027 AN |
10027 | /* |
10028 | * this function is expected to parse integer in the range of [0, 2^31-1] from | |
10029 | * given file using scanf format string fmt. If actual parsed value is | |
10030 | * negative, the result might be indistinguishable from error | |
10031 | */ | |
10032 | static int parse_uint_from_file(const char *file, const char *fmt) | |
10033 | { | |
10034 | char buf[STRERR_BUFSIZE]; | |
10035 | int err, ret; | |
10036 | FILE *f; | |
10037 | ||
10038 | f = fopen(file, "r"); | |
10039 | if (!f) { | |
10040 | err = -errno; | |
10041 | pr_debug("failed to open '%s': %s\n", file, | |
10042 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
10043 | return err; | |
10044 | } | |
10045 | err = fscanf(f, fmt, &ret); | |
10046 | if (err != 1) { | |
10047 | err = err == EOF ? -EIO : -errno; | |
10048 | pr_debug("failed to parse '%s': %s\n", file, | |
10049 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
10050 | fclose(f); | |
10051 | return err; | |
10052 | } | |
10053 | fclose(f); | |
10054 | return ret; | |
10055 | } | |
10056 | ||
10057 | static int determine_kprobe_perf_type(void) | |
10058 | { | |
10059 | const char *file = "/sys/bus/event_source/devices/kprobe/type"; | |
10060 | ||
10061 | return parse_uint_from_file(file, "%d\n"); | |
10062 | } | |
10063 | ||
10064 | static int determine_uprobe_perf_type(void) | |
10065 | { | |
10066 | const char *file = "/sys/bus/event_source/devices/uprobe/type"; | |
10067 | ||
10068 | return parse_uint_from_file(file, "%d\n"); | |
10069 | } | |
10070 | ||
10071 | static int determine_kprobe_retprobe_bit(void) | |
10072 | { | |
10073 | const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe"; | |
10074 | ||
10075 | return parse_uint_from_file(file, "config:%d\n"); | |
10076 | } | |
10077 | ||
10078 | static int determine_uprobe_retprobe_bit(void) | |
10079 | { | |
10080 | const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe"; | |
10081 | ||
10082 | return parse_uint_from_file(file, "config:%d\n"); | |
10083 | } | |
10084 | ||
5e3b8356 AN |
10085 | #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32 |
10086 | #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32 | |
10087 | ||
b2650027 | 10088 | static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name, |
5e3b8356 | 10089 | uint64_t offset, int pid, size_t ref_ctr_off) |
b2650027 AN |
10090 | { |
10091 | struct perf_event_attr attr = {}; | |
10092 | char errmsg[STRERR_BUFSIZE]; | |
10093 | int type, pfd, err; | |
10094 | ||
5e3b8356 AN |
10095 | if (ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS)) |
10096 | return -EINVAL; | |
10097 | ||
b2650027 AN |
10098 | type = uprobe ? determine_uprobe_perf_type() |
10099 | : determine_kprobe_perf_type(); | |
10100 | if (type < 0) { | |
be18010e KW |
10101 | pr_warn("failed to determine %s perf type: %s\n", |
10102 | uprobe ? "uprobe" : "kprobe", | |
10103 | libbpf_strerror_r(type, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10104 | return type; |
10105 | } | |
10106 | if (retprobe) { | |
10107 | int bit = uprobe ? determine_uprobe_retprobe_bit() | |
10108 | : determine_kprobe_retprobe_bit(); | |
10109 | ||
10110 | if (bit < 0) { | |
be18010e KW |
10111 | pr_warn("failed to determine %s retprobe bit: %s\n", |
10112 | uprobe ? "uprobe" : "kprobe", | |
10113 | libbpf_strerror_r(bit, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10114 | return bit; |
10115 | } | |
10116 | attr.config |= 1 << bit; | |
10117 | } | |
10118 | attr.size = sizeof(attr); | |
10119 | attr.type = type; | |
5e3b8356 | 10120 | attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT; |
36db2a94 AN |
10121 | attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */ |
10122 | attr.config2 = offset; /* kprobe_addr or probe_offset */ | |
b2650027 AN |
10123 | |
10124 | /* pid filter is meaningful only for uprobes */ | |
10125 | pfd = syscall(__NR_perf_event_open, &attr, | |
10126 | pid < 0 ? -1 : pid /* pid */, | |
10127 | pid == -1 ? 0 : -1 /* cpu */, | |
10128 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
10129 | if (pfd < 0) { | |
10130 | err = -errno; | |
be18010e KW |
10131 | pr_warn("%s perf_event_open() failed: %s\n", |
10132 | uprobe ? "uprobe" : "kprobe", | |
10133 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10134 | return err; |
10135 | } | |
10136 | return pfd; | |
10137 | } | |
10138 | ||
46ed5fc3 AN |
10139 | static int append_to_file(const char *file, const char *fmt, ...) |
10140 | { | |
10141 | int fd, n, err = 0; | |
10142 | va_list ap; | |
10143 | ||
92274e24 | 10144 | fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0); |
46ed5fc3 AN |
10145 | if (fd < 0) |
10146 | return -errno; | |
10147 | ||
10148 | va_start(ap, fmt); | |
10149 | n = vdprintf(fd, fmt, ap); | |
10150 | va_end(ap); | |
10151 | ||
10152 | if (n < 0) | |
10153 | err = -errno; | |
10154 | ||
10155 | close(fd); | |
10156 | return err; | |
10157 | } | |
10158 | ||
10159 | static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz, | |
10160 | const char *kfunc_name, size_t offset) | |
10161 | { | |
51a33c60 QW |
10162 | static int index = 0; |
10163 | ||
10164 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset, | |
10165 | __sync_fetch_and_add(&index, 1)); | |
46ed5fc3 AN |
10166 | } |
10167 | ||
10168 | static int add_kprobe_event_legacy(const char *probe_name, bool retprobe, | |
10169 | const char *kfunc_name, size_t offset) | |
10170 | { | |
10171 | const char *file = "/sys/kernel/debug/tracing/kprobe_events"; | |
10172 | ||
10173 | return append_to_file(file, "%c:%s/%s %s+0x%zx", | |
10174 | retprobe ? 'r' : 'p', | |
10175 | retprobe ? "kretprobes" : "kprobes", | |
10176 | probe_name, kfunc_name, offset); | |
10177 | } | |
10178 | ||
10179 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe) | |
10180 | { | |
10181 | const char *file = "/sys/kernel/debug/tracing/kprobe_events"; | |
10182 | ||
10183 | return append_to_file(file, "-:%s/%s", retprobe ? "kretprobes" : "kprobes", probe_name); | |
10184 | } | |
10185 | ||
10186 | static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
10187 | { | |
10188 | char file[256]; | |
10189 | ||
10190 | snprintf(file, sizeof(file), | |
10191 | "/sys/kernel/debug/tracing/events/%s/%s/id", | |
10192 | retprobe ? "kretprobes" : "kprobes", probe_name); | |
10193 | ||
10194 | return parse_uint_from_file(file, "%d\n"); | |
10195 | } | |
10196 | ||
10197 | static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe, | |
10198 | const char *kfunc_name, size_t offset, int pid) | |
ca304b40 RDT |
10199 | { |
10200 | struct perf_event_attr attr = {}; | |
10201 | char errmsg[STRERR_BUFSIZE]; | |
10202 | int type, pfd, err; | |
10203 | ||
46ed5fc3 | 10204 | err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset); |
ca304b40 | 10205 | if (err < 0) { |
46ed5fc3 AN |
10206 | pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n", |
10207 | kfunc_name, offset, | |
ca304b40 RDT |
10208 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10209 | return err; | |
10210 | } | |
46ed5fc3 | 10211 | type = determine_kprobe_perf_type_legacy(probe_name, retprobe); |
ca304b40 | 10212 | if (type < 0) { |
46ed5fc3 AN |
10213 | pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n", |
10214 | kfunc_name, offset, | |
ca304b40 RDT |
10215 | libbpf_strerror_r(type, errmsg, sizeof(errmsg))); |
10216 | return type; | |
10217 | } | |
10218 | attr.size = sizeof(attr); | |
10219 | attr.config = type; | |
10220 | attr.type = PERF_TYPE_TRACEPOINT; | |
10221 | ||
10222 | pfd = syscall(__NR_perf_event_open, &attr, | |
10223 | pid < 0 ? -1 : pid, /* pid */ | |
10224 | pid == -1 ? 0 : -1, /* cpu */ | |
10225 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
10226 | if (pfd < 0) { | |
10227 | err = -errno; | |
10228 | pr_warn("legacy kprobe perf_event_open() failed: %s\n", | |
10229 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10230 | return err; | |
10231 | } | |
10232 | return pfd; | |
10233 | } | |
10234 | ||
da97553e | 10235 | struct bpf_link * |
942025c9 | 10236 | bpf_program__attach_kprobe_opts(const struct bpf_program *prog, |
ac0ed488 | 10237 | const char *func_name, |
47faff37 | 10238 | const struct bpf_kprobe_opts *opts) |
b2650027 | 10239 | { |
47faff37 | 10240 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
b2650027 | 10241 | char errmsg[STRERR_BUFSIZE]; |
ca304b40 | 10242 | char *legacy_probe = NULL; |
b2650027 | 10243 | struct bpf_link *link; |
46ed5fc3 | 10244 | size_t offset; |
ca304b40 | 10245 | bool retprobe, legacy; |
b2650027 AN |
10246 | int pfd, err; |
10247 | ||
da97553e JO |
10248 | if (!OPTS_VALID(opts, bpf_kprobe_opts)) |
10249 | return libbpf_err_ptr(-EINVAL); | |
10250 | ||
10251 | retprobe = OPTS_GET(opts, retprobe, false); | |
10252 | offset = OPTS_GET(opts, offset, 0); | |
47faff37 | 10253 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
da97553e | 10254 | |
ca304b40 RDT |
10255 | legacy = determine_kprobe_perf_type() < 0; |
10256 | if (!legacy) { | |
10257 | pfd = perf_event_open_probe(false /* uprobe */, retprobe, | |
10258 | func_name, offset, | |
10259 | -1 /* pid */, 0 /* ref_ctr_off */); | |
10260 | } else { | |
46ed5fc3 AN |
10261 | char probe_name[256]; |
10262 | ||
10263 | gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
10264 | func_name, offset); | |
10265 | ||
71cff670 | 10266 | legacy_probe = strdup(probe_name); |
ca304b40 RDT |
10267 | if (!legacy_probe) |
10268 | return libbpf_err_ptr(-ENOMEM); | |
10269 | ||
46ed5fc3 | 10270 | pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name, |
ca304b40 RDT |
10271 | offset, -1 /* pid */); |
10272 | } | |
b2650027 | 10273 | if (pfd < 0) { |
46ed5fc3 AN |
10274 | err = -errno; |
10275 | pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n", | |
10276 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
10277 | func_name, offset, | |
303a2572 AN |
10278 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10279 | goto err_out; | |
b2650027 | 10280 | } |
47faff37 | 10281 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
10282 | err = libbpf_get_error(link); |
10283 | if (err) { | |
b2650027 | 10284 | close(pfd); |
46ed5fc3 AN |
10285 | pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n", |
10286 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
10287 | func_name, offset, | |
be18010e | 10288 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
303a2572 | 10289 | goto err_out; |
b2650027 | 10290 | } |
ca304b40 RDT |
10291 | if (legacy) { |
10292 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
10293 | ||
10294 | perf_link->legacy_probe_name = legacy_probe; | |
46ed5fc3 | 10295 | perf_link->legacy_is_kprobe = true; |
ca304b40 RDT |
10296 | perf_link->legacy_is_retprobe = retprobe; |
10297 | } | |
10298 | ||
b2650027 | 10299 | return link; |
303a2572 AN |
10300 | err_out: |
10301 | free(legacy_probe); | |
10302 | return libbpf_err_ptr(err); | |
b2650027 AN |
10303 | } |
10304 | ||
942025c9 | 10305 | struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog, |
ac0ed488 JO |
10306 | bool retprobe, |
10307 | const char *func_name) | |
10308 | { | |
da97553e | 10309 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts, |
ac0ed488 | 10310 | .retprobe = retprobe, |
da97553e | 10311 | ); |
ac0ed488 JO |
10312 | |
10313 | return bpf_program__attach_kprobe_opts(prog, func_name, &opts); | |
10314 | } | |
10315 | ||
ddc6b049 JO |
10316 | /* Adapted from perf/util/string.c */ |
10317 | static bool glob_match(const char *str, const char *pat) | |
10318 | { | |
10319 | while (*str && *pat && *pat != '*') { | |
10320 | if (*pat == '?') { /* Matches any single character */ | |
10321 | str++; | |
10322 | pat++; | |
10323 | continue; | |
10324 | } | |
10325 | if (*str != *pat) | |
10326 | return false; | |
10327 | str++; | |
10328 | pat++; | |
10329 | } | |
10330 | /* Check wild card */ | |
10331 | if (*pat == '*') { | |
10332 | while (*pat == '*') | |
10333 | pat++; | |
10334 | if (!*pat) /* Tail wild card matches all */ | |
10335 | return true; | |
10336 | while (*str) | |
10337 | if (glob_match(str++, pat)) | |
10338 | return true; | |
10339 | } | |
10340 | return !*str && !*pat; | |
10341 | } | |
10342 | ||
10343 | struct kprobe_multi_resolve { | |
10344 | const char *pattern; | |
10345 | unsigned long *addrs; | |
10346 | size_t cap; | |
10347 | size_t cnt; | |
10348 | }; | |
10349 | ||
10350 | static int | |
10351 | resolve_kprobe_multi_cb(unsigned long long sym_addr, char sym_type, | |
10352 | const char *sym_name, void *ctx) | |
10353 | { | |
10354 | struct kprobe_multi_resolve *res = ctx; | |
10355 | int err; | |
10356 | ||
10357 | if (!glob_match(sym_name, res->pattern)) | |
10358 | return 0; | |
10359 | ||
10360 | err = libbpf_ensure_mem((void **) &res->addrs, &res->cap, sizeof(unsigned long), | |
10361 | res->cnt + 1); | |
10362 | if (err) | |
10363 | return err; | |
10364 | ||
10365 | res->addrs[res->cnt++] = (unsigned long) sym_addr; | |
10366 | return 0; | |
10367 | } | |
10368 | ||
10369 | struct bpf_link * | |
10370 | bpf_program__attach_kprobe_multi_opts(const struct bpf_program *prog, | |
10371 | const char *pattern, | |
10372 | const struct bpf_kprobe_multi_opts *opts) | |
10373 | { | |
10374 | LIBBPF_OPTS(bpf_link_create_opts, lopts); | |
10375 | struct kprobe_multi_resolve res = { | |
10376 | .pattern = pattern, | |
10377 | }; | |
10378 | struct bpf_link *link = NULL; | |
10379 | char errmsg[STRERR_BUFSIZE]; | |
10380 | const unsigned long *addrs; | |
10381 | int err, link_fd, prog_fd; | |
10382 | const __u64 *cookies; | |
10383 | const char **syms; | |
10384 | bool retprobe; | |
10385 | size_t cnt; | |
10386 | ||
10387 | if (!OPTS_VALID(opts, bpf_kprobe_multi_opts)) | |
10388 | return libbpf_err_ptr(-EINVAL); | |
10389 | ||
10390 | syms = OPTS_GET(opts, syms, false); | |
10391 | addrs = OPTS_GET(opts, addrs, false); | |
10392 | cnt = OPTS_GET(opts, cnt, false); | |
10393 | cookies = OPTS_GET(opts, cookies, false); | |
10394 | ||
10395 | if (!pattern && !addrs && !syms) | |
10396 | return libbpf_err_ptr(-EINVAL); | |
10397 | if (pattern && (addrs || syms || cookies || cnt)) | |
10398 | return libbpf_err_ptr(-EINVAL); | |
10399 | if (!pattern && !cnt) | |
10400 | return libbpf_err_ptr(-EINVAL); | |
10401 | if (addrs && syms) | |
10402 | return libbpf_err_ptr(-EINVAL); | |
10403 | ||
10404 | if (pattern) { | |
10405 | err = libbpf_kallsyms_parse(resolve_kprobe_multi_cb, &res); | |
10406 | if (err) | |
10407 | goto error; | |
10408 | if (!res.cnt) { | |
10409 | err = -ENOENT; | |
10410 | goto error; | |
10411 | } | |
10412 | addrs = res.addrs; | |
10413 | cnt = res.cnt; | |
10414 | } | |
10415 | ||
10416 | retprobe = OPTS_GET(opts, retprobe, false); | |
10417 | ||
10418 | lopts.kprobe_multi.syms = syms; | |
10419 | lopts.kprobe_multi.addrs = addrs; | |
10420 | lopts.kprobe_multi.cookies = cookies; | |
10421 | lopts.kprobe_multi.cnt = cnt; | |
10422 | lopts.kprobe_multi.flags = retprobe ? BPF_F_KPROBE_MULTI_RETURN : 0; | |
10423 | ||
10424 | link = calloc(1, sizeof(*link)); | |
10425 | if (!link) { | |
10426 | err = -ENOMEM; | |
10427 | goto error; | |
10428 | } | |
10429 | link->detach = &bpf_link__detach_fd; | |
10430 | ||
10431 | prog_fd = bpf_program__fd(prog); | |
10432 | link_fd = bpf_link_create(prog_fd, 0, BPF_TRACE_KPROBE_MULTI, &lopts); | |
10433 | if (link_fd < 0) { | |
10434 | err = -errno; | |
10435 | pr_warn("prog '%s': failed to attach: %s\n", | |
10436 | prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10437 | goto error; | |
10438 | } | |
10439 | link->fd = link_fd; | |
10440 | free(res.addrs); | |
10441 | return link; | |
10442 | ||
10443 | error: | |
10444 | free(link); | |
10445 | free(res.addrs); | |
10446 | return libbpf_err_ptr(err); | |
10447 | } | |
10448 | ||
4fa5bcfe | 10449 | static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 10450 | { |
da97553e | 10451 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts); |
a2488b5f | 10452 | unsigned long offset = 0; |
d7a18ea7 | 10453 | const char *func_name; |
a2488b5f | 10454 | char *func; |
4fa5bcfe | 10455 | int n; |
d7a18ea7 | 10456 | |
13d35a0c AN |
10457 | opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/"); |
10458 | if (opts.retprobe) | |
10459 | func_name = prog->sec_name + sizeof("kretprobe/") - 1; | |
10460 | else | |
10461 | func_name = prog->sec_name + sizeof("kprobe/") - 1; | |
d7a18ea7 | 10462 | |
e3f9bc35 | 10463 | n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset); |
a2488b5f | 10464 | if (n < 1) { |
a2488b5f | 10465 | pr_warn("kprobe name is invalid: %s\n", func_name); |
4fa5bcfe | 10466 | return -EINVAL; |
a2488b5f AM |
10467 | } |
10468 | if (opts.retprobe && offset != 0) { | |
1f71a468 | 10469 | free(func); |
a2488b5f | 10470 | pr_warn("kretprobes do not support offset specification\n"); |
4fa5bcfe | 10471 | return -EINVAL; |
a2488b5f | 10472 | } |
d7a18ea7 | 10473 | |
a2488b5f | 10474 | opts.offset = offset; |
4fa5bcfe | 10475 | *link = bpf_program__attach_kprobe_opts(prog, func, &opts); |
a2488b5f | 10476 | free(func); |
4fa5bcfe | 10477 | return libbpf_get_error(*link); |
d7a18ea7 AN |
10478 | } |
10479 | ||
ddc6b049 JO |
10480 | static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
10481 | { | |
10482 | LIBBPF_OPTS(bpf_kprobe_multi_opts, opts); | |
10483 | const char *spec; | |
10484 | char *pattern; | |
10485 | int n; | |
10486 | ||
10487 | opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe.multi/"); | |
10488 | if (opts.retprobe) | |
10489 | spec = prog->sec_name + sizeof("kretprobe.multi/") - 1; | |
10490 | else | |
10491 | spec = prog->sec_name + sizeof("kprobe.multi/") - 1; | |
10492 | ||
10493 | n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern); | |
10494 | if (n < 1) { | |
10495 | pr_warn("kprobe multi pattern is invalid: %s\n", pattern); | |
10496 | return -EINVAL; | |
10497 | } | |
10498 | ||
10499 | *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts); | |
10500 | free(pattern); | |
10501 | return libbpf_get_error(*link); | |
10502 | } | |
10503 | ||
cc10623c AN |
10504 | static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz, |
10505 | const char *binary_path, uint64_t offset) | |
10506 | { | |
10507 | int i; | |
10508 | ||
10509 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset); | |
10510 | ||
10511 | /* sanitize binary_path in the probe name */ | |
10512 | for (i = 0; buf[i]; i++) { | |
10513 | if (!isalnum(buf[i])) | |
10514 | buf[i] = '_'; | |
10515 | } | |
10516 | } | |
10517 | ||
10518 | static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe, | |
10519 | const char *binary_path, size_t offset) | |
10520 | { | |
10521 | const char *file = "/sys/kernel/debug/tracing/uprobe_events"; | |
10522 | ||
10523 | return append_to_file(file, "%c:%s/%s %s:0x%zx", | |
10524 | retprobe ? 'r' : 'p', | |
10525 | retprobe ? "uretprobes" : "uprobes", | |
10526 | probe_name, binary_path, offset); | |
10527 | } | |
10528 | ||
10529 | static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe) | |
10530 | { | |
10531 | const char *file = "/sys/kernel/debug/tracing/uprobe_events"; | |
10532 | ||
10533 | return append_to_file(file, "-:%s/%s", retprobe ? "uretprobes" : "uprobes", probe_name); | |
10534 | } | |
10535 | ||
10536 | static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
10537 | { | |
10538 | char file[512]; | |
10539 | ||
10540 | snprintf(file, sizeof(file), | |
10541 | "/sys/kernel/debug/tracing/events/%s/%s/id", | |
10542 | retprobe ? "uretprobes" : "uprobes", probe_name); | |
10543 | ||
10544 | return parse_uint_from_file(file, "%d\n"); | |
10545 | } | |
10546 | ||
10547 | static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe, | |
10548 | const char *binary_path, size_t offset, int pid) | |
10549 | { | |
10550 | struct perf_event_attr attr; | |
10551 | int type, pfd, err; | |
10552 | ||
10553 | err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset); | |
10554 | if (err < 0) { | |
10555 | pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n", | |
10556 | binary_path, (size_t)offset, err); | |
10557 | return err; | |
10558 | } | |
10559 | type = determine_uprobe_perf_type_legacy(probe_name, retprobe); | |
10560 | if (type < 0) { | |
10561 | pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n", | |
10562 | binary_path, offset, err); | |
10563 | return type; | |
10564 | } | |
10565 | ||
10566 | memset(&attr, 0, sizeof(attr)); | |
10567 | attr.size = sizeof(attr); | |
10568 | attr.config = type; | |
10569 | attr.type = PERF_TYPE_TRACEPOINT; | |
10570 | ||
10571 | pfd = syscall(__NR_perf_event_open, &attr, | |
10572 | pid < 0 ? -1 : pid, /* pid */ | |
10573 | pid == -1 ? 0 : -1, /* cpu */ | |
10574 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
10575 | if (pfd < 0) { | |
10576 | err = -errno; | |
10577 | pr_warn("legacy uprobe perf_event_open() failed: %d\n", err); | |
10578 | return err; | |
10579 | } | |
10580 | return pfd; | |
10581 | } | |
10582 | ||
433966e3 AM |
10583 | /* uprobes deal in relative offsets; subtract the base address associated with |
10584 | * the mapped binary. See Documentation/trace/uprobetracer.rst for more | |
10585 | * details. | |
10586 | */ | |
10587 | static long elf_find_relative_offset(const char *filename, Elf *elf, long addr) | |
10588 | { | |
10589 | size_t n; | |
10590 | int i; | |
10591 | ||
10592 | if (elf_getphdrnum(elf, &n)) { | |
10593 | pr_warn("elf: failed to find program headers for '%s': %s\n", filename, | |
10594 | elf_errmsg(-1)); | |
10595 | return -ENOENT; | |
10596 | } | |
10597 | ||
10598 | for (i = 0; i < n; i++) { | |
10599 | int seg_start, seg_end, seg_offset; | |
10600 | GElf_Phdr phdr; | |
10601 | ||
10602 | if (!gelf_getphdr(elf, i, &phdr)) { | |
10603 | pr_warn("elf: failed to get program header %d from '%s': %s\n", i, filename, | |
10604 | elf_errmsg(-1)); | |
10605 | return -ENOENT; | |
10606 | } | |
10607 | if (phdr.p_type != PT_LOAD || !(phdr.p_flags & PF_X)) | |
10608 | continue; | |
10609 | ||
10610 | seg_start = phdr.p_vaddr; | |
10611 | seg_end = seg_start + phdr.p_memsz; | |
10612 | seg_offset = phdr.p_offset; | |
10613 | if (addr >= seg_start && addr < seg_end) | |
10614 | return addr - seg_start + seg_offset; | |
10615 | } | |
10616 | pr_warn("elf: failed to find prog header containing 0x%lx in '%s'\n", addr, filename); | |
10617 | return -ENOENT; | |
10618 | } | |
10619 | ||
10620 | /* Return next ELF section of sh_type after scn, or first of that type if scn is NULL. */ | |
10621 | static Elf_Scn *elf_find_next_scn_by_type(Elf *elf, int sh_type, Elf_Scn *scn) | |
10622 | { | |
10623 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
10624 | GElf_Shdr sh; | |
10625 | ||
10626 | if (!gelf_getshdr(scn, &sh)) | |
10627 | continue; | |
10628 | if (sh.sh_type == sh_type) | |
10629 | return scn; | |
10630 | } | |
10631 | return NULL; | |
10632 | } | |
10633 | ||
10634 | /* Find offset of function name in object specified by path. "name" matches | |
10635 | * symbol name or name@@LIB for library functions. | |
10636 | */ | |
10637 | static long elf_find_func_offset(const char *binary_path, const char *name) | |
10638 | { | |
10639 | int fd, i, sh_types[2] = { SHT_DYNSYM, SHT_SYMTAB }; | |
10640 | bool is_shared_lib, is_name_qualified; | |
10641 | char errmsg[STRERR_BUFSIZE]; | |
10642 | long ret = -ENOENT; | |
10643 | size_t name_len; | |
10644 | GElf_Ehdr ehdr; | |
10645 | Elf *elf; | |
10646 | ||
10647 | fd = open(binary_path, O_RDONLY | O_CLOEXEC); | |
10648 | if (fd < 0) { | |
10649 | ret = -errno; | |
10650 | pr_warn("failed to open %s: %s\n", binary_path, | |
10651 | libbpf_strerror_r(ret, errmsg, sizeof(errmsg))); | |
10652 | return ret; | |
10653 | } | |
10654 | elf = elf_begin(fd, ELF_C_READ_MMAP, NULL); | |
10655 | if (!elf) { | |
10656 | pr_warn("elf: could not read elf from %s: %s\n", binary_path, elf_errmsg(-1)); | |
10657 | close(fd); | |
10658 | return -LIBBPF_ERRNO__FORMAT; | |
10659 | } | |
10660 | if (!gelf_getehdr(elf, &ehdr)) { | |
10661 | pr_warn("elf: failed to get ehdr from %s: %s\n", binary_path, elf_errmsg(-1)); | |
10662 | ret = -LIBBPF_ERRNO__FORMAT; | |
10663 | goto out; | |
10664 | } | |
10665 | /* for shared lib case, we do not need to calculate relative offset */ | |
10666 | is_shared_lib = ehdr.e_type == ET_DYN; | |
10667 | ||
10668 | name_len = strlen(name); | |
10669 | /* Does name specify "@@LIB"? */ | |
10670 | is_name_qualified = strstr(name, "@@") != NULL; | |
10671 | ||
10672 | /* Search SHT_DYNSYM, SHT_SYMTAB for symbol. This search order is used because if | |
10673 | * a binary is stripped, it may only have SHT_DYNSYM, and a fully-statically | |
10674 | * linked binary may not have SHT_DYMSYM, so absence of a section should not be | |
10675 | * reported as a warning/error. | |
10676 | */ | |
10677 | for (i = 0; i < ARRAY_SIZE(sh_types); i++) { | |
10678 | size_t nr_syms, strtabidx, idx; | |
10679 | Elf_Data *symbols = NULL; | |
10680 | Elf_Scn *scn = NULL; | |
10681 | int last_bind = -1; | |
10682 | const char *sname; | |
10683 | GElf_Shdr sh; | |
10684 | ||
10685 | scn = elf_find_next_scn_by_type(elf, sh_types[i], NULL); | |
10686 | if (!scn) { | |
10687 | pr_debug("elf: failed to find symbol table ELF sections in '%s'\n", | |
10688 | binary_path); | |
10689 | continue; | |
10690 | } | |
10691 | if (!gelf_getshdr(scn, &sh)) | |
10692 | continue; | |
10693 | strtabidx = sh.sh_link; | |
10694 | symbols = elf_getdata(scn, 0); | |
10695 | if (!symbols) { | |
10696 | pr_warn("elf: failed to get symbols for symtab section in '%s': %s\n", | |
10697 | binary_path, elf_errmsg(-1)); | |
10698 | ret = -LIBBPF_ERRNO__FORMAT; | |
10699 | goto out; | |
10700 | } | |
10701 | nr_syms = symbols->d_size / sh.sh_entsize; | |
10702 | ||
10703 | for (idx = 0; idx < nr_syms; idx++) { | |
10704 | int curr_bind; | |
10705 | GElf_Sym sym; | |
10706 | ||
10707 | if (!gelf_getsym(symbols, idx, &sym)) | |
10708 | continue; | |
10709 | ||
10710 | if (GELF_ST_TYPE(sym.st_info) != STT_FUNC) | |
10711 | continue; | |
10712 | ||
10713 | sname = elf_strptr(elf, strtabidx, sym.st_name); | |
10714 | if (!sname) | |
10715 | continue; | |
10716 | ||
10717 | curr_bind = GELF_ST_BIND(sym.st_info); | |
10718 | ||
10719 | /* User can specify func, func@@LIB or func@@LIB_VERSION. */ | |
10720 | if (strncmp(sname, name, name_len) != 0) | |
10721 | continue; | |
10722 | /* ...but we don't want a search for "foo" to match 'foo2" also, so any | |
10723 | * additional characters in sname should be of the form "@@LIB". | |
10724 | */ | |
10725 | if (!is_name_qualified && sname[name_len] != '\0' && sname[name_len] != '@') | |
10726 | continue; | |
10727 | ||
10728 | if (ret >= 0) { | |
10729 | /* handle multiple matches */ | |
10730 | if (last_bind != STB_WEAK && curr_bind != STB_WEAK) { | |
10731 | /* Only accept one non-weak bind. */ | |
10732 | pr_warn("elf: ambiguous match for '%s', '%s' in '%s'\n", | |
10733 | sname, name, binary_path); | |
10734 | ret = -LIBBPF_ERRNO__FORMAT; | |
10735 | goto out; | |
10736 | } else if (curr_bind == STB_WEAK) { | |
10737 | /* already have a non-weak bind, and | |
10738 | * this is a weak bind, so ignore. | |
10739 | */ | |
10740 | continue; | |
10741 | } | |
10742 | } | |
10743 | ret = sym.st_value; | |
10744 | last_bind = curr_bind; | |
10745 | } | |
10746 | /* For binaries that are not shared libraries, we need relative offset */ | |
10747 | if (ret > 0 && !is_shared_lib) | |
10748 | ret = elf_find_relative_offset(binary_path, elf, ret); | |
10749 | if (ret > 0) | |
10750 | break; | |
10751 | } | |
10752 | ||
10753 | if (ret > 0) { | |
10754 | pr_debug("elf: symbol address match for '%s' in '%s': 0x%lx\n", name, binary_path, | |
10755 | ret); | |
10756 | } else { | |
10757 | if (ret == 0) { | |
10758 | pr_warn("elf: '%s' is 0 in symtab for '%s': %s\n", name, binary_path, | |
10759 | is_shared_lib ? "should not be 0 in a shared library" : | |
10760 | "try using shared library path instead"); | |
10761 | ret = -ENOENT; | |
10762 | } else { | |
10763 | pr_warn("elf: failed to find symbol '%s' in '%s'\n", name, binary_path); | |
10764 | } | |
10765 | } | |
10766 | out: | |
10767 | elf_end(elf); | |
10768 | close(fd); | |
10769 | return ret; | |
10770 | } | |
10771 | ||
56818931 IL |
10772 | static const char *arch_specific_lib_paths(void) |
10773 | { | |
10774 | /* | |
10775 | * Based on https://packages.debian.org/sid/libc6. | |
10776 | * | |
10777 | * Assume that the traced program is built for the same architecture | |
10778 | * as libbpf, which should cover the vast majority of cases. | |
10779 | */ | |
10780 | #if defined(__x86_64__) | |
10781 | return "/lib/x86_64-linux-gnu"; | |
10782 | #elif defined(__i386__) | |
10783 | return "/lib/i386-linux-gnu"; | |
10784 | #elif defined(__s390x__) | |
10785 | return "/lib/s390x-linux-gnu"; | |
10786 | #elif defined(__s390__) | |
10787 | return "/lib/s390-linux-gnu"; | |
10788 | #elif defined(__arm__) && defined(__SOFTFP__) | |
10789 | return "/lib/arm-linux-gnueabi"; | |
10790 | #elif defined(__arm__) && !defined(__SOFTFP__) | |
10791 | return "/lib/arm-linux-gnueabihf"; | |
10792 | #elif defined(__aarch64__) | |
10793 | return "/lib/aarch64-linux-gnu"; | |
10794 | #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 64 | |
10795 | return "/lib/mips64el-linux-gnuabi64"; | |
10796 | #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 32 | |
10797 | return "/lib/mipsel-linux-gnu"; | |
10798 | #elif defined(__powerpc64__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ | |
10799 | return "/lib/powerpc64le-linux-gnu"; | |
10800 | #elif defined(__sparc__) && defined(__arch64__) | |
10801 | return "/lib/sparc64-linux-gnu"; | |
10802 | #elif defined(__riscv) && __riscv_xlen == 64 | |
10803 | return "/lib/riscv64-linux-gnu"; | |
10804 | #else | |
10805 | return NULL; | |
10806 | #endif | |
10807 | } | |
10808 | ||
1ce3a60e AM |
10809 | /* Get full path to program/shared library. */ |
10810 | static int resolve_full_path(const char *file, char *result, size_t result_sz) | |
10811 | { | |
56818931 | 10812 | const char *search_paths[3] = {}; |
1ce3a60e AM |
10813 | int i; |
10814 | ||
a1c9d61b | 10815 | if (str_has_sfx(file, ".so") || strstr(file, ".so.")) { |
1ce3a60e AM |
10816 | search_paths[0] = getenv("LD_LIBRARY_PATH"); |
10817 | search_paths[1] = "/usr/lib64:/usr/lib"; | |
56818931 | 10818 | search_paths[2] = arch_specific_lib_paths(); |
1ce3a60e AM |
10819 | } else { |
10820 | search_paths[0] = getenv("PATH"); | |
10821 | search_paths[1] = "/usr/bin:/usr/sbin"; | |
10822 | } | |
10823 | ||
10824 | for (i = 0; i < ARRAY_SIZE(search_paths); i++) { | |
10825 | const char *s; | |
10826 | ||
10827 | if (!search_paths[i]) | |
10828 | continue; | |
10829 | for (s = search_paths[i]; s != NULL; s = strchr(s, ':')) { | |
10830 | char *next_path; | |
10831 | int seg_len; | |
10832 | ||
10833 | if (s[0] == ':') | |
10834 | s++; | |
10835 | next_path = strchr(s, ':'); | |
10836 | seg_len = next_path ? next_path - s : strlen(s); | |
10837 | if (!seg_len) | |
10838 | continue; | |
10839 | snprintf(result, result_sz, "%.*s/%s", seg_len, s, file); | |
10840 | /* ensure it is an executable file/link */ | |
10841 | if (access(result, R_OK | X_OK) < 0) | |
10842 | continue; | |
10843 | pr_debug("resolved '%s' to '%s'\n", file, result); | |
10844 | return 0; | |
10845 | } | |
10846 | } | |
10847 | return -ENOENT; | |
10848 | } | |
10849 | ||
47faff37 | 10850 | LIBBPF_API struct bpf_link * |
942025c9 | 10851 | bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid, |
47faff37 AN |
10852 | const char *binary_path, size_t func_offset, |
10853 | const struct bpf_uprobe_opts *opts) | |
b2650027 | 10854 | { |
47faff37 | 10855 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
cc10623c | 10856 | char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL; |
1ce3a60e | 10857 | char full_binary_path[PATH_MAX]; |
b2650027 | 10858 | struct bpf_link *link; |
5e3b8356 | 10859 | size_t ref_ctr_off; |
b2650027 | 10860 | int pfd, err; |
cc10623c | 10861 | bool retprobe, legacy; |
433966e3 | 10862 | const char *func_name; |
47faff37 AN |
10863 | |
10864 | if (!OPTS_VALID(opts, bpf_uprobe_opts)) | |
10865 | return libbpf_err_ptr(-EINVAL); | |
10866 | ||
10867 | retprobe = OPTS_GET(opts, retprobe, false); | |
5e3b8356 | 10868 | ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0); |
47faff37 | 10869 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
b2650027 | 10870 | |
1ce3a60e AM |
10871 | if (binary_path && !strchr(binary_path, '/')) { |
10872 | err = resolve_full_path(binary_path, full_binary_path, | |
10873 | sizeof(full_binary_path)); | |
10874 | if (err) { | |
2e4913e0 AN |
10875 | pr_warn("prog '%s': failed to resolve full path for '%s': %d\n", |
10876 | prog->name, binary_path, err); | |
1ce3a60e AM |
10877 | return libbpf_err_ptr(err); |
10878 | } | |
10879 | binary_path = full_binary_path; | |
10880 | } | |
433966e3 AM |
10881 | func_name = OPTS_GET(opts, func_name, NULL); |
10882 | if (func_name) { | |
10883 | long sym_off; | |
10884 | ||
10885 | if (!binary_path) { | |
10886 | pr_warn("prog '%s': name-based attach requires binary_path\n", | |
10887 | prog->name); | |
10888 | return libbpf_err_ptr(-EINVAL); | |
10889 | } | |
10890 | sym_off = elf_find_func_offset(binary_path, func_name); | |
10891 | if (sym_off < 0) | |
10892 | return libbpf_err_ptr(sym_off); | |
10893 | func_offset += sym_off; | |
10894 | } | |
1ce3a60e | 10895 | |
cc10623c AN |
10896 | legacy = determine_uprobe_perf_type() < 0; |
10897 | if (!legacy) { | |
10898 | pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path, | |
10899 | func_offset, pid, ref_ctr_off); | |
10900 | } else { | |
1ce3a60e | 10901 | char probe_name[PATH_MAX + 64]; |
cc10623c AN |
10902 | |
10903 | if (ref_ctr_off) | |
10904 | return libbpf_err_ptr(-EINVAL); | |
10905 | ||
10906 | gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
10907 | binary_path, func_offset); | |
10908 | ||
10909 | legacy_probe = strdup(probe_name); | |
10910 | if (!legacy_probe) | |
10911 | return libbpf_err_ptr(-ENOMEM); | |
10912 | ||
10913 | pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe, | |
10914 | binary_path, func_offset, pid); | |
10915 | } | |
b2650027 | 10916 | if (pfd < 0) { |
cc10623c | 10917 | err = -errno; |
52109584 AN |
10918 | pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n", |
10919 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e | 10920 | binary_path, func_offset, |
cc10623c AN |
10921 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10922 | goto err_out; | |
b2650027 | 10923 | } |
cc10623c | 10924 | |
47faff37 | 10925 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
10926 | err = libbpf_get_error(link); |
10927 | if (err) { | |
b2650027 | 10928 | close(pfd); |
52109584 AN |
10929 | pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n", |
10930 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e KW |
10931 | binary_path, func_offset, |
10932 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
cc10623c AN |
10933 | goto err_out; |
10934 | } | |
10935 | if (legacy) { | |
10936 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
10937 | ||
10938 | perf_link->legacy_probe_name = legacy_probe; | |
10939 | perf_link->legacy_is_kprobe = false; | |
10940 | perf_link->legacy_is_retprobe = retprobe; | |
b2650027 AN |
10941 | } |
10942 | return link; | |
cc10623c AN |
10943 | err_out: |
10944 | free(legacy_probe); | |
10945 | return libbpf_err_ptr(err); | |
10946 | ||
b2650027 AN |
10947 | } |
10948 | ||
39f8dc43 AM |
10949 | /* Format of u[ret]probe section definition supporting auto-attach: |
10950 | * u[ret]probe/binary:function[+offset] | |
10951 | * | |
10952 | * binary can be an absolute/relative path or a filename; the latter is resolved to a | |
10953 | * full binary path via bpf_program__attach_uprobe_opts. | |
10954 | * | |
10955 | * Specifying uprobe+ ensures we carry out strict matching; either "uprobe" must be | |
10956 | * specified (and auto-attach is not possible) or the above format is specified for | |
10957 | * auto-attach. | |
10958 | */ | |
10959 | static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link) | |
10960 | { | |
10961 | DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts); | |
90db26e6 AM |
10962 | char *probe_type = NULL, *binary_path = NULL, *func_name = NULL; |
10963 | int n, ret = -EINVAL; | |
10964 | long offset = 0; | |
39f8dc43 AM |
10965 | |
10966 | *link = NULL; | |
10967 | ||
90db26e6 AM |
10968 | n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[a-zA-Z0-9_.]+%li", |
10969 | &probe_type, &binary_path, &func_name, &offset); | |
10970 | switch (n) { | |
10971 | case 1: | |
10972 | /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */ | |
10973 | ret = 0; | |
10974 | break; | |
10975 | case 2: | |
10976 | pr_warn("prog '%s': section '%s' missing ':function[+offset]' specification\n", | |
10977 | prog->name, prog->sec_name); | |
10978 | break; | |
10979 | case 3: | |
10980 | case 4: | |
10981 | opts.retprobe = strcmp(probe_type, "uretprobe") == 0; | |
10982 | if (opts.retprobe && offset != 0) { | |
10983 | pr_warn("prog '%s': uretprobes do not support offset specification\n", | |
10984 | prog->name); | |
10985 | break; | |
10986 | } | |
10987 | opts.func_name = func_name; | |
10988 | *link = bpf_program__attach_uprobe_opts(prog, -1, binary_path, offset, &opts); | |
10989 | ret = libbpf_get_error(*link); | |
10990 | break; | |
10991 | default: | |
10992 | pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name, | |
39f8dc43 | 10993 | prog->sec_name); |
90db26e6 | 10994 | break; |
39f8dc43 | 10995 | } |
90db26e6 AM |
10996 | free(probe_type); |
10997 | free(binary_path); | |
10998 | free(func_name); | |
39f8dc43 | 10999 | |
90db26e6 | 11000 | return ret; |
39f8dc43 AM |
11001 | } |
11002 | ||
942025c9 | 11003 | struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog, |
47faff37 AN |
11004 | bool retprobe, pid_t pid, |
11005 | const char *binary_path, | |
11006 | size_t func_offset) | |
11007 | { | |
11008 | DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe); | |
11009 | ||
11010 | return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts); | |
11011 | } | |
11012 | ||
2e4913e0 AN |
11013 | struct bpf_link *bpf_program__attach_usdt(const struct bpf_program *prog, |
11014 | pid_t pid, const char *binary_path, | |
11015 | const char *usdt_provider, const char *usdt_name, | |
11016 | const struct bpf_usdt_opts *opts) | |
11017 | { | |
11018 | char resolved_path[512]; | |
11019 | struct bpf_object *obj = prog->obj; | |
11020 | struct bpf_link *link; | |
5af25a41 | 11021 | __u64 usdt_cookie; |
2e4913e0 AN |
11022 | int err; |
11023 | ||
11024 | if (!OPTS_VALID(opts, bpf_uprobe_opts)) | |
11025 | return libbpf_err_ptr(-EINVAL); | |
11026 | ||
11027 | if (bpf_program__fd(prog) < 0) { | |
11028 | pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n", | |
11029 | prog->name); | |
11030 | return libbpf_err_ptr(-EINVAL); | |
11031 | } | |
11032 | ||
11033 | if (!strchr(binary_path, '/')) { | |
11034 | err = resolve_full_path(binary_path, resolved_path, sizeof(resolved_path)); | |
11035 | if (err) { | |
11036 | pr_warn("prog '%s': failed to resolve full path for '%s': %d\n", | |
11037 | prog->name, binary_path, err); | |
11038 | return libbpf_err_ptr(err); | |
11039 | } | |
11040 | binary_path = resolved_path; | |
11041 | } | |
11042 | ||
11043 | /* USDT manager is instantiated lazily on first USDT attach. It will | |
11044 | * be destroyed together with BPF object in bpf_object__close(). | |
11045 | */ | |
11046 | if (IS_ERR(obj->usdt_man)) | |
11047 | return libbpf_ptr(obj->usdt_man); | |
11048 | if (!obj->usdt_man) { | |
11049 | obj->usdt_man = usdt_manager_new(obj); | |
11050 | if (IS_ERR(obj->usdt_man)) | |
11051 | return libbpf_ptr(obj->usdt_man); | |
11052 | } | |
11053 | ||
11054 | usdt_cookie = OPTS_GET(opts, usdt_cookie, 0); | |
11055 | link = usdt_manager_attach_usdt(obj->usdt_man, prog, pid, binary_path, | |
11056 | usdt_provider, usdt_name, usdt_cookie); | |
11057 | err = libbpf_get_error(link); | |
11058 | if (err) | |
11059 | return libbpf_err_ptr(err); | |
11060 | return link; | |
11061 | } | |
11062 | ||
11063 | static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link) | |
11064 | { | |
11065 | char *path = NULL, *provider = NULL, *name = NULL; | |
11066 | const char *sec_name; | |
11067 | int n, err; | |
11068 | ||
11069 | sec_name = bpf_program__section_name(prog); | |
11070 | if (strcmp(sec_name, "usdt") == 0) { | |
11071 | /* no auto-attach for just SEC("usdt") */ | |
11072 | *link = NULL; | |
11073 | return 0; | |
11074 | } | |
11075 | ||
11076 | n = sscanf(sec_name, "usdt/%m[^:]:%m[^:]:%m[^:]", &path, &provider, &name); | |
11077 | if (n != 3) { | |
11078 | pr_warn("invalid section '%s', expected SEC(\"usdt/<path>:<provider>:<name>\")\n", | |
11079 | sec_name); | |
11080 | err = -EINVAL; | |
11081 | } else { | |
11082 | *link = bpf_program__attach_usdt(prog, -1 /* any process */, path, | |
11083 | provider, name, NULL); | |
11084 | err = libbpf_get_error(*link); | |
11085 | } | |
11086 | free(path); | |
11087 | free(provider); | |
11088 | free(name); | |
11089 | return err; | |
11090 | } | |
11091 | ||
f6de59c1 AN |
11092 | static int determine_tracepoint_id(const char *tp_category, |
11093 | const char *tp_name) | |
11094 | { | |
11095 | char file[PATH_MAX]; | |
11096 | int ret; | |
11097 | ||
11098 | ret = snprintf(file, sizeof(file), | |
11099 | "/sys/kernel/debug/tracing/events/%s/%s/id", | |
11100 | tp_category, tp_name); | |
11101 | if (ret < 0) | |
11102 | return -errno; | |
11103 | if (ret >= sizeof(file)) { | |
11104 | pr_debug("tracepoint %s/%s path is too long\n", | |
11105 | tp_category, tp_name); | |
11106 | return -E2BIG; | |
11107 | } | |
11108 | return parse_uint_from_file(file, "%d\n"); | |
11109 | } | |
11110 | ||
11111 | static int perf_event_open_tracepoint(const char *tp_category, | |
11112 | const char *tp_name) | |
11113 | { | |
11114 | struct perf_event_attr attr = {}; | |
11115 | char errmsg[STRERR_BUFSIZE]; | |
11116 | int tp_id, pfd, err; | |
11117 | ||
11118 | tp_id = determine_tracepoint_id(tp_category, tp_name); | |
11119 | if (tp_id < 0) { | |
be18010e KW |
11120 | pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n", |
11121 | tp_category, tp_name, | |
11122 | libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
11123 | return tp_id; |
11124 | } | |
11125 | ||
11126 | attr.type = PERF_TYPE_TRACEPOINT; | |
11127 | attr.size = sizeof(attr); | |
11128 | attr.config = tp_id; | |
11129 | ||
11130 | pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */, | |
11131 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
11132 | if (pfd < 0) { | |
11133 | err = -errno; | |
be18010e KW |
11134 | pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n", |
11135 | tp_category, tp_name, | |
11136 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
11137 | return err; |
11138 | } | |
11139 | return pfd; | |
11140 | } | |
11141 | ||
942025c9 | 11142 | struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog, |
47faff37 AN |
11143 | const char *tp_category, |
11144 | const char *tp_name, | |
11145 | const struct bpf_tracepoint_opts *opts) | |
f6de59c1 | 11146 | { |
47faff37 | 11147 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f6de59c1 AN |
11148 | char errmsg[STRERR_BUFSIZE]; |
11149 | struct bpf_link *link; | |
11150 | int pfd, err; | |
11151 | ||
47faff37 AN |
11152 | if (!OPTS_VALID(opts, bpf_tracepoint_opts)) |
11153 | return libbpf_err_ptr(-EINVAL); | |
11154 | ||
11155 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); | |
11156 | ||
f6de59c1 AN |
11157 | pfd = perf_event_open_tracepoint(tp_category, tp_name); |
11158 | if (pfd < 0) { | |
52109584 AN |
11159 | pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n", |
11160 | prog->name, tp_category, tp_name, | |
be18010e | 11161 | libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 11162 | return libbpf_err_ptr(pfd); |
f6de59c1 | 11163 | } |
47faff37 | 11164 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
11165 | err = libbpf_get_error(link); |
11166 | if (err) { | |
f6de59c1 | 11167 | close(pfd); |
52109584 AN |
11168 | pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n", |
11169 | prog->name, tp_category, tp_name, | |
be18010e | 11170 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 11171 | return libbpf_err_ptr(err); |
f6de59c1 AN |
11172 | } |
11173 | return link; | |
11174 | } | |
11175 | ||
942025c9 | 11176 | struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog, |
47faff37 AN |
11177 | const char *tp_category, |
11178 | const char *tp_name) | |
11179 | { | |
11180 | return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL); | |
11181 | } | |
11182 | ||
4fa5bcfe | 11183 | static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 AN |
11184 | { |
11185 | char *sec_name, *tp_cat, *tp_name; | |
d7a18ea7 | 11186 | |
52109584 | 11187 | sec_name = strdup(prog->sec_name); |
d7a18ea7 | 11188 | if (!sec_name) |
4fa5bcfe | 11189 | return -ENOMEM; |
d7a18ea7 | 11190 | |
13d35a0c AN |
11191 | /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */ |
11192 | if (str_has_pfx(prog->sec_name, "tp/")) | |
11193 | tp_cat = sec_name + sizeof("tp/") - 1; | |
11194 | else | |
11195 | tp_cat = sec_name + sizeof("tracepoint/") - 1; | |
d7a18ea7 AN |
11196 | tp_name = strchr(tp_cat, '/'); |
11197 | if (!tp_name) { | |
e9fc3ce9 | 11198 | free(sec_name); |
4fa5bcfe | 11199 | return -EINVAL; |
d7a18ea7 AN |
11200 | } |
11201 | *tp_name = '\0'; | |
11202 | tp_name++; | |
11203 | ||
4fa5bcfe | 11204 | *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name); |
d7a18ea7 | 11205 | free(sec_name); |
4fa5bcfe | 11206 | return libbpf_get_error(*link); |
d7a18ea7 AN |
11207 | } |
11208 | ||
942025c9 | 11209 | struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog, |
84bf5e1f AN |
11210 | const char *tp_name) |
11211 | { | |
11212 | char errmsg[STRERR_BUFSIZE]; | |
c016b68e | 11213 | struct bpf_link *link; |
84bf5e1f AN |
11214 | int prog_fd, pfd; |
11215 | ||
11216 | prog_fd = bpf_program__fd(prog); | |
11217 | if (prog_fd < 0) { | |
52109584 | 11218 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 11219 | return libbpf_err_ptr(-EINVAL); |
84bf5e1f AN |
11220 | } |
11221 | ||
d6958706 | 11222 | link = calloc(1, sizeof(*link)); |
84bf5e1f | 11223 | if (!link) |
e9fc3ce9 | 11224 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 11225 | link->detach = &bpf_link__detach_fd; |
84bf5e1f AN |
11226 | |
11227 | pfd = bpf_raw_tracepoint_open(tp_name, prog_fd); | |
11228 | if (pfd < 0) { | |
11229 | pfd = -errno; | |
11230 | free(link); | |
52109584 AN |
11231 | pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n", |
11232 | prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 11233 | return libbpf_err_ptr(pfd); |
84bf5e1f AN |
11234 | } |
11235 | link->fd = pfd; | |
c016b68e | 11236 | return link; |
84bf5e1f AN |
11237 | } |
11238 | ||
4fa5bcfe | 11239 | static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 11240 | { |
ccaf12d6 HT |
11241 | static const char *const prefixes[] = { |
11242 | "raw_tp/", | |
11243 | "raw_tracepoint/", | |
11244 | "raw_tp.w/", | |
11245 | "raw_tracepoint.w/", | |
11246 | }; | |
11247 | size_t i; | |
11248 | const char *tp_name = NULL; | |
13d35a0c | 11249 | |
ccaf12d6 HT |
11250 | for (i = 0; i < ARRAY_SIZE(prefixes); i++) { |
11251 | if (str_has_pfx(prog->sec_name, prefixes[i])) { | |
11252 | tp_name = prog->sec_name + strlen(prefixes[i]); | |
11253 | break; | |
11254 | } | |
11255 | } | |
11256 | if (!tp_name) { | |
11257 | pr_warn("prog '%s': invalid section name '%s'\n", | |
11258 | prog->name, prog->sec_name); | |
4fa5bcfe | 11259 | return -EINVAL; |
ccaf12d6 | 11260 | } |
d7a18ea7 | 11261 | |
4fa5bcfe AN |
11262 | *link = bpf_program__attach_raw_tracepoint(prog, tp_name); |
11263 | return libbpf_get_error(link); | |
d7a18ea7 AN |
11264 | } |
11265 | ||
1e092a03 | 11266 | /* Common logic for all BPF program types that attach to a btf_id */ |
942025c9 | 11267 | static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog) |
b8c54ea4 AS |
11268 | { |
11269 | char errmsg[STRERR_BUFSIZE]; | |
c016b68e | 11270 | struct bpf_link *link; |
b8c54ea4 AS |
11271 | int prog_fd, pfd; |
11272 | ||
11273 | prog_fd = bpf_program__fd(prog); | |
11274 | if (prog_fd < 0) { | |
52109584 | 11275 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 11276 | return libbpf_err_ptr(-EINVAL); |
b8c54ea4 AS |
11277 | } |
11278 | ||
d6958706 | 11279 | link = calloc(1, sizeof(*link)); |
b8c54ea4 | 11280 | if (!link) |
e9fc3ce9 | 11281 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 11282 | link->detach = &bpf_link__detach_fd; |
b8c54ea4 | 11283 | |
8462e0b4 AN |
11284 | /* libbpf is smart enough to redirect to BPF_RAW_TRACEPOINT_OPEN on old kernels */ |
11285 | pfd = bpf_link_create(prog_fd, 0, bpf_program__expected_attach_type(prog), NULL); | |
b8c54ea4 AS |
11286 | if (pfd < 0) { |
11287 | pfd = -errno; | |
11288 | free(link); | |
52109584 AN |
11289 | pr_warn("prog '%s': failed to attach: %s\n", |
11290 | prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 11291 | return libbpf_err_ptr(pfd); |
b8c54ea4 AS |
11292 | } |
11293 | link->fd = pfd; | |
003fed59 | 11294 | return link; |
b8c54ea4 AS |
11295 | } |
11296 | ||
942025c9 | 11297 | struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog) |
1e092a03 KS |
11298 | { |
11299 | return bpf_program__attach_btf_id(prog); | |
11300 | } | |
11301 | ||
942025c9 | 11302 | struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog) |
1e092a03 KS |
11303 | { |
11304 | return bpf_program__attach_btf_id(prog); | |
11305 | } | |
11306 | ||
4fa5bcfe | 11307 | static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 11308 | { |
4fa5bcfe AN |
11309 | *link = bpf_program__attach_trace(prog); |
11310 | return libbpf_get_error(*link); | |
d7a18ea7 AN |
11311 | } |
11312 | ||
4fa5bcfe | 11313 | static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
1e092a03 | 11314 | { |
4fa5bcfe AN |
11315 | *link = bpf_program__attach_lsm(prog); |
11316 | return libbpf_get_error(*link); | |
1e092a03 KS |
11317 | } |
11318 | ||
d60d81ac | 11319 | static struct bpf_link * |
942025c9 | 11320 | bpf_program__attach_fd(const struct bpf_program *prog, int target_fd, int btf_id, |
d60d81ac | 11321 | const char *target_name) |
cc4f864b | 11322 | { |
a5359091 THJ |
11323 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts, |
11324 | .target_btf_id = btf_id); | |
cc4f864b AN |
11325 | enum bpf_attach_type attach_type; |
11326 | char errmsg[STRERR_BUFSIZE]; | |
11327 | struct bpf_link *link; | |
11328 | int prog_fd, link_fd; | |
11329 | ||
11330 | prog_fd = bpf_program__fd(prog); | |
11331 | if (prog_fd < 0) { | |
52109584 | 11332 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 11333 | return libbpf_err_ptr(-EINVAL); |
cc4f864b AN |
11334 | } |
11335 | ||
11336 | link = calloc(1, sizeof(*link)); | |
11337 | if (!link) | |
e9fc3ce9 | 11338 | return libbpf_err_ptr(-ENOMEM); |
cc4f864b AN |
11339 | link->detach = &bpf_link__detach_fd; |
11340 | ||
20eccf29 | 11341 | attach_type = bpf_program__expected_attach_type(prog); |
a5359091 | 11342 | link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts); |
cc4f864b AN |
11343 | if (link_fd < 0) { |
11344 | link_fd = -errno; | |
11345 | free(link); | |
52109584 AN |
11346 | pr_warn("prog '%s': failed to attach to %s: %s\n", |
11347 | prog->name, target_name, | |
cc4f864b | 11348 | libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 11349 | return libbpf_err_ptr(link_fd); |
cc4f864b AN |
11350 | } |
11351 | link->fd = link_fd; | |
11352 | return link; | |
11353 | } | |
11354 | ||
d60d81ac | 11355 | struct bpf_link * |
942025c9 | 11356 | bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd) |
d60d81ac | 11357 | { |
a5359091 | 11358 | return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup"); |
d60d81ac JS |
11359 | } |
11360 | ||
11361 | struct bpf_link * | |
942025c9 | 11362 | bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd) |
d60d81ac | 11363 | { |
a5359091 | 11364 | return bpf_program__attach_fd(prog, netns_fd, 0, "netns"); |
d60d81ac JS |
11365 | } |
11366 | ||
942025c9 | 11367 | struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex) |
dc8698ca AN |
11368 | { |
11369 | /* target_fd/target_ifindex use the same field in LINK_CREATE */ | |
a5359091 THJ |
11370 | return bpf_program__attach_fd(prog, ifindex, 0, "xdp"); |
11371 | } | |
11372 | ||
942025c9 | 11373 | struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog, |
a5359091 THJ |
11374 | int target_fd, |
11375 | const char *attach_func_name) | |
11376 | { | |
11377 | int btf_id; | |
11378 | ||
11379 | if (!!target_fd != !!attach_func_name) { | |
11380 | pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n", | |
11381 | prog->name); | |
e9fc3ce9 | 11382 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
11383 | } |
11384 | ||
11385 | if (prog->type != BPF_PROG_TYPE_EXT) { | |
11386 | pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace", | |
11387 | prog->name); | |
e9fc3ce9 | 11388 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
11389 | } |
11390 | ||
11391 | if (target_fd) { | |
11392 | btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd); | |
11393 | if (btf_id < 0) | |
e9fc3ce9 | 11394 | return libbpf_err_ptr(btf_id); |
a5359091 THJ |
11395 | |
11396 | return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace"); | |
11397 | } else { | |
11398 | /* no target, so use raw_tracepoint_open for compatibility | |
11399 | * with old kernels | |
11400 | */ | |
11401 | return bpf_program__attach_trace(prog); | |
11402 | } | |
dc8698ca AN |
11403 | } |
11404 | ||
c09add2f | 11405 | struct bpf_link * |
942025c9 | 11406 | bpf_program__attach_iter(const struct bpf_program *prog, |
c09add2f YS |
11407 | const struct bpf_iter_attach_opts *opts) |
11408 | { | |
cd31039a | 11409 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); |
c09add2f YS |
11410 | char errmsg[STRERR_BUFSIZE]; |
11411 | struct bpf_link *link; | |
11412 | int prog_fd, link_fd; | |
cd31039a | 11413 | __u32 target_fd = 0; |
c09add2f YS |
11414 | |
11415 | if (!OPTS_VALID(opts, bpf_iter_attach_opts)) | |
e9fc3ce9 | 11416 | return libbpf_err_ptr(-EINVAL); |
c09add2f | 11417 | |
74fc097d YS |
11418 | link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0); |
11419 | link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0); | |
cd31039a | 11420 | |
c09add2f YS |
11421 | prog_fd = bpf_program__fd(prog); |
11422 | if (prog_fd < 0) { | |
52109584 | 11423 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 11424 | return libbpf_err_ptr(-EINVAL); |
c09add2f YS |
11425 | } |
11426 | ||
11427 | link = calloc(1, sizeof(*link)); | |
11428 | if (!link) | |
e9fc3ce9 | 11429 | return libbpf_err_ptr(-ENOMEM); |
c09add2f YS |
11430 | link->detach = &bpf_link__detach_fd; |
11431 | ||
cd31039a YS |
11432 | link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER, |
11433 | &link_create_opts); | |
c09add2f YS |
11434 | if (link_fd < 0) { |
11435 | link_fd = -errno; | |
11436 | free(link); | |
52109584 AN |
11437 | pr_warn("prog '%s': failed to attach to iterator: %s\n", |
11438 | prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 11439 | return libbpf_err_ptr(link_fd); |
c09add2f YS |
11440 | } |
11441 | link->fd = link_fd; | |
11442 | return link; | |
11443 | } | |
11444 | ||
4fa5bcfe | 11445 | static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
e9fc3ce9 | 11446 | { |
4fa5bcfe AN |
11447 | *link = bpf_program__attach_iter(prog, NULL); |
11448 | return libbpf_get_error(*link); | |
e9fc3ce9 AN |
11449 | } |
11450 | ||
942025c9 | 11451 | struct bpf_link *bpf_program__attach(const struct bpf_program *prog) |
d7a18ea7 | 11452 | { |
4fa5bcfe AN |
11453 | struct bpf_link *link = NULL; |
11454 | int err; | |
11455 | ||
11456 | if (!prog->sec_def || !prog->sec_def->prog_attach_fn) | |
11457 | return libbpf_err_ptr(-EOPNOTSUPP); | |
d7a18ea7 | 11458 | |
4fa5bcfe AN |
11459 | err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link); |
11460 | if (err) | |
11461 | return libbpf_err_ptr(err); | |
11462 | ||
11463 | /* When calling bpf_program__attach() explicitly, auto-attach support | |
11464 | * is expected to work, so NULL returned link is considered an error. | |
11465 | * This is different for skeleton's attach, see comment in | |
11466 | * bpf_object__attach_skeleton(). | |
11467 | */ | |
11468 | if (!link) | |
11469 | return libbpf_err_ptr(-EOPNOTSUPP); | |
11470 | ||
11471 | return link; | |
d7a18ea7 AN |
11472 | } |
11473 | ||
590a0088 MKL |
11474 | static int bpf_link__detach_struct_ops(struct bpf_link *link) |
11475 | { | |
590a0088 MKL |
11476 | __u32 zero = 0; |
11477 | ||
c016b68e | 11478 | if (bpf_map_delete_elem(link->fd, &zero)) |
590a0088 MKL |
11479 | return -errno; |
11480 | ||
11481 | return 0; | |
11482 | } | |
11483 | ||
942025c9 | 11484 | struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map) |
590a0088 MKL |
11485 | { |
11486 | struct bpf_struct_ops *st_ops; | |
c016b68e | 11487 | struct bpf_link *link; |
590a0088 MKL |
11488 | __u32 i, zero = 0; |
11489 | int err; | |
11490 | ||
11491 | if (!bpf_map__is_struct_ops(map) || map->fd == -1) | |
e9fc3ce9 | 11492 | return libbpf_err_ptr(-EINVAL); |
590a0088 MKL |
11493 | |
11494 | link = calloc(1, sizeof(*link)); | |
11495 | if (!link) | |
e9fc3ce9 | 11496 | return libbpf_err_ptr(-EINVAL); |
590a0088 MKL |
11497 | |
11498 | st_ops = map->st_ops; | |
11499 | for (i = 0; i < btf_vlen(st_ops->type); i++) { | |
11500 | struct bpf_program *prog = st_ops->progs[i]; | |
11501 | void *kern_data; | |
11502 | int prog_fd; | |
11503 | ||
11504 | if (!prog) | |
11505 | continue; | |
11506 | ||
11507 | prog_fd = bpf_program__fd(prog); | |
11508 | kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i]; | |
11509 | *(unsigned long *)kern_data = prog_fd; | |
11510 | } | |
11511 | ||
11512 | err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0); | |
11513 | if (err) { | |
11514 | err = -errno; | |
11515 | free(link); | |
e9fc3ce9 | 11516 | return libbpf_err_ptr(err); |
590a0088 MKL |
11517 | } |
11518 | ||
c016b68e | 11519 | link->detach = bpf_link__detach_struct_ops; |
590a0088 MKL |
11520 | link->fd = map->fd; |
11521 | ||
c016b68e | 11522 | return link; |
590a0088 MKL |
11523 | } |
11524 | ||
7218c28c CL |
11525 | static enum bpf_perf_event_ret |
11526 | perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size, | |
11527 | void **copy_mem, size_t *copy_size, | |
11528 | bpf_perf_event_print_t fn, void *private_data) | |
d0cabbb0 | 11529 | { |
3dca2115 | 11530 | struct perf_event_mmap_page *header = mmap_mem; |
a64af0ef | 11531 | __u64 data_head = ring_buffer_read_head(header); |
d0cabbb0 | 11532 | __u64 data_tail = header->data_tail; |
3dca2115 DB |
11533 | void *base = ((__u8 *)header) + page_size; |
11534 | int ret = LIBBPF_PERF_EVENT_CONT; | |
11535 | struct perf_event_header *ehdr; | |
11536 | size_t ehdr_size; | |
11537 | ||
11538 | while (data_head != data_tail) { | |
11539 | ehdr = base + (data_tail & (mmap_size - 1)); | |
11540 | ehdr_size = ehdr->size; | |
11541 | ||
11542 | if (((void *)ehdr) + ehdr_size > base + mmap_size) { | |
11543 | void *copy_start = ehdr; | |
11544 | size_t len_first = base + mmap_size - copy_start; | |
11545 | size_t len_secnd = ehdr_size - len_first; | |
11546 | ||
11547 | if (*copy_size < ehdr_size) { | |
11548 | free(*copy_mem); | |
11549 | *copy_mem = malloc(ehdr_size); | |
11550 | if (!*copy_mem) { | |
11551 | *copy_size = 0; | |
d0cabbb0 JK |
11552 | ret = LIBBPF_PERF_EVENT_ERROR; |
11553 | break; | |
11554 | } | |
3dca2115 | 11555 | *copy_size = ehdr_size; |
d0cabbb0 JK |
11556 | } |
11557 | ||
3dca2115 DB |
11558 | memcpy(*copy_mem, copy_start, len_first); |
11559 | memcpy(*copy_mem + len_first, base, len_secnd); | |
11560 | ehdr = *copy_mem; | |
d0cabbb0 JK |
11561 | } |
11562 | ||
3dca2115 DB |
11563 | ret = fn(ehdr, private_data); |
11564 | data_tail += ehdr_size; | |
d0cabbb0 JK |
11565 | if (ret != LIBBPF_PERF_EVENT_CONT) |
11566 | break; | |
d0cabbb0 JK |
11567 | } |
11568 | ||
a64af0ef | 11569 | ring_buffer_write_tail(header, data_tail); |
e9fc3ce9 | 11570 | return libbpf_err(ret); |
d0cabbb0 | 11571 | } |
34be1646 | 11572 | |
7218c28c CL |
11573 | __attribute__((alias("perf_event_read_simple"))) |
11574 | enum bpf_perf_event_ret | |
11575 | bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size, | |
11576 | void **copy_mem, size_t *copy_size, | |
11577 | bpf_perf_event_print_t fn, void *private_data); | |
11578 | ||
fb84b822 AN |
11579 | struct perf_buffer; |
11580 | ||
11581 | struct perf_buffer_params { | |
11582 | struct perf_event_attr *attr; | |
11583 | /* if event_cb is specified, it takes precendence */ | |
11584 | perf_buffer_event_fn event_cb; | |
11585 | /* sample_cb and lost_cb are higher-level common-case callbacks */ | |
11586 | perf_buffer_sample_fn sample_cb; | |
11587 | perf_buffer_lost_fn lost_cb; | |
11588 | void *ctx; | |
11589 | int cpu_cnt; | |
11590 | int *cpus; | |
11591 | int *map_keys; | |
11592 | }; | |
11593 | ||
11594 | struct perf_cpu_buf { | |
11595 | struct perf_buffer *pb; | |
11596 | void *base; /* mmap()'ed memory */ | |
11597 | void *buf; /* for reconstructing segmented data */ | |
11598 | size_t buf_size; | |
11599 | int fd; | |
11600 | int cpu; | |
11601 | int map_key; | |
11602 | }; | |
11603 | ||
11604 | struct perf_buffer { | |
11605 | perf_buffer_event_fn event_cb; | |
11606 | perf_buffer_sample_fn sample_cb; | |
11607 | perf_buffer_lost_fn lost_cb; | |
11608 | void *ctx; /* passed into callbacks */ | |
11609 | ||
11610 | size_t page_size; | |
11611 | size_t mmap_size; | |
11612 | struct perf_cpu_buf **cpu_bufs; | |
11613 | struct epoll_event *events; | |
783b8f01 | 11614 | int cpu_cnt; /* number of allocated CPU buffers */ |
fb84b822 AN |
11615 | int epoll_fd; /* perf event FD */ |
11616 | int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */ | |
11617 | }; | |
11618 | ||
11619 | static void perf_buffer__free_cpu_buf(struct perf_buffer *pb, | |
11620 | struct perf_cpu_buf *cpu_buf) | |
11621 | { | |
11622 | if (!cpu_buf) | |
11623 | return; | |
11624 | if (cpu_buf->base && | |
11625 | munmap(cpu_buf->base, pb->mmap_size + pb->page_size)) | |
be18010e | 11626 | pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu); |
fb84b822 AN |
11627 | if (cpu_buf->fd >= 0) { |
11628 | ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0); | |
11629 | close(cpu_buf->fd); | |
11630 | } | |
11631 | free(cpu_buf->buf); | |
11632 | free(cpu_buf); | |
11633 | } | |
11634 | ||
11635 | void perf_buffer__free(struct perf_buffer *pb) | |
11636 | { | |
11637 | int i; | |
11638 | ||
50450fc7 | 11639 | if (IS_ERR_OR_NULL(pb)) |
fb84b822 AN |
11640 | return; |
11641 | if (pb->cpu_bufs) { | |
601b05ca | 11642 | for (i = 0; i < pb->cpu_cnt; i++) { |
fb84b822 AN |
11643 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; |
11644 | ||
601b05ca EC |
11645 | if (!cpu_buf) |
11646 | continue; | |
11647 | ||
fb84b822 AN |
11648 | bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key); |
11649 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
11650 | } | |
11651 | free(pb->cpu_bufs); | |
11652 | } | |
11653 | if (pb->epoll_fd >= 0) | |
11654 | close(pb->epoll_fd); | |
11655 | free(pb->events); | |
11656 | free(pb); | |
11657 | } | |
11658 | ||
11659 | static struct perf_cpu_buf * | |
11660 | perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr, | |
11661 | int cpu, int map_key) | |
11662 | { | |
11663 | struct perf_cpu_buf *cpu_buf; | |
11664 | char msg[STRERR_BUFSIZE]; | |
11665 | int err; | |
11666 | ||
11667 | cpu_buf = calloc(1, sizeof(*cpu_buf)); | |
11668 | if (!cpu_buf) | |
11669 | return ERR_PTR(-ENOMEM); | |
11670 | ||
11671 | cpu_buf->pb = pb; | |
11672 | cpu_buf->cpu = cpu; | |
11673 | cpu_buf->map_key = map_key; | |
11674 | ||
11675 | cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu, | |
11676 | -1, PERF_FLAG_FD_CLOEXEC); | |
11677 | if (cpu_buf->fd < 0) { | |
11678 | err = -errno; | |
be18010e KW |
11679 | pr_warn("failed to open perf buffer event on cpu #%d: %s\n", |
11680 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11681 | goto error; |
11682 | } | |
11683 | ||
11684 | cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size, | |
11685 | PROT_READ | PROT_WRITE, MAP_SHARED, | |
11686 | cpu_buf->fd, 0); | |
11687 | if (cpu_buf->base == MAP_FAILED) { | |
11688 | cpu_buf->base = NULL; | |
11689 | err = -errno; | |
be18010e KW |
11690 | pr_warn("failed to mmap perf buffer on cpu #%d: %s\n", |
11691 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11692 | goto error; |
11693 | } | |
11694 | ||
11695 | if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
11696 | err = -errno; | |
be18010e KW |
11697 | pr_warn("failed to enable perf buffer event on cpu #%d: %s\n", |
11698 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11699 | goto error; |
11700 | } | |
11701 | ||
11702 | return cpu_buf; | |
11703 | ||
11704 | error: | |
11705 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
11706 | return (struct perf_cpu_buf *)ERR_PTR(err); | |
11707 | } | |
11708 | ||
11709 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
11710 | struct perf_buffer_params *p); | |
11711 | ||
41788934 AN |
11712 | DEFAULT_VERSION(perf_buffer__new_v0_6_0, perf_buffer__new, LIBBPF_0.6.0) |
11713 | struct perf_buffer *perf_buffer__new_v0_6_0(int map_fd, size_t page_cnt, | |
11714 | perf_buffer_sample_fn sample_cb, | |
11715 | perf_buffer_lost_fn lost_cb, | |
11716 | void *ctx, | |
11717 | const struct perf_buffer_opts *opts) | |
fb84b822 AN |
11718 | { |
11719 | struct perf_buffer_params p = {}; | |
41788934 AN |
11720 | struct perf_event_attr attr = {}; |
11721 | ||
11722 | if (!OPTS_VALID(opts, perf_buffer_opts)) | |
11723 | return libbpf_err_ptr(-EINVAL); | |
4be6e05c | 11724 | |
65bb2e0f | 11725 | attr.config = PERF_COUNT_SW_BPF_OUTPUT; |
4be6e05c ACM |
11726 | attr.type = PERF_TYPE_SOFTWARE; |
11727 | attr.sample_type = PERF_SAMPLE_RAW; | |
11728 | attr.sample_period = 1; | |
11729 | attr.wakeup_events = 1; | |
fb84b822 AN |
11730 | |
11731 | p.attr = &attr; | |
41788934 AN |
11732 | p.sample_cb = sample_cb; |
11733 | p.lost_cb = lost_cb; | |
11734 | p.ctx = ctx; | |
fb84b822 | 11735 | |
e9fc3ce9 | 11736 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
11737 | } |
11738 | ||
41788934 AN |
11739 | COMPAT_VERSION(perf_buffer__new_deprecated, perf_buffer__new, LIBBPF_0.0.4) |
11740 | struct perf_buffer *perf_buffer__new_deprecated(int map_fd, size_t page_cnt, | |
11741 | const struct perf_buffer_opts *opts) | |
11742 | { | |
11743 | return perf_buffer__new_v0_6_0(map_fd, page_cnt, | |
11744 | opts ? opts->sample_cb : NULL, | |
11745 | opts ? opts->lost_cb : NULL, | |
11746 | opts ? opts->ctx : NULL, | |
11747 | NULL); | |
11748 | } | |
11749 | ||
11750 | DEFAULT_VERSION(perf_buffer__new_raw_v0_6_0, perf_buffer__new_raw, LIBBPF_0.6.0) | |
11751 | struct perf_buffer *perf_buffer__new_raw_v0_6_0(int map_fd, size_t page_cnt, | |
11752 | struct perf_event_attr *attr, | |
11753 | perf_buffer_event_fn event_cb, void *ctx, | |
11754 | const struct perf_buffer_raw_opts *opts) | |
fb84b822 AN |
11755 | { |
11756 | struct perf_buffer_params p = {}; | |
11757 | ||
41332d6e | 11758 | if (!attr) |
41788934 AN |
11759 | return libbpf_err_ptr(-EINVAL); |
11760 | ||
11761 | if (!OPTS_VALID(opts, perf_buffer_raw_opts)) | |
11762 | return libbpf_err_ptr(-EINVAL); | |
11763 | ||
11764 | p.attr = attr; | |
11765 | p.event_cb = event_cb; | |
11766 | p.ctx = ctx; | |
11767 | p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0); | |
11768 | p.cpus = OPTS_GET(opts, cpus, NULL); | |
11769 | p.map_keys = OPTS_GET(opts, map_keys, NULL); | |
fb84b822 | 11770 | |
e9fc3ce9 | 11771 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
11772 | } |
11773 | ||
41788934 AN |
11774 | COMPAT_VERSION(perf_buffer__new_raw_deprecated, perf_buffer__new_raw, LIBBPF_0.0.4) |
11775 | struct perf_buffer *perf_buffer__new_raw_deprecated(int map_fd, size_t page_cnt, | |
11776 | const struct perf_buffer_raw_opts *opts) | |
11777 | { | |
11778 | LIBBPF_OPTS(perf_buffer_raw_opts, inner_opts, | |
11779 | .cpu_cnt = opts->cpu_cnt, | |
11780 | .cpus = opts->cpus, | |
11781 | .map_keys = opts->map_keys, | |
11782 | ); | |
11783 | ||
11784 | return perf_buffer__new_raw_v0_6_0(map_fd, page_cnt, opts->attr, | |
11785 | opts->event_cb, opts->ctx, &inner_opts); | |
11786 | } | |
11787 | ||
fb84b822 AN |
11788 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, |
11789 | struct perf_buffer_params *p) | |
11790 | { | |
783b8f01 | 11791 | const char *online_cpus_file = "/sys/devices/system/cpu/online"; |
0e289487 | 11792 | struct bpf_map_info map; |
fb84b822 AN |
11793 | char msg[STRERR_BUFSIZE]; |
11794 | struct perf_buffer *pb; | |
783b8f01 | 11795 | bool *online = NULL; |
fb84b822 | 11796 | __u32 map_info_len; |
783b8f01 | 11797 | int err, i, j, n; |
fb84b822 | 11798 | |
41332d6e | 11799 | if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) { |
be18010e KW |
11800 | pr_warn("page count should be power of two, but is %zu\n", |
11801 | page_cnt); | |
fb84b822 AN |
11802 | return ERR_PTR(-EINVAL); |
11803 | } | |
11804 | ||
0e289487 AN |
11805 | /* best-effort sanity checks */ |
11806 | memset(&map, 0, sizeof(map)); | |
fb84b822 AN |
11807 | map_info_len = sizeof(map); |
11808 | err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len); | |
11809 | if (err) { | |
11810 | err = -errno; | |
0e289487 AN |
11811 | /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return |
11812 | * -EBADFD, -EFAULT, or -E2BIG on real error | |
11813 | */ | |
11814 | if (err != -EINVAL) { | |
11815 | pr_warn("failed to get map info for map FD %d: %s\n", | |
11816 | map_fd, libbpf_strerror_r(err, msg, sizeof(msg))); | |
11817 | return ERR_PTR(err); | |
11818 | } | |
11819 | pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n", | |
11820 | map_fd); | |
11821 | } else { | |
11822 | if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) { | |
11823 | pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n", | |
11824 | map.name); | |
11825 | return ERR_PTR(-EINVAL); | |
11826 | } | |
fb84b822 AN |
11827 | } |
11828 | ||
11829 | pb = calloc(1, sizeof(*pb)); | |
11830 | if (!pb) | |
11831 | return ERR_PTR(-ENOMEM); | |
11832 | ||
11833 | pb->event_cb = p->event_cb; | |
11834 | pb->sample_cb = p->sample_cb; | |
11835 | pb->lost_cb = p->lost_cb; | |
11836 | pb->ctx = p->ctx; | |
11837 | ||
11838 | pb->page_size = getpagesize(); | |
11839 | pb->mmap_size = pb->page_size * page_cnt; | |
11840 | pb->map_fd = map_fd; | |
11841 | ||
11842 | pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
11843 | if (pb->epoll_fd < 0) { | |
11844 | err = -errno; | |
be18010e KW |
11845 | pr_warn("failed to create epoll instance: %s\n", |
11846 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11847 | goto error; |
11848 | } | |
11849 | ||
11850 | if (p->cpu_cnt > 0) { | |
11851 | pb->cpu_cnt = p->cpu_cnt; | |
11852 | } else { | |
11853 | pb->cpu_cnt = libbpf_num_possible_cpus(); | |
11854 | if (pb->cpu_cnt < 0) { | |
11855 | err = pb->cpu_cnt; | |
11856 | goto error; | |
11857 | } | |
0e289487 | 11858 | if (map.max_entries && map.max_entries < pb->cpu_cnt) |
fb84b822 AN |
11859 | pb->cpu_cnt = map.max_entries; |
11860 | } | |
11861 | ||
11862 | pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events)); | |
11863 | if (!pb->events) { | |
11864 | err = -ENOMEM; | |
be18010e | 11865 | pr_warn("failed to allocate events: out of memory\n"); |
fb84b822 AN |
11866 | goto error; |
11867 | } | |
11868 | pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs)); | |
11869 | if (!pb->cpu_bufs) { | |
11870 | err = -ENOMEM; | |
be18010e | 11871 | pr_warn("failed to allocate buffers: out of memory\n"); |
fb84b822 AN |
11872 | goto error; |
11873 | } | |
11874 | ||
783b8f01 AN |
11875 | err = parse_cpu_mask_file(online_cpus_file, &online, &n); |
11876 | if (err) { | |
11877 | pr_warn("failed to get online CPU mask: %d\n", err); | |
11878 | goto error; | |
11879 | } | |
11880 | ||
11881 | for (i = 0, j = 0; i < pb->cpu_cnt; i++) { | |
fb84b822 AN |
11882 | struct perf_cpu_buf *cpu_buf; |
11883 | int cpu, map_key; | |
11884 | ||
11885 | cpu = p->cpu_cnt > 0 ? p->cpus[i] : i; | |
11886 | map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i; | |
11887 | ||
783b8f01 AN |
11888 | /* in case user didn't explicitly requested particular CPUs to |
11889 | * be attached to, skip offline/not present CPUs | |
11890 | */ | |
11891 | if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu])) | |
11892 | continue; | |
11893 | ||
fb84b822 AN |
11894 | cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key); |
11895 | if (IS_ERR(cpu_buf)) { | |
11896 | err = PTR_ERR(cpu_buf); | |
11897 | goto error; | |
11898 | } | |
11899 | ||
783b8f01 | 11900 | pb->cpu_bufs[j] = cpu_buf; |
fb84b822 AN |
11901 | |
11902 | err = bpf_map_update_elem(pb->map_fd, &map_key, | |
11903 | &cpu_buf->fd, 0); | |
11904 | if (err) { | |
11905 | err = -errno; | |
be18010e KW |
11906 | pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n", |
11907 | cpu, map_key, cpu_buf->fd, | |
11908 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11909 | goto error; |
11910 | } | |
11911 | ||
783b8f01 AN |
11912 | pb->events[j].events = EPOLLIN; |
11913 | pb->events[j].data.ptr = cpu_buf; | |
fb84b822 | 11914 | if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd, |
783b8f01 | 11915 | &pb->events[j]) < 0) { |
fb84b822 | 11916 | err = -errno; |
be18010e KW |
11917 | pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n", |
11918 | cpu, cpu_buf->fd, | |
11919 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11920 | goto error; |
11921 | } | |
783b8f01 | 11922 | j++; |
fb84b822 | 11923 | } |
783b8f01 AN |
11924 | pb->cpu_cnt = j; |
11925 | free(online); | |
fb84b822 AN |
11926 | |
11927 | return pb; | |
11928 | ||
11929 | error: | |
783b8f01 | 11930 | free(online); |
fb84b822 AN |
11931 | if (pb) |
11932 | perf_buffer__free(pb); | |
11933 | return ERR_PTR(err); | |
11934 | } | |
11935 | ||
11936 | struct perf_sample_raw { | |
11937 | struct perf_event_header header; | |
11938 | uint32_t size; | |
385bbf7b | 11939 | char data[]; |
fb84b822 AN |
11940 | }; |
11941 | ||
11942 | struct perf_sample_lost { | |
11943 | struct perf_event_header header; | |
11944 | uint64_t id; | |
11945 | uint64_t lost; | |
11946 | uint64_t sample_id; | |
11947 | }; | |
11948 | ||
11949 | static enum bpf_perf_event_ret | |
11950 | perf_buffer__process_record(struct perf_event_header *e, void *ctx) | |
11951 | { | |
11952 | struct perf_cpu_buf *cpu_buf = ctx; | |
11953 | struct perf_buffer *pb = cpu_buf->pb; | |
11954 | void *data = e; | |
11955 | ||
11956 | /* user wants full control over parsing perf event */ | |
11957 | if (pb->event_cb) | |
11958 | return pb->event_cb(pb->ctx, cpu_buf->cpu, e); | |
11959 | ||
11960 | switch (e->type) { | |
11961 | case PERF_RECORD_SAMPLE: { | |
11962 | struct perf_sample_raw *s = data; | |
11963 | ||
11964 | if (pb->sample_cb) | |
11965 | pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size); | |
11966 | break; | |
11967 | } | |
11968 | case PERF_RECORD_LOST: { | |
11969 | struct perf_sample_lost *s = data; | |
11970 | ||
11971 | if (pb->lost_cb) | |
11972 | pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost); | |
11973 | break; | |
11974 | } | |
11975 | default: | |
be18010e | 11976 | pr_warn("unknown perf sample type %d\n", e->type); |
fb84b822 AN |
11977 | return LIBBPF_PERF_EVENT_ERROR; |
11978 | } | |
11979 | return LIBBPF_PERF_EVENT_CONT; | |
11980 | } | |
11981 | ||
11982 | static int perf_buffer__process_records(struct perf_buffer *pb, | |
11983 | struct perf_cpu_buf *cpu_buf) | |
11984 | { | |
11985 | enum bpf_perf_event_ret ret; | |
11986 | ||
7218c28c CL |
11987 | ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size, |
11988 | pb->page_size, &cpu_buf->buf, | |
11989 | &cpu_buf->buf_size, | |
11990 | perf_buffer__process_record, cpu_buf); | |
fb84b822 AN |
11991 | if (ret != LIBBPF_PERF_EVENT_CONT) |
11992 | return ret; | |
11993 | return 0; | |
11994 | } | |
11995 | ||
dca5612f AN |
11996 | int perf_buffer__epoll_fd(const struct perf_buffer *pb) |
11997 | { | |
11998 | return pb->epoll_fd; | |
11999 | } | |
12000 | ||
fb84b822 AN |
12001 | int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms) |
12002 | { | |
12003 | int i, cnt, err; | |
12004 | ||
12005 | cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms); | |
e9fc3ce9 | 12006 | if (cnt < 0) |
af0efa05 | 12007 | return -errno; |
e9fc3ce9 | 12008 | |
fb84b822 AN |
12009 | for (i = 0; i < cnt; i++) { |
12010 | struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr; | |
12011 | ||
12012 | err = perf_buffer__process_records(pb, cpu_buf); | |
12013 | if (err) { | |
be18010e | 12014 | pr_warn("error while processing records: %d\n", err); |
e9fc3ce9 | 12015 | return libbpf_err(err); |
fb84b822 AN |
12016 | } |
12017 | } | |
e9fc3ce9 | 12018 | return cnt; |
fb84b822 AN |
12019 | } |
12020 | ||
dca5612f AN |
12021 | /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer |
12022 | * manager. | |
12023 | */ | |
12024 | size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb) | |
12025 | { | |
12026 | return pb->cpu_cnt; | |
12027 | } | |
12028 | ||
12029 | /* | |
12030 | * Return perf_event FD of a ring buffer in *buf_idx* slot of | |
12031 | * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using | |
12032 | * select()/poll()/epoll() Linux syscalls. | |
12033 | */ | |
12034 | int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx) | |
12035 | { | |
12036 | struct perf_cpu_buf *cpu_buf; | |
12037 | ||
12038 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 12039 | return libbpf_err(-EINVAL); |
dca5612f AN |
12040 | |
12041 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
12042 | if (!cpu_buf) | |
e9fc3ce9 | 12043 | return libbpf_err(-ENOENT); |
dca5612f AN |
12044 | |
12045 | return cpu_buf->fd; | |
12046 | } | |
12047 | ||
12048 | /* | |
12049 | * Consume data from perf ring buffer corresponding to slot *buf_idx* in | |
12050 | * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to | |
12051 | * consume, do nothing and return success. | |
12052 | * Returns: | |
12053 | * - 0 on success; | |
12054 | * - <0 on failure. | |
12055 | */ | |
12056 | int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx) | |
12057 | { | |
12058 | struct perf_cpu_buf *cpu_buf; | |
12059 | ||
12060 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 12061 | return libbpf_err(-EINVAL); |
dca5612f AN |
12062 | |
12063 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
12064 | if (!cpu_buf) | |
e9fc3ce9 | 12065 | return libbpf_err(-ENOENT); |
dca5612f AN |
12066 | |
12067 | return perf_buffer__process_records(pb, cpu_buf); | |
12068 | } | |
12069 | ||
272d51af EC |
12070 | int perf_buffer__consume(struct perf_buffer *pb) |
12071 | { | |
12072 | int i, err; | |
12073 | ||
12074 | for (i = 0; i < pb->cpu_cnt; i++) { | |
12075 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; | |
12076 | ||
12077 | if (!cpu_buf) | |
12078 | continue; | |
12079 | ||
12080 | err = perf_buffer__process_records(pb, cpu_buf); | |
12081 | if (err) { | |
dca5612f | 12082 | pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err); |
e9fc3ce9 | 12083 | return libbpf_err(err); |
272d51af EC |
12084 | } |
12085 | } | |
12086 | return 0; | |
12087 | } | |
12088 | ||
34be1646 SL |
12089 | struct bpf_prog_info_array_desc { |
12090 | int array_offset; /* e.g. offset of jited_prog_insns */ | |
12091 | int count_offset; /* e.g. offset of jited_prog_len */ | |
12092 | int size_offset; /* > 0: offset of rec size, | |
12093 | * < 0: fix size of -size_offset | |
12094 | */ | |
12095 | }; | |
12096 | ||
12097 | static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = { | |
12098 | [BPF_PROG_INFO_JITED_INSNS] = { | |
12099 | offsetof(struct bpf_prog_info, jited_prog_insns), | |
12100 | offsetof(struct bpf_prog_info, jited_prog_len), | |
12101 | -1, | |
12102 | }, | |
12103 | [BPF_PROG_INFO_XLATED_INSNS] = { | |
12104 | offsetof(struct bpf_prog_info, xlated_prog_insns), | |
12105 | offsetof(struct bpf_prog_info, xlated_prog_len), | |
12106 | -1, | |
12107 | }, | |
12108 | [BPF_PROG_INFO_MAP_IDS] = { | |
12109 | offsetof(struct bpf_prog_info, map_ids), | |
12110 | offsetof(struct bpf_prog_info, nr_map_ids), | |
12111 | -(int)sizeof(__u32), | |
12112 | }, | |
12113 | [BPF_PROG_INFO_JITED_KSYMS] = { | |
12114 | offsetof(struct bpf_prog_info, jited_ksyms), | |
12115 | offsetof(struct bpf_prog_info, nr_jited_ksyms), | |
12116 | -(int)sizeof(__u64), | |
12117 | }, | |
12118 | [BPF_PROG_INFO_JITED_FUNC_LENS] = { | |
12119 | offsetof(struct bpf_prog_info, jited_func_lens), | |
12120 | offsetof(struct bpf_prog_info, nr_jited_func_lens), | |
12121 | -(int)sizeof(__u32), | |
12122 | }, | |
12123 | [BPF_PROG_INFO_FUNC_INFO] = { | |
12124 | offsetof(struct bpf_prog_info, func_info), | |
12125 | offsetof(struct bpf_prog_info, nr_func_info), | |
12126 | offsetof(struct bpf_prog_info, func_info_rec_size), | |
12127 | }, | |
12128 | [BPF_PROG_INFO_LINE_INFO] = { | |
12129 | offsetof(struct bpf_prog_info, line_info), | |
12130 | offsetof(struct bpf_prog_info, nr_line_info), | |
12131 | offsetof(struct bpf_prog_info, line_info_rec_size), | |
12132 | }, | |
12133 | [BPF_PROG_INFO_JITED_LINE_INFO] = { | |
12134 | offsetof(struct bpf_prog_info, jited_line_info), | |
12135 | offsetof(struct bpf_prog_info, nr_jited_line_info), | |
12136 | offsetof(struct bpf_prog_info, jited_line_info_rec_size), | |
12137 | }, | |
12138 | [BPF_PROG_INFO_PROG_TAGS] = { | |
12139 | offsetof(struct bpf_prog_info, prog_tags), | |
12140 | offsetof(struct bpf_prog_info, nr_prog_tags), | |
12141 | -(int)sizeof(__u8) * BPF_TAG_SIZE, | |
12142 | }, | |
12143 | ||
12144 | }; | |
12145 | ||
8983b731 AN |
12146 | static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, |
12147 | int offset) | |
34be1646 SL |
12148 | { |
12149 | __u32 *array = (__u32 *)info; | |
12150 | ||
12151 | if (offset >= 0) | |
12152 | return array[offset / sizeof(__u32)]; | |
12153 | return -(int)offset; | |
12154 | } | |
12155 | ||
8983b731 AN |
12156 | static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, |
12157 | int offset) | |
34be1646 SL |
12158 | { |
12159 | __u64 *array = (__u64 *)info; | |
12160 | ||
12161 | if (offset >= 0) | |
12162 | return array[offset / sizeof(__u64)]; | |
12163 | return -(int)offset; | |
12164 | } | |
12165 | ||
12166 | static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset, | |
12167 | __u32 val) | |
12168 | { | |
12169 | __u32 *array = (__u32 *)info; | |
12170 | ||
12171 | if (offset >= 0) | |
12172 | array[offset / sizeof(__u32)] = val; | |
12173 | } | |
12174 | ||
12175 | static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset, | |
12176 | __u64 val) | |
12177 | { | |
12178 | __u64 *array = (__u64 *)info; | |
12179 | ||
12180 | if (offset >= 0) | |
12181 | array[offset / sizeof(__u64)] = val; | |
12182 | } | |
12183 | ||
12184 | struct bpf_prog_info_linear * | |
12185 | bpf_program__get_prog_info_linear(int fd, __u64 arrays) | |
12186 | { | |
12187 | struct bpf_prog_info_linear *info_linear; | |
12188 | struct bpf_prog_info info = {}; | |
12189 | __u32 info_len = sizeof(info); | |
12190 | __u32 data_len = 0; | |
12191 | int i, err; | |
12192 | void *ptr; | |
12193 | ||
12194 | if (arrays >> BPF_PROG_INFO_LAST_ARRAY) | |
e9fc3ce9 | 12195 | return libbpf_err_ptr(-EINVAL); |
34be1646 SL |
12196 | |
12197 | /* step 1: get array dimensions */ | |
12198 | err = bpf_obj_get_info_by_fd(fd, &info, &info_len); | |
12199 | if (err) { | |
12200 | pr_debug("can't get prog info: %s", strerror(errno)); | |
e9fc3ce9 | 12201 | return libbpf_err_ptr(-EFAULT); |
34be1646 SL |
12202 | } |
12203 | ||
12204 | /* step 2: calculate total size of all arrays */ | |
12205 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
12206 | bool include_array = (arrays & (1UL << i)) > 0; | |
12207 | struct bpf_prog_info_array_desc *desc; | |
12208 | __u32 count, size; | |
12209 | ||
12210 | desc = bpf_prog_info_array_desc + i; | |
12211 | ||
12212 | /* kernel is too old to support this field */ | |
12213 | if (info_len < desc->array_offset + sizeof(__u32) || | |
12214 | info_len < desc->count_offset + sizeof(__u32) || | |
12215 | (desc->size_offset > 0 && info_len < desc->size_offset)) | |
12216 | include_array = false; | |
12217 | ||
12218 | if (!include_array) { | |
12219 | arrays &= ~(1UL << i); /* clear the bit */ | |
12220 | continue; | |
12221 | } | |
12222 | ||
12223 | count = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
12224 | size = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
12225 | ||
12226 | data_len += count * size; | |
12227 | } | |
12228 | ||
12229 | /* step 3: allocate continuous memory */ | |
12230 | data_len = roundup(data_len, sizeof(__u64)); | |
12231 | info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len); | |
12232 | if (!info_linear) | |
e9fc3ce9 | 12233 | return libbpf_err_ptr(-ENOMEM); |
34be1646 SL |
12234 | |
12235 | /* step 4: fill data to info_linear->info */ | |
12236 | info_linear->arrays = arrays; | |
12237 | memset(&info_linear->info, 0, sizeof(info)); | |
12238 | ptr = info_linear->data; | |
12239 | ||
12240 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
12241 | struct bpf_prog_info_array_desc *desc; | |
12242 | __u32 count, size; | |
12243 | ||
12244 | if ((arrays & (1UL << i)) == 0) | |
12245 | continue; | |
12246 | ||
12247 | desc = bpf_prog_info_array_desc + i; | |
12248 | count = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
12249 | size = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
12250 | bpf_prog_info_set_offset_u32(&info_linear->info, | |
12251 | desc->count_offset, count); | |
12252 | bpf_prog_info_set_offset_u32(&info_linear->info, | |
12253 | desc->size_offset, size); | |
12254 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
12255 | desc->array_offset, | |
12256 | ptr_to_u64(ptr)); | |
12257 | ptr += count * size; | |
12258 | } | |
12259 | ||
12260 | /* step 5: call syscall again to get required arrays */ | |
12261 | err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len); | |
12262 | if (err) { | |
12263 | pr_debug("can't get prog info: %s", strerror(errno)); | |
12264 | free(info_linear); | |
e9fc3ce9 | 12265 | return libbpf_err_ptr(-EFAULT); |
34be1646 SL |
12266 | } |
12267 | ||
12268 | /* step 6: verify the data */ | |
12269 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
12270 | struct bpf_prog_info_array_desc *desc; | |
12271 | __u32 v1, v2; | |
12272 | ||
12273 | if ((arrays & (1UL << i)) == 0) | |
12274 | continue; | |
12275 | ||
12276 | desc = bpf_prog_info_array_desc + i; | |
12277 | v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
12278 | v2 = bpf_prog_info_read_offset_u32(&info_linear->info, | |
12279 | desc->count_offset); | |
12280 | if (v1 != v2) | |
be18010e | 12281 | pr_warn("%s: mismatch in element count\n", __func__); |
34be1646 SL |
12282 | |
12283 | v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
12284 | v2 = bpf_prog_info_read_offset_u32(&info_linear->info, | |
12285 | desc->size_offset); | |
12286 | if (v1 != v2) | |
be18010e | 12287 | pr_warn("%s: mismatch in rec size\n", __func__); |
34be1646 SL |
12288 | } |
12289 | ||
12290 | /* step 7: update info_len and data_len */ | |
12291 | info_linear->info_len = sizeof(struct bpf_prog_info); | |
12292 | info_linear->data_len = data_len; | |
12293 | ||
12294 | return info_linear; | |
12295 | } | |
12296 | ||
12297 | void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear) | |
12298 | { | |
12299 | int i; | |
12300 | ||
12301 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
12302 | struct bpf_prog_info_array_desc *desc; | |
12303 | __u64 addr, offs; | |
12304 | ||
12305 | if ((info_linear->arrays & (1UL << i)) == 0) | |
12306 | continue; | |
12307 | ||
12308 | desc = bpf_prog_info_array_desc + i; | |
12309 | addr = bpf_prog_info_read_offset_u64(&info_linear->info, | |
12310 | desc->array_offset); | |
12311 | offs = addr - ptr_to_u64(info_linear->data); | |
12312 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
12313 | desc->array_offset, offs); | |
12314 | } | |
12315 | } | |
12316 | ||
12317 | void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear) | |
12318 | { | |
12319 | int i; | |
12320 | ||
12321 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
12322 | struct bpf_prog_info_array_desc *desc; | |
12323 | __u64 addr, offs; | |
12324 | ||
12325 | if ((info_linear->arrays & (1UL << i)) == 0) | |
12326 | continue; | |
12327 | ||
12328 | desc = bpf_prog_info_array_desc + i; | |
12329 | offs = bpf_prog_info_read_offset_u64(&info_linear->info, | |
12330 | desc->array_offset); | |
12331 | addr = offs + ptr_to_u64(info_linear->data); | |
12332 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
12333 | desc->array_offset, addr); | |
12334 | } | |
12335 | } | |
6446b315 | 12336 | |
ff26ce5c EC |
12337 | int bpf_program__set_attach_target(struct bpf_program *prog, |
12338 | int attach_prog_fd, | |
12339 | const char *attach_func_name) | |
12340 | { | |
fe62de31 | 12341 | int btf_obj_fd = 0, btf_id = 0, err; |
ff26ce5c | 12342 | |
2d5ec1c6 | 12343 | if (!prog || attach_prog_fd < 0) |
e9fc3ce9 | 12344 | return libbpf_err(-EINVAL); |
ff26ce5c | 12345 | |
fe62de31 | 12346 | if (prog->obj->loaded) |
e9fc3ce9 | 12347 | return libbpf_err(-EINVAL); |
fe62de31 | 12348 | |
2d5ec1c6 AN |
12349 | if (attach_prog_fd && !attach_func_name) { |
12350 | /* remember attach_prog_fd and let bpf_program__load() find | |
12351 | * BTF ID during the program load | |
12352 | */ | |
12353 | prog->attach_prog_fd = attach_prog_fd; | |
12354 | return 0; | |
12355 | } | |
12356 | ||
fe62de31 | 12357 | if (attach_prog_fd) { |
ff26ce5c EC |
12358 | btf_id = libbpf_find_prog_btf_id(attach_func_name, |
12359 | attach_prog_fd); | |
fe62de31 | 12360 | if (btf_id < 0) |
e9fc3ce9 | 12361 | return libbpf_err(btf_id); |
fe62de31 | 12362 | } else { |
2d5ec1c6 AN |
12363 | if (!attach_func_name) |
12364 | return libbpf_err(-EINVAL); | |
12365 | ||
fe62de31 AN |
12366 | /* load btf_vmlinux, if not yet */ |
12367 | err = bpf_object__load_vmlinux_btf(prog->obj, true); | |
12368 | if (err) | |
e9fc3ce9 | 12369 | return libbpf_err(err); |
fe62de31 AN |
12370 | err = find_kernel_btf_id(prog->obj, attach_func_name, |
12371 | prog->expected_attach_type, | |
12372 | &btf_obj_fd, &btf_id); | |
12373 | if (err) | |
e9fc3ce9 | 12374 | return libbpf_err(err); |
fe62de31 | 12375 | } |
ff26ce5c EC |
12376 | |
12377 | prog->attach_btf_id = btf_id; | |
fe62de31 | 12378 | prog->attach_btf_obj_fd = btf_obj_fd; |
ff26ce5c EC |
12379 | prog->attach_prog_fd = attach_prog_fd; |
12380 | return 0; | |
12381 | } | |
12382 | ||
6803ee25 | 12383 | int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz) |
6446b315 | 12384 | { |
6803ee25 AN |
12385 | int err = 0, n, len, start, end = -1; |
12386 | bool *tmp; | |
6446b315 | 12387 | |
6803ee25 AN |
12388 | *mask = NULL; |
12389 | *mask_sz = 0; | |
12390 | ||
12391 | /* Each sub string separated by ',' has format \d+-\d+ or \d+ */ | |
12392 | while (*s) { | |
12393 | if (*s == ',' || *s == '\n') { | |
12394 | s++; | |
12395 | continue; | |
12396 | } | |
12397 | n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len); | |
12398 | if (n <= 0 || n > 2) { | |
12399 | pr_warn("Failed to get CPU range %s: %d\n", s, n); | |
12400 | err = -EINVAL; | |
12401 | goto cleanup; | |
12402 | } else if (n == 1) { | |
12403 | end = start; | |
12404 | } | |
12405 | if (start < 0 || start > end) { | |
12406 | pr_warn("Invalid CPU range [%d,%d] in %s\n", | |
12407 | start, end, s); | |
12408 | err = -EINVAL; | |
12409 | goto cleanup; | |
12410 | } | |
12411 | tmp = realloc(*mask, end + 1); | |
12412 | if (!tmp) { | |
12413 | err = -ENOMEM; | |
12414 | goto cleanup; | |
12415 | } | |
12416 | *mask = tmp; | |
12417 | memset(tmp + *mask_sz, 0, start - *mask_sz); | |
12418 | memset(tmp + start, 1, end - start + 1); | |
12419 | *mask_sz = end + 1; | |
12420 | s += len; | |
12421 | } | |
12422 | if (!*mask_sz) { | |
12423 | pr_warn("Empty CPU range\n"); | |
12424 | return -EINVAL; | |
12425 | } | |
12426 | return 0; | |
12427 | cleanup: | |
12428 | free(*mask); | |
12429 | *mask = NULL; | |
12430 | return err; | |
12431 | } | |
12432 | ||
12433 | int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz) | |
12434 | { | |
12435 | int fd, err = 0, len; | |
12436 | char buf[128]; | |
6446b315 | 12437 | |
92274e24 | 12438 | fd = open(fcpu, O_RDONLY | O_CLOEXEC); |
6446b315 | 12439 | if (fd < 0) { |
6803ee25 AN |
12440 | err = -errno; |
12441 | pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err); | |
12442 | return err; | |
6446b315 HL |
12443 | } |
12444 | len = read(fd, buf, sizeof(buf)); | |
12445 | close(fd); | |
12446 | if (len <= 0) { | |
6803ee25 AN |
12447 | err = len ? -errno : -EINVAL; |
12448 | pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err); | |
12449 | return err; | |
6446b315 | 12450 | } |
6803ee25 AN |
12451 | if (len >= sizeof(buf)) { |
12452 | pr_warn("CPU mask is too big in file %s\n", fcpu); | |
12453 | return -E2BIG; | |
6446b315 HL |
12454 | } |
12455 | buf[len] = '\0'; | |
12456 | ||
6803ee25 AN |
12457 | return parse_cpu_mask_str(buf, mask, mask_sz); |
12458 | } | |
12459 | ||
12460 | int libbpf_num_possible_cpus(void) | |
12461 | { | |
12462 | static const char *fcpu = "/sys/devices/system/cpu/possible"; | |
12463 | static int cpus; | |
12464 | int err, n, i, tmp_cpus; | |
12465 | bool *mask; | |
12466 | ||
12467 | tmp_cpus = READ_ONCE(cpus); | |
12468 | if (tmp_cpus > 0) | |
12469 | return tmp_cpus; | |
12470 | ||
12471 | err = parse_cpu_mask_file(fcpu, &mask, &n); | |
12472 | if (err) | |
e9fc3ce9 | 12473 | return libbpf_err(err); |
6803ee25 AN |
12474 | |
12475 | tmp_cpus = 0; | |
12476 | for (i = 0; i < n; i++) { | |
12477 | if (mask[i]) | |
12478 | tmp_cpus++; | |
6446b315 | 12479 | } |
6803ee25 | 12480 | free(mask); |
56fbc241 TC |
12481 | |
12482 | WRITE_ONCE(cpus, tmp_cpus); | |
12483 | return tmp_cpus; | |
6446b315 | 12484 | } |
d66562fb | 12485 | |
430025e5 DK |
12486 | static int populate_skeleton_maps(const struct bpf_object *obj, |
12487 | struct bpf_map_skeleton *maps, | |
12488 | size_t map_cnt) | |
12489 | { | |
12490 | int i; | |
12491 | ||
12492 | for (i = 0; i < map_cnt; i++) { | |
12493 | struct bpf_map **map = maps[i].map; | |
12494 | const char *name = maps[i].name; | |
12495 | void **mmaped = maps[i].mmaped; | |
12496 | ||
12497 | *map = bpf_object__find_map_by_name(obj, name); | |
12498 | if (!*map) { | |
12499 | pr_warn("failed to find skeleton map '%s'\n", name); | |
12500 | return -ESRCH; | |
12501 | } | |
12502 | ||
12503 | /* externs shouldn't be pre-setup from user code */ | |
12504 | if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG) | |
12505 | *mmaped = (*map)->mmaped; | |
12506 | } | |
12507 | return 0; | |
12508 | } | |
12509 | ||
12510 | static int populate_skeleton_progs(const struct bpf_object *obj, | |
12511 | struct bpf_prog_skeleton *progs, | |
12512 | size_t prog_cnt) | |
12513 | { | |
12514 | int i; | |
12515 | ||
12516 | for (i = 0; i < prog_cnt; i++) { | |
12517 | struct bpf_program **prog = progs[i].prog; | |
12518 | const char *name = progs[i].name; | |
12519 | ||
12520 | *prog = bpf_object__find_program_by_name(obj, name); | |
12521 | if (!*prog) { | |
12522 | pr_warn("failed to find skeleton program '%s'\n", name); | |
12523 | return -ESRCH; | |
12524 | } | |
12525 | } | |
12526 | return 0; | |
12527 | } | |
12528 | ||
d66562fb AN |
12529 | int bpf_object__open_skeleton(struct bpf_object_skeleton *s, |
12530 | const struct bpf_object_open_opts *opts) | |
12531 | { | |
12532 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts, | |
12533 | .object_name = s->name, | |
12534 | ); | |
12535 | struct bpf_object *obj; | |
430025e5 | 12536 | int err; |
d66562fb AN |
12537 | |
12538 | /* Attempt to preserve opts->object_name, unless overriden by user | |
12539 | * explicitly. Overwriting object name for skeletons is discouraged, | |
12540 | * as it breaks global data maps, because they contain object name | |
12541 | * prefix as their own map name prefix. When skeleton is generated, | |
12542 | * bpftool is making an assumption that this name will stay the same. | |
12543 | */ | |
12544 | if (opts) { | |
12545 | memcpy(&skel_opts, opts, sizeof(*opts)); | |
12546 | if (!opts->object_name) | |
12547 | skel_opts.object_name = s->name; | |
12548 | } | |
12549 | ||
12550 | obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts); | |
e9fc3ce9 AN |
12551 | err = libbpf_get_error(obj); |
12552 | if (err) { | |
12553 | pr_warn("failed to initialize skeleton BPF object '%s': %d\n", | |
12554 | s->name, err); | |
12555 | return libbpf_err(err); | |
d66562fb AN |
12556 | } |
12557 | ||
12558 | *s->obj = obj; | |
430025e5 DK |
12559 | err = populate_skeleton_maps(obj, s->maps, s->map_cnt); |
12560 | if (err) { | |
12561 | pr_warn("failed to populate skeleton maps for '%s': %d\n", s->name, err); | |
12562 | return libbpf_err(err); | |
12563 | } | |
12564 | ||
12565 | err = populate_skeleton_progs(obj, s->progs, s->prog_cnt); | |
12566 | if (err) { | |
12567 | pr_warn("failed to populate skeleton progs for '%s': %d\n", s->name, err); | |
12568 | return libbpf_err(err); | |
12569 | } | |
d66562fb | 12570 | |
430025e5 DK |
12571 | return 0; |
12572 | } | |
d66562fb | 12573 | |
430025e5 DK |
12574 | int bpf_object__open_subskeleton(struct bpf_object_subskeleton *s) |
12575 | { | |
12576 | int err, len, var_idx, i; | |
12577 | const char *var_name; | |
12578 | const struct bpf_map *map; | |
12579 | struct btf *btf; | |
12580 | __u32 map_type_id; | |
12581 | const struct btf_type *map_type, *var_type; | |
12582 | const struct bpf_var_skeleton *var_skel; | |
12583 | struct btf_var_secinfo *var; | |
d66562fb | 12584 | |
430025e5 DK |
12585 | if (!s->obj) |
12586 | return libbpf_err(-EINVAL); | |
12587 | ||
12588 | btf = bpf_object__btf(s->obj); | |
12589 | if (!btf) { | |
12590 | pr_warn("subskeletons require BTF at runtime (object %s)\n", | |
12591 | bpf_object__name(s->obj)); | |
12592 | return libbpf_err(-errno); | |
d66562fb AN |
12593 | } |
12594 | ||
430025e5 DK |
12595 | err = populate_skeleton_maps(s->obj, s->maps, s->map_cnt); |
12596 | if (err) { | |
12597 | pr_warn("failed to populate subskeleton maps: %d\n", err); | |
12598 | return libbpf_err(err); | |
12599 | } | |
d66562fb | 12600 | |
430025e5 DK |
12601 | err = populate_skeleton_progs(s->obj, s->progs, s->prog_cnt); |
12602 | if (err) { | |
12603 | pr_warn("failed to populate subskeleton maps: %d\n", err); | |
12604 | return libbpf_err(err); | |
d66562fb AN |
12605 | } |
12606 | ||
430025e5 DK |
12607 | for (var_idx = 0; var_idx < s->var_cnt; var_idx++) { |
12608 | var_skel = &s->vars[var_idx]; | |
12609 | map = *var_skel->map; | |
12610 | map_type_id = bpf_map__btf_value_type_id(map); | |
12611 | map_type = btf__type_by_id(btf, map_type_id); | |
12612 | ||
12613 | if (!btf_is_datasec(map_type)) { | |
12614 | pr_warn("type for map '%1$s' is not a datasec: %2$s", | |
12615 | bpf_map__name(map), | |
12616 | __btf_kind_str(btf_kind(map_type))); | |
12617 | return libbpf_err(-EINVAL); | |
12618 | } | |
12619 | ||
12620 | len = btf_vlen(map_type); | |
12621 | var = btf_var_secinfos(map_type); | |
12622 | for (i = 0; i < len; i++, var++) { | |
12623 | var_type = btf__type_by_id(btf, var->type); | |
12624 | var_name = btf__name_by_offset(btf, var_type->name_off); | |
12625 | if (strcmp(var_name, var_skel->name) == 0) { | |
12626 | *var_skel->addr = map->mmaped + var->offset; | |
12627 | break; | |
12628 | } | |
12629 | } | |
12630 | } | |
d66562fb AN |
12631 | return 0; |
12632 | } | |
12633 | ||
430025e5 DK |
12634 | void bpf_object__destroy_subskeleton(struct bpf_object_subskeleton *s) |
12635 | { | |
12636 | if (!s) | |
12637 | return; | |
12638 | free(s->maps); | |
12639 | free(s->progs); | |
12640 | free(s->vars); | |
12641 | free(s); | |
12642 | } | |
12643 | ||
d66562fb AN |
12644 | int bpf_object__load_skeleton(struct bpf_object_skeleton *s) |
12645 | { | |
12646 | int i, err; | |
12647 | ||
12648 | err = bpf_object__load(*s->obj); | |
12649 | if (err) { | |
12650 | pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err); | |
e9fc3ce9 | 12651 | return libbpf_err(err); |
d66562fb AN |
12652 | } |
12653 | ||
12654 | for (i = 0; i < s->map_cnt; i++) { | |
12655 | struct bpf_map *map = *s->maps[i].map; | |
12656 | size_t mmap_sz = bpf_map_mmap_sz(map); | |
12657 | int prot, map_fd = bpf_map__fd(map); | |
12658 | void **mmaped = s->maps[i].mmaped; | |
d66562fb AN |
12659 | |
12660 | if (!mmaped) | |
12661 | continue; | |
12662 | ||
12663 | if (!(map->def.map_flags & BPF_F_MMAPABLE)) { | |
12664 | *mmaped = NULL; | |
12665 | continue; | |
12666 | } | |
12667 | ||
12668 | if (map->def.map_flags & BPF_F_RDONLY_PROG) | |
12669 | prot = PROT_READ; | |
12670 | else | |
12671 | prot = PROT_READ | PROT_WRITE; | |
12672 | ||
12673 | /* Remap anonymous mmap()-ed "map initialization image" as | |
12674 | * a BPF map-backed mmap()-ed memory, but preserving the same | |
12675 | * memory address. This will cause kernel to change process' | |
12676 | * page table to point to a different piece of kernel memory, | |
12677 | * but from userspace point of view memory address (and its | |
12678 | * contents, being identical at this point) will stay the | |
12679 | * same. This mapping will be released by bpf_object__close() | |
12680 | * as per normal clean up procedure, so we don't need to worry | |
12681 | * about it from skeleton's clean up perspective. | |
12682 | */ | |
2ad97d47 AN |
12683 | *mmaped = mmap(map->mmaped, mmap_sz, prot, |
12684 | MAP_SHARED | MAP_FIXED, map_fd, 0); | |
12685 | if (*mmaped == MAP_FAILED) { | |
d66562fb AN |
12686 | err = -errno; |
12687 | *mmaped = NULL; | |
12688 | pr_warn("failed to re-mmap() map '%s': %d\n", | |
12689 | bpf_map__name(map), err); | |
e9fc3ce9 | 12690 | return libbpf_err(err); |
d66562fb AN |
12691 | } |
12692 | } | |
12693 | ||
12694 | return 0; | |
12695 | } | |
12696 | ||
12697 | int bpf_object__attach_skeleton(struct bpf_object_skeleton *s) | |
12698 | { | |
e9fc3ce9 | 12699 | int i, err; |
d66562fb AN |
12700 | |
12701 | for (i = 0; i < s->prog_cnt; i++) { | |
12702 | struct bpf_program *prog = *s->progs[i].prog; | |
12703 | struct bpf_link **link = s->progs[i].link; | |
d66562fb | 12704 | |
a3820c48 | 12705 | if (!prog->autoload) |
d9297581 AN |
12706 | continue; |
12707 | ||
5532dfd4 | 12708 | /* auto-attaching not supported for this program */ |
4fa5bcfe | 12709 | if (!prog->sec_def || !prog->sec_def->prog_attach_fn) |
d66562fb AN |
12710 | continue; |
12711 | ||
4fa5bcfe AN |
12712 | /* if user already set the link manually, don't attempt auto-attach */ |
12713 | if (*link) | |
12714 | continue; | |
12715 | ||
12716 | err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link); | |
e9fc3ce9 | 12717 | if (err) { |
4fa5bcfe | 12718 | pr_warn("prog '%s': failed to auto-attach: %d\n", |
e9fc3ce9 AN |
12719 | bpf_program__name(prog), err); |
12720 | return libbpf_err(err); | |
d66562fb | 12721 | } |
4fa5bcfe AN |
12722 | |
12723 | /* It's possible that for some SEC() definitions auto-attach | |
12724 | * is supported in some cases (e.g., if definition completely | |
12725 | * specifies target information), but is not in other cases. | |
12726 | * SEC("uprobe") is one such case. If user specified target | |
12727 | * binary and function name, such BPF program can be | |
12728 | * auto-attached. But if not, it shouldn't trigger skeleton's | |
12729 | * attach to fail. It should just be skipped. | |
12730 | * attach_fn signals such case with returning 0 (no error) and | |
12731 | * setting link to NULL. | |
12732 | */ | |
d66562fb AN |
12733 | } |
12734 | ||
12735 | return 0; | |
12736 | } | |
12737 | ||
12738 | void bpf_object__detach_skeleton(struct bpf_object_skeleton *s) | |
12739 | { | |
12740 | int i; | |
12741 | ||
12742 | for (i = 0; i < s->prog_cnt; i++) { | |
12743 | struct bpf_link **link = s->progs[i].link; | |
12744 | ||
50450fc7 | 12745 | bpf_link__destroy(*link); |
d66562fb AN |
12746 | *link = NULL; |
12747 | } | |
12748 | } | |
12749 | ||
12750 | void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s) | |
12751 | { | |
a32ea51a YS |
12752 | if (!s) |
12753 | return; | |
12754 | ||
d66562fb AN |
12755 | if (s->progs) |
12756 | bpf_object__detach_skeleton(s); | |
12757 | if (s->obj) | |
12758 | bpf_object__close(*s->obj); | |
12759 | free(s->maps); | |
12760 | free(s->progs); | |
12761 | free(s); | |
12762 | } |