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 | |
c1122392 | 305 | bool load; |
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 | ||
1a5e3fb1 | 486 | struct bpf_object { |
d859900c | 487 | char name[BPF_OBJ_NAME_LEN]; |
cb1e5e96 | 488 | char license[64]; |
438363c0 | 489 | __u32 kern_version; |
0b3d1efa | 490 | |
a5b8bd47 WN |
491 | struct bpf_program *programs; |
492 | size_t nr_programs; | |
9d759a9b WN |
493 | struct bpf_map *maps; |
494 | size_t nr_maps; | |
bf829271 | 495 | size_t maps_cap; |
9d759a9b | 496 | |
8601fd42 | 497 | char *kconfig; |
166750bc AN |
498 | struct extern_desc *externs; |
499 | int nr_extern; | |
81bfdd08 | 500 | int kconfig_map_idx; |
166750bc | 501 | |
52d3352e | 502 | bool loaded; |
c3c55696 | 503 | bool has_subcalls; |
25bbbd7a | 504 | bool has_rodata; |
a5b8bd47 | 505 | |
e2fa0156 AS |
506 | struct bpf_gen *gen_loader; |
507 | ||
29a30ff5 AN |
508 | /* Information when doing ELF related work. Only valid if efile.elf is not NULL */ |
509 | struct elf_state efile; | |
1a5e3fb1 | 510 | /* |
29a30ff5 | 511 | * All loaded bpf_object are linked in a list, which is |
9a208eff WN |
512 | * hidden to caller. bpf_objects__<func> handlers deal with |
513 | * all objects. | |
514 | */ | |
515 | struct list_head list; | |
10931d24 | 516 | |
8a138aed | 517 | struct btf *btf; |
0f7515ca AN |
518 | struct btf_ext *btf_ext; |
519 | ||
a6ed02ca KS |
520 | /* Parse and load BTF vmlinux if any of the programs in the object need |
521 | * it at load time. | |
522 | */ | |
523 | struct btf *btf_vmlinux; | |
1373ff59 SC |
524 | /* Path to the custom BTF to be used for BPF CO-RE relocations as an |
525 | * override for vmlinux BTF. | |
526 | */ | |
527 | char *btf_custom_path; | |
0f7515ca AN |
528 | /* vmlinux BTF override for CO-RE relocations */ |
529 | struct btf *btf_vmlinux_override; | |
4f33a53d AN |
530 | /* Lazily initialized kernel module BTFs */ |
531 | struct module_btf *btf_modules; | |
532 | bool btf_modules_loaded; | |
533 | size_t btf_module_cnt; | |
534 | size_t btf_module_cap; | |
8a138aed | 535 | |
e0e3ea88 AN |
536 | /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */ |
537 | char *log_buf; | |
538 | size_t log_size; | |
539 | __u32 log_level; | |
540 | ||
10931d24 WN |
541 | void *priv; |
542 | bpf_object_clear_priv_t clear_priv; | |
543 | ||
9dbe6015 KKD |
544 | int *fd_array; |
545 | size_t fd_array_cap; | |
546 | size_t fd_array_cnt; | |
547 | ||
1a5e3fb1 WN |
548 | char path[]; |
549 | }; | |
1a5e3fb1 | 550 | |
88a82120 AN |
551 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off); |
552 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off); | |
553 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx); | |
554 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name); | |
ad23b723 | 555 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn); |
88a82120 AN |
556 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn); |
557 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn); | |
ad23b723 AN |
558 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx); |
559 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx); | |
88a82120 | 560 | |
29cd77f4 | 561 | void bpf_program__unload(struct bpf_program *prog) |
55cffde2 | 562 | { |
b580563e WN |
563 | int i; |
564 | ||
55cffde2 WN |
565 | if (!prog) |
566 | return; | |
567 | ||
b580563e WN |
568 | /* |
569 | * If the object is opened but the program was never loaded, | |
570 | * it is possible that prog->instances.nr == -1. | |
571 | */ | |
572 | if (prog->instances.nr > 0) { | |
573 | for (i = 0; i < prog->instances.nr; i++) | |
574 | zclose(prog->instances.fds[i]); | |
575 | } else if (prog->instances.nr != -1) { | |
be18010e KW |
576 | pr_warn("Internal error: instances.nr is %d\n", |
577 | prog->instances.nr); | |
b580563e WN |
578 | } |
579 | ||
580 | prog->instances.nr = -1; | |
581 | zfree(&prog->instances.fds); | |
2993e051 | 582 | |
2993e051 | 583 | zfree(&prog->func_info); |
07a09d1b | 584 | zfree(&prog->line_info); |
55cffde2 WN |
585 | } |
586 | ||
a5b8bd47 WN |
587 | static void bpf_program__exit(struct bpf_program *prog) |
588 | { | |
589 | if (!prog) | |
590 | return; | |
591 | ||
aa9b1ac3 WN |
592 | if (prog->clear_priv) |
593 | prog->clear_priv(prog, prog->priv); | |
594 | ||
595 | prog->priv = NULL; | |
596 | prog->clear_priv = NULL; | |
597 | ||
55cffde2 | 598 | bpf_program__unload(prog); |
88cda1c9 | 599 | zfree(&prog->name); |
52109584 | 600 | zfree(&prog->sec_name); |
33a2c75c | 601 | zfree(&prog->pin_name); |
a5b8bd47 | 602 | zfree(&prog->insns); |
34090915 WN |
603 | zfree(&prog->reloc_desc); |
604 | ||
605 | prog->nr_reloc = 0; | |
a5b8bd47 | 606 | prog->insns_cnt = 0; |
c1122392 | 607 | prog->sec_idx = -1; |
a5b8bd47 WN |
608 | } |
609 | ||
33a2c75c SF |
610 | static char *__bpf_program__pin_name(struct bpf_program *prog) |
611 | { | |
612 | char *name, *p; | |
613 | ||
a77f879b SF |
614 | if (libbpf_mode & LIBBPF_STRICT_SEC_NAME) |
615 | name = strdup(prog->name); | |
616 | else | |
617 | name = strdup(prog->sec_name); | |
618 | ||
619 | if (!name) | |
620 | return NULL; | |
621 | ||
622 | p = name; | |
623 | ||
33a2c75c SF |
624 | while ((p = strchr(p, '/'))) |
625 | *p = '_'; | |
626 | ||
627 | return name; | |
628 | } | |
629 | ||
c3c55696 AN |
630 | static bool insn_is_subprog_call(const struct bpf_insn *insn) |
631 | { | |
632 | return BPF_CLASS(insn->code) == BPF_JMP && | |
633 | BPF_OP(insn->code) == BPF_CALL && | |
634 | BPF_SRC(insn->code) == BPF_K && | |
635 | insn->src_reg == BPF_PSEUDO_CALL && | |
636 | insn->dst_reg == 0 && | |
637 | insn->off == 0; | |
638 | } | |
639 | ||
aa0b8d43 MKL |
640 | static bool is_call_insn(const struct bpf_insn *insn) |
641 | { | |
642 | return insn->code == (BPF_JMP | BPF_CALL); | |
643 | } | |
644 | ||
53eddb5e YS |
645 | static bool insn_is_pseudo_func(struct bpf_insn *insn) |
646 | { | |
aa0b8d43 | 647 | return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC; |
53eddb5e YS |
648 | } |
649 | ||
a5b8bd47 | 650 | static int |
c3c55696 AN |
651 | bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog, |
652 | const char *name, size_t sec_idx, const char *sec_name, | |
653 | size_t sec_off, void *insn_data, size_t insn_data_sz) | |
a5b8bd47 | 654 | { |
c1122392 AN |
655 | if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) { |
656 | pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n", | |
657 | sec_name, name, sec_off, insn_data_sz); | |
a5b8bd47 WN |
658 | return -EINVAL; |
659 | } | |
660 | ||
1ad9cbb8 | 661 | memset(prog, 0, sizeof(*prog)); |
c3c55696 AN |
662 | prog->obj = obj; |
663 | ||
c1122392 AN |
664 | prog->sec_idx = sec_idx; |
665 | prog->sec_insn_off = sec_off / BPF_INSN_SZ; | |
666 | prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ; | |
667 | /* insns_cnt can later be increased by appending used subprograms */ | |
668 | prog->insns_cnt = prog->sec_insn_cnt; | |
a5b8bd47 | 669 | |
c1122392 AN |
670 | prog->type = BPF_PROG_TYPE_UNSPEC; |
671 | prog->load = true; | |
a5b8bd47 | 672 | |
c1122392 AN |
673 | prog->instances.fds = NULL; |
674 | prog->instances.nr = -1; | |
675 | ||
b3ce9079 AN |
676 | /* inherit object's log_level */ |
677 | prog->log_level = obj->log_level; | |
678 | ||
52109584 AN |
679 | prog->sec_name = strdup(sec_name); |
680 | if (!prog->sec_name) | |
c1122392 AN |
681 | goto errout; |
682 | ||
683 | prog->name = strdup(name); | |
684 | if (!prog->name) | |
a5b8bd47 | 685 | goto errout; |
a5b8bd47 | 686 | |
33a2c75c | 687 | prog->pin_name = __bpf_program__pin_name(prog); |
c1122392 | 688 | if (!prog->pin_name) |
33a2c75c | 689 | goto errout; |
33a2c75c | 690 | |
c1122392 AN |
691 | prog->insns = malloc(insn_data_sz); |
692 | if (!prog->insns) | |
a5b8bd47 | 693 | goto errout; |
c1122392 | 694 | memcpy(prog->insns, insn_data, insn_data_sz); |
a5b8bd47 WN |
695 | |
696 | return 0; | |
697 | errout: | |
c1122392 | 698 | pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name); |
a5b8bd47 WN |
699 | bpf_program__exit(prog); |
700 | return -ENOMEM; | |
701 | } | |
702 | ||
703 | static int | |
c1122392 AN |
704 | bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data, |
705 | const char *sec_name, int sec_idx) | |
a5b8bd47 | 706 | { |
6245947c | 707 | Elf_Data *symbols = obj->efile.symbols; |
c1122392 AN |
708 | struct bpf_program *prog, *progs; |
709 | void *data = sec_data->d_buf; | |
6245947c AN |
710 | size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms; |
711 | int nr_progs, err, i; | |
c1122392 | 712 | const char *name; |
ad23b723 | 713 | Elf64_Sym *sym; |
a5b8bd47 WN |
714 | |
715 | progs = obj->programs; | |
716 | nr_progs = obj->nr_programs; | |
ad23b723 | 717 | nr_syms = symbols->d_size / sizeof(Elf64_Sym); |
c1122392 | 718 | sec_off = 0; |
a5b8bd47 | 719 | |
6245947c | 720 | for (i = 0; i < nr_syms; i++) { |
ad23b723 AN |
721 | sym = elf_sym_by_idx(obj, i); |
722 | ||
723 | if (sym->st_shndx != sec_idx) | |
6245947c | 724 | continue; |
ad23b723 | 725 | if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC) |
6245947c | 726 | continue; |
88cda1c9 | 727 | |
ad23b723 AN |
728 | prog_sz = sym->st_size; |
729 | sec_off = sym->st_value; | |
88cda1c9 | 730 | |
ad23b723 | 731 | name = elf_sym_str(obj, sym->st_name); |
c1122392 AN |
732 | if (!name) { |
733 | pr_warn("sec '%s': failed to get symbol name for offset %zu\n", | |
734 | sec_name, sec_off); | |
735 | return -LIBBPF_ERRNO__FORMAT; | |
736 | } | |
88cda1c9 | 737 | |
c1122392 AN |
738 | if (sec_off + prog_sz > sec_sz) { |
739 | pr_warn("sec '%s': program at offset %zu crosses section boundary\n", | |
740 | sec_name, sec_off); | |
741 | return -LIBBPF_ERRNO__FORMAT; | |
742 | } | |
88cda1c9 | 743 | |
ad23b723 | 744 | if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) { |
513f485c AN |
745 | pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name); |
746 | return -ENOTSUP; | |
747 | } | |
748 | ||
c3c55696 AN |
749 | pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n", |
750 | sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz); | |
88cda1c9 | 751 | |
c3c55696 | 752 | progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs)); |
c1122392 AN |
753 | if (!progs) { |
754 | /* | |
755 | * In this case the original obj->programs | |
756 | * is still valid, so don't need special treat for | |
757 | * bpf_close_object(). | |
758 | */ | |
759 | pr_warn("sec '%s': failed to alloc memory for new program '%s'\n", | |
760 | sec_name, name); | |
761 | return -ENOMEM; | |
88cda1c9 | 762 | } |
c1122392 | 763 | obj->programs = progs; |
88cda1c9 | 764 | |
c1122392 | 765 | prog = &progs[nr_progs]; |
9a94f277 | 766 | |
c3c55696 AN |
767 | err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name, |
768 | sec_off, data + sec_off, prog_sz); | |
c1122392 AN |
769 | if (err) |
770 | return err; | |
9a94f277 | 771 | |
e5670fa0 AN |
772 | /* if function is a global/weak symbol, but has restricted |
773 | * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC | |
774 | * as static to enable more permissive BPF verification mode | |
775 | * with more outside context available to BPF verifier | |
aea28a60 | 776 | */ |
ad23b723 AN |
777 | if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL |
778 | && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN | |
779 | || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL)) | |
aea28a60 AN |
780 | prog->mark_btf_static = true; |
781 | ||
c1122392 AN |
782 | nr_progs++; |
783 | obj->nr_programs = nr_progs; | |
88cda1c9 MKL |
784 | } |
785 | ||
786 | return 0; | |
787 | } | |
788 | ||
5b3d7298 | 789 | __u32 get_kernel_version(void) |
5e61f270 | 790 | { |
5b3d7298 AN |
791 | /* On Ubuntu LINUX_VERSION_CODE doesn't correspond to info.release, |
792 | * but Ubuntu provides /proc/version_signature file, as described at | |
793 | * https://ubuntu.com/kernel, with an example contents below, which we | |
794 | * can use to get a proper LINUX_VERSION_CODE. | |
795 | * | |
796 | * Ubuntu 5.4.0-12.15-generic 5.4.8 | |
797 | * | |
798 | * In the above, 5.4.8 is what kernel is actually expecting, while | |
799 | * uname() call will return 5.4.0 in info.release. | |
800 | */ | |
801 | const char *ubuntu_kver_file = "/proc/version_signature"; | |
5e61f270 AN |
802 | __u32 major, minor, patch; |
803 | struct utsname info; | |
804 | ||
5b3d7298 AN |
805 | if (access(ubuntu_kver_file, R_OK) == 0) { |
806 | FILE *f; | |
807 | ||
808 | f = fopen(ubuntu_kver_file, "r"); | |
809 | if (f) { | |
810 | if (fscanf(f, "%*s %*s %d.%d.%d\n", &major, &minor, &patch) == 3) { | |
811 | fclose(f); | |
812 | return KERNEL_VERSION(major, minor, patch); | |
813 | } | |
814 | fclose(f); | |
815 | } | |
816 | /* something went wrong, fall back to uname() approach */ | |
817 | } | |
818 | ||
5e61f270 AN |
819 | uname(&info); |
820 | if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3) | |
821 | return 0; | |
822 | return KERNEL_VERSION(major, minor, patch); | |
823 | } | |
824 | ||
590a0088 MKL |
825 | static const struct btf_member * |
826 | find_member_by_offset(const struct btf_type *t, __u32 bit_offset) | |
827 | { | |
828 | struct btf_member *m; | |
829 | int i; | |
830 | ||
831 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
832 | if (btf_member_bit_offset(t, i) == bit_offset) | |
833 | return m; | |
834 | } | |
835 | ||
836 | return NULL; | |
837 | } | |
838 | ||
839 | static const struct btf_member * | |
840 | find_member_by_name(const struct btf *btf, const struct btf_type *t, | |
841 | const char *name) | |
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 (!strcmp(btf__name_by_offset(btf, m->name_off), name)) | |
848 | return m; | |
849 | } | |
850 | ||
851 | return NULL; | |
852 | } | |
853 | ||
854 | #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_" | |
a6ed02ca KS |
855 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
856 | const char *name, __u32 kind); | |
590a0088 MKL |
857 | |
858 | static int | |
859 | find_struct_ops_kern_types(const struct btf *btf, const char *tname, | |
860 | const struct btf_type **type, __u32 *type_id, | |
861 | const struct btf_type **vtype, __u32 *vtype_id, | |
862 | const struct btf_member **data_member) | |
863 | { | |
864 | const struct btf_type *kern_type, *kern_vtype; | |
865 | const struct btf_member *kern_data_member; | |
866 | __s32 kern_vtype_id, kern_type_id; | |
590a0088 MKL |
867 | __u32 i; |
868 | ||
869 | kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT); | |
870 | if (kern_type_id < 0) { | |
871 | pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n", | |
872 | tname); | |
873 | return kern_type_id; | |
874 | } | |
875 | kern_type = btf__type_by_id(btf, kern_type_id); | |
876 | ||
877 | /* Find the corresponding "map_value" type that will be used | |
878 | * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example, | |
879 | * find "struct bpf_struct_ops_tcp_congestion_ops" from the | |
880 | * btf_vmlinux. | |
881 | */ | |
a6ed02ca KS |
882 | kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX, |
883 | tname, BTF_KIND_STRUCT); | |
590a0088 | 884 | if (kern_vtype_id < 0) { |
a6ed02ca KS |
885 | pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n", |
886 | STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
887 | return kern_vtype_id; |
888 | } | |
889 | kern_vtype = btf__type_by_id(btf, kern_vtype_id); | |
890 | ||
891 | /* Find "struct tcp_congestion_ops" from | |
892 | * struct bpf_struct_ops_tcp_congestion_ops { | |
893 | * [ ... ] | |
894 | * struct tcp_congestion_ops data; | |
895 | * } | |
896 | */ | |
897 | kern_data_member = btf_members(kern_vtype); | |
898 | for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) { | |
899 | if (kern_data_member->type == kern_type_id) | |
900 | break; | |
901 | } | |
902 | if (i == btf_vlen(kern_vtype)) { | |
a6ed02ca KS |
903 | pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n", |
904 | tname, STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
905 | return -EINVAL; |
906 | } | |
907 | ||
908 | *type = kern_type; | |
909 | *type_id = kern_type_id; | |
910 | *vtype = kern_vtype; | |
911 | *vtype_id = kern_vtype_id; | |
912 | *data_member = kern_data_member; | |
913 | ||
914 | return 0; | |
915 | } | |
916 | ||
917 | static bool bpf_map__is_struct_ops(const struct bpf_map *map) | |
918 | { | |
919 | return map->def.type == BPF_MAP_TYPE_STRUCT_OPS; | |
920 | } | |
921 | ||
922 | /* Init the map's fields that depend on kern_btf */ | |
923 | static int bpf_map__init_kern_struct_ops(struct bpf_map *map, | |
924 | const struct btf *btf, | |
925 | const struct btf *kern_btf) | |
926 | { | |
927 | const struct btf_member *member, *kern_member, *kern_data_member; | |
928 | const struct btf_type *type, *kern_type, *kern_vtype; | |
929 | __u32 i, kern_type_id, kern_vtype_id, kern_data_off; | |
930 | struct bpf_struct_ops *st_ops; | |
931 | void *data, *kern_data; | |
932 | const char *tname; | |
933 | int err; | |
934 | ||
935 | st_ops = map->st_ops; | |
936 | type = st_ops->type; | |
937 | tname = st_ops->tname; | |
938 | err = find_struct_ops_kern_types(kern_btf, tname, | |
939 | &kern_type, &kern_type_id, | |
940 | &kern_vtype, &kern_vtype_id, | |
941 | &kern_data_member); | |
942 | if (err) | |
943 | return err; | |
944 | ||
945 | pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n", | |
946 | map->name, st_ops->type_id, kern_type_id, kern_vtype_id); | |
947 | ||
948 | map->def.value_size = kern_vtype->size; | |
949 | map->btf_vmlinux_value_type_id = kern_vtype_id; | |
950 | ||
951 | st_ops->kern_vdata = calloc(1, kern_vtype->size); | |
952 | if (!st_ops->kern_vdata) | |
953 | return -ENOMEM; | |
954 | ||
955 | data = st_ops->data; | |
956 | kern_data_off = kern_data_member->offset / 8; | |
957 | kern_data = st_ops->kern_vdata + kern_data_off; | |
958 | ||
959 | member = btf_members(type); | |
960 | for (i = 0; i < btf_vlen(type); i++, member++) { | |
961 | const struct btf_type *mtype, *kern_mtype; | |
962 | __u32 mtype_id, kern_mtype_id; | |
963 | void *mdata, *kern_mdata; | |
964 | __s64 msize, kern_msize; | |
965 | __u32 moff, kern_moff; | |
966 | __u32 kern_member_idx; | |
967 | const char *mname; | |
968 | ||
969 | mname = btf__name_by_offset(btf, member->name_off); | |
970 | kern_member = find_member_by_name(kern_btf, kern_type, mname); | |
971 | if (!kern_member) { | |
972 | pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n", | |
973 | map->name, mname); | |
974 | return -ENOTSUP; | |
975 | } | |
976 | ||
977 | kern_member_idx = kern_member - btf_members(kern_type); | |
978 | if (btf_member_bitfield_size(type, i) || | |
979 | btf_member_bitfield_size(kern_type, kern_member_idx)) { | |
980 | pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n", | |
981 | map->name, mname); | |
982 | return -ENOTSUP; | |
983 | } | |
984 | ||
985 | moff = member->offset / 8; | |
986 | kern_moff = kern_member->offset / 8; | |
987 | ||
988 | mdata = data + moff; | |
989 | kern_mdata = kern_data + kern_moff; | |
990 | ||
991 | mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id); | |
992 | kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type, | |
993 | &kern_mtype_id); | |
994 | if (BTF_INFO_KIND(mtype->info) != | |
995 | BTF_INFO_KIND(kern_mtype->info)) { | |
996 | pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n", | |
997 | map->name, mname, BTF_INFO_KIND(mtype->info), | |
998 | BTF_INFO_KIND(kern_mtype->info)); | |
999 | return -ENOTSUP; | |
1000 | } | |
1001 | ||
1002 | if (btf_is_ptr(mtype)) { | |
1003 | struct bpf_program *prog; | |
1004 | ||
d2836ddd MKL |
1005 | prog = st_ops->progs[i]; |
1006 | if (!prog) | |
1007 | continue; | |
1008 | ||
590a0088 MKL |
1009 | kern_mtype = skip_mods_and_typedefs(kern_btf, |
1010 | kern_mtype->type, | |
1011 | &kern_mtype_id); | |
d2836ddd MKL |
1012 | |
1013 | /* mtype->type must be a func_proto which was | |
1014 | * guaranteed in bpf_object__collect_st_ops_relos(), | |
1015 | * so only check kern_mtype for func_proto here. | |
1016 | */ | |
1017 | if (!btf_is_func_proto(kern_mtype)) { | |
1018 | pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n", | |
590a0088 MKL |
1019 | map->name, mname); |
1020 | return -ENOTSUP; | |
1021 | } | |
1022 | ||
590a0088 MKL |
1023 | prog->attach_btf_id = kern_type_id; |
1024 | prog->expected_attach_type = kern_member_idx; | |
1025 | ||
1026 | st_ops->kern_func_off[i] = kern_data_off + kern_moff; | |
1027 | ||
1028 | pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n", | |
1029 | map->name, mname, prog->name, moff, | |
1030 | kern_moff); | |
1031 | ||
1032 | continue; | |
1033 | } | |
1034 | ||
1035 | msize = btf__resolve_size(btf, mtype_id); | |
1036 | kern_msize = btf__resolve_size(kern_btf, kern_mtype_id); | |
1037 | if (msize < 0 || kern_msize < 0 || msize != kern_msize) { | |
1038 | pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n", | |
1039 | map->name, mname, (ssize_t)msize, | |
1040 | (ssize_t)kern_msize); | |
1041 | return -ENOTSUP; | |
1042 | } | |
1043 | ||
1044 | pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n", | |
1045 | map->name, mname, (unsigned int)msize, | |
1046 | moff, kern_moff); | |
1047 | memcpy(kern_mdata, mdata, msize); | |
1048 | } | |
1049 | ||
1050 | return 0; | |
1051 | } | |
1052 | ||
1053 | static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj) | |
1054 | { | |
590a0088 MKL |
1055 | struct bpf_map *map; |
1056 | size_t i; | |
1057 | int err; | |
1058 | ||
1059 | for (i = 0; i < obj->nr_maps; i++) { | |
1060 | map = &obj->maps[i]; | |
1061 | ||
1062 | if (!bpf_map__is_struct_ops(map)) | |
1063 | continue; | |
1064 | ||
a6ed02ca KS |
1065 | err = bpf_map__init_kern_struct_ops(map, obj->btf, |
1066 | obj->btf_vmlinux); | |
1067 | if (err) | |
590a0088 | 1068 | return err; |
590a0088 MKL |
1069 | } |
1070 | ||
590a0088 MKL |
1071 | return 0; |
1072 | } | |
1073 | ||
1074 | static int bpf_object__init_struct_ops_maps(struct bpf_object *obj) | |
1075 | { | |
1076 | const struct btf_type *type, *datasec; | |
1077 | const struct btf_var_secinfo *vsi; | |
1078 | struct bpf_struct_ops *st_ops; | |
1079 | const char *tname, *var_name; | |
1080 | __s32 type_id, datasec_id; | |
1081 | const struct btf *btf; | |
1082 | struct bpf_map *map; | |
1083 | __u32 i; | |
1084 | ||
1085 | if (obj->efile.st_ops_shndx == -1) | |
1086 | return 0; | |
1087 | ||
1088 | btf = obj->btf; | |
1089 | datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC, | |
1090 | BTF_KIND_DATASEC); | |
1091 | if (datasec_id < 0) { | |
1092 | pr_warn("struct_ops init: DATASEC %s not found\n", | |
1093 | STRUCT_OPS_SEC); | |
1094 | return -EINVAL; | |
1095 | } | |
1096 | ||
1097 | datasec = btf__type_by_id(btf, datasec_id); | |
1098 | vsi = btf_var_secinfos(datasec); | |
1099 | for (i = 0; i < btf_vlen(datasec); i++, vsi++) { | |
1100 | type = btf__type_by_id(obj->btf, vsi->type); | |
1101 | var_name = btf__name_by_offset(obj->btf, type->name_off); | |
1102 | ||
1103 | type_id = btf__resolve_type(obj->btf, vsi->type); | |
1104 | if (type_id < 0) { | |
1105 | pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n", | |
1106 | vsi->type, STRUCT_OPS_SEC); | |
1107 | return -EINVAL; | |
1108 | } | |
1109 | ||
1110 | type = btf__type_by_id(obj->btf, type_id); | |
1111 | tname = btf__name_by_offset(obj->btf, type->name_off); | |
1112 | if (!tname[0]) { | |
1113 | pr_warn("struct_ops init: anonymous type is not supported\n"); | |
1114 | return -ENOTSUP; | |
1115 | } | |
1116 | if (!btf_is_struct(type)) { | |
1117 | pr_warn("struct_ops init: %s is not a struct\n", tname); | |
1118 | return -EINVAL; | |
1119 | } | |
1120 | ||
1121 | map = bpf_object__add_map(obj); | |
1122 | if (IS_ERR(map)) | |
1123 | return PTR_ERR(map); | |
1124 | ||
1125 | map->sec_idx = obj->efile.st_ops_shndx; | |
1126 | map->sec_offset = vsi->offset; | |
1127 | map->name = strdup(var_name); | |
1128 | if (!map->name) | |
1129 | return -ENOMEM; | |
1130 | ||
1131 | map->def.type = BPF_MAP_TYPE_STRUCT_OPS; | |
1132 | map->def.key_size = sizeof(int); | |
1133 | map->def.value_size = type->size; | |
1134 | map->def.max_entries = 1; | |
1135 | ||
1136 | map->st_ops = calloc(1, sizeof(*map->st_ops)); | |
1137 | if (!map->st_ops) | |
1138 | return -ENOMEM; | |
1139 | st_ops = map->st_ops; | |
1140 | st_ops->data = malloc(type->size); | |
1141 | st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs)); | |
1142 | st_ops->kern_func_off = malloc(btf_vlen(type) * | |
1143 | sizeof(*st_ops->kern_func_off)); | |
1144 | if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off) | |
1145 | return -ENOMEM; | |
1146 | ||
1147 | if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) { | |
1148 | pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n", | |
1149 | var_name, STRUCT_OPS_SEC); | |
1150 | return -EINVAL; | |
1151 | } | |
1152 | ||
1153 | memcpy(st_ops->data, | |
1154 | obj->efile.st_ops_data->d_buf + vsi->offset, | |
1155 | type->size); | |
1156 | st_ops->tname = tname; | |
1157 | st_ops->type = type; | |
1158 | st_ops->type_id = type_id; | |
1159 | ||
1160 | pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n", | |
1161 | tname, type_id, var_name, vsi->offset); | |
1162 | } | |
1163 | ||
1164 | return 0; | |
1165 | } | |
1166 | ||
6c956392 | 1167 | static struct bpf_object *bpf_object__new(const char *path, |
5e61f270 | 1168 | const void *obj_buf, |
2ce8450e AN |
1169 | size_t obj_buf_sz, |
1170 | const char *obj_name) | |
1a5e3fb1 | 1171 | { |
689624f0 | 1172 | bool strict = (libbpf_mode & LIBBPF_STRICT_NO_OBJECT_LIST); |
1a5e3fb1 | 1173 | struct bpf_object *obj; |
d859900c | 1174 | char *end; |
1a5e3fb1 WN |
1175 | |
1176 | obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1); | |
1177 | if (!obj) { | |
be18010e | 1178 | pr_warn("alloc memory failed for %s\n", path); |
6371ca3b | 1179 | return ERR_PTR(-ENOMEM); |
1a5e3fb1 WN |
1180 | } |
1181 | ||
1182 | strcpy(obj->path, path); | |
2ce8450e | 1183 | if (obj_name) { |
9fc205b4 | 1184 | libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name)); |
2ce8450e AN |
1185 | } else { |
1186 | /* Using basename() GNU version which doesn't modify arg. */ | |
9fc205b4 | 1187 | libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name)); |
2ce8450e AN |
1188 | end = strchr(obj->name, '.'); |
1189 | if (end) | |
1190 | *end = 0; | |
1191 | } | |
6c956392 | 1192 | |
d859900c | 1193 | obj->efile.fd = -1; |
6c956392 | 1194 | /* |
76e1022b | 1195 | * Caller of this function should also call |
6c956392 WN |
1196 | * bpf_object__elf_finish() after data collection to return |
1197 | * obj_buf to user. If not, we should duplicate the buffer to | |
1198 | * avoid user freeing them before elf finish. | |
1199 | */ | |
1200 | obj->efile.obj_buf = obj_buf; | |
1201 | obj->efile.obj_buf_sz = obj_buf_sz; | |
666810e8 | 1202 | obj->efile.maps_shndx = -1; |
abd29c93 | 1203 | obj->efile.btf_maps_shndx = -1; |
590a0088 | 1204 | obj->efile.st_ops_shndx = -1; |
81bfdd08 | 1205 | obj->kconfig_map_idx = -1; |
6c956392 | 1206 | |
5e61f270 | 1207 | obj->kern_version = get_kernel_version(); |
52d3352e | 1208 | obj->loaded = false; |
9a208eff WN |
1209 | |
1210 | INIT_LIST_HEAD(&obj->list); | |
689624f0 JB |
1211 | if (!strict) |
1212 | list_add(&obj->list, &bpf_objects_list); | |
1a5e3fb1 WN |
1213 | return obj; |
1214 | } | |
1215 | ||
1216 | static void bpf_object__elf_finish(struct bpf_object *obj) | |
1217 | { | |
29a30ff5 | 1218 | if (!obj->efile.elf) |
1a5e3fb1 WN |
1219 | return; |
1220 | ||
1221 | if (obj->efile.elf) { | |
1222 | elf_end(obj->efile.elf); | |
1223 | obj->efile.elf = NULL; | |
1224 | } | |
bec7d68c | 1225 | obj->efile.symbols = NULL; |
590a0088 | 1226 | obj->efile.st_ops_data = NULL; |
b62f06e8 | 1227 | |
25bbbd7a AN |
1228 | zfree(&obj->efile.secs); |
1229 | obj->efile.sec_cnt = 0; | |
1a5e3fb1 | 1230 | zclose(obj->efile.fd); |
6c956392 WN |
1231 | obj->efile.obj_buf = NULL; |
1232 | obj->efile.obj_buf_sz = 0; | |
1a5e3fb1 WN |
1233 | } |
1234 | ||
1235 | static int bpf_object__elf_init(struct bpf_object *obj) | |
1236 | { | |
ad23b723 | 1237 | Elf64_Ehdr *ehdr; |
1a5e3fb1 | 1238 | int err = 0; |
ad23b723 | 1239 | Elf *elf; |
1a5e3fb1 | 1240 | |
29a30ff5 | 1241 | if (obj->efile.elf) { |
88a82120 | 1242 | pr_warn("elf: init internal error\n"); |
6371ca3b | 1243 | return -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1244 | } |
1245 | ||
6c956392 WN |
1246 | if (obj->efile.obj_buf_sz > 0) { |
1247 | /* | |
1248 | * obj_buf should have been validated by | |
1249 | * bpf_object__open_buffer(). | |
1250 | */ | |
ad23b723 | 1251 | elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz); |
6c956392 | 1252 | } else { |
92274e24 | 1253 | obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC); |
6c956392 | 1254 | if (obj->efile.fd < 0) { |
be5c5d4e | 1255 | char errmsg[STRERR_BUFSIZE], *cp; |
1ce6a9fc | 1256 | |
be5c5d4e AN |
1257 | err = -errno; |
1258 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
88a82120 | 1259 | pr_warn("elf: failed to open %s: %s\n", obj->path, cp); |
be5c5d4e | 1260 | return err; |
6c956392 WN |
1261 | } |
1262 | ||
ad23b723 | 1263 | elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL); |
1a5e3fb1 WN |
1264 | } |
1265 | ||
ad23b723 | 1266 | if (!elf) { |
88a82120 | 1267 | pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1268 | err = -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1269 | goto errout; |
1270 | } | |
1271 | ||
ad23b723 AN |
1272 | obj->efile.elf = elf; |
1273 | ||
1274 | if (elf_kind(elf) != ELF_K_ELF) { | |
1275 | err = -LIBBPF_ERRNO__FORMAT; | |
1276 | pr_warn("elf: '%s' is not a proper ELF object\n", obj->path); | |
1277 | goto errout; | |
1278 | } | |
1279 | ||
1280 | if (gelf_getclass(elf) != ELFCLASS64) { | |
1281 | err = -LIBBPF_ERRNO__FORMAT; | |
1282 | pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path); | |
1283 | goto errout; | |
1284 | } | |
1285 | ||
1286 | obj->efile.ehdr = ehdr = elf64_getehdr(elf); | |
1287 | if (!obj->efile.ehdr) { | |
88a82120 | 1288 | pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1289 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1290 | goto errout; |
1291 | } | |
1a5e3fb1 | 1292 | |
ad23b723 | 1293 | if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) { |
88a82120 AN |
1294 | pr_warn("elf: failed to get section names section index for %s: %s\n", |
1295 | obj->path, elf_errmsg(-1)); | |
1296 | err = -LIBBPF_ERRNO__FORMAT; | |
1297 | goto errout; | |
1298 | } | |
1299 | ||
1300 | /* Elf is corrupted/truncated, avoid calling elf_strptr. */ | |
ad23b723 | 1301 | if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) { |
88a82120 AN |
1302 | pr_warn("elf: failed to get section names strings from %s: %s\n", |
1303 | obj->path, elf_errmsg(-1)); | |
8f3f5792 NK |
1304 | err = -LIBBPF_ERRNO__FORMAT; |
1305 | goto errout; | |
88a82120 AN |
1306 | } |
1307 | ||
9b16137a | 1308 | /* Old LLVM set e_machine to EM_NONE */ |
ad23b723 | 1309 | if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) { |
88a82120 | 1310 | pr_warn("elf: %s is not a valid eBPF object file\n", obj->path); |
6371ca3b | 1311 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1312 | goto errout; |
1313 | } | |
1314 | ||
1315 | return 0; | |
1316 | errout: | |
1317 | bpf_object__elf_finish(obj); | |
1318 | return err; | |
1319 | } | |
1320 | ||
12ef5634 | 1321 | static int bpf_object__check_endianness(struct bpf_object *obj) |
cc4228d5 | 1322 | { |
3930198d | 1323 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
ad23b723 | 1324 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB) |
12ef5634 | 1325 | return 0; |
3930198d | 1326 | #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
ad23b723 | 1327 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB) |
12ef5634 AN |
1328 | return 0; |
1329 | #else | |
1330 | # error "Unrecognized __BYTE_ORDER__" | |
1331 | #endif | |
88a82120 | 1332 | pr_warn("elf: endianness mismatch in %s.\n", obj->path); |
6371ca3b | 1333 | return -LIBBPF_ERRNO__ENDIAN; |
cc4228d5 WN |
1334 | } |
1335 | ||
cb1e5e96 | 1336 | static int |
399dc65e | 1337 | bpf_object__init_license(struct bpf_object *obj, void *data, size_t size) |
cb1e5e96 | 1338 | { |
f9798239 AN |
1339 | /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't |
1340 | * go over allowed ELF data section buffer | |
1341 | */ | |
1342 | libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license))); | |
cb1e5e96 WN |
1343 | pr_debug("license of %s is %s\n", obj->path, obj->license); |
1344 | return 0; | |
1345 | } | |
1346 | ||
54b8625c JF |
1347 | static int |
1348 | bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size) | |
1349 | { | |
1350 | __u32 kver; | |
1351 | ||
1352 | if (size != sizeof(kver)) { | |
be18010e | 1353 | pr_warn("invalid kver section in %s\n", obj->path); |
54b8625c JF |
1354 | return -LIBBPF_ERRNO__FORMAT; |
1355 | } | |
1356 | memcpy(&kver, data, sizeof(kver)); | |
1357 | obj->kern_version = kver; | |
1358 | pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version); | |
1359 | return 0; | |
1360 | } | |
1361 | ||
addb9fc9 NS |
1362 | static bool bpf_map_type__is_map_in_map(enum bpf_map_type type) |
1363 | { | |
1364 | if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1365 | type == BPF_MAP_TYPE_HASH_OF_MAPS) | |
1366 | return true; | |
1367 | return false; | |
1368 | } | |
1369 | ||
b96c07f3 | 1370 | static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size) |
1713d68b | 1371 | { |
25bbbd7a AN |
1372 | Elf_Data *data; |
1373 | Elf_Scn *scn; | |
1713d68b | 1374 | |
25bbbd7a | 1375 | if (!name) |
1713d68b | 1376 | return -EINVAL; |
88a82120 | 1377 | |
25bbbd7a AN |
1378 | scn = elf_sec_by_name(obj, name); |
1379 | data = elf_sec_data(obj, scn); | |
1380 | if (data) { | |
25bbbd7a | 1381 | *size = data->d_size; |
08894d9c | 1382 | return 0; /* found it */ |
1713d68b DB |
1383 | } |
1384 | ||
08894d9c | 1385 | return -ENOENT; |
1713d68b DB |
1386 | } |
1387 | ||
b96c07f3 | 1388 | static int find_elf_var_offset(const struct bpf_object *obj, const char *name, __u32 *off) |
1713d68b DB |
1389 | { |
1390 | Elf_Data *symbols = obj->efile.symbols; | |
1391 | const char *sname; | |
1392 | size_t si; | |
1393 | ||
1394 | if (!name || !off) | |
1395 | return -EINVAL; | |
1396 | ||
ad23b723 AN |
1397 | for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) { |
1398 | Elf64_Sym *sym = elf_sym_by_idx(obj, si); | |
1713d68b | 1399 | |
ad23b723 AN |
1400 | if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL || |
1401 | ELF64_ST_TYPE(sym->st_info) != STT_OBJECT) | |
1713d68b DB |
1402 | continue; |
1403 | ||
ad23b723 | 1404 | sname = elf_sym_str(obj, sym->st_name); |
1713d68b | 1405 | if (!sname) { |
ad23b723 | 1406 | pr_warn("failed to get sym name string for var %s\n", name); |
1713d68b DB |
1407 | return -EIO; |
1408 | } | |
1409 | if (strcmp(name, sname) == 0) { | |
ad23b723 | 1410 | *off = sym->st_value; |
1713d68b DB |
1411 | return 0; |
1412 | } | |
1413 | } | |
1414 | ||
1415 | return -ENOENT; | |
1416 | } | |
1417 | ||
bf829271 | 1418 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj) |
d859900c | 1419 | { |
bf829271 AN |
1420 | struct bpf_map *new_maps; |
1421 | size_t new_cap; | |
1422 | int i; | |
1423 | ||
1424 | if (obj->nr_maps < obj->maps_cap) | |
1425 | return &obj->maps[obj->nr_maps++]; | |
1426 | ||
95064979 | 1427 | new_cap = max((size_t)4, obj->maps_cap * 3 / 2); |
029258d7 | 1428 | new_maps = libbpf_reallocarray(obj->maps, new_cap, sizeof(*obj->maps)); |
bf829271 | 1429 | if (!new_maps) { |
be18010e | 1430 | pr_warn("alloc maps for object failed\n"); |
bf829271 AN |
1431 | return ERR_PTR(-ENOMEM); |
1432 | } | |
1433 | ||
1434 | obj->maps_cap = new_cap; | |
1435 | obj->maps = new_maps; | |
1436 | ||
1437 | /* zero out new maps */ | |
1438 | memset(obj->maps + obj->nr_maps, 0, | |
1439 | (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps)); | |
1440 | /* | |
1441 | * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin) | |
1442 | * when failure (zclose won't close negative fd)). | |
1443 | */ | |
1444 | for (i = obj->nr_maps; i < obj->maps_cap; i++) { | |
1445 | obj->maps[i].fd = -1; | |
1446 | obj->maps[i].inner_map_fd = -1; | |
1447 | } | |
1448 | ||
1449 | return &obj->maps[obj->nr_maps++]; | |
d859900c DB |
1450 | } |
1451 | ||
eba9c5f4 AN |
1452 | static size_t bpf_map_mmap_sz(const struct bpf_map *map) |
1453 | { | |
1454 | long page_sz = sysconf(_SC_PAGE_SIZE); | |
1455 | size_t map_sz; | |
1456 | ||
c701917e | 1457 | map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries; |
eba9c5f4 AN |
1458 | map_sz = roundup(map_sz, page_sz); |
1459 | return map_sz; | |
1460 | } | |
1461 | ||
aed65917 | 1462 | static char *internal_map_name(struct bpf_object *obj, const char *real_name) |
81bfdd08 | 1463 | { |
113e6b7e | 1464 | char map_name[BPF_OBJ_NAME_LEN], *p; |
aed65917 AN |
1465 | int pfx_len, sfx_len = max((size_t)7, strlen(real_name)); |
1466 | ||
1467 | /* This is one of the more confusing parts of libbpf for various | |
1468 | * reasons, some of which are historical. The original idea for naming | |
1469 | * internal names was to include as much of BPF object name prefix as | |
1470 | * possible, so that it can be distinguished from similar internal | |
1471 | * maps of a different BPF object. | |
1472 | * As an example, let's say we have bpf_object named 'my_object_name' | |
1473 | * and internal map corresponding to '.rodata' ELF section. The final | |
1474 | * map name advertised to user and to the kernel will be | |
1475 | * 'my_objec.rodata', taking first 8 characters of object name and | |
1476 | * entire 7 characters of '.rodata'. | |
1477 | * Somewhat confusingly, if internal map ELF section name is shorter | |
1478 | * than 7 characters, e.g., '.bss', we still reserve 7 characters | |
1479 | * for the suffix, even though we only have 4 actual characters, and | |
1480 | * resulting map will be called 'my_objec.bss', not even using all 15 | |
1481 | * characters allowed by the kernel. Oh well, at least the truncated | |
1482 | * object name is somewhat consistent in this case. But if the map | |
1483 | * name is '.kconfig', we'll still have entirety of '.kconfig' added | |
1484 | * (8 chars) and thus will be left with only first 7 characters of the | |
1485 | * object name ('my_obje'). Happy guessing, user, that the final map | |
1486 | * name will be "my_obje.kconfig". | |
1487 | * Now, with libbpf starting to support arbitrarily named .rodata.* | |
1488 | * and .data.* data sections, it's possible that ELF section name is | |
1489 | * longer than allowed 15 chars, so we now need to be careful to take | |
1490 | * only up to 15 first characters of ELF name, taking no BPF object | |
1491 | * name characters at all. So '.rodata.abracadabra' will result in | |
1492 | * '.rodata.abracad' kernel and user-visible name. | |
1493 | * We need to keep this convoluted logic intact for .data, .bss and | |
1494 | * .rodata maps, but for new custom .data.custom and .rodata.custom | |
1495 | * maps we use their ELF names as is, not prepending bpf_object name | |
1496 | * in front. We still need to truncate them to 15 characters for the | |
1497 | * kernel. Full name can be recovered for such maps by using DATASEC | |
1498 | * BTF type associated with such map's value type, though. | |
1499 | */ | |
1500 | if (sfx_len >= BPF_OBJ_NAME_LEN) | |
1501 | sfx_len = BPF_OBJ_NAME_LEN - 1; | |
1502 | ||
1503 | /* if there are two or more dots in map name, it's a custom dot map */ | |
1504 | if (strchr(real_name + 1, '.') != NULL) | |
1505 | pfx_len = 0; | |
1506 | else | |
1507 | pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name)); | |
81bfdd08 AN |
1508 | |
1509 | snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name, | |
aed65917 | 1510 | sfx_len, real_name); |
81bfdd08 | 1511 | |
113e6b7e THJ |
1512 | /* sanitise map name to characters allowed by kernel */ |
1513 | for (p = map_name; *p && p < map_name + sizeof(map_name); p++) | |
1514 | if (!isalnum(*p) && *p != '_' && *p != '.') | |
1515 | *p = '_'; | |
1516 | ||
81bfdd08 AN |
1517 | return strdup(map_name); |
1518 | } | |
1519 | ||
d859900c | 1520 | static int |
bf829271 | 1521 | bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type, |
aed65917 | 1522 | const char *real_name, int sec_idx, void *data, size_t data_sz) |
d859900c | 1523 | { |
bf829271 AN |
1524 | struct bpf_map_def *def; |
1525 | struct bpf_map *map; | |
eba9c5f4 | 1526 | int err; |
bf829271 AN |
1527 | |
1528 | map = bpf_object__add_map(obj); | |
1529 | if (IS_ERR(map)) | |
1530 | return PTR_ERR(map); | |
d859900c DB |
1531 | |
1532 | map->libbpf_type = type; | |
db48814b AN |
1533 | map->sec_idx = sec_idx; |
1534 | map->sec_offset = 0; | |
aed65917 AN |
1535 | map->real_name = strdup(real_name); |
1536 | map->name = internal_map_name(obj, real_name); | |
1537 | if (!map->real_name || !map->name) { | |
1538 | zfree(&map->real_name); | |
1539 | zfree(&map->name); | |
d859900c DB |
1540 | return -ENOMEM; |
1541 | } | |
1542 | ||
bf829271 | 1543 | def = &map->def; |
d859900c DB |
1544 | def->type = BPF_MAP_TYPE_ARRAY; |
1545 | def->key_size = sizeof(int); | |
eba9c5f4 | 1546 | def->value_size = data_sz; |
d859900c | 1547 | def->max_entries = 1; |
81bfdd08 | 1548 | def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG |
166750bc | 1549 | ? BPF_F_RDONLY_PROG : 0; |
0d13bfce | 1550 | def->map_flags |= BPF_F_MMAPABLE; |
7fe74b43 AN |
1551 | |
1552 | pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n", | |
81bfdd08 | 1553 | map->name, map->sec_idx, map->sec_offset, def->map_flags); |
7fe74b43 | 1554 | |
eba9c5f4 AN |
1555 | map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE, |
1556 | MAP_SHARED | MAP_ANONYMOUS, -1, 0); | |
1557 | if (map->mmaped == MAP_FAILED) { | |
1558 | err = -errno; | |
1559 | map->mmaped = NULL; | |
1560 | pr_warn("failed to alloc map '%s' content buffer: %d\n", | |
1561 | map->name, err); | |
aed65917 | 1562 | zfree(&map->real_name); |
eba9c5f4 AN |
1563 | zfree(&map->name); |
1564 | return err; | |
d859900c DB |
1565 | } |
1566 | ||
166750bc | 1567 | if (data) |
eba9c5f4 AN |
1568 | memcpy(map->mmaped, data, data_sz); |
1569 | ||
e1d1dc46 | 1570 | pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name); |
d859900c DB |
1571 | return 0; |
1572 | } | |
1573 | ||
bf829271 AN |
1574 | static int bpf_object__init_global_data_maps(struct bpf_object *obj) |
1575 | { | |
25bbbd7a | 1576 | struct elf_sec_desc *sec_desc; |
aed65917 | 1577 | const char *sec_name; |
25bbbd7a | 1578 | int err = 0, sec_idx; |
bf829271 | 1579 | |
bf829271 AN |
1580 | /* |
1581 | * Populate obj->maps with libbpf internal maps. | |
1582 | */ | |
25bbbd7a AN |
1583 | for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) { |
1584 | sec_desc = &obj->efile.secs[sec_idx]; | |
1585 | ||
1586 | switch (sec_desc->sec_type) { | |
1587 | case SEC_DATA: | |
aed65917 | 1588 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1589 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA, |
aed65917 | 1590 | sec_name, sec_idx, |
25bbbd7a AN |
1591 | sec_desc->data->d_buf, |
1592 | sec_desc->data->d_size); | |
1593 | break; | |
1594 | case SEC_RODATA: | |
1595 | obj->has_rodata = true; | |
aed65917 | 1596 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1597 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA, |
aed65917 | 1598 | sec_name, sec_idx, |
25bbbd7a AN |
1599 | sec_desc->data->d_buf, |
1600 | sec_desc->data->d_size); | |
1601 | break; | |
1602 | case SEC_BSS: | |
aed65917 | 1603 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1604 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS, |
aed65917 | 1605 | sec_name, sec_idx, |
25bbbd7a AN |
1606 | NULL, |
1607 | sec_desc->data->d_size); | |
1608 | break; | |
1609 | default: | |
1610 | /* skip */ | |
1611 | break; | |
1612 | } | |
bf829271 AN |
1613 | if (err) |
1614 | return err; | |
1615 | } | |
1616 | return 0; | |
1617 | } | |
1618 | ||
166750bc AN |
1619 | |
1620 | static struct extern_desc *find_extern_by_name(const struct bpf_object *obj, | |
1621 | const void *name) | |
1622 | { | |
1623 | int i; | |
1624 | ||
1625 | for (i = 0; i < obj->nr_extern; i++) { | |
1626 | if (strcmp(obj->externs[i].name, name) == 0) | |
1627 | return &obj->externs[i]; | |
1628 | } | |
1629 | return NULL; | |
1630 | } | |
1631 | ||
2e33efe3 AN |
1632 | static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val, |
1633 | char value) | |
166750bc | 1634 | { |
2e33efe3 AN |
1635 | switch (ext->kcfg.type) { |
1636 | case KCFG_BOOL: | |
166750bc | 1637 | if (value == 'm') { |
2e33efe3 | 1638 | pr_warn("extern (kcfg) %s=%c should be tristate or char\n", |
166750bc AN |
1639 | ext->name, value); |
1640 | return -EINVAL; | |
1641 | } | |
1642 | *(bool *)ext_val = value == 'y' ? true : false; | |
1643 | break; | |
2e33efe3 | 1644 | case KCFG_TRISTATE: |
166750bc AN |
1645 | if (value == 'y') |
1646 | *(enum libbpf_tristate *)ext_val = TRI_YES; | |
1647 | else if (value == 'm') | |
1648 | *(enum libbpf_tristate *)ext_val = TRI_MODULE; | |
1649 | else /* value == 'n' */ | |
1650 | *(enum libbpf_tristate *)ext_val = TRI_NO; | |
1651 | break; | |
2e33efe3 | 1652 | case KCFG_CHAR: |
166750bc AN |
1653 | *(char *)ext_val = value; |
1654 | break; | |
2e33efe3 AN |
1655 | case KCFG_UNKNOWN: |
1656 | case KCFG_INT: | |
1657 | case KCFG_CHAR_ARR: | |
166750bc | 1658 | default: |
2e33efe3 | 1659 | pr_warn("extern (kcfg) %s=%c should be bool, tristate, or char\n", |
166750bc AN |
1660 | ext->name, value); |
1661 | return -EINVAL; | |
1662 | } | |
1663 | ext->is_set = true; | |
1664 | return 0; | |
1665 | } | |
1666 | ||
2e33efe3 AN |
1667 | static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val, |
1668 | const char *value) | |
166750bc AN |
1669 | { |
1670 | size_t len; | |
1671 | ||
2e33efe3 AN |
1672 | if (ext->kcfg.type != KCFG_CHAR_ARR) { |
1673 | pr_warn("extern (kcfg) %s=%s should be char array\n", ext->name, value); | |
166750bc AN |
1674 | return -EINVAL; |
1675 | } | |
1676 | ||
1677 | len = strlen(value); | |
1678 | if (value[len - 1] != '"') { | |
2e33efe3 | 1679 | pr_warn("extern (kcfg) '%s': invalid string config '%s'\n", |
166750bc AN |
1680 | ext->name, value); |
1681 | return -EINVAL; | |
1682 | } | |
1683 | ||
1684 | /* strip quotes */ | |
1685 | len -= 2; | |
2e33efe3 AN |
1686 | if (len >= ext->kcfg.sz) { |
1687 | pr_warn("extern (kcfg) '%s': long string config %s of (%zu bytes) truncated to %d bytes\n", | |
1688 | ext->name, value, len, ext->kcfg.sz - 1); | |
1689 | len = ext->kcfg.sz - 1; | |
166750bc AN |
1690 | } |
1691 | memcpy(ext_val, value + 1, len); | |
1692 | ext_val[len] = '\0'; | |
1693 | ext->is_set = true; | |
1694 | return 0; | |
1695 | } | |
1696 | ||
1697 | static int parse_u64(const char *value, __u64 *res) | |
1698 | { | |
1699 | char *value_end; | |
1700 | int err; | |
1701 | ||
1702 | errno = 0; | |
1703 | *res = strtoull(value, &value_end, 0); | |
1704 | if (errno) { | |
1705 | err = -errno; | |
1706 | pr_warn("failed to parse '%s' as integer: %d\n", value, err); | |
1707 | return err; | |
1708 | } | |
1709 | if (*value_end) { | |
1710 | pr_warn("failed to parse '%s' as integer completely\n", value); | |
1711 | return -EINVAL; | |
1712 | } | |
1713 | return 0; | |
1714 | } | |
1715 | ||
2e33efe3 | 1716 | static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v) |
166750bc | 1717 | { |
2e33efe3 | 1718 | int bit_sz = ext->kcfg.sz * 8; |
166750bc | 1719 | |
2e33efe3 | 1720 | if (ext->kcfg.sz == 8) |
166750bc AN |
1721 | return true; |
1722 | ||
1723 | /* Validate that value stored in u64 fits in integer of `ext->sz` | |
1724 | * bytes size without any loss of information. If the target integer | |
1725 | * is signed, we rely on the following limits of integer type of | |
1726 | * Y bits and subsequent transformation: | |
1727 | * | |
1728 | * -2^(Y-1) <= X <= 2^(Y-1) - 1 | |
1729 | * 0 <= X + 2^(Y-1) <= 2^Y - 1 | |
1730 | * 0 <= X + 2^(Y-1) < 2^Y | |
1731 | * | |
1732 | * For unsigned target integer, check that all the (64 - Y) bits are | |
1733 | * zero. | |
1734 | */ | |
2e33efe3 | 1735 | if (ext->kcfg.is_signed) |
166750bc AN |
1736 | return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz); |
1737 | else | |
1738 | return (v >> bit_sz) == 0; | |
1739 | } | |
1740 | ||
2e33efe3 AN |
1741 | static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val, |
1742 | __u64 value) | |
166750bc | 1743 | { |
2e33efe3 AN |
1744 | if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) { |
1745 | pr_warn("extern (kcfg) %s=%llu should be integer\n", | |
7745ff98 | 1746 | ext->name, (unsigned long long)value); |
166750bc AN |
1747 | return -EINVAL; |
1748 | } | |
2e33efe3 AN |
1749 | if (!is_kcfg_value_in_range(ext, value)) { |
1750 | pr_warn("extern (kcfg) %s=%llu value doesn't fit in %d bytes\n", | |
1751 | ext->name, (unsigned long long)value, ext->kcfg.sz); | |
166750bc AN |
1752 | return -ERANGE; |
1753 | } | |
2e33efe3 | 1754 | switch (ext->kcfg.sz) { |
166750bc AN |
1755 | case 1: *(__u8 *)ext_val = value; break; |
1756 | case 2: *(__u16 *)ext_val = value; break; | |
1757 | case 4: *(__u32 *)ext_val = value; break; | |
1758 | case 8: *(__u64 *)ext_val = value; break; | |
1759 | default: | |
1760 | return -EINVAL; | |
1761 | } | |
1762 | ext->is_set = true; | |
1763 | return 0; | |
1764 | } | |
1765 | ||
8601fd42 AN |
1766 | static int bpf_object__process_kconfig_line(struct bpf_object *obj, |
1767 | char *buf, void *data) | |
166750bc | 1768 | { |
166750bc | 1769 | struct extern_desc *ext; |
8601fd42 | 1770 | char *sep, *value; |
166750bc AN |
1771 | int len, err = 0; |
1772 | void *ext_val; | |
1773 | __u64 num; | |
166750bc | 1774 | |
13d35a0c | 1775 | if (!str_has_pfx(buf, "CONFIG_")) |
8601fd42 | 1776 | return 0; |
166750bc | 1777 | |
8601fd42 AN |
1778 | sep = strchr(buf, '='); |
1779 | if (!sep) { | |
1780 | pr_warn("failed to parse '%s': no separator\n", buf); | |
1781 | return -EINVAL; | |
1782 | } | |
1783 | ||
1784 | /* Trim ending '\n' */ | |
1785 | len = strlen(buf); | |
1786 | if (buf[len - 1] == '\n') | |
1787 | buf[len - 1] = '\0'; | |
1788 | /* Split on '=' and ensure that a value is present. */ | |
1789 | *sep = '\0'; | |
1790 | if (!sep[1]) { | |
1791 | *sep = '='; | |
1792 | pr_warn("failed to parse '%s': no value\n", buf); | |
1793 | return -EINVAL; | |
1794 | } | |
1795 | ||
1796 | ext = find_extern_by_name(obj, buf); | |
1797 | if (!ext || ext->is_set) | |
1798 | return 0; | |
1799 | ||
2e33efe3 | 1800 | ext_val = data + ext->kcfg.data_off; |
8601fd42 AN |
1801 | value = sep + 1; |
1802 | ||
1803 | switch (*value) { | |
1804 | case 'y': case 'n': case 'm': | |
2e33efe3 | 1805 | err = set_kcfg_value_tri(ext, ext_val, *value); |
8601fd42 AN |
1806 | break; |
1807 | case '"': | |
2e33efe3 | 1808 | err = set_kcfg_value_str(ext, ext_val, value); |
8601fd42 AN |
1809 | break; |
1810 | default: | |
1811 | /* assume integer */ | |
1812 | err = parse_u64(value, &num); | |
1813 | if (err) { | |
2e33efe3 | 1814 | pr_warn("extern (kcfg) %s=%s should be integer\n", |
8601fd42 AN |
1815 | ext->name, value); |
1816 | return err; | |
1817 | } | |
2e33efe3 | 1818 | err = set_kcfg_value_num(ext, ext_val, num); |
8601fd42 | 1819 | break; |
166750bc | 1820 | } |
8601fd42 AN |
1821 | if (err) |
1822 | return err; | |
2e33efe3 | 1823 | pr_debug("extern (kcfg) %s=%s\n", ext->name, value); |
8601fd42 AN |
1824 | return 0; |
1825 | } | |
1826 | ||
1827 | static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data) | |
1828 | { | |
1829 | char buf[PATH_MAX]; | |
1830 | struct utsname uts; | |
1831 | int len, err = 0; | |
1832 | gzFile file; | |
1833 | ||
1834 | uname(&uts); | |
1835 | len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release); | |
1836 | if (len < 0) | |
1837 | return -EINVAL; | |
1838 | else if (len >= PATH_MAX) | |
1839 | return -ENAMETOOLONG; | |
1840 | ||
1841 | /* gzopen also accepts uncompressed files. */ | |
1842 | file = gzopen(buf, "r"); | |
1843 | if (!file) | |
1844 | file = gzopen("/proc/config.gz", "r"); | |
1845 | ||
166750bc | 1846 | if (!file) { |
8601fd42 | 1847 | pr_warn("failed to open system Kconfig\n"); |
166750bc AN |
1848 | return -ENOENT; |
1849 | } | |
1850 | ||
1851 | while (gzgets(file, buf, sizeof(buf))) { | |
8601fd42 AN |
1852 | err = bpf_object__process_kconfig_line(obj, buf, data); |
1853 | if (err) { | |
1854 | pr_warn("error parsing system Kconfig line '%s': %d\n", | |
1855 | buf, err); | |
166750bc AN |
1856 | goto out; |
1857 | } | |
8601fd42 | 1858 | } |
166750bc | 1859 | |
8601fd42 AN |
1860 | out: |
1861 | gzclose(file); | |
1862 | return err; | |
1863 | } | |
166750bc | 1864 | |
8601fd42 AN |
1865 | static int bpf_object__read_kconfig_mem(struct bpf_object *obj, |
1866 | const char *config, void *data) | |
1867 | { | |
1868 | char buf[PATH_MAX]; | |
1869 | int err = 0; | |
1870 | FILE *file; | |
166750bc | 1871 | |
8601fd42 AN |
1872 | file = fmemopen((void *)config, strlen(config), "r"); |
1873 | if (!file) { | |
1874 | err = -errno; | |
1875 | pr_warn("failed to open in-memory Kconfig: %d\n", err); | |
1876 | return err; | |
1877 | } | |
1878 | ||
1879 | while (fgets(buf, sizeof(buf), file)) { | |
1880 | err = bpf_object__process_kconfig_line(obj, buf, data); | |
1881 | if (err) { | |
1882 | pr_warn("error parsing in-memory Kconfig line '%s': %d\n", | |
1883 | buf, err); | |
166750bc AN |
1884 | break; |
1885 | } | |
166750bc AN |
1886 | } |
1887 | ||
8601fd42 | 1888 | fclose(file); |
166750bc AN |
1889 | return err; |
1890 | } | |
1891 | ||
81bfdd08 | 1892 | static int bpf_object__init_kconfig_map(struct bpf_object *obj) |
166750bc | 1893 | { |
2e33efe3 | 1894 | struct extern_desc *last_ext = NULL, *ext; |
166750bc | 1895 | size_t map_sz; |
2e33efe3 | 1896 | int i, err; |
166750bc | 1897 | |
2e33efe3 AN |
1898 | for (i = 0; i < obj->nr_extern; i++) { |
1899 | ext = &obj->externs[i]; | |
1900 | if (ext->type == EXT_KCFG) | |
1901 | last_ext = ext; | |
1902 | } | |
166750bc | 1903 | |
2e33efe3 AN |
1904 | if (!last_ext) |
1905 | return 0; | |
166750bc | 1906 | |
2e33efe3 | 1907 | map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz; |
81bfdd08 | 1908 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG, |
aed65917 | 1909 | ".kconfig", obj->efile.symbols_shndx, |
166750bc AN |
1910 | NULL, map_sz); |
1911 | if (err) | |
1912 | return err; | |
1913 | ||
81bfdd08 | 1914 | obj->kconfig_map_idx = obj->nr_maps - 1; |
166750bc AN |
1915 | |
1916 | return 0; | |
1917 | } | |
1918 | ||
bf829271 | 1919 | static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict) |
561bbcca | 1920 | { |
561bbcca | 1921 | Elf_Data *symbols = obj->efile.symbols; |
bf829271 | 1922 | int i, map_def_sz = 0, nr_maps = 0, nr_syms; |
d859900c | 1923 | Elf_Data *data = NULL; |
bf829271 AN |
1924 | Elf_Scn *scn; |
1925 | ||
1926 | if (obj->efile.maps_shndx < 0) | |
1927 | return 0; | |
561bbcca | 1928 | |
93b8952d AN |
1929 | if (libbpf_mode & LIBBPF_STRICT_MAP_DEFINITIONS) { |
1930 | pr_warn("legacy map definitions in SEC(\"maps\") are not supported\n"); | |
1931 | return -EOPNOTSUPP; | |
1932 | } | |
1933 | ||
4708bbda EL |
1934 | if (!symbols) |
1935 | return -EINVAL; | |
1936 | ||
88a82120 AN |
1937 | scn = elf_sec_by_idx(obj, obj->efile.maps_shndx); |
1938 | data = elf_sec_data(obj, scn); | |
bf829271 | 1939 | if (!scn || !data) { |
88a82120 AN |
1940 | pr_warn("elf: failed to get legacy map definitions for %s\n", |
1941 | obj->path); | |
bf829271 | 1942 | return -EINVAL; |
4708bbda | 1943 | } |
561bbcca | 1944 | |
4708bbda EL |
1945 | /* |
1946 | * Count number of maps. Each map has a name. | |
1947 | * Array of maps is not supported: only the first element is | |
1948 | * considered. | |
1949 | * | |
1950 | * TODO: Detect array of map and report error. | |
1951 | */ | |
ad23b723 | 1952 | nr_syms = symbols->d_size / sizeof(Elf64_Sym); |
bf829271 | 1953 | for (i = 0; i < nr_syms; i++) { |
ad23b723 | 1954 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
4708bbda | 1955 | |
ad23b723 | 1956 | if (sym->st_shndx != obj->efile.maps_shndx) |
4708bbda | 1957 | continue; |
ad23b723 | 1958 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION) |
161ecd53 | 1959 | continue; |
4708bbda EL |
1960 | nr_maps++; |
1961 | } | |
b13c5c14 | 1962 | /* Assume equally sized map definitions */ |
88a82120 AN |
1963 | pr_debug("elf: found %d legacy map definitions (%zd bytes) in %s\n", |
1964 | nr_maps, data->d_size, obj->path); | |
bf829271 | 1965 | |
98e527af | 1966 | if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) { |
88a82120 AN |
1967 | pr_warn("elf: unable to determine legacy map definition size in %s\n", |
1968 | obj->path); | |
bf829271 | 1969 | return -EINVAL; |
addb9fc9 | 1970 | } |
98e527af | 1971 | map_def_sz = data->d_size / nr_maps; |
4708bbda | 1972 | |
bf829271 AN |
1973 | /* Fill obj->maps using data in "maps" section. */ |
1974 | for (i = 0; i < nr_syms; i++) { | |
ad23b723 | 1975 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
561bbcca | 1976 | const char *map_name; |
4708bbda | 1977 | struct bpf_map_def *def; |
bf829271 | 1978 | struct bpf_map *map; |
561bbcca | 1979 | |
ad23b723 | 1980 | if (sym->st_shndx != obj->efile.maps_shndx) |
561bbcca | 1981 | continue; |
ad23b723 | 1982 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION) |
c3e8c44a | 1983 | continue; |
561bbcca | 1984 | |
bf829271 AN |
1985 | map = bpf_object__add_map(obj); |
1986 | if (IS_ERR(map)) | |
1987 | return PTR_ERR(map); | |
1988 | ||
ad23b723 | 1989 | map_name = elf_sym_str(obj, sym->st_name); |
c51829bb | 1990 | if (!map_name) { |
be18010e KW |
1991 | pr_warn("failed to get map #%d name sym string for obj %s\n", |
1992 | i, obj->path); | |
c51829bb AN |
1993 | return -LIBBPF_ERRNO__FORMAT; |
1994 | } | |
d859900c | 1995 | |
93b8952d AN |
1996 | pr_warn("map '%s' (legacy): legacy map definitions are deprecated, use BTF-defined maps instead\n", map_name); |
1997 | ||
ad23b723 | 1998 | if (ELF64_ST_BIND(sym->st_info) == STB_LOCAL) { |
c1cccec9 AN |
1999 | pr_warn("map '%s' (legacy): static maps are not supported\n", map_name); |
2000 | return -ENOTSUP; | |
2001 | } | |
2002 | ||
bf829271 | 2003 | map->libbpf_type = LIBBPF_MAP_UNSPEC; |
ad23b723 AN |
2004 | map->sec_idx = sym->st_shndx; |
2005 | map->sec_offset = sym->st_value; | |
db48814b AN |
2006 | pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n", |
2007 | map_name, map->sec_idx, map->sec_offset); | |
ad23b723 | 2008 | if (sym->st_value + map_def_sz > data->d_size) { |
be18010e KW |
2009 | pr_warn("corrupted maps section in %s: last map \"%s\" too small\n", |
2010 | obj->path, map_name); | |
4708bbda | 2011 | return -EINVAL; |
561bbcca | 2012 | } |
4708bbda | 2013 | |
bf829271 AN |
2014 | map->name = strdup(map_name); |
2015 | if (!map->name) { | |
aed65917 | 2016 | pr_warn("map '%s': failed to alloc map name\n", map_name); |
973170e6 WN |
2017 | return -ENOMEM; |
2018 | } | |
bf829271 | 2019 | pr_debug("map %d is \"%s\"\n", i, map->name); |
ad23b723 | 2020 | def = (struct bpf_map_def *)(data->d_buf + sym->st_value); |
b13c5c14 CG |
2021 | /* |
2022 | * If the definition of the map in the object file fits in | |
2023 | * bpf_map_def, copy it. Any extra fields in our version | |
2024 | * of bpf_map_def will default to zero as a result of the | |
2025 | * calloc above. | |
2026 | */ | |
2027 | if (map_def_sz <= sizeof(struct bpf_map_def)) { | |
bf829271 | 2028 | memcpy(&map->def, def, map_def_sz); |
b13c5c14 CG |
2029 | } else { |
2030 | /* | |
2031 | * Here the map structure being read is bigger than what | |
2032 | * we expect, truncate if the excess bits are all zero. | |
2033 | * If they are not zero, reject this map as | |
2034 | * incompatible. | |
2035 | */ | |
2036 | char *b; | |
8983b731 | 2037 | |
b13c5c14 CG |
2038 | for (b = ((char *)def) + sizeof(struct bpf_map_def); |
2039 | b < ((char *)def) + map_def_sz; b++) { | |
2040 | if (*b != 0) { | |
8983b731 | 2041 | pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n", |
be18010e | 2042 | obj->path, map_name); |
c034a177 JF |
2043 | if (strict) |
2044 | return -EINVAL; | |
b13c5c14 CG |
2045 | } |
2046 | } | |
bf829271 | 2047 | memcpy(&map->def, def, sizeof(struct bpf_map_def)); |
b13c5c14 | 2048 | } |
561bbcca | 2049 | } |
bf829271 AN |
2050 | return 0; |
2051 | } | |
4708bbda | 2052 | |
42869d28 | 2053 | const struct btf_type * |
ddc7c304 | 2054 | skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id) |
abd29c93 AN |
2055 | { |
2056 | const struct btf_type *t = btf__type_by_id(btf, id); | |
8837fe5d | 2057 | |
ddc7c304 AN |
2058 | if (res_id) |
2059 | *res_id = id; | |
2060 | ||
2061 | while (btf_is_mod(t) || btf_is_typedef(t)) { | |
2062 | if (res_id) | |
2063 | *res_id = t->type; | |
2064 | t = btf__type_by_id(btf, t->type); | |
abd29c93 | 2065 | } |
ddc7c304 AN |
2066 | |
2067 | return t; | |
abd29c93 AN |
2068 | } |
2069 | ||
590a0088 MKL |
2070 | static const struct btf_type * |
2071 | resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id) | |
2072 | { | |
2073 | const struct btf_type *t; | |
2074 | ||
2075 | t = skip_mods_and_typedefs(btf, id, NULL); | |
2076 | if (!btf_is_ptr(t)) | |
2077 | return NULL; | |
2078 | ||
2079 | t = skip_mods_and_typedefs(btf, t->type, res_id); | |
2080 | ||
2081 | return btf_is_func_proto(t) ? t : NULL; | |
2082 | } | |
2083 | ||
774e132e | 2084 | static const char *__btf_kind_str(__u16 kind) |
81ba0889 | 2085 | { |
774e132e | 2086 | switch (kind) { |
81ba0889 AN |
2087 | case BTF_KIND_UNKN: return "void"; |
2088 | case BTF_KIND_INT: return "int"; | |
2089 | case BTF_KIND_PTR: return "ptr"; | |
2090 | case BTF_KIND_ARRAY: return "array"; | |
2091 | case BTF_KIND_STRUCT: return "struct"; | |
2092 | case BTF_KIND_UNION: return "union"; | |
2093 | case BTF_KIND_ENUM: return "enum"; | |
2094 | case BTF_KIND_FWD: return "fwd"; | |
2095 | case BTF_KIND_TYPEDEF: return "typedef"; | |
2096 | case BTF_KIND_VOLATILE: return "volatile"; | |
2097 | case BTF_KIND_CONST: return "const"; | |
2098 | case BTF_KIND_RESTRICT: return "restrict"; | |
2099 | case BTF_KIND_FUNC: return "func"; | |
2100 | case BTF_KIND_FUNC_PROTO: return "func_proto"; | |
2101 | case BTF_KIND_VAR: return "var"; | |
2102 | case BTF_KIND_DATASEC: return "datasec"; | |
22541a9e | 2103 | case BTF_KIND_FLOAT: return "float"; |
223f903e | 2104 | case BTF_KIND_DECL_TAG: return "decl_tag"; |
2dc1e488 | 2105 | case BTF_KIND_TYPE_TAG: return "type_tag"; |
81ba0889 AN |
2106 | default: return "unknown"; |
2107 | } | |
2108 | } | |
2109 | ||
42869d28 | 2110 | const char *btf_kind_str(const struct btf_type *t) |
774e132e MKL |
2111 | { |
2112 | return __btf_kind_str(btf_kind(t)); | |
2113 | } | |
2114 | ||
ef99b02b AN |
2115 | /* |
2116 | * Fetch integer attribute of BTF map definition. Such attributes are | |
2117 | * represented using a pointer to an array, in which dimensionality of array | |
2118 | * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY]; | |
2119 | * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF | |
2120 | * type definition, while using only sizeof(void *) space in ELF data section. | |
2121 | */ | |
2122 | static bool get_map_field_int(const char *map_name, const struct btf *btf, | |
8983b731 AN |
2123 | const struct btf_member *m, __u32 *res) |
2124 | { | |
ddc7c304 | 2125 | const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL); |
abd29c93 | 2126 | const char *name = btf__name_by_offset(btf, m->name_off); |
ef99b02b AN |
2127 | const struct btf_array *arr_info; |
2128 | const struct btf_type *arr_t; | |
abd29c93 | 2129 | |
b03bc685 | 2130 | if (!btf_is_ptr(t)) { |
81ba0889 AN |
2131 | pr_warn("map '%s': attr '%s': expected PTR, got %s.\n", |
2132 | map_name, name, btf_kind_str(t)); | |
abd29c93 AN |
2133 | return false; |
2134 | } | |
ef99b02b AN |
2135 | |
2136 | arr_t = btf__type_by_id(btf, t->type); | |
2137 | if (!arr_t) { | |
be18010e KW |
2138 | pr_warn("map '%s': attr '%s': type [%u] not found.\n", |
2139 | map_name, name, t->type); | |
abd29c93 AN |
2140 | return false; |
2141 | } | |
b03bc685 | 2142 | if (!btf_is_array(arr_t)) { |
81ba0889 AN |
2143 | pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n", |
2144 | map_name, name, btf_kind_str(arr_t)); | |
abd29c93 AN |
2145 | return false; |
2146 | } | |
b03bc685 | 2147 | arr_info = btf_array(arr_t); |
ef99b02b | 2148 | *res = arr_info->nelems; |
abd29c93 AN |
2149 | return true; |
2150 | } | |
2151 | ||
57a00f41 THJ |
2152 | static int build_map_pin_path(struct bpf_map *map, const char *path) |
2153 | { | |
2154 | char buf[PATH_MAX]; | |
6e9cab2e | 2155 | int len; |
57a00f41 THJ |
2156 | |
2157 | if (!path) | |
2158 | path = "/sys/fs/bpf"; | |
2159 | ||
2160 | len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map)); | |
2161 | if (len < 0) | |
2162 | return -EINVAL; | |
2163 | else if (len >= PATH_MAX) | |
2164 | return -ENAMETOOLONG; | |
2165 | ||
6e9cab2e | 2166 | return bpf_map__set_pin_path(map, buf); |
57a00f41 THJ |
2167 | } |
2168 | ||
c7ef5ec9 AN |
2169 | int parse_btf_map_def(const char *map_name, struct btf *btf, |
2170 | const struct btf_type *def_t, bool strict, | |
2171 | struct btf_map_def *map_def, struct btf_map_def *inner_def) | |
abd29c93 | 2172 | { |
41017e56 | 2173 | const struct btf_type *t; |
abd29c93 | 2174 | const struct btf_member *m; |
c7ef5ec9 | 2175 | bool is_inner = inner_def == NULL; |
abd29c93 AN |
2176 | int vlen, i; |
2177 | ||
c7ef5ec9 AN |
2178 | vlen = btf_vlen(def_t); |
2179 | m = btf_members(def_t); | |
abd29c93 | 2180 | for (i = 0; i < vlen; i++, m++) { |
c7ef5ec9 | 2181 | const char *name = btf__name_by_offset(btf, m->name_off); |
abd29c93 AN |
2182 | |
2183 | if (!name) { | |
c7ef5ec9 | 2184 | pr_warn("map '%s': invalid field #%d.\n", map_name, i); |
abd29c93 AN |
2185 | return -EINVAL; |
2186 | } | |
2187 | if (strcmp(name, "type") == 0) { | |
c7ef5ec9 | 2188 | if (!get_map_field_int(map_name, btf, m, &map_def->map_type)) |
abd29c93 | 2189 | return -EINVAL; |
c7ef5ec9 | 2190 | map_def->parts |= MAP_DEF_MAP_TYPE; |
abd29c93 | 2191 | } else if (strcmp(name, "max_entries") == 0) { |
c7ef5ec9 | 2192 | if (!get_map_field_int(map_name, btf, m, &map_def->max_entries)) |
abd29c93 | 2193 | return -EINVAL; |
c7ef5ec9 | 2194 | map_def->parts |= MAP_DEF_MAX_ENTRIES; |
abd29c93 | 2195 | } else if (strcmp(name, "map_flags") == 0) { |
c7ef5ec9 | 2196 | if (!get_map_field_int(map_name, btf, m, &map_def->map_flags)) |
abd29c93 | 2197 | return -EINVAL; |
c7ef5ec9 | 2198 | map_def->parts |= MAP_DEF_MAP_FLAGS; |
1bdb6c9a | 2199 | } else if (strcmp(name, "numa_node") == 0) { |
c7ef5ec9 | 2200 | if (!get_map_field_int(map_name, btf, m, &map_def->numa_node)) |
1bdb6c9a | 2201 | return -EINVAL; |
c7ef5ec9 | 2202 | map_def->parts |= MAP_DEF_NUMA_NODE; |
abd29c93 AN |
2203 | } else if (strcmp(name, "key_size") == 0) { |
2204 | __u32 sz; | |
2205 | ||
c7ef5ec9 | 2206 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2207 | return -EINVAL; |
c7ef5ec9 | 2208 | if (map_def->key_size && map_def->key_size != sz) { |
be18010e | 2209 | pr_warn("map '%s': conflicting key size %u != %u.\n", |
c7ef5ec9 | 2210 | map_name, map_def->key_size, sz); |
abd29c93 AN |
2211 | return -EINVAL; |
2212 | } | |
c7ef5ec9 AN |
2213 | map_def->key_size = sz; |
2214 | map_def->parts |= MAP_DEF_KEY_SIZE; | |
abd29c93 AN |
2215 | } else if (strcmp(name, "key") == 0) { |
2216 | __s64 sz; | |
2217 | ||
c7ef5ec9 | 2218 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2219 | if (!t) { |
be18010e | 2220 | pr_warn("map '%s': key type [%d] not found.\n", |
c7ef5ec9 | 2221 | map_name, m->type); |
abd29c93 AN |
2222 | return -EINVAL; |
2223 | } | |
b03bc685 | 2224 | if (!btf_is_ptr(t)) { |
81ba0889 | 2225 | pr_warn("map '%s': key spec is not PTR: %s.\n", |
c7ef5ec9 | 2226 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2227 | return -EINVAL; |
2228 | } | |
c7ef5ec9 | 2229 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2230 | if (sz < 0) { |
679152d3 | 2231 | pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n", |
c7ef5ec9 | 2232 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2233 | return sz; |
2234 | } | |
c7ef5ec9 | 2235 | if (map_def->key_size && map_def->key_size != sz) { |
679152d3 | 2236 | pr_warn("map '%s': conflicting key size %u != %zd.\n", |
c7ef5ec9 | 2237 | map_name, map_def->key_size, (ssize_t)sz); |
abd29c93 AN |
2238 | return -EINVAL; |
2239 | } | |
c7ef5ec9 AN |
2240 | map_def->key_size = sz; |
2241 | map_def->key_type_id = t->type; | |
2242 | map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE; | |
abd29c93 AN |
2243 | } else if (strcmp(name, "value_size") == 0) { |
2244 | __u32 sz; | |
2245 | ||
c7ef5ec9 | 2246 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2247 | return -EINVAL; |
c7ef5ec9 | 2248 | if (map_def->value_size && map_def->value_size != sz) { |
be18010e | 2249 | pr_warn("map '%s': conflicting value size %u != %u.\n", |
c7ef5ec9 | 2250 | map_name, map_def->value_size, sz); |
abd29c93 AN |
2251 | return -EINVAL; |
2252 | } | |
c7ef5ec9 AN |
2253 | map_def->value_size = sz; |
2254 | map_def->parts |= MAP_DEF_VALUE_SIZE; | |
abd29c93 AN |
2255 | } else if (strcmp(name, "value") == 0) { |
2256 | __s64 sz; | |
2257 | ||
c7ef5ec9 | 2258 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2259 | if (!t) { |
be18010e | 2260 | pr_warn("map '%s': value type [%d] not found.\n", |
c7ef5ec9 | 2261 | map_name, m->type); |
abd29c93 AN |
2262 | return -EINVAL; |
2263 | } | |
b03bc685 | 2264 | if (!btf_is_ptr(t)) { |
81ba0889 | 2265 | pr_warn("map '%s': value spec is not PTR: %s.\n", |
c7ef5ec9 | 2266 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2267 | return -EINVAL; |
2268 | } | |
c7ef5ec9 | 2269 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2270 | if (sz < 0) { |
679152d3 | 2271 | pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n", |
c7ef5ec9 | 2272 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2273 | return sz; |
2274 | } | |
c7ef5ec9 | 2275 | if (map_def->value_size && map_def->value_size != sz) { |
679152d3 | 2276 | pr_warn("map '%s': conflicting value size %u != %zd.\n", |
c7ef5ec9 | 2277 | map_name, map_def->value_size, (ssize_t)sz); |
abd29c93 AN |
2278 | return -EINVAL; |
2279 | } | |
c7ef5ec9 AN |
2280 | map_def->value_size = sz; |
2281 | map_def->value_type_id = t->type; | |
2282 | map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE; | |
646f02ff AN |
2283 | } |
2284 | else if (strcmp(name, "values") == 0) { | |
341ac5ff HC |
2285 | bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type); |
2286 | bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY; | |
2287 | const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value"; | |
c7ef5ec9 | 2288 | char inner_map_name[128]; |
646f02ff AN |
2289 | int err; |
2290 | ||
2291 | if (is_inner) { | |
2292 | pr_warn("map '%s': multi-level inner maps not supported.\n", | |
c7ef5ec9 | 2293 | map_name); |
646f02ff AN |
2294 | return -ENOTSUP; |
2295 | } | |
2296 | if (i != vlen - 1) { | |
2297 | pr_warn("map '%s': '%s' member should be last.\n", | |
c7ef5ec9 | 2298 | map_name, name); |
646f02ff AN |
2299 | return -EINVAL; |
2300 | } | |
341ac5ff HC |
2301 | if (!is_map_in_map && !is_prog_array) { |
2302 | pr_warn("map '%s': should be map-in-map or prog-array.\n", | |
c7ef5ec9 | 2303 | map_name); |
646f02ff AN |
2304 | return -ENOTSUP; |
2305 | } | |
c7ef5ec9 | 2306 | if (map_def->value_size && map_def->value_size != 4) { |
646f02ff | 2307 | pr_warn("map '%s': conflicting value size %u != 4.\n", |
c7ef5ec9 | 2308 | map_name, map_def->value_size); |
646f02ff AN |
2309 | return -EINVAL; |
2310 | } | |
c7ef5ec9 AN |
2311 | map_def->value_size = 4; |
2312 | t = btf__type_by_id(btf, m->type); | |
646f02ff | 2313 | if (!t) { |
341ac5ff HC |
2314 | pr_warn("map '%s': %s type [%d] not found.\n", |
2315 | map_name, desc, m->type); | |
646f02ff AN |
2316 | return -EINVAL; |
2317 | } | |
2318 | if (!btf_is_array(t) || btf_array(t)->nelems) { | |
341ac5ff HC |
2319 | pr_warn("map '%s': %s spec is not a zero-sized array.\n", |
2320 | map_name, desc); | |
646f02ff AN |
2321 | return -EINVAL; |
2322 | } | |
c7ef5ec9 | 2323 | t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL); |
646f02ff | 2324 | if (!btf_is_ptr(t)) { |
341ac5ff HC |
2325 | pr_warn("map '%s': %s def is of unexpected kind %s.\n", |
2326 | map_name, desc, btf_kind_str(t)); | |
646f02ff AN |
2327 | return -EINVAL; |
2328 | } | |
c7ef5ec9 | 2329 | t = skip_mods_and_typedefs(btf, t->type, NULL); |
341ac5ff HC |
2330 | if (is_prog_array) { |
2331 | if (!btf_is_func_proto(t)) { | |
2332 | pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n", | |
2333 | map_name, btf_kind_str(t)); | |
2334 | return -EINVAL; | |
2335 | } | |
2336 | continue; | |
2337 | } | |
646f02ff | 2338 | if (!btf_is_struct(t)) { |
81ba0889 | 2339 | pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n", |
c7ef5ec9 | 2340 | map_name, btf_kind_str(t)); |
646f02ff AN |
2341 | return -EINVAL; |
2342 | } | |
2343 | ||
c7ef5ec9 AN |
2344 | snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name); |
2345 | err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL); | |
646f02ff AN |
2346 | if (err) |
2347 | return err; | |
c7ef5ec9 AN |
2348 | |
2349 | map_def->parts |= MAP_DEF_INNER_MAP; | |
57a00f41 THJ |
2350 | } else if (strcmp(name, "pinning") == 0) { |
2351 | __u32 val; | |
57a00f41 | 2352 | |
646f02ff | 2353 | if (is_inner) { |
c7ef5ec9 | 2354 | pr_warn("map '%s': inner def can't be pinned.\n", map_name); |
646f02ff AN |
2355 | return -EINVAL; |
2356 | } | |
c7ef5ec9 | 2357 | if (!get_map_field_int(map_name, btf, m, &val)) |
57a00f41 | 2358 | return -EINVAL; |
c7ef5ec9 | 2359 | if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) { |
57a00f41 | 2360 | pr_warn("map '%s': invalid pinning value %u.\n", |
c7ef5ec9 | 2361 | map_name, val); |
57a00f41 THJ |
2362 | return -EINVAL; |
2363 | } | |
c7ef5ec9 AN |
2364 | map_def->pinning = val; |
2365 | map_def->parts |= MAP_DEF_PINNING; | |
47512102 JK |
2366 | } else if (strcmp(name, "map_extra") == 0) { |
2367 | __u32 map_extra; | |
2368 | ||
2369 | if (!get_map_field_int(map_name, btf, m, &map_extra)) | |
2370 | return -EINVAL; | |
2371 | map_def->map_extra = map_extra; | |
2372 | map_def->parts |= MAP_DEF_MAP_EXTRA; | |
abd29c93 AN |
2373 | } else { |
2374 | if (strict) { | |
c7ef5ec9 | 2375 | pr_warn("map '%s': unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2376 | return -ENOTSUP; |
2377 | } | |
c7ef5ec9 | 2378 | pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2379 | } |
2380 | } | |
2381 | ||
c7ef5ec9 AN |
2382 | if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) { |
2383 | pr_warn("map '%s': map type isn't specified.\n", map_name); | |
abd29c93 AN |
2384 | return -EINVAL; |
2385 | } | |
2386 | ||
2387 | return 0; | |
2388 | } | |
2389 | ||
c7ef5ec9 AN |
2390 | static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def) |
2391 | { | |
2392 | map->def.type = def->map_type; | |
2393 | map->def.key_size = def->key_size; | |
2394 | map->def.value_size = def->value_size; | |
2395 | map->def.max_entries = def->max_entries; | |
2396 | map->def.map_flags = def->map_flags; | |
47512102 | 2397 | map->map_extra = def->map_extra; |
c7ef5ec9 AN |
2398 | |
2399 | map->numa_node = def->numa_node; | |
2400 | map->btf_key_type_id = def->key_type_id; | |
2401 | map->btf_value_type_id = def->value_type_id; | |
2402 | ||
2403 | if (def->parts & MAP_DEF_MAP_TYPE) | |
2404 | pr_debug("map '%s': found type = %u.\n", map->name, def->map_type); | |
2405 | ||
2406 | if (def->parts & MAP_DEF_KEY_TYPE) | |
2407 | pr_debug("map '%s': found key [%u], sz = %u.\n", | |
2408 | map->name, def->key_type_id, def->key_size); | |
2409 | else if (def->parts & MAP_DEF_KEY_SIZE) | |
2410 | pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size); | |
2411 | ||
2412 | if (def->parts & MAP_DEF_VALUE_TYPE) | |
2413 | pr_debug("map '%s': found value [%u], sz = %u.\n", | |
2414 | map->name, def->value_type_id, def->value_size); | |
2415 | else if (def->parts & MAP_DEF_VALUE_SIZE) | |
2416 | pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size); | |
2417 | ||
2418 | if (def->parts & MAP_DEF_MAX_ENTRIES) | |
2419 | pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries); | |
2420 | if (def->parts & MAP_DEF_MAP_FLAGS) | |
47512102 JK |
2421 | pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags); |
2422 | if (def->parts & MAP_DEF_MAP_EXTRA) | |
2423 | pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name, | |
2424 | (unsigned long long)def->map_extra); | |
c7ef5ec9 AN |
2425 | if (def->parts & MAP_DEF_PINNING) |
2426 | pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning); | |
2427 | if (def->parts & MAP_DEF_NUMA_NODE) | |
2428 | pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node); | |
2429 | ||
2430 | if (def->parts & MAP_DEF_INNER_MAP) | |
2431 | pr_debug("map '%s': found inner map definition.\n", map->name); | |
2432 | } | |
2433 | ||
c1cccec9 AN |
2434 | static const char *btf_var_linkage_str(__u32 linkage) |
2435 | { | |
2436 | switch (linkage) { | |
2437 | case BTF_VAR_STATIC: return "static"; | |
2438 | case BTF_VAR_GLOBAL_ALLOCATED: return "global"; | |
2439 | case BTF_VAR_GLOBAL_EXTERN: return "extern"; | |
2440 | default: return "unknown"; | |
2441 | } | |
2442 | } | |
2443 | ||
41017e56 AN |
2444 | static int bpf_object__init_user_btf_map(struct bpf_object *obj, |
2445 | const struct btf_type *sec, | |
2446 | int var_idx, int sec_idx, | |
2447 | const Elf_Data *data, bool strict, | |
2448 | const char *pin_root_path) | |
2449 | { | |
c7ef5ec9 | 2450 | struct btf_map_def map_def = {}, inner_def = {}; |
41017e56 AN |
2451 | const struct btf_type *var, *def; |
2452 | const struct btf_var_secinfo *vi; | |
2453 | const struct btf_var *var_extra; | |
2454 | const char *map_name; | |
2455 | struct bpf_map *map; | |
c7ef5ec9 | 2456 | int err; |
41017e56 AN |
2457 | |
2458 | vi = btf_var_secinfos(sec) + var_idx; | |
2459 | var = btf__type_by_id(obj->btf, vi->type); | |
2460 | var_extra = btf_var(var); | |
2461 | map_name = btf__name_by_offset(obj->btf, var->name_off); | |
2462 | ||
2463 | if (map_name == NULL || map_name[0] == '\0') { | |
2464 | pr_warn("map #%d: empty name.\n", var_idx); | |
2465 | return -EINVAL; | |
2466 | } | |
2467 | if ((__u64)vi->offset + vi->size > data->d_size) { | |
2468 | pr_warn("map '%s' BTF data is corrupted.\n", map_name); | |
2469 | return -EINVAL; | |
2470 | } | |
2471 | if (!btf_is_var(var)) { | |
81ba0889 AN |
2472 | pr_warn("map '%s': unexpected var kind %s.\n", |
2473 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2474 | return -EINVAL; |
2475 | } | |
c1cccec9 AN |
2476 | if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) { |
2477 | pr_warn("map '%s': unsupported map linkage %s.\n", | |
2478 | map_name, btf_var_linkage_str(var_extra->linkage)); | |
41017e56 AN |
2479 | return -EOPNOTSUPP; |
2480 | } | |
2481 | ||
2482 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
2483 | if (!btf_is_struct(def)) { | |
81ba0889 AN |
2484 | pr_warn("map '%s': unexpected def kind %s.\n", |
2485 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2486 | return -EINVAL; |
2487 | } | |
2488 | if (def->size > vi->size) { | |
2489 | pr_warn("map '%s': invalid def size.\n", map_name); | |
2490 | return -EINVAL; | |
2491 | } | |
2492 | ||
2493 | map = bpf_object__add_map(obj); | |
2494 | if (IS_ERR(map)) | |
2495 | return PTR_ERR(map); | |
2496 | map->name = strdup(map_name); | |
2497 | if (!map->name) { | |
2498 | pr_warn("map '%s': failed to alloc map name.\n", map_name); | |
2499 | return -ENOMEM; | |
2500 | } | |
2501 | map->libbpf_type = LIBBPF_MAP_UNSPEC; | |
2502 | map->def.type = BPF_MAP_TYPE_UNSPEC; | |
2503 | map->sec_idx = sec_idx; | |
2504 | map->sec_offset = vi->offset; | |
646f02ff | 2505 | map->btf_var_idx = var_idx; |
41017e56 AN |
2506 | pr_debug("map '%s': at sec_idx %d, offset %zu.\n", |
2507 | map_name, map->sec_idx, map->sec_offset); | |
2508 | ||
c7ef5ec9 AN |
2509 | err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def); |
2510 | if (err) | |
2511 | return err; | |
2512 | ||
2513 | fill_map_from_def(map, &map_def); | |
2514 | ||
2515 | if (map_def.pinning == LIBBPF_PIN_BY_NAME) { | |
2516 | err = build_map_pin_path(map, pin_root_path); | |
2517 | if (err) { | |
2518 | pr_warn("map '%s': couldn't build pin path.\n", map->name); | |
2519 | return err; | |
2520 | } | |
2521 | } | |
2522 | ||
2523 | if (map_def.parts & MAP_DEF_INNER_MAP) { | |
2524 | map->inner_map = calloc(1, sizeof(*map->inner_map)); | |
2525 | if (!map->inner_map) | |
2526 | return -ENOMEM; | |
2527 | map->inner_map->fd = -1; | |
2528 | map->inner_map->sec_idx = sec_idx; | |
2529 | map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1); | |
2530 | if (!map->inner_map->name) | |
2531 | return -ENOMEM; | |
2532 | sprintf(map->inner_map->name, "%s.inner", map_name); | |
2533 | ||
2534 | fill_map_from_def(map->inner_map, &inner_def); | |
2535 | } | |
2536 | ||
2537 | return 0; | |
41017e56 AN |
2538 | } |
2539 | ||
57a00f41 THJ |
2540 | static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict, |
2541 | const char *pin_root_path) | |
abd29c93 AN |
2542 | { |
2543 | const struct btf_type *sec = NULL; | |
2544 | int nr_types, i, vlen, err; | |
2545 | const struct btf_type *t; | |
2546 | const char *name; | |
2547 | Elf_Data *data; | |
2548 | Elf_Scn *scn; | |
2549 | ||
2550 | if (obj->efile.btf_maps_shndx < 0) | |
2551 | return 0; | |
2552 | ||
88a82120 AN |
2553 | scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx); |
2554 | data = elf_sec_data(obj, scn); | |
abd29c93 | 2555 | if (!scn || !data) { |
88a82120 AN |
2556 | pr_warn("elf: failed to get %s map definitions for %s\n", |
2557 | MAPS_ELF_SEC, obj->path); | |
abd29c93 AN |
2558 | return -EINVAL; |
2559 | } | |
2560 | ||
6a886de0 HC |
2561 | nr_types = btf__type_cnt(obj->btf); |
2562 | for (i = 1; i < nr_types; i++) { | |
abd29c93 | 2563 | t = btf__type_by_id(obj->btf, i); |
b03bc685 | 2564 | if (!btf_is_datasec(t)) |
abd29c93 AN |
2565 | continue; |
2566 | name = btf__name_by_offset(obj->btf, t->name_off); | |
2567 | if (strcmp(name, MAPS_ELF_SEC) == 0) { | |
2568 | sec = t; | |
646f02ff | 2569 | obj->efile.btf_maps_sec_btf_id = i; |
abd29c93 AN |
2570 | break; |
2571 | } | |
2572 | } | |
2573 | ||
2574 | if (!sec) { | |
be18010e | 2575 | pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC); |
abd29c93 AN |
2576 | return -ENOENT; |
2577 | } | |
2578 | ||
b03bc685 | 2579 | vlen = btf_vlen(sec); |
abd29c93 AN |
2580 | for (i = 0; i < vlen; i++) { |
2581 | err = bpf_object__init_user_btf_map(obj, sec, i, | |
2582 | obj->efile.btf_maps_shndx, | |
8983b731 AN |
2583 | data, strict, |
2584 | pin_root_path); | |
abd29c93 AN |
2585 | if (err) |
2586 | return err; | |
2587 | } | |
2588 | ||
2589 | return 0; | |
2590 | } | |
2591 | ||
0d13bfce | 2592 | static int bpf_object__init_maps(struct bpf_object *obj, |
01af3bf0 | 2593 | const struct bpf_object_open_opts *opts) |
bf829271 | 2594 | { |
166750bc AN |
2595 | const char *pin_root_path; |
2596 | bool strict; | |
bf829271 | 2597 | int err; |
8837fe5d | 2598 | |
166750bc AN |
2599 | strict = !OPTS_GET(opts, relaxed_maps, false); |
2600 | pin_root_path = OPTS_GET(opts, pin_root_path, NULL); | |
bf829271 | 2601 | |
166750bc AN |
2602 | err = bpf_object__init_user_maps(obj, strict); |
2603 | err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path); | |
2604 | err = err ?: bpf_object__init_global_data_maps(obj); | |
81bfdd08 | 2605 | err = err ?: bpf_object__init_kconfig_map(obj); |
590a0088 | 2606 | err = err ?: bpf_object__init_struct_ops_maps(obj); |
bf829271 | 2607 | |
3b3af91c | 2608 | return err; |
561bbcca WN |
2609 | } |
2610 | ||
e3d91b0c JDB |
2611 | static bool section_have_execinstr(struct bpf_object *obj, int idx) |
2612 | { | |
ad23b723 | 2613 | Elf64_Shdr *sh; |
e3d91b0c | 2614 | |
ad23b723 AN |
2615 | sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx)); |
2616 | if (!sh) | |
e3d91b0c JDB |
2617 | return false; |
2618 | ||
ad23b723 | 2619 | return sh->sh_flags & SHF_EXECINSTR; |
e3d91b0c JDB |
2620 | } |
2621 | ||
0f0e55d8 AN |
2622 | static bool btf_needs_sanitization(struct bpf_object *obj) |
2623 | { | |
9ca1f56a AS |
2624 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
2625 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
2626 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
2627 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 2628 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
2dc1e488 | 2629 | bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG); |
0f0e55d8 | 2630 | |
2dc1e488 YS |
2631 | return !has_func || !has_datasec || !has_func_global || !has_float || |
2632 | !has_decl_tag || !has_type_tag; | |
0f0e55d8 AN |
2633 | } |
2634 | ||
2635 | static void bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf) | |
d7c4b398 | 2636 | { |
9ca1f56a AS |
2637 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
2638 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
2639 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
2640 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 2641 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
2dc1e488 | 2642 | bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG); |
d7c4b398 AN |
2643 | struct btf_type *t; |
2644 | int i, j, vlen; | |
d7c4b398 | 2645 | |
6a886de0 | 2646 | for (i = 1; i < btf__type_cnt(btf); i++) { |
d7c4b398 | 2647 | t = (struct btf_type *)btf__type_by_id(btf, i); |
d7c4b398 | 2648 | |
223f903e YS |
2649 | if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) { |
2650 | /* replace VAR/DECL_TAG with INT */ | |
d7c4b398 | 2651 | t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0); |
1d4126c4 AN |
2652 | /* |
2653 | * using size = 1 is the safest choice, 4 will be too | |
2654 | * big and cause kernel BTF validation failure if | |
2655 | * original variable took less than 4 bytes | |
2656 | */ | |
2657 | t->size = 1; | |
708852dc | 2658 | *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8); |
b03bc685 | 2659 | } else if (!has_datasec && btf_is_datasec(t)) { |
d7c4b398 | 2660 | /* replace DATASEC with STRUCT */ |
b03bc685 AN |
2661 | const struct btf_var_secinfo *v = btf_var_secinfos(t); |
2662 | struct btf_member *m = btf_members(t); | |
d7c4b398 AN |
2663 | struct btf_type *vt; |
2664 | char *name; | |
2665 | ||
2666 | name = (char *)btf__name_by_offset(btf, t->name_off); | |
2667 | while (*name) { | |
2668 | if (*name == '.') | |
2669 | *name = '_'; | |
2670 | name++; | |
2671 | } | |
2672 | ||
b03bc685 | 2673 | vlen = btf_vlen(t); |
d7c4b398 AN |
2674 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen); |
2675 | for (j = 0; j < vlen; j++, v++, m++) { | |
2676 | /* order of field assignments is important */ | |
2677 | m->offset = v->offset * 8; | |
2678 | m->type = v->type; | |
2679 | /* preserve variable name as member name */ | |
2680 | vt = (void *)btf__type_by_id(btf, v->type); | |
2681 | m->name_off = vt->name_off; | |
2682 | } | |
b03bc685 | 2683 | } else if (!has_func && btf_is_func_proto(t)) { |
d7c4b398 | 2684 | /* replace FUNC_PROTO with ENUM */ |
b03bc685 | 2685 | vlen = btf_vlen(t); |
d7c4b398 AN |
2686 | t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen); |
2687 | t->size = sizeof(__u32); /* kernel enforced */ | |
b03bc685 | 2688 | } else if (!has_func && btf_is_func(t)) { |
d7c4b398 AN |
2689 | /* replace FUNC with TYPEDEF */ |
2690 | t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0); | |
2d3eb67f AS |
2691 | } else if (!has_func_global && btf_is_func(t)) { |
2692 | /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */ | |
2693 | t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0); | |
22541a9e IL |
2694 | } else if (!has_float && btf_is_float(t)) { |
2695 | /* replace FLOAT with an equally-sized empty STRUCT; | |
2696 | * since C compilers do not accept e.g. "float" as a | |
2697 | * valid struct name, make it anonymous | |
2698 | */ | |
2699 | t->name_off = 0; | |
2700 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0); | |
2dc1e488 YS |
2701 | } else if (!has_type_tag && btf_is_type_tag(t)) { |
2702 | /* replace TYPE_TAG with a CONST */ | |
2703 | t->name_off = 0; | |
2704 | t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0); | |
d7c4b398 AN |
2705 | } |
2706 | } | |
2707 | } | |
2708 | ||
b35f14f4 | 2709 | static bool libbpf_needs_btf(const struct bpf_object *obj) |
abd29c93 | 2710 | { |
b35f14f4 AN |
2711 | return obj->efile.btf_maps_shndx >= 0 || |
2712 | obj->efile.st_ops_shndx >= 0 || | |
2713 | obj->nr_extern > 0; | |
2714 | } | |
2715 | ||
2716 | static bool kernel_needs_btf(const struct bpf_object *obj) | |
2717 | { | |
2718 | return obj->efile.st_ops_shndx >= 0; | |
abd29c93 AN |
2719 | } |
2720 | ||
063183bf | 2721 | static int bpf_object__init_btf(struct bpf_object *obj, |
9c6660d0 AN |
2722 | Elf_Data *btf_data, |
2723 | Elf_Data *btf_ext_data) | |
2724 | { | |
b7d7f3e1 | 2725 | int err = -ENOENT; |
9c6660d0 AN |
2726 | |
2727 | if (btf_data) { | |
2728 | obj->btf = btf__new(btf_data->d_buf, btf_data->d_size); | |
e9fc3ce9 AN |
2729 | err = libbpf_get_error(obj->btf); |
2730 | if (err) { | |
b7d7f3e1 | 2731 | obj->btf = NULL; |
e9fc3ce9 | 2732 | pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err); |
9c6660d0 AN |
2733 | goto out; |
2734 | } | |
4c01925f AN |
2735 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
2736 | btf__set_pointer_size(obj->btf, 8); | |
9c6660d0 AN |
2737 | } |
2738 | if (btf_ext_data) { | |
2739 | if (!obj->btf) { | |
2740 | pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n", | |
2741 | BTF_EXT_ELF_SEC, BTF_ELF_SEC); | |
2742 | goto out; | |
2743 | } | |
e9fc3ce9 AN |
2744 | obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size); |
2745 | err = libbpf_get_error(obj->btf_ext); | |
2746 | if (err) { | |
2747 | pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n", | |
2748 | BTF_EXT_ELF_SEC, err); | |
9c6660d0 AN |
2749 | obj->btf_ext = NULL; |
2750 | goto out; | |
2751 | } | |
9c6660d0 AN |
2752 | } |
2753 | out: | |
b35f14f4 | 2754 | if (err && libbpf_needs_btf(obj)) { |
be18010e | 2755 | pr_warn("BTF is required, but is missing or corrupted.\n"); |
b7d7f3e1 | 2756 | return err; |
abd29c93 | 2757 | } |
9c6660d0 AN |
2758 | return 0; |
2759 | } | |
2760 | ||
b96c07f3 AN |
2761 | static int compare_vsi_off(const void *_a, const void *_b) |
2762 | { | |
2763 | const struct btf_var_secinfo *a = _a; | |
2764 | const struct btf_var_secinfo *b = _b; | |
2765 | ||
2766 | return a->offset - b->offset; | |
2767 | } | |
2768 | ||
2769 | static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf, | |
2770 | struct btf_type *t) | |
2771 | { | |
2772 | __u32 size = 0, off = 0, i, vars = btf_vlen(t); | |
2773 | const char *name = btf__name_by_offset(btf, t->name_off); | |
2774 | const struct btf_type *t_var; | |
2775 | struct btf_var_secinfo *vsi; | |
2776 | const struct btf_var *var; | |
2777 | int ret; | |
2778 | ||
2779 | if (!name) { | |
2780 | pr_debug("No name found in string section for DATASEC kind.\n"); | |
2781 | return -ENOENT; | |
2782 | } | |
2783 | ||
2784 | /* .extern datasec size and var offsets were set correctly during | |
2785 | * extern collection step, so just skip straight to sorting variables | |
2786 | */ | |
2787 | if (t->size) | |
2788 | goto sort_vars; | |
2789 | ||
2790 | ret = find_elf_sec_sz(obj, name, &size); | |
6966d4c4 | 2791 | if (ret || !size) { |
b96c07f3 AN |
2792 | pr_debug("Invalid size for section %s: %u bytes\n", name, size); |
2793 | return -ENOENT; | |
2794 | } | |
2795 | ||
2796 | t->size = size; | |
2797 | ||
2798 | for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) { | |
2799 | t_var = btf__type_by_id(btf, vsi->type); | |
88918dc1 | 2800 | if (!t_var || !btf_is_var(t_var)) { |
b96c07f3 AN |
2801 | pr_debug("Non-VAR type seen in section %s\n", name); |
2802 | return -EINVAL; | |
2803 | } | |
2804 | ||
88918dc1 | 2805 | var = btf_var(t_var); |
b96c07f3 AN |
2806 | if (var->linkage == BTF_VAR_STATIC) |
2807 | continue; | |
2808 | ||
2809 | name = btf__name_by_offset(btf, t_var->name_off); | |
2810 | if (!name) { | |
2811 | pr_debug("No name found in string section for VAR kind\n"); | |
2812 | return -ENOENT; | |
2813 | } | |
2814 | ||
2815 | ret = find_elf_var_offset(obj, name, &off); | |
2816 | if (ret) { | |
2817 | pr_debug("No offset found in symbol table for VAR %s\n", | |
2818 | name); | |
2819 | return -ENOENT; | |
2820 | } | |
2821 | ||
2822 | vsi->offset = off; | |
2823 | } | |
2824 | ||
2825 | sort_vars: | |
2826 | qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off); | |
2827 | return 0; | |
2828 | } | |
2829 | ||
2830 | static int btf_finalize_data(struct bpf_object *obj, struct btf *btf) | |
2831 | { | |
2832 | int err = 0; | |
6a886de0 | 2833 | __u32 i, n = btf__type_cnt(btf); |
b96c07f3 | 2834 | |
6a886de0 | 2835 | for (i = 1; i < n; i++) { |
b96c07f3 AN |
2836 | struct btf_type *t = btf_type_by_id(btf, i); |
2837 | ||
2838 | /* Loader needs to fix up some of the things compiler | |
2839 | * couldn't get its hands on while emitting BTF. This | |
2840 | * is section size and global variable offset. We use | |
2841 | * the info from the ELF itself for this purpose. | |
2842 | */ | |
2843 | if (btf_is_datasec(t)) { | |
2844 | err = btf_fixup_datasec(obj, btf, t); | |
2845 | if (err) | |
2846 | break; | |
2847 | } | |
2848 | } | |
2849 | ||
2850 | return libbpf_err(err); | |
2851 | } | |
2852 | ||
2853 | int btf__finalize_data(struct bpf_object *obj, struct btf *btf) | |
2854 | { | |
2855 | return btf_finalize_data(obj, btf); | |
2856 | } | |
2857 | ||
166750bc AN |
2858 | static int bpf_object__finalize_btf(struct bpf_object *obj) |
2859 | { | |
2860 | int err; | |
2861 | ||
2862 | if (!obj->btf) | |
2863 | return 0; | |
2864 | ||
b96c07f3 | 2865 | err = btf_finalize_data(obj, obj->btf); |
bfc96656 AN |
2866 | if (err) { |
2867 | pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err); | |
2868 | return err; | |
166750bc | 2869 | } |
bfc96656 | 2870 | |
166750bc AN |
2871 | return 0; |
2872 | } | |
2873 | ||
fe62de31 | 2874 | static bool prog_needs_vmlinux_btf(struct bpf_program *prog) |
a6ed02ca | 2875 | { |
1e092a03 KS |
2876 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS || |
2877 | prog->type == BPF_PROG_TYPE_LSM) | |
a6ed02ca KS |
2878 | return true; |
2879 | ||
2880 | /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs | |
2881 | * also need vmlinux BTF | |
2882 | */ | |
2883 | if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd) | |
2884 | return true; | |
2885 | ||
2886 | return false; | |
2887 | } | |
2888 | ||
fe62de31 | 2889 | static bool obj_needs_vmlinux_btf(const struct bpf_object *obj) |
a6ed02ca KS |
2890 | { |
2891 | struct bpf_program *prog; | |
fe62de31 | 2892 | int i; |
a6ed02ca | 2893 | |
1373ff59 SC |
2894 | /* CO-RE relocations need kernel BTF, only when btf_custom_path |
2895 | * is not specified | |
2896 | */ | |
2897 | if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path) | |
fe62de31 | 2898 | return true; |
192b6638 | 2899 | |
d370bbe1 HL |
2900 | /* Support for typed ksyms needs kernel BTF */ |
2901 | for (i = 0; i < obj->nr_extern; i++) { | |
2902 | const struct extern_desc *ext; | |
2903 | ||
2904 | ext = &obj->externs[i]; | |
fe62de31 AN |
2905 | if (ext->type == EXT_KSYM && ext->ksym.type_id) |
2906 | return true; | |
d370bbe1 HL |
2907 | } |
2908 | ||
a6ed02ca | 2909 | bpf_object__for_each_program(prog, obj) { |
d9297581 AN |
2910 | if (!prog->load) |
2911 | continue; | |
fe62de31 AN |
2912 | if (prog_needs_vmlinux_btf(prog)) |
2913 | return true; | |
a6ed02ca KS |
2914 | } |
2915 | ||
fe62de31 AN |
2916 | return false; |
2917 | } | |
2918 | ||
2919 | static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force) | |
2920 | { | |
2921 | int err; | |
2922 | ||
2923 | /* btf_vmlinux could be loaded earlier */ | |
67234743 | 2924 | if (obj->btf_vmlinux || obj->gen_loader) |
fe62de31 AN |
2925 | return 0; |
2926 | ||
2927 | if (!force && !obj_needs_vmlinux_btf(obj)) | |
192b6638 AN |
2928 | return 0; |
2929 | ||
a710eed3 | 2930 | obj->btf_vmlinux = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
2931 | err = libbpf_get_error(obj->btf_vmlinux); |
2932 | if (err) { | |
192b6638 AN |
2933 | pr_warn("Error loading vmlinux BTF: %d\n", err); |
2934 | obj->btf_vmlinux = NULL; | |
2935 | return err; | |
2936 | } | |
a6ed02ca KS |
2937 | return 0; |
2938 | } | |
2939 | ||
063183bf AN |
2940 | static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj) |
2941 | { | |
0f0e55d8 AN |
2942 | struct btf *kern_btf = obj->btf; |
2943 | bool btf_mandatory, sanitize; | |
aea28a60 | 2944 | int i, err = 0; |
063183bf AN |
2945 | |
2946 | if (!obj->btf) | |
2947 | return 0; | |
2948 | ||
9ca1f56a | 2949 | if (!kernel_supports(obj, FEAT_BTF)) { |
68b08647 AN |
2950 | if (kernel_needs_btf(obj)) { |
2951 | err = -EOPNOTSUPP; | |
2952 | goto report; | |
2953 | } | |
2954 | pr_debug("Kernel doesn't support BTF, skipping uploading it.\n"); | |
2955 | return 0; | |
2956 | } | |
2957 | ||
aea28a60 AN |
2958 | /* Even though some subprogs are global/weak, user might prefer more |
2959 | * permissive BPF verification process that BPF verifier performs for | |
2960 | * static functions, taking into account more context from the caller | |
2961 | * functions. In such case, they need to mark such subprogs with | |
2962 | * __attribute__((visibility("hidden"))) and libbpf will adjust | |
2963 | * corresponding FUNC BTF type to be marked as static and trigger more | |
2964 | * involved BPF verification process. | |
2965 | */ | |
2966 | for (i = 0; i < obj->nr_programs; i++) { | |
2967 | struct bpf_program *prog = &obj->programs[i]; | |
2968 | struct btf_type *t; | |
2969 | const char *name; | |
2970 | int j, n; | |
2971 | ||
2972 | if (!prog->mark_btf_static || !prog_is_subprog(obj, prog)) | |
2973 | continue; | |
2974 | ||
6a886de0 HC |
2975 | n = btf__type_cnt(obj->btf); |
2976 | for (j = 1; j < n; j++) { | |
aea28a60 AN |
2977 | t = btf_type_by_id(obj->btf, j); |
2978 | if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) | |
2979 | continue; | |
2980 | ||
2981 | name = btf__str_by_offset(obj->btf, t->name_off); | |
2982 | if (strcmp(name, prog->name) != 0) | |
2983 | continue; | |
2984 | ||
2985 | t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0); | |
2986 | break; | |
2987 | } | |
2988 | } | |
2989 | ||
0f0e55d8 AN |
2990 | sanitize = btf_needs_sanitization(obj); |
2991 | if (sanitize) { | |
5c3320d7 | 2992 | const void *raw_data; |
0f0e55d8 | 2993 | __u32 sz; |
063183bf | 2994 | |
0f0e55d8 | 2995 | /* clone BTF to sanitize a copy and leave the original intact */ |
6a886de0 | 2996 | raw_data = btf__raw_data(obj->btf, &sz); |
5c3320d7 | 2997 | kern_btf = btf__new(raw_data, sz); |
e9fc3ce9 AN |
2998 | err = libbpf_get_error(kern_btf); |
2999 | if (err) | |
3000 | return err; | |
04efe591 | 3001 | |
4c01925f AN |
3002 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
3003 | btf__set_pointer_size(obj->btf, 8); | |
0f0e55d8 | 3004 | bpf_object__sanitize_btf(obj, kern_btf); |
063183bf | 3005 | } |
0f0e55d8 | 3006 | |
67234743 AS |
3007 | if (obj->gen_loader) { |
3008 | __u32 raw_size = 0; | |
6a886de0 | 3009 | const void *raw_data = btf__raw_data(kern_btf, &raw_size); |
67234743 AS |
3010 | |
3011 | if (!raw_data) | |
3012 | return -ENOMEM; | |
3013 | bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size); | |
3014 | /* Pretend to have valid FD to pass various fd >= 0 checks. | |
3015 | * This fd == 0 will not be used with any syscall and will be reset to -1 eventually. | |
3016 | */ | |
3017 | btf__set_fd(kern_btf, 0); | |
3018 | } else { | |
e0e3ea88 AN |
3019 | /* currently BPF_BTF_LOAD only supports log_level 1 */ |
3020 | err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size, | |
3021 | obj->log_level ? 1 : 0); | |
67234743 | 3022 | } |
0f0e55d8 AN |
3023 | if (sanitize) { |
3024 | if (!err) { | |
3025 | /* move fd to libbpf's BTF */ | |
3026 | btf__set_fd(obj->btf, btf__fd(kern_btf)); | |
3027 | btf__set_fd(kern_btf, -1); | |
3028 | } | |
3029 | btf__free(kern_btf); | |
3030 | } | |
68b08647 | 3031 | report: |
0f0e55d8 AN |
3032 | if (err) { |
3033 | btf_mandatory = kernel_needs_btf(obj); | |
3034 | pr_warn("Error loading .BTF into kernel: %d. %s\n", err, | |
3035 | btf_mandatory ? "BTF is mandatory, can't proceed." | |
3036 | : "BTF is optional, ignoring."); | |
3037 | if (!btf_mandatory) | |
3038 | err = 0; | |
3039 | } | |
3040 | return err; | |
063183bf AN |
3041 | } |
3042 | ||
88a82120 AN |
3043 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off) |
3044 | { | |
3045 | const char *name; | |
3046 | ||
3047 | name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off); | |
3048 | if (!name) { | |
3049 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3050 | off, obj->path, elf_errmsg(-1)); | |
3051 | return NULL; | |
3052 | } | |
3053 | ||
3054 | return name; | |
3055 | } | |
3056 | ||
3057 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off) | |
3058 | { | |
3059 | const char *name; | |
3060 | ||
3061 | name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off); | |
3062 | if (!name) { | |
3063 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3064 | off, obj->path, elf_errmsg(-1)); | |
3065 | return NULL; | |
3066 | } | |
3067 | ||
3068 | return name; | |
3069 | } | |
3070 | ||
3071 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx) | |
3072 | { | |
3073 | Elf_Scn *scn; | |
3074 | ||
3075 | scn = elf_getscn(obj->efile.elf, idx); | |
3076 | if (!scn) { | |
3077 | pr_warn("elf: failed to get section(%zu) from %s: %s\n", | |
3078 | idx, obj->path, elf_errmsg(-1)); | |
3079 | return NULL; | |
3080 | } | |
3081 | return scn; | |
3082 | } | |
3083 | ||
3084 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name) | |
3085 | { | |
3086 | Elf_Scn *scn = NULL; | |
3087 | Elf *elf = obj->efile.elf; | |
3088 | const char *sec_name; | |
3089 | ||
3090 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
3091 | sec_name = elf_sec_name(obj, scn); | |
3092 | if (!sec_name) | |
3093 | return NULL; | |
3094 | ||
3095 | if (strcmp(sec_name, name) != 0) | |
3096 | continue; | |
3097 | ||
3098 | return scn; | |
3099 | } | |
3100 | return NULL; | |
3101 | } | |
3102 | ||
ad23b723 | 3103 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn) |
88a82120 | 3104 | { |
ad23b723 AN |
3105 | Elf64_Shdr *shdr; |
3106 | ||
88a82120 | 3107 | if (!scn) |
ad23b723 | 3108 | return NULL; |
88a82120 | 3109 | |
ad23b723 AN |
3110 | shdr = elf64_getshdr(scn); |
3111 | if (!shdr) { | |
88a82120 AN |
3112 | pr_warn("elf: failed to get section(%zu) header from %s: %s\n", |
3113 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
ad23b723 | 3114 | return NULL; |
88a82120 AN |
3115 | } |
3116 | ||
ad23b723 | 3117 | return shdr; |
88a82120 AN |
3118 | } |
3119 | ||
3120 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn) | |
3121 | { | |
3122 | const char *name; | |
ad23b723 | 3123 | Elf64_Shdr *sh; |
88a82120 AN |
3124 | |
3125 | if (!scn) | |
3126 | return NULL; | |
3127 | ||
ad23b723 AN |
3128 | sh = elf_sec_hdr(obj, scn); |
3129 | if (!sh) | |
88a82120 AN |
3130 | return NULL; |
3131 | ||
ad23b723 | 3132 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 AN |
3133 | if (!name) { |
3134 | pr_warn("elf: failed to get section(%zu) name from %s: %s\n", | |
3135 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
3136 | return NULL; | |
3137 | } | |
3138 | ||
3139 | return name; | |
3140 | } | |
3141 | ||
3142 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn) | |
3143 | { | |
3144 | Elf_Data *data; | |
3145 | ||
3146 | if (!scn) | |
3147 | return NULL; | |
3148 | ||
3149 | data = elf_getdata(scn, 0); | |
3150 | if (!data) { | |
3151 | pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n", | |
3152 | elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>", | |
3153 | obj->path, elf_errmsg(-1)); | |
3154 | return NULL; | |
3155 | } | |
3156 | ||
3157 | return data; | |
3158 | } | |
3159 | ||
ad23b723 AN |
3160 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx) |
3161 | { | |
3162 | if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym)) | |
3163 | return NULL; | |
3164 | ||
3165 | return (Elf64_Sym *)obj->efile.symbols->d_buf + idx; | |
3166 | } | |
3167 | ||
3168 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx) | |
3169 | { | |
3170 | if (idx >= data->d_size / sizeof(Elf64_Rel)) | |
3171 | return NULL; | |
3172 | ||
3173 | return (Elf64_Rel *)data->d_buf + idx; | |
3174 | } | |
3175 | ||
50e09460 AN |
3176 | static bool is_sec_name_dwarf(const char *name) |
3177 | { | |
3178 | /* approximation, but the actual list is too long */ | |
13d35a0c | 3179 | return str_has_pfx(name, ".debug_"); |
50e09460 AN |
3180 | } |
3181 | ||
ad23b723 | 3182 | static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name) |
50e09460 AN |
3183 | { |
3184 | /* no special handling of .strtab */ | |
3185 | if (hdr->sh_type == SHT_STRTAB) | |
3186 | return true; | |
3187 | ||
3188 | /* ignore .llvm_addrsig section as well */ | |
faf6ed32 | 3189 | if (hdr->sh_type == SHT_LLVM_ADDRSIG) |
50e09460 AN |
3190 | return true; |
3191 | ||
3192 | /* no subprograms will lead to an empty .text section, ignore it */ | |
3193 | if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 && | |
3194 | strcmp(name, ".text") == 0) | |
3195 | return true; | |
3196 | ||
3197 | /* DWARF sections */ | |
3198 | if (is_sec_name_dwarf(name)) | |
3199 | return true; | |
3200 | ||
13d35a0c | 3201 | if (str_has_pfx(name, ".rel")) { |
50e09460 AN |
3202 | name += sizeof(".rel") - 1; |
3203 | /* DWARF section relocations */ | |
3204 | if (is_sec_name_dwarf(name)) | |
3205 | return true; | |
3206 | ||
3207 | /* .BTF and .BTF.ext don't need relocations */ | |
3208 | if (strcmp(name, BTF_ELF_SEC) == 0 || | |
3209 | strcmp(name, BTF_EXT_ELF_SEC) == 0) | |
3210 | return true; | |
3211 | } | |
3212 | ||
3213 | return false; | |
3214 | } | |
3215 | ||
db2b8b06 AN |
3216 | static int cmp_progs(const void *_a, const void *_b) |
3217 | { | |
3218 | const struct bpf_program *a = _a; | |
3219 | const struct bpf_program *b = _b; | |
3220 | ||
3221 | if (a->sec_idx != b->sec_idx) | |
3222 | return a->sec_idx < b->sec_idx ? -1 : 1; | |
3223 | ||
3224 | /* sec_insn_off can't be the same within the section */ | |
3225 | return a->sec_insn_off < b->sec_insn_off ? -1 : 1; | |
3226 | } | |
3227 | ||
0d13bfce | 3228 | static int bpf_object__elf_collect(struct bpf_object *obj) |
29603665 | 3229 | { |
25bbbd7a | 3230 | struct elf_sec_desc *sec_desc; |
29603665 | 3231 | Elf *elf = obj->efile.elf; |
f0187f0b | 3232 | Elf_Data *btf_ext_data = NULL; |
1713d68b | 3233 | Elf_Data *btf_data = NULL; |
666810e8 | 3234 | int idx = 0, err = 0; |
0201c575 AN |
3235 | const char *name; |
3236 | Elf_Data *data; | |
3237 | Elf_Scn *scn; | |
ad23b723 | 3238 | Elf64_Shdr *sh; |
29603665 | 3239 | |
0d6988e1 AN |
3240 | /* ELF section indices are 0-based, but sec #0 is special "invalid" |
3241 | * section. e_shnum does include sec #0, so e_shnum is the necessary | |
3242 | * size of an array to keep all the sections. | |
25bbbd7a | 3243 | */ |
0d6988e1 | 3244 | obj->efile.sec_cnt = obj->efile.ehdr->e_shnum; |
25bbbd7a AN |
3245 | obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs)); |
3246 | if (!obj->efile.secs) | |
3247 | return -ENOMEM; | |
29603665 | 3248 | |
0201c575 AN |
3249 | /* a bunch of ELF parsing functionality depends on processing symbols, |
3250 | * so do the first pass and find the symbol table | |
3251 | */ | |
3252 | scn = NULL; | |
29603665 | 3253 | while ((scn = elf_nextscn(elf, scn)) != NULL) { |
ad23b723 AN |
3254 | sh = elf_sec_hdr(obj, scn); |
3255 | if (!sh) | |
0201c575 AN |
3256 | return -LIBBPF_ERRNO__FORMAT; |
3257 | ||
ad23b723 | 3258 | if (sh->sh_type == SHT_SYMTAB) { |
0201c575 AN |
3259 | if (obj->efile.symbols) { |
3260 | pr_warn("elf: multiple symbol tables in %s\n", obj->path); | |
3261 | return -LIBBPF_ERRNO__FORMAT; | |
3262 | } | |
29603665 | 3263 | |
0201c575 AN |
3264 | data = elf_sec_data(obj, scn); |
3265 | if (!data) | |
3266 | return -LIBBPF_ERRNO__FORMAT; | |
3267 | ||
25bbbd7a AN |
3268 | idx = elf_ndxscn(scn); |
3269 | ||
0201c575 | 3270 | obj->efile.symbols = data; |
25bbbd7a | 3271 | obj->efile.symbols_shndx = idx; |
ad23b723 | 3272 | obj->efile.strtabidx = sh->sh_link; |
0201c575 AN |
3273 | } |
3274 | } | |
3275 | ||
03e601f4 THJ |
3276 | if (!obj->efile.symbols) { |
3277 | pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n", | |
3278 | obj->path); | |
3279 | return -ENOENT; | |
3280 | } | |
3281 | ||
0201c575 AN |
3282 | scn = NULL; |
3283 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
25bbbd7a AN |
3284 | idx = elf_ndxscn(scn); |
3285 | sec_desc = &obj->efile.secs[idx]; | |
88a82120 | 3286 | |
ad23b723 AN |
3287 | sh = elf_sec_hdr(obj, scn); |
3288 | if (!sh) | |
01b29d1d | 3289 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3290 | |
ad23b723 | 3291 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 | 3292 | if (!name) |
01b29d1d | 3293 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3294 | |
ad23b723 | 3295 | if (ignore_elf_section(sh, name)) |
50e09460 AN |
3296 | continue; |
3297 | ||
88a82120 AN |
3298 | data = elf_sec_data(obj, scn); |
3299 | if (!data) | |
01b29d1d | 3300 | return -LIBBPF_ERRNO__FORMAT; |
88a82120 AN |
3301 | |
3302 | pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n", | |
077c066a | 3303 | idx, name, (unsigned long)data->d_size, |
ad23b723 AN |
3304 | (int)sh->sh_link, (unsigned long)sh->sh_flags, |
3305 | (int)sh->sh_type); | |
cb1e5e96 | 3306 | |
1713d68b | 3307 | if (strcmp(name, "license") == 0) { |
88a82120 | 3308 | err = bpf_object__init_license(obj, data->d_buf, data->d_size); |
01b29d1d AN |
3309 | if (err) |
3310 | return err; | |
1713d68b | 3311 | } else if (strcmp(name, "version") == 0) { |
88a82120 | 3312 | err = bpf_object__init_kversion(obj, data->d_buf, data->d_size); |
54b8625c JF |
3313 | if (err) |
3314 | return err; | |
1713d68b | 3315 | } else if (strcmp(name, "maps") == 0) { |
666810e8 | 3316 | obj->efile.maps_shndx = idx; |
abd29c93 AN |
3317 | } else if (strcmp(name, MAPS_ELF_SEC) == 0) { |
3318 | obj->efile.btf_maps_shndx = idx; | |
1713d68b | 3319 | } else if (strcmp(name, BTF_ELF_SEC) == 0) { |
62554d52 AN |
3320 | if (sh->sh_type != SHT_PROGBITS) |
3321 | return -LIBBPF_ERRNO__FORMAT; | |
1713d68b | 3322 | btf_data = data; |
2993e051 | 3323 | } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) { |
62554d52 AN |
3324 | if (sh->sh_type != SHT_PROGBITS) |
3325 | return -LIBBPF_ERRNO__FORMAT; | |
f0187f0b | 3326 | btf_ext_data = data; |
ad23b723 | 3327 | } else if (sh->sh_type == SHT_SYMTAB) { |
0201c575 | 3328 | /* already processed during the first pass above */ |
ad23b723 AN |
3329 | } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) { |
3330 | if (sh->sh_flags & SHF_EXECINSTR) { | |
f8c7a4d4 JS |
3331 | if (strcmp(name, ".text") == 0) |
3332 | obj->efile.text_shndx = idx; | |
c1122392 | 3333 | err = bpf_object__add_programs(obj, data, name, idx); |
88a82120 | 3334 | if (err) |
01b29d1d | 3335 | return err; |
aed65917 AN |
3336 | } else if (strcmp(name, DATA_SEC) == 0 || |
3337 | str_has_pfx(name, DATA_SEC ".")) { | |
25bbbd7a AN |
3338 | sec_desc->sec_type = SEC_DATA; |
3339 | sec_desc->shdr = sh; | |
3340 | sec_desc->data = data; | |
aed65917 AN |
3341 | } else if (strcmp(name, RODATA_SEC) == 0 || |
3342 | str_has_pfx(name, RODATA_SEC ".")) { | |
25bbbd7a AN |
3343 | sec_desc->sec_type = SEC_RODATA; |
3344 | sec_desc->shdr = sh; | |
3345 | sec_desc->data = data; | |
590a0088 MKL |
3346 | } else if (strcmp(name, STRUCT_OPS_SEC) == 0) { |
3347 | obj->efile.st_ops_data = data; | |
3348 | obj->efile.st_ops_shndx = idx; | |
d859900c | 3349 | } else { |
50e09460 AN |
3350 | pr_info("elf: skipping unrecognized data section(%d) %s\n", |
3351 | idx, name); | |
a5b8bd47 | 3352 | } |
ad23b723 | 3353 | } else if (sh->sh_type == SHT_REL) { |
25bbbd7a | 3354 | int targ_sec_idx = sh->sh_info; /* points to other section */ |
e3d91b0c | 3355 | |
b7332d28 AN |
3356 | if (sh->sh_entsize != sizeof(Elf64_Rel) || |
3357 | targ_sec_idx >= obj->efile.sec_cnt) | |
3358 | return -LIBBPF_ERRNO__FORMAT; | |
3359 | ||
e3d91b0c | 3360 | /* Only do relo for section with exec instructions */ |
25bbbd7a | 3361 | if (!section_have_execinstr(obj, targ_sec_idx) && |
646f02ff AN |
3362 | strcmp(name, ".rel" STRUCT_OPS_SEC) && |
3363 | strcmp(name, ".rel" MAPS_ELF_SEC)) { | |
50e09460 | 3364 | pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n", |
25bbbd7a AN |
3365 | idx, name, targ_sec_idx, |
3366 | elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>"); | |
e3d91b0c JDB |
3367 | continue; |
3368 | } | |
b62f06e8 | 3369 | |
25bbbd7a AN |
3370 | sec_desc->sec_type = SEC_RELO; |
3371 | sec_desc->shdr = sh; | |
3372 | sec_desc->data = data; | |
ad23b723 | 3373 | } else if (sh->sh_type == SHT_NOBITS && strcmp(name, BSS_SEC) == 0) { |
25bbbd7a AN |
3374 | sec_desc->sec_type = SEC_BSS; |
3375 | sec_desc->shdr = sh; | |
3376 | sec_desc->data = data; | |
077c066a | 3377 | } else { |
2e80be60 | 3378 | pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name, |
ad23b723 | 3379 | (size_t)sh->sh_size); |
bec7d68c | 3380 | } |
29603665 | 3381 | } |
561bbcca | 3382 | |
d3a3aa0c | 3383 | if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) { |
88a82120 | 3384 | pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path); |
f102154d | 3385 | return -LIBBPF_ERRNO__FORMAT; |
77ba9a5b | 3386 | } |
db2b8b06 AN |
3387 | |
3388 | /* sort BPF programs by section name and in-section instruction offset | |
3389 | * for faster search */ | |
2a6a9bf2 AN |
3390 | if (obj->nr_programs) |
3391 | qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs); | |
db2b8b06 | 3392 | |
0d13bfce | 3393 | return bpf_object__init_btf(obj, btf_data, btf_ext_data); |
29603665 WN |
3394 | } |
3395 | ||
ad23b723 | 3396 | static bool sym_is_extern(const Elf64_Sym *sym) |
166750bc | 3397 | { |
ad23b723 | 3398 | int bind = ELF64_ST_BIND(sym->st_info); |
166750bc AN |
3399 | /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */ |
3400 | return sym->st_shndx == SHN_UNDEF && | |
3401 | (bind == STB_GLOBAL || bind == STB_WEAK) && | |
ad23b723 | 3402 | ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE; |
166750bc AN |
3403 | } |
3404 | ||
ad23b723 | 3405 | static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx) |
53eddb5e | 3406 | { |
ad23b723 AN |
3407 | int bind = ELF64_ST_BIND(sym->st_info); |
3408 | int type = ELF64_ST_TYPE(sym->st_info); | |
53eddb5e YS |
3409 | |
3410 | /* in .text section */ | |
3411 | if (sym->st_shndx != text_shndx) | |
3412 | return false; | |
3413 | ||
3414 | /* local function */ | |
3415 | if (bind == STB_LOCAL && type == STT_SECTION) | |
3416 | return true; | |
3417 | ||
3418 | /* global function */ | |
3419 | return bind == STB_GLOBAL && type == STT_FUNC; | |
3420 | } | |
3421 | ||
166750bc AN |
3422 | static int find_extern_btf_id(const struct btf *btf, const char *ext_name) |
3423 | { | |
3424 | const struct btf_type *t; | |
5bd022ec | 3425 | const char *tname; |
166750bc AN |
3426 | int i, n; |
3427 | ||
3428 | if (!btf) | |
3429 | return -ESRCH; | |
3430 | ||
6a886de0 HC |
3431 | n = btf__type_cnt(btf); |
3432 | for (i = 1; i < n; i++) { | |
166750bc AN |
3433 | t = btf__type_by_id(btf, i); |
3434 | ||
5bd022ec | 3435 | if (!btf_is_var(t) && !btf_is_func(t)) |
166750bc AN |
3436 | continue; |
3437 | ||
5bd022ec MKL |
3438 | tname = btf__name_by_offset(btf, t->name_off); |
3439 | if (strcmp(tname, ext_name)) | |
166750bc AN |
3440 | continue; |
3441 | ||
5bd022ec MKL |
3442 | if (btf_is_var(t) && |
3443 | btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN) | |
3444 | return -EINVAL; | |
3445 | ||
3446 | if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN) | |
166750bc AN |
3447 | return -EINVAL; |
3448 | ||
3449 | return i; | |
3450 | } | |
3451 | ||
3452 | return -ENOENT; | |
3453 | } | |
3454 | ||
2e33efe3 AN |
3455 | static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) { |
3456 | const struct btf_var_secinfo *vs; | |
3457 | const struct btf_type *t; | |
3458 | int i, j, n; | |
3459 | ||
3460 | if (!btf) | |
3461 | return -ESRCH; | |
3462 | ||
6a886de0 HC |
3463 | n = btf__type_cnt(btf); |
3464 | for (i = 1; i < n; i++) { | |
2e33efe3 AN |
3465 | t = btf__type_by_id(btf, i); |
3466 | ||
3467 | if (!btf_is_datasec(t)) | |
3468 | continue; | |
3469 | ||
3470 | vs = btf_var_secinfos(t); | |
3471 | for (j = 0; j < btf_vlen(t); j++, vs++) { | |
3472 | if (vs->type == ext_btf_id) | |
3473 | return i; | |
3474 | } | |
3475 | } | |
3476 | ||
3477 | return -ENOENT; | |
3478 | } | |
3479 | ||
3480 | static enum kcfg_type find_kcfg_type(const struct btf *btf, int id, | |
3481 | bool *is_signed) | |
166750bc AN |
3482 | { |
3483 | const struct btf_type *t; | |
3484 | const char *name; | |
3485 | ||
3486 | t = skip_mods_and_typedefs(btf, id, NULL); | |
3487 | name = btf__name_by_offset(btf, t->name_off); | |
3488 | ||
3489 | if (is_signed) | |
3490 | *is_signed = false; | |
3491 | switch (btf_kind(t)) { | |
3492 | case BTF_KIND_INT: { | |
3493 | int enc = btf_int_encoding(t); | |
3494 | ||
3495 | if (enc & BTF_INT_BOOL) | |
2e33efe3 | 3496 | return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN; |
166750bc AN |
3497 | if (is_signed) |
3498 | *is_signed = enc & BTF_INT_SIGNED; | |
3499 | if (t->size == 1) | |
2e33efe3 | 3500 | return KCFG_CHAR; |
166750bc | 3501 | if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1))) |
2e33efe3 AN |
3502 | return KCFG_UNKNOWN; |
3503 | return KCFG_INT; | |
166750bc AN |
3504 | } |
3505 | case BTF_KIND_ENUM: | |
3506 | if (t->size != 4) | |
2e33efe3 | 3507 | return KCFG_UNKNOWN; |
166750bc | 3508 | if (strcmp(name, "libbpf_tristate")) |
2e33efe3 AN |
3509 | return KCFG_UNKNOWN; |
3510 | return KCFG_TRISTATE; | |
166750bc AN |
3511 | case BTF_KIND_ARRAY: |
3512 | if (btf_array(t)->nelems == 0) | |
2e33efe3 AN |
3513 | return KCFG_UNKNOWN; |
3514 | if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR) | |
3515 | return KCFG_UNKNOWN; | |
3516 | return KCFG_CHAR_ARR; | |
166750bc | 3517 | default: |
2e33efe3 | 3518 | return KCFG_UNKNOWN; |
166750bc AN |
3519 | } |
3520 | } | |
3521 | ||
3522 | static int cmp_externs(const void *_a, const void *_b) | |
3523 | { | |
3524 | const struct extern_desc *a = _a; | |
3525 | const struct extern_desc *b = _b; | |
3526 | ||
2e33efe3 AN |
3527 | if (a->type != b->type) |
3528 | return a->type < b->type ? -1 : 1; | |
3529 | ||
3530 | if (a->type == EXT_KCFG) { | |
3531 | /* descending order by alignment requirements */ | |
3532 | if (a->kcfg.align != b->kcfg.align) | |
3533 | return a->kcfg.align > b->kcfg.align ? -1 : 1; | |
3534 | /* ascending order by size, within same alignment class */ | |
3535 | if (a->kcfg.sz != b->kcfg.sz) | |
3536 | return a->kcfg.sz < b->kcfg.sz ? -1 : 1; | |
3537 | } | |
3538 | ||
166750bc AN |
3539 | /* resolve ties by name */ |
3540 | return strcmp(a->name, b->name); | |
3541 | } | |
3542 | ||
1c0c7074 AN |
3543 | static int find_int_btf_id(const struct btf *btf) |
3544 | { | |
3545 | const struct btf_type *t; | |
3546 | int i, n; | |
3547 | ||
6a886de0 HC |
3548 | n = btf__type_cnt(btf); |
3549 | for (i = 1; i < n; i++) { | |
1c0c7074 AN |
3550 | t = btf__type_by_id(btf, i); |
3551 | ||
3552 | if (btf_is_int(t) && btf_int_bits(t) == 32) | |
3553 | return i; | |
3554 | } | |
3555 | ||
3556 | return 0; | |
3557 | } | |
3558 | ||
5bd022ec MKL |
3559 | static int add_dummy_ksym_var(struct btf *btf) |
3560 | { | |
3561 | int i, int_btf_id, sec_btf_id, dummy_var_btf_id; | |
3562 | const struct btf_var_secinfo *vs; | |
3563 | const struct btf_type *sec; | |
3564 | ||
9683e577 IR |
3565 | if (!btf) |
3566 | return 0; | |
3567 | ||
5bd022ec MKL |
3568 | sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC, |
3569 | BTF_KIND_DATASEC); | |
3570 | if (sec_btf_id < 0) | |
3571 | return 0; | |
3572 | ||
3573 | sec = btf__type_by_id(btf, sec_btf_id); | |
3574 | vs = btf_var_secinfos(sec); | |
3575 | for (i = 0; i < btf_vlen(sec); i++, vs++) { | |
3576 | const struct btf_type *vt; | |
3577 | ||
3578 | vt = btf__type_by_id(btf, vs->type); | |
3579 | if (btf_is_func(vt)) | |
3580 | break; | |
3581 | } | |
3582 | ||
3583 | /* No func in ksyms sec. No need to add dummy var. */ | |
3584 | if (i == btf_vlen(sec)) | |
3585 | return 0; | |
3586 | ||
3587 | int_btf_id = find_int_btf_id(btf); | |
3588 | dummy_var_btf_id = btf__add_var(btf, | |
3589 | "dummy_ksym", | |
3590 | BTF_VAR_GLOBAL_ALLOCATED, | |
3591 | int_btf_id); | |
3592 | if (dummy_var_btf_id < 0) | |
3593 | pr_warn("cannot create a dummy_ksym var\n"); | |
3594 | ||
3595 | return dummy_var_btf_id; | |
3596 | } | |
3597 | ||
166750bc AN |
3598 | static int bpf_object__collect_externs(struct bpf_object *obj) |
3599 | { | |
1c0c7074 | 3600 | struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL; |
166750bc AN |
3601 | const struct btf_type *t; |
3602 | struct extern_desc *ext; | |
5bd022ec | 3603 | int i, n, off, dummy_var_btf_id; |
2e33efe3 | 3604 | const char *ext_name, *sec_name; |
166750bc | 3605 | Elf_Scn *scn; |
ad23b723 | 3606 | Elf64_Shdr *sh; |
166750bc AN |
3607 | |
3608 | if (!obj->efile.symbols) | |
3609 | return 0; | |
3610 | ||
88a82120 | 3611 | scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx); |
ad23b723 | 3612 | sh = elf_sec_hdr(obj, scn); |
83390787 | 3613 | if (!sh || sh->sh_entsize != sizeof(Elf64_Sym)) |
166750bc | 3614 | return -LIBBPF_ERRNO__FORMAT; |
166750bc | 3615 | |
5bd022ec MKL |
3616 | dummy_var_btf_id = add_dummy_ksym_var(obj->btf); |
3617 | if (dummy_var_btf_id < 0) | |
3618 | return dummy_var_btf_id; | |
3619 | ||
ad23b723 | 3620 | n = sh->sh_size / sh->sh_entsize; |
166750bc | 3621 | pr_debug("looking for externs among %d symbols...\n", n); |
88a82120 | 3622 | |
166750bc | 3623 | for (i = 0; i < n; i++) { |
ad23b723 | 3624 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
166750bc | 3625 | |
ad23b723 | 3626 | if (!sym) |
166750bc | 3627 | return -LIBBPF_ERRNO__FORMAT; |
ad23b723 | 3628 | if (!sym_is_extern(sym)) |
166750bc | 3629 | continue; |
ad23b723 | 3630 | ext_name = elf_sym_str(obj, sym->st_name); |
166750bc AN |
3631 | if (!ext_name || !ext_name[0]) |
3632 | continue; | |
3633 | ||
3634 | ext = obj->externs; | |
029258d7 | 3635 | ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext)); |
166750bc AN |
3636 | if (!ext) |
3637 | return -ENOMEM; | |
3638 | obj->externs = ext; | |
3639 | ext = &ext[obj->nr_extern]; | |
3640 | memset(ext, 0, sizeof(*ext)); | |
3641 | obj->nr_extern++; | |
3642 | ||
3643 | ext->btf_id = find_extern_btf_id(obj->btf, ext_name); | |
3644 | if (ext->btf_id <= 0) { | |
3645 | pr_warn("failed to find BTF for extern '%s': %d\n", | |
3646 | ext_name, ext->btf_id); | |
3647 | return ext->btf_id; | |
3648 | } | |
3649 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
3650 | ext->name = btf__name_by_offset(obj->btf, t->name_off); | |
3651 | ext->sym_idx = i; | |
ad23b723 | 3652 | ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK; |
2e33efe3 AN |
3653 | |
3654 | ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id); | |
3655 | if (ext->sec_btf_id <= 0) { | |
3656 | pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n", | |
3657 | ext_name, ext->btf_id, ext->sec_btf_id); | |
3658 | return ext->sec_btf_id; | |
166750bc | 3659 | } |
2e33efe3 AN |
3660 | sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id); |
3661 | sec_name = btf__name_by_offset(obj->btf, sec->name_off); | |
3662 | ||
3663 | if (strcmp(sec_name, KCONFIG_SEC) == 0) { | |
5bd022ec MKL |
3664 | if (btf_is_func(t)) { |
3665 | pr_warn("extern function %s is unsupported under %s section\n", | |
3666 | ext->name, KCONFIG_SEC); | |
3667 | return -ENOTSUP; | |
3668 | } | |
2e33efe3 AN |
3669 | kcfg_sec = sec; |
3670 | ext->type = EXT_KCFG; | |
3671 | ext->kcfg.sz = btf__resolve_size(obj->btf, t->type); | |
3672 | if (ext->kcfg.sz <= 0) { | |
3673 | pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n", | |
3674 | ext_name, ext->kcfg.sz); | |
3675 | return ext->kcfg.sz; | |
3676 | } | |
3677 | ext->kcfg.align = btf__align_of(obj->btf, t->type); | |
3678 | if (ext->kcfg.align <= 0) { | |
3679 | pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n", | |
3680 | ext_name, ext->kcfg.align); | |
3681 | return -EINVAL; | |
3682 | } | |
3683 | ext->kcfg.type = find_kcfg_type(obj->btf, t->type, | |
3684 | &ext->kcfg.is_signed); | |
3685 | if (ext->kcfg.type == KCFG_UNKNOWN) { | |
3686 | pr_warn("extern (kcfg) '%s' type is unsupported\n", ext_name); | |
3687 | return -ENOTSUP; | |
3688 | } | |
1c0c7074 | 3689 | } else if (strcmp(sec_name, KSYMS_SEC) == 0) { |
1c0c7074 AN |
3690 | ksym_sec = sec; |
3691 | ext->type = EXT_KSYM; | |
d370bbe1 HL |
3692 | skip_mods_and_typedefs(obj->btf, t->type, |
3693 | &ext->ksym.type_id); | |
2e33efe3 AN |
3694 | } else { |
3695 | pr_warn("unrecognized extern section '%s'\n", sec_name); | |
166750bc AN |
3696 | return -ENOTSUP; |
3697 | } | |
3698 | } | |
3699 | pr_debug("collected %d externs total\n", obj->nr_extern); | |
3700 | ||
3701 | if (!obj->nr_extern) | |
3702 | return 0; | |
3703 | ||
2e33efe3 | 3704 | /* sort externs by type, for kcfg ones also by (align, size, name) */ |
166750bc | 3705 | qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs); |
166750bc | 3706 | |
1c0c7074 AN |
3707 | /* for .ksyms section, we need to turn all externs into allocated |
3708 | * variables in BTF to pass kernel verification; we do this by | |
3709 | * pretending that each extern is a 8-byte variable | |
3710 | */ | |
3711 | if (ksym_sec) { | |
3712 | /* find existing 4-byte integer type in BTF to use for fake | |
3713 | * extern variables in DATASEC | |
3714 | */ | |
3715 | int int_btf_id = find_int_btf_id(obj->btf); | |
5bd022ec MKL |
3716 | /* For extern function, a dummy_var added earlier |
3717 | * will be used to replace the vs->type and | |
3718 | * its name string will be used to refill | |
3719 | * the missing param's name. | |
3720 | */ | |
3721 | const struct btf_type *dummy_var; | |
1c0c7074 | 3722 | |
5bd022ec | 3723 | dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id); |
1c0c7074 AN |
3724 | for (i = 0; i < obj->nr_extern; i++) { |
3725 | ext = &obj->externs[i]; | |
3726 | if (ext->type != EXT_KSYM) | |
3727 | continue; | |
3728 | pr_debug("extern (ksym) #%d: symbol %d, name %s\n", | |
3729 | i, ext->sym_idx, ext->name); | |
3730 | } | |
3731 | ||
3732 | sec = ksym_sec; | |
3733 | n = btf_vlen(sec); | |
3734 | for (i = 0, off = 0; i < n; i++, off += sizeof(int)) { | |
3735 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
3736 | struct btf_type *vt; | |
3737 | ||
3738 | vt = (void *)btf__type_by_id(obj->btf, vs->type); | |
3739 | ext_name = btf__name_by_offset(obj->btf, vt->name_off); | |
3740 | ext = find_extern_by_name(obj, ext_name); | |
3741 | if (!ext) { | |
5bd022ec MKL |
3742 | pr_warn("failed to find extern definition for BTF %s '%s'\n", |
3743 | btf_kind_str(vt), ext_name); | |
1c0c7074 AN |
3744 | return -ESRCH; |
3745 | } | |
5bd022ec MKL |
3746 | if (btf_is_func(vt)) { |
3747 | const struct btf_type *func_proto; | |
3748 | struct btf_param *param; | |
3749 | int j; | |
3750 | ||
3751 | func_proto = btf__type_by_id(obj->btf, | |
3752 | vt->type); | |
3753 | param = btf_params(func_proto); | |
3754 | /* Reuse the dummy_var string if the | |
3755 | * func proto does not have param name. | |
3756 | */ | |
3757 | for (j = 0; j < btf_vlen(func_proto); j++) | |
3758 | if (param[j].type && !param[j].name_off) | |
3759 | param[j].name_off = | |
3760 | dummy_var->name_off; | |
3761 | vs->type = dummy_var_btf_id; | |
3762 | vt->info &= ~0xffff; | |
3763 | vt->info |= BTF_FUNC_GLOBAL; | |
3764 | } else { | |
3765 | btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
3766 | vt->type = int_btf_id; | |
3767 | } | |
1c0c7074 AN |
3768 | vs->offset = off; |
3769 | vs->size = sizeof(int); | |
3770 | } | |
3771 | sec->size = off; | |
3772 | } | |
3773 | ||
2e33efe3 AN |
3774 | if (kcfg_sec) { |
3775 | sec = kcfg_sec; | |
3776 | /* for kcfg externs calculate their offsets within a .kconfig map */ | |
3777 | off = 0; | |
3778 | for (i = 0; i < obj->nr_extern; i++) { | |
3779 | ext = &obj->externs[i]; | |
3780 | if (ext->type != EXT_KCFG) | |
3781 | continue; | |
166750bc | 3782 | |
2e33efe3 AN |
3783 | ext->kcfg.data_off = roundup(off, ext->kcfg.align); |
3784 | off = ext->kcfg.data_off + ext->kcfg.sz; | |
1c0c7074 | 3785 | pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n", |
2e33efe3 AN |
3786 | i, ext->sym_idx, ext->kcfg.data_off, ext->name); |
3787 | } | |
3788 | sec->size = off; | |
3789 | n = btf_vlen(sec); | |
3790 | for (i = 0; i < n; i++) { | |
3791 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
3792 | ||
3793 | t = btf__type_by_id(obj->btf, vs->type); | |
3794 | ext_name = btf__name_by_offset(obj->btf, t->name_off); | |
3795 | ext = find_extern_by_name(obj, ext_name); | |
3796 | if (!ext) { | |
3797 | pr_warn("failed to find extern definition for BTF var '%s'\n", | |
3798 | ext_name); | |
3799 | return -ESRCH; | |
3800 | } | |
3801 | btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
3802 | vs->offset = ext->kcfg.data_off; | |
166750bc | 3803 | } |
166750bc | 3804 | } |
166750bc AN |
3805 | return 0; |
3806 | } | |
3807 | ||
6d4b198b | 3808 | struct bpf_program * |
a324aae3 AN |
3809 | bpf_object__find_program_by_title(const struct bpf_object *obj, |
3810 | const char *title) | |
6d4b198b JK |
3811 | { |
3812 | struct bpf_program *pos; | |
3813 | ||
3814 | bpf_object__for_each_program(pos, obj) { | |
52109584 | 3815 | if (pos->sec_name && !strcmp(pos->sec_name, title)) |
6d4b198b JK |
3816 | return pos; |
3817 | } | |
e9fc3ce9 | 3818 | return errno = ENOENT, NULL; |
6d4b198b JK |
3819 | } |
3820 | ||
c3c55696 AN |
3821 | static bool prog_is_subprog(const struct bpf_object *obj, |
3822 | const struct bpf_program *prog) | |
3823 | { | |
197afc63 AN |
3824 | /* For legacy reasons, libbpf supports an entry-point BPF programs |
3825 | * without SEC() attribute, i.e., those in the .text section. But if | |
3826 | * there are 2 or more such programs in the .text section, they all | |
3827 | * must be subprograms called from entry-point BPF programs in | |
3828 | * designated SEC()'tions, otherwise there is no way to distinguish | |
3829 | * which of those programs should be loaded vs which are a subprogram. | |
3830 | * Similarly, if there is a function/program in .text and at least one | |
3831 | * other BPF program with custom SEC() attribute, then we just assume | |
3832 | * .text programs are subprograms (even if they are not called from | |
3833 | * other programs), because libbpf never explicitly supported mixing | |
3834 | * SEC()-designated BPF programs and .text entry-point BPF programs. | |
3835 | */ | |
3836 | return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1; | |
c3c55696 AN |
3837 | } |
3838 | ||
01af3bf0 AN |
3839 | struct bpf_program * |
3840 | bpf_object__find_program_by_name(const struct bpf_object *obj, | |
3841 | const char *name) | |
3842 | { | |
3843 | struct bpf_program *prog; | |
3844 | ||
3845 | bpf_object__for_each_program(prog, obj) { | |
c3c55696 AN |
3846 | if (prog_is_subprog(obj, prog)) |
3847 | continue; | |
01af3bf0 AN |
3848 | if (!strcmp(prog->name, name)) |
3849 | return prog; | |
3850 | } | |
e9fc3ce9 | 3851 | return errno = ENOENT, NULL; |
01af3bf0 AN |
3852 | } |
3853 | ||
d859900c DB |
3854 | static bool bpf_object__shndx_is_data(const struct bpf_object *obj, |
3855 | int shndx) | |
3856 | { | |
25bbbd7a AN |
3857 | switch (obj->efile.secs[shndx].sec_type) { |
3858 | case SEC_BSS: | |
3859 | case SEC_DATA: | |
3860 | case SEC_RODATA: | |
3861 | return true; | |
3862 | default: | |
3863 | return false; | |
3864 | } | |
d859900c DB |
3865 | } |
3866 | ||
3867 | static bool bpf_object__shndx_is_maps(const struct bpf_object *obj, | |
3868 | int shndx) | |
3869 | { | |
abd29c93 AN |
3870 | return shndx == obj->efile.maps_shndx || |
3871 | shndx == obj->efile.btf_maps_shndx; | |
d859900c DB |
3872 | } |
3873 | ||
d859900c DB |
3874 | static enum libbpf_map_type |
3875 | bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx) | |
3876 | { | |
25bbbd7a AN |
3877 | if (shndx == obj->efile.symbols_shndx) |
3878 | return LIBBPF_MAP_KCONFIG; | |
3879 | ||
3880 | switch (obj->efile.secs[shndx].sec_type) { | |
3881 | case SEC_BSS: | |
d859900c | 3882 | return LIBBPF_MAP_BSS; |
25bbbd7a AN |
3883 | case SEC_DATA: |
3884 | return LIBBPF_MAP_DATA; | |
3885 | case SEC_RODATA: | |
d859900c | 3886 | return LIBBPF_MAP_RODATA; |
25bbbd7a | 3887 | default: |
d859900c | 3888 | return LIBBPF_MAP_UNSPEC; |
25bbbd7a | 3889 | } |
d859900c DB |
3890 | } |
3891 | ||
1f8e2bcb AN |
3892 | static int bpf_program__record_reloc(struct bpf_program *prog, |
3893 | struct reloc_desc *reloc_desc, | |
9c0f8cbd | 3894 | __u32 insn_idx, const char *sym_name, |
ad23b723 | 3895 | const Elf64_Sym *sym, const Elf64_Rel *rel) |
1f8e2bcb AN |
3896 | { |
3897 | struct bpf_insn *insn = &prog->insns[insn_idx]; | |
3898 | size_t map_idx, nr_maps = prog->obj->nr_maps; | |
3899 | struct bpf_object *obj = prog->obj; | |
3900 | __u32 shdr_idx = sym->st_shndx; | |
3901 | enum libbpf_map_type type; | |
9c0f8cbd | 3902 | const char *sym_sec_name; |
1f8e2bcb AN |
3903 | struct bpf_map *map; |
3904 | ||
aa0b8d43 | 3905 | if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) { |
9c0f8cbd AN |
3906 | pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n", |
3907 | prog->name, sym_name, insn_idx, insn->code); | |
1f8e2bcb AN |
3908 | return -LIBBPF_ERRNO__RELOC; |
3909 | } | |
166750bc AN |
3910 | |
3911 | if (sym_is_extern(sym)) { | |
ad23b723 | 3912 | int sym_idx = ELF64_R_SYM(rel->r_info); |
166750bc AN |
3913 | int i, n = obj->nr_extern; |
3914 | struct extern_desc *ext; | |
3915 | ||
3916 | for (i = 0; i < n; i++) { | |
3917 | ext = &obj->externs[i]; | |
3918 | if (ext->sym_idx == sym_idx) | |
3919 | break; | |
3920 | } | |
3921 | if (i >= n) { | |
9c0f8cbd AN |
3922 | pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n", |
3923 | prog->name, sym_name, sym_idx); | |
166750bc AN |
3924 | return -LIBBPF_ERRNO__RELOC; |
3925 | } | |
9c0f8cbd AN |
3926 | pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n", |
3927 | prog->name, i, ext->name, ext->sym_idx, insn_idx); | |
5bd022ec MKL |
3928 | if (insn->code == (BPF_JMP | BPF_CALL)) |
3929 | reloc_desc->type = RELO_EXTERN_FUNC; | |
3930 | else | |
3931 | reloc_desc->type = RELO_EXTERN_VAR; | |
166750bc | 3932 | reloc_desc->insn_idx = insn_idx; |
2e33efe3 | 3933 | reloc_desc->sym_off = i; /* sym_off stores extern index */ |
166750bc AN |
3934 | return 0; |
3935 | } | |
3936 | ||
aa0b8d43 MKL |
3937 | /* sub-program call relocation */ |
3938 | if (is_call_insn(insn)) { | |
3939 | if (insn->src_reg != BPF_PSEUDO_CALL) { | |
3940 | pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name); | |
3941 | return -LIBBPF_ERRNO__RELOC; | |
3942 | } | |
3943 | /* text_shndx can be 0, if no default "main" program exists */ | |
3944 | if (!shdr_idx || shdr_idx != obj->efile.text_shndx) { | |
3945 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); | |
3946 | pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n", | |
3947 | prog->name, sym_name, sym_sec_name); | |
3948 | return -LIBBPF_ERRNO__RELOC; | |
3949 | } | |
3950 | if (sym->st_value % BPF_INSN_SZ) { | |
3951 | pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n", | |
3952 | prog->name, sym_name, (size_t)sym->st_value); | |
3953 | return -LIBBPF_ERRNO__RELOC; | |
3954 | } | |
3955 | reloc_desc->type = RELO_CALL; | |
3956 | reloc_desc->insn_idx = insn_idx; | |
3957 | reloc_desc->sym_off = sym->st_value; | |
3958 | return 0; | |
3959 | } | |
3960 | ||
1f8e2bcb | 3961 | if (!shdr_idx || shdr_idx >= SHN_LORESERVE) { |
9c0f8cbd AN |
3962 | pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n", |
3963 | prog->name, sym_name, shdr_idx); | |
1f8e2bcb AN |
3964 | return -LIBBPF_ERRNO__RELOC; |
3965 | } | |
3966 | ||
53eddb5e YS |
3967 | /* loading subprog addresses */ |
3968 | if (sym_is_subprog(sym, obj->efile.text_shndx)) { | |
3969 | /* global_func: sym->st_value = offset in the section, insn->imm = 0. | |
3970 | * local_func: sym->st_value = 0, insn->imm = offset in the section. | |
3971 | */ | |
3972 | if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) { | |
3973 | pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n", | |
3974 | prog->name, sym_name, (size_t)sym->st_value, insn->imm); | |
3975 | return -LIBBPF_ERRNO__RELOC; | |
3976 | } | |
3977 | ||
3978 | reloc_desc->type = RELO_SUBPROG_ADDR; | |
3979 | reloc_desc->insn_idx = insn_idx; | |
3980 | reloc_desc->sym_off = sym->st_value; | |
3981 | return 0; | |
3982 | } | |
3983 | ||
1f8e2bcb | 3984 | type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx); |
9c0f8cbd | 3985 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); |
1f8e2bcb AN |
3986 | |
3987 | /* generic map reference relocation */ | |
3988 | if (type == LIBBPF_MAP_UNSPEC) { | |
3989 | if (!bpf_object__shndx_is_maps(obj, shdr_idx)) { | |
9c0f8cbd AN |
3990 | pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n", |
3991 | prog->name, sym_name, sym_sec_name); | |
1f8e2bcb AN |
3992 | return -LIBBPF_ERRNO__RELOC; |
3993 | } | |
3994 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { | |
3995 | map = &obj->maps[map_idx]; | |
3996 | if (map->libbpf_type != type || | |
3997 | map->sec_idx != sym->st_shndx || | |
3998 | map->sec_offset != sym->st_value) | |
3999 | continue; | |
9c0f8cbd AN |
4000 | pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n", |
4001 | prog->name, map_idx, map->name, map->sec_idx, | |
1f8e2bcb AN |
4002 | map->sec_offset, insn_idx); |
4003 | break; | |
4004 | } | |
4005 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
4006 | pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n", |
4007 | prog->name, sym_sec_name, (size_t)sym->st_value); | |
1f8e2bcb AN |
4008 | return -LIBBPF_ERRNO__RELOC; |
4009 | } | |
4010 | reloc_desc->type = RELO_LD64; | |
4011 | reloc_desc->insn_idx = insn_idx; | |
4012 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 4013 | reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */ |
1f8e2bcb AN |
4014 | return 0; |
4015 | } | |
4016 | ||
4017 | /* global data map relocation */ | |
4018 | if (!bpf_object__shndx_is_data(obj, shdr_idx)) { | |
9c0f8cbd AN |
4019 | pr_warn("prog '%s': bad data relo against section '%s'\n", |
4020 | prog->name, sym_sec_name); | |
1f8e2bcb | 4021 | return -LIBBPF_ERRNO__RELOC; |
1f8e2bcb | 4022 | } |
1f8e2bcb AN |
4023 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { |
4024 | map = &obj->maps[map_idx]; | |
25bbbd7a | 4025 | if (map->libbpf_type != type || map->sec_idx != sym->st_shndx) |
1f8e2bcb | 4026 | continue; |
9c0f8cbd AN |
4027 | pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n", |
4028 | prog->name, map_idx, map->name, map->sec_idx, | |
4029 | map->sec_offset, insn_idx); | |
1f8e2bcb AN |
4030 | break; |
4031 | } | |
4032 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
4033 | pr_warn("prog '%s': data relo failed to find map for section '%s'\n", |
4034 | prog->name, sym_sec_name); | |
1f8e2bcb AN |
4035 | return -LIBBPF_ERRNO__RELOC; |
4036 | } | |
4037 | ||
4038 | reloc_desc->type = RELO_DATA; | |
4039 | reloc_desc->insn_idx = insn_idx; | |
4040 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 4041 | reloc_desc->sym_off = sym->st_value; |
1f8e2bcb AN |
4042 | return 0; |
4043 | } | |
4044 | ||
db2b8b06 AN |
4045 | static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx) |
4046 | { | |
4047 | return insn_idx >= prog->sec_insn_off && | |
4048 | insn_idx < prog->sec_insn_off + prog->sec_insn_cnt; | |
4049 | } | |
4050 | ||
4051 | static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj, | |
4052 | size_t sec_idx, size_t insn_idx) | |
4053 | { | |
4054 | int l = 0, r = obj->nr_programs - 1, m; | |
4055 | struct bpf_program *prog; | |
4056 | ||
4057 | while (l < r) { | |
4058 | m = l + (r - l + 1) / 2; | |
4059 | prog = &obj->programs[m]; | |
4060 | ||
4061 | if (prog->sec_idx < sec_idx || | |
4062 | (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx)) | |
4063 | l = m; | |
4064 | else | |
4065 | r = m - 1; | |
4066 | } | |
4067 | /* matching program could be at index l, but it still might be the | |
4068 | * wrong one, so we need to double check conditions for the last time | |
4069 | */ | |
4070 | prog = &obj->programs[l]; | |
4071 | if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx)) | |
4072 | return prog; | |
4073 | return NULL; | |
4074 | } | |
4075 | ||
34090915 | 4076 | static int |
ad23b723 | 4077 | bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data) |
34090915 | 4078 | { |
9c0f8cbd | 4079 | const char *relo_sec_name, *sec_name; |
b7332d28 | 4080 | size_t sec_idx = shdr->sh_info, sym_idx; |
c3c55696 AN |
4081 | struct bpf_program *prog; |
4082 | struct reloc_desc *relos; | |
1f8e2bcb | 4083 | int err, i, nrels; |
c3c55696 AN |
4084 | const char *sym_name; |
4085 | __u32 insn_idx; | |
6245947c AN |
4086 | Elf_Scn *scn; |
4087 | Elf_Data *scn_data; | |
ad23b723 AN |
4088 | Elf64_Sym *sym; |
4089 | Elf64_Rel *rel; | |
34090915 | 4090 | |
b7332d28 AN |
4091 | if (sec_idx >= obj->efile.sec_cnt) |
4092 | return -EINVAL; | |
4093 | ||
6245947c AN |
4094 | scn = elf_sec_by_idx(obj, sec_idx); |
4095 | scn_data = elf_sec_data(obj, scn); | |
4096 | ||
9c0f8cbd | 4097 | relo_sec_name = elf_sec_str(obj, shdr->sh_name); |
6245947c | 4098 | sec_name = elf_sec_name(obj, scn); |
9c0f8cbd AN |
4099 | if (!relo_sec_name || !sec_name) |
4100 | return -EINVAL; | |
4101 | ||
4102 | pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n", | |
4103 | relo_sec_name, sec_idx, sec_name); | |
34090915 WN |
4104 | nrels = shdr->sh_size / shdr->sh_entsize; |
4105 | ||
34090915 | 4106 | for (i = 0; i < nrels; i++) { |
ad23b723 AN |
4107 | rel = elf_rel_by_idx(data, i); |
4108 | if (!rel) { | |
9c0f8cbd | 4109 | pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i); |
6371ca3b | 4110 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4111 | } |
ad23b723 | 4112 | |
b7332d28 AN |
4113 | sym_idx = ELF64_R_SYM(rel->r_info); |
4114 | sym = elf_sym_by_idx(obj, sym_idx); | |
ad23b723 | 4115 | if (!sym) { |
b7332d28 AN |
4116 | pr_warn("sec '%s': symbol #%zu not found for relo #%d\n", |
4117 | relo_sec_name, sym_idx, i); | |
4118 | return -LIBBPF_ERRNO__FORMAT; | |
4119 | } | |
4120 | ||
4121 | if (sym->st_shndx >= obj->efile.sec_cnt) { | |
4122 | pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n", | |
4123 | relo_sec_name, sym_idx, (size_t)sym->st_shndx, i); | |
6371ca3b | 4124 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4125 | } |
6245947c | 4126 | |
ad23b723 | 4127 | if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) { |
9c0f8cbd | 4128 | pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n", |
b7332d28 | 4129 | relo_sec_name, (size_t)rel->r_offset, i); |
1f8e2bcb | 4130 | return -LIBBPF_ERRNO__FORMAT; |
9c0f8cbd | 4131 | } |
d859900c | 4132 | |
ad23b723 | 4133 | insn_idx = rel->r_offset / BPF_INSN_SZ; |
c3c55696 AN |
4134 | /* relocations against static functions are recorded as |
4135 | * relocations against the section that contains a function; | |
4136 | * in such case, symbol will be STT_SECTION and sym.st_name | |
4137 | * will point to empty string (0), so fetch section name | |
4138 | * instead | |
4139 | */ | |
ad23b723 AN |
4140 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0) |
4141 | sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx)); | |
c3c55696 | 4142 | else |
ad23b723 | 4143 | sym_name = elf_sym_str(obj, sym->st_name); |
c3c55696 | 4144 | sym_name = sym_name ?: "<?"; |
d859900c | 4145 | |
9c0f8cbd AN |
4146 | pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n", |
4147 | relo_sec_name, i, insn_idx, sym_name); | |
666810e8 | 4148 | |
c3c55696 AN |
4149 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); |
4150 | if (!prog) { | |
6245947c | 4151 | pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n", |
c3c55696 | 4152 | relo_sec_name, i, sec_name, insn_idx); |
6245947c | 4153 | continue; |
c3c55696 AN |
4154 | } |
4155 | ||
4156 | relos = libbpf_reallocarray(prog->reloc_desc, | |
4157 | prog->nr_reloc + 1, sizeof(*relos)); | |
4158 | if (!relos) | |
4159 | return -ENOMEM; | |
4160 | prog->reloc_desc = relos; | |
4161 | ||
4162 | /* adjust insn_idx to local BPF program frame of reference */ | |
4163 | insn_idx -= prog->sec_insn_off; | |
4164 | err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc], | |
ad23b723 | 4165 | insn_idx, sym_name, sym, rel); |
1f8e2bcb AN |
4166 | if (err) |
4167 | return err; | |
c3c55696 AN |
4168 | |
4169 | prog->nr_reloc++; | |
34090915 WN |
4170 | } |
4171 | return 0; | |
4172 | } | |
4173 | ||
abd29c93 | 4174 | static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map) |
8a138aed MKL |
4175 | { |
4176 | struct bpf_map_def *def = &map->def; | |
d859900c | 4177 | __u32 key_type_id = 0, value_type_id = 0; |
96408c43 | 4178 | int ret; |
8a138aed | 4179 | |
590a0088 MKL |
4180 | /* if it's BTF-defined map, we don't need to search for type IDs. |
4181 | * For struct_ops map, it does not need btf_key_type_id and | |
4182 | * btf_value_type_id. | |
4183 | */ | |
4184 | if (map->sec_idx == obj->efile.btf_maps_shndx || | |
4185 | bpf_map__is_struct_ops(map)) | |
abd29c93 AN |
4186 | return 0; |
4187 | ||
d859900c | 4188 | if (!bpf_map__is_internal(map)) { |
93b8952d | 4189 | pr_warn("Use of BPF_ANNOTATE_KV_PAIR is deprecated, use BTF-defined maps in .maps section instead\n"); |
227a0713 AN |
4190 | #pragma GCC diagnostic push |
4191 | #pragma GCC diagnostic ignored "-Wdeprecated-declarations" | |
abd29c93 | 4192 | ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size, |
d859900c DB |
4193 | def->value_size, &key_type_id, |
4194 | &value_type_id); | |
227a0713 | 4195 | #pragma GCC diagnostic pop |
d859900c DB |
4196 | } else { |
4197 | /* | |
4198 | * LLVM annotates global data differently in BTF, that is, | |
4199 | * only as '.data', '.bss' or '.rodata'. | |
4200 | */ | |
aed65917 | 4201 | ret = btf__find_by_name(obj->btf, map->real_name); |
d859900c DB |
4202 | } |
4203 | if (ret < 0) | |
96408c43 | 4204 | return ret; |
8a138aed | 4205 | |
96408c43 | 4206 | map->btf_key_type_id = key_type_id; |
d859900c DB |
4207 | map->btf_value_type_id = bpf_map__is_internal(map) ? |
4208 | ret : value_type_id; | |
8a138aed MKL |
4209 | return 0; |
4210 | } | |
4211 | ||
97eb3138 MP |
4212 | static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info) |
4213 | { | |
4214 | char file[PATH_MAX], buff[4096]; | |
4215 | FILE *fp; | |
4216 | __u32 val; | |
4217 | int err; | |
4218 | ||
4219 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
4220 | memset(info, 0, sizeof(*info)); | |
4221 | ||
4222 | fp = fopen(file, "r"); | |
4223 | if (!fp) { | |
4224 | err = -errno; | |
4225 | pr_warn("failed to open %s: %d. No procfs support?\n", file, | |
4226 | err); | |
4227 | return err; | |
4228 | } | |
4229 | ||
4230 | while (fgets(buff, sizeof(buff), fp)) { | |
4231 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
4232 | info->type = val; | |
4233 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) | |
4234 | info->key_size = val; | |
4235 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) | |
4236 | info->value_size = val; | |
4237 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) | |
4238 | info->max_entries = val; | |
4239 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) | |
4240 | info->map_flags = val; | |
4241 | } | |
4242 | ||
4243 | fclose(fp); | |
4244 | ||
4245 | return 0; | |
4246 | } | |
4247 | ||
26736eb9 JK |
4248 | int bpf_map__reuse_fd(struct bpf_map *map, int fd) |
4249 | { | |
4250 | struct bpf_map_info info = {}; | |
4251 | __u32 len = sizeof(info); | |
4252 | int new_fd, err; | |
4253 | char *new_name; | |
4254 | ||
4255 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
97eb3138 MP |
4256 | if (err && errno == EINVAL) |
4257 | err = bpf_get_map_info_from_fdinfo(fd, &info); | |
26736eb9 | 4258 | if (err) |
e9fc3ce9 | 4259 | return libbpf_err(err); |
26736eb9 JK |
4260 | |
4261 | new_name = strdup(info.name); | |
4262 | if (!new_name) | |
e9fc3ce9 | 4263 | return libbpf_err(-errno); |
26736eb9 JK |
4264 | |
4265 | new_fd = open("/", O_RDONLY | O_CLOEXEC); | |
d1b4574a THJ |
4266 | if (new_fd < 0) { |
4267 | err = -errno; | |
26736eb9 | 4268 | goto err_free_new_name; |
d1b4574a | 4269 | } |
26736eb9 JK |
4270 | |
4271 | new_fd = dup3(fd, new_fd, O_CLOEXEC); | |
d1b4574a THJ |
4272 | if (new_fd < 0) { |
4273 | err = -errno; | |
26736eb9 | 4274 | goto err_close_new_fd; |
d1b4574a | 4275 | } |
26736eb9 JK |
4276 | |
4277 | err = zclose(map->fd); | |
d1b4574a THJ |
4278 | if (err) { |
4279 | err = -errno; | |
26736eb9 | 4280 | goto err_close_new_fd; |
d1b4574a | 4281 | } |
26736eb9 JK |
4282 | free(map->name); |
4283 | ||
4284 | map->fd = new_fd; | |
4285 | map->name = new_name; | |
4286 | map->def.type = info.type; | |
4287 | map->def.key_size = info.key_size; | |
4288 | map->def.value_size = info.value_size; | |
4289 | map->def.max_entries = info.max_entries; | |
4290 | map->def.map_flags = info.map_flags; | |
4291 | map->btf_key_type_id = info.btf_key_type_id; | |
4292 | map->btf_value_type_id = info.btf_value_type_id; | |
ec6d5f47 | 4293 | map->reused = true; |
47512102 | 4294 | map->map_extra = info.map_extra; |
26736eb9 JK |
4295 | |
4296 | return 0; | |
4297 | ||
4298 | err_close_new_fd: | |
4299 | close(new_fd); | |
4300 | err_free_new_name: | |
4301 | free(new_name); | |
e9fc3ce9 | 4302 | return libbpf_err(err); |
26736eb9 JK |
4303 | } |
4304 | ||
1bdb6c9a | 4305 | __u32 bpf_map__max_entries(const struct bpf_map *map) |
1a11a4c7 | 4306 | { |
1bdb6c9a AN |
4307 | return map->def.max_entries; |
4308 | } | |
1a11a4c7 | 4309 | |
b3278099 AN |
4310 | struct bpf_map *bpf_map__inner_map(struct bpf_map *map) |
4311 | { | |
4312 | if (!bpf_map_type__is_map_in_map(map->def.type)) | |
e9fc3ce9 | 4313 | return errno = EINVAL, NULL; |
b3278099 AN |
4314 | |
4315 | return map->inner_map; | |
4316 | } | |
4317 | ||
1bdb6c9a AN |
4318 | int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries) |
4319 | { | |
1a11a4c7 | 4320 | if (map->fd >= 0) |
e9fc3ce9 | 4321 | return libbpf_err(-EBUSY); |
1a11a4c7 | 4322 | map->def.max_entries = max_entries; |
1a11a4c7 AI |
4323 | return 0; |
4324 | } | |
4325 | ||
1bdb6c9a AN |
4326 | int bpf_map__resize(struct bpf_map *map, __u32 max_entries) |
4327 | { | |
4328 | if (!map || !max_entries) | |
e9fc3ce9 | 4329 | return libbpf_err(-EINVAL); |
1bdb6c9a AN |
4330 | |
4331 | return bpf_map__set_max_entries(map, max_entries); | |
4332 | } | |
4333 | ||
47eff617 | 4334 | static int |
fd9eef1a | 4335 | bpf_object__probe_loading(struct bpf_object *obj) |
47eff617 | 4336 | { |
47eff617 SF |
4337 | char *cp, errmsg[STRERR_BUFSIZE]; |
4338 | struct bpf_insn insns[] = { | |
4339 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4340 | BPF_EXIT_INSN(), | |
4341 | }; | |
e32660ac | 4342 | int ret, insn_cnt = ARRAY_SIZE(insns); |
47eff617 | 4343 | |
f9bceaa5 SF |
4344 | if (obj->gen_loader) |
4345 | return 0; | |
4346 | ||
e542f2c4 AN |
4347 | ret = bump_rlimit_memlock(); |
4348 | if (ret) | |
4349 | pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret); | |
4350 | ||
47eff617 | 4351 | /* make sure basic loading works */ |
e32660ac AN |
4352 | ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL); |
4353 | if (ret < 0) | |
4354 | ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL); | |
47eff617 | 4355 | if (ret < 0) { |
fd9eef1a EC |
4356 | ret = errno; |
4357 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4358 | pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF " | |
4359 | "program. Make sure your kernel supports BPF " | |
4360 | "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is " | |
4361 | "set to big enough value.\n", __func__, cp, ret); | |
4362 | return -ret; | |
47eff617 SF |
4363 | } |
4364 | close(ret); | |
4365 | ||
fd9eef1a EC |
4366 | return 0; |
4367 | } | |
4368 | ||
bb180fb2 AN |
4369 | static int probe_fd(int fd) |
4370 | { | |
4371 | if (fd >= 0) | |
4372 | close(fd); | |
4373 | return fd >= 0; | |
4374 | } | |
4375 | ||
47b6cb4d | 4376 | static int probe_kern_prog_name(void) |
fd9eef1a | 4377 | { |
fd9eef1a EC |
4378 | struct bpf_insn insns[] = { |
4379 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4380 | BPF_EXIT_INSN(), | |
4381 | }; | |
e32660ac | 4382 | int ret, insn_cnt = ARRAY_SIZE(insns); |
fd9eef1a EC |
4383 | |
4384 | /* make sure loading with name works */ | |
e32660ac | 4385 | ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, "test", "GPL", insns, insn_cnt, NULL); |
bb180fb2 | 4386 | return probe_fd(ret); |
47eff617 SF |
4387 | } |
4388 | ||
47b6cb4d | 4389 | static int probe_kern_global_data(void) |
8837fe5d | 4390 | { |
8837fe5d DB |
4391 | char *cp, errmsg[STRERR_BUFSIZE]; |
4392 | struct bpf_insn insns[] = { | |
4393 | BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16), | |
4394 | BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42), | |
4395 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4396 | BPF_EXIT_INSN(), | |
4397 | }; | |
e32660ac | 4398 | int ret, map, insn_cnt = ARRAY_SIZE(insns); |
8837fe5d | 4399 | |
a9606f40 | 4400 | map = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), 32, 1, NULL); |
8837fe5d | 4401 | if (map < 0) { |
23ab656b THJ |
4402 | ret = -errno; |
4403 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
be18010e | 4404 | pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n", |
23ab656b THJ |
4405 | __func__, cp, -ret); |
4406 | return ret; | |
8837fe5d DB |
4407 | } |
4408 | ||
4409 | insns[0].imm = map; | |
4410 | ||
e32660ac | 4411 | ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL); |
47b6cb4d | 4412 | close(map); |
bb180fb2 | 4413 | return probe_fd(ret); |
8837fe5d DB |
4414 | } |
4415 | ||
68b08647 AN |
4416 | static int probe_kern_btf(void) |
4417 | { | |
4418 | static const char strs[] = "\0int"; | |
4419 | __u32 types[] = { | |
4420 | /* int */ | |
4421 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), | |
4422 | }; | |
4423 | ||
4424 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4425 | strs, sizeof(strs))); | |
4426 | } | |
4427 | ||
47b6cb4d | 4428 | static int probe_kern_btf_func(void) |
d7c4b398 | 4429 | { |
8983b731 | 4430 | static const char strs[] = "\0int\0x\0a"; |
d7c4b398 AN |
4431 | /* void x(int a) {} */ |
4432 | __u32 types[] = { | |
4433 | /* int */ | |
4434 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4435 | /* FUNC_PROTO */ /* [2] */ | |
4436 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0), | |
4437 | BTF_PARAM_ENC(7, 1), | |
4438 | /* FUNC x */ /* [3] */ | |
4439 | BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2), | |
4440 | }; | |
d7c4b398 | 4441 | |
bb180fb2 AN |
4442 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4443 | strs, sizeof(strs))); | |
d7c4b398 AN |
4444 | } |
4445 | ||
47b6cb4d | 4446 | static int probe_kern_btf_func_global(void) |
2d3eb67f AS |
4447 | { |
4448 | static const char strs[] = "\0int\0x\0a"; | |
4449 | /* static void x(int a) {} */ | |
4450 | __u32 types[] = { | |
4451 | /* int */ | |
4452 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4453 | /* FUNC_PROTO */ /* [2] */ | |
4454 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0), | |
4455 | BTF_PARAM_ENC(7, 1), | |
4456 | /* FUNC x BTF_FUNC_GLOBAL */ /* [3] */ | |
4457 | BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2), | |
4458 | }; | |
2d3eb67f | 4459 | |
bb180fb2 AN |
4460 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4461 | strs, sizeof(strs))); | |
2d3eb67f AS |
4462 | } |
4463 | ||
47b6cb4d | 4464 | static int probe_kern_btf_datasec(void) |
d7c4b398 | 4465 | { |
8983b731 | 4466 | static const char strs[] = "\0x\0.data"; |
d7c4b398 AN |
4467 | /* static int a; */ |
4468 | __u32 types[] = { | |
4469 | /* int */ | |
4470 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4471 | /* VAR x */ /* [2] */ | |
4472 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1), | |
4473 | BTF_VAR_STATIC, | |
4474 | /* DATASEC val */ /* [3] */ | |
4475 | BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4), | |
4476 | BTF_VAR_SECINFO_ENC(2, 0, 4), | |
4477 | }; | |
cfd49210 | 4478 | |
bb180fb2 AN |
4479 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4480 | strs, sizeof(strs))); | |
d7c4b398 AN |
4481 | } |
4482 | ||
22541a9e IL |
4483 | static int probe_kern_btf_float(void) |
4484 | { | |
4485 | static const char strs[] = "\0float"; | |
4486 | __u32 types[] = { | |
4487 | /* float */ | |
4488 | BTF_TYPE_FLOAT_ENC(1, 4), | |
4489 | }; | |
4490 | ||
4491 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4492 | strs, sizeof(strs))); | |
4493 | } | |
4494 | ||
223f903e | 4495 | static int probe_kern_btf_decl_tag(void) |
5b84bd10 YS |
4496 | { |
4497 | static const char strs[] = "\0tag"; | |
4498 | __u32 types[] = { | |
4499 | /* int */ | |
4500 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4501 | /* VAR x */ /* [2] */ | |
4502 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1), | |
4503 | BTF_VAR_STATIC, | |
4504 | /* attr */ | |
223f903e | 4505 | BTF_TYPE_DECL_TAG_ENC(1, 2, -1), |
5b84bd10 YS |
4506 | }; |
4507 | ||
4508 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4509 | strs, sizeof(strs))); | |
4510 | } | |
4511 | ||
2dc1e488 YS |
4512 | static int probe_kern_btf_type_tag(void) |
4513 | { | |
4514 | static const char strs[] = "\0tag"; | |
4515 | __u32 types[] = { | |
4516 | /* int */ | |
4517 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4518 | /* attr */ | |
4519 | BTF_TYPE_TYPE_TAG_ENC(1, 1), /* [2] */ | |
4520 | /* ptr */ | |
4521 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_PTR, 0, 0), 2), /* [3] */ | |
4522 | }; | |
4523 | ||
4524 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4525 | strs, sizeof(strs))); | |
4526 | } | |
4527 | ||
47b6cb4d | 4528 | static int probe_kern_array_mmap(void) |
7fe74b43 | 4529 | { |
a9606f40 AN |
4530 | LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); |
4531 | int fd; | |
7fe74b43 | 4532 | |
a9606f40 AN |
4533 | fd = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), sizeof(int), 1, &opts); |
4534 | return probe_fd(fd); | |
7fe74b43 AN |
4535 | } |
4536 | ||
47b6cb4d | 4537 | static int probe_kern_exp_attach_type(void) |
25498a19 | 4538 | { |
e32660ac | 4539 | LIBBPF_OPTS(bpf_prog_load_opts, opts, .expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE); |
25498a19 AN |
4540 | struct bpf_insn insns[] = { |
4541 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4542 | BPF_EXIT_INSN(), | |
4543 | }; | |
e32660ac | 4544 | int fd, insn_cnt = ARRAY_SIZE(insns); |
25498a19 | 4545 | |
25498a19 AN |
4546 | /* use any valid combination of program type and (optional) |
4547 | * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS) | |
4548 | * to see if kernel supports expected_attach_type field for | |
4549 | * BPF_PROG_LOAD command | |
4550 | */ | |
e32660ac AN |
4551 | fd = bpf_prog_load(BPF_PROG_TYPE_CGROUP_SOCK, NULL, "GPL", insns, insn_cnt, &opts); |
4552 | return probe_fd(fd); | |
25498a19 AN |
4553 | } |
4554 | ||
109cea5a AN |
4555 | static int probe_kern_probe_read_kernel(void) |
4556 | { | |
109cea5a AN |
4557 | struct bpf_insn insns[] = { |
4558 | BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), /* r1 = r10 (fp) */ | |
4559 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8), /* r1 += -8 */ | |
4560 | BPF_MOV64_IMM(BPF_REG_2, 8), /* r2 = 8 */ | |
4561 | BPF_MOV64_IMM(BPF_REG_3, 0), /* r3 = 0 */ | |
4562 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel), | |
4563 | BPF_EXIT_INSN(), | |
4564 | }; | |
e32660ac | 4565 | int fd, insn_cnt = ARRAY_SIZE(insns); |
109cea5a | 4566 | |
e32660ac AN |
4567 | fd = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL, "GPL", insns, insn_cnt, NULL); |
4568 | return probe_fd(fd); | |
109cea5a AN |
4569 | } |
4570 | ||
5d23328d YZ |
4571 | static int probe_prog_bind_map(void) |
4572 | { | |
5d23328d YZ |
4573 | char *cp, errmsg[STRERR_BUFSIZE]; |
4574 | struct bpf_insn insns[] = { | |
4575 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4576 | BPF_EXIT_INSN(), | |
4577 | }; | |
e32660ac | 4578 | int ret, map, prog, insn_cnt = ARRAY_SIZE(insns); |
5d23328d | 4579 | |
a9606f40 | 4580 | map = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), 32, 1, NULL); |
5d23328d YZ |
4581 | if (map < 0) { |
4582 | ret = -errno; | |
4583 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4584 | pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n", | |
4585 | __func__, cp, -ret); | |
4586 | return ret; | |
4587 | } | |
4588 | ||
e32660ac | 4589 | prog = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL); |
5d23328d YZ |
4590 | if (prog < 0) { |
4591 | close(map); | |
4592 | return 0; | |
4593 | } | |
4594 | ||
4595 | ret = bpf_prog_bind_map(prog, map, NULL); | |
4596 | ||
4597 | close(map); | |
4598 | close(prog); | |
4599 | ||
4600 | return ret >= 0; | |
4601 | } | |
4602 | ||
4f33a53d AN |
4603 | static int probe_module_btf(void) |
4604 | { | |
4605 | static const char strs[] = "\0int"; | |
4606 | __u32 types[] = { | |
4607 | /* int */ | |
4608 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), | |
4609 | }; | |
4610 | struct bpf_btf_info info; | |
4611 | __u32 len = sizeof(info); | |
4612 | char name[16]; | |
4613 | int fd, err; | |
4614 | ||
4615 | fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs)); | |
4616 | if (fd < 0) | |
4617 | return 0; /* BTF not supported at all */ | |
4618 | ||
4619 | memset(&info, 0, sizeof(info)); | |
4620 | info.name = ptr_to_u64(name); | |
4621 | info.name_len = sizeof(name); | |
4622 | ||
4623 | /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer; | |
4624 | * kernel's module BTF support coincides with support for | |
4625 | * name/name_len fields in struct bpf_btf_info. | |
4626 | */ | |
4627 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
4628 | close(fd); | |
4629 | return !err; | |
4630 | } | |
4631 | ||
668ace0e AN |
4632 | static int probe_perf_link(void) |
4633 | { | |
668ace0e AN |
4634 | struct bpf_insn insns[] = { |
4635 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4636 | BPF_EXIT_INSN(), | |
4637 | }; | |
4638 | int prog_fd, link_fd, err; | |
4639 | ||
e32660ac AN |
4640 | prog_fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", |
4641 | insns, ARRAY_SIZE(insns), NULL); | |
668ace0e AN |
4642 | if (prog_fd < 0) |
4643 | return -errno; | |
4644 | ||
4645 | /* use invalid perf_event FD to get EBADF, if link is supported; | |
4646 | * otherwise EINVAL should be returned | |
4647 | */ | |
4648 | link_fd = bpf_link_create(prog_fd, -1, BPF_PERF_EVENT, NULL); | |
4649 | err = -errno; /* close() can clobber errno */ | |
4650 | ||
4651 | if (link_fd >= 0) | |
4652 | close(link_fd); | |
4653 | close(prog_fd); | |
4654 | ||
4655 | return link_fd < 0 && err == -EBADF; | |
4656 | } | |
4657 | ||
47b6cb4d AN |
4658 | enum kern_feature_result { |
4659 | FEAT_UNKNOWN = 0, | |
4660 | FEAT_SUPPORTED = 1, | |
4661 | FEAT_MISSING = 2, | |
4662 | }; | |
4663 | ||
4664 | typedef int (*feature_probe_fn)(void); | |
4665 | ||
4666 | static struct kern_feature_desc { | |
4667 | const char *desc; | |
4668 | feature_probe_fn probe; | |
4669 | enum kern_feature_result res; | |
4670 | } feature_probes[__FEAT_CNT] = { | |
4671 | [FEAT_PROG_NAME] = { | |
4672 | "BPF program name", probe_kern_prog_name, | |
4673 | }, | |
4674 | [FEAT_GLOBAL_DATA] = { | |
4675 | "global variables", probe_kern_global_data, | |
4676 | }, | |
68b08647 AN |
4677 | [FEAT_BTF] = { |
4678 | "minimal BTF", probe_kern_btf, | |
4679 | }, | |
47b6cb4d AN |
4680 | [FEAT_BTF_FUNC] = { |
4681 | "BTF functions", probe_kern_btf_func, | |
4682 | }, | |
4683 | [FEAT_BTF_GLOBAL_FUNC] = { | |
4684 | "BTF global function", probe_kern_btf_func_global, | |
4685 | }, | |
4686 | [FEAT_BTF_DATASEC] = { | |
4687 | "BTF data section and variable", probe_kern_btf_datasec, | |
4688 | }, | |
4689 | [FEAT_ARRAY_MMAP] = { | |
4690 | "ARRAY map mmap()", probe_kern_array_mmap, | |
4691 | }, | |
4692 | [FEAT_EXP_ATTACH_TYPE] = { | |
4693 | "BPF_PROG_LOAD expected_attach_type attribute", | |
4694 | probe_kern_exp_attach_type, | |
4695 | }, | |
109cea5a AN |
4696 | [FEAT_PROBE_READ_KERN] = { |
4697 | "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel, | |
5d23328d YZ |
4698 | }, |
4699 | [FEAT_PROG_BIND_MAP] = { | |
4700 | "BPF_PROG_BIND_MAP support", probe_prog_bind_map, | |
4f33a53d AN |
4701 | }, |
4702 | [FEAT_MODULE_BTF] = { | |
4703 | "module BTF support", probe_module_btf, | |
4704 | }, | |
22541a9e IL |
4705 | [FEAT_BTF_FLOAT] = { |
4706 | "BTF_KIND_FLOAT support", probe_kern_btf_float, | |
4707 | }, | |
668ace0e AN |
4708 | [FEAT_PERF_LINK] = { |
4709 | "BPF perf link support", probe_perf_link, | |
4710 | }, | |
223f903e YS |
4711 | [FEAT_BTF_DECL_TAG] = { |
4712 | "BTF_KIND_DECL_TAG support", probe_kern_btf_decl_tag, | |
5b84bd10 | 4713 | }, |
2dc1e488 YS |
4714 | [FEAT_BTF_TYPE_TAG] = { |
4715 | "BTF_KIND_TYPE_TAG support", probe_kern_btf_type_tag, | |
4716 | }, | |
e542f2c4 AN |
4717 | [FEAT_MEMCG_ACCOUNT] = { |
4718 | "memcg-based memory accounting", probe_memcg_account, | |
4719 | }, | |
47b6cb4d | 4720 | }; |
8837fe5d | 4721 | |
e542f2c4 | 4722 | bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id) |
47b6cb4d AN |
4723 | { |
4724 | struct kern_feature_desc *feat = &feature_probes[feat_id]; | |
4725 | int ret; | |
4726 | ||
e542f2c4 | 4727 | if (obj && obj->gen_loader) |
67234743 AS |
4728 | /* To generate loader program assume the latest kernel |
4729 | * to avoid doing extra prog_load, map_create syscalls. | |
4730 | */ | |
4731 | return true; | |
4732 | ||
47b6cb4d AN |
4733 | if (READ_ONCE(feat->res) == FEAT_UNKNOWN) { |
4734 | ret = feat->probe(); | |
4735 | if (ret > 0) { | |
4736 | WRITE_ONCE(feat->res, FEAT_SUPPORTED); | |
4737 | } else if (ret == 0) { | |
4738 | WRITE_ONCE(feat->res, FEAT_MISSING); | |
4739 | } else { | |
4740 | pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret); | |
4741 | WRITE_ONCE(feat->res, FEAT_MISSING); | |
4742 | } | |
8837fe5d DB |
4743 | } |
4744 | ||
47b6cb4d | 4745 | return READ_ONCE(feat->res) == FEAT_SUPPORTED; |
47eff617 SF |
4746 | } |
4747 | ||
57a00f41 THJ |
4748 | static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd) |
4749 | { | |
4750 | struct bpf_map_info map_info = {}; | |
4751 | char msg[STRERR_BUFSIZE]; | |
4752 | __u32 map_info_len; | |
97eb3138 | 4753 | int err; |
57a00f41 THJ |
4754 | |
4755 | map_info_len = sizeof(map_info); | |
4756 | ||
97eb3138 MP |
4757 | err = bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len); |
4758 | if (err && errno == EINVAL) | |
4759 | err = bpf_get_map_info_from_fdinfo(map_fd, &map_info); | |
4760 | if (err) { | |
4761 | pr_warn("failed to get map info for map FD %d: %s\n", map_fd, | |
4762 | libbpf_strerror_r(errno, msg, sizeof(msg))); | |
57a00f41 THJ |
4763 | return false; |
4764 | } | |
4765 | ||
4766 | return (map_info.type == map->def.type && | |
4767 | map_info.key_size == map->def.key_size && | |
4768 | map_info.value_size == map->def.value_size && | |
4769 | map_info.max_entries == map->def.max_entries && | |
47512102 JK |
4770 | map_info.map_flags == map->def.map_flags && |
4771 | map_info.map_extra == map->map_extra); | |
57a00f41 THJ |
4772 | } |
4773 | ||
4774 | static int | |
4775 | bpf_object__reuse_map(struct bpf_map *map) | |
4776 | { | |
4777 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4778 | int err, pin_fd; | |
4779 | ||
4780 | pin_fd = bpf_obj_get(map->pin_path); | |
4781 | if (pin_fd < 0) { | |
4782 | err = -errno; | |
4783 | if (err == -ENOENT) { | |
4784 | pr_debug("found no pinned map to reuse at '%s'\n", | |
4785 | map->pin_path); | |
4786 | return 0; | |
4787 | } | |
4788 | ||
4789 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
4790 | pr_warn("couldn't retrieve pinned map '%s': %s\n", | |
4791 | map->pin_path, cp); | |
4792 | return err; | |
4793 | } | |
4794 | ||
4795 | if (!map_is_reuse_compat(map, pin_fd)) { | |
4796 | pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n", | |
4797 | map->pin_path); | |
4798 | close(pin_fd); | |
4799 | return -EINVAL; | |
4800 | } | |
4801 | ||
4802 | err = bpf_map__reuse_fd(map, pin_fd); | |
4803 | if (err) { | |
4804 | close(pin_fd); | |
4805 | return err; | |
4806 | } | |
4807 | map->pinned = true; | |
4808 | pr_debug("reused pinned map at '%s'\n", map->pin_path); | |
4809 | ||
4810 | return 0; | |
4811 | } | |
4812 | ||
d859900c DB |
4813 | static int |
4814 | bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map) | |
4815 | { | |
166750bc | 4816 | enum libbpf_map_type map_type = map->libbpf_type; |
d859900c DB |
4817 | char *cp, errmsg[STRERR_BUFSIZE]; |
4818 | int err, zero = 0; | |
d859900c | 4819 | |
67234743 AS |
4820 | if (obj->gen_loader) { |
4821 | bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps, | |
4822 | map->mmaped, map->def.value_size); | |
4823 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) | |
4824 | bpf_gen__map_freeze(obj->gen_loader, map - obj->maps); | |
4825 | return 0; | |
4826 | } | |
eba9c5f4 AN |
4827 | err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0); |
4828 | if (err) { | |
4829 | err = -errno; | |
4830 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
4831 | pr_warn("Error setting initial map(%s) contents: %s\n", | |
4832 | map->name, cp); | |
4833 | return err; | |
4834 | } | |
d859900c | 4835 | |
81bfdd08 AN |
4836 | /* Freeze .rodata and .kconfig map as read-only from syscall side. */ |
4837 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) { | |
d859900c DB |
4838 | err = bpf_map_freeze(map->fd); |
4839 | if (err) { | |
eba9c5f4 AN |
4840 | err = -errno; |
4841 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e KW |
4842 | pr_warn("Error freezing map(%s) as read-only: %s\n", |
4843 | map->name, cp); | |
eba9c5f4 | 4844 | return err; |
d859900c DB |
4845 | } |
4846 | } | |
eba9c5f4 | 4847 | return 0; |
d859900c DB |
4848 | } |
4849 | ||
2d39d7c5 AN |
4850 | static void bpf_map__destroy(struct bpf_map *map); |
4851 | ||
67234743 | 4852 | static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner) |
2d39d7c5 | 4853 | { |
992c4225 | 4854 | LIBBPF_OPTS(bpf_map_create_opts, create_attr); |
2d39d7c5 | 4855 | struct bpf_map_def *def = &map->def; |
992c4225 | 4856 | const char *map_name = NULL; |
a21ab4c5 | 4857 | int err = 0; |
2d39d7c5 | 4858 | |
9ca1f56a | 4859 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
992c4225 | 4860 | map_name = map->name; |
2d39d7c5 | 4861 | create_attr.map_ifindex = map->map_ifindex; |
2d39d7c5 | 4862 | create_attr.map_flags = def->map_flags; |
1bdb6c9a | 4863 | create_attr.numa_node = map->numa_node; |
47512102 | 4864 | create_attr.map_extra = map->map_extra; |
2d39d7c5 | 4865 | |
2d39d7c5 | 4866 | if (bpf_map__is_struct_ops(map)) |
992c4225 | 4867 | create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id; |
2d39d7c5 | 4868 | |
0f0e55d8 | 4869 | if (obj->btf && btf__fd(obj->btf) >= 0 && !bpf_map_find_btf_info(obj, map)) { |
2d39d7c5 AN |
4870 | create_attr.btf_fd = btf__fd(obj->btf); |
4871 | create_attr.btf_key_type_id = map->btf_key_type_id; | |
4872 | create_attr.btf_value_type_id = map->btf_value_type_id; | |
4873 | } | |
4874 | ||
646f02ff AN |
4875 | if (bpf_map_type__is_map_in_map(def->type)) { |
4876 | if (map->inner_map) { | |
67234743 | 4877 | err = bpf_object__create_map(obj, map->inner_map, true); |
646f02ff AN |
4878 | if (err) { |
4879 | pr_warn("map '%s': failed to create inner map: %d\n", | |
4880 | map->name, err); | |
4881 | return err; | |
4882 | } | |
4883 | map->inner_map_fd = bpf_map__fd(map->inner_map); | |
4884 | } | |
4885 | if (map->inner_map_fd >= 0) | |
4886 | create_attr.inner_map_fd = map->inner_map_fd; | |
4887 | } | |
4888 | ||
f7310523 HC |
4889 | switch (def->type) { |
4890 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
4891 | case BPF_MAP_TYPE_CGROUP_ARRAY: | |
4892 | case BPF_MAP_TYPE_STACK_TRACE: | |
4893 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
4894 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
4895 | case BPF_MAP_TYPE_DEVMAP: | |
4896 | case BPF_MAP_TYPE_DEVMAP_HASH: | |
4897 | case BPF_MAP_TYPE_CPUMAP: | |
4898 | case BPF_MAP_TYPE_XSKMAP: | |
4899 | case BPF_MAP_TYPE_SOCKMAP: | |
4900 | case BPF_MAP_TYPE_SOCKHASH: | |
4901 | case BPF_MAP_TYPE_QUEUE: | |
4902 | case BPF_MAP_TYPE_STACK: | |
4903 | case BPF_MAP_TYPE_RINGBUF: | |
4904 | create_attr.btf_fd = 0; | |
4905 | create_attr.btf_key_type_id = 0; | |
4906 | create_attr.btf_value_type_id = 0; | |
4907 | map->btf_key_type_id = 0; | |
4908 | map->btf_value_type_id = 0; | |
4909 | default: | |
4910 | break; | |
4911 | } | |
4912 | ||
67234743 | 4913 | if (obj->gen_loader) { |
992c4225 | 4914 | bpf_gen__map_create(obj->gen_loader, def->type, map_name, |
a4fbfdd7 | 4915 | def->key_size, def->value_size, def->max_entries, |
992c4225 | 4916 | &create_attr, is_inner ? -1 : map - obj->maps); |
67234743 AS |
4917 | /* Pretend to have valid FD to pass various fd >= 0 checks. |
4918 | * This fd == 0 will not be used with any syscall and will be reset to -1 eventually. | |
4919 | */ | |
4920 | map->fd = 0; | |
4921 | } else { | |
992c4225 AN |
4922 | map->fd = bpf_map_create(def->type, map_name, |
4923 | def->key_size, def->value_size, | |
a4fbfdd7 | 4924 | def->max_entries, &create_attr); |
67234743 | 4925 | } |
2d39d7c5 AN |
4926 | if (map->fd < 0 && (create_attr.btf_key_type_id || |
4927 | create_attr.btf_value_type_id)) { | |
4928 | char *cp, errmsg[STRERR_BUFSIZE]; | |
2d39d7c5 | 4929 | |
a21ab4c5 | 4930 | err = -errno; |
2d39d7c5 AN |
4931 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); |
4932 | pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n", | |
4933 | map->name, cp, err); | |
4934 | create_attr.btf_fd = 0; | |
4935 | create_attr.btf_key_type_id = 0; | |
4936 | create_attr.btf_value_type_id = 0; | |
4937 | map->btf_key_type_id = 0; | |
4938 | map->btf_value_type_id = 0; | |
992c4225 AN |
4939 | map->fd = bpf_map_create(def->type, map_name, |
4940 | def->key_size, def->value_size, | |
a4fbfdd7 | 4941 | def->max_entries, &create_attr); |
2d39d7c5 AN |
4942 | } |
4943 | ||
a21ab4c5 | 4944 | err = map->fd < 0 ? -errno : 0; |
2d39d7c5 | 4945 | |
646f02ff | 4946 | if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) { |
67234743 AS |
4947 | if (obj->gen_loader) |
4948 | map->inner_map->fd = -1; | |
646f02ff AN |
4949 | bpf_map__destroy(map->inner_map); |
4950 | zfree(&map->inner_map); | |
4951 | } | |
4952 | ||
a21ab4c5 | 4953 | return err; |
2d39d7c5 AN |
4954 | } |
4955 | ||
341ac5ff | 4956 | static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map) |
a0f2b7ac HL |
4957 | { |
4958 | const struct bpf_map *targ_map; | |
4959 | unsigned int i; | |
67234743 | 4960 | int fd, err = 0; |
a0f2b7ac HL |
4961 | |
4962 | for (i = 0; i < map->init_slots_sz; i++) { | |
4963 | if (!map->init_slots[i]) | |
4964 | continue; | |
4965 | ||
4966 | targ_map = map->init_slots[i]; | |
4967 | fd = bpf_map__fd(targ_map); | |
341ac5ff | 4968 | |
67234743 | 4969 | if (obj->gen_loader) { |
be05c944 AS |
4970 | bpf_gen__populate_outer_map(obj->gen_loader, |
4971 | map - obj->maps, i, | |
4972 | targ_map - obj->maps); | |
67234743 AS |
4973 | } else { |
4974 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
4975 | } | |
a0f2b7ac HL |
4976 | if (err) { |
4977 | err = -errno; | |
4978 | pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n", | |
341ac5ff | 4979 | map->name, i, targ_map->name, fd, err); |
a0f2b7ac HL |
4980 | return err; |
4981 | } | |
4982 | pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n", | |
4983 | map->name, i, targ_map->name, fd); | |
4984 | } | |
4985 | ||
4986 | zfree(&map->init_slots); | |
4987 | map->init_slots_sz = 0; | |
4988 | ||
4989 | return 0; | |
4990 | } | |
4991 | ||
341ac5ff HC |
4992 | static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map) |
4993 | { | |
4994 | const struct bpf_program *targ_prog; | |
4995 | unsigned int i; | |
4996 | int fd, err; | |
4997 | ||
4998 | if (obj->gen_loader) | |
4999 | return -ENOTSUP; | |
5000 | ||
5001 | for (i = 0; i < map->init_slots_sz; i++) { | |
5002 | if (!map->init_slots[i]) | |
5003 | continue; | |
5004 | ||
5005 | targ_prog = map->init_slots[i]; | |
5006 | fd = bpf_program__fd(targ_prog); | |
5007 | ||
5008 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
5009 | if (err) { | |
5010 | err = -errno; | |
5011 | pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %d\n", | |
5012 | map->name, i, targ_prog->name, fd, err); | |
5013 | return err; | |
5014 | } | |
5015 | pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n", | |
5016 | map->name, i, targ_prog->name, fd); | |
5017 | } | |
5018 | ||
5019 | zfree(&map->init_slots); | |
5020 | map->init_slots_sz = 0; | |
5021 | ||
5022 | return 0; | |
5023 | } | |
5024 | ||
5025 | static int bpf_object_init_prog_arrays(struct bpf_object *obj) | |
5026 | { | |
5027 | struct bpf_map *map; | |
5028 | int i, err; | |
5029 | ||
5030 | for (i = 0; i < obj->nr_maps; i++) { | |
5031 | map = &obj->maps[i]; | |
5032 | ||
5033 | if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY) | |
5034 | continue; | |
5035 | ||
5036 | err = init_prog_array_slots(obj, map); | |
5037 | if (err < 0) { | |
5038 | zclose(map->fd); | |
5039 | return err; | |
5040 | } | |
5041 | } | |
5042 | return 0; | |
5043 | } | |
5044 | ||
a4fbfdd7 ST |
5045 | static int map_set_def_max_entries(struct bpf_map *map) |
5046 | { | |
5047 | if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) { | |
5048 | int nr_cpus; | |
5049 | ||
5050 | nr_cpus = libbpf_num_possible_cpus(); | |
5051 | if (nr_cpus < 0) { | |
5052 | pr_warn("map '%s': failed to determine number of system CPUs: %d\n", | |
5053 | map->name, nr_cpus); | |
5054 | return nr_cpus; | |
5055 | } | |
5056 | pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus); | |
5057 | map->def.max_entries = nr_cpus; | |
5058 | } | |
5059 | ||
5060 | return 0; | |
5061 | } | |
5062 | ||
52d3352e WN |
5063 | static int |
5064 | bpf_object__create_maps(struct bpf_object *obj) | |
5065 | { | |
2d39d7c5 AN |
5066 | struct bpf_map *map; |
5067 | char *cp, errmsg[STRERR_BUFSIZE]; | |
5068 | unsigned int i, j; | |
8a138aed | 5069 | int err; |
043c5bb3 | 5070 | bool retried; |
52d3352e | 5071 | |
9d759a9b | 5072 | for (i = 0; i < obj->nr_maps; i++) { |
2d39d7c5 | 5073 | map = &obj->maps[i]; |
8a138aed | 5074 | |
16e0c35c AN |
5075 | /* To support old kernels, we skip creating global data maps |
5076 | * (.rodata, .data, .kconfig, etc); later on, during program | |
5077 | * loading, if we detect that at least one of the to-be-loaded | |
5078 | * programs is referencing any global data map, we'll error | |
5079 | * out with program name and relocation index logged. | |
5080 | * This approach allows to accommodate Clang emitting | |
5081 | * unnecessary .rodata.str1.1 sections for string literals, | |
5082 | * but also it allows to have CO-RE applications that use | |
5083 | * global variables in some of BPF programs, but not others. | |
5084 | * If those global variable-using programs are not loaded at | |
5085 | * runtime due to bpf_program__set_autoload(prog, false), | |
5086 | * bpf_object loading will succeed just fine even on old | |
5087 | * kernels. | |
5088 | */ | |
5089 | if (bpf_map__is_internal(map) && | |
229fae38 SC |
5090 | !kernel_supports(obj, FEAT_GLOBAL_DATA)) { |
5091 | map->skipped = true; | |
16e0c35c | 5092 | continue; |
229fae38 | 5093 | } |
16e0c35c | 5094 | |
a4fbfdd7 ST |
5095 | err = map_set_def_max_entries(map); |
5096 | if (err) | |
5097 | goto err_out; | |
5098 | ||
043c5bb3 MP |
5099 | retried = false; |
5100 | retry: | |
57a00f41 THJ |
5101 | if (map->pin_path) { |
5102 | err = bpf_object__reuse_map(map); | |
5103 | if (err) { | |
2d39d7c5 | 5104 | pr_warn("map '%s': error reusing pinned map\n", |
57a00f41 | 5105 | map->name); |
2d39d7c5 | 5106 | goto err_out; |
57a00f41 | 5107 | } |
043c5bb3 MP |
5108 | if (retried && map->fd < 0) { |
5109 | pr_warn("map '%s': cannot find pinned map\n", | |
5110 | map->name); | |
5111 | err = -ENOENT; | |
5112 | goto err_out; | |
5113 | } | |
57a00f41 THJ |
5114 | } |
5115 | ||
26736eb9 | 5116 | if (map->fd >= 0) { |
2d39d7c5 | 5117 | pr_debug("map '%s': skipping creation (preset fd=%d)\n", |
26736eb9 | 5118 | map->name, map->fd); |
2c193d32 | 5119 | } else { |
67234743 | 5120 | err = bpf_object__create_map(obj, map, false); |
2c193d32 | 5121 | if (err) |
d859900c | 5122 | goto err_out; |
d859900c | 5123 | |
2c193d32 HL |
5124 | pr_debug("map '%s': created successfully, fd=%d\n", |
5125 | map->name, map->fd); | |
646f02ff | 5126 | |
2c193d32 HL |
5127 | if (bpf_map__is_internal(map)) { |
5128 | err = bpf_object__populate_internal_map(obj, map); | |
5129 | if (err < 0) { | |
5130 | zclose(map->fd); | |
5131 | goto err_out; | |
5132 | } | |
d859900c | 5133 | } |
646f02ff | 5134 | |
341ac5ff HC |
5135 | if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) { |
5136 | err = init_map_in_map_slots(obj, map); | |
2c193d32 HL |
5137 | if (err < 0) { |
5138 | zclose(map->fd); | |
646f02ff AN |
5139 | goto err_out; |
5140 | } | |
646f02ff | 5141 | } |
646f02ff AN |
5142 | } |
5143 | ||
57a00f41 THJ |
5144 | if (map->pin_path && !map->pinned) { |
5145 | err = bpf_map__pin(map, NULL); | |
5146 | if (err) { | |
043c5bb3 MP |
5147 | zclose(map->fd); |
5148 | if (!retried && err == -EEXIST) { | |
5149 | retried = true; | |
5150 | goto retry; | |
5151 | } | |
2d39d7c5 AN |
5152 | pr_warn("map '%s': failed to auto-pin at '%s': %d\n", |
5153 | map->name, map->pin_path, err); | |
2d39d7c5 | 5154 | goto err_out; |
57a00f41 THJ |
5155 | } |
5156 | } | |
52d3352e WN |
5157 | } |
5158 | ||
52d3352e | 5159 | return 0; |
2d39d7c5 AN |
5160 | |
5161 | err_out: | |
5162 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
5163 | pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err); | |
5164 | pr_perm_msg(err); | |
5165 | for (j = 0; j < i; j++) | |
5166 | zclose(obj->maps[j].fd); | |
5167 | return err; | |
52d3352e WN |
5168 | } |
5169 | ||
ddc7c304 AN |
5170 | static bool bpf_core_is_flavor_sep(const char *s) |
5171 | { | |
5172 | /* check X___Y name pattern, where X and Y are not underscores */ | |
5173 | return s[0] != '_' && /* X */ | |
5174 | s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */ | |
5175 | s[4] != '_'; /* Y */ | |
5176 | } | |
5177 | ||
5178 | /* Given 'some_struct_name___with_flavor' return the length of a name prefix | |
5179 | * before last triple underscore. Struct name part after last triple | |
5180 | * underscore is ignored by BPF CO-RE relocation during relocation matching. | |
5181 | */ | |
b0588390 | 5182 | size_t bpf_core_essential_name_len(const char *name) |
ddc7c304 AN |
5183 | { |
5184 | size_t n = strlen(name); | |
5185 | int i; | |
5186 | ||
5187 | for (i = n - 5; i >= 0; i--) { | |
5188 | if (bpf_core_is_flavor_sep(name + i)) | |
5189 | return i + 1; | |
5190 | } | |
5191 | return n; | |
5192 | } | |
5193 | ||
8de6cae4 | 5194 | void bpf_core_free_cands(struct bpf_core_cand_list *cands) |
ddc7c304 | 5195 | { |
8de6cae4 MV |
5196 | if (!cands) |
5197 | return; | |
5198 | ||
0f7515ca AN |
5199 | free(cands->cands); |
5200 | free(cands); | |
ddc7c304 AN |
5201 | } |
5202 | ||
8de6cae4 MV |
5203 | int bpf_core_add_cands(struct bpf_core_cand *local_cand, |
5204 | size_t local_essent_len, | |
5205 | const struct btf *targ_btf, | |
5206 | const char *targ_btf_name, | |
5207 | int targ_start_id, | |
5208 | struct bpf_core_cand_list *cands) | |
ddc7c304 | 5209 | { |
301ba4d7 | 5210 | struct bpf_core_cand *new_cands, *cand; |
03d5b991 AN |
5211 | const struct btf_type *t, *local_t; |
5212 | const char *targ_name, *local_name; | |
0f7515ca AN |
5213 | size_t targ_essent_len; |
5214 | int n, i; | |
ddc7c304 | 5215 | |
03d5b991 AN |
5216 | local_t = btf__type_by_id(local_cand->btf, local_cand->id); |
5217 | local_name = btf__str_by_offset(local_cand->btf, local_t->name_off); | |
5218 | ||
6a886de0 HC |
5219 | n = btf__type_cnt(targ_btf); |
5220 | for (i = targ_start_id; i < n; i++) { | |
ddc7c304 | 5221 | t = btf__type_by_id(targ_btf, i); |
03d5b991 | 5222 | if (btf_kind(t) != btf_kind(local_t)) |
ddc7c304 AN |
5223 | continue; |
5224 | ||
3fc32f40 AN |
5225 | targ_name = btf__name_by_offset(targ_btf, t->name_off); |
5226 | if (str_is_empty(targ_name)) | |
d121e1d3 AN |
5227 | continue; |
5228 | ||
ddc7c304 AN |
5229 | targ_essent_len = bpf_core_essential_name_len(targ_name); |
5230 | if (targ_essent_len != local_essent_len) | |
5231 | continue; | |
5232 | ||
03d5b991 | 5233 | if (strncmp(local_name, targ_name, local_essent_len) != 0) |
0f7515ca AN |
5234 | continue; |
5235 | ||
5236 | pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n", | |
03d5b991 AN |
5237 | local_cand->id, btf_kind_str(local_t), |
5238 | local_name, i, btf_kind_str(t), targ_name, | |
0f7515ca AN |
5239 | targ_btf_name); |
5240 | new_cands = libbpf_reallocarray(cands->cands, cands->len + 1, | |
5241 | sizeof(*cands->cands)); | |
5242 | if (!new_cands) | |
5243 | return -ENOMEM; | |
5244 | ||
5245 | cand = &new_cands[cands->len]; | |
5246 | cand->btf = targ_btf; | |
0f7515ca AN |
5247 | cand->id = i; |
5248 | ||
5249 | cands->cands = new_cands; | |
5250 | cands->len++; | |
ddc7c304 | 5251 | } |
0f7515ca AN |
5252 | return 0; |
5253 | } | |
5254 | ||
4f33a53d AN |
5255 | static int load_module_btfs(struct bpf_object *obj) |
5256 | { | |
5257 | struct bpf_btf_info info; | |
5258 | struct module_btf *mod_btf; | |
5259 | struct btf *btf; | |
5260 | char name[64]; | |
5261 | __u32 id = 0, len; | |
5262 | int err, fd; | |
5263 | ||
5264 | if (obj->btf_modules_loaded) | |
5265 | return 0; | |
5266 | ||
67234743 AS |
5267 | if (obj->gen_loader) |
5268 | return 0; | |
5269 | ||
4f33a53d AN |
5270 | /* don't do this again, even if we find no module BTFs */ |
5271 | obj->btf_modules_loaded = true; | |
5272 | ||
5273 | /* kernel too old to support module BTFs */ | |
9ca1f56a | 5274 | if (!kernel_supports(obj, FEAT_MODULE_BTF)) |
4f33a53d AN |
5275 | return 0; |
5276 | ||
5277 | while (true) { | |
5278 | err = bpf_btf_get_next_id(id, &id); | |
5279 | if (err && errno == ENOENT) | |
5280 | return 0; | |
5281 | if (err) { | |
5282 | err = -errno; | |
5283 | pr_warn("failed to iterate BTF objects: %d\n", err); | |
5284 | return err; | |
ddc7c304 | 5285 | } |
4f33a53d AN |
5286 | |
5287 | fd = bpf_btf_get_fd_by_id(id); | |
5288 | if (fd < 0) { | |
5289 | if (errno == ENOENT) | |
5290 | continue; /* expected race: BTF was unloaded */ | |
5291 | err = -errno; | |
5292 | pr_warn("failed to get BTF object #%d FD: %d\n", id, err); | |
5293 | return err; | |
5294 | } | |
5295 | ||
5296 | len = sizeof(info); | |
5297 | memset(&info, 0, sizeof(info)); | |
5298 | info.name = ptr_to_u64(name); | |
5299 | info.name_len = sizeof(name); | |
5300 | ||
5301 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
5302 | if (err) { | |
5303 | err = -errno; | |
5304 | pr_warn("failed to get BTF object #%d info: %d\n", id, err); | |
91abb4a6 | 5305 | goto err_out; |
4f33a53d AN |
5306 | } |
5307 | ||
5308 | /* ignore non-module BTFs */ | |
5309 | if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) { | |
5310 | close(fd); | |
5311 | continue; | |
5312 | } | |
5313 | ||
5314 | btf = btf_get_from_fd(fd, obj->btf_vmlinux); | |
e9fc3ce9 AN |
5315 | err = libbpf_get_error(btf); |
5316 | if (err) { | |
5317 | pr_warn("failed to load module [%s]'s BTF object #%d: %d\n", | |
5318 | name, id, err); | |
91abb4a6 | 5319 | goto err_out; |
4f33a53d AN |
5320 | } |
5321 | ||
3b029e06 AN |
5322 | err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap, |
5323 | sizeof(*obj->btf_modules), obj->btf_module_cnt + 1); | |
4f33a53d | 5324 | if (err) |
91abb4a6 | 5325 | goto err_out; |
4f33a53d AN |
5326 | |
5327 | mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; | |
5328 | ||
5329 | mod_btf->btf = btf; | |
5330 | mod_btf->id = id; | |
91abb4a6 | 5331 | mod_btf->fd = fd; |
4f33a53d | 5332 | mod_btf->name = strdup(name); |
91abb4a6 AN |
5333 | if (!mod_btf->name) { |
5334 | err = -ENOMEM; | |
5335 | goto err_out; | |
5336 | } | |
5337 | continue; | |
5338 | ||
5339 | err_out: | |
5340 | close(fd); | |
5341 | return err; | |
ddc7c304 | 5342 | } |
4f33a53d AN |
5343 | |
5344 | return 0; | |
5345 | } | |
5346 | ||
301ba4d7 | 5347 | static struct bpf_core_cand_list * |
0f7515ca AN |
5348 | bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id) |
5349 | { | |
301ba4d7 AS |
5350 | struct bpf_core_cand local_cand = {}; |
5351 | struct bpf_core_cand_list *cands; | |
4f33a53d | 5352 | const struct btf *main_btf; |
03d5b991 AN |
5353 | const struct btf_type *local_t; |
5354 | const char *local_name; | |
0f7515ca | 5355 | size_t local_essent_len; |
4f33a53d | 5356 | int err, i; |
0f7515ca AN |
5357 | |
5358 | local_cand.btf = local_btf; | |
03d5b991 AN |
5359 | local_cand.id = local_type_id; |
5360 | local_t = btf__type_by_id(local_btf, local_type_id); | |
5361 | if (!local_t) | |
0f7515ca AN |
5362 | return ERR_PTR(-EINVAL); |
5363 | ||
03d5b991 AN |
5364 | local_name = btf__name_by_offset(local_btf, local_t->name_off); |
5365 | if (str_is_empty(local_name)) | |
0f7515ca | 5366 | return ERR_PTR(-EINVAL); |
03d5b991 | 5367 | local_essent_len = bpf_core_essential_name_len(local_name); |
0f7515ca AN |
5368 | |
5369 | cands = calloc(1, sizeof(*cands)); | |
5370 | if (!cands) | |
5371 | return ERR_PTR(-ENOMEM); | |
5372 | ||
5373 | /* Attempt to find target candidates in vmlinux BTF first */ | |
4f33a53d AN |
5374 | main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux; |
5375 | err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands); | |
5376 | if (err) | |
5377 | goto err_out; | |
5378 | ||
5379 | /* if vmlinux BTF has any candidate, don't got for module BTFs */ | |
5380 | if (cands->len) | |
5381 | return cands; | |
5382 | ||
5383 | /* if vmlinux BTF was overridden, don't attempt to load module BTFs */ | |
5384 | if (obj->btf_vmlinux_override) | |
5385 | return cands; | |
5386 | ||
5387 | /* now look through module BTFs, trying to still find candidates */ | |
5388 | err = load_module_btfs(obj); | |
5389 | if (err) | |
5390 | goto err_out; | |
5391 | ||
5392 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
5393 | err = bpf_core_add_cands(&local_cand, local_essent_len, | |
5394 | obj->btf_modules[i].btf, | |
5395 | obj->btf_modules[i].name, | |
6a886de0 | 5396 | btf__type_cnt(obj->btf_vmlinux), |
4f33a53d AN |
5397 | cands); |
5398 | if (err) | |
5399 | goto err_out; | |
0f7515ca AN |
5400 | } |
5401 | ||
5402 | return cands; | |
ddc7c304 | 5403 | err_out: |
4f33a53d | 5404 | bpf_core_free_cands(cands); |
ddc7c304 AN |
5405 | return ERR_PTR(err); |
5406 | } | |
5407 | ||
3fc32f40 AN |
5408 | /* Check local and target types for compatibility. This check is used for |
5409 | * type-based CO-RE relocations and follow slightly different rules than | |
5410 | * field-based relocations. This function assumes that root types were already | |
5411 | * checked for name match. Beyond that initial root-level name check, names | |
5412 | * are completely ignored. Compatibility rules are as follows: | |
5413 | * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but | |
5414 | * kind should match for local and target types (i.e., STRUCT is not | |
5415 | * compatible with UNION); | |
5416 | * - for ENUMs, the size is ignored; | |
5417 | * - for INT, size and signedness are ignored; | |
5418 | * - for ARRAY, dimensionality is ignored, element types are checked for | |
5419 | * compatibility recursively; | |
5420 | * - CONST/VOLATILE/RESTRICT modifiers are ignored; | |
5421 | * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible; | |
5422 | * - FUNC_PROTOs are compatible if they have compatible signature: same | |
5423 | * number of input args and compatible return and argument types. | |
5424 | * These rules are not set in stone and probably will be adjusted as we get | |
5425 | * more experience with using BPF CO-RE relocations. | |
5426 | */ | |
b0588390 AS |
5427 | int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id, |
5428 | const struct btf *targ_btf, __u32 targ_id) | |
3fc32f40 AN |
5429 | { |
5430 | const struct btf_type *local_type, *targ_type; | |
5431 | int depth = 32; /* max recursion depth */ | |
5432 | ||
5433 | /* caller made sure that names match (ignoring flavor suffix) */ | |
5434 | local_type = btf__type_by_id(local_btf, local_id); | |
f872e4bc | 5435 | targ_type = btf__type_by_id(targ_btf, targ_id); |
3fc32f40 AN |
5436 | if (btf_kind(local_type) != btf_kind(targ_type)) |
5437 | return 0; | |
5438 | ||
5439 | recur: | |
5440 | depth--; | |
5441 | if (depth < 0) | |
5442 | return -EINVAL; | |
5443 | ||
5444 | local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id); | |
5445 | targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id); | |
5446 | if (!local_type || !targ_type) | |
5447 | return -EINVAL; | |
5448 | ||
5449 | if (btf_kind(local_type) != btf_kind(targ_type)) | |
5450 | return 0; | |
5451 | ||
5452 | switch (btf_kind(local_type)) { | |
5453 | case BTF_KIND_UNKN: | |
5454 | case BTF_KIND_STRUCT: | |
5455 | case BTF_KIND_UNION: | |
5456 | case BTF_KIND_ENUM: | |
5457 | case BTF_KIND_FWD: | |
5458 | return 1; | |
5459 | case BTF_KIND_INT: | |
5460 | /* just reject deprecated bitfield-like integers; all other | |
5461 | * integers are by default compatible between each other | |
5462 | */ | |
5463 | return btf_int_offset(local_type) == 0 && btf_int_offset(targ_type) == 0; | |
5464 | case BTF_KIND_PTR: | |
5465 | local_id = local_type->type; | |
5466 | targ_id = targ_type->type; | |
5467 | goto recur; | |
5468 | case BTF_KIND_ARRAY: | |
5469 | local_id = btf_array(local_type)->type; | |
5470 | targ_id = btf_array(targ_type)->type; | |
5471 | goto recur; | |
5472 | case BTF_KIND_FUNC_PROTO: { | |
5473 | struct btf_param *local_p = btf_params(local_type); | |
5474 | struct btf_param *targ_p = btf_params(targ_type); | |
5475 | __u16 local_vlen = btf_vlen(local_type); | |
5476 | __u16 targ_vlen = btf_vlen(targ_type); | |
5477 | int i, err; | |
5478 | ||
5479 | if (local_vlen != targ_vlen) | |
5480 | return 0; | |
5481 | ||
5482 | for (i = 0; i < local_vlen; i++, local_p++, targ_p++) { | |
5483 | skip_mods_and_typedefs(local_btf, local_p->type, &local_id); | |
5484 | skip_mods_and_typedefs(targ_btf, targ_p->type, &targ_id); | |
5485 | err = bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id); | |
5486 | if (err <= 0) | |
5487 | return err; | |
5488 | } | |
5489 | ||
5490 | /* tail recurse for return type check */ | |
5491 | skip_mods_and_typedefs(local_btf, local_type->type, &local_id); | |
5492 | skip_mods_and_typedefs(targ_btf, targ_type->type, &targ_id); | |
5493 | goto recur; | |
5494 | } | |
5495 | default: | |
5496 | pr_warn("unexpected kind %s relocated, local [%d], target [%d]\n", | |
5497 | btf_kind_str(local_type), local_id, targ_id); | |
5498 | return 0; | |
5499 | } | |
5500 | } | |
5501 | ||
ddc7c304 AN |
5502 | static size_t bpf_core_hash_fn(const void *key, void *ctx) |
5503 | { | |
5504 | return (size_t)key; | |
5505 | } | |
5506 | ||
5507 | static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx) | |
5508 | { | |
5509 | return k1 == k2; | |
5510 | } | |
5511 | ||
5512 | static void *u32_as_hash_key(__u32 x) | |
5513 | { | |
5514 | return (void *)(uintptr_t)x; | |
5515 | } | |
5516 | ||
d0e92887 AS |
5517 | static int record_relo_core(struct bpf_program *prog, |
5518 | const struct bpf_core_relo *core_relo, int insn_idx) | |
5519 | { | |
5520 | struct reloc_desc *relos, *relo; | |
5521 | ||
5522 | relos = libbpf_reallocarray(prog->reloc_desc, | |
5523 | prog->nr_reloc + 1, sizeof(*relos)); | |
5524 | if (!relos) | |
5525 | return -ENOMEM; | |
5526 | relo = &relos[prog->nr_reloc]; | |
5527 | relo->type = RELO_CORE; | |
5528 | relo->insn_idx = insn_idx; | |
5529 | relo->core_relo = core_relo; | |
5530 | prog->reloc_desc = relos; | |
5531 | prog->nr_reloc++; | |
5532 | return 0; | |
5533 | } | |
5534 | ||
adb8fa19 MV |
5535 | static int bpf_core_resolve_relo(struct bpf_program *prog, |
5536 | const struct bpf_core_relo *relo, | |
5537 | int relo_idx, | |
5538 | const struct btf *local_btf, | |
5539 | struct hashmap *cand_cache, | |
5540 | struct bpf_core_relo_res *targ_res) | |
3ee4f533 | 5541 | { |
78c1f8d0 | 5542 | struct bpf_core_spec specs_scratch[3] = {}; |
3ee4f533 | 5543 | const void *type_key = u32_as_hash_key(relo->type_id); |
301ba4d7 | 5544 | struct bpf_core_cand_list *cands = NULL; |
3ee4f533 AS |
5545 | const char *prog_name = prog->name; |
5546 | const struct btf_type *local_type; | |
5547 | const char *local_name; | |
5548 | __u32 local_id = relo->type_id; | |
adb8fa19 | 5549 | int err; |
3ee4f533 AS |
5550 | |
5551 | local_type = btf__type_by_id(local_btf, local_id); | |
5552 | if (!local_type) | |
5553 | return -EINVAL; | |
5554 | ||
5555 | local_name = btf__name_by_offset(local_btf, local_type->name_off); | |
5556 | if (!local_name) | |
5557 | return -EINVAL; | |
5558 | ||
46334a0c | 5559 | if (relo->kind != BPF_CORE_TYPE_ID_LOCAL && |
3ee4f533 AS |
5560 | !hashmap__find(cand_cache, type_key, (void **)&cands)) { |
5561 | cands = bpf_core_find_cands(prog->obj, local_btf, local_id); | |
5562 | if (IS_ERR(cands)) { | |
5563 | pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n", | |
5564 | prog_name, relo_idx, local_id, btf_kind_str(local_type), | |
5565 | local_name, PTR_ERR(cands)); | |
5566 | return PTR_ERR(cands); | |
5567 | } | |
5568 | err = hashmap__set(cand_cache, type_key, cands, NULL, NULL); | |
5569 | if (err) { | |
5570 | bpf_core_free_cands(cands); | |
5571 | return err; | |
5572 | } | |
5573 | } | |
5574 | ||
adb8fa19 MV |
5575 | return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch, |
5576 | targ_res); | |
3ee4f533 AS |
5577 | } |
5578 | ||
ddc7c304 | 5579 | static int |
28b93c64 | 5580 | bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) |
ddc7c304 AN |
5581 | { |
5582 | const struct btf_ext_info_sec *sec; | |
adb8fa19 | 5583 | struct bpf_core_relo_res targ_res; |
28b93c64 | 5584 | const struct bpf_core_relo *rec; |
ddc7c304 AN |
5585 | const struct btf_ext_info *seg; |
5586 | struct hashmap_entry *entry; | |
5587 | struct hashmap *cand_cache = NULL; | |
5588 | struct bpf_program *prog; | |
adb8fa19 | 5589 | struct bpf_insn *insn; |
ddc7c304 | 5590 | const char *sec_name; |
db2b8b06 | 5591 | int i, err = 0, insn_idx, sec_idx; |
ddc7c304 | 5592 | |
28b93c64 AN |
5593 | if (obj->btf_ext->core_relo_info.len == 0) |
5594 | return 0; | |
5595 | ||
0f7515ca AN |
5596 | if (targ_btf_path) { |
5597 | obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL); | |
e9fc3ce9 AN |
5598 | err = libbpf_get_error(obj->btf_vmlinux_override); |
5599 | if (err) { | |
0f7515ca AN |
5600 | pr_warn("failed to parse target BTF: %d\n", err); |
5601 | return err; | |
5602 | } | |
ddc7c304 AN |
5603 | } |
5604 | ||
5605 | cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL); | |
5606 | if (IS_ERR(cand_cache)) { | |
5607 | err = PTR_ERR(cand_cache); | |
5608 | goto out; | |
5609 | } | |
5610 | ||
28b93c64 | 5611 | seg = &obj->btf_ext->core_relo_info; |
ddc7c304 AN |
5612 | for_each_btf_ext_sec(seg, sec) { |
5613 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); | |
5614 | if (str_is_empty(sec_name)) { | |
5615 | err = -EINVAL; | |
5616 | goto out; | |
5617 | } | |
db2b8b06 AN |
5618 | /* bpf_object's ELF is gone by now so it's not easy to find |
5619 | * section index by section name, but we can find *any* | |
5620 | * bpf_program within desired section name and use it's | |
5621 | * prog->sec_idx to do a proper search by section index and | |
5622 | * instruction offset | |
5623 | */ | |
9c82a63c AN |
5624 | prog = NULL; |
5625 | for (i = 0; i < obj->nr_programs; i++) { | |
db2b8b06 | 5626 | prog = &obj->programs[i]; |
52109584 | 5627 | if (strcmp(prog->sec_name, sec_name) == 0) |
9c82a63c | 5628 | break; |
9c82a63c | 5629 | } |
ddc7c304 | 5630 | if (!prog) { |
db2b8b06 AN |
5631 | pr_warn("sec '%s': failed to find a BPF program\n", sec_name); |
5632 | return -ENOENT; | |
ddc7c304 | 5633 | } |
db2b8b06 | 5634 | sec_idx = prog->sec_idx; |
ddc7c304 | 5635 | |
9c0f8cbd | 5636 | pr_debug("sec '%s': found %d CO-RE relocations\n", |
ddc7c304 AN |
5637 | sec_name, sec->num_info); |
5638 | ||
5639 | for_each_btf_ext_rec(seg, sec, i, rec) { | |
adb8fa19 MV |
5640 | if (rec->insn_off % BPF_INSN_SZ) |
5641 | return -EINVAL; | |
db2b8b06 AN |
5642 | insn_idx = rec->insn_off / BPF_INSN_SZ; |
5643 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); | |
5644 | if (!prog) { | |
5645 | pr_warn("sec '%s': failed to find program at insn #%d for CO-RE offset relocation #%d\n", | |
5646 | sec_name, insn_idx, i); | |
5647 | err = -EINVAL; | |
5648 | goto out; | |
5649 | } | |
47f7cf63 AN |
5650 | /* no need to apply CO-RE relocation if the program is |
5651 | * not going to be loaded | |
5652 | */ | |
5653 | if (!prog->load) | |
5654 | continue; | |
db2b8b06 | 5655 | |
adb8fa19 MV |
5656 | /* adjust insn_idx from section frame of reference to the local |
5657 | * program's frame of reference; (sub-)program code is not yet | |
5658 | * relocated, so it's enough to just subtract in-section offset | |
5659 | */ | |
5660 | insn_idx = insn_idx - prog->sec_insn_off; | |
5661 | if (insn_idx >= prog->insns_cnt) | |
5662 | return -EINVAL; | |
5663 | insn = &prog->insns[insn_idx]; | |
5664 | ||
5665 | if (prog->obj->gen_loader) { | |
5666 | err = record_relo_core(prog, rec, insn_idx); | |
5667 | if (err) { | |
5668 | pr_warn("prog '%s': relo #%d: failed to record relocation: %d\n", | |
5669 | prog->name, i, err); | |
5670 | goto out; | |
5671 | } | |
5672 | continue; | |
5673 | } | |
5674 | ||
5675 | err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res); | |
ddc7c304 | 5676 | if (err) { |
be18010e | 5677 | pr_warn("prog '%s': relo #%d: failed to relocate: %d\n", |
9c0f8cbd | 5678 | prog->name, i, err); |
ddc7c304 AN |
5679 | goto out; |
5680 | } | |
adb8fa19 MV |
5681 | |
5682 | err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res); | |
5683 | if (err) { | |
5684 | pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n", | |
5685 | prog->name, i, insn_idx, err); | |
5686 | goto out; | |
5687 | } | |
ddc7c304 AN |
5688 | } |
5689 | } | |
5690 | ||
5691 | out: | |
4f33a53d | 5692 | /* obj->btf_vmlinux and module BTFs are freed after object load */ |
0f7515ca AN |
5693 | btf__free(obj->btf_vmlinux_override); |
5694 | obj->btf_vmlinux_override = NULL; | |
5695 | ||
ddc7c304 AN |
5696 | if (!IS_ERR_OR_NULL(cand_cache)) { |
5697 | hashmap__for_each_entry(cand_cache, entry, i) { | |
5698 | bpf_core_free_cands(entry->value); | |
5699 | } | |
5700 | hashmap__free(cand_cache); | |
5701 | } | |
5702 | return err; | |
5703 | } | |
5704 | ||
c3c55696 AN |
5705 | /* Relocate data references within program code: |
5706 | * - map references; | |
5707 | * - global variable references; | |
5708 | * - extern references. | |
5709 | */ | |
48cca7e4 | 5710 | static int |
c3c55696 | 5711 | bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog) |
8a47a6c5 | 5712 | { |
c3c55696 | 5713 | int i; |
8a47a6c5 WN |
5714 | |
5715 | for (i = 0; i < prog->nr_reloc; i++) { | |
53f8dd43 | 5716 | struct reloc_desc *relo = &prog->reloc_desc[i]; |
166750bc | 5717 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; |
2e33efe3 | 5718 | struct extern_desc *ext; |
8a47a6c5 | 5719 | |
166750bc AN |
5720 | switch (relo->type) { |
5721 | case RELO_LD64: | |
e2fa0156 AS |
5722 | if (obj->gen_loader) { |
5723 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX; | |
5724 | insn[0].imm = relo->map_idx; | |
5725 | } else { | |
5726 | insn[0].src_reg = BPF_PSEUDO_MAP_FD; | |
5727 | insn[0].imm = obj->maps[relo->map_idx].fd; | |
5728 | } | |
166750bc AN |
5729 | break; |
5730 | case RELO_DATA: | |
166750bc | 5731 | insn[1].imm = insn[0].imm + relo->sym_off; |
e2fa0156 AS |
5732 | if (obj->gen_loader) { |
5733 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
5734 | insn[0].imm = relo->map_idx; | |
5735 | } else { | |
16e0c35c AN |
5736 | const struct bpf_map *map = &obj->maps[relo->map_idx]; |
5737 | ||
229fae38 | 5738 | if (map->skipped) { |
16e0c35c AN |
5739 | pr_warn("prog '%s': relo #%d: kernel doesn't support global data\n", |
5740 | prog->name, i); | |
5741 | return -ENOTSUP; | |
5742 | } | |
e2fa0156 AS |
5743 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; |
5744 | insn[0].imm = obj->maps[relo->map_idx].fd; | |
5745 | } | |
166750bc | 5746 | break; |
0c091e5c | 5747 | case RELO_EXTERN_VAR: |
2e33efe3 | 5748 | ext = &obj->externs[relo->sym_off]; |
1c0c7074 | 5749 | if (ext->type == EXT_KCFG) { |
e2fa0156 AS |
5750 | if (obj->gen_loader) { |
5751 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
5752 | insn[0].imm = obj->kconfig_map_idx; | |
5753 | } else { | |
5754 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; | |
5755 | insn[0].imm = obj->maps[obj->kconfig_map_idx].fd; | |
5756 | } | |
1c0c7074 AN |
5757 | insn[1].imm = ext->kcfg.data_off; |
5758 | } else /* EXT_KSYM */ { | |
2211c825 | 5759 | if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */ |
d370bbe1 | 5760 | insn[0].src_reg = BPF_PSEUDO_BTF_ID; |
284d2587 AN |
5761 | insn[0].imm = ext->ksym.kernel_btf_id; |
5762 | insn[1].imm = ext->ksym.kernel_btf_obj_fd; | |
2211c825 | 5763 | } else { /* typeless ksyms or unresolved typed ksyms */ |
d370bbe1 HL |
5764 | insn[0].imm = (__u32)ext->ksym.addr; |
5765 | insn[1].imm = ext->ksym.addr >> 32; | |
5766 | } | |
1c0c7074 | 5767 | } |
166750bc | 5768 | break; |
5bd022ec MKL |
5769 | case RELO_EXTERN_FUNC: |
5770 | ext = &obj->externs[relo->sym_off]; | |
5771 | insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL; | |
466b2e13 KKD |
5772 | if (ext->is_set) { |
5773 | insn[0].imm = ext->ksym.kernel_btf_id; | |
5774 | insn[0].off = ext->ksym.btf_fd_idx; | |
5775 | } else { /* unresolved weak kfunc */ | |
5776 | insn[0].imm = 0; | |
5777 | insn[0].off = 0; | |
5778 | } | |
5bd022ec | 5779 | break; |
53eddb5e | 5780 | case RELO_SUBPROG_ADDR: |
b1268826 AS |
5781 | if (insn[0].src_reg != BPF_PSEUDO_FUNC) { |
5782 | pr_warn("prog '%s': relo #%d: bad insn\n", | |
5783 | prog->name, i); | |
5784 | return -EINVAL; | |
5785 | } | |
5786 | /* handled already */ | |
53eddb5e | 5787 | break; |
166750bc | 5788 | case RELO_CALL: |
b1268826 | 5789 | /* handled already */ |
166750bc | 5790 | break; |
d0e92887 AS |
5791 | case RELO_CORE: |
5792 | /* will be handled by bpf_program_record_relos() */ | |
5793 | break; | |
166750bc | 5794 | default: |
9c0f8cbd AN |
5795 | pr_warn("prog '%s': relo #%d: bad relo type %d\n", |
5796 | prog->name, i, relo->type); | |
166750bc | 5797 | return -EINVAL; |
8a47a6c5 | 5798 | } |
8a47a6c5 WN |
5799 | } |
5800 | ||
c3c55696 AN |
5801 | return 0; |
5802 | } | |
5803 | ||
8505e870 AN |
5804 | static int adjust_prog_btf_ext_info(const struct bpf_object *obj, |
5805 | const struct bpf_program *prog, | |
5806 | const struct btf_ext_info *ext_info, | |
5807 | void **prog_info, __u32 *prog_rec_cnt, | |
5808 | __u32 *prog_rec_sz) | |
5809 | { | |
5810 | void *copy_start = NULL, *copy_end = NULL; | |
5811 | void *rec, *rec_end, *new_prog_info; | |
5812 | const struct btf_ext_info_sec *sec; | |
5813 | size_t old_sz, new_sz; | |
5814 | const char *sec_name; | |
5815 | int i, off_adj; | |
5816 | ||
5817 | for_each_btf_ext_sec(ext_info, sec) { | |
5818 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); | |
5819 | if (!sec_name) | |
5820 | return -EINVAL; | |
52109584 | 5821 | if (strcmp(sec_name, prog->sec_name) != 0) |
8505e870 AN |
5822 | continue; |
5823 | ||
5824 | for_each_btf_ext_rec(ext_info, sec, i, rec) { | |
5825 | __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ; | |
5826 | ||
5827 | if (insn_off < prog->sec_insn_off) | |
5828 | continue; | |
5829 | if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt) | |
5830 | break; | |
5831 | ||
5832 | if (!copy_start) | |
5833 | copy_start = rec; | |
5834 | copy_end = rec + ext_info->rec_size; | |
5835 | } | |
5836 | ||
5837 | if (!copy_start) | |
5838 | return -ENOENT; | |
5839 | ||
5840 | /* append func/line info of a given (sub-)program to the main | |
5841 | * program func/line info | |
5842 | */ | |
8eb62958 | 5843 | old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size; |
8505e870 AN |
5844 | new_sz = old_sz + (copy_end - copy_start); |
5845 | new_prog_info = realloc(*prog_info, new_sz); | |
5846 | if (!new_prog_info) | |
5847 | return -ENOMEM; | |
5848 | *prog_info = new_prog_info; | |
5849 | *prog_rec_cnt = new_sz / ext_info->rec_size; | |
5850 | memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start); | |
5851 | ||
5852 | /* Kernel instruction offsets are in units of 8-byte | |
5853 | * instructions, while .BTF.ext instruction offsets generated | |
5854 | * by Clang are in units of bytes. So convert Clang offsets | |
5855 | * into kernel offsets and adjust offset according to program | |
5856 | * relocated position. | |
5857 | */ | |
5858 | off_adj = prog->sub_insn_off - prog->sec_insn_off; | |
5859 | rec = new_prog_info + old_sz; | |
5860 | rec_end = new_prog_info + new_sz; | |
5861 | for (; rec < rec_end; rec += ext_info->rec_size) { | |
5862 | __u32 *insn_off = rec; | |
5863 | ||
5864 | *insn_off = *insn_off / BPF_INSN_SZ + off_adj; | |
5865 | } | |
5866 | *prog_rec_sz = ext_info->rec_size; | |
5867 | return 0; | |
5868 | } | |
5869 | ||
5870 | return -ENOENT; | |
5871 | } | |
5872 | ||
5873 | static int | |
5874 | reloc_prog_func_and_line_info(const struct bpf_object *obj, | |
5875 | struct bpf_program *main_prog, | |
5876 | const struct bpf_program *prog) | |
5877 | { | |
5878 | int err; | |
5879 | ||
5880 | /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't | |
5881 | * supprot func/line info | |
5882 | */ | |
9ca1f56a | 5883 | if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC)) |
8505e870 AN |
5884 | return 0; |
5885 | ||
5886 | /* only attempt func info relocation if main program's func_info | |
5887 | * relocation was successful | |
5888 | */ | |
5889 | if (main_prog != prog && !main_prog->func_info) | |
5890 | goto line_info; | |
5891 | ||
5892 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info, | |
5893 | &main_prog->func_info, | |
5894 | &main_prog->func_info_cnt, | |
5895 | &main_prog->func_info_rec_size); | |
5896 | if (err) { | |
5897 | if (err != -ENOENT) { | |
5898 | pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n", | |
5899 | prog->name, err); | |
5900 | return err; | |
5901 | } | |
5902 | if (main_prog->func_info) { | |
5903 | /* | |
5904 | * Some info has already been found but has problem | |
5905 | * in the last btf_ext reloc. Must have to error out. | |
5906 | */ | |
5907 | pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name); | |
5908 | return err; | |
5909 | } | |
5910 | /* Have problem loading the very first info. Ignore the rest. */ | |
5911 | pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n", | |
5912 | prog->name); | |
5913 | } | |
5914 | ||
5915 | line_info: | |
5916 | /* don't relocate line info if main program's relocation failed */ | |
5917 | if (main_prog != prog && !main_prog->line_info) | |
5918 | return 0; | |
5919 | ||
5920 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info, | |
5921 | &main_prog->line_info, | |
5922 | &main_prog->line_info_cnt, | |
5923 | &main_prog->line_info_rec_size); | |
5924 | if (err) { | |
5925 | if (err != -ENOENT) { | |
5926 | pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n", | |
5927 | prog->name, err); | |
5928 | return err; | |
5929 | } | |
5930 | if (main_prog->line_info) { | |
5931 | /* | |
5932 | * Some info has already been found but has problem | |
5933 | * in the last btf_ext reloc. Must have to error out. | |
5934 | */ | |
5935 | pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name); | |
5936 | return err; | |
5937 | } | |
5938 | /* Have problem loading the very first info. Ignore the rest. */ | |
5939 | pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n", | |
5940 | prog->name); | |
5941 | } | |
5942 | return 0; | |
5943 | } | |
5944 | ||
c3c55696 AN |
5945 | static int cmp_relo_by_insn_idx(const void *key, const void *elem) |
5946 | { | |
5947 | size_t insn_idx = *(const size_t *)key; | |
5948 | const struct reloc_desc *relo = elem; | |
5949 | ||
5950 | if (insn_idx == relo->insn_idx) | |
5951 | return 0; | |
5952 | return insn_idx < relo->insn_idx ? -1 : 1; | |
5953 | } | |
5954 | ||
5955 | static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx) | |
5956 | { | |
2a6a9bf2 AN |
5957 | if (!prog->nr_reloc) |
5958 | return NULL; | |
c3c55696 AN |
5959 | return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc, |
5960 | sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx); | |
5961 | } | |
5962 | ||
b1268826 AS |
5963 | static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog) |
5964 | { | |
5965 | int new_cnt = main_prog->nr_reloc + subprog->nr_reloc; | |
5966 | struct reloc_desc *relos; | |
5967 | int i; | |
5968 | ||
5969 | if (main_prog == subprog) | |
5970 | return 0; | |
5971 | relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos)); | |
5972 | if (!relos) | |
5973 | return -ENOMEM; | |
2a6a9bf2 AN |
5974 | if (subprog->nr_reloc) |
5975 | memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc, | |
5976 | sizeof(*relos) * subprog->nr_reloc); | |
b1268826 AS |
5977 | |
5978 | for (i = main_prog->nr_reloc; i < new_cnt; i++) | |
5979 | relos[i].insn_idx += subprog->sub_insn_off; | |
5980 | /* After insn_idx adjustment the 'relos' array is still sorted | |
5981 | * by insn_idx and doesn't break bsearch. | |
5982 | */ | |
5983 | main_prog->reloc_desc = relos; | |
5984 | main_prog->nr_reloc = new_cnt; | |
5985 | return 0; | |
5986 | } | |
5987 | ||
c3c55696 AN |
5988 | static int |
5989 | bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog, | |
5990 | struct bpf_program *prog) | |
5991 | { | |
5992 | size_t sub_insn_idx, insn_idx, new_cnt; | |
5993 | struct bpf_program *subprog; | |
5994 | struct bpf_insn *insns, *insn; | |
5995 | struct reloc_desc *relo; | |
5996 | int err; | |
5997 | ||
5998 | err = reloc_prog_func_and_line_info(obj, main_prog, prog); | |
5999 | if (err) | |
6000 | return err; | |
6001 | ||
6002 | for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) { | |
6003 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
53eddb5e | 6004 | if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn)) |
c3c55696 AN |
6005 | continue; |
6006 | ||
6007 | relo = find_prog_insn_relo(prog, insn_idx); | |
b1268826 AS |
6008 | if (relo && relo->type == RELO_EXTERN_FUNC) |
6009 | /* kfunc relocations will be handled later | |
6010 | * in bpf_object__relocate_data() | |
6011 | */ | |
6012 | continue; | |
53eddb5e | 6013 | if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) { |
c3c55696 AN |
6014 | pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n", |
6015 | prog->name, insn_idx, relo->type); | |
6016 | return -LIBBPF_ERRNO__RELOC; | |
6017 | } | |
6018 | if (relo) { | |
6019 | /* sub-program instruction index is a combination of | |
6020 | * an offset of a symbol pointed to by relocation and | |
6021 | * call instruction's imm field; for global functions, | |
6022 | * call always has imm = -1, but for static functions | |
6023 | * relocation is against STT_SECTION and insn->imm | |
6024 | * points to a start of a static function | |
53eddb5e YS |
6025 | * |
6026 | * for subprog addr relocation, the relo->sym_off + insn->imm is | |
6027 | * the byte offset in the corresponding section. | |
c3c55696 | 6028 | */ |
53eddb5e YS |
6029 | if (relo->type == RELO_CALL) |
6030 | sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1; | |
6031 | else | |
6032 | sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ; | |
6033 | } else if (insn_is_pseudo_func(insn)) { | |
6034 | /* | |
6035 | * RELO_SUBPROG_ADDR relo is always emitted even if both | |
6036 | * functions are in the same section, so it shouldn't reach here. | |
6037 | */ | |
6038 | pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n", | |
6039 | prog->name, insn_idx); | |
6040 | return -LIBBPF_ERRNO__RELOC; | |
c3c55696 AN |
6041 | } else { |
6042 | /* if subprogram call is to a static function within | |
6043 | * the same ELF section, there won't be any relocation | |
6044 | * emitted, but it also means there is no additional | |
6045 | * offset necessary, insns->imm is relative to | |
6046 | * instruction's original position within the section | |
6047 | */ | |
6048 | sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1; | |
6049 | } | |
6050 | ||
6051 | /* we enforce that sub-programs should be in .text section */ | |
6052 | subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx); | |
6053 | if (!subprog) { | |
6054 | pr_warn("prog '%s': no .text section found yet sub-program call exists\n", | |
6055 | prog->name); | |
6056 | return -LIBBPF_ERRNO__RELOC; | |
6057 | } | |
6058 | ||
6059 | /* if it's the first call instruction calling into this | |
6060 | * subprogram (meaning this subprog hasn't been processed | |
6061 | * yet) within the context of current main program: | |
6062 | * - append it at the end of main program's instructions blog; | |
6063 | * - process is recursively, while current program is put on hold; | |
6064 | * - if that subprogram calls some other not yet processes | |
6065 | * subprogram, same thing will happen recursively until | |
6066 | * there are no more unprocesses subprograms left to append | |
6067 | * and relocate. | |
6068 | */ | |
6069 | if (subprog->sub_insn_off == 0) { | |
6070 | subprog->sub_insn_off = main_prog->insns_cnt; | |
6071 | ||
6072 | new_cnt = main_prog->insns_cnt + subprog->insns_cnt; | |
6073 | insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns)); | |
6074 | if (!insns) { | |
6075 | pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name); | |
6076 | return -ENOMEM; | |
6077 | } | |
6078 | main_prog->insns = insns; | |
6079 | main_prog->insns_cnt = new_cnt; | |
6080 | ||
6081 | memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns, | |
6082 | subprog->insns_cnt * sizeof(*insns)); | |
6083 | ||
6084 | pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n", | |
6085 | main_prog->name, subprog->insns_cnt, subprog->name); | |
6086 | ||
b1268826 AS |
6087 | /* The subprog insns are now appended. Append its relos too. */ |
6088 | err = append_subprog_relos(main_prog, subprog); | |
6089 | if (err) | |
6090 | return err; | |
c3c55696 AN |
6091 | err = bpf_object__reloc_code(obj, main_prog, subprog); |
6092 | if (err) | |
6093 | return err; | |
6094 | } | |
6095 | ||
6096 | /* main_prog->insns memory could have been re-allocated, so | |
6097 | * calculate pointer again | |
6098 | */ | |
6099 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
6100 | /* calculate correct instruction position within current main | |
6101 | * prog; each main prog can have a different set of | |
6102 | * subprograms appended (potentially in different order as | |
6103 | * well), so position of any subprog can be different for | |
6104 | * different main programs */ | |
6105 | insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1; | |
6106 | ||
c3c55696 AN |
6107 | pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n", |
6108 | prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off); | |
6109 | } | |
6110 | ||
6111 | return 0; | |
6112 | } | |
6113 | ||
6114 | /* | |
6115 | * Relocate sub-program calls. | |
6116 | * | |
6117 | * Algorithm operates as follows. Each entry-point BPF program (referred to as | |
6118 | * main prog) is processed separately. For each subprog (non-entry functions, | |
6119 | * that can be called from either entry progs or other subprogs) gets their | |
6120 | * sub_insn_off reset to zero. This serves as indicator that this subprogram | |
6121 | * hasn't been yet appended and relocated within current main prog. Once its | |
6122 | * relocated, sub_insn_off will point at the position within current main prog | |
6123 | * where given subprog was appended. This will further be used to relocate all | |
6124 | * the call instructions jumping into this subprog. | |
6125 | * | |
6126 | * We start with main program and process all call instructions. If the call | |
6127 | * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off | |
6128 | * is zero), subprog instructions are appended at the end of main program's | |
6129 | * instruction array. Then main program is "put on hold" while we recursively | |
6130 | * process newly appended subprogram. If that subprogram calls into another | |
6131 | * subprogram that hasn't been appended, new subprogram is appended again to | |
6132 | * the *main* prog's instructions (subprog's instructions are always left | |
6133 | * untouched, as they need to be in unmodified state for subsequent main progs | |
6134 | * and subprog instructions are always sent only as part of a main prog) and | |
6135 | * the process continues recursively. Once all the subprogs called from a main | |
6136 | * prog or any of its subprogs are appended (and relocated), all their | |
6137 | * positions within finalized instructions array are known, so it's easy to | |
6138 | * rewrite call instructions with correct relative offsets, corresponding to | |
6139 | * desired target subprog. | |
6140 | * | |
6141 | * Its important to realize that some subprogs might not be called from some | |
6142 | * main prog and any of its called/used subprogs. Those will keep their | |
6143 | * subprog->sub_insn_off as zero at all times and won't be appended to current | |
6144 | * main prog and won't be relocated within the context of current main prog. | |
6145 | * They might still be used from other main progs later. | |
6146 | * | |
6147 | * Visually this process can be shown as below. Suppose we have two main | |
6148 | * programs mainA and mainB and BPF object contains three subprogs: subA, | |
6149 | * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and | |
6150 | * subC both call subB: | |
6151 | * | |
6152 | * +--------+ +-------+ | |
6153 | * | v v | | |
6154 | * +--+---+ +--+-+-+ +---+--+ | |
6155 | * | subA | | subB | | subC | | |
6156 | * +--+---+ +------+ +---+--+ | |
6157 | * ^ ^ | |
6158 | * | | | |
6159 | * +---+-------+ +------+----+ | |
6160 | * | mainA | | mainB | | |
6161 | * +-----------+ +-----------+ | |
6162 | * | |
6163 | * We'll start relocating mainA, will find subA, append it and start | |
6164 | * processing sub A recursively: | |
6165 | * | |
6166 | * +-----------+------+ | |
6167 | * | mainA | subA | | |
6168 | * +-----------+------+ | |
6169 | * | |
6170 | * At this point we notice that subB is used from subA, so we append it and | |
6171 | * relocate (there are no further subcalls from subB): | |
6172 | * | |
6173 | * +-----------+------+------+ | |
6174 | * | mainA | subA | subB | | |
6175 | * +-----------+------+------+ | |
6176 | * | |
6177 | * At this point, we relocate subA calls, then go one level up and finish with | |
6178 | * relocatin mainA calls. mainA is done. | |
6179 | * | |
6180 | * For mainB process is similar but results in different order. We start with | |
6181 | * mainB and skip subA and subB, as mainB never calls them (at least | |
6182 | * directly), but we see subC is needed, so we append and start processing it: | |
6183 | * | |
6184 | * +-----------+------+ | |
6185 | * | mainB | subC | | |
6186 | * +-----------+------+ | |
6187 | * Now we see subC needs subB, so we go back to it, append and relocate it: | |
6188 | * | |
6189 | * +-----------+------+------+ | |
6190 | * | mainB | subC | subB | | |
6191 | * +-----------+------+------+ | |
6192 | * | |
6193 | * At this point we unwind recursion, relocate calls in subC, then in mainB. | |
6194 | */ | |
6195 | static int | |
6196 | bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog) | |
6197 | { | |
6198 | struct bpf_program *subprog; | |
d3d93e34 | 6199 | int i, err; |
c3c55696 | 6200 | |
c3c55696 AN |
6201 | /* mark all subprogs as not relocated (yet) within the context of |
6202 | * current main program | |
6203 | */ | |
6204 | for (i = 0; i < obj->nr_programs; i++) { | |
6205 | subprog = &obj->programs[i]; | |
6206 | if (!prog_is_subprog(obj, subprog)) | |
6207 | continue; | |
6208 | ||
6209 | subprog->sub_insn_off = 0; | |
c3c55696 AN |
6210 | } |
6211 | ||
6212 | err = bpf_object__reloc_code(obj, prog, prog); | |
6213 | if (err) | |
6214 | return err; | |
6215 | ||
6216 | ||
8a47a6c5 WN |
6217 | return 0; |
6218 | } | |
6219 | ||
67234743 AS |
6220 | static void |
6221 | bpf_object__free_relocs(struct bpf_object *obj) | |
6222 | { | |
6223 | struct bpf_program *prog; | |
6224 | int i; | |
6225 | ||
6226 | /* free up relocation descriptors */ | |
6227 | for (i = 0; i < obj->nr_programs; i++) { | |
6228 | prog = &obj->programs[i]; | |
6229 | zfree(&prog->reloc_desc); | |
6230 | prog->nr_reloc = 0; | |
6231 | } | |
6232 | } | |
6233 | ||
d0e92887 AS |
6234 | static int cmp_relocs(const void *_a, const void *_b) |
6235 | { | |
6236 | const struct reloc_desc *a = _a; | |
6237 | const struct reloc_desc *b = _b; | |
6238 | ||
6239 | if (a->insn_idx != b->insn_idx) | |
6240 | return a->insn_idx < b->insn_idx ? -1 : 1; | |
6241 | ||
6242 | /* no two relocations should have the same insn_idx, but ... */ | |
6243 | if (a->type != b->type) | |
6244 | return a->type < b->type ? -1 : 1; | |
6245 | ||
6246 | return 0; | |
6247 | } | |
6248 | ||
6249 | static void bpf_object__sort_relos(struct bpf_object *obj) | |
6250 | { | |
6251 | int i; | |
6252 | ||
6253 | for (i = 0; i < obj->nr_programs; i++) { | |
6254 | struct bpf_program *p = &obj->programs[i]; | |
6255 | ||
6256 | if (!p->nr_reloc) | |
6257 | continue; | |
6258 | ||
6259 | qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs); | |
6260 | } | |
6261 | } | |
6262 | ||
8a47a6c5 | 6263 | static int |
ddc7c304 | 6264 | bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path) |
8a47a6c5 WN |
6265 | { |
6266 | struct bpf_program *prog; | |
b1268826 | 6267 | size_t i, j; |
8a47a6c5 WN |
6268 | int err; |
6269 | ||
ddc7c304 AN |
6270 | if (obj->btf_ext) { |
6271 | err = bpf_object__relocate_core(obj, targ_btf_path); | |
6272 | if (err) { | |
be18010e KW |
6273 | pr_warn("failed to perform CO-RE relocations: %d\n", |
6274 | err); | |
ddc7c304 AN |
6275 | return err; |
6276 | } | |
d0e92887 AS |
6277 | if (obj->gen_loader) |
6278 | bpf_object__sort_relos(obj); | |
ddc7c304 | 6279 | } |
b1268826 AS |
6280 | |
6281 | /* Before relocating calls pre-process relocations and mark | |
6282 | * few ld_imm64 instructions that points to subprogs. | |
6283 | * Otherwise bpf_object__reloc_code() later would have to consider | |
6284 | * all ld_imm64 insns as relocation candidates. That would | |
6285 | * reduce relocation speed, since amount of find_prog_insn_relo() | |
6286 | * would increase and most of them will fail to find a relo. | |
9173cac3 AN |
6287 | */ |
6288 | for (i = 0; i < obj->nr_programs; i++) { | |
6289 | prog = &obj->programs[i]; | |
b1268826 AS |
6290 | for (j = 0; j < prog->nr_reloc; j++) { |
6291 | struct reloc_desc *relo = &prog->reloc_desc[j]; | |
6292 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; | |
6293 | ||
6294 | /* mark the insn, so it's recognized by insn_is_pseudo_func() */ | |
6295 | if (relo->type == RELO_SUBPROG_ADDR) | |
6296 | insn[0].src_reg = BPF_PSEUDO_FUNC; | |
9173cac3 | 6297 | } |
9173cac3 | 6298 | } |
b1268826 AS |
6299 | |
6300 | /* relocate subprogram calls and append used subprograms to main | |
c3c55696 AN |
6301 | * programs; each copy of subprogram code needs to be relocated |
6302 | * differently for each main program, because its code location might | |
b1268826 AS |
6303 | * have changed. |
6304 | * Append subprog relos to main programs to allow data relos to be | |
6305 | * processed after text is completely relocated. | |
9173cac3 | 6306 | */ |
8a47a6c5 WN |
6307 | for (i = 0; i < obj->nr_programs; i++) { |
6308 | prog = &obj->programs[i]; | |
c3c55696 AN |
6309 | /* sub-program's sub-calls are relocated within the context of |
6310 | * its main program only | |
6311 | */ | |
6312 | if (prog_is_subprog(obj, prog)) | |
9173cac3 | 6313 | continue; |
16e0c35c AN |
6314 | if (!prog->load) |
6315 | continue; | |
8a47a6c5 | 6316 | |
c3c55696 | 6317 | err = bpf_object__relocate_calls(obj, prog); |
8a47a6c5 | 6318 | if (err) { |
9c0f8cbd AN |
6319 | pr_warn("prog '%s': failed to relocate calls: %d\n", |
6320 | prog->name, err); | |
8a47a6c5 WN |
6321 | return err; |
6322 | } | |
6323 | } | |
b1268826 AS |
6324 | /* Process data relos for main programs */ |
6325 | for (i = 0; i < obj->nr_programs; i++) { | |
6326 | prog = &obj->programs[i]; | |
6327 | if (prog_is_subprog(obj, prog)) | |
6328 | continue; | |
16e0c35c AN |
6329 | if (!prog->load) |
6330 | continue; | |
b1268826 AS |
6331 | err = bpf_object__relocate_data(obj, prog); |
6332 | if (err) { | |
6333 | pr_warn("prog '%s': failed to relocate data references: %d\n", | |
6334 | prog->name, err); | |
6335 | return err; | |
6336 | } | |
6337 | } | |
67234743 AS |
6338 | if (!obj->gen_loader) |
6339 | bpf_object__free_relocs(obj); | |
8a47a6c5 WN |
6340 | return 0; |
6341 | } | |
6342 | ||
646f02ff | 6343 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 6344 | Elf64_Shdr *shdr, Elf_Data *data); |
646f02ff AN |
6345 | |
6346 | static int bpf_object__collect_map_relos(struct bpf_object *obj, | |
ad23b723 | 6347 | Elf64_Shdr *shdr, Elf_Data *data) |
646f02ff | 6348 | { |
15728ad3 AN |
6349 | const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *); |
6350 | int i, j, nrels, new_sz; | |
063e6881 | 6351 | const struct btf_var_secinfo *vi = NULL; |
646f02ff | 6352 | const struct btf_type *sec, *var, *def; |
341ac5ff HC |
6353 | struct bpf_map *map = NULL, *targ_map = NULL; |
6354 | struct bpf_program *targ_prog = NULL; | |
6355 | bool is_prog_array, is_map_in_map; | |
646f02ff | 6356 | const struct btf_member *member; |
341ac5ff | 6357 | const char *name, *mname, *type; |
646f02ff | 6358 | unsigned int moff; |
ad23b723 AN |
6359 | Elf64_Sym *sym; |
6360 | Elf64_Rel *rel; | |
646f02ff AN |
6361 | void *tmp; |
6362 | ||
6363 | if (!obj->efile.btf_maps_sec_btf_id || !obj->btf) | |
6364 | return -EINVAL; | |
6365 | sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id); | |
6366 | if (!sec) | |
6367 | return -EINVAL; | |
6368 | ||
646f02ff AN |
6369 | nrels = shdr->sh_size / shdr->sh_entsize; |
6370 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
6371 | rel = elf_rel_by_idx(data, i); |
6372 | if (!rel) { | |
646f02ff AN |
6373 | pr_warn(".maps relo #%d: failed to get ELF relo\n", i); |
6374 | return -LIBBPF_ERRNO__FORMAT; | |
6375 | } | |
ad23b723 AN |
6376 | |
6377 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); | |
6378 | if (!sym) { | |
646f02ff | 6379 | pr_warn(".maps relo #%d: symbol %zx not found\n", |
ad23b723 | 6380 | i, (size_t)ELF64_R_SYM(rel->r_info)); |
646f02ff AN |
6381 | return -LIBBPF_ERRNO__FORMAT; |
6382 | } | |
ad23b723 | 6383 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
646f02ff | 6384 | |
ad23b723 AN |
6385 | pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n", |
6386 | i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value, | |
6387 | (size_t)rel->r_offset, sym->st_name, name); | |
646f02ff AN |
6388 | |
6389 | for (j = 0; j < obj->nr_maps; j++) { | |
6390 | map = &obj->maps[j]; | |
6391 | if (map->sec_idx != obj->efile.btf_maps_shndx) | |
6392 | continue; | |
6393 | ||
6394 | vi = btf_var_secinfos(sec) + map->btf_var_idx; | |
ad23b723 AN |
6395 | if (vi->offset <= rel->r_offset && |
6396 | rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size) | |
646f02ff AN |
6397 | break; |
6398 | } | |
6399 | if (j == obj->nr_maps) { | |
ad23b723 AN |
6400 | pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n", |
6401 | i, name, (size_t)rel->r_offset); | |
646f02ff AN |
6402 | return -EINVAL; |
6403 | } | |
6404 | ||
341ac5ff HC |
6405 | is_map_in_map = bpf_map_type__is_map_in_map(map->def.type); |
6406 | is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY; | |
6407 | type = is_map_in_map ? "map" : "prog"; | |
6408 | if (is_map_in_map) { | |
6409 | if (sym->st_shndx != obj->efile.btf_maps_shndx) { | |
6410 | pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n", | |
6411 | i, name); | |
6412 | return -LIBBPF_ERRNO__RELOC; | |
6413 | } | |
6414 | if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS && | |
6415 | map->def.key_size != sizeof(int)) { | |
6416 | pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n", | |
6417 | i, map->name, sizeof(int)); | |
6418 | return -EINVAL; | |
6419 | } | |
6420 | targ_map = bpf_object__find_map_by_name(obj, name); | |
6421 | if (!targ_map) { | |
6422 | pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n", | |
6423 | i, name); | |
6424 | return -ESRCH; | |
6425 | } | |
6426 | } else if (is_prog_array) { | |
6427 | targ_prog = bpf_object__find_program_by_name(obj, name); | |
6428 | if (!targ_prog) { | |
6429 | pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n", | |
6430 | i, name); | |
6431 | return -ESRCH; | |
6432 | } | |
6433 | if (targ_prog->sec_idx != sym->st_shndx || | |
6434 | targ_prog->sec_insn_off * 8 != sym->st_value || | |
6435 | prog_is_subprog(obj, targ_prog)) { | |
6436 | pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n", | |
6437 | i, name); | |
6438 | return -LIBBPF_ERRNO__RELOC; | |
6439 | } | |
6440 | } else { | |
646f02ff AN |
6441 | return -EINVAL; |
6442 | } | |
6443 | ||
646f02ff AN |
6444 | var = btf__type_by_id(obj->btf, vi->type); |
6445 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
6446 | if (btf_vlen(def) == 0) | |
6447 | return -EINVAL; | |
6448 | member = btf_members(def) + btf_vlen(def) - 1; | |
6449 | mname = btf__name_by_offset(obj->btf, member->name_off); | |
6450 | if (strcmp(mname, "values")) | |
6451 | return -EINVAL; | |
6452 | ||
6453 | moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8; | |
ad23b723 | 6454 | if (rel->r_offset - vi->offset < moff) |
646f02ff AN |
6455 | return -EINVAL; |
6456 | ||
ad23b723 | 6457 | moff = rel->r_offset - vi->offset - moff; |
15728ad3 AN |
6458 | /* here we use BPF pointer size, which is always 64 bit, as we |
6459 | * are parsing ELF that was built for BPF target | |
6460 | */ | |
6461 | if (moff % bpf_ptr_sz) | |
646f02ff | 6462 | return -EINVAL; |
15728ad3 | 6463 | moff /= bpf_ptr_sz; |
646f02ff AN |
6464 | if (moff >= map->init_slots_sz) { |
6465 | new_sz = moff + 1; | |
029258d7 | 6466 | tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz); |
646f02ff AN |
6467 | if (!tmp) |
6468 | return -ENOMEM; | |
6469 | map->init_slots = tmp; | |
6470 | memset(map->init_slots + map->init_slots_sz, 0, | |
15728ad3 | 6471 | (new_sz - map->init_slots_sz) * host_ptr_sz); |
646f02ff AN |
6472 | map->init_slots_sz = new_sz; |
6473 | } | |
341ac5ff | 6474 | map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog; |
646f02ff | 6475 | |
341ac5ff HC |
6476 | pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n", |
6477 | i, map->name, moff, type, name); | |
646f02ff AN |
6478 | } |
6479 | ||
6480 | return 0; | |
6481 | } | |
590a0088 | 6482 | |
c3c55696 AN |
6483 | static int bpf_object__collect_relos(struct bpf_object *obj) |
6484 | { | |
6485 | int i, err; | |
34090915 | 6486 | |
25bbbd7a AN |
6487 | for (i = 0; i < obj->efile.sec_cnt; i++) { |
6488 | struct elf_sec_desc *sec_desc = &obj->efile.secs[i]; | |
6489 | Elf64_Shdr *shdr; | |
6490 | Elf_Data *data; | |
6491 | int idx; | |
6492 | ||
6493 | if (sec_desc->sec_type != SEC_RELO) | |
6494 | continue; | |
6495 | ||
6496 | shdr = sec_desc->shdr; | |
6497 | data = sec_desc->data; | |
6498 | idx = shdr->sh_info; | |
34090915 WN |
6499 | |
6500 | if (shdr->sh_type != SHT_REL) { | |
be18010e | 6501 | pr_warn("internal error at %d\n", __LINE__); |
6371ca3b | 6502 | return -LIBBPF_ERRNO__INTERNAL; |
34090915 WN |
6503 | } |
6504 | ||
c3c55696 | 6505 | if (idx == obj->efile.st_ops_shndx) |
646f02ff | 6506 | err = bpf_object__collect_st_ops_relos(obj, shdr, data); |
c3c55696 | 6507 | else if (idx == obj->efile.btf_maps_shndx) |
646f02ff | 6508 | err = bpf_object__collect_map_relos(obj, shdr, data); |
c3c55696 AN |
6509 | else |
6510 | err = bpf_object__collect_prog_relos(obj, shdr, data); | |
34090915 | 6511 | if (err) |
6371ca3b | 6512 | return err; |
34090915 | 6513 | } |
c3c55696 | 6514 | |
d0e92887 | 6515 | bpf_object__sort_relos(obj); |
34090915 WN |
6516 | return 0; |
6517 | } | |
6518 | ||
109cea5a AN |
6519 | static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id) |
6520 | { | |
9b2f6fec | 6521 | if (BPF_CLASS(insn->code) == BPF_JMP && |
109cea5a AN |
6522 | BPF_OP(insn->code) == BPF_CALL && |
6523 | BPF_SRC(insn->code) == BPF_K && | |
9b2f6fec AN |
6524 | insn->src_reg == 0 && |
6525 | insn->dst_reg == 0) { | |
6526 | *func_id = insn->imm; | |
109cea5a AN |
6527 | return true; |
6528 | } | |
6529 | return false; | |
6530 | } | |
6531 | ||
42869d28 | 6532 | static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog) |
109cea5a AN |
6533 | { |
6534 | struct bpf_insn *insn = prog->insns; | |
6535 | enum bpf_func_id func_id; | |
6536 | int i; | |
6537 | ||
67234743 AS |
6538 | if (obj->gen_loader) |
6539 | return 0; | |
6540 | ||
109cea5a AN |
6541 | for (i = 0; i < prog->insns_cnt; i++, insn++) { |
6542 | if (!insn_is_helper_call(insn, &func_id)) | |
6543 | continue; | |
6544 | ||
6545 | /* on kernels that don't yet support | |
6546 | * bpf_probe_read_{kernel,user}[_str] helpers, fall back | |
6547 | * to bpf_probe_read() which works well for old kernels | |
6548 | */ | |
6549 | switch (func_id) { | |
6550 | case BPF_FUNC_probe_read_kernel: | |
6551 | case BPF_FUNC_probe_read_user: | |
9ca1f56a | 6552 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
6553 | insn->imm = BPF_FUNC_probe_read; |
6554 | break; | |
6555 | case BPF_FUNC_probe_read_kernel_str: | |
6556 | case BPF_FUNC_probe_read_user_str: | |
9ca1f56a | 6557 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
6558 | insn->imm = BPF_FUNC_probe_read_str; |
6559 | break; | |
6560 | default: | |
6561 | break; | |
6562 | } | |
6563 | } | |
6564 | return 0; | |
6565 | } | |
6566 | ||
15ea31fa AN |
6567 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
6568 | int *btf_obj_fd, int *btf_type_id); | |
12d9466d | 6569 | |
4fa5bcfe AN |
6570 | /* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */ |
6571 | static int libbpf_prepare_prog_load(struct bpf_program *prog, | |
6572 | struct bpf_prog_load_opts *opts, long cookie) | |
12d9466d | 6573 | { |
15ea31fa AN |
6574 | enum sec_def_flags def = cookie; |
6575 | ||
12d9466d | 6576 | /* old kernels might not support specifying expected_attach_type */ |
15ea31fa | 6577 | if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE)) |
d10ef2b8 | 6578 | opts->expected_attach_type = 0; |
12d9466d | 6579 | |
15ea31fa | 6580 | if (def & SEC_SLEEPABLE) |
d10ef2b8 | 6581 | opts->prog_flags |= BPF_F_SLEEPABLE; |
12d9466d | 6582 | |
082c4bfb LB |
6583 | if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS)) |
6584 | opts->prog_flags |= BPF_F_XDP_HAS_FRAGS; | |
6585 | ||
4a4d4cee LB |
6586 | if (def & SEC_DEPRECATED) |
6587 | 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", | |
6588 | prog->sec_name); | |
6589 | ||
12d9466d AN |
6590 | if ((prog->type == BPF_PROG_TYPE_TRACING || |
6591 | prog->type == BPF_PROG_TYPE_LSM || | |
6592 | prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) { | |
6593 | int btf_obj_fd = 0, btf_type_id = 0, err; | |
15ea31fa | 6594 | const char *attach_name; |
12d9466d | 6595 | |
15ea31fa AN |
6596 | attach_name = strchr(prog->sec_name, '/') + 1; |
6597 | err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id); | |
12d9466d AN |
6598 | if (err) |
6599 | return err; | |
6600 | ||
6601 | /* cache resolved BTF FD and BTF type ID in the prog */ | |
6602 | prog->attach_btf_obj_fd = btf_obj_fd; | |
6603 | prog->attach_btf_id = btf_type_id; | |
6604 | ||
6605 | /* but by now libbpf common logic is not utilizing | |
6606 | * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because | |
d10ef2b8 AN |
6607 | * this callback is called after opts were populated by |
6608 | * libbpf, so this callback has to update opts explicitly here | |
12d9466d | 6609 | */ |
d10ef2b8 AN |
6610 | opts->attach_btf_obj_fd = btf_obj_fd; |
6611 | opts->attach_btf_id = btf_type_id; | |
12d9466d AN |
6612 | } |
6613 | return 0; | |
6614 | } | |
6615 | ||
be2f2d16 AN |
6616 | static int bpf_object_load_prog_instance(struct bpf_object *obj, struct bpf_program *prog, |
6617 | struct bpf_insn *insns, int insns_cnt, | |
6618 | const char *license, __u32 kern_version, | |
6619 | int *prog_fd) | |
55cffde2 | 6620 | { |
d10ef2b8 AN |
6621 | LIBBPF_OPTS(bpf_prog_load_opts, load_attr); |
6622 | const char *prog_name = NULL; | |
1ce6a9fc | 6623 | char *cp, errmsg[STRERR_BUFSIZE]; |
8395f320 | 6624 | size_t log_buf_size = 0; |
b3ce9079 | 6625 | char *log_buf = NULL, *tmp; |
12d9466d | 6626 | int btf_fd, ret, err; |
b3ce9079 AN |
6627 | bool own_log_buf = true; |
6628 | __u32 log_level = prog->log_level; | |
55cffde2 | 6629 | |
80b2b5c3 AM |
6630 | if (prog->type == BPF_PROG_TYPE_UNSPEC) { |
6631 | /* | |
6632 | * The program type must be set. Most likely we couldn't find a proper | |
6633 | * section definition at load time, and thus we didn't infer the type. | |
6634 | */ | |
6635 | pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n", | |
6636 | prog->name, prog->sec_name); | |
6637 | return -EINVAL; | |
6638 | } | |
6639 | ||
fba01a06 AN |
6640 | if (!insns || !insns_cnt) |
6641 | return -EINVAL; | |
6642 | ||
12d9466d | 6643 | load_attr.expected_attach_type = prog->expected_attach_type; |
25bbbd7a | 6644 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
d10ef2b8 | 6645 | prog_name = prog->name; |
12d9466d AN |
6646 | load_attr.attach_prog_fd = prog->attach_prog_fd; |
6647 | load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd; | |
6aef10a4 AN |
6648 | load_attr.attach_btf_id = prog->attach_btf_id; |
6649 | load_attr.kern_version = kern_version; | |
6650 | load_attr.prog_ifindex = prog->prog_ifindex; | |
6651 | ||
0f0e55d8 | 6652 | /* specify func_info/line_info only if kernel supports them */ |
25bbbd7a AN |
6653 | btf_fd = bpf_object__btf_fd(obj); |
6654 | if (btf_fd >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) { | |
0f0e55d8 AN |
6655 | load_attr.prog_btf_fd = btf_fd; |
6656 | load_attr.func_info = prog->func_info; | |
6657 | load_attr.func_info_rec_size = prog->func_info_rec_size; | |
6658 | load_attr.func_info_cnt = prog->func_info_cnt; | |
6659 | load_attr.line_info = prog->line_info; | |
6660 | load_attr.line_info_rec_size = prog->line_info_rec_size; | |
6661 | load_attr.line_info_cnt = prog->line_info_cnt; | |
6662 | } | |
b3ce9079 | 6663 | load_attr.log_level = log_level; |
04656198 | 6664 | load_attr.prog_flags = prog->prog_flags; |
25bbbd7a | 6665 | load_attr.fd_array = obj->fd_array; |
55cffde2 | 6666 | |
12d9466d | 6667 | /* adjust load_attr if sec_def provides custom preload callback */ |
4fa5bcfe AN |
6668 | if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) { |
6669 | err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie); | |
12d9466d AN |
6670 | if (err < 0) { |
6671 | pr_warn("prog '%s': failed to prepare load attributes: %d\n", | |
6672 | prog->name, err); | |
6673 | return err; | |
6674 | } | |
6675 | } | |
6676 | ||
25bbbd7a | 6677 | if (obj->gen_loader) { |
d10ef2b8 AN |
6678 | bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name, |
6679 | license, insns, insns_cnt, &load_attr, | |
25bbbd7a | 6680 | prog - obj->programs); |
be2f2d16 | 6681 | *prog_fd = -1; |
67234743 AS |
6682 | return 0; |
6683 | } | |
8395f320 | 6684 | |
b3ce9079 AN |
6685 | retry_load: |
6686 | /* if log_level is zero, we don't request logs initiallly even if | |
6687 | * custom log_buf is specified; if the program load fails, then we'll | |
6688 | * bump log_level to 1 and use either custom log_buf or we'll allocate | |
6689 | * our own and retry the load to get details on what failed | |
6690 | */ | |
6691 | if (log_level) { | |
6692 | if (prog->log_buf) { | |
6693 | log_buf = prog->log_buf; | |
6694 | log_buf_size = prog->log_size; | |
6695 | own_log_buf = false; | |
6696 | } else if (obj->log_buf) { | |
6697 | log_buf = obj->log_buf; | |
6698 | log_buf_size = obj->log_size; | |
6699 | own_log_buf = false; | |
6700 | } else { | |
6701 | log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2); | |
6702 | tmp = realloc(log_buf, log_buf_size); | |
6703 | if (!tmp) { | |
6704 | ret = -ENOMEM; | |
6705 | goto out; | |
6706 | } | |
6707 | log_buf = tmp; | |
6708 | log_buf[0] = '\0'; | |
6709 | own_log_buf = true; | |
6710 | } | |
8395f320 | 6711 | } |
55cffde2 | 6712 | |
6aef10a4 | 6713 | load_attr.log_buf = log_buf; |
d10ef2b8 | 6714 | load_attr.log_size = log_buf_size; |
b3ce9079 | 6715 | load_attr.log_level = log_level; |
55cffde2 | 6716 | |
b3ce9079 | 6717 | ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr); |
55cffde2 | 6718 | if (ret >= 0) { |
b3ce9079 | 6719 | if (log_level && own_log_buf) { |
ad9a7f96 AN |
6720 | pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n", |
6721 | prog->name, log_buf); | |
6722 | } | |
5d23328d | 6723 | |
25bbbd7a AN |
6724 | if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) { |
6725 | struct bpf_map *map; | |
6726 | int i; | |
6727 | ||
6728 | for (i = 0; i < obj->nr_maps; i++) { | |
6729 | map = &prog->obj->maps[i]; | |
6730 | if (map->libbpf_type != LIBBPF_MAP_RODATA) | |
6731 | continue; | |
5d23328d | 6732 | |
25bbbd7a AN |
6733 | if (bpf_prog_bind_map(ret, bpf_map__fd(map), NULL)) { |
6734 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); | |
ad9a7f96 AN |
6735 | pr_warn("prog '%s': failed to bind map '%s': %s\n", |
6736 | prog->name, map->real_name, cp); | |
25bbbd7a AN |
6737 | /* Don't fail hard if can't bind rodata. */ |
6738 | } | |
5d23328d YZ |
6739 | } |
6740 | } | |
6741 | ||
be2f2d16 | 6742 | *prog_fd = ret; |
55cffde2 WN |
6743 | ret = 0; |
6744 | goto out; | |
6745 | } | |
6746 | ||
b3ce9079 AN |
6747 | if (log_level == 0) { |
6748 | log_level = 1; | |
da11b417 AS |
6749 | goto retry_load; |
6750 | } | |
b3ce9079 AN |
6751 | /* On ENOSPC, increase log buffer size and retry, unless custom |
6752 | * log_buf is specified. | |
6753 | * Be careful to not overflow u32, though. Kernel's log buf size limit | |
6754 | * isn't part of UAPI so it can always be bumped to full 4GB. So don't | |
6755 | * multiply by 2 unless we are sure we'll fit within 32 bits. | |
6756 | * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2). | |
6757 | */ | |
6758 | if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2) | |
6759 | goto retry_load; | |
2eda2145 AN |
6760 | |
6761 | ret = -errno; | |
24d6a808 | 6762 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
ad9a7f96 | 6763 | pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, cp); |
dc3a2d25 | 6764 | pr_perm_msg(ret); |
55cffde2 | 6765 | |
b3ce9079 | 6766 | if (own_log_buf && log_buf && log_buf[0] != '\0') { |
ad9a7f96 AN |
6767 | pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n", |
6768 | prog->name, log_buf); | |
6769 | } | |
6770 | if (insns_cnt >= BPF_MAXINSNS) { | |
6771 | pr_warn("prog '%s': program too large (%d insns), at most %d insns\n", | |
6772 | prog->name, insns_cnt, BPF_MAXINSNS); | |
55cffde2 WN |
6773 | } |
6774 | ||
6775 | out: | |
b3ce9079 AN |
6776 | if (own_log_buf) |
6777 | free(log_buf); | |
55cffde2 WN |
6778 | return ret; |
6779 | } | |
6780 | ||
d0e92887 | 6781 | static int bpf_program_record_relos(struct bpf_program *prog) |
67234743 AS |
6782 | { |
6783 | struct bpf_object *obj = prog->obj; | |
6784 | int i; | |
6785 | ||
6786 | for (i = 0; i < prog->nr_reloc; i++) { | |
6787 | struct reloc_desc *relo = &prog->reloc_desc[i]; | |
6788 | struct extern_desc *ext = &obj->externs[relo->sym_off]; | |
6789 | ||
6790 | switch (relo->type) { | |
6791 | case RELO_EXTERN_VAR: | |
6792 | if (ext->type != EXT_KSYM) | |
6793 | continue; | |
c24941cd KKD |
6794 | bpf_gen__record_extern(obj->gen_loader, ext->name, |
6795 | ext->is_weak, !ext->ksym.type_id, | |
18f4fccb | 6796 | BTF_KIND_VAR, relo->insn_idx); |
67234743 AS |
6797 | break; |
6798 | case RELO_EXTERN_FUNC: | |
c24941cd KKD |
6799 | bpf_gen__record_extern(obj->gen_loader, ext->name, |
6800 | ext->is_weak, false, BTF_KIND_FUNC, | |
67234743 AS |
6801 | relo->insn_idx); |
6802 | break; | |
d0e92887 AS |
6803 | case RELO_CORE: { |
6804 | struct bpf_core_relo cr = { | |
6805 | .insn_off = relo->insn_idx * 8, | |
6806 | .type_id = relo->core_relo->type_id, | |
6807 | .access_str_off = relo->core_relo->access_str_off, | |
6808 | .kind = relo->core_relo->kind, | |
6809 | }; | |
6810 | ||
6811 | bpf_gen__record_relo_core(obj->gen_loader, &cr); | |
6812 | break; | |
6813 | } | |
67234743 AS |
6814 | default: |
6815 | continue; | |
6816 | } | |
6817 | } | |
6818 | return 0; | |
6819 | } | |
6820 | ||
be2f2d16 AN |
6821 | static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog, |
6822 | const char *license, __u32 kern_ver) | |
55cffde2 | 6823 | { |
91abb4a6 | 6824 | int err = 0, fd, i; |
13acb508 | 6825 | |
be2f2d16 | 6826 | if (obj->loaded) { |
52109584 | 6827 | pr_warn("prog '%s': can't load after object was loaded\n", prog->name); |
e9fc3ce9 | 6828 | return libbpf_err(-EINVAL); |
d9297581 AN |
6829 | } |
6830 | ||
b580563e WN |
6831 | if (prog->instances.nr < 0 || !prog->instances.fds) { |
6832 | if (prog->preprocessor) { | |
be18010e | 6833 | pr_warn("Internal error: can't load program '%s'\n", |
52109584 | 6834 | prog->name); |
e9fc3ce9 | 6835 | return libbpf_err(-LIBBPF_ERRNO__INTERNAL); |
b580563e | 6836 | } |
55cffde2 | 6837 | |
b580563e WN |
6838 | prog->instances.fds = malloc(sizeof(int)); |
6839 | if (!prog->instances.fds) { | |
be18010e | 6840 | pr_warn("Not enough memory for BPF fds\n"); |
e9fc3ce9 | 6841 | return libbpf_err(-ENOMEM); |
b580563e WN |
6842 | } |
6843 | prog->instances.nr = 1; | |
6844 | prog->instances.fds[0] = -1; | |
6845 | } | |
6846 | ||
6847 | if (!prog->preprocessor) { | |
6848 | if (prog->instances.nr != 1) { | |
52109584 AN |
6849 | pr_warn("prog '%s': inconsistent nr(%d) != 1\n", |
6850 | prog->name, prog->instances.nr); | |
b580563e | 6851 | } |
be2f2d16 | 6852 | if (obj->gen_loader) |
d0e92887 | 6853 | bpf_program_record_relos(prog); |
be2f2d16 AN |
6854 | err = bpf_object_load_prog_instance(obj, prog, |
6855 | prog->insns, prog->insns_cnt, | |
6856 | license, kern_ver, &fd); | |
b580563e WN |
6857 | if (!err) |
6858 | prog->instances.fds[0] = fd; | |
6859 | goto out; | |
6860 | } | |
6861 | ||
6862 | for (i = 0; i < prog->instances.nr; i++) { | |
6863 | struct bpf_prog_prep_result result; | |
6864 | bpf_program_prep_t preprocessor = prog->preprocessor; | |
6865 | ||
1ad9cbb8 | 6866 | memset(&result, 0, sizeof(result)); |
b580563e WN |
6867 | err = preprocessor(prog, i, prog->insns, |
6868 | prog->insns_cnt, &result); | |
6869 | if (err) { | |
be18010e | 6870 | pr_warn("Preprocessing the %dth instance of program '%s' failed\n", |
52109584 | 6871 | i, prog->name); |
b580563e WN |
6872 | goto out; |
6873 | } | |
6874 | ||
6875 | if (!result.new_insn_ptr || !result.new_insn_cnt) { | |
6876 | pr_debug("Skip loading the %dth instance of program '%s'\n", | |
52109584 | 6877 | i, prog->name); |
b580563e WN |
6878 | prog->instances.fds[i] = -1; |
6879 | if (result.pfd) | |
6880 | *result.pfd = -1; | |
6881 | continue; | |
6882 | } | |
6883 | ||
be2f2d16 AN |
6884 | err = bpf_object_load_prog_instance(obj, prog, |
6885 | result.new_insn_ptr, result.new_insn_cnt, | |
6886 | license, kern_ver, &fd); | |
b580563e | 6887 | if (err) { |
be18010e | 6888 | pr_warn("Loading the %dth instance of program '%s' failed\n", |
52109584 | 6889 | i, prog->name); |
b580563e WN |
6890 | goto out; |
6891 | } | |
6892 | ||
6893 | if (result.pfd) | |
6894 | *result.pfd = fd; | |
6895 | prog->instances.fds[i] = fd; | |
6896 | } | |
6897 | out: | |
55cffde2 | 6898 | if (err) |
52109584 | 6899 | pr_warn("failed to load program '%s'\n", prog->name); |
e9fc3ce9 | 6900 | return libbpf_err(err); |
55cffde2 WN |
6901 | } |
6902 | ||
be2f2d16 AN |
6903 | int bpf_program__load(struct bpf_program *prog, const char *license, __u32 kern_ver) |
6904 | { | |
6905 | return bpf_object_load_prog(prog->obj, prog, license, kern_ver); | |
6906 | } | |
6907 | ||
55cffde2 | 6908 | static int |
60276f98 | 6909 | bpf_object__load_progs(struct bpf_object *obj, int log_level) |
55cffde2 | 6910 | { |
d9297581 | 6911 | struct bpf_program *prog; |
55cffde2 WN |
6912 | size_t i; |
6913 | int err; | |
6914 | ||
109cea5a AN |
6915 | for (i = 0; i < obj->nr_programs; i++) { |
6916 | prog = &obj->programs[i]; | |
6917 | err = bpf_object__sanitize_prog(obj, prog); | |
6918 | if (err) | |
6919 | return err; | |
6920 | } | |
6921 | ||
55cffde2 | 6922 | for (i = 0; i < obj->nr_programs; i++) { |
d9297581 | 6923 | prog = &obj->programs[i]; |
c3c55696 | 6924 | if (prog_is_subprog(obj, prog)) |
48cca7e4 | 6925 | continue; |
d9297581 | 6926 | if (!prog->load) { |
9c0f8cbd | 6927 | pr_debug("prog '%s': skipped loading\n", prog->name); |
d9297581 AN |
6928 | continue; |
6929 | } | |
6930 | prog->log_level |= log_level; | |
be2f2d16 | 6931 | err = bpf_object_load_prog(obj, prog, obj->license, obj->kern_version); |
55cffde2 WN |
6932 | if (err) |
6933 | return err; | |
6934 | } | |
67234743 AS |
6935 | if (obj->gen_loader) |
6936 | bpf_object__free_relocs(obj); | |
55cffde2 WN |
6937 | return 0; |
6938 | } | |
6939 | ||
25498a19 AN |
6940 | static const struct bpf_sec_def *find_sec_def(const char *sec_name); |
6941 | ||
91b4d1d1 AN |
6942 | static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts) |
6943 | { | |
6944 | struct bpf_program *prog; | |
12d9466d | 6945 | int err; |
91b4d1d1 AN |
6946 | |
6947 | bpf_object__for_each_program(prog, obj) { | |
6948 | prog->sec_def = find_sec_def(prog->sec_name); | |
6949 | if (!prog->sec_def) { | |
6950 | /* couldn't guess, but user might manually specify */ | |
6951 | pr_debug("prog '%s': unrecognized ELF section name '%s'\n", | |
6952 | prog->name, prog->sec_name); | |
6953 | continue; | |
6954 | } | |
6955 | ||
91b4d1d1 AN |
6956 | bpf_program__set_type(prog, prog->sec_def->prog_type); |
6957 | bpf_program__set_expected_attach_type(prog, prog->sec_def->expected_attach_type); | |
6958 | ||
91b555d7 AN |
6959 | #pragma GCC diagnostic push |
6960 | #pragma GCC diagnostic ignored "-Wdeprecated-declarations" | |
91b4d1d1 AN |
6961 | if (prog->sec_def->prog_type == BPF_PROG_TYPE_TRACING || |
6962 | prog->sec_def->prog_type == BPF_PROG_TYPE_EXT) | |
6963 | prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0); | |
91b555d7 | 6964 | #pragma GCC diagnostic pop |
12d9466d AN |
6965 | |
6966 | /* sec_def can have custom callback which should be called | |
6967 | * after bpf_program is initialized to adjust its properties | |
6968 | */ | |
4fa5bcfe AN |
6969 | if (prog->sec_def->prog_setup_fn) { |
6970 | err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie); | |
12d9466d AN |
6971 | if (err < 0) { |
6972 | pr_warn("prog '%s': failed to initialize: %d\n", | |
6973 | prog->name, err); | |
6974 | return err; | |
6975 | } | |
6976 | } | |
91b4d1d1 AN |
6977 | } |
6978 | ||
6979 | return 0; | |
6980 | } | |
6981 | ||
ad9a7f96 AN |
6982 | static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz, |
6983 | const struct bpf_object_open_opts *opts) | |
1a5e3fb1 | 6984 | { |
1373ff59 | 6985 | const char *obj_name, *kconfig, *btf_tmp_path; |
1a5e3fb1 | 6986 | struct bpf_object *obj; |
291ee02b | 6987 | char tmp_name[64]; |
6371ca3b | 6988 | int err; |
e0e3ea88 AN |
6989 | char *log_buf; |
6990 | size_t log_size; | |
6991 | __u32 log_level; | |
1a5e3fb1 WN |
6992 | |
6993 | if (elf_version(EV_CURRENT) == EV_NONE) { | |
be18010e KW |
6994 | pr_warn("failed to init libelf for %s\n", |
6995 | path ? : "(mem buf)"); | |
6371ca3b | 6996 | return ERR_PTR(-LIBBPF_ERRNO__LIBELF); |
1a5e3fb1 WN |
6997 | } |
6998 | ||
291ee02b AN |
6999 | if (!OPTS_VALID(opts, bpf_object_open_opts)) |
7000 | return ERR_PTR(-EINVAL); | |
7001 | ||
1aace10f | 7002 | obj_name = OPTS_GET(opts, object_name, NULL); |
291ee02b AN |
7003 | if (obj_buf) { |
7004 | if (!obj_name) { | |
7005 | snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx", | |
7006 | (unsigned long)obj_buf, | |
7007 | (unsigned long)obj_buf_sz); | |
7008 | obj_name = tmp_name; | |
7009 | } | |
7010 | path = obj_name; | |
7011 | pr_debug("loading object '%s' from buffer\n", obj_name); | |
7012 | } | |
7013 | ||
e0e3ea88 AN |
7014 | log_buf = OPTS_GET(opts, kernel_log_buf, NULL); |
7015 | log_size = OPTS_GET(opts, kernel_log_size, 0); | |
7016 | log_level = OPTS_GET(opts, kernel_log_level, 0); | |
7017 | if (log_size > UINT_MAX) | |
7018 | return ERR_PTR(-EINVAL); | |
7019 | if (log_size && !log_buf) | |
7020 | return ERR_PTR(-EINVAL); | |
7021 | ||
2ce8450e | 7022 | obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name); |
6371ca3b WN |
7023 | if (IS_ERR(obj)) |
7024 | return obj; | |
1a5e3fb1 | 7025 | |
e0e3ea88 AN |
7026 | obj->log_buf = log_buf; |
7027 | obj->log_size = log_size; | |
7028 | obj->log_level = log_level; | |
7029 | ||
1373ff59 SC |
7030 | btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL); |
7031 | if (btf_tmp_path) { | |
7032 | if (strlen(btf_tmp_path) >= PATH_MAX) { | |
7033 | err = -ENAMETOOLONG; | |
7034 | goto out; | |
7035 | } | |
7036 | obj->btf_custom_path = strdup(btf_tmp_path); | |
7037 | if (!obj->btf_custom_path) { | |
7038 | err = -ENOMEM; | |
7039 | goto out; | |
7040 | } | |
7041 | } | |
7042 | ||
8601fd42 AN |
7043 | kconfig = OPTS_GET(opts, kconfig, NULL); |
7044 | if (kconfig) { | |
7045 | obj->kconfig = strdup(kconfig); | |
18353c87 SC |
7046 | if (!obj->kconfig) { |
7047 | err = -ENOMEM; | |
7048 | goto out; | |
7049 | } | |
166750bc | 7050 | } |
291ee02b | 7051 | |
0d13bfce AN |
7052 | err = bpf_object__elf_init(obj); |
7053 | err = err ? : bpf_object__check_endianness(obj); | |
7054 | err = err ? : bpf_object__elf_collect(obj); | |
166750bc AN |
7055 | err = err ? : bpf_object__collect_externs(obj); |
7056 | err = err ? : bpf_object__finalize_btf(obj); | |
0d13bfce | 7057 | err = err ? : bpf_object__init_maps(obj, opts); |
91b4d1d1 | 7058 | err = err ? : bpf_object_init_progs(obj, opts); |
c3c55696 | 7059 | err = err ? : bpf_object__collect_relos(obj); |
0d13bfce AN |
7060 | if (err) |
7061 | goto out; | |
dd4436bb | 7062 | |
91b4d1d1 | 7063 | bpf_object__elf_finish(obj); |
dd4436bb | 7064 | |
1a5e3fb1 WN |
7065 | return obj; |
7066 | out: | |
7067 | bpf_object__close(obj); | |
6371ca3b | 7068 | return ERR_PTR(err); |
1a5e3fb1 WN |
7069 | } |
7070 | ||
5e61f270 AN |
7071 | static struct bpf_object * |
7072 | __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags) | |
1a5e3fb1 | 7073 | { |
e00aca65 | 7074 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts, |
291ee02b AN |
7075 | .relaxed_maps = flags & MAPS_RELAX_COMPAT, |
7076 | ); | |
7077 | ||
1a5e3fb1 | 7078 | /* param validation */ |
07f2d4ea | 7079 | if (!attr->file) |
1a5e3fb1 WN |
7080 | return NULL; |
7081 | ||
07f2d4ea | 7082 | pr_debug("loading %s\n", attr->file); |
ad9a7f96 | 7083 | return bpf_object_open(attr->file, NULL, 0, &opts); |
c034a177 JF |
7084 | } |
7085 | ||
7086 | struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr) | |
7087 | { | |
e9fc3ce9 | 7088 | return libbpf_ptr(__bpf_object__open_xattr(attr, 0)); |
07f2d4ea JK |
7089 | } |
7090 | ||
7091 | struct bpf_object *bpf_object__open(const char *path) | |
7092 | { | |
7093 | struct bpf_object_open_attr attr = { | |
7094 | .file = path, | |
7095 | .prog_type = BPF_PROG_TYPE_UNSPEC, | |
7096 | }; | |
1a5e3fb1 | 7097 | |
e9fc3ce9 | 7098 | return libbpf_ptr(__bpf_object__open_xattr(&attr, 0)); |
6c956392 WN |
7099 | } |
7100 | ||
2ce8450e | 7101 | struct bpf_object * |
01af3bf0 | 7102 | bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts) |
2ce8450e | 7103 | { |
2ce8450e | 7104 | if (!path) |
e9fc3ce9 | 7105 | return libbpf_err_ptr(-EINVAL); |
2ce8450e AN |
7106 | |
7107 | pr_debug("loading %s\n", path); | |
7108 | ||
ad9a7f96 | 7109 | return libbpf_ptr(bpf_object_open(path, NULL, 0, opts)); |
2ce8450e AN |
7110 | } |
7111 | ||
7112 | struct bpf_object * | |
7113 | bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz, | |
01af3bf0 | 7114 | const struct bpf_object_open_opts *opts) |
6c956392 | 7115 | { |
2ce8450e | 7116 | if (!obj_buf || obj_buf_sz == 0) |
e9fc3ce9 | 7117 | return libbpf_err_ptr(-EINVAL); |
6c956392 | 7118 | |
ad9a7f96 | 7119 | return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, opts)); |
2ce8450e AN |
7120 | } |
7121 | ||
7122 | struct bpf_object * | |
7123 | bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz, | |
7124 | const char *name) | |
7125 | { | |
e00aca65 | 7126 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts, |
2ce8450e AN |
7127 | .object_name = name, |
7128 | /* wrong default, but backwards-compatible */ | |
7129 | .relaxed_maps = true, | |
7130 | ); | |
7131 | ||
7132 | /* returning NULL is wrong, but backwards-compatible */ | |
7133 | if (!obj_buf || obj_buf_sz == 0) | |
e9fc3ce9 | 7134 | return errno = EINVAL, NULL; |
6c956392 | 7135 | |
ad9a7f96 | 7136 | return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, &opts)); |
1a5e3fb1 WN |
7137 | } |
7138 | ||
4a404a7e | 7139 | static int bpf_object_unload(struct bpf_object *obj) |
52d3352e WN |
7140 | { |
7141 | size_t i; | |
7142 | ||
7143 | if (!obj) | |
e9fc3ce9 | 7144 | return libbpf_err(-EINVAL); |
52d3352e | 7145 | |
590a0088 | 7146 | for (i = 0; i < obj->nr_maps; i++) { |
9d759a9b | 7147 | zclose(obj->maps[i].fd); |
590a0088 MKL |
7148 | if (obj->maps[i].st_ops) |
7149 | zfree(&obj->maps[i].st_ops->kern_vdata); | |
7150 | } | |
52d3352e | 7151 | |
55cffde2 WN |
7152 | for (i = 0; i < obj->nr_programs; i++) |
7153 | bpf_program__unload(&obj->programs[i]); | |
7154 | ||
52d3352e WN |
7155 | return 0; |
7156 | } | |
7157 | ||
4a404a7e HC |
7158 | int bpf_object__unload(struct bpf_object *obj) __attribute__((alias("bpf_object_unload"))); |
7159 | ||
0d13bfce AN |
7160 | static int bpf_object__sanitize_maps(struct bpf_object *obj) |
7161 | { | |
7162 | struct bpf_map *m; | |
7163 | ||
7164 | bpf_object__for_each_map(m, obj) { | |
7165 | if (!bpf_map__is_internal(m)) | |
7166 | continue; | |
9ca1f56a | 7167 | if (!kernel_supports(obj, FEAT_ARRAY_MMAP)) |
0d13bfce AN |
7168 | m->def.map_flags ^= BPF_F_MMAPABLE; |
7169 | } | |
7170 | ||
7171 | return 0; | |
7172 | } | |
7173 | ||
1c0c7074 AN |
7174 | static int bpf_object__read_kallsyms_file(struct bpf_object *obj) |
7175 | { | |
7176 | char sym_type, sym_name[500]; | |
7177 | unsigned long long sym_addr; | |
5bd022ec | 7178 | const struct btf_type *t; |
1c0c7074 AN |
7179 | struct extern_desc *ext; |
7180 | int ret, err = 0; | |
7181 | FILE *f; | |
7182 | ||
7183 | f = fopen("/proc/kallsyms", "r"); | |
7184 | if (!f) { | |
7185 | err = -errno; | |
7186 | pr_warn("failed to open /proc/kallsyms: %d\n", err); | |
7187 | return err; | |
7188 | } | |
7189 | ||
7190 | while (true) { | |
7191 | ret = fscanf(f, "%llx %c %499s%*[^\n]\n", | |
7192 | &sym_addr, &sym_type, sym_name); | |
7193 | if (ret == EOF && feof(f)) | |
7194 | break; | |
7195 | if (ret != 3) { | |
135c783f | 7196 | pr_warn("failed to read kallsyms entry: %d\n", ret); |
1c0c7074 AN |
7197 | err = -EINVAL; |
7198 | goto out; | |
7199 | } | |
7200 | ||
7201 | ext = find_extern_by_name(obj, sym_name); | |
7202 | if (!ext || ext->type != EXT_KSYM) | |
7203 | continue; | |
7204 | ||
5bd022ec MKL |
7205 | t = btf__type_by_id(obj->btf, ext->btf_id); |
7206 | if (!btf_is_var(t)) | |
7207 | continue; | |
7208 | ||
1c0c7074 AN |
7209 | if (ext->is_set && ext->ksym.addr != sym_addr) { |
7210 | pr_warn("extern (ksym) '%s' resolution is ambiguous: 0x%llx or 0x%llx\n", | |
7211 | sym_name, ext->ksym.addr, sym_addr); | |
7212 | err = -EINVAL; | |
7213 | goto out; | |
7214 | } | |
7215 | if (!ext->is_set) { | |
7216 | ext->is_set = true; | |
7217 | ext->ksym.addr = sym_addr; | |
7218 | pr_debug("extern (ksym) %s=0x%llx\n", sym_name, sym_addr); | |
7219 | } | |
7220 | } | |
7221 | ||
7222 | out: | |
7223 | fclose(f); | |
7224 | return err; | |
7225 | } | |
7226 | ||
774e132e MKL |
7227 | static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name, |
7228 | __u16 kind, struct btf **res_btf, | |
9dbe6015 | 7229 | struct module_btf **res_mod_btf) |
d370bbe1 | 7230 | { |
9dbe6015 | 7231 | struct module_btf *mod_btf; |
284d2587 | 7232 | struct btf *btf; |
9dbe6015 | 7233 | int i, id, err; |
d370bbe1 | 7234 | |
933d1aa3 | 7235 | btf = obj->btf_vmlinux; |
9dbe6015 | 7236 | mod_btf = NULL; |
774e132e MKL |
7237 | id = btf__find_by_name_kind(btf, ksym_name, kind); |
7238 | ||
933d1aa3 MKL |
7239 | if (id == -ENOENT) { |
7240 | err = load_module_btfs(obj); | |
7241 | if (err) | |
7242 | return err; | |
d370bbe1 | 7243 | |
933d1aa3 | 7244 | for (i = 0; i < obj->btf_module_cnt; i++) { |
9dbe6015 KKD |
7245 | /* we assume module_btf's BTF FD is always >0 */ |
7246 | mod_btf = &obj->btf_modules[i]; | |
7247 | btf = mod_btf->btf; | |
7248 | id = btf__find_by_name_kind_own(btf, ksym_name, kind); | |
933d1aa3 MKL |
7249 | if (id != -ENOENT) |
7250 | break; | |
7251 | } | |
7252 | } | |
2211c825 | 7253 | if (id <= 0) |
933d1aa3 | 7254 | return -ESRCH; |
d370bbe1 | 7255 | |
774e132e | 7256 | *res_btf = btf; |
9dbe6015 | 7257 | *res_mod_btf = mod_btf; |
774e132e MKL |
7258 | return id; |
7259 | } | |
7260 | ||
7261 | static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj, | |
7262 | struct extern_desc *ext) | |
7263 | { | |
7264 | const struct btf_type *targ_var, *targ_type; | |
7265 | __u32 targ_type_id, local_type_id; | |
9dbe6015 | 7266 | struct module_btf *mod_btf = NULL; |
774e132e | 7267 | const char *targ_var_name; |
774e132e | 7268 | struct btf *btf = NULL; |
9dbe6015 | 7269 | int id, err; |
774e132e | 7270 | |
9dbe6015 | 7271 | id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf); |
466b2e13 KKD |
7272 | if (id < 0) { |
7273 | if (id == -ESRCH && ext->is_weak) | |
7274 | return 0; | |
2211c825 HL |
7275 | pr_warn("extern (var ksym) '%s': not found in kernel BTF\n", |
7276 | ext->name); | |
774e132e | 7277 | return id; |
2211c825 | 7278 | } |
774e132e | 7279 | |
933d1aa3 MKL |
7280 | /* find local type_id */ |
7281 | local_type_id = ext->ksym.type_id; | |
284d2587 | 7282 | |
933d1aa3 MKL |
7283 | /* find target type_id */ |
7284 | targ_var = btf__type_by_id(btf, id); | |
7285 | targ_var_name = btf__name_by_offset(btf, targ_var->name_off); | |
7286 | targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id); | |
d370bbe1 | 7287 | |
933d1aa3 MKL |
7288 | err = bpf_core_types_are_compat(obj->btf, local_type_id, |
7289 | btf, targ_type_id); | |
7290 | if (err <= 0) { | |
7291 | const struct btf_type *local_type; | |
7292 | const char *targ_name, *local_name; | |
d370bbe1 | 7293 | |
933d1aa3 MKL |
7294 | local_type = btf__type_by_id(obj->btf, local_type_id); |
7295 | local_name = btf__name_by_offset(obj->btf, local_type->name_off); | |
7296 | targ_name = btf__name_by_offset(btf, targ_type->name_off); | |
d370bbe1 | 7297 | |
933d1aa3 MKL |
7298 | pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n", |
7299 | ext->name, local_type_id, | |
7300 | btf_kind_str(local_type), local_name, targ_type_id, | |
7301 | btf_kind_str(targ_type), targ_name); | |
7302 | return -EINVAL; | |
7303 | } | |
d370bbe1 | 7304 | |
933d1aa3 | 7305 | ext->is_set = true; |
9dbe6015 | 7306 | ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0; |
933d1aa3 MKL |
7307 | ext->ksym.kernel_btf_id = id; |
7308 | pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n", | |
7309 | ext->name, id, btf_kind_str(targ_var), targ_var_name); | |
d370bbe1 | 7310 | |
933d1aa3 MKL |
7311 | return 0; |
7312 | } | |
d370bbe1 | 7313 | |
5bd022ec MKL |
7314 | static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj, |
7315 | struct extern_desc *ext) | |
7316 | { | |
7317 | int local_func_proto_id, kfunc_proto_id, kfunc_id; | |
9dbe6015 | 7318 | struct module_btf *mod_btf = NULL; |
5bd022ec MKL |
7319 | const struct btf_type *kern_func; |
7320 | struct btf *kern_btf = NULL; | |
9dbe6015 | 7321 | int ret; |
5bd022ec MKL |
7322 | |
7323 | local_func_proto_id = ext->ksym.type_id; | |
7324 | ||
9dbe6015 | 7325 | kfunc_id = find_ksym_btf_id(obj, ext->name, BTF_KIND_FUNC, &kern_btf, &mod_btf); |
5bd022ec | 7326 | if (kfunc_id < 0) { |
466b2e13 KKD |
7327 | if (kfunc_id == -ESRCH && ext->is_weak) |
7328 | return 0; | |
7329 | pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n", | |
5bd022ec MKL |
7330 | ext->name); |
7331 | return kfunc_id; | |
7332 | } | |
7333 | ||
5bd022ec MKL |
7334 | kern_func = btf__type_by_id(kern_btf, kfunc_id); |
7335 | kfunc_proto_id = kern_func->type; | |
7336 | ||
7337 | ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id, | |
7338 | kern_btf, kfunc_proto_id); | |
7339 | if (ret <= 0) { | |
7340 | pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with kernel [%d]\n", | |
7341 | ext->name, local_func_proto_id, kfunc_proto_id); | |
7342 | return -EINVAL; | |
7343 | } | |
7344 | ||
9dbe6015 KKD |
7345 | /* set index for module BTF fd in fd_array, if unset */ |
7346 | if (mod_btf && !mod_btf->fd_array_idx) { | |
7347 | /* insn->off is s16 */ | |
7348 | if (obj->fd_array_cnt == INT16_MAX) { | |
7349 | pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n", | |
7350 | ext->name, mod_btf->fd_array_idx); | |
7351 | return -E2BIG; | |
7352 | } | |
7353 | /* Cannot use index 0 for module BTF fd */ | |
7354 | if (!obj->fd_array_cnt) | |
7355 | obj->fd_array_cnt = 1; | |
7356 | ||
7357 | ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int), | |
7358 | obj->fd_array_cnt + 1); | |
7359 | if (ret) | |
7360 | return ret; | |
7361 | mod_btf->fd_array_idx = obj->fd_array_cnt; | |
7362 | /* we assume module BTF FD is always >0 */ | |
7363 | obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd; | |
7364 | } | |
7365 | ||
5bd022ec | 7366 | ext->is_set = true; |
5bd022ec | 7367 | ext->ksym.kernel_btf_id = kfunc_id; |
9dbe6015 | 7368 | ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0; |
5bd022ec MKL |
7369 | pr_debug("extern (func ksym) '%s': resolved to kernel [%d]\n", |
7370 | ext->name, kfunc_id); | |
7371 | ||
7372 | return 0; | |
7373 | } | |
7374 | ||
933d1aa3 MKL |
7375 | static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj) |
7376 | { | |
5bd022ec | 7377 | const struct btf_type *t; |
933d1aa3 MKL |
7378 | struct extern_desc *ext; |
7379 | int i, err; | |
7380 | ||
7381 | for (i = 0; i < obj->nr_extern; i++) { | |
7382 | ext = &obj->externs[i]; | |
7383 | if (ext->type != EXT_KSYM || !ext->ksym.type_id) | |
7384 | continue; | |
7385 | ||
67234743 AS |
7386 | if (obj->gen_loader) { |
7387 | ext->is_set = true; | |
7388 | ext->ksym.kernel_btf_obj_fd = 0; | |
7389 | ext->ksym.kernel_btf_id = 0; | |
7390 | continue; | |
7391 | } | |
5bd022ec MKL |
7392 | t = btf__type_by_id(obj->btf, ext->btf_id); |
7393 | if (btf_is_var(t)) | |
7394 | err = bpf_object__resolve_ksym_var_btf_id(obj, ext); | |
7395 | else | |
7396 | err = bpf_object__resolve_ksym_func_btf_id(obj, ext); | |
933d1aa3 MKL |
7397 | if (err) |
7398 | return err; | |
d370bbe1 HL |
7399 | } |
7400 | return 0; | |
7401 | } | |
7402 | ||
166750bc | 7403 | static int bpf_object__resolve_externs(struct bpf_object *obj, |
8601fd42 | 7404 | const char *extra_kconfig) |
166750bc | 7405 | { |
1c0c7074 | 7406 | bool need_config = false, need_kallsyms = false; |
d370bbe1 | 7407 | bool need_vmlinux_btf = false; |
166750bc | 7408 | struct extern_desc *ext; |
2e33efe3 | 7409 | void *kcfg_data = NULL; |
166750bc | 7410 | int err, i; |
166750bc AN |
7411 | |
7412 | if (obj->nr_extern == 0) | |
7413 | return 0; | |
7414 | ||
2e33efe3 AN |
7415 | if (obj->kconfig_map_idx >= 0) |
7416 | kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped; | |
166750bc AN |
7417 | |
7418 | for (i = 0; i < obj->nr_extern; i++) { | |
7419 | ext = &obj->externs[i]; | |
7420 | ||
2e33efe3 AN |
7421 | if (ext->type == EXT_KCFG && |
7422 | strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) { | |
7423 | void *ext_val = kcfg_data + ext->kcfg.data_off; | |
166750bc AN |
7424 | __u32 kver = get_kernel_version(); |
7425 | ||
7426 | if (!kver) { | |
7427 | pr_warn("failed to get kernel version\n"); | |
7428 | return -EINVAL; | |
7429 | } | |
2e33efe3 | 7430 | err = set_kcfg_value_num(ext, ext_val, kver); |
166750bc AN |
7431 | if (err) |
7432 | return err; | |
2e33efe3 | 7433 | pr_debug("extern (kcfg) %s=0x%x\n", ext->name, kver); |
13d35a0c | 7434 | } else if (ext->type == EXT_KCFG && str_has_pfx(ext->name, "CONFIG_")) { |
166750bc | 7435 | need_config = true; |
1c0c7074 | 7436 | } else if (ext->type == EXT_KSYM) { |
d370bbe1 HL |
7437 | if (ext->ksym.type_id) |
7438 | need_vmlinux_btf = true; | |
7439 | else | |
7440 | need_kallsyms = true; | |
166750bc AN |
7441 | } else { |
7442 | pr_warn("unrecognized extern '%s'\n", ext->name); | |
7443 | return -EINVAL; | |
7444 | } | |
7445 | } | |
8601fd42 | 7446 | if (need_config && extra_kconfig) { |
2e33efe3 | 7447 | err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data); |
8601fd42 AN |
7448 | if (err) |
7449 | return -EINVAL; | |
7450 | need_config = false; | |
7451 | for (i = 0; i < obj->nr_extern; i++) { | |
7452 | ext = &obj->externs[i]; | |
2e33efe3 | 7453 | if (ext->type == EXT_KCFG && !ext->is_set) { |
8601fd42 AN |
7454 | need_config = true; |
7455 | break; | |
7456 | } | |
7457 | } | |
7458 | } | |
166750bc | 7459 | if (need_config) { |
2e33efe3 | 7460 | err = bpf_object__read_kconfig_file(obj, kcfg_data); |
166750bc AN |
7461 | if (err) |
7462 | return -EINVAL; | |
7463 | } | |
1c0c7074 AN |
7464 | if (need_kallsyms) { |
7465 | err = bpf_object__read_kallsyms_file(obj); | |
7466 | if (err) | |
7467 | return -EINVAL; | |
7468 | } | |
d370bbe1 HL |
7469 | if (need_vmlinux_btf) { |
7470 | err = bpf_object__resolve_ksyms_btf_id(obj); | |
7471 | if (err) | |
7472 | return -EINVAL; | |
7473 | } | |
166750bc AN |
7474 | for (i = 0; i < obj->nr_extern; i++) { |
7475 | ext = &obj->externs[i]; | |
7476 | ||
7477 | if (!ext->is_set && !ext->is_weak) { | |
7478 | pr_warn("extern %s (strong) not resolved\n", ext->name); | |
7479 | return -ESRCH; | |
7480 | } else if (!ext->is_set) { | |
7481 | pr_debug("extern %s (weak) not resolved, defaulting to zero\n", | |
7482 | ext->name); | |
7483 | } | |
7484 | } | |
7485 | ||
7486 | return 0; | |
7487 | } | |
7488 | ||
e7b924ca | 7489 | static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path) |
52d3352e | 7490 | { |
ec6d5f47 | 7491 | int err, i; |
6371ca3b | 7492 | |
52d3352e | 7493 | if (!obj) |
e9fc3ce9 | 7494 | return libbpf_err(-EINVAL); |
52d3352e WN |
7495 | |
7496 | if (obj->loaded) { | |
d9297581 | 7497 | pr_warn("object '%s': load can't be attempted twice\n", obj->name); |
e9fc3ce9 | 7498 | return libbpf_err(-EINVAL); |
52d3352e WN |
7499 | } |
7500 | ||
67234743 | 7501 | if (obj->gen_loader) |
be315829 | 7502 | bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps); |
67234743 | 7503 | |
fd9eef1a | 7504 | err = bpf_object__probe_loading(obj); |
fe62de31 | 7505 | err = err ? : bpf_object__load_vmlinux_btf(obj, false); |
8601fd42 | 7506 | err = err ? : bpf_object__resolve_externs(obj, obj->kconfig); |
0d13bfce AN |
7507 | err = err ? : bpf_object__sanitize_and_load_btf(obj); |
7508 | err = err ? : bpf_object__sanitize_maps(obj); | |
590a0088 | 7509 | err = err ? : bpf_object__init_kern_struct_ops_maps(obj); |
0d13bfce | 7510 | err = err ? : bpf_object__create_maps(obj); |
e7b924ca AN |
7511 | err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path); |
7512 | err = err ? : bpf_object__load_progs(obj, extra_log_level); | |
341ac5ff | 7513 | err = err ? : bpf_object_init_prog_arrays(obj); |
a6ed02ca | 7514 | |
67234743 AS |
7515 | if (obj->gen_loader) { |
7516 | /* reset FDs */ | |
4729445b KKD |
7517 | if (obj->btf) |
7518 | btf__set_fd(obj->btf, -1); | |
67234743 AS |
7519 | for (i = 0; i < obj->nr_maps; i++) |
7520 | obj->maps[i].fd = -1; | |
7521 | if (!err) | |
ba05fd36 | 7522 | err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps); |
67234743 AS |
7523 | } |
7524 | ||
9dbe6015 KKD |
7525 | /* clean up fd_array */ |
7526 | zfree(&obj->fd_array); | |
7527 | ||
4f33a53d AN |
7528 | /* clean up module BTFs */ |
7529 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
91abb4a6 | 7530 | close(obj->btf_modules[i].fd); |
4f33a53d AN |
7531 | btf__free(obj->btf_modules[i].btf); |
7532 | free(obj->btf_modules[i].name); | |
7533 | } | |
7534 | free(obj->btf_modules); | |
7535 | ||
7536 | /* clean up vmlinux BTF */ | |
a6ed02ca KS |
7537 | btf__free(obj->btf_vmlinux); |
7538 | obj->btf_vmlinux = NULL; | |
7539 | ||
d9297581 AN |
7540 | obj->loaded = true; /* doesn't matter if successfully or not */ |
7541 | ||
0d13bfce AN |
7542 | if (err) |
7543 | goto out; | |
52d3352e WN |
7544 | |
7545 | return 0; | |
7546 | out: | |
ec6d5f47 THJ |
7547 | /* unpin any maps that were auto-pinned during load */ |
7548 | for (i = 0; i < obj->nr_maps; i++) | |
7549 | if (obj->maps[i].pinned && !obj->maps[i].reused) | |
7550 | bpf_map__unpin(&obj->maps[i], NULL); | |
7551 | ||
4a404a7e | 7552 | bpf_object_unload(obj); |
be18010e | 7553 | pr_warn("failed to load object '%s'\n", obj->path); |
e9fc3ce9 | 7554 | return libbpf_err(err); |
52d3352e WN |
7555 | } |
7556 | ||
e7b924ca | 7557 | int bpf_object__load_xattr(struct bpf_object_load_attr *attr) |
60276f98 | 7558 | { |
e7b924ca AN |
7559 | return bpf_object_load(attr->obj, attr->log_level, attr->target_btf_path); |
7560 | } | |
60276f98 | 7561 | |
e7b924ca AN |
7562 | int bpf_object__load(struct bpf_object *obj) |
7563 | { | |
7564 | return bpf_object_load(obj, 0, NULL); | |
60276f98 QM |
7565 | } |
7566 | ||
196f8487 THJ |
7567 | static int make_parent_dir(const char *path) |
7568 | { | |
7569 | char *cp, errmsg[STRERR_BUFSIZE]; | |
7570 | char *dname, *dir; | |
7571 | int err = 0; | |
7572 | ||
7573 | dname = strdup(path); | |
7574 | if (dname == NULL) | |
7575 | return -ENOMEM; | |
7576 | ||
7577 | dir = dirname(dname); | |
7578 | if (mkdir(dir, 0700) && errno != EEXIST) | |
7579 | err = -errno; | |
7580 | ||
7581 | free(dname); | |
7582 | if (err) { | |
7583 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
7584 | pr_warn("failed to mkdir %s: %s\n", path, cp); | |
7585 | } | |
7586 | return err; | |
7587 | } | |
7588 | ||
f367540c JS |
7589 | static int check_path(const char *path) |
7590 | { | |
1ce6a9fc | 7591 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
7592 | struct statfs st_fs; |
7593 | char *dname, *dir; | |
7594 | int err = 0; | |
7595 | ||
7596 | if (path == NULL) | |
7597 | return -EINVAL; | |
7598 | ||
7599 | dname = strdup(path); | |
7600 | if (dname == NULL) | |
7601 | return -ENOMEM; | |
7602 | ||
7603 | dir = dirname(dname); | |
7604 | if (statfs(dir, &st_fs)) { | |
24d6a808 | 7605 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
be18010e | 7606 | pr_warn("failed to statfs %s: %s\n", dir, cp); |
f367540c JS |
7607 | err = -errno; |
7608 | } | |
7609 | free(dname); | |
7610 | ||
7611 | if (!err && st_fs.f_type != BPF_FS_MAGIC) { | |
be18010e | 7612 | pr_warn("specified path %s is not on BPF FS\n", path); |
f367540c JS |
7613 | err = -EINVAL; |
7614 | } | |
7615 | ||
7616 | return err; | |
7617 | } | |
7618 | ||
e21d585c | 7619 | static int bpf_program_pin_instance(struct bpf_program *prog, const char *path, int instance) |
f367540c | 7620 | { |
1ce6a9fc | 7621 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
7622 | int err; |
7623 | ||
196f8487 THJ |
7624 | err = make_parent_dir(path); |
7625 | if (err) | |
e9fc3ce9 | 7626 | return libbpf_err(err); |
196f8487 | 7627 | |
f367540c JS |
7628 | err = check_path(path); |
7629 | if (err) | |
e9fc3ce9 | 7630 | return libbpf_err(err); |
f367540c JS |
7631 | |
7632 | if (prog == NULL) { | |
be18010e | 7633 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7634 | return libbpf_err(-EINVAL); |
f367540c JS |
7635 | } |
7636 | ||
7637 | if (instance < 0 || instance >= prog->instances.nr) { | |
be18010e | 7638 | pr_warn("invalid prog instance %d of prog %s (max %d)\n", |
52109584 | 7639 | instance, prog->name, prog->instances.nr); |
e9fc3ce9 | 7640 | return libbpf_err(-EINVAL); |
f367540c JS |
7641 | } |
7642 | ||
7643 | if (bpf_obj_pin(prog->instances.fds[instance], path)) { | |
23ab656b THJ |
7644 | err = -errno; |
7645 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e | 7646 | pr_warn("failed to pin program: %s\n", cp); |
e9fc3ce9 | 7647 | return libbpf_err(err); |
f367540c JS |
7648 | } |
7649 | pr_debug("pinned program '%s'\n", path); | |
7650 | ||
7651 | return 0; | |
7652 | } | |
7653 | ||
e21d585c | 7654 | static int bpf_program_unpin_instance(struct bpf_program *prog, const char *path, int instance) |
0c19a9fb SF |
7655 | { |
7656 | int err; | |
7657 | ||
7658 | err = check_path(path); | |
7659 | if (err) | |
e9fc3ce9 | 7660 | return libbpf_err(err); |
0c19a9fb SF |
7661 | |
7662 | if (prog == NULL) { | |
be18010e | 7663 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7664 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7665 | } |
7666 | ||
7667 | if (instance < 0 || instance >= prog->instances.nr) { | |
be18010e | 7668 | pr_warn("invalid prog instance %d of prog %s (max %d)\n", |
52109584 | 7669 | instance, prog->name, prog->instances.nr); |
e9fc3ce9 | 7670 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7671 | } |
7672 | ||
7673 | err = unlink(path); | |
7674 | if (err != 0) | |
e9fc3ce9 AN |
7675 | return libbpf_err(-errno); |
7676 | ||
0c19a9fb SF |
7677 | pr_debug("unpinned program '%s'\n", path); |
7678 | ||
7679 | return 0; | |
7680 | } | |
7681 | ||
e21d585c AN |
7682 | __attribute__((alias("bpf_program_pin_instance"))) |
7683 | int bpf_object__pin_instance(struct bpf_program *prog, const char *path, int instance); | |
7684 | ||
7685 | __attribute__((alias("bpf_program_unpin_instance"))) | |
7686 | int bpf_program__unpin_instance(struct bpf_program *prog, const char *path, int instance); | |
7687 | ||
f367540c JS |
7688 | int bpf_program__pin(struct bpf_program *prog, const char *path) |
7689 | { | |
7690 | int i, err; | |
7691 | ||
196f8487 THJ |
7692 | err = make_parent_dir(path); |
7693 | if (err) | |
e9fc3ce9 | 7694 | return libbpf_err(err); |
196f8487 | 7695 | |
f367540c JS |
7696 | err = check_path(path); |
7697 | if (err) | |
e9fc3ce9 | 7698 | return libbpf_err(err); |
f367540c JS |
7699 | |
7700 | if (prog == NULL) { | |
be18010e | 7701 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7702 | return libbpf_err(-EINVAL); |
f367540c JS |
7703 | } |
7704 | ||
7705 | if (prog->instances.nr <= 0) { | |
52109584 | 7706 | pr_warn("no instances of prog %s to pin\n", prog->name); |
e9fc3ce9 | 7707 | return libbpf_err(-EINVAL); |
f367540c JS |
7708 | } |
7709 | ||
fd734c5c SF |
7710 | if (prog->instances.nr == 1) { |
7711 | /* don't create subdirs when pinning single instance */ | |
e21d585c | 7712 | return bpf_program_pin_instance(prog, path, 0); |
fd734c5c SF |
7713 | } |
7714 | ||
0c19a9fb SF |
7715 | for (i = 0; i < prog->instances.nr; i++) { |
7716 | char buf[PATH_MAX]; | |
7717 | int len; | |
7718 | ||
7719 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7720 | if (len < 0) { | |
7721 | err = -EINVAL; | |
7722 | goto err_unpin; | |
7723 | } else if (len >= PATH_MAX) { | |
7724 | err = -ENAMETOOLONG; | |
7725 | goto err_unpin; | |
7726 | } | |
7727 | ||
e21d585c | 7728 | err = bpf_program_pin_instance(prog, buf, i); |
0c19a9fb SF |
7729 | if (err) |
7730 | goto err_unpin; | |
7731 | } | |
7732 | ||
7733 | return 0; | |
7734 | ||
7735 | err_unpin: | |
7736 | for (i = i - 1; i >= 0; i--) { | |
7737 | char buf[PATH_MAX]; | |
7738 | int len; | |
7739 | ||
7740 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7741 | if (len < 0) | |
7742 | continue; | |
7743 | else if (len >= PATH_MAX) | |
7744 | continue; | |
7745 | ||
e21d585c | 7746 | bpf_program_unpin_instance(prog, buf, i); |
0c19a9fb SF |
7747 | } |
7748 | ||
7749 | rmdir(path); | |
7750 | ||
e9fc3ce9 | 7751 | return libbpf_err(err); |
0c19a9fb SF |
7752 | } |
7753 | ||
7754 | int bpf_program__unpin(struct bpf_program *prog, const char *path) | |
7755 | { | |
7756 | int i, err; | |
7757 | ||
7758 | err = check_path(path); | |
7759 | if (err) | |
e9fc3ce9 | 7760 | return libbpf_err(err); |
0c19a9fb SF |
7761 | |
7762 | if (prog == NULL) { | |
be18010e | 7763 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7764 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7765 | } |
7766 | ||
7767 | if (prog->instances.nr <= 0) { | |
52109584 | 7768 | pr_warn("no instances of prog %s to pin\n", prog->name); |
e9fc3ce9 | 7769 | return libbpf_err(-EINVAL); |
fd734c5c SF |
7770 | } |
7771 | ||
7772 | if (prog->instances.nr == 1) { | |
7773 | /* don't create subdirs when pinning single instance */ | |
e21d585c | 7774 | return bpf_program_unpin_instance(prog, path, 0); |
0c19a9fb SF |
7775 | } |
7776 | ||
f367540c JS |
7777 | for (i = 0; i < prog->instances.nr; i++) { |
7778 | char buf[PATH_MAX]; | |
7779 | int len; | |
7780 | ||
7781 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7782 | if (len < 0) | |
e9fc3ce9 | 7783 | return libbpf_err(-EINVAL); |
f367540c | 7784 | else if (len >= PATH_MAX) |
e9fc3ce9 | 7785 | return libbpf_err(-ENAMETOOLONG); |
f367540c | 7786 | |
e21d585c | 7787 | err = bpf_program_unpin_instance(prog, buf, i); |
f367540c JS |
7788 | if (err) |
7789 | return err; | |
7790 | } | |
7791 | ||
0c19a9fb SF |
7792 | err = rmdir(path); |
7793 | if (err) | |
e9fc3ce9 | 7794 | return libbpf_err(-errno); |
0c19a9fb | 7795 | |
f367540c JS |
7796 | return 0; |
7797 | } | |
7798 | ||
b6989f35 JS |
7799 | int bpf_map__pin(struct bpf_map *map, const char *path) |
7800 | { | |
1ce6a9fc | 7801 | char *cp, errmsg[STRERR_BUFSIZE]; |
b6989f35 JS |
7802 | int err; |
7803 | ||
b6989f35 | 7804 | if (map == NULL) { |
be18010e | 7805 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 7806 | return libbpf_err(-EINVAL); |
b6989f35 JS |
7807 | } |
7808 | ||
4580b25f THJ |
7809 | if (map->pin_path) { |
7810 | if (path && strcmp(path, map->pin_path)) { | |
7811 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
7812 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 7813 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7814 | } else if (map->pinned) { |
7815 | pr_debug("map '%s' already pinned at '%s'; not re-pinning\n", | |
7816 | bpf_map__name(map), map->pin_path); | |
7817 | return 0; | |
7818 | } | |
7819 | } else { | |
7820 | if (!path) { | |
7821 | pr_warn("missing a path to pin map '%s' at\n", | |
7822 | bpf_map__name(map)); | |
e9fc3ce9 | 7823 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7824 | } else if (map->pinned) { |
7825 | pr_warn("map '%s' already pinned\n", bpf_map__name(map)); | |
e9fc3ce9 | 7826 | return libbpf_err(-EEXIST); |
4580b25f THJ |
7827 | } |
7828 | ||
7829 | map->pin_path = strdup(path); | |
7830 | if (!map->pin_path) { | |
7831 | err = -errno; | |
7832 | goto out_err; | |
7833 | } | |
b6989f35 JS |
7834 | } |
7835 | ||
196f8487 THJ |
7836 | err = make_parent_dir(map->pin_path); |
7837 | if (err) | |
e9fc3ce9 | 7838 | return libbpf_err(err); |
196f8487 | 7839 | |
4580b25f THJ |
7840 | err = check_path(map->pin_path); |
7841 | if (err) | |
e9fc3ce9 | 7842 | return libbpf_err(err); |
4580b25f THJ |
7843 | |
7844 | if (bpf_obj_pin(map->fd, map->pin_path)) { | |
7845 | err = -errno; | |
7846 | goto out_err; | |
7847 | } | |
7848 | ||
7849 | map->pinned = true; | |
7850 | pr_debug("pinned map '%s'\n", map->pin_path); | |
0c19a9fb | 7851 | |
b6989f35 | 7852 | return 0; |
4580b25f THJ |
7853 | |
7854 | out_err: | |
7855 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
7856 | pr_warn("failed to pin map: %s\n", cp); | |
e9fc3ce9 | 7857 | return libbpf_err(err); |
b6989f35 JS |
7858 | } |
7859 | ||
0c19a9fb SF |
7860 | int bpf_map__unpin(struct bpf_map *map, const char *path) |
7861 | { | |
7862 | int err; | |
7863 | ||
0c19a9fb | 7864 | if (map == NULL) { |
be18010e | 7865 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 7866 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7867 | } |
7868 | ||
4580b25f THJ |
7869 | if (map->pin_path) { |
7870 | if (path && strcmp(path, map->pin_path)) { | |
7871 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
7872 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 7873 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7874 | } |
7875 | path = map->pin_path; | |
7876 | } else if (!path) { | |
7877 | pr_warn("no path to unpin map '%s' from\n", | |
7878 | bpf_map__name(map)); | |
e9fc3ce9 | 7879 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7880 | } |
7881 | ||
7882 | err = check_path(path); | |
7883 | if (err) | |
e9fc3ce9 | 7884 | return libbpf_err(err); |
4580b25f | 7885 | |
0c19a9fb SF |
7886 | err = unlink(path); |
7887 | if (err != 0) | |
e9fc3ce9 | 7888 | return libbpf_err(-errno); |
4580b25f THJ |
7889 | |
7890 | map->pinned = false; | |
7891 | pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path); | |
0c19a9fb SF |
7892 | |
7893 | return 0; | |
7894 | } | |
7895 | ||
4580b25f THJ |
7896 | int bpf_map__set_pin_path(struct bpf_map *map, const char *path) |
7897 | { | |
7898 | char *new = NULL; | |
7899 | ||
7900 | if (path) { | |
7901 | new = strdup(path); | |
7902 | if (!new) | |
e9fc3ce9 | 7903 | return libbpf_err(-errno); |
4580b25f THJ |
7904 | } |
7905 | ||
7906 | free(map->pin_path); | |
7907 | map->pin_path = new; | |
7908 | return 0; | |
7909 | } | |
7910 | ||
20eccf29 AN |
7911 | __alias(bpf_map__pin_path) |
7912 | const char *bpf_map__get_pin_path(const struct bpf_map *map); | |
4580b25f | 7913 | |
e244d34d EL |
7914 | const char *bpf_map__pin_path(const struct bpf_map *map) |
7915 | { | |
7916 | return map->pin_path; | |
7917 | } | |
7918 | ||
4580b25f THJ |
7919 | bool bpf_map__is_pinned(const struct bpf_map *map) |
7920 | { | |
7921 | return map->pinned; | |
7922 | } | |
7923 | ||
9cf309c5 THJ |
7924 | static void sanitize_pin_path(char *s) |
7925 | { | |
7926 | /* bpffs disallows periods in path names */ | |
7927 | while (*s) { | |
7928 | if (*s == '.') | |
7929 | *s = '_'; | |
7930 | s++; | |
7931 | } | |
7932 | } | |
7933 | ||
0c19a9fb | 7934 | int bpf_object__pin_maps(struct bpf_object *obj, const char *path) |
d5148d85 | 7935 | { |
d5148d85 JS |
7936 | struct bpf_map *map; |
7937 | int err; | |
7938 | ||
7939 | if (!obj) | |
e9fc3ce9 | 7940 | return libbpf_err(-ENOENT); |
d5148d85 JS |
7941 | |
7942 | if (!obj->loaded) { | |
be18010e | 7943 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 7944 | return libbpf_err(-ENOENT); |
d5148d85 JS |
7945 | } |
7946 | ||
f74a53d9 | 7947 | bpf_object__for_each_map(map, obj) { |
4580b25f | 7948 | char *pin_path = NULL; |
0c19a9fb | 7949 | char buf[PATH_MAX]; |
0c19a9fb | 7950 | |
229fae38 SC |
7951 | if (map->skipped) |
7952 | continue; | |
7953 | ||
4580b25f THJ |
7954 | if (path) { |
7955 | int len; | |
7956 | ||
7957 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
7958 | bpf_map__name(map)); | |
7959 | if (len < 0) { | |
7960 | err = -EINVAL; | |
7961 | goto err_unpin_maps; | |
7962 | } else if (len >= PATH_MAX) { | |
7963 | err = -ENAMETOOLONG; | |
7964 | goto err_unpin_maps; | |
7965 | } | |
9cf309c5 | 7966 | sanitize_pin_path(buf); |
4580b25f THJ |
7967 | pin_path = buf; |
7968 | } else if (!map->pin_path) { | |
7969 | continue; | |
0c19a9fb SF |
7970 | } |
7971 | ||
4580b25f | 7972 | err = bpf_map__pin(map, pin_path); |
0c19a9fb SF |
7973 | if (err) |
7974 | goto err_unpin_maps; | |
7975 | } | |
7976 | ||
7977 | return 0; | |
7978 | ||
7979 | err_unpin_maps: | |
bcc40fc0 | 7980 | while ((map = bpf_object__prev_map(obj, map))) { |
4580b25f | 7981 | if (!map->pin_path) |
0c19a9fb SF |
7982 | continue; |
7983 | ||
4580b25f | 7984 | bpf_map__unpin(map, NULL); |
0c19a9fb SF |
7985 | } |
7986 | ||
e9fc3ce9 | 7987 | return libbpf_err(err); |
0c19a9fb SF |
7988 | } |
7989 | ||
7990 | int bpf_object__unpin_maps(struct bpf_object *obj, const char *path) | |
7991 | { | |
7992 | struct bpf_map *map; | |
7993 | int err; | |
7994 | ||
7995 | if (!obj) | |
e9fc3ce9 | 7996 | return libbpf_err(-ENOENT); |
0c19a9fb | 7997 | |
f74a53d9 | 7998 | bpf_object__for_each_map(map, obj) { |
4580b25f | 7999 | char *pin_path = NULL; |
d5148d85 | 8000 | char buf[PATH_MAX]; |
d5148d85 | 8001 | |
4580b25f THJ |
8002 | if (path) { |
8003 | int len; | |
8004 | ||
8005 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
8006 | bpf_map__name(map)); | |
8007 | if (len < 0) | |
e9fc3ce9 | 8008 | return libbpf_err(-EINVAL); |
4580b25f | 8009 | else if (len >= PATH_MAX) |
e9fc3ce9 | 8010 | return libbpf_err(-ENAMETOOLONG); |
9cf309c5 | 8011 | sanitize_pin_path(buf); |
4580b25f THJ |
8012 | pin_path = buf; |
8013 | } else if (!map->pin_path) { | |
8014 | continue; | |
8015 | } | |
d5148d85 | 8016 | |
4580b25f | 8017 | err = bpf_map__unpin(map, pin_path); |
d5148d85 | 8018 | if (err) |
e9fc3ce9 | 8019 | return libbpf_err(err); |
d5148d85 JS |
8020 | } |
8021 | ||
0c19a9fb SF |
8022 | return 0; |
8023 | } | |
8024 | ||
8025 | int bpf_object__pin_programs(struct bpf_object *obj, const char *path) | |
8026 | { | |
8027 | struct bpf_program *prog; | |
8028 | int err; | |
8029 | ||
8030 | if (!obj) | |
e9fc3ce9 | 8031 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
8032 | |
8033 | if (!obj->loaded) { | |
be18010e | 8034 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 8035 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
8036 | } |
8037 | ||
0c19a9fb SF |
8038 | bpf_object__for_each_program(prog, obj) { |
8039 | char buf[PATH_MAX]; | |
8040 | int len; | |
8041 | ||
8042 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 8043 | prog->pin_name); |
0c19a9fb SF |
8044 | if (len < 0) { |
8045 | err = -EINVAL; | |
8046 | goto err_unpin_programs; | |
8047 | } else if (len >= PATH_MAX) { | |
8048 | err = -ENAMETOOLONG; | |
8049 | goto err_unpin_programs; | |
8050 | } | |
8051 | ||
8052 | err = bpf_program__pin(prog, buf); | |
8053 | if (err) | |
8054 | goto err_unpin_programs; | |
8055 | } | |
8056 | ||
8057 | return 0; | |
8058 | ||
8059 | err_unpin_programs: | |
bcc40fc0 | 8060 | while ((prog = bpf_object__prev_program(obj, prog))) { |
0c19a9fb SF |
8061 | char buf[PATH_MAX]; |
8062 | int len; | |
8063 | ||
8064 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 8065 | prog->pin_name); |
0c19a9fb SF |
8066 | if (len < 0) |
8067 | continue; | |
8068 | else if (len >= PATH_MAX) | |
8069 | continue; | |
8070 | ||
8071 | bpf_program__unpin(prog, buf); | |
8072 | } | |
8073 | ||
e9fc3ce9 | 8074 | return libbpf_err(err); |
0c19a9fb SF |
8075 | } |
8076 | ||
8077 | int bpf_object__unpin_programs(struct bpf_object *obj, const char *path) | |
8078 | { | |
8079 | struct bpf_program *prog; | |
8080 | int err; | |
8081 | ||
8082 | if (!obj) | |
e9fc3ce9 | 8083 | return libbpf_err(-ENOENT); |
0c19a9fb | 8084 | |
d5148d85 JS |
8085 | bpf_object__for_each_program(prog, obj) { |
8086 | char buf[PATH_MAX]; | |
8087 | int len; | |
8088 | ||
8089 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 8090 | prog->pin_name); |
d5148d85 | 8091 | if (len < 0) |
e9fc3ce9 | 8092 | return libbpf_err(-EINVAL); |
d5148d85 | 8093 | else if (len >= PATH_MAX) |
e9fc3ce9 | 8094 | return libbpf_err(-ENAMETOOLONG); |
d5148d85 | 8095 | |
0c19a9fb | 8096 | err = bpf_program__unpin(prog, buf); |
d5148d85 | 8097 | if (err) |
e9fc3ce9 | 8098 | return libbpf_err(err); |
d5148d85 JS |
8099 | } |
8100 | ||
8101 | return 0; | |
8102 | } | |
8103 | ||
0c19a9fb SF |
8104 | int bpf_object__pin(struct bpf_object *obj, const char *path) |
8105 | { | |
8106 | int err; | |
8107 | ||
8108 | err = bpf_object__pin_maps(obj, path); | |
8109 | if (err) | |
e9fc3ce9 | 8110 | return libbpf_err(err); |
0c19a9fb SF |
8111 | |
8112 | err = bpf_object__pin_programs(obj, path); | |
8113 | if (err) { | |
8114 | bpf_object__unpin_maps(obj, path); | |
e9fc3ce9 | 8115 | return libbpf_err(err); |
0c19a9fb SF |
8116 | } |
8117 | ||
8118 | return 0; | |
8119 | } | |
8120 | ||
2d39d7c5 AN |
8121 | static void bpf_map__destroy(struct bpf_map *map) |
8122 | { | |
8123 | if (map->clear_priv) | |
8124 | map->clear_priv(map, map->priv); | |
8125 | map->priv = NULL; | |
8126 | map->clear_priv = NULL; | |
8127 | ||
646f02ff AN |
8128 | if (map->inner_map) { |
8129 | bpf_map__destroy(map->inner_map); | |
8130 | zfree(&map->inner_map); | |
8131 | } | |
8132 | ||
8133 | zfree(&map->init_slots); | |
8134 | map->init_slots_sz = 0; | |
8135 | ||
2d39d7c5 AN |
8136 | if (map->mmaped) { |
8137 | munmap(map->mmaped, bpf_map_mmap_sz(map)); | |
8138 | map->mmaped = NULL; | |
8139 | } | |
8140 | ||
8141 | if (map->st_ops) { | |
8142 | zfree(&map->st_ops->data); | |
8143 | zfree(&map->st_ops->progs); | |
8144 | zfree(&map->st_ops->kern_func_off); | |
8145 | zfree(&map->st_ops); | |
8146 | } | |
8147 | ||
8148 | zfree(&map->name); | |
aed65917 | 8149 | zfree(&map->real_name); |
2d39d7c5 AN |
8150 | zfree(&map->pin_path); |
8151 | ||
8152 | if (map->fd >= 0) | |
8153 | zclose(map->fd); | |
8154 | } | |
8155 | ||
1a5e3fb1 WN |
8156 | void bpf_object__close(struct bpf_object *obj) |
8157 | { | |
a5b8bd47 WN |
8158 | size_t i; |
8159 | ||
50450fc7 | 8160 | if (IS_ERR_OR_NULL(obj)) |
1a5e3fb1 WN |
8161 | return; |
8162 | ||
10931d24 WN |
8163 | if (obj->clear_priv) |
8164 | obj->clear_priv(obj, obj->priv); | |
8165 | ||
67234743 | 8166 | bpf_gen__free(obj->gen_loader); |
1a5e3fb1 | 8167 | bpf_object__elf_finish(obj); |
4a404a7e | 8168 | bpf_object_unload(obj); |
8a138aed | 8169 | btf__free(obj->btf); |
2993e051 | 8170 | btf_ext__free(obj->btf_ext); |
1a5e3fb1 | 8171 | |
2d39d7c5 AN |
8172 | for (i = 0; i < obj->nr_maps; i++) |
8173 | bpf_map__destroy(&obj->maps[i]); | |
d859900c | 8174 | |
1373ff59 | 8175 | zfree(&obj->btf_custom_path); |
8601fd42 | 8176 | zfree(&obj->kconfig); |
166750bc AN |
8177 | zfree(&obj->externs); |
8178 | obj->nr_extern = 0; | |
8179 | ||
9d759a9b WN |
8180 | zfree(&obj->maps); |
8181 | obj->nr_maps = 0; | |
a5b8bd47 WN |
8182 | |
8183 | if (obj->programs && obj->nr_programs) { | |
8184 | for (i = 0; i < obj->nr_programs; i++) | |
8185 | bpf_program__exit(&obj->programs[i]); | |
8186 | } | |
8187 | zfree(&obj->programs); | |
8188 | ||
9a208eff | 8189 | list_del(&obj->list); |
1a5e3fb1 WN |
8190 | free(obj); |
8191 | } | |
aa9b1ac3 | 8192 | |
9a208eff WN |
8193 | struct bpf_object * |
8194 | bpf_object__next(struct bpf_object *prev) | |
8195 | { | |
8196 | struct bpf_object *next; | |
689624f0 JB |
8197 | bool strict = (libbpf_mode & LIBBPF_STRICT_NO_OBJECT_LIST); |
8198 | ||
8199 | if (strict) | |
8200 | return NULL; | |
9a208eff WN |
8201 | |
8202 | if (!prev) | |
8203 | next = list_first_entry(&bpf_objects_list, | |
8204 | struct bpf_object, | |
8205 | list); | |
8206 | else | |
8207 | next = list_next_entry(prev, list); | |
8208 | ||
8209 | /* Empty list is noticed here so don't need checking on entry. */ | |
8210 | if (&next->list == &bpf_objects_list) | |
8211 | return NULL; | |
8212 | ||
8213 | return next; | |
8214 | } | |
8215 | ||
a324aae3 | 8216 | const char *bpf_object__name(const struct bpf_object *obj) |
acf860ae | 8217 | { |
e9fc3ce9 | 8218 | return obj ? obj->name : libbpf_err_ptr(-EINVAL); |
acf860ae WN |
8219 | } |
8220 | ||
a324aae3 | 8221 | unsigned int bpf_object__kversion(const struct bpf_object *obj) |
45825d8a | 8222 | { |
a7fe0450 | 8223 | return obj ? obj->kern_version : 0; |
45825d8a WN |
8224 | } |
8225 | ||
a324aae3 | 8226 | struct btf *bpf_object__btf(const struct bpf_object *obj) |
789f6bab AI |
8227 | { |
8228 | return obj ? obj->btf : NULL; | |
8229 | } | |
8230 | ||
8a138aed MKL |
8231 | int bpf_object__btf_fd(const struct bpf_object *obj) |
8232 | { | |
8233 | return obj->btf ? btf__fd(obj->btf) : -1; | |
8234 | } | |
8235 | ||
155f556d RDT |
8236 | int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version) |
8237 | { | |
8238 | if (obj->loaded) | |
e9fc3ce9 | 8239 | return libbpf_err(-EINVAL); |
155f556d RDT |
8240 | |
8241 | obj->kern_version = kern_version; | |
8242 | ||
8243 | return 0; | |
8244 | } | |
8245 | ||
10931d24 WN |
8246 | int bpf_object__set_priv(struct bpf_object *obj, void *priv, |
8247 | bpf_object_clear_priv_t clear_priv) | |
8248 | { | |
8249 | if (obj->priv && obj->clear_priv) | |
8250 | obj->clear_priv(obj, obj->priv); | |
8251 | ||
8252 | obj->priv = priv; | |
8253 | obj->clear_priv = clear_priv; | |
8254 | return 0; | |
8255 | } | |
8256 | ||
a324aae3 | 8257 | void *bpf_object__priv(const struct bpf_object *obj) |
10931d24 | 8258 | { |
e9fc3ce9 | 8259 | return obj ? obj->priv : libbpf_err_ptr(-EINVAL); |
10931d24 WN |
8260 | } |
8261 | ||
67234743 AS |
8262 | int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts) |
8263 | { | |
8264 | struct bpf_gen *gen; | |
8265 | ||
8266 | if (!opts) | |
8267 | return -EFAULT; | |
8268 | if (!OPTS_VALID(opts, gen_loader_opts)) | |
8269 | return -EINVAL; | |
8270 | gen = calloc(sizeof(*gen), 1); | |
8271 | if (!gen) | |
8272 | return -ENOMEM; | |
8273 | gen->opts = opts; | |
8274 | obj->gen_loader = gen; | |
8275 | return 0; | |
8276 | } | |
8277 | ||
eac7d845 | 8278 | static struct bpf_program * |
a324aae3 AN |
8279 | __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj, |
8280 | bool forward) | |
aa9b1ac3 | 8281 | { |
a83d6e76 | 8282 | size_t nr_programs = obj->nr_programs; |
0c19a9fb | 8283 | ssize_t idx; |
aa9b1ac3 | 8284 | |
a83d6e76 | 8285 | if (!nr_programs) |
aa9b1ac3 | 8286 | return NULL; |
aa9b1ac3 | 8287 | |
a83d6e76 MKL |
8288 | if (!p) |
8289 | /* Iter from the beginning */ | |
8290 | return forward ? &obj->programs[0] : | |
8291 | &obj->programs[nr_programs - 1]; | |
8292 | ||
0c19a9fb | 8293 | if (p->obj != obj) { |
be18010e | 8294 | pr_warn("error: program handler doesn't match object\n"); |
e9fc3ce9 | 8295 | return errno = EINVAL, NULL; |
aa9b1ac3 WN |
8296 | } |
8297 | ||
a83d6e76 | 8298 | idx = (p - obj->programs) + (forward ? 1 : -1); |
0c19a9fb | 8299 | if (idx >= obj->nr_programs || idx < 0) |
aa9b1ac3 WN |
8300 | return NULL; |
8301 | return &obj->programs[idx]; | |
8302 | } | |
8303 | ||
eac7d845 | 8304 | struct bpf_program * |
a324aae3 | 8305 | bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj) |
2088a3a7 HC |
8306 | { |
8307 | return bpf_object__next_program(obj, prev); | |
8308 | } | |
8309 | ||
8310 | struct bpf_program * | |
8311 | bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev) | |
eac7d845 JK |
8312 | { |
8313 | struct bpf_program *prog = prev; | |
8314 | ||
8315 | do { | |
a83d6e76 | 8316 | prog = __bpf_program__iter(prog, obj, true); |
c3c55696 | 8317 | } while (prog && prog_is_subprog(obj, prog)); |
0c19a9fb SF |
8318 | |
8319 | return prog; | |
8320 | } | |
8321 | ||
8322 | struct bpf_program * | |
a324aae3 | 8323 | bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj) |
2088a3a7 HC |
8324 | { |
8325 | return bpf_object__prev_program(obj, next); | |
8326 | } | |
8327 | ||
8328 | struct bpf_program * | |
8329 | bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next) | |
0c19a9fb SF |
8330 | { |
8331 | struct bpf_program *prog = next; | |
8332 | ||
0c19a9fb | 8333 | do { |
a83d6e76 | 8334 | prog = __bpf_program__iter(prog, obj, false); |
c3c55696 | 8335 | } while (prog && prog_is_subprog(obj, prog)); |
eac7d845 JK |
8336 | |
8337 | return prog; | |
8338 | } | |
8339 | ||
edb13ed4 ACM |
8340 | int bpf_program__set_priv(struct bpf_program *prog, void *priv, |
8341 | bpf_program_clear_priv_t clear_priv) | |
aa9b1ac3 WN |
8342 | { |
8343 | if (prog->priv && prog->clear_priv) | |
8344 | prog->clear_priv(prog, prog->priv); | |
8345 | ||
8346 | prog->priv = priv; | |
8347 | prog->clear_priv = clear_priv; | |
8348 | return 0; | |
8349 | } | |
8350 | ||
a324aae3 | 8351 | void *bpf_program__priv(const struct bpf_program *prog) |
aa9b1ac3 | 8352 | { |
e9fc3ce9 | 8353 | return prog ? prog->priv : libbpf_err_ptr(-EINVAL); |
aa9b1ac3 WN |
8354 | } |
8355 | ||
9aba3613 JK |
8356 | void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex) |
8357 | { | |
8358 | prog->prog_ifindex = ifindex; | |
8359 | } | |
8360 | ||
01af3bf0 AN |
8361 | const char *bpf_program__name(const struct bpf_program *prog) |
8362 | { | |
8363 | return prog->name; | |
8364 | } | |
8365 | ||
52109584 AN |
8366 | const char *bpf_program__section_name(const struct bpf_program *prog) |
8367 | { | |
8368 | return prog->sec_name; | |
8369 | } | |
8370 | ||
a324aae3 | 8371 | const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy) |
aa9b1ac3 WN |
8372 | { |
8373 | const char *title; | |
8374 | ||
52109584 | 8375 | title = prog->sec_name; |
715f8db9 | 8376 | if (needs_copy) { |
aa9b1ac3 WN |
8377 | title = strdup(title); |
8378 | if (!title) { | |
be18010e | 8379 | pr_warn("failed to strdup program title\n"); |
e9fc3ce9 | 8380 | return libbpf_err_ptr(-ENOMEM); |
aa9b1ac3 WN |
8381 | } |
8382 | } | |
8383 | ||
8384 | return title; | |
8385 | } | |
8386 | ||
d9297581 AN |
8387 | bool bpf_program__autoload(const struct bpf_program *prog) |
8388 | { | |
8389 | return prog->load; | |
8390 | } | |
8391 | ||
8392 | int bpf_program__set_autoload(struct bpf_program *prog, bool autoload) | |
8393 | { | |
8394 | if (prog->obj->loaded) | |
e9fc3ce9 | 8395 | return libbpf_err(-EINVAL); |
d9297581 AN |
8396 | |
8397 | prog->load = autoload; | |
8398 | return 0; | |
8399 | } | |
8400 | ||
bcc40fc0 AN |
8401 | static int bpf_program_nth_fd(const struct bpf_program *prog, int n); |
8402 | ||
a324aae3 | 8403 | int bpf_program__fd(const struct bpf_program *prog) |
aa9b1ac3 | 8404 | { |
bcc40fc0 | 8405 | return bpf_program_nth_fd(prog, 0); |
b580563e WN |
8406 | } |
8407 | ||
1a734efe THJ |
8408 | size_t bpf_program__size(const struct bpf_program *prog) |
8409 | { | |
9c0f8cbd | 8410 | return prog->insns_cnt * BPF_INSN_SZ; |
1a734efe THJ |
8411 | } |
8412 | ||
65a7fa2e AN |
8413 | const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog) |
8414 | { | |
8415 | return prog->insns; | |
8416 | } | |
8417 | ||
8418 | size_t bpf_program__insn_cnt(const struct bpf_program *prog) | |
8419 | { | |
8420 | return prog->insns_cnt; | |
8421 | } | |
8422 | ||
b580563e WN |
8423 | int bpf_program__set_prep(struct bpf_program *prog, int nr_instances, |
8424 | bpf_program_prep_t prep) | |
8425 | { | |
8426 | int *instances_fds; | |
8427 | ||
8428 | if (nr_instances <= 0 || !prep) | |
e9fc3ce9 | 8429 | return libbpf_err(-EINVAL); |
b580563e WN |
8430 | |
8431 | if (prog->instances.nr > 0 || prog->instances.fds) { | |
be18010e | 8432 | pr_warn("Can't set pre-processor after loading\n"); |
e9fc3ce9 | 8433 | return libbpf_err(-EINVAL); |
b580563e WN |
8434 | } |
8435 | ||
8436 | instances_fds = malloc(sizeof(int) * nr_instances); | |
8437 | if (!instances_fds) { | |
be18010e | 8438 | pr_warn("alloc memory failed for fds\n"); |
e9fc3ce9 | 8439 | return libbpf_err(-ENOMEM); |
b580563e WN |
8440 | } |
8441 | ||
8442 | /* fill all fd with -1 */ | |
8443 | memset(instances_fds, -1, sizeof(int) * nr_instances); | |
8444 | ||
8445 | prog->instances.nr = nr_instances; | |
8446 | prog->instances.fds = instances_fds; | |
8447 | prog->preprocessor = prep; | |
8448 | return 0; | |
8449 | } | |
8450 | ||
bcc40fc0 AN |
8451 | __attribute__((alias("bpf_program_nth_fd"))) |
8452 | int bpf_program__nth_fd(const struct bpf_program *prog, int n); | |
8453 | ||
8454 | static int bpf_program_nth_fd(const struct bpf_program *prog, int n) | |
b580563e WN |
8455 | { |
8456 | int fd; | |
8457 | ||
1e960043 | 8458 | if (!prog) |
e9fc3ce9 | 8459 | return libbpf_err(-EINVAL); |
1e960043 | 8460 | |
b580563e | 8461 | if (n >= prog->instances.nr || n < 0) { |
be18010e | 8462 | pr_warn("Can't get the %dth fd from program %s: only %d instances\n", |
52109584 | 8463 | n, prog->name, prog->instances.nr); |
e9fc3ce9 | 8464 | return libbpf_err(-EINVAL); |
b580563e WN |
8465 | } |
8466 | ||
8467 | fd = prog->instances.fds[n]; | |
8468 | if (fd < 0) { | |
be18010e | 8469 | pr_warn("%dth instance of program '%s' is invalid\n", |
52109584 | 8470 | n, prog->name); |
e9fc3ce9 | 8471 | return libbpf_err(-ENOENT); |
b580563e WN |
8472 | } |
8473 | ||
8474 | return fd; | |
aa9b1ac3 | 8475 | } |
9d759a9b | 8476 | |
20eccf29 AN |
8477 | __alias(bpf_program__type) |
8478 | enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog); | |
8479 | ||
8480 | enum bpf_prog_type bpf_program__type(const struct bpf_program *prog) | |
f1eead9e AN |
8481 | { |
8482 | return prog->type; | |
8483 | } | |
8484 | ||
dd26b7f5 | 8485 | void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type) |
5f44e4c8 WN |
8486 | { |
8487 | prog->type = type; | |
8488 | } | |
8489 | ||
a324aae3 | 8490 | static bool bpf_program__is_type(const struct bpf_program *prog, |
5f44e4c8 WN |
8491 | enum bpf_prog_type type) |
8492 | { | |
8493 | return prog ? (prog->type == type) : false; | |
8494 | } | |
8495 | ||
a324aae3 AN |
8496 | #define BPF_PROG_TYPE_FNS(NAME, TYPE) \ |
8497 | int bpf_program__set_##NAME(struct bpf_program *prog) \ | |
8498 | { \ | |
8499 | if (!prog) \ | |
e9fc3ce9 | 8500 | return libbpf_err(-EINVAL); \ |
a324aae3 AN |
8501 | bpf_program__set_type(prog, TYPE); \ |
8502 | return 0; \ | |
8503 | } \ | |
8504 | \ | |
8505 | bool bpf_program__is_##NAME(const struct bpf_program *prog) \ | |
8506 | { \ | |
8507 | return bpf_program__is_type(prog, TYPE); \ | |
8508 | } \ | |
ed794073 | 8509 | |
7803ba73 | 8510 | BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER); |
1e092a03 | 8511 | BPF_PROG_TYPE_FNS(lsm, BPF_PROG_TYPE_LSM); |
ed794073 | 8512 | BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE); |
7803ba73 JS |
8513 | BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS); |
8514 | BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT); | |
ed794073 | 8515 | BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT); |
e14c93fd | 8516 | BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT); |
7803ba73 JS |
8517 | BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP); |
8518 | BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT); | |
12a8654b | 8519 | BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING); |
590a0088 | 8520 | BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS); |
2db6eab1 | 8521 | BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT); |
499dd29d | 8522 | BPF_PROG_TYPE_FNS(sk_lookup, BPF_PROG_TYPE_SK_LOOKUP); |
5f44e4c8 | 8523 | |
20eccf29 AN |
8524 | __alias(bpf_program__expected_attach_type) |
8525 | enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog); | |
8526 | ||
8527 | enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog) | |
f1eead9e AN |
8528 | { |
8529 | return prog->expected_attach_type; | |
8530 | } | |
8531 | ||
16962b24 JF |
8532 | void bpf_program__set_expected_attach_type(struct bpf_program *prog, |
8533 | enum bpf_attach_type type) | |
d7be143b AI |
8534 | { |
8535 | prog->expected_attach_type = type; | |
8536 | } | |
8537 | ||
a6ca7158 AN |
8538 | __u32 bpf_program__flags(const struct bpf_program *prog) |
8539 | { | |
8540 | return prog->prog_flags; | |
8541 | } | |
8542 | ||
8cccee9e | 8543 | int bpf_program__set_flags(struct bpf_program *prog, __u32 flags) |
a6ca7158 AN |
8544 | { |
8545 | if (prog->obj->loaded) | |
8546 | return libbpf_err(-EBUSY); | |
8547 | ||
8cccee9e | 8548 | prog->prog_flags = flags; |
a6ca7158 AN |
8549 | return 0; |
8550 | } | |
8551 | ||
dbdd2c7f AN |
8552 | __u32 bpf_program__log_level(const struct bpf_program *prog) |
8553 | { | |
8554 | return prog->log_level; | |
8555 | } | |
8556 | ||
8557 | int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level) | |
8558 | { | |
8559 | if (prog->obj->loaded) | |
8560 | return libbpf_err(-EBUSY); | |
8561 | ||
8562 | prog->log_level = log_level; | |
8563 | return 0; | |
8564 | } | |
8565 | ||
b3ce9079 AN |
8566 | const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size) |
8567 | { | |
8568 | *log_size = prog->log_size; | |
8569 | return prog->log_buf; | |
8570 | } | |
8571 | ||
8572 | int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size) | |
8573 | { | |
8574 | if (log_size && !log_buf) | |
8575 | return -EINVAL; | |
8576 | if (prog->log_size > UINT_MAX) | |
8577 | return -EINVAL; | |
8578 | if (prog->obj->loaded) | |
8579 | return -EBUSY; | |
8580 | ||
8581 | prog->log_buf = log_buf; | |
8582 | prog->log_size = log_size; | |
a6ca7158 AN |
8583 | return 0; |
8584 | } | |
8585 | ||
15ea31fa | 8586 | #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \ |
697f104d | 8587 | .sec = (char *)sec_pfx, \ |
d7a18ea7 | 8588 | .prog_type = BPF_PROG_TYPE_##ptype, \ |
15ea31fa AN |
8589 | .expected_attach_type = atype, \ |
8590 | .cookie = (long)(flags), \ | |
4fa5bcfe | 8591 | .prog_prepare_load_fn = libbpf_prepare_prog_load, \ |
d7a18ea7 AN |
8592 | __VA_ARGS__ \ |
8593 | } | |
8594 | ||
4fa5bcfe AN |
8595 | static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
8596 | static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
8597 | static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
8598 | static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
8599 | static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
8600 | static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
d7a18ea7 | 8601 | |
d7a18ea7 | 8602 | static const struct bpf_sec_def section_defs[] = { |
dd94d45c AN |
8603 | SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE | SEC_SLOPPY_PFX), |
8604 | SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8605 | SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
15ea31fa AN |
8606 | SEC_DEF("kprobe/", KPROBE, 0, SEC_NONE, attach_kprobe), |
8607 | SEC_DEF("uprobe/", KPROBE, 0, SEC_NONE), | |
8608 | SEC_DEF("kretprobe/", KPROBE, 0, SEC_NONE, attach_kprobe), | |
8609 | SEC_DEF("uretprobe/", KPROBE, 0, SEC_NONE), | |
15ea31fa | 8610 | SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE), |
4a4d4cee | 8611 | SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE | SEC_SLOPPY_PFX | SEC_DEPRECATED), |
dd94d45c | 8612 | SEC_DEF("action", SCHED_ACT, 0, SEC_NONE | SEC_SLOPPY_PFX), |
15ea31fa AN |
8613 | SEC_DEF("tracepoint/", TRACEPOINT, 0, SEC_NONE, attach_tp), |
8614 | SEC_DEF("tp/", TRACEPOINT, 0, SEC_NONE, attach_tp), | |
8615 | SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
8616 | SEC_DEF("raw_tp/", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
ccaf12d6 HT |
8617 | SEC_DEF("raw_tracepoint.w/", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), |
8618 | SEC_DEF("raw_tp.w/", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), | |
15ea31fa AN |
8619 | SEC_DEF("tp_btf/", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace), |
8620 | SEC_DEF("fentry/", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace), | |
8621 | SEC_DEF("fmod_ret/", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace), | |
8622 | SEC_DEF("fexit/", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace), | |
8623 | SEC_DEF("fentry.s/", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8624 | SEC_DEF("fmod_ret.s/", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8625 | SEC_DEF("fexit.s/", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8626 | SEC_DEF("freplace/", EXT, 0, SEC_ATTACH_BTF, attach_trace), | |
8627 | SEC_DEF("lsm/", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm), | |
8628 | SEC_DEF("lsm.s/", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm), | |
8629 | SEC_DEF("iter/", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter), | |
a8b77f74 | 8630 | SEC_DEF("iter.s/", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter), |
15ea31fa | 8631 | SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE), |
082c4bfb | 8632 | SEC_DEF("xdp.frags/devmap", XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS), |
4a4d4cee LB |
8633 | SEC_DEF("xdp/devmap", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE), |
8634 | SEC_DEF("xdp_devmap/", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE | SEC_DEPRECATED), | |
082c4bfb | 8635 | SEC_DEF("xdp.frags/cpumap", XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS), |
4a4d4cee LB |
8636 | SEC_DEF("xdp/cpumap", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE), |
8637 | SEC_DEF("xdp_cpumap/", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE | SEC_DEPRECATED), | |
082c4bfb | 8638 | SEC_DEF("xdp.frags", XDP, BPF_XDP, SEC_XDP_FRAGS), |
dd94d45c AN |
8639 | SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), |
8640 | SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8641 | SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8642 | SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8643 | SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8644 | SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8645 | SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8646 | SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8647 | SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8648 | SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8649 | SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8650 | SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8651 | SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8652 | SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8653 | SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8654 | SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8655 | SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8656 | SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8657 | SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8658 | SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8659 | SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8660 | SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8661 | SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8662 | SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8663 | SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8664 | SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8665 | SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8666 | SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8667 | SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8668 | SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8669 | SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8670 | SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8671 | SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8672 | SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8673 | SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8674 | SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8675 | SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8676 | SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), | |
7c80c87a | 8677 | SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX), |
583c9009 | 8678 | }; |
d7be143b | 8679 | |
697f104d AN |
8680 | static size_t custom_sec_def_cnt; |
8681 | static struct bpf_sec_def *custom_sec_defs; | |
8682 | static struct bpf_sec_def custom_fallback_def; | |
8683 | static bool has_custom_fallback_def; | |
c76e4c22 | 8684 | |
697f104d AN |
8685 | static int last_custom_sec_def_handler_id; |
8686 | ||
8687 | int libbpf_register_prog_handler(const char *sec, | |
8688 | enum bpf_prog_type prog_type, | |
8689 | enum bpf_attach_type exp_attach_type, | |
8690 | const struct libbpf_prog_handler_opts *opts) | |
d7a18ea7 | 8691 | { |
697f104d | 8692 | struct bpf_sec_def *sec_def; |
d7a18ea7 | 8693 | |
697f104d AN |
8694 | if (!OPTS_VALID(opts, libbpf_prog_handler_opts)) |
8695 | return libbpf_err(-EINVAL); | |
dd94d45c | 8696 | |
697f104d AN |
8697 | if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */ |
8698 | return libbpf_err(-E2BIG); | |
dd94d45c | 8699 | |
697f104d AN |
8700 | if (sec) { |
8701 | sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1, | |
8702 | sizeof(*sec_def)); | |
8703 | if (!sec_def) | |
8704 | return libbpf_err(-ENOMEM); | |
dd94d45c | 8705 | |
697f104d AN |
8706 | custom_sec_defs = sec_def; |
8707 | sec_def = &custom_sec_defs[custom_sec_def_cnt]; | |
8708 | } else { | |
8709 | if (has_custom_fallback_def) | |
8710 | return libbpf_err(-EBUSY); | |
dd94d45c | 8711 | |
697f104d AN |
8712 | sec_def = &custom_fallback_def; |
8713 | } | |
8714 | ||
8715 | sec_def->sec = sec ? strdup(sec) : NULL; | |
8716 | if (sec && !sec_def->sec) | |
8717 | return libbpf_err(-ENOMEM); | |
8718 | ||
8719 | sec_def->prog_type = prog_type; | |
8720 | sec_def->expected_attach_type = exp_attach_type; | |
8721 | sec_def->cookie = OPTS_GET(opts, cookie, 0); | |
8722 | ||
8723 | sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL); | |
8724 | sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL); | |
8725 | sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL); | |
8726 | ||
8727 | sec_def->handler_id = ++last_custom_sec_def_handler_id; | |
8728 | ||
8729 | if (sec) | |
8730 | custom_sec_def_cnt++; | |
8731 | else | |
8732 | has_custom_fallback_def = true; | |
8733 | ||
8734 | return sec_def->handler_id; | |
8735 | } | |
8736 | ||
8737 | int libbpf_unregister_prog_handler(int handler_id) | |
8738 | { | |
8739 | struct bpf_sec_def *sec_defs; | |
8740 | int i; | |
8741 | ||
8742 | if (handler_id <= 0) | |
8743 | return libbpf_err(-EINVAL); | |
8744 | ||
8745 | if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) { | |
8746 | memset(&custom_fallback_def, 0, sizeof(custom_fallback_def)); | |
8747 | has_custom_fallback_def = false; | |
8748 | return 0; | |
8749 | } | |
8750 | ||
8751 | for (i = 0; i < custom_sec_def_cnt; i++) { | |
8752 | if (custom_sec_defs[i].handler_id == handler_id) | |
8753 | break; | |
8754 | } | |
8755 | ||
8756 | if (i == custom_sec_def_cnt) | |
8757 | return libbpf_err(-ENOENT); | |
8758 | ||
8759 | free(custom_sec_defs[i].sec); | |
8760 | for (i = i + 1; i < custom_sec_def_cnt; i++) | |
8761 | custom_sec_defs[i - 1] = custom_sec_defs[i]; | |
8762 | custom_sec_def_cnt--; | |
8763 | ||
8764 | /* try to shrink the array, but it's ok if we couldn't */ | |
8765 | sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs)); | |
8766 | if (sec_defs) | |
8767 | custom_sec_defs = sec_defs; | |
8768 | ||
8769 | return 0; | |
8770 | } | |
8771 | ||
8772 | static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name, | |
8773 | bool allow_sloppy) | |
8774 | { | |
8775 | size_t len = strlen(sec_def->sec); | |
8776 | ||
8777 | /* "type/" always has to have proper SEC("type/extras") form */ | |
8778 | if (sec_def->sec[len - 1] == '/') { | |
8779 | if (str_has_pfx(sec_name, sec_def->sec)) | |
8780 | return true; | |
8781 | return false; | |
8782 | } | |
8783 | ||
8784 | /* "type+" means it can be either exact SEC("type") or | |
8785 | * well-formed SEC("type/extras") with proper '/' separator | |
8786 | */ | |
8787 | if (sec_def->sec[len - 1] == '+') { | |
8788 | len--; | |
8789 | /* not even a prefix */ | |
8790 | if (strncmp(sec_name, sec_def->sec, len) != 0) | |
8791 | return false; | |
8792 | /* exact match or has '/' separator */ | |
8793 | if (sec_name[len] == '\0' || sec_name[len] == '/') | |
8794 | return true; | |
8795 | return false; | |
8796 | } | |
8797 | ||
8798 | /* SEC_SLOPPY_PFX definitions are allowed to be just prefix | |
8799 | * matches, unless strict section name mode | |
8800 | * (LIBBPF_STRICT_SEC_NAME) is enabled, in which case the | |
8801 | * match has to be exact. | |
8802 | */ | |
8803 | if (allow_sloppy && str_has_pfx(sec_name, sec_def->sec)) | |
8804 | return true; | |
8805 | ||
8806 | /* Definitions not marked SEC_SLOPPY_PFX (e.g., | |
8807 | * SEC("syscall")) are exact matches in both modes. | |
8808 | */ | |
8809 | return strcmp(sec_name, sec_def->sec) == 0; | |
8810 | } | |
8811 | ||
8812 | static const struct bpf_sec_def *find_sec_def(const char *sec_name) | |
8813 | { | |
8814 | const struct bpf_sec_def *sec_def; | |
8815 | int i, n; | |
8816 | bool strict = libbpf_mode & LIBBPF_STRICT_SEC_NAME, allow_sloppy; | |
8817 | ||
8818 | n = custom_sec_def_cnt; | |
8819 | for (i = 0; i < n; i++) { | |
8820 | sec_def = &custom_sec_defs[i]; | |
8821 | if (sec_def_matches(sec_def, sec_name, false)) | |
8822 | return sec_def; | |
8823 | } | |
8824 | ||
8825 | n = ARRAY_SIZE(section_defs); | |
8826 | for (i = 0; i < n; i++) { | |
8827 | sec_def = §ion_defs[i]; | |
8828 | allow_sloppy = (sec_def->cookie & SEC_SLOPPY_PFX) && !strict; | |
8829 | if (sec_def_matches(sec_def, sec_name, allow_sloppy)) | |
dd94d45c | 8830 | return sec_def; |
d7a18ea7 | 8831 | } |
697f104d AN |
8832 | |
8833 | if (has_custom_fallback_def) | |
8834 | return &custom_fallback_def; | |
8835 | ||
d7a18ea7 AN |
8836 | return NULL; |
8837 | } | |
8838 | ||
697f104d AN |
8839 | #define MAX_TYPE_NAME_SIZE 32 |
8840 | ||
c76e4c22 TS |
8841 | static char *libbpf_get_type_names(bool attach_type) |
8842 | { | |
d7a18ea7 | 8843 | int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE; |
c76e4c22 TS |
8844 | char *buf; |
8845 | ||
8846 | buf = malloc(len); | |
8847 | if (!buf) | |
8848 | return NULL; | |
8849 | ||
8850 | buf[0] = '\0'; | |
8851 | /* Forge string buf with all available names */ | |
d7a18ea7 | 8852 | for (i = 0; i < ARRAY_SIZE(section_defs); i++) { |
15ea31fa AN |
8853 | const struct bpf_sec_def *sec_def = §ion_defs[i]; |
8854 | ||
8855 | if (attach_type) { | |
4fa5bcfe | 8856 | if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load) |
15ea31fa AN |
8857 | continue; |
8858 | ||
8859 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
8860 | continue; | |
8861 | } | |
c76e4c22 | 8862 | |
d7a18ea7 | 8863 | if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) { |
c76e4c22 TS |
8864 | free(buf); |
8865 | return NULL; | |
8866 | } | |
8867 | strcat(buf, " "); | |
d7a18ea7 | 8868 | strcat(buf, section_defs[i].sec); |
c76e4c22 TS |
8869 | } |
8870 | ||
8871 | return buf; | |
8872 | } | |
8873 | ||
b60df2a0 JK |
8874 | int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type, |
8875 | enum bpf_attach_type *expected_attach_type) | |
583c9009 | 8876 | { |
d7a18ea7 | 8877 | const struct bpf_sec_def *sec_def; |
c76e4c22 | 8878 | char *type_names; |
583c9009 | 8879 | |
b60df2a0 | 8880 | if (!name) |
e9fc3ce9 | 8881 | return libbpf_err(-EINVAL); |
583c9009 | 8882 | |
d7a18ea7 AN |
8883 | sec_def = find_sec_def(name); |
8884 | if (sec_def) { | |
8885 | *prog_type = sec_def->prog_type; | |
8886 | *expected_attach_type = sec_def->expected_attach_type; | |
b60df2a0 JK |
8887 | return 0; |
8888 | } | |
d7a18ea7 | 8889 | |
4a3d6c6a | 8890 | pr_debug("failed to guess program type from ELF section '%s'\n", name); |
c76e4c22 TS |
8891 | type_names = libbpf_get_type_names(false); |
8892 | if (type_names != NULL) { | |
3f519353 | 8893 | pr_debug("supported section(type) names are:%s\n", type_names); |
c76e4c22 TS |
8894 | free(type_names); |
8895 | } | |
8896 | ||
e9fc3ce9 | 8897 | return libbpf_err(-ESRCH); |
b60df2a0 | 8898 | } |
583c9009 | 8899 | |
590a0088 MKL |
8900 | static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj, |
8901 | size_t offset) | |
8902 | { | |
8903 | struct bpf_map *map; | |
8904 | size_t i; | |
8905 | ||
8906 | for (i = 0; i < obj->nr_maps; i++) { | |
8907 | map = &obj->maps[i]; | |
8908 | if (!bpf_map__is_struct_ops(map)) | |
8909 | continue; | |
8910 | if (map->sec_offset <= offset && | |
8911 | offset - map->sec_offset < map->def.value_size) | |
8912 | return map; | |
8913 | } | |
8914 | ||
8915 | return NULL; | |
8916 | } | |
8917 | ||
8918 | /* Collect the reloc from ELF and populate the st_ops->progs[] */ | |
646f02ff | 8919 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 8920 | Elf64_Shdr *shdr, Elf_Data *data) |
590a0088 MKL |
8921 | { |
8922 | const struct btf_member *member; | |
8923 | struct bpf_struct_ops *st_ops; | |
8924 | struct bpf_program *prog; | |
8925 | unsigned int shdr_idx; | |
8926 | const struct btf *btf; | |
8927 | struct bpf_map *map; | |
7e06aad5 | 8928 | unsigned int moff, insn_idx; |
590a0088 | 8929 | const char *name; |
1d1a3bcf | 8930 | __u32 member_idx; |
ad23b723 AN |
8931 | Elf64_Sym *sym; |
8932 | Elf64_Rel *rel; | |
590a0088 MKL |
8933 | int i, nrels; |
8934 | ||
590a0088 MKL |
8935 | btf = obj->btf; |
8936 | nrels = shdr->sh_size / shdr->sh_entsize; | |
8937 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
8938 | rel = elf_rel_by_idx(data, i); |
8939 | if (!rel) { | |
590a0088 MKL |
8940 | pr_warn("struct_ops reloc: failed to get %d reloc\n", i); |
8941 | return -LIBBPF_ERRNO__FORMAT; | |
8942 | } | |
8943 | ||
ad23b723 AN |
8944 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); |
8945 | if (!sym) { | |
590a0088 | 8946 | pr_warn("struct_ops reloc: symbol %zx not found\n", |
ad23b723 | 8947 | (size_t)ELF64_R_SYM(rel->r_info)); |
590a0088 MKL |
8948 | return -LIBBPF_ERRNO__FORMAT; |
8949 | } | |
8950 | ||
ad23b723 AN |
8951 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
8952 | map = find_struct_ops_map_by_offset(obj, rel->r_offset); | |
590a0088 | 8953 | if (!map) { |
ad23b723 AN |
8954 | pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n", |
8955 | (size_t)rel->r_offset); | |
590a0088 MKL |
8956 | return -EINVAL; |
8957 | } | |
8958 | ||
ad23b723 AN |
8959 | moff = rel->r_offset - map->sec_offset; |
8960 | shdr_idx = sym->st_shndx; | |
590a0088 | 8961 | st_ops = map->st_ops; |
ad23b723 | 8962 | 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 | 8963 | map->name, |
ad23b723 AN |
8964 | (long long)(rel->r_info >> 32), |
8965 | (long long)sym->st_value, | |
8966 | shdr_idx, (size_t)rel->r_offset, | |
8967 | map->sec_offset, sym->st_name, name); | |
590a0088 MKL |
8968 | |
8969 | if (shdr_idx >= SHN_LORESERVE) { | |
ad23b723 AN |
8970 | pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n", |
8971 | map->name, (size_t)rel->r_offset, shdr_idx); | |
590a0088 MKL |
8972 | return -LIBBPF_ERRNO__RELOC; |
8973 | } | |
ad23b723 | 8974 | if (sym->st_value % BPF_INSN_SZ) { |
7e06aad5 | 8975 | pr_warn("struct_ops reloc %s: invalid target program offset %llu\n", |
ad23b723 | 8976 | map->name, (unsigned long long)sym->st_value); |
7e06aad5 AN |
8977 | return -LIBBPF_ERRNO__FORMAT; |
8978 | } | |
ad23b723 | 8979 | insn_idx = sym->st_value / BPF_INSN_SZ; |
590a0088 MKL |
8980 | |
8981 | member = find_member_by_offset(st_ops->type, moff * 8); | |
8982 | if (!member) { | |
8983 | pr_warn("struct_ops reloc %s: cannot find member at moff %u\n", | |
8984 | map->name, moff); | |
8985 | return -EINVAL; | |
8986 | } | |
8987 | member_idx = member - btf_members(st_ops->type); | |
8988 | name = btf__name_by_offset(btf, member->name_off); | |
8989 | ||
8990 | if (!resolve_func_ptr(btf, member->type, NULL)) { | |
8991 | pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n", | |
8992 | map->name, name); | |
8993 | return -EINVAL; | |
8994 | } | |
8995 | ||
7e06aad5 | 8996 | prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx); |
590a0088 MKL |
8997 | if (!prog) { |
8998 | pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n", | |
8999 | map->name, shdr_idx, name); | |
9000 | return -EINVAL; | |
9001 | } | |
9002 | ||
91b4d1d1 AN |
9003 | /* prevent the use of BPF prog with invalid type */ |
9004 | if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) { | |
9005 | pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n", | |
9006 | map->name, prog->name); | |
9007 | return -EINVAL; | |
9008 | } | |
590a0088 | 9009 | |
91b4d1d1 AN |
9010 | /* if we haven't yet processed this BPF program, record proper |
9011 | * attach_btf_id and member_idx | |
9012 | */ | |
9013 | if (!prog->attach_btf_id) { | |
590a0088 MKL |
9014 | prog->attach_btf_id = st_ops->type_id; |
9015 | prog->expected_attach_type = member_idx; | |
590a0088 | 9016 | } |
91b4d1d1 AN |
9017 | |
9018 | /* struct_ops BPF prog can be re-used between multiple | |
9019 | * .struct_ops as long as it's the same struct_ops struct | |
9020 | * definition and the same function pointer field | |
9021 | */ | |
9022 | if (prog->attach_btf_id != st_ops->type_id || | |
9023 | prog->expected_attach_type != member_idx) { | |
9024 | 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", | |
9025 | map->name, prog->name, prog->sec_name, prog->type, | |
9026 | prog->attach_btf_id, prog->expected_attach_type, name); | |
9027 | return -EINVAL; | |
9028 | } | |
9029 | ||
590a0088 MKL |
9030 | st_ops->progs[member_idx] = prog; |
9031 | } | |
9032 | ||
9033 | return 0; | |
590a0088 MKL |
9034 | } |
9035 | ||
a6ed02ca | 9036 | #define BTF_TRACE_PREFIX "btf_trace_" |
1e092a03 | 9037 | #define BTF_LSM_PREFIX "bpf_lsm_" |
21aef70e | 9038 | #define BTF_ITER_PREFIX "bpf_iter_" |
a6ed02ca KS |
9039 | #define BTF_MAX_NAME_SIZE 128 |
9040 | ||
67234743 AS |
9041 | void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type, |
9042 | const char **prefix, int *kind) | |
9043 | { | |
9044 | switch (attach_type) { | |
9045 | case BPF_TRACE_RAW_TP: | |
9046 | *prefix = BTF_TRACE_PREFIX; | |
9047 | *kind = BTF_KIND_TYPEDEF; | |
9048 | break; | |
9049 | case BPF_LSM_MAC: | |
9050 | *prefix = BTF_LSM_PREFIX; | |
9051 | *kind = BTF_KIND_FUNC; | |
9052 | break; | |
9053 | case BPF_TRACE_ITER: | |
9054 | *prefix = BTF_ITER_PREFIX; | |
9055 | *kind = BTF_KIND_FUNC; | |
9056 | break; | |
9057 | default: | |
9058 | *prefix = ""; | |
9059 | *kind = BTF_KIND_FUNC; | |
9060 | } | |
9061 | } | |
9062 | ||
a6ed02ca KS |
9063 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
9064 | const char *name, __u32 kind) | |
9065 | { | |
9066 | char btf_type_name[BTF_MAX_NAME_SIZE]; | |
9067 | int ret; | |
9068 | ||
9069 | ret = snprintf(btf_type_name, sizeof(btf_type_name), | |
9070 | "%s%s", prefix, name); | |
9071 | /* snprintf returns the number of characters written excluding the | |
c139e40a | 9072 | * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it |
a6ed02ca KS |
9073 | * indicates truncation. |
9074 | */ | |
9075 | if (ret < 0 || ret >= sizeof(btf_type_name)) | |
9076 | return -ENAMETOOLONG; | |
9077 | return btf__find_by_name_kind(btf, btf_type_name, kind); | |
9078 | } | |
9079 | ||
91abb4a6 AN |
9080 | static inline int find_attach_btf_id(struct btf *btf, const char *name, |
9081 | enum bpf_attach_type attach_type) | |
a6ed02ca | 9082 | { |
67234743 AS |
9083 | const char *prefix; |
9084 | int kind; | |
a6ed02ca | 9085 | |
67234743 AS |
9086 | btf_get_kernel_prefix_kind(attach_type, &prefix, &kind); |
9087 | return find_btf_by_prefix_kind(btf, prefix, name, kind); | |
a6ed02ca KS |
9088 | } |
9089 | ||
b8c54ea4 AS |
9090 | int libbpf_find_vmlinux_btf_id(const char *name, |
9091 | enum bpf_attach_type attach_type) | |
12a8654b | 9092 | { |
a6ed02ca | 9093 | struct btf *btf; |
3521ffa2 | 9094 | int err; |
12a8654b | 9095 | |
a710eed3 | 9096 | btf = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
9097 | err = libbpf_get_error(btf); |
9098 | if (err) { | |
12a8654b | 9099 | pr_warn("vmlinux BTF is not found\n"); |
e9fc3ce9 | 9100 | return libbpf_err(err); |
12a8654b AS |
9101 | } |
9102 | ||
91abb4a6 AN |
9103 | err = find_attach_btf_id(btf, name, attach_type); |
9104 | if (err <= 0) | |
9105 | pr_warn("%s is not found in vmlinux BTF\n", name); | |
9106 | ||
3521ffa2 | 9107 | btf__free(btf); |
e9fc3ce9 | 9108 | return libbpf_err(err); |
b8c54ea4 AS |
9109 | } |
9110 | ||
e7bf94db AS |
9111 | static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd) |
9112 | { | |
ebc7b50a DM |
9113 | struct bpf_prog_info info = {}; |
9114 | __u32 info_len = sizeof(info); | |
6cc93e2f | 9115 | struct btf *btf; |
6d2d73cd | 9116 | int err; |
e7bf94db | 9117 | |
ebc7b50a | 9118 | err = bpf_obj_get_info_by_fd(attach_prog_fd, &info, &info_len); |
e9fc3ce9 | 9119 | if (err) { |
ebc7b50a DM |
9120 | pr_warn("failed bpf_obj_get_info_by_fd for FD %d: %d\n", |
9121 | attach_prog_fd, err); | |
e9fc3ce9 | 9122 | return err; |
e7bf94db | 9123 | } |
6d2d73cd QM |
9124 | |
9125 | err = -EINVAL; | |
ebc7b50a | 9126 | if (!info.btf_id) { |
e7bf94db AS |
9127 | pr_warn("The target program doesn't have BTF\n"); |
9128 | goto out; | |
9129 | } | |
ebc7b50a DM |
9130 | btf = btf__load_from_kernel_by_id(info.btf_id); |
9131 | err = libbpf_get_error(btf); | |
9132 | if (err) { | |
9133 | pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err); | |
e7bf94db AS |
9134 | goto out; |
9135 | } | |
9136 | err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC); | |
9137 | btf__free(btf); | |
9138 | if (err <= 0) { | |
9139 | pr_warn("%s is not found in prog's BTF\n", name); | |
9140 | goto out; | |
9141 | } | |
9142 | out: | |
e7bf94db AS |
9143 | return err; |
9144 | } | |
9145 | ||
91abb4a6 AN |
9146 | static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name, |
9147 | enum bpf_attach_type attach_type, | |
9148 | int *btf_obj_fd, int *btf_type_id) | |
9149 | { | |
9150 | int ret, i; | |
9151 | ||
9152 | ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type); | |
9153 | if (ret > 0) { | |
9154 | *btf_obj_fd = 0; /* vmlinux BTF */ | |
9155 | *btf_type_id = ret; | |
9156 | return 0; | |
9157 | } | |
9158 | if (ret != -ENOENT) | |
9159 | return ret; | |
9160 | ||
9161 | ret = load_module_btfs(obj); | |
9162 | if (ret) | |
9163 | return ret; | |
9164 | ||
9165 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
9166 | const struct module_btf *mod = &obj->btf_modules[i]; | |
9167 | ||
9168 | ret = find_attach_btf_id(mod->btf, attach_name, attach_type); | |
9169 | if (ret > 0) { | |
9170 | *btf_obj_fd = mod->fd; | |
9171 | *btf_type_id = ret; | |
9172 | return 0; | |
9173 | } | |
9174 | if (ret == -ENOENT) | |
9175 | continue; | |
9176 | ||
9177 | return ret; | |
9178 | } | |
9179 | ||
9180 | return -ESRCH; | |
9181 | } | |
9182 | ||
15ea31fa AN |
9183 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
9184 | int *btf_obj_fd, int *btf_type_id) | |
b8c54ea4 | 9185 | { |
a6ed02ca KS |
9186 | enum bpf_attach_type attach_type = prog->expected_attach_type; |
9187 | __u32 attach_prog_fd = prog->attach_prog_fd; | |
b6291a6f | 9188 | int err = 0; |
b8c54ea4 | 9189 | |
91abb4a6 AN |
9190 | /* BPF program's BTF ID */ |
9191 | if (attach_prog_fd) { | |
9192 | err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd); | |
9193 | if (err < 0) { | |
9194 | pr_warn("failed to find BPF program (FD %d) BTF ID for '%s': %d\n", | |
9195 | attach_prog_fd, attach_name, err); | |
9196 | return err; | |
9197 | } | |
9198 | *btf_obj_fd = 0; | |
9199 | *btf_type_id = err; | |
9200 | return 0; | |
9201 | } | |
9202 | ||
9203 | /* kernel/module BTF ID */ | |
67234743 AS |
9204 | if (prog->obj->gen_loader) { |
9205 | bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type); | |
9206 | *btf_obj_fd = 0; | |
9207 | *btf_type_id = 1; | |
9208 | } else { | |
9209 | err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id); | |
9210 | } | |
91abb4a6 AN |
9211 | if (err) { |
9212 | pr_warn("failed to find kernel BTF type ID of '%s': %d\n", attach_name, err); | |
b8c54ea4 | 9213 | return err; |
12a8654b | 9214 | } |
91abb4a6 | 9215 | return 0; |
12a8654b AS |
9216 | } |
9217 | ||
956b620f AI |
9218 | int libbpf_attach_type_by_name(const char *name, |
9219 | enum bpf_attach_type *attach_type) | |
9220 | { | |
c76e4c22 | 9221 | char *type_names; |
b6291a6f | 9222 | const struct bpf_sec_def *sec_def; |
956b620f AI |
9223 | |
9224 | if (!name) | |
e9fc3ce9 | 9225 | return libbpf_err(-EINVAL); |
956b620f | 9226 | |
b6291a6f AN |
9227 | sec_def = find_sec_def(name); |
9228 | if (!sec_def) { | |
9229 | pr_debug("failed to guess attach type based on ELF section name '%s'\n", name); | |
9230 | type_names = libbpf_get_type_names(true); | |
9231 | if (type_names != NULL) { | |
9232 | pr_debug("attachable section(type) names are:%s\n", type_names); | |
9233 | free(type_names); | |
9234 | } | |
9235 | ||
9236 | return libbpf_err(-EINVAL); | |
c76e4c22 TS |
9237 | } |
9238 | ||
4fa5bcfe | 9239 | if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load) |
15ea31fa AN |
9240 | return libbpf_err(-EINVAL); |
9241 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
b6291a6f AN |
9242 | return libbpf_err(-EINVAL); |
9243 | ||
9244 | *attach_type = sec_def->expected_attach_type; | |
9245 | return 0; | |
956b620f AI |
9246 | } |
9247 | ||
a324aae3 | 9248 | int bpf_map__fd(const struct bpf_map *map) |
9d759a9b | 9249 | { |
e9fc3ce9 | 9250 | return map ? map->fd : libbpf_err(-EINVAL); |
9d759a9b WN |
9251 | } |
9252 | ||
a324aae3 | 9253 | const struct bpf_map_def *bpf_map__def(const struct bpf_map *map) |
9d759a9b | 9254 | { |
e9fc3ce9 | 9255 | return map ? &map->def : libbpf_err_ptr(-EINVAL); |
9d759a9b WN |
9256 | } |
9257 | ||
aed65917 AN |
9258 | static bool map_uses_real_name(const struct bpf_map *map) |
9259 | { | |
9260 | /* Since libbpf started to support custom .data.* and .rodata.* maps, | |
9261 | * their user-visible name differs from kernel-visible name. Users see | |
9262 | * such map's corresponding ELF section name as a map name. | |
9263 | * This check distinguishes .data/.rodata from .data.* and .rodata.* | |
9264 | * maps to know which name has to be returned to the user. | |
9265 | */ | |
9266 | if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0) | |
9267 | return true; | |
9268 | if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0) | |
9269 | return true; | |
9270 | return false; | |
9271 | } | |
9272 | ||
a324aae3 | 9273 | const char *bpf_map__name(const struct bpf_map *map) |
561bbcca | 9274 | { |
aed65917 AN |
9275 | if (!map) |
9276 | return NULL; | |
9277 | ||
9278 | if (map_uses_real_name(map)) | |
9279 | return map->real_name; | |
9280 | ||
9281 | return map->name; | |
561bbcca WN |
9282 | } |
9283 | ||
1bdb6c9a AN |
9284 | enum bpf_map_type bpf_map__type(const struct bpf_map *map) |
9285 | { | |
9286 | return map->def.type; | |
9287 | } | |
9288 | ||
9289 | int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type) | |
9290 | { | |
9291 | if (map->fd >= 0) | |
e9fc3ce9 | 9292 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9293 | map->def.type = type; |
9294 | return 0; | |
9295 | } | |
9296 | ||
9297 | __u32 bpf_map__map_flags(const struct bpf_map *map) | |
9298 | { | |
9299 | return map->def.map_flags; | |
9300 | } | |
9301 | ||
9302 | int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags) | |
9303 | { | |
9304 | if (map->fd >= 0) | |
e9fc3ce9 | 9305 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9306 | map->def.map_flags = flags; |
9307 | return 0; | |
9308 | } | |
9309 | ||
47512102 JK |
9310 | __u64 bpf_map__map_extra(const struct bpf_map *map) |
9311 | { | |
9312 | return map->map_extra; | |
9313 | } | |
9314 | ||
9315 | int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra) | |
9316 | { | |
9317 | if (map->fd >= 0) | |
9318 | return libbpf_err(-EBUSY); | |
9319 | map->map_extra = map_extra; | |
9320 | return 0; | |
9321 | } | |
9322 | ||
1bdb6c9a AN |
9323 | __u32 bpf_map__numa_node(const struct bpf_map *map) |
9324 | { | |
9325 | return map->numa_node; | |
9326 | } | |
9327 | ||
9328 | int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node) | |
9329 | { | |
9330 | if (map->fd >= 0) | |
e9fc3ce9 | 9331 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9332 | map->numa_node = numa_node; |
9333 | return 0; | |
9334 | } | |
9335 | ||
9336 | __u32 bpf_map__key_size(const struct bpf_map *map) | |
9337 | { | |
9338 | return map->def.key_size; | |
9339 | } | |
9340 | ||
9341 | int bpf_map__set_key_size(struct bpf_map *map, __u32 size) | |
9342 | { | |
9343 | if (map->fd >= 0) | |
e9fc3ce9 | 9344 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9345 | map->def.key_size = size; |
9346 | return 0; | |
9347 | } | |
9348 | ||
9349 | __u32 bpf_map__value_size(const struct bpf_map *map) | |
9350 | { | |
9351 | return map->def.value_size; | |
9352 | } | |
9353 | ||
9354 | int bpf_map__set_value_size(struct bpf_map *map, __u32 size) | |
9355 | { | |
9356 | if (map->fd >= 0) | |
e9fc3ce9 | 9357 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9358 | map->def.value_size = size; |
9359 | return 0; | |
9360 | } | |
9361 | ||
5b891af7 | 9362 | __u32 bpf_map__btf_key_type_id(const struct bpf_map *map) |
8a138aed | 9363 | { |
61746dbe | 9364 | return map ? map->btf_key_type_id : 0; |
8a138aed MKL |
9365 | } |
9366 | ||
5b891af7 | 9367 | __u32 bpf_map__btf_value_type_id(const struct bpf_map *map) |
8a138aed | 9368 | { |
61746dbe | 9369 | return map ? map->btf_value_type_id : 0; |
8a138aed MKL |
9370 | } |
9371 | ||
edb13ed4 ACM |
9372 | int bpf_map__set_priv(struct bpf_map *map, void *priv, |
9373 | bpf_map_clear_priv_t clear_priv) | |
9d759a9b WN |
9374 | { |
9375 | if (!map) | |
e9fc3ce9 | 9376 | return libbpf_err(-EINVAL); |
9d759a9b WN |
9377 | |
9378 | if (map->priv) { | |
9379 | if (map->clear_priv) | |
9380 | map->clear_priv(map, map->priv); | |
9381 | } | |
9382 | ||
9383 | map->priv = priv; | |
9384 | map->clear_priv = clear_priv; | |
9385 | return 0; | |
9386 | } | |
9387 | ||
a324aae3 | 9388 | void *bpf_map__priv(const struct bpf_map *map) |
9d759a9b | 9389 | { |
e9fc3ce9 | 9390 | return map ? map->priv : libbpf_err_ptr(-EINVAL); |
9d759a9b WN |
9391 | } |
9392 | ||
e2842be5 THJ |
9393 | int bpf_map__set_initial_value(struct bpf_map *map, |
9394 | const void *data, size_t size) | |
9395 | { | |
9396 | if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG || | |
9397 | size != map->def.value_size || map->fd >= 0) | |
e9fc3ce9 | 9398 | return libbpf_err(-EINVAL); |
e2842be5 THJ |
9399 | |
9400 | memcpy(map->mmaped, data, size); | |
9401 | return 0; | |
9402 | } | |
9403 | ||
7723256b AS |
9404 | const void *bpf_map__initial_value(struct bpf_map *map, size_t *psize) |
9405 | { | |
9406 | if (!map->mmaped) | |
9407 | return NULL; | |
9408 | *psize = map->def.value_size; | |
9409 | return map->mmaped; | |
9410 | } | |
9411 | ||
a324aae3 | 9412 | bool bpf_map__is_offload_neutral(const struct bpf_map *map) |
f83fb22c JK |
9413 | { |
9414 | return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; | |
9415 | } | |
9416 | ||
a324aae3 | 9417 | bool bpf_map__is_internal(const struct bpf_map *map) |
d859900c DB |
9418 | { |
9419 | return map->libbpf_type != LIBBPF_MAP_UNSPEC; | |
9420 | } | |
9421 | ||
1bdb6c9a AN |
9422 | __u32 bpf_map__ifindex(const struct bpf_map *map) |
9423 | { | |
9424 | return map->map_ifindex; | |
9425 | } | |
9426 | ||
9427 | int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex) | |
9aba3613 | 9428 | { |
1bdb6c9a | 9429 | if (map->fd >= 0) |
e9fc3ce9 | 9430 | return libbpf_err(-EBUSY); |
9aba3613 | 9431 | map->map_ifindex = ifindex; |
1bdb6c9a | 9432 | return 0; |
9aba3613 JK |
9433 | } |
9434 | ||
addb9fc9 NS |
9435 | int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd) |
9436 | { | |
9437 | if (!bpf_map_type__is_map_in_map(map->def.type)) { | |
be18010e | 9438 | pr_warn("error: unsupported map type\n"); |
e9fc3ce9 | 9439 | return libbpf_err(-EINVAL); |
addb9fc9 NS |
9440 | } |
9441 | if (map->inner_map_fd != -1) { | |
be18010e | 9442 | pr_warn("error: inner_map_fd already specified\n"); |
e9fc3ce9 | 9443 | return libbpf_err(-EINVAL); |
addb9fc9 | 9444 | } |
8f7b239e AN |
9445 | if (map->inner_map) { |
9446 | bpf_map__destroy(map->inner_map); | |
9447 | zfree(&map->inner_map); | |
9448 | } | |
addb9fc9 NS |
9449 | map->inner_map_fd = fd; |
9450 | return 0; | |
9451 | } | |
9452 | ||
0c19a9fb | 9453 | static struct bpf_map * |
a324aae3 | 9454 | __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i) |
9d759a9b | 9455 | { |
0c19a9fb | 9456 | ssize_t idx; |
9d759a9b WN |
9457 | struct bpf_map *s, *e; |
9458 | ||
9459 | if (!obj || !obj->maps) | |
e9fc3ce9 | 9460 | return errno = EINVAL, NULL; |
9d759a9b WN |
9461 | |
9462 | s = obj->maps; | |
9463 | e = obj->maps + obj->nr_maps; | |
9464 | ||
0c19a9fb | 9465 | if ((m < s) || (m >= e)) { |
be18010e KW |
9466 | pr_warn("error in %s: map handler doesn't belong to object\n", |
9467 | __func__); | |
e9fc3ce9 | 9468 | return errno = EINVAL, NULL; |
9d759a9b WN |
9469 | } |
9470 | ||
0c19a9fb SF |
9471 | idx = (m - obj->maps) + i; |
9472 | if (idx >= obj->nr_maps || idx < 0) | |
9d759a9b WN |
9473 | return NULL; |
9474 | return &obj->maps[idx]; | |
9475 | } | |
561bbcca | 9476 | |
0c19a9fb | 9477 | struct bpf_map * |
a324aae3 | 9478 | bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj) |
2088a3a7 HC |
9479 | { |
9480 | return bpf_object__next_map(obj, prev); | |
9481 | } | |
9482 | ||
9483 | struct bpf_map * | |
9484 | bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev) | |
0c19a9fb SF |
9485 | { |
9486 | if (prev == NULL) | |
9487 | return obj->maps; | |
9488 | ||
9489 | return __bpf_map__iter(prev, obj, 1); | |
9490 | } | |
9491 | ||
9492 | struct bpf_map * | |
a324aae3 | 9493 | bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj) |
2088a3a7 HC |
9494 | { |
9495 | return bpf_object__prev_map(obj, next); | |
9496 | } | |
9497 | ||
9498 | struct bpf_map * | |
9499 | bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next) | |
0c19a9fb SF |
9500 | { |
9501 | if (next == NULL) { | |
9502 | if (!obj->nr_maps) | |
9503 | return NULL; | |
9504 | return obj->maps + obj->nr_maps - 1; | |
9505 | } | |
9506 | ||
9507 | return __bpf_map__iter(next, obj, -1); | |
9508 | } | |
9509 | ||
561bbcca | 9510 | struct bpf_map * |
a324aae3 | 9511 | bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name) |
561bbcca WN |
9512 | { |
9513 | struct bpf_map *pos; | |
9514 | ||
f74a53d9 | 9515 | bpf_object__for_each_map(pos, obj) { |
26071635 AN |
9516 | /* if it's a special internal map name (which always starts |
9517 | * with dot) then check if that special name matches the | |
9518 | * real map name (ELF section name) | |
9519 | */ | |
9520 | if (name[0] == '.') { | |
9521 | if (pos->real_name && strcmp(pos->real_name, name) == 0) | |
9522 | return pos; | |
9523 | continue; | |
9524 | } | |
9525 | /* otherwise map name has to be an exact match */ | |
aed65917 AN |
9526 | if (map_uses_real_name(pos)) { |
9527 | if (strcmp(pos->real_name, name) == 0) | |
9528 | return pos; | |
9529 | continue; | |
9530 | } | |
9531 | if (strcmp(pos->name, name) == 0) | |
561bbcca WN |
9532 | return pos; |
9533 | } | |
e9fc3ce9 | 9534 | return errno = ENOENT, NULL; |
561bbcca | 9535 | } |
5a6acad1 | 9536 | |
f3cea32d | 9537 | int |
a324aae3 | 9538 | bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name) |
f3cea32d MF |
9539 | { |
9540 | return bpf_map__fd(bpf_object__find_map_by_name(obj, name)); | |
9541 | } | |
9542 | ||
5a6acad1 WN |
9543 | struct bpf_map * |
9544 | bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset) | |
9545 | { | |
e9fc3ce9 | 9546 | return libbpf_err_ptr(-ENOTSUP); |
5a6acad1 | 9547 | } |
e28ff1a8 JS |
9548 | |
9549 | long libbpf_get_error(const void *ptr) | |
9550 | { | |
e9fc3ce9 AN |
9551 | if (!IS_ERR_OR_NULL(ptr)) |
9552 | return 0; | |
9553 | ||
9554 | if (IS_ERR(ptr)) | |
9555 | errno = -PTR_ERR(ptr); | |
9556 | ||
9557 | /* If ptr == NULL, then errno should be already set by the failing | |
9558 | * API, because libbpf never returns NULL on success and it now always | |
9559 | * sets errno on error. So no extra errno handling for ptr == NULL | |
9560 | * case. | |
9561 | */ | |
9562 | return -errno; | |
e28ff1a8 | 9563 | } |
6f6d33f3 | 9564 | |
e32660ac | 9565 | __attribute__((alias("bpf_prog_load_xattr2"))) |
d7be143b | 9566 | int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr, |
e32660ac AN |
9567 | struct bpf_object **pobj, int *prog_fd); |
9568 | ||
9569 | static int bpf_prog_load_xattr2(const struct bpf_prog_load_attr *attr, | |
9570 | struct bpf_object **pobj, int *prog_fd) | |
6f6d33f3 | 9571 | { |
33bae185 | 9572 | struct bpf_object_open_attr open_attr = {}; |
48cca7e4 | 9573 | struct bpf_program *prog, *first_prog = NULL; |
6f6d33f3 | 9574 | struct bpf_object *obj; |
f0307a7e | 9575 | struct bpf_map *map; |
6f6d33f3 JF |
9576 | int err; |
9577 | ||
d7be143b | 9578 | if (!attr) |
e9fc3ce9 | 9579 | return libbpf_err(-EINVAL); |
17387dd5 | 9580 | if (!attr->file) |
e9fc3ce9 | 9581 | return libbpf_err(-EINVAL); |
d7be143b | 9582 | |
33bae185 LY |
9583 | open_attr.file = attr->file; |
9584 | open_attr.prog_type = attr->prog_type; | |
9585 | ||
fc763870 | 9586 | obj = __bpf_object__open_xattr(&open_attr, 0); |
e9fc3ce9 AN |
9587 | err = libbpf_get_error(obj); |
9588 | if (err) | |
9589 | return libbpf_err(-ENOENT); | |
6f6d33f3 | 9590 | |
48cca7e4 | 9591 | bpf_object__for_each_program(prog, obj) { |
dd4436bb | 9592 | enum bpf_attach_type attach_type = attr->expected_attach_type; |
48cca7e4 | 9593 | /* |
dd4436bb AN |
9594 | * to preserve backwards compatibility, bpf_prog_load treats |
9595 | * attr->prog_type, if specified, as an override to whatever | |
9596 | * bpf_object__open guessed | |
48cca7e4 | 9597 | */ |
dd4436bb AN |
9598 | if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) { |
9599 | bpf_program__set_type(prog, attr->prog_type); | |
9600 | bpf_program__set_expected_attach_type(prog, | |
9601 | attach_type); | |
9602 | } | |
20eccf29 | 9603 | if (bpf_program__type(prog) == BPF_PROG_TYPE_UNSPEC) { |
dd4436bb AN |
9604 | /* |
9605 | * we haven't guessed from section name and user | |
9606 | * didn't provide a fallback type, too bad... | |
9607 | */ | |
9608 | bpf_object__close(obj); | |
e9fc3ce9 | 9609 | return libbpf_err(-EINVAL); |
583c9009 | 9610 | } |
48cca7e4 | 9611 | |
dd4436bb | 9612 | prog->prog_ifindex = attr->ifindex; |
da11b417 | 9613 | prog->log_level = attr->log_level; |
2b288740 | 9614 | prog->prog_flags |= attr->prog_flags; |
69495d2a | 9615 | if (!first_prog) |
48cca7e4 AS |
9616 | first_prog = prog; |
9617 | } | |
9618 | ||
f74a53d9 | 9619 | bpf_object__for_each_map(map, obj) { |
a5dd9589 | 9620 | if (map->def.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
f83fb22c | 9621 | map->map_ifindex = attr->ifindex; |
f0307a7e DB |
9622 | } |
9623 | ||
48cca7e4 | 9624 | if (!first_prog) { |
be18010e | 9625 | pr_warn("object file doesn't contain bpf program\n"); |
48cca7e4 | 9626 | bpf_object__close(obj); |
e9fc3ce9 | 9627 | return libbpf_err(-ENOENT); |
583c9009 RG |
9628 | } |
9629 | ||
6f6d33f3 JF |
9630 | err = bpf_object__load(obj); |
9631 | if (err) { | |
9632 | bpf_object__close(obj); | |
e9fc3ce9 | 9633 | return libbpf_err(err); |
6f6d33f3 JF |
9634 | } |
9635 | ||
9636 | *pobj = obj; | |
48cca7e4 | 9637 | *prog_fd = bpf_program__fd(first_prog); |
6f6d33f3 JF |
9638 | return 0; |
9639 | } | |
d0cabbb0 | 9640 | |
e32660ac AN |
9641 | COMPAT_VERSION(bpf_prog_load_deprecated, bpf_prog_load, LIBBPF_0.0.1) |
9642 | int bpf_prog_load_deprecated(const char *file, enum bpf_prog_type type, | |
9643 | struct bpf_object **pobj, int *prog_fd) | |
9644 | { | |
9645 | struct bpf_prog_load_attr attr; | |
9646 | ||
9647 | memset(&attr, 0, sizeof(struct bpf_prog_load_attr)); | |
9648 | attr.file = file; | |
9649 | attr.prog_type = type; | |
9650 | attr.expected_attach_type = 0; | |
9651 | ||
9652 | return bpf_prog_load_xattr2(&attr, pobj, prog_fd); | |
9653 | } | |
9654 | ||
1c2e9efc | 9655 | struct bpf_link { |
d6958706 | 9656 | int (*detach)(struct bpf_link *link); |
d88b71d4 | 9657 | void (*dealloc)(struct bpf_link *link); |
c016b68e AN |
9658 | char *pin_path; /* NULL, if not pinned */ |
9659 | int fd; /* hook FD, -1 if not applicable */ | |
d6958706 | 9660 | bool disconnected; |
1c2e9efc AN |
9661 | }; |
9662 | ||
cc4f864b AN |
9663 | /* Replace link's underlying BPF program with the new one */ |
9664 | int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog) | |
9665 | { | |
e9fc3ce9 | 9666 | int ret; |
c139e40a | 9667 | |
e9fc3ce9 AN |
9668 | ret = bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL); |
9669 | return libbpf_err_errno(ret); | |
cc4f864b AN |
9670 | } |
9671 | ||
d6958706 AN |
9672 | /* Release "ownership" of underlying BPF resource (typically, BPF program |
9673 | * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected | |
9674 | * link, when destructed through bpf_link__destroy() call won't attempt to | |
9675 | * detach/unregisted that BPF resource. This is useful in situations where, | |
9676 | * say, attached BPF program has to outlive userspace program that attached it | |
9677 | * in the system. Depending on type of BPF program, though, there might be | |
9678 | * additional steps (like pinning BPF program in BPF FS) necessary to ensure | |
9679 | * exit of userspace program doesn't trigger automatic detachment and clean up | |
9680 | * inside the kernel. | |
9681 | */ | |
9682 | void bpf_link__disconnect(struct bpf_link *link) | |
9683 | { | |
9684 | link->disconnected = true; | |
9685 | } | |
9686 | ||
1c2e9efc AN |
9687 | int bpf_link__destroy(struct bpf_link *link) |
9688 | { | |
d6958706 | 9689 | int err = 0; |
1c2e9efc | 9690 | |
50450fc7 | 9691 | if (IS_ERR_OR_NULL(link)) |
1c2e9efc AN |
9692 | return 0; |
9693 | ||
d6958706 AN |
9694 | if (!link->disconnected && link->detach) |
9695 | err = link->detach(link); | |
c016b68e AN |
9696 | if (link->pin_path) |
9697 | free(link->pin_path); | |
d88b71d4 AN |
9698 | if (link->dealloc) |
9699 | link->dealloc(link); | |
9700 | else | |
9701 | free(link); | |
1c2e9efc | 9702 | |
e9fc3ce9 | 9703 | return libbpf_err(err); |
1c2e9efc AN |
9704 | } |
9705 | ||
c016b68e AN |
9706 | int bpf_link__fd(const struct bpf_link *link) |
9707 | { | |
9708 | return link->fd; | |
9709 | } | |
9710 | ||
9711 | const char *bpf_link__pin_path(const struct bpf_link *link) | |
9712 | { | |
9713 | return link->pin_path; | |
9714 | } | |
9715 | ||
9716 | static int bpf_link__detach_fd(struct bpf_link *link) | |
9717 | { | |
e9fc3ce9 | 9718 | return libbpf_err_errno(close(link->fd)); |
c016b68e AN |
9719 | } |
9720 | ||
9721 | struct bpf_link *bpf_link__open(const char *path) | |
9722 | { | |
9723 | struct bpf_link *link; | |
9724 | int fd; | |
9725 | ||
9726 | fd = bpf_obj_get(path); | |
9727 | if (fd < 0) { | |
9728 | fd = -errno; | |
9729 | pr_warn("failed to open link at %s: %d\n", path, fd); | |
e9fc3ce9 | 9730 | return libbpf_err_ptr(fd); |
c016b68e AN |
9731 | } |
9732 | ||
9733 | link = calloc(1, sizeof(*link)); | |
9734 | if (!link) { | |
9735 | close(fd); | |
e9fc3ce9 | 9736 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
9737 | } |
9738 | link->detach = &bpf_link__detach_fd; | |
9739 | link->fd = fd; | |
9740 | ||
9741 | link->pin_path = strdup(path); | |
9742 | if (!link->pin_path) { | |
9743 | bpf_link__destroy(link); | |
e9fc3ce9 | 9744 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
9745 | } |
9746 | ||
9747 | return link; | |
9748 | } | |
9749 | ||
2e49527e AN |
9750 | int bpf_link__detach(struct bpf_link *link) |
9751 | { | |
9752 | return bpf_link_detach(link->fd) ? -errno : 0; | |
9753 | } | |
9754 | ||
c016b68e AN |
9755 | int bpf_link__pin(struct bpf_link *link, const char *path) |
9756 | { | |
9757 | int err; | |
9758 | ||
9759 | if (link->pin_path) | |
e9fc3ce9 | 9760 | return libbpf_err(-EBUSY); |
c016b68e AN |
9761 | err = make_parent_dir(path); |
9762 | if (err) | |
e9fc3ce9 | 9763 | return libbpf_err(err); |
c016b68e AN |
9764 | err = check_path(path); |
9765 | if (err) | |
e9fc3ce9 | 9766 | return libbpf_err(err); |
c016b68e AN |
9767 | |
9768 | link->pin_path = strdup(path); | |
9769 | if (!link->pin_path) | |
e9fc3ce9 | 9770 | return libbpf_err(-ENOMEM); |
c016b68e AN |
9771 | |
9772 | if (bpf_obj_pin(link->fd, link->pin_path)) { | |
9773 | err = -errno; | |
9774 | zfree(&link->pin_path); | |
e9fc3ce9 | 9775 | return libbpf_err(err); |
c016b68e AN |
9776 | } |
9777 | ||
9778 | pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path); | |
9779 | return 0; | |
9780 | } | |
9781 | ||
9782 | int bpf_link__unpin(struct bpf_link *link) | |
9783 | { | |
9784 | int err; | |
9785 | ||
9786 | if (!link->pin_path) | |
e9fc3ce9 | 9787 | return libbpf_err(-EINVAL); |
c016b68e AN |
9788 | |
9789 | err = unlink(link->pin_path); | |
9790 | if (err != 0) | |
af0efa05 | 9791 | return -errno; |
c016b68e AN |
9792 | |
9793 | pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path); | |
9794 | zfree(&link->pin_path); | |
9795 | return 0; | |
9796 | } | |
63f2f5ee | 9797 | |
668ace0e AN |
9798 | struct bpf_link_perf { |
9799 | struct bpf_link link; | |
9800 | int perf_event_fd; | |
ca304b40 RDT |
9801 | /* legacy kprobe support: keep track of probe identifier and type */ |
9802 | char *legacy_probe_name; | |
46ed5fc3 | 9803 | bool legacy_is_kprobe; |
ca304b40 | 9804 | bool legacy_is_retprobe; |
668ace0e AN |
9805 | }; |
9806 | ||
46ed5fc3 | 9807 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe); |
cc10623c | 9808 | static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe); |
46ed5fc3 | 9809 | |
668ace0e | 9810 | static int bpf_link_perf_detach(struct bpf_link *link) |
63f2f5ee | 9811 | { |
668ace0e AN |
9812 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); |
9813 | int err = 0; | |
63f2f5ee | 9814 | |
668ace0e | 9815 | if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0) |
63f2f5ee AN |
9816 | err = -errno; |
9817 | ||
668ace0e AN |
9818 | if (perf_link->perf_event_fd != link->fd) |
9819 | close(perf_link->perf_event_fd); | |
c016b68e | 9820 | close(link->fd); |
668ace0e | 9821 | |
cc10623c | 9822 | /* legacy uprobe/kprobe needs to be removed after perf event fd closure */ |
46ed5fc3 AN |
9823 | if (perf_link->legacy_probe_name) { |
9824 | if (perf_link->legacy_is_kprobe) { | |
9825 | err = remove_kprobe_event_legacy(perf_link->legacy_probe_name, | |
9826 | perf_link->legacy_is_retprobe); | |
cc10623c AN |
9827 | } else { |
9828 | err = remove_uprobe_event_legacy(perf_link->legacy_probe_name, | |
9829 | perf_link->legacy_is_retprobe); | |
46ed5fc3 AN |
9830 | } |
9831 | } | |
ca304b40 RDT |
9832 | |
9833 | return err; | |
63f2f5ee AN |
9834 | } |
9835 | ||
668ace0e AN |
9836 | static void bpf_link_perf_dealloc(struct bpf_link *link) |
9837 | { | |
9838 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
9839 | ||
ca304b40 | 9840 | free(perf_link->legacy_probe_name); |
668ace0e AN |
9841 | free(perf_link); |
9842 | } | |
9843 | ||
942025c9 | 9844 | struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd, |
47faff37 | 9845 | const struct bpf_perf_event_opts *opts) |
63f2f5ee AN |
9846 | { |
9847 | char errmsg[STRERR_BUFSIZE]; | |
668ace0e AN |
9848 | struct bpf_link_perf *link; |
9849 | int prog_fd, link_fd = -1, err; | |
63f2f5ee | 9850 | |
47faff37 AN |
9851 | if (!OPTS_VALID(opts, bpf_perf_event_opts)) |
9852 | return libbpf_err_ptr(-EINVAL); | |
9853 | ||
63f2f5ee | 9854 | if (pfd < 0) { |
52109584 AN |
9855 | pr_warn("prog '%s': invalid perf event FD %d\n", |
9856 | prog->name, pfd); | |
e9fc3ce9 | 9857 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
9858 | } |
9859 | prog_fd = bpf_program__fd(prog); | |
9860 | if (prog_fd < 0) { | |
52109584 AN |
9861 | pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n", |
9862 | prog->name); | |
e9fc3ce9 | 9863 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
9864 | } |
9865 | ||
d6958706 | 9866 | link = calloc(1, sizeof(*link)); |
63f2f5ee | 9867 | if (!link) |
e9fc3ce9 | 9868 | return libbpf_err_ptr(-ENOMEM); |
668ace0e AN |
9869 | link->link.detach = &bpf_link_perf_detach; |
9870 | link->link.dealloc = &bpf_link_perf_dealloc; | |
9871 | link->perf_event_fd = pfd; | |
63f2f5ee | 9872 | |
668ace0e | 9873 | if (kernel_supports(prog->obj, FEAT_PERF_LINK)) { |
47faff37 AN |
9874 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts, |
9875 | .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0)); | |
9876 | ||
9877 | link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts); | |
668ace0e AN |
9878 | if (link_fd < 0) { |
9879 | err = -errno; | |
9880 | pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n", | |
9881 | prog->name, pfd, | |
9882 | err, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
9883 | goto err_out; | |
9884 | } | |
9885 | link->link.fd = link_fd; | |
9886 | } else { | |
47faff37 AN |
9887 | if (OPTS_GET(opts, bpf_cookie, 0)) { |
9888 | pr_warn("prog '%s': user context value is not supported\n", prog->name); | |
9889 | err = -EOPNOTSUPP; | |
9890 | goto err_out; | |
9891 | } | |
9892 | ||
668ace0e AN |
9893 | if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) { |
9894 | err = -errno; | |
9895 | pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n", | |
9896 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
9897 | if (err == -EPROTO) | |
9898 | pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n", | |
9899 | prog->name, pfd); | |
9900 | goto err_out; | |
9901 | } | |
9902 | link->link.fd = pfd; | |
63f2f5ee AN |
9903 | } |
9904 | if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
9905 | err = -errno; | |
668ace0e | 9906 | pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n", |
52109584 | 9907 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
668ace0e | 9908 | goto err_out; |
63f2f5ee | 9909 | } |
668ace0e AN |
9910 | |
9911 | return &link->link; | |
9912 | err_out: | |
9913 | if (link_fd >= 0) | |
9914 | close(link_fd); | |
9915 | free(link); | |
9916 | return libbpf_err_ptr(err); | |
63f2f5ee AN |
9917 | } |
9918 | ||
942025c9 | 9919 | struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd) |
47faff37 AN |
9920 | { |
9921 | return bpf_program__attach_perf_event_opts(prog, pfd, NULL); | |
9922 | } | |
9923 | ||
b2650027 AN |
9924 | /* |
9925 | * this function is expected to parse integer in the range of [0, 2^31-1] from | |
9926 | * given file using scanf format string fmt. If actual parsed value is | |
9927 | * negative, the result might be indistinguishable from error | |
9928 | */ | |
9929 | static int parse_uint_from_file(const char *file, const char *fmt) | |
9930 | { | |
9931 | char buf[STRERR_BUFSIZE]; | |
9932 | int err, ret; | |
9933 | FILE *f; | |
9934 | ||
9935 | f = fopen(file, "r"); | |
9936 | if (!f) { | |
9937 | err = -errno; | |
9938 | pr_debug("failed to open '%s': %s\n", file, | |
9939 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
9940 | return err; | |
9941 | } | |
9942 | err = fscanf(f, fmt, &ret); | |
9943 | if (err != 1) { | |
9944 | err = err == EOF ? -EIO : -errno; | |
9945 | pr_debug("failed to parse '%s': %s\n", file, | |
9946 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
9947 | fclose(f); | |
9948 | return err; | |
9949 | } | |
9950 | fclose(f); | |
9951 | return ret; | |
9952 | } | |
9953 | ||
9954 | static int determine_kprobe_perf_type(void) | |
9955 | { | |
9956 | const char *file = "/sys/bus/event_source/devices/kprobe/type"; | |
9957 | ||
9958 | return parse_uint_from_file(file, "%d\n"); | |
9959 | } | |
9960 | ||
9961 | static int determine_uprobe_perf_type(void) | |
9962 | { | |
9963 | const char *file = "/sys/bus/event_source/devices/uprobe/type"; | |
9964 | ||
9965 | return parse_uint_from_file(file, "%d\n"); | |
9966 | } | |
9967 | ||
9968 | static int determine_kprobe_retprobe_bit(void) | |
9969 | { | |
9970 | const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe"; | |
9971 | ||
9972 | return parse_uint_from_file(file, "config:%d\n"); | |
9973 | } | |
9974 | ||
9975 | static int determine_uprobe_retprobe_bit(void) | |
9976 | { | |
9977 | const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe"; | |
9978 | ||
9979 | return parse_uint_from_file(file, "config:%d\n"); | |
9980 | } | |
9981 | ||
5e3b8356 AN |
9982 | #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32 |
9983 | #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32 | |
9984 | ||
b2650027 | 9985 | static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name, |
5e3b8356 | 9986 | uint64_t offset, int pid, size_t ref_ctr_off) |
b2650027 AN |
9987 | { |
9988 | struct perf_event_attr attr = {}; | |
9989 | char errmsg[STRERR_BUFSIZE]; | |
9990 | int type, pfd, err; | |
9991 | ||
5e3b8356 AN |
9992 | if (ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS)) |
9993 | return -EINVAL; | |
9994 | ||
b2650027 AN |
9995 | type = uprobe ? determine_uprobe_perf_type() |
9996 | : determine_kprobe_perf_type(); | |
9997 | if (type < 0) { | |
be18010e KW |
9998 | pr_warn("failed to determine %s perf type: %s\n", |
9999 | uprobe ? "uprobe" : "kprobe", | |
10000 | libbpf_strerror_r(type, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10001 | return type; |
10002 | } | |
10003 | if (retprobe) { | |
10004 | int bit = uprobe ? determine_uprobe_retprobe_bit() | |
10005 | : determine_kprobe_retprobe_bit(); | |
10006 | ||
10007 | if (bit < 0) { | |
be18010e KW |
10008 | pr_warn("failed to determine %s retprobe bit: %s\n", |
10009 | uprobe ? "uprobe" : "kprobe", | |
10010 | libbpf_strerror_r(bit, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10011 | return bit; |
10012 | } | |
10013 | attr.config |= 1 << bit; | |
10014 | } | |
10015 | attr.size = sizeof(attr); | |
10016 | attr.type = type; | |
5e3b8356 | 10017 | attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT; |
36db2a94 AN |
10018 | attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */ |
10019 | attr.config2 = offset; /* kprobe_addr or probe_offset */ | |
b2650027 AN |
10020 | |
10021 | /* pid filter is meaningful only for uprobes */ | |
10022 | pfd = syscall(__NR_perf_event_open, &attr, | |
10023 | pid < 0 ? -1 : pid /* pid */, | |
10024 | pid == -1 ? 0 : -1 /* cpu */, | |
10025 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
10026 | if (pfd < 0) { | |
10027 | err = -errno; | |
be18010e KW |
10028 | pr_warn("%s perf_event_open() failed: %s\n", |
10029 | uprobe ? "uprobe" : "kprobe", | |
10030 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10031 | return err; |
10032 | } | |
10033 | return pfd; | |
10034 | } | |
10035 | ||
46ed5fc3 AN |
10036 | static int append_to_file(const char *file, const char *fmt, ...) |
10037 | { | |
10038 | int fd, n, err = 0; | |
10039 | va_list ap; | |
10040 | ||
92274e24 | 10041 | fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0); |
46ed5fc3 AN |
10042 | if (fd < 0) |
10043 | return -errno; | |
10044 | ||
10045 | va_start(ap, fmt); | |
10046 | n = vdprintf(fd, fmt, ap); | |
10047 | va_end(ap); | |
10048 | ||
10049 | if (n < 0) | |
10050 | err = -errno; | |
10051 | ||
10052 | close(fd); | |
10053 | return err; | |
10054 | } | |
10055 | ||
10056 | static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz, | |
10057 | const char *kfunc_name, size_t offset) | |
10058 | { | |
51a33c60 QW |
10059 | static int index = 0; |
10060 | ||
10061 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset, | |
10062 | __sync_fetch_and_add(&index, 1)); | |
46ed5fc3 AN |
10063 | } |
10064 | ||
10065 | static int add_kprobe_event_legacy(const char *probe_name, bool retprobe, | |
10066 | const char *kfunc_name, size_t offset) | |
10067 | { | |
10068 | const char *file = "/sys/kernel/debug/tracing/kprobe_events"; | |
10069 | ||
10070 | return append_to_file(file, "%c:%s/%s %s+0x%zx", | |
10071 | retprobe ? 'r' : 'p', | |
10072 | retprobe ? "kretprobes" : "kprobes", | |
10073 | probe_name, kfunc_name, offset); | |
10074 | } | |
10075 | ||
10076 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe) | |
10077 | { | |
10078 | const char *file = "/sys/kernel/debug/tracing/kprobe_events"; | |
10079 | ||
10080 | return append_to_file(file, "-:%s/%s", retprobe ? "kretprobes" : "kprobes", probe_name); | |
10081 | } | |
10082 | ||
10083 | static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
10084 | { | |
10085 | char file[256]; | |
10086 | ||
10087 | snprintf(file, sizeof(file), | |
10088 | "/sys/kernel/debug/tracing/events/%s/%s/id", | |
10089 | retprobe ? "kretprobes" : "kprobes", probe_name); | |
10090 | ||
10091 | return parse_uint_from_file(file, "%d\n"); | |
10092 | } | |
10093 | ||
10094 | static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe, | |
10095 | const char *kfunc_name, size_t offset, int pid) | |
ca304b40 RDT |
10096 | { |
10097 | struct perf_event_attr attr = {}; | |
10098 | char errmsg[STRERR_BUFSIZE]; | |
10099 | int type, pfd, err; | |
10100 | ||
46ed5fc3 | 10101 | err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset); |
ca304b40 | 10102 | if (err < 0) { |
46ed5fc3 AN |
10103 | pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n", |
10104 | kfunc_name, offset, | |
ca304b40 RDT |
10105 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10106 | return err; | |
10107 | } | |
46ed5fc3 | 10108 | type = determine_kprobe_perf_type_legacy(probe_name, retprobe); |
ca304b40 | 10109 | if (type < 0) { |
46ed5fc3 AN |
10110 | pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n", |
10111 | kfunc_name, offset, | |
ca304b40 RDT |
10112 | libbpf_strerror_r(type, errmsg, sizeof(errmsg))); |
10113 | return type; | |
10114 | } | |
10115 | attr.size = sizeof(attr); | |
10116 | attr.config = type; | |
10117 | attr.type = PERF_TYPE_TRACEPOINT; | |
10118 | ||
10119 | pfd = syscall(__NR_perf_event_open, &attr, | |
10120 | pid < 0 ? -1 : pid, /* pid */ | |
10121 | pid == -1 ? 0 : -1, /* cpu */ | |
10122 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
10123 | if (pfd < 0) { | |
10124 | err = -errno; | |
10125 | pr_warn("legacy kprobe perf_event_open() failed: %s\n", | |
10126 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10127 | return err; | |
10128 | } | |
10129 | return pfd; | |
10130 | } | |
10131 | ||
da97553e | 10132 | struct bpf_link * |
942025c9 | 10133 | bpf_program__attach_kprobe_opts(const struct bpf_program *prog, |
ac0ed488 | 10134 | const char *func_name, |
47faff37 | 10135 | const struct bpf_kprobe_opts *opts) |
b2650027 | 10136 | { |
47faff37 | 10137 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
b2650027 | 10138 | char errmsg[STRERR_BUFSIZE]; |
ca304b40 | 10139 | char *legacy_probe = NULL; |
b2650027 | 10140 | struct bpf_link *link; |
46ed5fc3 | 10141 | size_t offset; |
ca304b40 | 10142 | bool retprobe, legacy; |
b2650027 AN |
10143 | int pfd, err; |
10144 | ||
da97553e JO |
10145 | if (!OPTS_VALID(opts, bpf_kprobe_opts)) |
10146 | return libbpf_err_ptr(-EINVAL); | |
10147 | ||
10148 | retprobe = OPTS_GET(opts, retprobe, false); | |
10149 | offset = OPTS_GET(opts, offset, 0); | |
47faff37 | 10150 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
da97553e | 10151 | |
ca304b40 RDT |
10152 | legacy = determine_kprobe_perf_type() < 0; |
10153 | if (!legacy) { | |
10154 | pfd = perf_event_open_probe(false /* uprobe */, retprobe, | |
10155 | func_name, offset, | |
10156 | -1 /* pid */, 0 /* ref_ctr_off */); | |
10157 | } else { | |
46ed5fc3 AN |
10158 | char probe_name[256]; |
10159 | ||
10160 | gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
10161 | func_name, offset); | |
10162 | ||
71cff670 | 10163 | legacy_probe = strdup(probe_name); |
ca304b40 RDT |
10164 | if (!legacy_probe) |
10165 | return libbpf_err_ptr(-ENOMEM); | |
10166 | ||
46ed5fc3 | 10167 | pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name, |
ca304b40 RDT |
10168 | offset, -1 /* pid */); |
10169 | } | |
b2650027 | 10170 | if (pfd < 0) { |
46ed5fc3 AN |
10171 | err = -errno; |
10172 | pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n", | |
10173 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
10174 | func_name, offset, | |
303a2572 AN |
10175 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10176 | goto err_out; | |
b2650027 | 10177 | } |
47faff37 | 10178 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
10179 | err = libbpf_get_error(link); |
10180 | if (err) { | |
b2650027 | 10181 | close(pfd); |
46ed5fc3 AN |
10182 | pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n", |
10183 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
10184 | func_name, offset, | |
be18010e | 10185 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
303a2572 | 10186 | goto err_out; |
b2650027 | 10187 | } |
ca304b40 RDT |
10188 | if (legacy) { |
10189 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
10190 | ||
10191 | perf_link->legacy_probe_name = legacy_probe; | |
46ed5fc3 | 10192 | perf_link->legacy_is_kprobe = true; |
ca304b40 RDT |
10193 | perf_link->legacy_is_retprobe = retprobe; |
10194 | } | |
10195 | ||
b2650027 | 10196 | return link; |
303a2572 AN |
10197 | err_out: |
10198 | free(legacy_probe); | |
10199 | return libbpf_err_ptr(err); | |
b2650027 AN |
10200 | } |
10201 | ||
942025c9 | 10202 | struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog, |
ac0ed488 JO |
10203 | bool retprobe, |
10204 | const char *func_name) | |
10205 | { | |
da97553e | 10206 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts, |
ac0ed488 | 10207 | .retprobe = retprobe, |
da97553e | 10208 | ); |
ac0ed488 JO |
10209 | |
10210 | return bpf_program__attach_kprobe_opts(prog, func_name, &opts); | |
10211 | } | |
10212 | ||
4fa5bcfe | 10213 | static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 10214 | { |
da97553e | 10215 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts); |
a2488b5f | 10216 | unsigned long offset = 0; |
d7a18ea7 | 10217 | const char *func_name; |
a2488b5f | 10218 | char *func; |
4fa5bcfe | 10219 | int n; |
d7a18ea7 | 10220 | |
13d35a0c AN |
10221 | opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/"); |
10222 | if (opts.retprobe) | |
10223 | func_name = prog->sec_name + sizeof("kretprobe/") - 1; | |
10224 | else | |
10225 | func_name = prog->sec_name + sizeof("kprobe/") - 1; | |
d7a18ea7 | 10226 | |
e3f9bc35 | 10227 | n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset); |
a2488b5f | 10228 | if (n < 1) { |
a2488b5f | 10229 | pr_warn("kprobe name is invalid: %s\n", func_name); |
4fa5bcfe | 10230 | return -EINVAL; |
a2488b5f AM |
10231 | } |
10232 | if (opts.retprobe && offset != 0) { | |
1f71a468 | 10233 | free(func); |
a2488b5f | 10234 | pr_warn("kretprobes do not support offset specification\n"); |
4fa5bcfe | 10235 | return -EINVAL; |
a2488b5f | 10236 | } |
d7a18ea7 | 10237 | |
a2488b5f | 10238 | opts.offset = offset; |
4fa5bcfe | 10239 | *link = bpf_program__attach_kprobe_opts(prog, func, &opts); |
a2488b5f | 10240 | free(func); |
4fa5bcfe | 10241 | return libbpf_get_error(*link); |
d7a18ea7 AN |
10242 | } |
10243 | ||
cc10623c AN |
10244 | static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz, |
10245 | const char *binary_path, uint64_t offset) | |
10246 | { | |
10247 | int i; | |
10248 | ||
10249 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset); | |
10250 | ||
10251 | /* sanitize binary_path in the probe name */ | |
10252 | for (i = 0; buf[i]; i++) { | |
10253 | if (!isalnum(buf[i])) | |
10254 | buf[i] = '_'; | |
10255 | } | |
10256 | } | |
10257 | ||
10258 | static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe, | |
10259 | const char *binary_path, size_t offset) | |
10260 | { | |
10261 | const char *file = "/sys/kernel/debug/tracing/uprobe_events"; | |
10262 | ||
10263 | return append_to_file(file, "%c:%s/%s %s:0x%zx", | |
10264 | retprobe ? 'r' : 'p', | |
10265 | retprobe ? "uretprobes" : "uprobes", | |
10266 | probe_name, binary_path, offset); | |
10267 | } | |
10268 | ||
10269 | static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe) | |
10270 | { | |
10271 | const char *file = "/sys/kernel/debug/tracing/uprobe_events"; | |
10272 | ||
10273 | return append_to_file(file, "-:%s/%s", retprobe ? "uretprobes" : "uprobes", probe_name); | |
10274 | } | |
10275 | ||
10276 | static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
10277 | { | |
10278 | char file[512]; | |
10279 | ||
10280 | snprintf(file, sizeof(file), | |
10281 | "/sys/kernel/debug/tracing/events/%s/%s/id", | |
10282 | retprobe ? "uretprobes" : "uprobes", probe_name); | |
10283 | ||
10284 | return parse_uint_from_file(file, "%d\n"); | |
10285 | } | |
10286 | ||
10287 | static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe, | |
10288 | const char *binary_path, size_t offset, int pid) | |
10289 | { | |
10290 | struct perf_event_attr attr; | |
10291 | int type, pfd, err; | |
10292 | ||
10293 | err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset); | |
10294 | if (err < 0) { | |
10295 | pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n", | |
10296 | binary_path, (size_t)offset, err); | |
10297 | return err; | |
10298 | } | |
10299 | type = determine_uprobe_perf_type_legacy(probe_name, retprobe); | |
10300 | if (type < 0) { | |
10301 | pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n", | |
10302 | binary_path, offset, err); | |
10303 | return type; | |
10304 | } | |
10305 | ||
10306 | memset(&attr, 0, sizeof(attr)); | |
10307 | attr.size = sizeof(attr); | |
10308 | attr.config = type; | |
10309 | attr.type = PERF_TYPE_TRACEPOINT; | |
10310 | ||
10311 | pfd = syscall(__NR_perf_event_open, &attr, | |
10312 | pid < 0 ? -1 : pid, /* pid */ | |
10313 | pid == -1 ? 0 : -1, /* cpu */ | |
10314 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
10315 | if (pfd < 0) { | |
10316 | err = -errno; | |
10317 | pr_warn("legacy uprobe perf_event_open() failed: %d\n", err); | |
10318 | return err; | |
10319 | } | |
10320 | return pfd; | |
10321 | } | |
10322 | ||
47faff37 | 10323 | LIBBPF_API struct bpf_link * |
942025c9 | 10324 | bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid, |
47faff37 AN |
10325 | const char *binary_path, size_t func_offset, |
10326 | const struct bpf_uprobe_opts *opts) | |
b2650027 | 10327 | { |
47faff37 | 10328 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
cc10623c | 10329 | char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL; |
b2650027 | 10330 | struct bpf_link *link; |
5e3b8356 | 10331 | size_t ref_ctr_off; |
b2650027 | 10332 | int pfd, err; |
cc10623c | 10333 | bool retprobe, legacy; |
47faff37 AN |
10334 | |
10335 | if (!OPTS_VALID(opts, bpf_uprobe_opts)) | |
10336 | return libbpf_err_ptr(-EINVAL); | |
10337 | ||
10338 | retprobe = OPTS_GET(opts, retprobe, false); | |
5e3b8356 | 10339 | ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0); |
47faff37 | 10340 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
b2650027 | 10341 | |
cc10623c AN |
10342 | legacy = determine_uprobe_perf_type() < 0; |
10343 | if (!legacy) { | |
10344 | pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path, | |
10345 | func_offset, pid, ref_ctr_off); | |
10346 | } else { | |
10347 | char probe_name[512]; | |
10348 | ||
10349 | if (ref_ctr_off) | |
10350 | return libbpf_err_ptr(-EINVAL); | |
10351 | ||
10352 | gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
10353 | binary_path, func_offset); | |
10354 | ||
10355 | legacy_probe = strdup(probe_name); | |
10356 | if (!legacy_probe) | |
10357 | return libbpf_err_ptr(-ENOMEM); | |
10358 | ||
10359 | pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe, | |
10360 | binary_path, func_offset, pid); | |
10361 | } | |
b2650027 | 10362 | if (pfd < 0) { |
cc10623c | 10363 | err = -errno; |
52109584 AN |
10364 | pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n", |
10365 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e | 10366 | binary_path, func_offset, |
cc10623c AN |
10367 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10368 | goto err_out; | |
b2650027 | 10369 | } |
cc10623c | 10370 | |
47faff37 | 10371 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
10372 | err = libbpf_get_error(link); |
10373 | if (err) { | |
b2650027 | 10374 | close(pfd); |
52109584 AN |
10375 | pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n", |
10376 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e KW |
10377 | binary_path, func_offset, |
10378 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
cc10623c AN |
10379 | goto err_out; |
10380 | } | |
10381 | if (legacy) { | |
10382 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
10383 | ||
10384 | perf_link->legacy_probe_name = legacy_probe; | |
10385 | perf_link->legacy_is_kprobe = false; | |
10386 | perf_link->legacy_is_retprobe = retprobe; | |
b2650027 AN |
10387 | } |
10388 | return link; | |
cc10623c AN |
10389 | err_out: |
10390 | free(legacy_probe); | |
10391 | return libbpf_err_ptr(err); | |
10392 | ||
b2650027 AN |
10393 | } |
10394 | ||
942025c9 | 10395 | struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog, |
47faff37 AN |
10396 | bool retprobe, pid_t pid, |
10397 | const char *binary_path, | |
10398 | size_t func_offset) | |
10399 | { | |
10400 | DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe); | |
10401 | ||
10402 | return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts); | |
10403 | } | |
10404 | ||
f6de59c1 AN |
10405 | static int determine_tracepoint_id(const char *tp_category, |
10406 | const char *tp_name) | |
10407 | { | |
10408 | char file[PATH_MAX]; | |
10409 | int ret; | |
10410 | ||
10411 | ret = snprintf(file, sizeof(file), | |
10412 | "/sys/kernel/debug/tracing/events/%s/%s/id", | |
10413 | tp_category, tp_name); | |
10414 | if (ret < 0) | |
10415 | return -errno; | |
10416 | if (ret >= sizeof(file)) { | |
10417 | pr_debug("tracepoint %s/%s path is too long\n", | |
10418 | tp_category, tp_name); | |
10419 | return -E2BIG; | |
10420 | } | |
10421 | return parse_uint_from_file(file, "%d\n"); | |
10422 | } | |
10423 | ||
10424 | static int perf_event_open_tracepoint(const char *tp_category, | |
10425 | const char *tp_name) | |
10426 | { | |
10427 | struct perf_event_attr attr = {}; | |
10428 | char errmsg[STRERR_BUFSIZE]; | |
10429 | int tp_id, pfd, err; | |
10430 | ||
10431 | tp_id = determine_tracepoint_id(tp_category, tp_name); | |
10432 | if (tp_id < 0) { | |
be18010e KW |
10433 | pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n", |
10434 | tp_category, tp_name, | |
10435 | libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
10436 | return tp_id; |
10437 | } | |
10438 | ||
10439 | attr.type = PERF_TYPE_TRACEPOINT; | |
10440 | attr.size = sizeof(attr); | |
10441 | attr.config = tp_id; | |
10442 | ||
10443 | pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */, | |
10444 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
10445 | if (pfd < 0) { | |
10446 | err = -errno; | |
be18010e KW |
10447 | pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n", |
10448 | tp_category, tp_name, | |
10449 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
10450 | return err; |
10451 | } | |
10452 | return pfd; | |
10453 | } | |
10454 | ||
942025c9 | 10455 | struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog, |
47faff37 AN |
10456 | const char *tp_category, |
10457 | const char *tp_name, | |
10458 | const struct bpf_tracepoint_opts *opts) | |
f6de59c1 | 10459 | { |
47faff37 | 10460 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f6de59c1 AN |
10461 | char errmsg[STRERR_BUFSIZE]; |
10462 | struct bpf_link *link; | |
10463 | int pfd, err; | |
10464 | ||
47faff37 AN |
10465 | if (!OPTS_VALID(opts, bpf_tracepoint_opts)) |
10466 | return libbpf_err_ptr(-EINVAL); | |
10467 | ||
10468 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); | |
10469 | ||
f6de59c1 AN |
10470 | pfd = perf_event_open_tracepoint(tp_category, tp_name); |
10471 | if (pfd < 0) { | |
52109584 AN |
10472 | pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n", |
10473 | prog->name, tp_category, tp_name, | |
be18010e | 10474 | libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 10475 | return libbpf_err_ptr(pfd); |
f6de59c1 | 10476 | } |
47faff37 | 10477 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
10478 | err = libbpf_get_error(link); |
10479 | if (err) { | |
f6de59c1 | 10480 | close(pfd); |
52109584 AN |
10481 | pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n", |
10482 | prog->name, tp_category, tp_name, | |
be18010e | 10483 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 10484 | return libbpf_err_ptr(err); |
f6de59c1 AN |
10485 | } |
10486 | return link; | |
10487 | } | |
10488 | ||
942025c9 | 10489 | struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog, |
47faff37 AN |
10490 | const char *tp_category, |
10491 | const char *tp_name) | |
10492 | { | |
10493 | return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL); | |
10494 | } | |
10495 | ||
4fa5bcfe | 10496 | static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 AN |
10497 | { |
10498 | char *sec_name, *tp_cat, *tp_name; | |
d7a18ea7 | 10499 | |
52109584 | 10500 | sec_name = strdup(prog->sec_name); |
d7a18ea7 | 10501 | if (!sec_name) |
4fa5bcfe | 10502 | return -ENOMEM; |
d7a18ea7 | 10503 | |
13d35a0c AN |
10504 | /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */ |
10505 | if (str_has_pfx(prog->sec_name, "tp/")) | |
10506 | tp_cat = sec_name + sizeof("tp/") - 1; | |
10507 | else | |
10508 | tp_cat = sec_name + sizeof("tracepoint/") - 1; | |
d7a18ea7 AN |
10509 | tp_name = strchr(tp_cat, '/'); |
10510 | if (!tp_name) { | |
e9fc3ce9 | 10511 | free(sec_name); |
4fa5bcfe | 10512 | return -EINVAL; |
d7a18ea7 AN |
10513 | } |
10514 | *tp_name = '\0'; | |
10515 | tp_name++; | |
10516 | ||
4fa5bcfe | 10517 | *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name); |
d7a18ea7 | 10518 | free(sec_name); |
4fa5bcfe | 10519 | return libbpf_get_error(*link); |
d7a18ea7 AN |
10520 | } |
10521 | ||
942025c9 | 10522 | struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog, |
84bf5e1f AN |
10523 | const char *tp_name) |
10524 | { | |
10525 | char errmsg[STRERR_BUFSIZE]; | |
c016b68e | 10526 | struct bpf_link *link; |
84bf5e1f AN |
10527 | int prog_fd, pfd; |
10528 | ||
10529 | prog_fd = bpf_program__fd(prog); | |
10530 | if (prog_fd < 0) { | |
52109584 | 10531 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 10532 | return libbpf_err_ptr(-EINVAL); |
84bf5e1f AN |
10533 | } |
10534 | ||
d6958706 | 10535 | link = calloc(1, sizeof(*link)); |
84bf5e1f | 10536 | if (!link) |
e9fc3ce9 | 10537 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 10538 | link->detach = &bpf_link__detach_fd; |
84bf5e1f AN |
10539 | |
10540 | pfd = bpf_raw_tracepoint_open(tp_name, prog_fd); | |
10541 | if (pfd < 0) { | |
10542 | pfd = -errno; | |
10543 | free(link); | |
52109584 AN |
10544 | pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n", |
10545 | prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 10546 | return libbpf_err_ptr(pfd); |
84bf5e1f AN |
10547 | } |
10548 | link->fd = pfd; | |
c016b68e | 10549 | return link; |
84bf5e1f AN |
10550 | } |
10551 | ||
4fa5bcfe | 10552 | static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 10553 | { |
ccaf12d6 HT |
10554 | static const char *const prefixes[] = { |
10555 | "raw_tp/", | |
10556 | "raw_tracepoint/", | |
10557 | "raw_tp.w/", | |
10558 | "raw_tracepoint.w/", | |
10559 | }; | |
10560 | size_t i; | |
10561 | const char *tp_name = NULL; | |
13d35a0c | 10562 | |
ccaf12d6 HT |
10563 | for (i = 0; i < ARRAY_SIZE(prefixes); i++) { |
10564 | if (str_has_pfx(prog->sec_name, prefixes[i])) { | |
10565 | tp_name = prog->sec_name + strlen(prefixes[i]); | |
10566 | break; | |
10567 | } | |
10568 | } | |
10569 | if (!tp_name) { | |
10570 | pr_warn("prog '%s': invalid section name '%s'\n", | |
10571 | prog->name, prog->sec_name); | |
4fa5bcfe | 10572 | return -EINVAL; |
ccaf12d6 | 10573 | } |
d7a18ea7 | 10574 | |
4fa5bcfe AN |
10575 | *link = bpf_program__attach_raw_tracepoint(prog, tp_name); |
10576 | return libbpf_get_error(link); | |
d7a18ea7 AN |
10577 | } |
10578 | ||
1e092a03 | 10579 | /* Common logic for all BPF program types that attach to a btf_id */ |
942025c9 | 10580 | static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog) |
b8c54ea4 AS |
10581 | { |
10582 | char errmsg[STRERR_BUFSIZE]; | |
c016b68e | 10583 | struct bpf_link *link; |
b8c54ea4 AS |
10584 | int prog_fd, pfd; |
10585 | ||
10586 | prog_fd = bpf_program__fd(prog); | |
10587 | if (prog_fd < 0) { | |
52109584 | 10588 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 10589 | return libbpf_err_ptr(-EINVAL); |
b8c54ea4 AS |
10590 | } |
10591 | ||
d6958706 | 10592 | link = calloc(1, sizeof(*link)); |
b8c54ea4 | 10593 | if (!link) |
e9fc3ce9 | 10594 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 10595 | link->detach = &bpf_link__detach_fd; |
b8c54ea4 AS |
10596 | |
10597 | pfd = bpf_raw_tracepoint_open(NULL, prog_fd); | |
10598 | if (pfd < 0) { | |
10599 | pfd = -errno; | |
10600 | free(link); | |
52109584 AN |
10601 | pr_warn("prog '%s': failed to attach: %s\n", |
10602 | prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 10603 | return libbpf_err_ptr(pfd); |
b8c54ea4 AS |
10604 | } |
10605 | link->fd = pfd; | |
10606 | return (struct bpf_link *)link; | |
10607 | } | |
10608 | ||
942025c9 | 10609 | struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog) |
1e092a03 KS |
10610 | { |
10611 | return bpf_program__attach_btf_id(prog); | |
10612 | } | |
10613 | ||
942025c9 | 10614 | struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog) |
1e092a03 KS |
10615 | { |
10616 | return bpf_program__attach_btf_id(prog); | |
10617 | } | |
10618 | ||
4fa5bcfe | 10619 | static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 10620 | { |
4fa5bcfe AN |
10621 | *link = bpf_program__attach_trace(prog); |
10622 | return libbpf_get_error(*link); | |
d7a18ea7 AN |
10623 | } |
10624 | ||
4fa5bcfe | 10625 | static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
1e092a03 | 10626 | { |
4fa5bcfe AN |
10627 | *link = bpf_program__attach_lsm(prog); |
10628 | return libbpf_get_error(*link); | |
1e092a03 KS |
10629 | } |
10630 | ||
d60d81ac | 10631 | static struct bpf_link * |
942025c9 | 10632 | bpf_program__attach_fd(const struct bpf_program *prog, int target_fd, int btf_id, |
d60d81ac | 10633 | const char *target_name) |
cc4f864b | 10634 | { |
a5359091 THJ |
10635 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts, |
10636 | .target_btf_id = btf_id); | |
cc4f864b AN |
10637 | enum bpf_attach_type attach_type; |
10638 | char errmsg[STRERR_BUFSIZE]; | |
10639 | struct bpf_link *link; | |
10640 | int prog_fd, link_fd; | |
10641 | ||
10642 | prog_fd = bpf_program__fd(prog); | |
10643 | if (prog_fd < 0) { | |
52109584 | 10644 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 10645 | return libbpf_err_ptr(-EINVAL); |
cc4f864b AN |
10646 | } |
10647 | ||
10648 | link = calloc(1, sizeof(*link)); | |
10649 | if (!link) | |
e9fc3ce9 | 10650 | return libbpf_err_ptr(-ENOMEM); |
cc4f864b AN |
10651 | link->detach = &bpf_link__detach_fd; |
10652 | ||
20eccf29 | 10653 | attach_type = bpf_program__expected_attach_type(prog); |
a5359091 | 10654 | link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts); |
cc4f864b AN |
10655 | if (link_fd < 0) { |
10656 | link_fd = -errno; | |
10657 | free(link); | |
52109584 AN |
10658 | pr_warn("prog '%s': failed to attach to %s: %s\n", |
10659 | prog->name, target_name, | |
cc4f864b | 10660 | libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 10661 | return libbpf_err_ptr(link_fd); |
cc4f864b AN |
10662 | } |
10663 | link->fd = link_fd; | |
10664 | return link; | |
10665 | } | |
10666 | ||
d60d81ac | 10667 | struct bpf_link * |
942025c9 | 10668 | bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd) |
d60d81ac | 10669 | { |
a5359091 | 10670 | return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup"); |
d60d81ac JS |
10671 | } |
10672 | ||
10673 | struct bpf_link * | |
942025c9 | 10674 | bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd) |
d60d81ac | 10675 | { |
a5359091 | 10676 | return bpf_program__attach_fd(prog, netns_fd, 0, "netns"); |
d60d81ac JS |
10677 | } |
10678 | ||
942025c9 | 10679 | struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex) |
dc8698ca AN |
10680 | { |
10681 | /* target_fd/target_ifindex use the same field in LINK_CREATE */ | |
a5359091 THJ |
10682 | return bpf_program__attach_fd(prog, ifindex, 0, "xdp"); |
10683 | } | |
10684 | ||
942025c9 | 10685 | struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog, |
a5359091 THJ |
10686 | int target_fd, |
10687 | const char *attach_func_name) | |
10688 | { | |
10689 | int btf_id; | |
10690 | ||
10691 | if (!!target_fd != !!attach_func_name) { | |
10692 | pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n", | |
10693 | prog->name); | |
e9fc3ce9 | 10694 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
10695 | } |
10696 | ||
10697 | if (prog->type != BPF_PROG_TYPE_EXT) { | |
10698 | pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace", | |
10699 | prog->name); | |
e9fc3ce9 | 10700 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
10701 | } |
10702 | ||
10703 | if (target_fd) { | |
10704 | btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd); | |
10705 | if (btf_id < 0) | |
e9fc3ce9 | 10706 | return libbpf_err_ptr(btf_id); |
a5359091 THJ |
10707 | |
10708 | return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace"); | |
10709 | } else { | |
10710 | /* no target, so use raw_tracepoint_open for compatibility | |
10711 | * with old kernels | |
10712 | */ | |
10713 | return bpf_program__attach_trace(prog); | |
10714 | } | |
dc8698ca AN |
10715 | } |
10716 | ||
c09add2f | 10717 | struct bpf_link * |
942025c9 | 10718 | bpf_program__attach_iter(const struct bpf_program *prog, |
c09add2f YS |
10719 | const struct bpf_iter_attach_opts *opts) |
10720 | { | |
cd31039a | 10721 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); |
c09add2f YS |
10722 | char errmsg[STRERR_BUFSIZE]; |
10723 | struct bpf_link *link; | |
10724 | int prog_fd, link_fd; | |
cd31039a | 10725 | __u32 target_fd = 0; |
c09add2f YS |
10726 | |
10727 | if (!OPTS_VALID(opts, bpf_iter_attach_opts)) | |
e9fc3ce9 | 10728 | return libbpf_err_ptr(-EINVAL); |
c09add2f | 10729 | |
74fc097d YS |
10730 | link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0); |
10731 | link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0); | |
cd31039a | 10732 | |
c09add2f YS |
10733 | prog_fd = bpf_program__fd(prog); |
10734 | if (prog_fd < 0) { | |
52109584 | 10735 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 10736 | return libbpf_err_ptr(-EINVAL); |
c09add2f YS |
10737 | } |
10738 | ||
10739 | link = calloc(1, sizeof(*link)); | |
10740 | if (!link) | |
e9fc3ce9 | 10741 | return libbpf_err_ptr(-ENOMEM); |
c09add2f YS |
10742 | link->detach = &bpf_link__detach_fd; |
10743 | ||
cd31039a YS |
10744 | link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER, |
10745 | &link_create_opts); | |
c09add2f YS |
10746 | if (link_fd < 0) { |
10747 | link_fd = -errno; | |
10748 | free(link); | |
52109584 AN |
10749 | pr_warn("prog '%s': failed to attach to iterator: %s\n", |
10750 | prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 10751 | return libbpf_err_ptr(link_fd); |
c09add2f YS |
10752 | } |
10753 | link->fd = link_fd; | |
10754 | return link; | |
10755 | } | |
10756 | ||
4fa5bcfe | 10757 | static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
e9fc3ce9 | 10758 | { |
4fa5bcfe AN |
10759 | *link = bpf_program__attach_iter(prog, NULL); |
10760 | return libbpf_get_error(*link); | |
e9fc3ce9 AN |
10761 | } |
10762 | ||
942025c9 | 10763 | struct bpf_link *bpf_program__attach(const struct bpf_program *prog) |
d7a18ea7 | 10764 | { |
4fa5bcfe AN |
10765 | struct bpf_link *link = NULL; |
10766 | int err; | |
10767 | ||
10768 | if (!prog->sec_def || !prog->sec_def->prog_attach_fn) | |
10769 | return libbpf_err_ptr(-EOPNOTSUPP); | |
d7a18ea7 | 10770 | |
4fa5bcfe AN |
10771 | err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link); |
10772 | if (err) | |
10773 | return libbpf_err_ptr(err); | |
10774 | ||
10775 | /* When calling bpf_program__attach() explicitly, auto-attach support | |
10776 | * is expected to work, so NULL returned link is considered an error. | |
10777 | * This is different for skeleton's attach, see comment in | |
10778 | * bpf_object__attach_skeleton(). | |
10779 | */ | |
10780 | if (!link) | |
10781 | return libbpf_err_ptr(-EOPNOTSUPP); | |
10782 | ||
10783 | return link; | |
d7a18ea7 AN |
10784 | } |
10785 | ||
590a0088 MKL |
10786 | static int bpf_link__detach_struct_ops(struct bpf_link *link) |
10787 | { | |
590a0088 MKL |
10788 | __u32 zero = 0; |
10789 | ||
c016b68e | 10790 | if (bpf_map_delete_elem(link->fd, &zero)) |
590a0088 MKL |
10791 | return -errno; |
10792 | ||
10793 | return 0; | |
10794 | } | |
10795 | ||
942025c9 | 10796 | struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map) |
590a0088 MKL |
10797 | { |
10798 | struct bpf_struct_ops *st_ops; | |
c016b68e | 10799 | struct bpf_link *link; |
590a0088 MKL |
10800 | __u32 i, zero = 0; |
10801 | int err; | |
10802 | ||
10803 | if (!bpf_map__is_struct_ops(map) || map->fd == -1) | |
e9fc3ce9 | 10804 | return libbpf_err_ptr(-EINVAL); |
590a0088 MKL |
10805 | |
10806 | link = calloc(1, sizeof(*link)); | |
10807 | if (!link) | |
e9fc3ce9 | 10808 | return libbpf_err_ptr(-EINVAL); |
590a0088 MKL |
10809 | |
10810 | st_ops = map->st_ops; | |
10811 | for (i = 0; i < btf_vlen(st_ops->type); i++) { | |
10812 | struct bpf_program *prog = st_ops->progs[i]; | |
10813 | void *kern_data; | |
10814 | int prog_fd; | |
10815 | ||
10816 | if (!prog) | |
10817 | continue; | |
10818 | ||
10819 | prog_fd = bpf_program__fd(prog); | |
10820 | kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i]; | |
10821 | *(unsigned long *)kern_data = prog_fd; | |
10822 | } | |
10823 | ||
10824 | err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0); | |
10825 | if (err) { | |
10826 | err = -errno; | |
10827 | free(link); | |
e9fc3ce9 | 10828 | return libbpf_err_ptr(err); |
590a0088 MKL |
10829 | } |
10830 | ||
c016b68e | 10831 | link->detach = bpf_link__detach_struct_ops; |
590a0088 MKL |
10832 | link->fd = map->fd; |
10833 | ||
c016b68e | 10834 | return link; |
590a0088 MKL |
10835 | } |
10836 | ||
7218c28c CL |
10837 | static enum bpf_perf_event_ret |
10838 | perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size, | |
10839 | void **copy_mem, size_t *copy_size, | |
10840 | bpf_perf_event_print_t fn, void *private_data) | |
d0cabbb0 | 10841 | { |
3dca2115 | 10842 | struct perf_event_mmap_page *header = mmap_mem; |
a64af0ef | 10843 | __u64 data_head = ring_buffer_read_head(header); |
d0cabbb0 | 10844 | __u64 data_tail = header->data_tail; |
3dca2115 DB |
10845 | void *base = ((__u8 *)header) + page_size; |
10846 | int ret = LIBBPF_PERF_EVENT_CONT; | |
10847 | struct perf_event_header *ehdr; | |
10848 | size_t ehdr_size; | |
10849 | ||
10850 | while (data_head != data_tail) { | |
10851 | ehdr = base + (data_tail & (mmap_size - 1)); | |
10852 | ehdr_size = ehdr->size; | |
10853 | ||
10854 | if (((void *)ehdr) + ehdr_size > base + mmap_size) { | |
10855 | void *copy_start = ehdr; | |
10856 | size_t len_first = base + mmap_size - copy_start; | |
10857 | size_t len_secnd = ehdr_size - len_first; | |
10858 | ||
10859 | if (*copy_size < ehdr_size) { | |
10860 | free(*copy_mem); | |
10861 | *copy_mem = malloc(ehdr_size); | |
10862 | if (!*copy_mem) { | |
10863 | *copy_size = 0; | |
d0cabbb0 JK |
10864 | ret = LIBBPF_PERF_EVENT_ERROR; |
10865 | break; | |
10866 | } | |
3dca2115 | 10867 | *copy_size = ehdr_size; |
d0cabbb0 JK |
10868 | } |
10869 | ||
3dca2115 DB |
10870 | memcpy(*copy_mem, copy_start, len_first); |
10871 | memcpy(*copy_mem + len_first, base, len_secnd); | |
10872 | ehdr = *copy_mem; | |
d0cabbb0 JK |
10873 | } |
10874 | ||
3dca2115 DB |
10875 | ret = fn(ehdr, private_data); |
10876 | data_tail += ehdr_size; | |
d0cabbb0 JK |
10877 | if (ret != LIBBPF_PERF_EVENT_CONT) |
10878 | break; | |
d0cabbb0 JK |
10879 | } |
10880 | ||
a64af0ef | 10881 | ring_buffer_write_tail(header, data_tail); |
e9fc3ce9 | 10882 | return libbpf_err(ret); |
d0cabbb0 | 10883 | } |
34be1646 | 10884 | |
7218c28c CL |
10885 | __attribute__((alias("perf_event_read_simple"))) |
10886 | enum bpf_perf_event_ret | |
10887 | bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size, | |
10888 | void **copy_mem, size_t *copy_size, | |
10889 | bpf_perf_event_print_t fn, void *private_data); | |
10890 | ||
fb84b822 AN |
10891 | struct perf_buffer; |
10892 | ||
10893 | struct perf_buffer_params { | |
10894 | struct perf_event_attr *attr; | |
10895 | /* if event_cb is specified, it takes precendence */ | |
10896 | perf_buffer_event_fn event_cb; | |
10897 | /* sample_cb and lost_cb are higher-level common-case callbacks */ | |
10898 | perf_buffer_sample_fn sample_cb; | |
10899 | perf_buffer_lost_fn lost_cb; | |
10900 | void *ctx; | |
10901 | int cpu_cnt; | |
10902 | int *cpus; | |
10903 | int *map_keys; | |
10904 | }; | |
10905 | ||
10906 | struct perf_cpu_buf { | |
10907 | struct perf_buffer *pb; | |
10908 | void *base; /* mmap()'ed memory */ | |
10909 | void *buf; /* for reconstructing segmented data */ | |
10910 | size_t buf_size; | |
10911 | int fd; | |
10912 | int cpu; | |
10913 | int map_key; | |
10914 | }; | |
10915 | ||
10916 | struct perf_buffer { | |
10917 | perf_buffer_event_fn event_cb; | |
10918 | perf_buffer_sample_fn sample_cb; | |
10919 | perf_buffer_lost_fn lost_cb; | |
10920 | void *ctx; /* passed into callbacks */ | |
10921 | ||
10922 | size_t page_size; | |
10923 | size_t mmap_size; | |
10924 | struct perf_cpu_buf **cpu_bufs; | |
10925 | struct epoll_event *events; | |
783b8f01 | 10926 | int cpu_cnt; /* number of allocated CPU buffers */ |
fb84b822 AN |
10927 | int epoll_fd; /* perf event FD */ |
10928 | int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */ | |
10929 | }; | |
10930 | ||
10931 | static void perf_buffer__free_cpu_buf(struct perf_buffer *pb, | |
10932 | struct perf_cpu_buf *cpu_buf) | |
10933 | { | |
10934 | if (!cpu_buf) | |
10935 | return; | |
10936 | if (cpu_buf->base && | |
10937 | munmap(cpu_buf->base, pb->mmap_size + pb->page_size)) | |
be18010e | 10938 | pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu); |
fb84b822 AN |
10939 | if (cpu_buf->fd >= 0) { |
10940 | ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0); | |
10941 | close(cpu_buf->fd); | |
10942 | } | |
10943 | free(cpu_buf->buf); | |
10944 | free(cpu_buf); | |
10945 | } | |
10946 | ||
10947 | void perf_buffer__free(struct perf_buffer *pb) | |
10948 | { | |
10949 | int i; | |
10950 | ||
50450fc7 | 10951 | if (IS_ERR_OR_NULL(pb)) |
fb84b822 AN |
10952 | return; |
10953 | if (pb->cpu_bufs) { | |
601b05ca | 10954 | for (i = 0; i < pb->cpu_cnt; i++) { |
fb84b822 AN |
10955 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; |
10956 | ||
601b05ca EC |
10957 | if (!cpu_buf) |
10958 | continue; | |
10959 | ||
fb84b822 AN |
10960 | bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key); |
10961 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
10962 | } | |
10963 | free(pb->cpu_bufs); | |
10964 | } | |
10965 | if (pb->epoll_fd >= 0) | |
10966 | close(pb->epoll_fd); | |
10967 | free(pb->events); | |
10968 | free(pb); | |
10969 | } | |
10970 | ||
10971 | static struct perf_cpu_buf * | |
10972 | perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr, | |
10973 | int cpu, int map_key) | |
10974 | { | |
10975 | struct perf_cpu_buf *cpu_buf; | |
10976 | char msg[STRERR_BUFSIZE]; | |
10977 | int err; | |
10978 | ||
10979 | cpu_buf = calloc(1, sizeof(*cpu_buf)); | |
10980 | if (!cpu_buf) | |
10981 | return ERR_PTR(-ENOMEM); | |
10982 | ||
10983 | cpu_buf->pb = pb; | |
10984 | cpu_buf->cpu = cpu; | |
10985 | cpu_buf->map_key = map_key; | |
10986 | ||
10987 | cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu, | |
10988 | -1, PERF_FLAG_FD_CLOEXEC); | |
10989 | if (cpu_buf->fd < 0) { | |
10990 | err = -errno; | |
be18010e KW |
10991 | pr_warn("failed to open perf buffer event on cpu #%d: %s\n", |
10992 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10993 | goto error; |
10994 | } | |
10995 | ||
10996 | cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size, | |
10997 | PROT_READ | PROT_WRITE, MAP_SHARED, | |
10998 | cpu_buf->fd, 0); | |
10999 | if (cpu_buf->base == MAP_FAILED) { | |
11000 | cpu_buf->base = NULL; | |
11001 | err = -errno; | |
be18010e KW |
11002 | pr_warn("failed to mmap perf buffer on cpu #%d: %s\n", |
11003 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11004 | goto error; |
11005 | } | |
11006 | ||
11007 | if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
11008 | err = -errno; | |
be18010e KW |
11009 | pr_warn("failed to enable perf buffer event on cpu #%d: %s\n", |
11010 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11011 | goto error; |
11012 | } | |
11013 | ||
11014 | return cpu_buf; | |
11015 | ||
11016 | error: | |
11017 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
11018 | return (struct perf_cpu_buf *)ERR_PTR(err); | |
11019 | } | |
11020 | ||
11021 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
11022 | struct perf_buffer_params *p); | |
11023 | ||
41788934 AN |
11024 | DEFAULT_VERSION(perf_buffer__new_v0_6_0, perf_buffer__new, LIBBPF_0.6.0) |
11025 | struct perf_buffer *perf_buffer__new_v0_6_0(int map_fd, size_t page_cnt, | |
11026 | perf_buffer_sample_fn sample_cb, | |
11027 | perf_buffer_lost_fn lost_cb, | |
11028 | void *ctx, | |
11029 | const struct perf_buffer_opts *opts) | |
fb84b822 AN |
11030 | { |
11031 | struct perf_buffer_params p = {}; | |
41788934 AN |
11032 | struct perf_event_attr attr = {}; |
11033 | ||
11034 | if (!OPTS_VALID(opts, perf_buffer_opts)) | |
11035 | return libbpf_err_ptr(-EINVAL); | |
4be6e05c | 11036 | |
65bb2e0f | 11037 | attr.config = PERF_COUNT_SW_BPF_OUTPUT; |
4be6e05c ACM |
11038 | attr.type = PERF_TYPE_SOFTWARE; |
11039 | attr.sample_type = PERF_SAMPLE_RAW; | |
11040 | attr.sample_period = 1; | |
11041 | attr.wakeup_events = 1; | |
fb84b822 AN |
11042 | |
11043 | p.attr = &attr; | |
41788934 AN |
11044 | p.sample_cb = sample_cb; |
11045 | p.lost_cb = lost_cb; | |
11046 | p.ctx = ctx; | |
fb84b822 | 11047 | |
e9fc3ce9 | 11048 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
11049 | } |
11050 | ||
41788934 AN |
11051 | COMPAT_VERSION(perf_buffer__new_deprecated, perf_buffer__new, LIBBPF_0.0.4) |
11052 | struct perf_buffer *perf_buffer__new_deprecated(int map_fd, size_t page_cnt, | |
11053 | const struct perf_buffer_opts *opts) | |
11054 | { | |
11055 | return perf_buffer__new_v0_6_0(map_fd, page_cnt, | |
11056 | opts ? opts->sample_cb : NULL, | |
11057 | opts ? opts->lost_cb : NULL, | |
11058 | opts ? opts->ctx : NULL, | |
11059 | NULL); | |
11060 | } | |
11061 | ||
11062 | DEFAULT_VERSION(perf_buffer__new_raw_v0_6_0, perf_buffer__new_raw, LIBBPF_0.6.0) | |
11063 | struct perf_buffer *perf_buffer__new_raw_v0_6_0(int map_fd, size_t page_cnt, | |
11064 | struct perf_event_attr *attr, | |
11065 | perf_buffer_event_fn event_cb, void *ctx, | |
11066 | const struct perf_buffer_raw_opts *opts) | |
fb84b822 AN |
11067 | { |
11068 | struct perf_buffer_params p = {}; | |
11069 | ||
41332d6e | 11070 | if (!attr) |
41788934 AN |
11071 | return libbpf_err_ptr(-EINVAL); |
11072 | ||
11073 | if (!OPTS_VALID(opts, perf_buffer_raw_opts)) | |
11074 | return libbpf_err_ptr(-EINVAL); | |
11075 | ||
11076 | p.attr = attr; | |
11077 | p.event_cb = event_cb; | |
11078 | p.ctx = ctx; | |
11079 | p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0); | |
11080 | p.cpus = OPTS_GET(opts, cpus, NULL); | |
11081 | p.map_keys = OPTS_GET(opts, map_keys, NULL); | |
fb84b822 | 11082 | |
e9fc3ce9 | 11083 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
11084 | } |
11085 | ||
41788934 AN |
11086 | COMPAT_VERSION(perf_buffer__new_raw_deprecated, perf_buffer__new_raw, LIBBPF_0.0.4) |
11087 | struct perf_buffer *perf_buffer__new_raw_deprecated(int map_fd, size_t page_cnt, | |
11088 | const struct perf_buffer_raw_opts *opts) | |
11089 | { | |
11090 | LIBBPF_OPTS(perf_buffer_raw_opts, inner_opts, | |
11091 | .cpu_cnt = opts->cpu_cnt, | |
11092 | .cpus = opts->cpus, | |
11093 | .map_keys = opts->map_keys, | |
11094 | ); | |
11095 | ||
11096 | return perf_buffer__new_raw_v0_6_0(map_fd, page_cnt, opts->attr, | |
11097 | opts->event_cb, opts->ctx, &inner_opts); | |
11098 | } | |
11099 | ||
fb84b822 AN |
11100 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, |
11101 | struct perf_buffer_params *p) | |
11102 | { | |
783b8f01 | 11103 | const char *online_cpus_file = "/sys/devices/system/cpu/online"; |
0e289487 | 11104 | struct bpf_map_info map; |
fb84b822 AN |
11105 | char msg[STRERR_BUFSIZE]; |
11106 | struct perf_buffer *pb; | |
783b8f01 | 11107 | bool *online = NULL; |
fb84b822 | 11108 | __u32 map_info_len; |
783b8f01 | 11109 | int err, i, j, n; |
fb84b822 | 11110 | |
41332d6e | 11111 | if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) { |
be18010e KW |
11112 | pr_warn("page count should be power of two, but is %zu\n", |
11113 | page_cnt); | |
fb84b822 AN |
11114 | return ERR_PTR(-EINVAL); |
11115 | } | |
11116 | ||
0e289487 AN |
11117 | /* best-effort sanity checks */ |
11118 | memset(&map, 0, sizeof(map)); | |
fb84b822 AN |
11119 | map_info_len = sizeof(map); |
11120 | err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len); | |
11121 | if (err) { | |
11122 | err = -errno; | |
0e289487 AN |
11123 | /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return |
11124 | * -EBADFD, -EFAULT, or -E2BIG on real error | |
11125 | */ | |
11126 | if (err != -EINVAL) { | |
11127 | pr_warn("failed to get map info for map FD %d: %s\n", | |
11128 | map_fd, libbpf_strerror_r(err, msg, sizeof(msg))); | |
11129 | return ERR_PTR(err); | |
11130 | } | |
11131 | pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n", | |
11132 | map_fd); | |
11133 | } else { | |
11134 | if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) { | |
11135 | pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n", | |
11136 | map.name); | |
11137 | return ERR_PTR(-EINVAL); | |
11138 | } | |
fb84b822 AN |
11139 | } |
11140 | ||
11141 | pb = calloc(1, sizeof(*pb)); | |
11142 | if (!pb) | |
11143 | return ERR_PTR(-ENOMEM); | |
11144 | ||
11145 | pb->event_cb = p->event_cb; | |
11146 | pb->sample_cb = p->sample_cb; | |
11147 | pb->lost_cb = p->lost_cb; | |
11148 | pb->ctx = p->ctx; | |
11149 | ||
11150 | pb->page_size = getpagesize(); | |
11151 | pb->mmap_size = pb->page_size * page_cnt; | |
11152 | pb->map_fd = map_fd; | |
11153 | ||
11154 | pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
11155 | if (pb->epoll_fd < 0) { | |
11156 | err = -errno; | |
be18010e KW |
11157 | pr_warn("failed to create epoll instance: %s\n", |
11158 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11159 | goto error; |
11160 | } | |
11161 | ||
11162 | if (p->cpu_cnt > 0) { | |
11163 | pb->cpu_cnt = p->cpu_cnt; | |
11164 | } else { | |
11165 | pb->cpu_cnt = libbpf_num_possible_cpus(); | |
11166 | if (pb->cpu_cnt < 0) { | |
11167 | err = pb->cpu_cnt; | |
11168 | goto error; | |
11169 | } | |
0e289487 | 11170 | if (map.max_entries && map.max_entries < pb->cpu_cnt) |
fb84b822 AN |
11171 | pb->cpu_cnt = map.max_entries; |
11172 | } | |
11173 | ||
11174 | pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events)); | |
11175 | if (!pb->events) { | |
11176 | err = -ENOMEM; | |
be18010e | 11177 | pr_warn("failed to allocate events: out of memory\n"); |
fb84b822 AN |
11178 | goto error; |
11179 | } | |
11180 | pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs)); | |
11181 | if (!pb->cpu_bufs) { | |
11182 | err = -ENOMEM; | |
be18010e | 11183 | pr_warn("failed to allocate buffers: out of memory\n"); |
fb84b822 AN |
11184 | goto error; |
11185 | } | |
11186 | ||
783b8f01 AN |
11187 | err = parse_cpu_mask_file(online_cpus_file, &online, &n); |
11188 | if (err) { | |
11189 | pr_warn("failed to get online CPU mask: %d\n", err); | |
11190 | goto error; | |
11191 | } | |
11192 | ||
11193 | for (i = 0, j = 0; i < pb->cpu_cnt; i++) { | |
fb84b822 AN |
11194 | struct perf_cpu_buf *cpu_buf; |
11195 | int cpu, map_key; | |
11196 | ||
11197 | cpu = p->cpu_cnt > 0 ? p->cpus[i] : i; | |
11198 | map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i; | |
11199 | ||
783b8f01 AN |
11200 | /* in case user didn't explicitly requested particular CPUs to |
11201 | * be attached to, skip offline/not present CPUs | |
11202 | */ | |
11203 | if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu])) | |
11204 | continue; | |
11205 | ||
fb84b822 AN |
11206 | cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key); |
11207 | if (IS_ERR(cpu_buf)) { | |
11208 | err = PTR_ERR(cpu_buf); | |
11209 | goto error; | |
11210 | } | |
11211 | ||
783b8f01 | 11212 | pb->cpu_bufs[j] = cpu_buf; |
fb84b822 AN |
11213 | |
11214 | err = bpf_map_update_elem(pb->map_fd, &map_key, | |
11215 | &cpu_buf->fd, 0); | |
11216 | if (err) { | |
11217 | err = -errno; | |
be18010e KW |
11218 | pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n", |
11219 | cpu, map_key, cpu_buf->fd, | |
11220 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11221 | goto error; |
11222 | } | |
11223 | ||
783b8f01 AN |
11224 | pb->events[j].events = EPOLLIN; |
11225 | pb->events[j].data.ptr = cpu_buf; | |
fb84b822 | 11226 | if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd, |
783b8f01 | 11227 | &pb->events[j]) < 0) { |
fb84b822 | 11228 | err = -errno; |
be18010e KW |
11229 | pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n", |
11230 | cpu, cpu_buf->fd, | |
11231 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
11232 | goto error; |
11233 | } | |
783b8f01 | 11234 | j++; |
fb84b822 | 11235 | } |
783b8f01 AN |
11236 | pb->cpu_cnt = j; |
11237 | free(online); | |
fb84b822 AN |
11238 | |
11239 | return pb; | |
11240 | ||
11241 | error: | |
783b8f01 | 11242 | free(online); |
fb84b822 AN |
11243 | if (pb) |
11244 | perf_buffer__free(pb); | |
11245 | return ERR_PTR(err); | |
11246 | } | |
11247 | ||
11248 | struct perf_sample_raw { | |
11249 | struct perf_event_header header; | |
11250 | uint32_t size; | |
385bbf7b | 11251 | char data[]; |
fb84b822 AN |
11252 | }; |
11253 | ||
11254 | struct perf_sample_lost { | |
11255 | struct perf_event_header header; | |
11256 | uint64_t id; | |
11257 | uint64_t lost; | |
11258 | uint64_t sample_id; | |
11259 | }; | |
11260 | ||
11261 | static enum bpf_perf_event_ret | |
11262 | perf_buffer__process_record(struct perf_event_header *e, void *ctx) | |
11263 | { | |
11264 | struct perf_cpu_buf *cpu_buf = ctx; | |
11265 | struct perf_buffer *pb = cpu_buf->pb; | |
11266 | void *data = e; | |
11267 | ||
11268 | /* user wants full control over parsing perf event */ | |
11269 | if (pb->event_cb) | |
11270 | return pb->event_cb(pb->ctx, cpu_buf->cpu, e); | |
11271 | ||
11272 | switch (e->type) { | |
11273 | case PERF_RECORD_SAMPLE: { | |
11274 | struct perf_sample_raw *s = data; | |
11275 | ||
11276 | if (pb->sample_cb) | |
11277 | pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size); | |
11278 | break; | |
11279 | } | |
11280 | case PERF_RECORD_LOST: { | |
11281 | struct perf_sample_lost *s = data; | |
11282 | ||
11283 | if (pb->lost_cb) | |
11284 | pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost); | |
11285 | break; | |
11286 | } | |
11287 | default: | |
be18010e | 11288 | pr_warn("unknown perf sample type %d\n", e->type); |
fb84b822 AN |
11289 | return LIBBPF_PERF_EVENT_ERROR; |
11290 | } | |
11291 | return LIBBPF_PERF_EVENT_CONT; | |
11292 | } | |
11293 | ||
11294 | static int perf_buffer__process_records(struct perf_buffer *pb, | |
11295 | struct perf_cpu_buf *cpu_buf) | |
11296 | { | |
11297 | enum bpf_perf_event_ret ret; | |
11298 | ||
7218c28c CL |
11299 | ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size, |
11300 | pb->page_size, &cpu_buf->buf, | |
11301 | &cpu_buf->buf_size, | |
11302 | perf_buffer__process_record, cpu_buf); | |
fb84b822 AN |
11303 | if (ret != LIBBPF_PERF_EVENT_CONT) |
11304 | return ret; | |
11305 | return 0; | |
11306 | } | |
11307 | ||
dca5612f AN |
11308 | int perf_buffer__epoll_fd(const struct perf_buffer *pb) |
11309 | { | |
11310 | return pb->epoll_fd; | |
11311 | } | |
11312 | ||
fb84b822 AN |
11313 | int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms) |
11314 | { | |
11315 | int i, cnt, err; | |
11316 | ||
11317 | cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms); | |
e9fc3ce9 | 11318 | if (cnt < 0) |
af0efa05 | 11319 | return -errno; |
e9fc3ce9 | 11320 | |
fb84b822 AN |
11321 | for (i = 0; i < cnt; i++) { |
11322 | struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr; | |
11323 | ||
11324 | err = perf_buffer__process_records(pb, cpu_buf); | |
11325 | if (err) { | |
be18010e | 11326 | pr_warn("error while processing records: %d\n", err); |
e9fc3ce9 | 11327 | return libbpf_err(err); |
fb84b822 AN |
11328 | } |
11329 | } | |
e9fc3ce9 | 11330 | return cnt; |
fb84b822 AN |
11331 | } |
11332 | ||
dca5612f AN |
11333 | /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer |
11334 | * manager. | |
11335 | */ | |
11336 | size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb) | |
11337 | { | |
11338 | return pb->cpu_cnt; | |
11339 | } | |
11340 | ||
11341 | /* | |
11342 | * Return perf_event FD of a ring buffer in *buf_idx* slot of | |
11343 | * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using | |
11344 | * select()/poll()/epoll() Linux syscalls. | |
11345 | */ | |
11346 | int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx) | |
11347 | { | |
11348 | struct perf_cpu_buf *cpu_buf; | |
11349 | ||
11350 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 11351 | return libbpf_err(-EINVAL); |
dca5612f AN |
11352 | |
11353 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
11354 | if (!cpu_buf) | |
e9fc3ce9 | 11355 | return libbpf_err(-ENOENT); |
dca5612f AN |
11356 | |
11357 | return cpu_buf->fd; | |
11358 | } | |
11359 | ||
11360 | /* | |
11361 | * Consume data from perf ring buffer corresponding to slot *buf_idx* in | |
11362 | * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to | |
11363 | * consume, do nothing and return success. | |
11364 | * Returns: | |
11365 | * - 0 on success; | |
11366 | * - <0 on failure. | |
11367 | */ | |
11368 | int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx) | |
11369 | { | |
11370 | struct perf_cpu_buf *cpu_buf; | |
11371 | ||
11372 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 11373 | return libbpf_err(-EINVAL); |
dca5612f AN |
11374 | |
11375 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
11376 | if (!cpu_buf) | |
e9fc3ce9 | 11377 | return libbpf_err(-ENOENT); |
dca5612f AN |
11378 | |
11379 | return perf_buffer__process_records(pb, cpu_buf); | |
11380 | } | |
11381 | ||
272d51af EC |
11382 | int perf_buffer__consume(struct perf_buffer *pb) |
11383 | { | |
11384 | int i, err; | |
11385 | ||
11386 | for (i = 0; i < pb->cpu_cnt; i++) { | |
11387 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; | |
11388 | ||
11389 | if (!cpu_buf) | |
11390 | continue; | |
11391 | ||
11392 | err = perf_buffer__process_records(pb, cpu_buf); | |
11393 | if (err) { | |
dca5612f | 11394 | pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err); |
e9fc3ce9 | 11395 | return libbpf_err(err); |
272d51af EC |
11396 | } |
11397 | } | |
11398 | return 0; | |
11399 | } | |
11400 | ||
34be1646 SL |
11401 | struct bpf_prog_info_array_desc { |
11402 | int array_offset; /* e.g. offset of jited_prog_insns */ | |
11403 | int count_offset; /* e.g. offset of jited_prog_len */ | |
11404 | int size_offset; /* > 0: offset of rec size, | |
11405 | * < 0: fix size of -size_offset | |
11406 | */ | |
11407 | }; | |
11408 | ||
11409 | static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = { | |
11410 | [BPF_PROG_INFO_JITED_INSNS] = { | |
11411 | offsetof(struct bpf_prog_info, jited_prog_insns), | |
11412 | offsetof(struct bpf_prog_info, jited_prog_len), | |
11413 | -1, | |
11414 | }, | |
11415 | [BPF_PROG_INFO_XLATED_INSNS] = { | |
11416 | offsetof(struct bpf_prog_info, xlated_prog_insns), | |
11417 | offsetof(struct bpf_prog_info, xlated_prog_len), | |
11418 | -1, | |
11419 | }, | |
11420 | [BPF_PROG_INFO_MAP_IDS] = { | |
11421 | offsetof(struct bpf_prog_info, map_ids), | |
11422 | offsetof(struct bpf_prog_info, nr_map_ids), | |
11423 | -(int)sizeof(__u32), | |
11424 | }, | |
11425 | [BPF_PROG_INFO_JITED_KSYMS] = { | |
11426 | offsetof(struct bpf_prog_info, jited_ksyms), | |
11427 | offsetof(struct bpf_prog_info, nr_jited_ksyms), | |
11428 | -(int)sizeof(__u64), | |
11429 | }, | |
11430 | [BPF_PROG_INFO_JITED_FUNC_LENS] = { | |
11431 | offsetof(struct bpf_prog_info, jited_func_lens), | |
11432 | offsetof(struct bpf_prog_info, nr_jited_func_lens), | |
11433 | -(int)sizeof(__u32), | |
11434 | }, | |
11435 | [BPF_PROG_INFO_FUNC_INFO] = { | |
11436 | offsetof(struct bpf_prog_info, func_info), | |
11437 | offsetof(struct bpf_prog_info, nr_func_info), | |
11438 | offsetof(struct bpf_prog_info, func_info_rec_size), | |
11439 | }, | |
11440 | [BPF_PROG_INFO_LINE_INFO] = { | |
11441 | offsetof(struct bpf_prog_info, line_info), | |
11442 | offsetof(struct bpf_prog_info, nr_line_info), | |
11443 | offsetof(struct bpf_prog_info, line_info_rec_size), | |
11444 | }, | |
11445 | [BPF_PROG_INFO_JITED_LINE_INFO] = { | |
11446 | offsetof(struct bpf_prog_info, jited_line_info), | |
11447 | offsetof(struct bpf_prog_info, nr_jited_line_info), | |
11448 | offsetof(struct bpf_prog_info, jited_line_info_rec_size), | |
11449 | }, | |
11450 | [BPF_PROG_INFO_PROG_TAGS] = { | |
11451 | offsetof(struct bpf_prog_info, prog_tags), | |
11452 | offsetof(struct bpf_prog_info, nr_prog_tags), | |
11453 | -(int)sizeof(__u8) * BPF_TAG_SIZE, | |
11454 | }, | |
11455 | ||
11456 | }; | |
11457 | ||
8983b731 AN |
11458 | static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, |
11459 | int offset) | |
34be1646 SL |
11460 | { |
11461 | __u32 *array = (__u32 *)info; | |
11462 | ||
11463 | if (offset >= 0) | |
11464 | return array[offset / sizeof(__u32)]; | |
11465 | return -(int)offset; | |
11466 | } | |
11467 | ||
8983b731 AN |
11468 | static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, |
11469 | int offset) | |
34be1646 SL |
11470 | { |
11471 | __u64 *array = (__u64 *)info; | |
11472 | ||
11473 | if (offset >= 0) | |
11474 | return array[offset / sizeof(__u64)]; | |
11475 | return -(int)offset; | |
11476 | } | |
11477 | ||
11478 | static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset, | |
11479 | __u32 val) | |
11480 | { | |
11481 | __u32 *array = (__u32 *)info; | |
11482 | ||
11483 | if (offset >= 0) | |
11484 | array[offset / sizeof(__u32)] = val; | |
11485 | } | |
11486 | ||
11487 | static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset, | |
11488 | __u64 val) | |
11489 | { | |
11490 | __u64 *array = (__u64 *)info; | |
11491 | ||
11492 | if (offset >= 0) | |
11493 | array[offset / sizeof(__u64)] = val; | |
11494 | } | |
11495 | ||
11496 | struct bpf_prog_info_linear * | |
11497 | bpf_program__get_prog_info_linear(int fd, __u64 arrays) | |
11498 | { | |
11499 | struct bpf_prog_info_linear *info_linear; | |
11500 | struct bpf_prog_info info = {}; | |
11501 | __u32 info_len = sizeof(info); | |
11502 | __u32 data_len = 0; | |
11503 | int i, err; | |
11504 | void *ptr; | |
11505 | ||
11506 | if (arrays >> BPF_PROG_INFO_LAST_ARRAY) | |
e9fc3ce9 | 11507 | return libbpf_err_ptr(-EINVAL); |
34be1646 SL |
11508 | |
11509 | /* step 1: get array dimensions */ | |
11510 | err = bpf_obj_get_info_by_fd(fd, &info, &info_len); | |
11511 | if (err) { | |
11512 | pr_debug("can't get prog info: %s", strerror(errno)); | |
e9fc3ce9 | 11513 | return libbpf_err_ptr(-EFAULT); |
34be1646 SL |
11514 | } |
11515 | ||
11516 | /* step 2: calculate total size of all arrays */ | |
11517 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
11518 | bool include_array = (arrays & (1UL << i)) > 0; | |
11519 | struct bpf_prog_info_array_desc *desc; | |
11520 | __u32 count, size; | |
11521 | ||
11522 | desc = bpf_prog_info_array_desc + i; | |
11523 | ||
11524 | /* kernel is too old to support this field */ | |
11525 | if (info_len < desc->array_offset + sizeof(__u32) || | |
11526 | info_len < desc->count_offset + sizeof(__u32) || | |
11527 | (desc->size_offset > 0 && info_len < desc->size_offset)) | |
11528 | include_array = false; | |
11529 | ||
11530 | if (!include_array) { | |
11531 | arrays &= ~(1UL << i); /* clear the bit */ | |
11532 | continue; | |
11533 | } | |
11534 | ||
11535 | count = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
11536 | size = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
11537 | ||
11538 | data_len += count * size; | |
11539 | } | |
11540 | ||
11541 | /* step 3: allocate continuous memory */ | |
11542 | data_len = roundup(data_len, sizeof(__u64)); | |
11543 | info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len); | |
11544 | if (!info_linear) | |
e9fc3ce9 | 11545 | return libbpf_err_ptr(-ENOMEM); |
34be1646 SL |
11546 | |
11547 | /* step 4: fill data to info_linear->info */ | |
11548 | info_linear->arrays = arrays; | |
11549 | memset(&info_linear->info, 0, sizeof(info)); | |
11550 | ptr = info_linear->data; | |
11551 | ||
11552 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
11553 | struct bpf_prog_info_array_desc *desc; | |
11554 | __u32 count, size; | |
11555 | ||
11556 | if ((arrays & (1UL << i)) == 0) | |
11557 | continue; | |
11558 | ||
11559 | desc = bpf_prog_info_array_desc + i; | |
11560 | count = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
11561 | size = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
11562 | bpf_prog_info_set_offset_u32(&info_linear->info, | |
11563 | desc->count_offset, count); | |
11564 | bpf_prog_info_set_offset_u32(&info_linear->info, | |
11565 | desc->size_offset, size); | |
11566 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
11567 | desc->array_offset, | |
11568 | ptr_to_u64(ptr)); | |
11569 | ptr += count * size; | |
11570 | } | |
11571 | ||
11572 | /* step 5: call syscall again to get required arrays */ | |
11573 | err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len); | |
11574 | if (err) { | |
11575 | pr_debug("can't get prog info: %s", strerror(errno)); | |
11576 | free(info_linear); | |
e9fc3ce9 | 11577 | return libbpf_err_ptr(-EFAULT); |
34be1646 SL |
11578 | } |
11579 | ||
11580 | /* step 6: verify the data */ | |
11581 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
11582 | struct bpf_prog_info_array_desc *desc; | |
11583 | __u32 v1, v2; | |
11584 | ||
11585 | if ((arrays & (1UL << i)) == 0) | |
11586 | continue; | |
11587 | ||
11588 | desc = bpf_prog_info_array_desc + i; | |
11589 | v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
11590 | v2 = bpf_prog_info_read_offset_u32(&info_linear->info, | |
11591 | desc->count_offset); | |
11592 | if (v1 != v2) | |
be18010e | 11593 | pr_warn("%s: mismatch in element count\n", __func__); |
34be1646 SL |
11594 | |
11595 | v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
11596 | v2 = bpf_prog_info_read_offset_u32(&info_linear->info, | |
11597 | desc->size_offset); | |
11598 | if (v1 != v2) | |
be18010e | 11599 | pr_warn("%s: mismatch in rec size\n", __func__); |
34be1646 SL |
11600 | } |
11601 | ||
11602 | /* step 7: update info_len and data_len */ | |
11603 | info_linear->info_len = sizeof(struct bpf_prog_info); | |
11604 | info_linear->data_len = data_len; | |
11605 | ||
11606 | return info_linear; | |
11607 | } | |
11608 | ||
11609 | void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear) | |
11610 | { | |
11611 | int i; | |
11612 | ||
11613 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
11614 | struct bpf_prog_info_array_desc *desc; | |
11615 | __u64 addr, offs; | |
11616 | ||
11617 | if ((info_linear->arrays & (1UL << i)) == 0) | |
11618 | continue; | |
11619 | ||
11620 | desc = bpf_prog_info_array_desc + i; | |
11621 | addr = bpf_prog_info_read_offset_u64(&info_linear->info, | |
11622 | desc->array_offset); | |
11623 | offs = addr - ptr_to_u64(info_linear->data); | |
11624 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
11625 | desc->array_offset, offs); | |
11626 | } | |
11627 | } | |
11628 | ||
11629 | void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear) | |
11630 | { | |
11631 | int i; | |
11632 | ||
11633 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
11634 | struct bpf_prog_info_array_desc *desc; | |
11635 | __u64 addr, offs; | |
11636 | ||
11637 | if ((info_linear->arrays & (1UL << i)) == 0) | |
11638 | continue; | |
11639 | ||
11640 | desc = bpf_prog_info_array_desc + i; | |
11641 | offs = bpf_prog_info_read_offset_u64(&info_linear->info, | |
11642 | desc->array_offset); | |
11643 | addr = offs + ptr_to_u64(info_linear->data); | |
11644 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
11645 | desc->array_offset, addr); | |
11646 | } | |
11647 | } | |
6446b315 | 11648 | |
ff26ce5c EC |
11649 | int bpf_program__set_attach_target(struct bpf_program *prog, |
11650 | int attach_prog_fd, | |
11651 | const char *attach_func_name) | |
11652 | { | |
fe62de31 | 11653 | int btf_obj_fd = 0, btf_id = 0, err; |
ff26ce5c | 11654 | |
2d5ec1c6 | 11655 | if (!prog || attach_prog_fd < 0) |
e9fc3ce9 | 11656 | return libbpf_err(-EINVAL); |
ff26ce5c | 11657 | |
fe62de31 | 11658 | if (prog->obj->loaded) |
e9fc3ce9 | 11659 | return libbpf_err(-EINVAL); |
fe62de31 | 11660 | |
2d5ec1c6 AN |
11661 | if (attach_prog_fd && !attach_func_name) { |
11662 | /* remember attach_prog_fd and let bpf_program__load() find | |
11663 | * BTF ID during the program load | |
11664 | */ | |
11665 | prog->attach_prog_fd = attach_prog_fd; | |
11666 | return 0; | |
11667 | } | |
11668 | ||
fe62de31 | 11669 | if (attach_prog_fd) { |
ff26ce5c EC |
11670 | btf_id = libbpf_find_prog_btf_id(attach_func_name, |
11671 | attach_prog_fd); | |
fe62de31 | 11672 | if (btf_id < 0) |
e9fc3ce9 | 11673 | return libbpf_err(btf_id); |
fe62de31 | 11674 | } else { |
2d5ec1c6 AN |
11675 | if (!attach_func_name) |
11676 | return libbpf_err(-EINVAL); | |
11677 | ||
fe62de31 AN |
11678 | /* load btf_vmlinux, if not yet */ |
11679 | err = bpf_object__load_vmlinux_btf(prog->obj, true); | |
11680 | if (err) | |
e9fc3ce9 | 11681 | return libbpf_err(err); |
fe62de31 AN |
11682 | err = find_kernel_btf_id(prog->obj, attach_func_name, |
11683 | prog->expected_attach_type, | |
11684 | &btf_obj_fd, &btf_id); | |
11685 | if (err) | |
e9fc3ce9 | 11686 | return libbpf_err(err); |
fe62de31 | 11687 | } |
ff26ce5c EC |
11688 | |
11689 | prog->attach_btf_id = btf_id; | |
fe62de31 | 11690 | prog->attach_btf_obj_fd = btf_obj_fd; |
ff26ce5c EC |
11691 | prog->attach_prog_fd = attach_prog_fd; |
11692 | return 0; | |
11693 | } | |
11694 | ||
6803ee25 | 11695 | int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz) |
6446b315 | 11696 | { |
6803ee25 AN |
11697 | int err = 0, n, len, start, end = -1; |
11698 | bool *tmp; | |
6446b315 | 11699 | |
6803ee25 AN |
11700 | *mask = NULL; |
11701 | *mask_sz = 0; | |
11702 | ||
11703 | /* Each sub string separated by ',' has format \d+-\d+ or \d+ */ | |
11704 | while (*s) { | |
11705 | if (*s == ',' || *s == '\n') { | |
11706 | s++; | |
11707 | continue; | |
11708 | } | |
11709 | n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len); | |
11710 | if (n <= 0 || n > 2) { | |
11711 | pr_warn("Failed to get CPU range %s: %d\n", s, n); | |
11712 | err = -EINVAL; | |
11713 | goto cleanup; | |
11714 | } else if (n == 1) { | |
11715 | end = start; | |
11716 | } | |
11717 | if (start < 0 || start > end) { | |
11718 | pr_warn("Invalid CPU range [%d,%d] in %s\n", | |
11719 | start, end, s); | |
11720 | err = -EINVAL; | |
11721 | goto cleanup; | |
11722 | } | |
11723 | tmp = realloc(*mask, end + 1); | |
11724 | if (!tmp) { | |
11725 | err = -ENOMEM; | |
11726 | goto cleanup; | |
11727 | } | |
11728 | *mask = tmp; | |
11729 | memset(tmp + *mask_sz, 0, start - *mask_sz); | |
11730 | memset(tmp + start, 1, end - start + 1); | |
11731 | *mask_sz = end + 1; | |
11732 | s += len; | |
11733 | } | |
11734 | if (!*mask_sz) { | |
11735 | pr_warn("Empty CPU range\n"); | |
11736 | return -EINVAL; | |
11737 | } | |
11738 | return 0; | |
11739 | cleanup: | |
11740 | free(*mask); | |
11741 | *mask = NULL; | |
11742 | return err; | |
11743 | } | |
11744 | ||
11745 | int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz) | |
11746 | { | |
11747 | int fd, err = 0, len; | |
11748 | char buf[128]; | |
6446b315 | 11749 | |
92274e24 | 11750 | fd = open(fcpu, O_RDONLY | O_CLOEXEC); |
6446b315 | 11751 | if (fd < 0) { |
6803ee25 AN |
11752 | err = -errno; |
11753 | pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err); | |
11754 | return err; | |
6446b315 HL |
11755 | } |
11756 | len = read(fd, buf, sizeof(buf)); | |
11757 | close(fd); | |
11758 | if (len <= 0) { | |
6803ee25 AN |
11759 | err = len ? -errno : -EINVAL; |
11760 | pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err); | |
11761 | return err; | |
6446b315 | 11762 | } |
6803ee25 AN |
11763 | if (len >= sizeof(buf)) { |
11764 | pr_warn("CPU mask is too big in file %s\n", fcpu); | |
11765 | return -E2BIG; | |
6446b315 HL |
11766 | } |
11767 | buf[len] = '\0'; | |
11768 | ||
6803ee25 AN |
11769 | return parse_cpu_mask_str(buf, mask, mask_sz); |
11770 | } | |
11771 | ||
11772 | int libbpf_num_possible_cpus(void) | |
11773 | { | |
11774 | static const char *fcpu = "/sys/devices/system/cpu/possible"; | |
11775 | static int cpus; | |
11776 | int err, n, i, tmp_cpus; | |
11777 | bool *mask; | |
11778 | ||
11779 | tmp_cpus = READ_ONCE(cpus); | |
11780 | if (tmp_cpus > 0) | |
11781 | return tmp_cpus; | |
11782 | ||
11783 | err = parse_cpu_mask_file(fcpu, &mask, &n); | |
11784 | if (err) | |
e9fc3ce9 | 11785 | return libbpf_err(err); |
6803ee25 AN |
11786 | |
11787 | tmp_cpus = 0; | |
11788 | for (i = 0; i < n; i++) { | |
11789 | if (mask[i]) | |
11790 | tmp_cpus++; | |
6446b315 | 11791 | } |
6803ee25 | 11792 | free(mask); |
56fbc241 TC |
11793 | |
11794 | WRITE_ONCE(cpus, tmp_cpus); | |
11795 | return tmp_cpus; | |
6446b315 | 11796 | } |
d66562fb AN |
11797 | |
11798 | int bpf_object__open_skeleton(struct bpf_object_skeleton *s, | |
11799 | const struct bpf_object_open_opts *opts) | |
11800 | { | |
11801 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts, | |
11802 | .object_name = s->name, | |
11803 | ); | |
11804 | struct bpf_object *obj; | |
e9fc3ce9 | 11805 | int i, err; |
d66562fb AN |
11806 | |
11807 | /* Attempt to preserve opts->object_name, unless overriden by user | |
11808 | * explicitly. Overwriting object name for skeletons is discouraged, | |
11809 | * as it breaks global data maps, because they contain object name | |
11810 | * prefix as their own map name prefix. When skeleton is generated, | |
11811 | * bpftool is making an assumption that this name will stay the same. | |
11812 | */ | |
11813 | if (opts) { | |
11814 | memcpy(&skel_opts, opts, sizeof(*opts)); | |
11815 | if (!opts->object_name) | |
11816 | skel_opts.object_name = s->name; | |
11817 | } | |
11818 | ||
11819 | obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts); | |
e9fc3ce9 AN |
11820 | err = libbpf_get_error(obj); |
11821 | if (err) { | |
11822 | pr_warn("failed to initialize skeleton BPF object '%s': %d\n", | |
11823 | s->name, err); | |
11824 | return libbpf_err(err); | |
d66562fb AN |
11825 | } |
11826 | ||
11827 | *s->obj = obj; | |
11828 | ||
11829 | for (i = 0; i < s->map_cnt; i++) { | |
11830 | struct bpf_map **map = s->maps[i].map; | |
11831 | const char *name = s->maps[i].name; | |
11832 | void **mmaped = s->maps[i].mmaped; | |
11833 | ||
11834 | *map = bpf_object__find_map_by_name(obj, name); | |
11835 | if (!*map) { | |
11836 | pr_warn("failed to find skeleton map '%s'\n", name); | |
e9fc3ce9 | 11837 | return libbpf_err(-ESRCH); |
d66562fb AN |
11838 | } |
11839 | ||
2ad97d47 | 11840 | /* externs shouldn't be pre-setup from user code */ |
81bfdd08 | 11841 | if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG) |
d66562fb AN |
11842 | *mmaped = (*map)->mmaped; |
11843 | } | |
11844 | ||
11845 | for (i = 0; i < s->prog_cnt; i++) { | |
11846 | struct bpf_program **prog = s->progs[i].prog; | |
11847 | const char *name = s->progs[i].name; | |
11848 | ||
11849 | *prog = bpf_object__find_program_by_name(obj, name); | |
11850 | if (!*prog) { | |
11851 | pr_warn("failed to find skeleton program '%s'\n", name); | |
e9fc3ce9 | 11852 | return libbpf_err(-ESRCH); |
d66562fb AN |
11853 | } |
11854 | } | |
11855 | ||
11856 | return 0; | |
11857 | } | |
11858 | ||
11859 | int bpf_object__load_skeleton(struct bpf_object_skeleton *s) | |
11860 | { | |
11861 | int i, err; | |
11862 | ||
11863 | err = bpf_object__load(*s->obj); | |
11864 | if (err) { | |
11865 | pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err); | |
e9fc3ce9 | 11866 | return libbpf_err(err); |
d66562fb AN |
11867 | } |
11868 | ||
11869 | for (i = 0; i < s->map_cnt; i++) { | |
11870 | struct bpf_map *map = *s->maps[i].map; | |
11871 | size_t mmap_sz = bpf_map_mmap_sz(map); | |
11872 | int prot, map_fd = bpf_map__fd(map); | |
11873 | void **mmaped = s->maps[i].mmaped; | |
d66562fb AN |
11874 | |
11875 | if (!mmaped) | |
11876 | continue; | |
11877 | ||
11878 | if (!(map->def.map_flags & BPF_F_MMAPABLE)) { | |
11879 | *mmaped = NULL; | |
11880 | continue; | |
11881 | } | |
11882 | ||
11883 | if (map->def.map_flags & BPF_F_RDONLY_PROG) | |
11884 | prot = PROT_READ; | |
11885 | else | |
11886 | prot = PROT_READ | PROT_WRITE; | |
11887 | ||
11888 | /* Remap anonymous mmap()-ed "map initialization image" as | |
11889 | * a BPF map-backed mmap()-ed memory, but preserving the same | |
11890 | * memory address. This will cause kernel to change process' | |
11891 | * page table to point to a different piece of kernel memory, | |
11892 | * but from userspace point of view memory address (and its | |
11893 | * contents, being identical at this point) will stay the | |
11894 | * same. This mapping will be released by bpf_object__close() | |
11895 | * as per normal clean up procedure, so we don't need to worry | |
11896 | * about it from skeleton's clean up perspective. | |
11897 | */ | |
2ad97d47 AN |
11898 | *mmaped = mmap(map->mmaped, mmap_sz, prot, |
11899 | MAP_SHARED | MAP_FIXED, map_fd, 0); | |
11900 | if (*mmaped == MAP_FAILED) { | |
d66562fb AN |
11901 | err = -errno; |
11902 | *mmaped = NULL; | |
11903 | pr_warn("failed to re-mmap() map '%s': %d\n", | |
11904 | bpf_map__name(map), err); | |
e9fc3ce9 | 11905 | return libbpf_err(err); |
d66562fb AN |
11906 | } |
11907 | } | |
11908 | ||
11909 | return 0; | |
11910 | } | |
11911 | ||
11912 | int bpf_object__attach_skeleton(struct bpf_object_skeleton *s) | |
11913 | { | |
e9fc3ce9 | 11914 | int i, err; |
d66562fb AN |
11915 | |
11916 | for (i = 0; i < s->prog_cnt; i++) { | |
11917 | struct bpf_program *prog = *s->progs[i].prog; | |
11918 | struct bpf_link **link = s->progs[i].link; | |
d66562fb | 11919 | |
d9297581 AN |
11920 | if (!prog->load) |
11921 | continue; | |
11922 | ||
5532dfd4 | 11923 | /* auto-attaching not supported for this program */ |
4fa5bcfe | 11924 | if (!prog->sec_def || !prog->sec_def->prog_attach_fn) |
d66562fb AN |
11925 | continue; |
11926 | ||
4fa5bcfe AN |
11927 | /* if user already set the link manually, don't attempt auto-attach */ |
11928 | if (*link) | |
11929 | continue; | |
11930 | ||
11931 | err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link); | |
e9fc3ce9 | 11932 | if (err) { |
4fa5bcfe | 11933 | pr_warn("prog '%s': failed to auto-attach: %d\n", |
e9fc3ce9 AN |
11934 | bpf_program__name(prog), err); |
11935 | return libbpf_err(err); | |
d66562fb | 11936 | } |
4fa5bcfe AN |
11937 | |
11938 | /* It's possible that for some SEC() definitions auto-attach | |
11939 | * is supported in some cases (e.g., if definition completely | |
11940 | * specifies target information), but is not in other cases. | |
11941 | * SEC("uprobe") is one such case. If user specified target | |
11942 | * binary and function name, such BPF program can be | |
11943 | * auto-attached. But if not, it shouldn't trigger skeleton's | |
11944 | * attach to fail. It should just be skipped. | |
11945 | * attach_fn signals such case with returning 0 (no error) and | |
11946 | * setting link to NULL. | |
11947 | */ | |
d66562fb AN |
11948 | } |
11949 | ||
11950 | return 0; | |
11951 | } | |
11952 | ||
11953 | void bpf_object__detach_skeleton(struct bpf_object_skeleton *s) | |
11954 | { | |
11955 | int i; | |
11956 | ||
11957 | for (i = 0; i < s->prog_cnt; i++) { | |
11958 | struct bpf_link **link = s->progs[i].link; | |
11959 | ||
50450fc7 | 11960 | bpf_link__destroy(*link); |
d66562fb AN |
11961 | *link = NULL; |
11962 | } | |
11963 | } | |
11964 | ||
11965 | void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s) | |
11966 | { | |
a32ea51a YS |
11967 | if (!s) |
11968 | return; | |
11969 | ||
d66562fb AN |
11970 | if (s->progs) |
11971 | bpf_object__detach_skeleton(s); | |
11972 | if (s->obj) | |
11973 | bpf_object__close(*s->obj); | |
11974 | free(s->maps); | |
11975 | free(s->progs); | |
11976 | free(s); | |
11977 | } |