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" |
b3f59d66 | 57 | |
9b16137a WN |
58 | #ifndef EM_BPF |
59 | #define EM_BPF 247 | |
60 | #endif | |
61 | ||
f367540c JS |
62 | #ifndef BPF_FS_MAGIC |
63 | #define BPF_FS_MAGIC 0xcafe4a11 | |
64 | #endif | |
65 | ||
9c0f8cbd AN |
66 | #define BPF_INSN_SZ (sizeof(struct bpf_insn)) |
67 | ||
ff466b58 AI |
68 | /* vsprintf() in __base_pr() uses nonliteral format string. It may break |
69 | * compilation if user enables corresponding warning. Disable it explicitly. | |
70 | */ | |
71 | #pragma GCC diagnostic ignored "-Wformat-nonliteral" | |
72 | ||
b3f59d66 WN |
73 | #define __printf(a, b) __attribute__((format(printf, a, b))) |
74 | ||
590a0088 | 75 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj); |
590a0088 MKL |
76 | static const struct btf_type * |
77 | skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id); | |
78 | ||
a8a1f7d0 SF |
79 | static int __base_pr(enum libbpf_print_level level, const char *format, |
80 | va_list args) | |
b3f59d66 | 81 | { |
6f1ae8b6 YS |
82 | if (level == LIBBPF_DEBUG) |
83 | return 0; | |
84 | ||
a8a1f7d0 | 85 | return vfprintf(stderr, format, args); |
b3f59d66 WN |
86 | } |
87 | ||
a8a1f7d0 | 88 | static libbpf_print_fn_t __libbpf_pr = __base_pr; |
b3f59d66 | 89 | |
e87fd8ba | 90 | libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn) |
b3f59d66 | 91 | { |
e87fd8ba AN |
92 | libbpf_print_fn_t old_print_fn = __libbpf_pr; |
93 | ||
6f1ae8b6 | 94 | __libbpf_pr = fn; |
e87fd8ba | 95 | return old_print_fn; |
b3f59d66 | 96 | } |
1a5e3fb1 | 97 | |
8461ef8b YS |
98 | __printf(2, 3) |
99 | void libbpf_print(enum libbpf_print_level level, const char *format, ...) | |
100 | { | |
101 | va_list args; | |
102 | ||
6f1ae8b6 YS |
103 | if (!__libbpf_pr) |
104 | return; | |
105 | ||
8461ef8b | 106 | va_start(args, format); |
6f1ae8b6 | 107 | __libbpf_pr(level, format, args); |
8461ef8b YS |
108 | va_end(args); |
109 | } | |
110 | ||
dc3a2d25 THJ |
111 | static void pr_perm_msg(int err) |
112 | { | |
113 | struct rlimit limit; | |
114 | char buf[100]; | |
115 | ||
116 | if (err != -EPERM || geteuid() != 0) | |
117 | return; | |
118 | ||
119 | err = getrlimit(RLIMIT_MEMLOCK, &limit); | |
120 | if (err) | |
121 | return; | |
122 | ||
123 | if (limit.rlim_cur == RLIM_INFINITY) | |
124 | return; | |
125 | ||
126 | if (limit.rlim_cur < 1024) | |
b5c7d0d0 | 127 | snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur); |
dc3a2d25 THJ |
128 | else if (limit.rlim_cur < 1024*1024) |
129 | snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024); | |
130 | else | |
131 | snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024)); | |
132 | ||
133 | pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n", | |
134 | buf); | |
135 | } | |
136 | ||
6371ca3b WN |
137 | #define STRERR_BUFSIZE 128 |
138 | ||
1a5e3fb1 WN |
139 | /* Copied from tools/perf/util/util.h */ |
140 | #ifndef zfree | |
141 | # define zfree(ptr) ({ free(*ptr); *ptr = NULL; }) | |
142 | #endif | |
143 | ||
144 | #ifndef zclose | |
145 | # define zclose(fd) ({ \ | |
146 | int ___err = 0; \ | |
147 | if ((fd) >= 0) \ | |
148 | ___err = close((fd)); \ | |
149 | fd = -1; \ | |
150 | ___err; }) | |
151 | #endif | |
152 | ||
34be1646 SL |
153 | static inline __u64 ptr_to_u64(const void *ptr) |
154 | { | |
155 | return (__u64) (unsigned long) ptr; | |
156 | } | |
157 | ||
47b6cb4d | 158 | enum kern_feature_id { |
47eff617 | 159 | /* v4.14: kernel support for program & map names. */ |
47b6cb4d | 160 | FEAT_PROG_NAME, |
8837fe5d | 161 | /* v5.2: kernel support for global data sections. */ |
47b6cb4d | 162 | FEAT_GLOBAL_DATA, |
68b08647 AN |
163 | /* BTF support */ |
164 | FEAT_BTF, | |
d7c4b398 | 165 | /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */ |
47b6cb4d | 166 | FEAT_BTF_FUNC, |
d7c4b398 | 167 | /* BTF_KIND_VAR and BTF_KIND_DATASEC support */ |
47b6cb4d | 168 | FEAT_BTF_DATASEC, |
2d3eb67f | 169 | /* BTF_FUNC_GLOBAL is supported */ |
47b6cb4d AN |
170 | FEAT_BTF_GLOBAL_FUNC, |
171 | /* BPF_F_MMAPABLE is supported for arrays */ | |
172 | FEAT_ARRAY_MMAP, | |
25498a19 | 173 | /* kernel support for expected_attach_type in BPF_PROG_LOAD */ |
47b6cb4d | 174 | FEAT_EXP_ATTACH_TYPE, |
109cea5a AN |
175 | /* bpf_probe_read_{kernel,user}[_str] helpers */ |
176 | FEAT_PROBE_READ_KERN, | |
5d23328d YZ |
177 | /* BPF_PROG_BIND_MAP is supported */ |
178 | FEAT_PROG_BIND_MAP, | |
4f33a53d AN |
179 | /* Kernel support for module BTFs */ |
180 | FEAT_MODULE_BTF, | |
22541a9e IL |
181 | /* BTF_KIND_FLOAT support */ |
182 | FEAT_BTF_FLOAT, | |
47b6cb4d | 183 | __FEAT_CNT, |
47eff617 SF |
184 | }; |
185 | ||
47b6cb4d AN |
186 | static bool kernel_supports(enum kern_feature_id feat_id); |
187 | ||
166750bc AN |
188 | enum reloc_type { |
189 | RELO_LD64, | |
190 | RELO_CALL, | |
191 | RELO_DATA, | |
192 | RELO_EXTERN, | |
53eddb5e | 193 | RELO_SUBPROG_ADDR, |
166750bc AN |
194 | }; |
195 | ||
196 | struct reloc_desc { | |
197 | enum reloc_type type; | |
198 | int insn_idx; | |
199 | int map_idx; | |
200 | int sym_off; | |
c3c55696 | 201 | bool processed; |
166750bc AN |
202 | }; |
203 | ||
25498a19 AN |
204 | struct bpf_sec_def; |
205 | ||
206 | typedef struct bpf_link *(*attach_fn_t)(const struct bpf_sec_def *sec, | |
207 | struct bpf_program *prog); | |
208 | ||
209 | struct bpf_sec_def { | |
210 | const char *sec; | |
211 | size_t len; | |
212 | enum bpf_prog_type prog_type; | |
213 | enum bpf_attach_type expected_attach_type; | |
214 | bool is_exp_attach_type_optional; | |
215 | bool is_attachable; | |
216 | bool is_attach_btf; | |
2b288740 | 217 | bool is_sleepable; |
25498a19 AN |
218 | attach_fn_t attach_fn; |
219 | }; | |
220 | ||
a5b8bd47 WN |
221 | /* |
222 | * bpf_prog should be a better name but it has been used in | |
223 | * linux/filter.h. | |
224 | */ | |
225 | struct bpf_program { | |
25498a19 | 226 | const struct bpf_sec_def *sec_def; |
52109584 | 227 | char *sec_name; |
c1122392 AN |
228 | size_t sec_idx; |
229 | /* this program's instruction offset (in number of instructions) | |
230 | * within its containing ELF section | |
231 | */ | |
232 | size_t sec_insn_off; | |
233 | /* number of original instructions in ELF section belonging to this | |
234 | * program, not taking into account subprogram instructions possible | |
235 | * appended later during relocation | |
236 | */ | |
237 | size_t sec_insn_cnt; | |
238 | /* Offset (in number of instructions) of the start of instruction | |
239 | * belonging to this BPF program within its containing main BPF | |
240 | * program. For the entry-point (main) BPF program, this is always | |
241 | * zero. For a sub-program, this gets reset before each of main BPF | |
242 | * programs are processed and relocated and is used to determined | |
243 | * whether sub-program was already appended to the main program, and | |
244 | * if yes, at which instruction offset. | |
245 | */ | |
246 | size_t sub_insn_off; | |
247 | ||
248 | char *name; | |
52109584 | 249 | /* sec_name with / replaced by _; makes recursive pinning |
33a2c75c SF |
250 | * in bpf_object__pin_programs easier |
251 | */ | |
252 | char *pin_name; | |
c1122392 AN |
253 | |
254 | /* instructions that belong to BPF program; insns[0] is located at | |
255 | * sec_insn_off instruction within its ELF section in ELF file, so | |
256 | * when mapping ELF file instruction index to the local instruction, | |
257 | * one needs to subtract sec_insn_off; and vice versa. | |
258 | */ | |
a5b8bd47 | 259 | struct bpf_insn *insns; |
c1122392 AN |
260 | /* actual number of instruction in this BPF program's image; for |
261 | * entry-point BPF programs this includes the size of main program | |
262 | * itself plus all the used sub-programs, appended at the end | |
263 | */ | |
c3c55696 | 264 | size_t insns_cnt; |
34090915 | 265 | |
166750bc | 266 | struct reloc_desc *reloc_desc; |
34090915 | 267 | int nr_reloc; |
da11b417 | 268 | int log_level; |
55cffde2 | 269 | |
b580563e WN |
270 | struct { |
271 | int nr; | |
272 | int *fds; | |
273 | } instances; | |
274 | bpf_program_prep_t preprocessor; | |
aa9b1ac3 WN |
275 | |
276 | struct bpf_object *obj; | |
277 | void *priv; | |
278 | bpf_program_clear_priv_t clear_priv; | |
d7be143b | 279 | |
c1122392 AN |
280 | bool load; |
281 | enum bpf_prog_type type; | |
d7be143b | 282 | enum bpf_attach_type expected_attach_type; |
c1122392 | 283 | int prog_ifindex; |
91abb4a6 | 284 | __u32 attach_btf_obj_fd; |
12a8654b | 285 | __u32 attach_btf_id; |
e7bf94db | 286 | __u32 attach_prog_fd; |
2993e051 YS |
287 | void *func_info; |
288 | __u32 func_info_rec_size; | |
f0187f0b | 289 | __u32 func_info_cnt; |
47eff617 | 290 | |
3d650141 MKL |
291 | void *line_info; |
292 | __u32 line_info_rec_size; | |
293 | __u32 line_info_cnt; | |
04656198 | 294 | __u32 prog_flags; |
a5b8bd47 WN |
295 | }; |
296 | ||
590a0088 MKL |
297 | struct bpf_struct_ops { |
298 | const char *tname; | |
299 | const struct btf_type *type; | |
300 | struct bpf_program **progs; | |
301 | __u32 *kern_func_off; | |
302 | /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */ | |
303 | void *data; | |
304 | /* e.g. struct bpf_struct_ops_tcp_congestion_ops in | |
305 | * btf_vmlinux's format. | |
306 | * struct bpf_struct_ops_tcp_congestion_ops { | |
307 | * [... some other kernel fields ...] | |
308 | * struct tcp_congestion_ops data; | |
309 | * } | |
310 | * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops) | |
311 | * bpf_map__init_kern_struct_ops() will populate the "kern_vdata" | |
312 | * from "data". | |
313 | */ | |
314 | void *kern_vdata; | |
315 | __u32 type_id; | |
316 | }; | |
317 | ||
ac9d1389 AN |
318 | #define DATA_SEC ".data" |
319 | #define BSS_SEC ".bss" | |
320 | #define RODATA_SEC ".rodata" | |
81bfdd08 | 321 | #define KCONFIG_SEC ".kconfig" |
1c0c7074 | 322 | #define KSYMS_SEC ".ksyms" |
590a0088 | 323 | #define STRUCT_OPS_SEC ".struct_ops" |
ac9d1389 | 324 | |
d859900c DB |
325 | enum libbpf_map_type { |
326 | LIBBPF_MAP_UNSPEC, | |
327 | LIBBPF_MAP_DATA, | |
328 | LIBBPF_MAP_BSS, | |
329 | LIBBPF_MAP_RODATA, | |
81bfdd08 | 330 | LIBBPF_MAP_KCONFIG, |
d859900c DB |
331 | }; |
332 | ||
333 | static const char * const libbpf_type_to_btf_name[] = { | |
ac9d1389 AN |
334 | [LIBBPF_MAP_DATA] = DATA_SEC, |
335 | [LIBBPF_MAP_BSS] = BSS_SEC, | |
336 | [LIBBPF_MAP_RODATA] = RODATA_SEC, | |
81bfdd08 | 337 | [LIBBPF_MAP_KCONFIG] = KCONFIG_SEC, |
d859900c DB |
338 | }; |
339 | ||
9d759a9b | 340 | struct bpf_map { |
561bbcca | 341 | char *name; |
01af3bf0 | 342 | int fd; |
db48814b AN |
343 | int sec_idx; |
344 | size_t sec_offset; | |
f0307a7e | 345 | int map_ifindex; |
addb9fc9 | 346 | int inner_map_fd; |
9d759a9b | 347 | struct bpf_map_def def; |
1bdb6c9a | 348 | __u32 numa_node; |
646f02ff | 349 | __u32 btf_var_idx; |
5b891af7 MKL |
350 | __u32 btf_key_type_id; |
351 | __u32 btf_value_type_id; | |
590a0088 | 352 | __u32 btf_vmlinux_value_type_id; |
9d759a9b WN |
353 | void *priv; |
354 | bpf_map_clear_priv_t clear_priv; | |
d859900c | 355 | enum libbpf_map_type libbpf_type; |
eba9c5f4 | 356 | void *mmaped; |
590a0088 | 357 | struct bpf_struct_ops *st_ops; |
646f02ff AN |
358 | struct bpf_map *inner_map; |
359 | void **init_slots; | |
360 | int init_slots_sz; | |
4580b25f THJ |
361 | char *pin_path; |
362 | bool pinned; | |
ec6d5f47 | 363 | bool reused; |
d859900c DB |
364 | }; |
365 | ||
166750bc AN |
366 | enum extern_type { |
367 | EXT_UNKNOWN, | |
2e33efe3 | 368 | EXT_KCFG, |
1c0c7074 | 369 | EXT_KSYM, |
2e33efe3 AN |
370 | }; |
371 | ||
372 | enum kcfg_type { | |
373 | KCFG_UNKNOWN, | |
374 | KCFG_CHAR, | |
375 | KCFG_BOOL, | |
376 | KCFG_INT, | |
377 | KCFG_TRISTATE, | |
378 | KCFG_CHAR_ARR, | |
166750bc AN |
379 | }; |
380 | ||
381 | struct extern_desc { | |
2e33efe3 | 382 | enum extern_type type; |
166750bc AN |
383 | int sym_idx; |
384 | int btf_id; | |
2e33efe3 AN |
385 | int sec_btf_id; |
386 | const char *name; | |
166750bc | 387 | bool is_set; |
2e33efe3 AN |
388 | bool is_weak; |
389 | union { | |
390 | struct { | |
391 | enum kcfg_type type; | |
392 | int sz; | |
393 | int align; | |
394 | int data_off; | |
395 | bool is_signed; | |
396 | } kcfg; | |
1c0c7074 AN |
397 | struct { |
398 | unsigned long long addr; | |
d370bbe1 HL |
399 | |
400 | /* target btf_id of the corresponding kernel var. */ | |
284d2587 AN |
401 | int kernel_btf_obj_fd; |
402 | int kernel_btf_id; | |
d370bbe1 HL |
403 | |
404 | /* local btf_id of the ksym extern's type. */ | |
405 | __u32 type_id; | |
1c0c7074 | 406 | } ksym; |
2e33efe3 | 407 | }; |
166750bc AN |
408 | }; |
409 | ||
9a208eff WN |
410 | static LIST_HEAD(bpf_objects_list); |
411 | ||
4f33a53d AN |
412 | struct module_btf { |
413 | struct btf *btf; | |
414 | char *name; | |
415 | __u32 id; | |
91abb4a6 | 416 | int fd; |
4f33a53d AN |
417 | }; |
418 | ||
1a5e3fb1 | 419 | struct bpf_object { |
d859900c | 420 | char name[BPF_OBJ_NAME_LEN]; |
cb1e5e96 | 421 | char license[64]; |
438363c0 | 422 | __u32 kern_version; |
0b3d1efa | 423 | |
a5b8bd47 WN |
424 | struct bpf_program *programs; |
425 | size_t nr_programs; | |
9d759a9b WN |
426 | struct bpf_map *maps; |
427 | size_t nr_maps; | |
bf829271 | 428 | size_t maps_cap; |
9d759a9b | 429 | |
8601fd42 | 430 | char *kconfig; |
166750bc AN |
431 | struct extern_desc *externs; |
432 | int nr_extern; | |
81bfdd08 | 433 | int kconfig_map_idx; |
5d23328d | 434 | int rodata_map_idx; |
166750bc | 435 | |
52d3352e | 436 | bool loaded; |
c3c55696 | 437 | bool has_subcalls; |
a5b8bd47 | 438 | |
1a5e3fb1 WN |
439 | /* |
440 | * Information when doing elf related work. Only valid if fd | |
441 | * is valid. | |
442 | */ | |
443 | struct { | |
444 | int fd; | |
5e61f270 | 445 | const void *obj_buf; |
6c956392 | 446 | size_t obj_buf_sz; |
1a5e3fb1 WN |
447 | Elf *elf; |
448 | GElf_Ehdr ehdr; | |
bec7d68c | 449 | Elf_Data *symbols; |
d859900c DB |
450 | Elf_Data *data; |
451 | Elf_Data *rodata; | |
452 | Elf_Data *bss; | |
590a0088 | 453 | Elf_Data *st_ops_data; |
88a82120 | 454 | size_t shstrndx; /* section index for section name strings */ |
77ba9a5b | 455 | size_t strtabidx; |
b62f06e8 WN |
456 | struct { |
457 | GElf_Shdr shdr; | |
458 | Elf_Data *data; | |
1f8e2bcb AN |
459 | } *reloc_sects; |
460 | int nr_reloc_sects; | |
666810e8 | 461 | int maps_shndx; |
abd29c93 | 462 | int btf_maps_shndx; |
646f02ff | 463 | __u32 btf_maps_sec_btf_id; |
48cca7e4 | 464 | int text_shndx; |
166750bc | 465 | int symbols_shndx; |
d859900c DB |
466 | int data_shndx; |
467 | int rodata_shndx; | |
468 | int bss_shndx; | |
590a0088 | 469 | int st_ops_shndx; |
1a5e3fb1 | 470 | } efile; |
9a208eff WN |
471 | /* |
472 | * All loaded bpf_object is linked in a list, which is | |
473 | * hidden to caller. bpf_objects__<func> handlers deal with | |
474 | * all objects. | |
475 | */ | |
476 | struct list_head list; | |
10931d24 | 477 | |
8a138aed | 478 | struct btf *btf; |
0f7515ca AN |
479 | struct btf_ext *btf_ext; |
480 | ||
a6ed02ca KS |
481 | /* Parse and load BTF vmlinux if any of the programs in the object need |
482 | * it at load time. | |
483 | */ | |
484 | struct btf *btf_vmlinux; | |
0f7515ca AN |
485 | /* vmlinux BTF override for CO-RE relocations */ |
486 | struct btf *btf_vmlinux_override; | |
4f33a53d AN |
487 | /* Lazily initialized kernel module BTFs */ |
488 | struct module_btf *btf_modules; | |
489 | bool btf_modules_loaded; | |
490 | size_t btf_module_cnt; | |
491 | size_t btf_module_cap; | |
8a138aed | 492 | |
10931d24 WN |
493 | void *priv; |
494 | bpf_object_clear_priv_t clear_priv; | |
495 | ||
1a5e3fb1 WN |
496 | char path[]; |
497 | }; | |
498 | #define obj_elf_valid(o) ((o)->efile.elf) | |
499 | ||
88a82120 AN |
500 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off); |
501 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off); | |
502 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx); | |
503 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name); | |
504 | static int elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn, GElf_Shdr *hdr); | |
505 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn); | |
506 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn); | |
c1122392 AN |
507 | static int elf_sym_by_sec_off(const struct bpf_object *obj, size_t sec_idx, |
508 | size_t off, __u32 sym_type, GElf_Sym *sym); | |
88a82120 | 509 | |
29cd77f4 | 510 | void bpf_program__unload(struct bpf_program *prog) |
55cffde2 | 511 | { |
b580563e WN |
512 | int i; |
513 | ||
55cffde2 WN |
514 | if (!prog) |
515 | return; | |
516 | ||
b580563e WN |
517 | /* |
518 | * If the object is opened but the program was never loaded, | |
519 | * it is possible that prog->instances.nr == -1. | |
520 | */ | |
521 | if (prog->instances.nr > 0) { | |
522 | for (i = 0; i < prog->instances.nr; i++) | |
523 | zclose(prog->instances.fds[i]); | |
524 | } else if (prog->instances.nr != -1) { | |
be18010e KW |
525 | pr_warn("Internal error: instances.nr is %d\n", |
526 | prog->instances.nr); | |
b580563e WN |
527 | } |
528 | ||
529 | prog->instances.nr = -1; | |
530 | zfree(&prog->instances.fds); | |
2993e051 | 531 | |
2993e051 | 532 | zfree(&prog->func_info); |
07a09d1b | 533 | zfree(&prog->line_info); |
55cffde2 WN |
534 | } |
535 | ||
a5b8bd47 WN |
536 | static void bpf_program__exit(struct bpf_program *prog) |
537 | { | |
538 | if (!prog) | |
539 | return; | |
540 | ||
aa9b1ac3 WN |
541 | if (prog->clear_priv) |
542 | prog->clear_priv(prog, prog->priv); | |
543 | ||
544 | prog->priv = NULL; | |
545 | prog->clear_priv = NULL; | |
546 | ||
55cffde2 | 547 | bpf_program__unload(prog); |
88cda1c9 | 548 | zfree(&prog->name); |
52109584 | 549 | zfree(&prog->sec_name); |
33a2c75c | 550 | zfree(&prog->pin_name); |
a5b8bd47 | 551 | zfree(&prog->insns); |
34090915 WN |
552 | zfree(&prog->reloc_desc); |
553 | ||
554 | prog->nr_reloc = 0; | |
a5b8bd47 | 555 | prog->insns_cnt = 0; |
c1122392 | 556 | prog->sec_idx = -1; |
a5b8bd47 WN |
557 | } |
558 | ||
33a2c75c SF |
559 | static char *__bpf_program__pin_name(struct bpf_program *prog) |
560 | { | |
561 | char *name, *p; | |
562 | ||
52109584 | 563 | name = p = strdup(prog->sec_name); |
33a2c75c SF |
564 | while ((p = strchr(p, '/'))) |
565 | *p = '_'; | |
566 | ||
567 | return name; | |
568 | } | |
569 | ||
c3c55696 AN |
570 | static bool insn_is_subprog_call(const struct bpf_insn *insn) |
571 | { | |
572 | return BPF_CLASS(insn->code) == BPF_JMP && | |
573 | BPF_OP(insn->code) == BPF_CALL && | |
574 | BPF_SRC(insn->code) == BPF_K && | |
575 | insn->src_reg == BPF_PSEUDO_CALL && | |
576 | insn->dst_reg == 0 && | |
577 | insn->off == 0; | |
578 | } | |
579 | ||
b8f871fa YS |
580 | static bool is_ldimm64(struct bpf_insn *insn) |
581 | { | |
582 | return insn->code == (BPF_LD | BPF_IMM | BPF_DW); | |
583 | } | |
584 | ||
53eddb5e YS |
585 | static bool insn_is_pseudo_func(struct bpf_insn *insn) |
586 | { | |
587 | return is_ldimm64(insn) && insn->src_reg == BPF_PSEUDO_FUNC; | |
588 | } | |
589 | ||
a5b8bd47 | 590 | static int |
c3c55696 AN |
591 | bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog, |
592 | const char *name, size_t sec_idx, const char *sec_name, | |
593 | size_t sec_off, void *insn_data, size_t insn_data_sz) | |
a5b8bd47 | 594 | { |
c1122392 AN |
595 | if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) { |
596 | pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n", | |
597 | sec_name, name, sec_off, insn_data_sz); | |
a5b8bd47 WN |
598 | return -EINVAL; |
599 | } | |
600 | ||
1ad9cbb8 | 601 | memset(prog, 0, sizeof(*prog)); |
c3c55696 AN |
602 | prog->obj = obj; |
603 | ||
c1122392 AN |
604 | prog->sec_idx = sec_idx; |
605 | prog->sec_insn_off = sec_off / BPF_INSN_SZ; | |
606 | prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ; | |
607 | /* insns_cnt can later be increased by appending used subprograms */ | |
608 | prog->insns_cnt = prog->sec_insn_cnt; | |
a5b8bd47 | 609 | |
c1122392 AN |
610 | prog->type = BPF_PROG_TYPE_UNSPEC; |
611 | prog->load = true; | |
a5b8bd47 | 612 | |
c1122392 AN |
613 | prog->instances.fds = NULL; |
614 | prog->instances.nr = -1; | |
615 | ||
52109584 AN |
616 | prog->sec_name = strdup(sec_name); |
617 | if (!prog->sec_name) | |
c1122392 AN |
618 | goto errout; |
619 | ||
620 | prog->name = strdup(name); | |
621 | if (!prog->name) | |
a5b8bd47 | 622 | goto errout; |
a5b8bd47 | 623 | |
33a2c75c | 624 | prog->pin_name = __bpf_program__pin_name(prog); |
c1122392 | 625 | if (!prog->pin_name) |
33a2c75c | 626 | goto errout; |
33a2c75c | 627 | |
c1122392 AN |
628 | prog->insns = malloc(insn_data_sz); |
629 | if (!prog->insns) | |
a5b8bd47 | 630 | goto errout; |
c1122392 | 631 | memcpy(prog->insns, insn_data, insn_data_sz); |
a5b8bd47 WN |
632 | |
633 | return 0; | |
634 | errout: | |
c1122392 | 635 | pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name); |
a5b8bd47 WN |
636 | bpf_program__exit(prog); |
637 | return -ENOMEM; | |
638 | } | |
639 | ||
640 | static int | |
c1122392 AN |
641 | bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data, |
642 | const char *sec_name, int sec_idx) | |
a5b8bd47 | 643 | { |
c1122392 AN |
644 | struct bpf_program *prog, *progs; |
645 | void *data = sec_data->d_buf; | |
646 | size_t sec_sz = sec_data->d_size, sec_off, prog_sz; | |
a5b8bd47 | 647 | int nr_progs, err; |
c1122392 AN |
648 | const char *name; |
649 | GElf_Sym sym; | |
a5b8bd47 WN |
650 | |
651 | progs = obj->programs; | |
652 | nr_progs = obj->nr_programs; | |
c1122392 | 653 | sec_off = 0; |
a5b8bd47 | 654 | |
c1122392 AN |
655 | while (sec_off < sec_sz) { |
656 | if (elf_sym_by_sec_off(obj, sec_idx, sec_off, STT_FUNC, &sym)) { | |
657 | pr_warn("sec '%s': failed to find program symbol at offset %zu\n", | |
658 | sec_name, sec_off); | |
659 | return -LIBBPF_ERRNO__FORMAT; | |
660 | } | |
88cda1c9 | 661 | |
c1122392 | 662 | prog_sz = sym.st_size; |
88cda1c9 | 663 | |
c1122392 AN |
664 | name = elf_sym_str(obj, sym.st_name); |
665 | if (!name) { | |
666 | pr_warn("sec '%s': failed to get symbol name for offset %zu\n", | |
667 | sec_name, sec_off); | |
668 | return -LIBBPF_ERRNO__FORMAT; | |
669 | } | |
88cda1c9 | 670 | |
c1122392 AN |
671 | if (sec_off + prog_sz > sec_sz) { |
672 | pr_warn("sec '%s': program at offset %zu crosses section boundary\n", | |
673 | sec_name, sec_off); | |
674 | return -LIBBPF_ERRNO__FORMAT; | |
675 | } | |
88cda1c9 | 676 | |
c3c55696 AN |
677 | pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n", |
678 | sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz); | |
88cda1c9 | 679 | |
c3c55696 | 680 | progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs)); |
c1122392 AN |
681 | if (!progs) { |
682 | /* | |
683 | * In this case the original obj->programs | |
684 | * is still valid, so don't need special treat for | |
685 | * bpf_close_object(). | |
686 | */ | |
687 | pr_warn("sec '%s': failed to alloc memory for new program '%s'\n", | |
688 | sec_name, name); | |
689 | return -ENOMEM; | |
88cda1c9 | 690 | } |
c1122392 | 691 | obj->programs = progs; |
88cda1c9 | 692 | |
c1122392 | 693 | prog = &progs[nr_progs]; |
9a94f277 | 694 | |
c3c55696 AN |
695 | err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name, |
696 | sec_off, data + sec_off, prog_sz); | |
c1122392 AN |
697 | if (err) |
698 | return err; | |
9a94f277 | 699 | |
c1122392 AN |
700 | nr_progs++; |
701 | obj->nr_programs = nr_progs; | |
702 | ||
703 | sec_off += prog_sz; | |
88cda1c9 MKL |
704 | } |
705 | ||
706 | return 0; | |
707 | } | |
708 | ||
5e61f270 AN |
709 | static __u32 get_kernel_version(void) |
710 | { | |
711 | __u32 major, minor, patch; | |
712 | struct utsname info; | |
713 | ||
714 | uname(&info); | |
715 | if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3) | |
716 | return 0; | |
717 | return KERNEL_VERSION(major, minor, patch); | |
718 | } | |
719 | ||
590a0088 MKL |
720 | static const struct btf_member * |
721 | find_member_by_offset(const struct btf_type *t, __u32 bit_offset) | |
722 | { | |
723 | struct btf_member *m; | |
724 | int i; | |
725 | ||
726 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
727 | if (btf_member_bit_offset(t, i) == bit_offset) | |
728 | return m; | |
729 | } | |
730 | ||
731 | return NULL; | |
732 | } | |
733 | ||
734 | static const struct btf_member * | |
735 | find_member_by_name(const struct btf *btf, const struct btf_type *t, | |
736 | const char *name) | |
737 | { | |
738 | struct btf_member *m; | |
739 | int i; | |
740 | ||
741 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
742 | if (!strcmp(btf__name_by_offset(btf, m->name_off), name)) | |
743 | return m; | |
744 | } | |
745 | ||
746 | return NULL; | |
747 | } | |
748 | ||
749 | #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_" | |
a6ed02ca KS |
750 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
751 | const char *name, __u32 kind); | |
590a0088 MKL |
752 | |
753 | static int | |
754 | find_struct_ops_kern_types(const struct btf *btf, const char *tname, | |
755 | const struct btf_type **type, __u32 *type_id, | |
756 | const struct btf_type **vtype, __u32 *vtype_id, | |
757 | const struct btf_member **data_member) | |
758 | { | |
759 | const struct btf_type *kern_type, *kern_vtype; | |
760 | const struct btf_member *kern_data_member; | |
761 | __s32 kern_vtype_id, kern_type_id; | |
590a0088 MKL |
762 | __u32 i; |
763 | ||
764 | kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT); | |
765 | if (kern_type_id < 0) { | |
766 | pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n", | |
767 | tname); | |
768 | return kern_type_id; | |
769 | } | |
770 | kern_type = btf__type_by_id(btf, kern_type_id); | |
771 | ||
772 | /* Find the corresponding "map_value" type that will be used | |
773 | * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example, | |
774 | * find "struct bpf_struct_ops_tcp_congestion_ops" from the | |
775 | * btf_vmlinux. | |
776 | */ | |
a6ed02ca KS |
777 | kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX, |
778 | tname, BTF_KIND_STRUCT); | |
590a0088 | 779 | if (kern_vtype_id < 0) { |
a6ed02ca KS |
780 | pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n", |
781 | STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
782 | return kern_vtype_id; |
783 | } | |
784 | kern_vtype = btf__type_by_id(btf, kern_vtype_id); | |
785 | ||
786 | /* Find "struct tcp_congestion_ops" from | |
787 | * struct bpf_struct_ops_tcp_congestion_ops { | |
788 | * [ ... ] | |
789 | * struct tcp_congestion_ops data; | |
790 | * } | |
791 | */ | |
792 | kern_data_member = btf_members(kern_vtype); | |
793 | for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) { | |
794 | if (kern_data_member->type == kern_type_id) | |
795 | break; | |
796 | } | |
797 | if (i == btf_vlen(kern_vtype)) { | |
a6ed02ca KS |
798 | pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n", |
799 | tname, STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
800 | return -EINVAL; |
801 | } | |
802 | ||
803 | *type = kern_type; | |
804 | *type_id = kern_type_id; | |
805 | *vtype = kern_vtype; | |
806 | *vtype_id = kern_vtype_id; | |
807 | *data_member = kern_data_member; | |
808 | ||
809 | return 0; | |
810 | } | |
811 | ||
812 | static bool bpf_map__is_struct_ops(const struct bpf_map *map) | |
813 | { | |
814 | return map->def.type == BPF_MAP_TYPE_STRUCT_OPS; | |
815 | } | |
816 | ||
817 | /* Init the map's fields that depend on kern_btf */ | |
818 | static int bpf_map__init_kern_struct_ops(struct bpf_map *map, | |
819 | const struct btf *btf, | |
820 | const struct btf *kern_btf) | |
821 | { | |
822 | const struct btf_member *member, *kern_member, *kern_data_member; | |
823 | const struct btf_type *type, *kern_type, *kern_vtype; | |
824 | __u32 i, kern_type_id, kern_vtype_id, kern_data_off; | |
825 | struct bpf_struct_ops *st_ops; | |
826 | void *data, *kern_data; | |
827 | const char *tname; | |
828 | int err; | |
829 | ||
830 | st_ops = map->st_ops; | |
831 | type = st_ops->type; | |
832 | tname = st_ops->tname; | |
833 | err = find_struct_ops_kern_types(kern_btf, tname, | |
834 | &kern_type, &kern_type_id, | |
835 | &kern_vtype, &kern_vtype_id, | |
836 | &kern_data_member); | |
837 | if (err) | |
838 | return err; | |
839 | ||
840 | pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n", | |
841 | map->name, st_ops->type_id, kern_type_id, kern_vtype_id); | |
842 | ||
843 | map->def.value_size = kern_vtype->size; | |
844 | map->btf_vmlinux_value_type_id = kern_vtype_id; | |
845 | ||
846 | st_ops->kern_vdata = calloc(1, kern_vtype->size); | |
847 | if (!st_ops->kern_vdata) | |
848 | return -ENOMEM; | |
849 | ||
850 | data = st_ops->data; | |
851 | kern_data_off = kern_data_member->offset / 8; | |
852 | kern_data = st_ops->kern_vdata + kern_data_off; | |
853 | ||
854 | member = btf_members(type); | |
855 | for (i = 0; i < btf_vlen(type); i++, member++) { | |
856 | const struct btf_type *mtype, *kern_mtype; | |
857 | __u32 mtype_id, kern_mtype_id; | |
858 | void *mdata, *kern_mdata; | |
859 | __s64 msize, kern_msize; | |
860 | __u32 moff, kern_moff; | |
861 | __u32 kern_member_idx; | |
862 | const char *mname; | |
863 | ||
864 | mname = btf__name_by_offset(btf, member->name_off); | |
865 | kern_member = find_member_by_name(kern_btf, kern_type, mname); | |
866 | if (!kern_member) { | |
867 | pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n", | |
868 | map->name, mname); | |
869 | return -ENOTSUP; | |
870 | } | |
871 | ||
872 | kern_member_idx = kern_member - btf_members(kern_type); | |
873 | if (btf_member_bitfield_size(type, i) || | |
874 | btf_member_bitfield_size(kern_type, kern_member_idx)) { | |
875 | pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n", | |
876 | map->name, mname); | |
877 | return -ENOTSUP; | |
878 | } | |
879 | ||
880 | moff = member->offset / 8; | |
881 | kern_moff = kern_member->offset / 8; | |
882 | ||
883 | mdata = data + moff; | |
884 | kern_mdata = kern_data + kern_moff; | |
885 | ||
886 | mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id); | |
887 | kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type, | |
888 | &kern_mtype_id); | |
889 | if (BTF_INFO_KIND(mtype->info) != | |
890 | BTF_INFO_KIND(kern_mtype->info)) { | |
891 | pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n", | |
892 | map->name, mname, BTF_INFO_KIND(mtype->info), | |
893 | BTF_INFO_KIND(kern_mtype->info)); | |
894 | return -ENOTSUP; | |
895 | } | |
896 | ||
897 | if (btf_is_ptr(mtype)) { | |
898 | struct bpf_program *prog; | |
899 | ||
d2836ddd MKL |
900 | prog = st_ops->progs[i]; |
901 | if (!prog) | |
902 | continue; | |
903 | ||
590a0088 MKL |
904 | kern_mtype = skip_mods_and_typedefs(kern_btf, |
905 | kern_mtype->type, | |
906 | &kern_mtype_id); | |
d2836ddd MKL |
907 | |
908 | /* mtype->type must be a func_proto which was | |
909 | * guaranteed in bpf_object__collect_st_ops_relos(), | |
910 | * so only check kern_mtype for func_proto here. | |
911 | */ | |
912 | if (!btf_is_func_proto(kern_mtype)) { | |
913 | pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n", | |
590a0088 MKL |
914 | map->name, mname); |
915 | return -ENOTSUP; | |
916 | } | |
917 | ||
590a0088 MKL |
918 | prog->attach_btf_id = kern_type_id; |
919 | prog->expected_attach_type = kern_member_idx; | |
920 | ||
921 | st_ops->kern_func_off[i] = kern_data_off + kern_moff; | |
922 | ||
923 | pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n", | |
924 | map->name, mname, prog->name, moff, | |
925 | kern_moff); | |
926 | ||
927 | continue; | |
928 | } | |
929 | ||
930 | msize = btf__resolve_size(btf, mtype_id); | |
931 | kern_msize = btf__resolve_size(kern_btf, kern_mtype_id); | |
932 | if (msize < 0 || kern_msize < 0 || msize != kern_msize) { | |
933 | pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n", | |
934 | map->name, mname, (ssize_t)msize, | |
935 | (ssize_t)kern_msize); | |
936 | return -ENOTSUP; | |
937 | } | |
938 | ||
939 | pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n", | |
940 | map->name, mname, (unsigned int)msize, | |
941 | moff, kern_moff); | |
942 | memcpy(kern_mdata, mdata, msize); | |
943 | } | |
944 | ||
945 | return 0; | |
946 | } | |
947 | ||
948 | static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj) | |
949 | { | |
590a0088 MKL |
950 | struct bpf_map *map; |
951 | size_t i; | |
952 | int err; | |
953 | ||
954 | for (i = 0; i < obj->nr_maps; i++) { | |
955 | map = &obj->maps[i]; | |
956 | ||
957 | if (!bpf_map__is_struct_ops(map)) | |
958 | continue; | |
959 | ||
a6ed02ca KS |
960 | err = bpf_map__init_kern_struct_ops(map, obj->btf, |
961 | obj->btf_vmlinux); | |
962 | if (err) | |
590a0088 | 963 | return err; |
590a0088 MKL |
964 | } |
965 | ||
590a0088 MKL |
966 | return 0; |
967 | } | |
968 | ||
969 | static int bpf_object__init_struct_ops_maps(struct bpf_object *obj) | |
970 | { | |
971 | const struct btf_type *type, *datasec; | |
972 | const struct btf_var_secinfo *vsi; | |
973 | struct bpf_struct_ops *st_ops; | |
974 | const char *tname, *var_name; | |
975 | __s32 type_id, datasec_id; | |
976 | const struct btf *btf; | |
977 | struct bpf_map *map; | |
978 | __u32 i; | |
979 | ||
980 | if (obj->efile.st_ops_shndx == -1) | |
981 | return 0; | |
982 | ||
983 | btf = obj->btf; | |
984 | datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC, | |
985 | BTF_KIND_DATASEC); | |
986 | if (datasec_id < 0) { | |
987 | pr_warn("struct_ops init: DATASEC %s not found\n", | |
988 | STRUCT_OPS_SEC); | |
989 | return -EINVAL; | |
990 | } | |
991 | ||
992 | datasec = btf__type_by_id(btf, datasec_id); | |
993 | vsi = btf_var_secinfos(datasec); | |
994 | for (i = 0; i < btf_vlen(datasec); i++, vsi++) { | |
995 | type = btf__type_by_id(obj->btf, vsi->type); | |
996 | var_name = btf__name_by_offset(obj->btf, type->name_off); | |
997 | ||
998 | type_id = btf__resolve_type(obj->btf, vsi->type); | |
999 | if (type_id < 0) { | |
1000 | pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n", | |
1001 | vsi->type, STRUCT_OPS_SEC); | |
1002 | return -EINVAL; | |
1003 | } | |
1004 | ||
1005 | type = btf__type_by_id(obj->btf, type_id); | |
1006 | tname = btf__name_by_offset(obj->btf, type->name_off); | |
1007 | if (!tname[0]) { | |
1008 | pr_warn("struct_ops init: anonymous type is not supported\n"); | |
1009 | return -ENOTSUP; | |
1010 | } | |
1011 | if (!btf_is_struct(type)) { | |
1012 | pr_warn("struct_ops init: %s is not a struct\n", tname); | |
1013 | return -EINVAL; | |
1014 | } | |
1015 | ||
1016 | map = bpf_object__add_map(obj); | |
1017 | if (IS_ERR(map)) | |
1018 | return PTR_ERR(map); | |
1019 | ||
1020 | map->sec_idx = obj->efile.st_ops_shndx; | |
1021 | map->sec_offset = vsi->offset; | |
1022 | map->name = strdup(var_name); | |
1023 | if (!map->name) | |
1024 | return -ENOMEM; | |
1025 | ||
1026 | map->def.type = BPF_MAP_TYPE_STRUCT_OPS; | |
1027 | map->def.key_size = sizeof(int); | |
1028 | map->def.value_size = type->size; | |
1029 | map->def.max_entries = 1; | |
1030 | ||
1031 | map->st_ops = calloc(1, sizeof(*map->st_ops)); | |
1032 | if (!map->st_ops) | |
1033 | return -ENOMEM; | |
1034 | st_ops = map->st_ops; | |
1035 | st_ops->data = malloc(type->size); | |
1036 | st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs)); | |
1037 | st_ops->kern_func_off = malloc(btf_vlen(type) * | |
1038 | sizeof(*st_ops->kern_func_off)); | |
1039 | if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off) | |
1040 | return -ENOMEM; | |
1041 | ||
1042 | if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) { | |
1043 | pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n", | |
1044 | var_name, STRUCT_OPS_SEC); | |
1045 | return -EINVAL; | |
1046 | } | |
1047 | ||
1048 | memcpy(st_ops->data, | |
1049 | obj->efile.st_ops_data->d_buf + vsi->offset, | |
1050 | type->size); | |
1051 | st_ops->tname = tname; | |
1052 | st_ops->type = type; | |
1053 | st_ops->type_id = type_id; | |
1054 | ||
1055 | pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n", | |
1056 | tname, type_id, var_name, vsi->offset); | |
1057 | } | |
1058 | ||
1059 | return 0; | |
1060 | } | |
1061 | ||
6c956392 | 1062 | static struct bpf_object *bpf_object__new(const char *path, |
5e61f270 | 1063 | const void *obj_buf, |
2ce8450e AN |
1064 | size_t obj_buf_sz, |
1065 | const char *obj_name) | |
1a5e3fb1 WN |
1066 | { |
1067 | struct bpf_object *obj; | |
d859900c | 1068 | char *end; |
1a5e3fb1 WN |
1069 | |
1070 | obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1); | |
1071 | if (!obj) { | |
be18010e | 1072 | pr_warn("alloc memory failed for %s\n", path); |
6371ca3b | 1073 | return ERR_PTR(-ENOMEM); |
1a5e3fb1 WN |
1074 | } |
1075 | ||
1076 | strcpy(obj->path, path); | |
2ce8450e AN |
1077 | if (obj_name) { |
1078 | strncpy(obj->name, obj_name, sizeof(obj->name) - 1); | |
1079 | obj->name[sizeof(obj->name) - 1] = 0; | |
1080 | } else { | |
1081 | /* Using basename() GNU version which doesn't modify arg. */ | |
1082 | strncpy(obj->name, basename((void *)path), | |
1083 | sizeof(obj->name) - 1); | |
1084 | end = strchr(obj->name, '.'); | |
1085 | if (end) | |
1086 | *end = 0; | |
1087 | } | |
6c956392 | 1088 | |
d859900c | 1089 | obj->efile.fd = -1; |
6c956392 | 1090 | /* |
76e1022b | 1091 | * Caller of this function should also call |
6c956392 WN |
1092 | * bpf_object__elf_finish() after data collection to return |
1093 | * obj_buf to user. If not, we should duplicate the buffer to | |
1094 | * avoid user freeing them before elf finish. | |
1095 | */ | |
1096 | obj->efile.obj_buf = obj_buf; | |
1097 | obj->efile.obj_buf_sz = obj_buf_sz; | |
666810e8 | 1098 | obj->efile.maps_shndx = -1; |
abd29c93 | 1099 | obj->efile.btf_maps_shndx = -1; |
d859900c DB |
1100 | obj->efile.data_shndx = -1; |
1101 | obj->efile.rodata_shndx = -1; | |
1102 | obj->efile.bss_shndx = -1; | |
590a0088 | 1103 | obj->efile.st_ops_shndx = -1; |
81bfdd08 | 1104 | obj->kconfig_map_idx = -1; |
5d23328d | 1105 | obj->rodata_map_idx = -1; |
6c956392 | 1106 | |
5e61f270 | 1107 | obj->kern_version = get_kernel_version(); |
52d3352e | 1108 | obj->loaded = false; |
9a208eff WN |
1109 | |
1110 | INIT_LIST_HEAD(&obj->list); | |
1111 | list_add(&obj->list, &bpf_objects_list); | |
1a5e3fb1 WN |
1112 | return obj; |
1113 | } | |
1114 | ||
1115 | static void bpf_object__elf_finish(struct bpf_object *obj) | |
1116 | { | |
1117 | if (!obj_elf_valid(obj)) | |
1118 | return; | |
1119 | ||
1120 | if (obj->efile.elf) { | |
1121 | elf_end(obj->efile.elf); | |
1122 | obj->efile.elf = NULL; | |
1123 | } | |
bec7d68c | 1124 | obj->efile.symbols = NULL; |
d859900c DB |
1125 | obj->efile.data = NULL; |
1126 | obj->efile.rodata = NULL; | |
1127 | obj->efile.bss = NULL; | |
590a0088 | 1128 | obj->efile.st_ops_data = NULL; |
b62f06e8 | 1129 | |
1f8e2bcb AN |
1130 | zfree(&obj->efile.reloc_sects); |
1131 | obj->efile.nr_reloc_sects = 0; | |
1a5e3fb1 | 1132 | zclose(obj->efile.fd); |
6c956392 WN |
1133 | obj->efile.obj_buf = NULL; |
1134 | obj->efile.obj_buf_sz = 0; | |
1a5e3fb1 WN |
1135 | } |
1136 | ||
22dd1ac9 AN |
1137 | /* if libelf is old and doesn't support mmap(), fall back to read() */ |
1138 | #ifndef ELF_C_READ_MMAP | |
1139 | #define ELF_C_READ_MMAP ELF_C_READ | |
1140 | #endif | |
1141 | ||
1a5e3fb1 WN |
1142 | static int bpf_object__elf_init(struct bpf_object *obj) |
1143 | { | |
1144 | int err = 0; | |
1145 | GElf_Ehdr *ep; | |
1146 | ||
1147 | if (obj_elf_valid(obj)) { | |
88a82120 | 1148 | pr_warn("elf: init internal error\n"); |
6371ca3b | 1149 | return -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1150 | } |
1151 | ||
6c956392 WN |
1152 | if (obj->efile.obj_buf_sz > 0) { |
1153 | /* | |
1154 | * obj_buf should have been validated by | |
1155 | * bpf_object__open_buffer(). | |
1156 | */ | |
5e61f270 | 1157 | obj->efile.elf = elf_memory((char *)obj->efile.obj_buf, |
6c956392 WN |
1158 | obj->efile.obj_buf_sz); |
1159 | } else { | |
1160 | obj->efile.fd = open(obj->path, O_RDONLY); | |
1161 | if (obj->efile.fd < 0) { | |
be5c5d4e | 1162 | char errmsg[STRERR_BUFSIZE], *cp; |
1ce6a9fc | 1163 | |
be5c5d4e AN |
1164 | err = -errno; |
1165 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
88a82120 | 1166 | pr_warn("elf: failed to open %s: %s\n", obj->path, cp); |
be5c5d4e | 1167 | return err; |
6c956392 WN |
1168 | } |
1169 | ||
22dd1ac9 | 1170 | obj->efile.elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL); |
1a5e3fb1 WN |
1171 | } |
1172 | ||
1a5e3fb1 | 1173 | if (!obj->efile.elf) { |
88a82120 | 1174 | pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1175 | err = -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1176 | goto errout; |
1177 | } | |
1178 | ||
1179 | if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) { | |
88a82120 | 1180 | pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1181 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1182 | goto errout; |
1183 | } | |
1184 | ep = &obj->efile.ehdr; | |
1185 | ||
88a82120 AN |
1186 | if (elf_getshdrstrndx(obj->efile.elf, &obj->efile.shstrndx)) { |
1187 | pr_warn("elf: failed to get section names section index for %s: %s\n", | |
1188 | obj->path, elf_errmsg(-1)); | |
1189 | err = -LIBBPF_ERRNO__FORMAT; | |
1190 | goto errout; | |
1191 | } | |
1192 | ||
1193 | /* Elf is corrupted/truncated, avoid calling elf_strptr. */ | |
1194 | if (!elf_rawdata(elf_getscn(obj->efile.elf, obj->efile.shstrndx), NULL)) { | |
1195 | pr_warn("elf: failed to get section names strings from %s: %s\n", | |
1196 | obj->path, elf_errmsg(-1)); | |
1197 | return -LIBBPF_ERRNO__FORMAT; | |
1198 | } | |
1199 | ||
9b16137a | 1200 | /* Old LLVM set e_machine to EM_NONE */ |
76e1022b AN |
1201 | if (ep->e_type != ET_REL || |
1202 | (ep->e_machine && ep->e_machine != EM_BPF)) { | |
88a82120 | 1203 | pr_warn("elf: %s is not a valid eBPF object file\n", obj->path); |
6371ca3b | 1204 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1205 | goto errout; |
1206 | } | |
1207 | ||
1208 | return 0; | |
1209 | errout: | |
1210 | bpf_object__elf_finish(obj); | |
1211 | return err; | |
1212 | } | |
1213 | ||
12ef5634 | 1214 | static int bpf_object__check_endianness(struct bpf_object *obj) |
cc4228d5 | 1215 | { |
cdb2f920 | 1216 | #if __BYTE_ORDER == __LITTLE_ENDIAN |
12ef5634 AN |
1217 | if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB) |
1218 | return 0; | |
cdb2f920 | 1219 | #elif __BYTE_ORDER == __BIG_ENDIAN |
12ef5634 AN |
1220 | if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB) |
1221 | return 0; | |
1222 | #else | |
1223 | # error "Unrecognized __BYTE_ORDER__" | |
1224 | #endif | |
88a82120 | 1225 | pr_warn("elf: endianness mismatch in %s.\n", obj->path); |
6371ca3b | 1226 | return -LIBBPF_ERRNO__ENDIAN; |
cc4228d5 WN |
1227 | } |
1228 | ||
cb1e5e96 | 1229 | static int |
399dc65e | 1230 | bpf_object__init_license(struct bpf_object *obj, void *data, size_t size) |
cb1e5e96 | 1231 | { |
399dc65e | 1232 | memcpy(obj->license, data, min(size, sizeof(obj->license) - 1)); |
cb1e5e96 WN |
1233 | pr_debug("license of %s is %s\n", obj->path, obj->license); |
1234 | return 0; | |
1235 | } | |
1236 | ||
54b8625c JF |
1237 | static int |
1238 | bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size) | |
1239 | { | |
1240 | __u32 kver; | |
1241 | ||
1242 | if (size != sizeof(kver)) { | |
be18010e | 1243 | pr_warn("invalid kver section in %s\n", obj->path); |
54b8625c JF |
1244 | return -LIBBPF_ERRNO__FORMAT; |
1245 | } | |
1246 | memcpy(&kver, data, sizeof(kver)); | |
1247 | obj->kern_version = kver; | |
1248 | pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version); | |
1249 | return 0; | |
1250 | } | |
1251 | ||
addb9fc9 NS |
1252 | static bool bpf_map_type__is_map_in_map(enum bpf_map_type type) |
1253 | { | |
1254 | if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1255 | type == BPF_MAP_TYPE_HASH_OF_MAPS) | |
1256 | return true; | |
1257 | return false; | |
1258 | } | |
1259 | ||
1713d68b DB |
1260 | int bpf_object__section_size(const struct bpf_object *obj, const char *name, |
1261 | __u32 *size) | |
1262 | { | |
1263 | int ret = -ENOENT; | |
1713d68b DB |
1264 | |
1265 | *size = 0; | |
1266 | if (!name) { | |
1267 | return -EINVAL; | |
ac9d1389 | 1268 | } else if (!strcmp(name, DATA_SEC)) { |
1713d68b DB |
1269 | if (obj->efile.data) |
1270 | *size = obj->efile.data->d_size; | |
ac9d1389 | 1271 | } else if (!strcmp(name, BSS_SEC)) { |
1713d68b DB |
1272 | if (obj->efile.bss) |
1273 | *size = obj->efile.bss->d_size; | |
ac9d1389 | 1274 | } else if (!strcmp(name, RODATA_SEC)) { |
1713d68b DB |
1275 | if (obj->efile.rodata) |
1276 | *size = obj->efile.rodata->d_size; | |
590a0088 MKL |
1277 | } else if (!strcmp(name, STRUCT_OPS_SEC)) { |
1278 | if (obj->efile.st_ops_data) | |
1279 | *size = obj->efile.st_ops_data->d_size; | |
1713d68b | 1280 | } else { |
88a82120 AN |
1281 | Elf_Scn *scn = elf_sec_by_name(obj, name); |
1282 | Elf_Data *data = elf_sec_data(obj, scn); | |
1283 | ||
1284 | if (data) { | |
1285 | ret = 0; /* found it */ | |
1286 | *size = data->d_size; | |
1287 | } | |
1713d68b DB |
1288 | } |
1289 | ||
1290 | return *size ? 0 : ret; | |
1291 | } | |
1292 | ||
1293 | int bpf_object__variable_offset(const struct bpf_object *obj, const char *name, | |
1294 | __u32 *off) | |
1295 | { | |
1296 | Elf_Data *symbols = obj->efile.symbols; | |
1297 | const char *sname; | |
1298 | size_t si; | |
1299 | ||
1300 | if (!name || !off) | |
1301 | return -EINVAL; | |
1302 | ||
1303 | for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) { | |
1304 | GElf_Sym sym; | |
1305 | ||
1306 | if (!gelf_getsym(symbols, si, &sym)) | |
1307 | continue; | |
1308 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || | |
1309 | GELF_ST_TYPE(sym.st_info) != STT_OBJECT) | |
1310 | continue; | |
1311 | ||
88a82120 | 1312 | sname = elf_sym_str(obj, sym.st_name); |
1713d68b | 1313 | if (!sname) { |
be18010e KW |
1314 | pr_warn("failed to get sym name string for var %s\n", |
1315 | name); | |
1713d68b DB |
1316 | return -EIO; |
1317 | } | |
1318 | if (strcmp(name, sname) == 0) { | |
1319 | *off = sym.st_value; | |
1320 | return 0; | |
1321 | } | |
1322 | } | |
1323 | ||
1324 | return -ENOENT; | |
1325 | } | |
1326 | ||
bf829271 | 1327 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj) |
d859900c | 1328 | { |
bf829271 AN |
1329 | struct bpf_map *new_maps; |
1330 | size_t new_cap; | |
1331 | int i; | |
1332 | ||
1333 | if (obj->nr_maps < obj->maps_cap) | |
1334 | return &obj->maps[obj->nr_maps++]; | |
1335 | ||
95064979 | 1336 | new_cap = max((size_t)4, obj->maps_cap * 3 / 2); |
029258d7 | 1337 | new_maps = libbpf_reallocarray(obj->maps, new_cap, sizeof(*obj->maps)); |
bf829271 | 1338 | if (!new_maps) { |
be18010e | 1339 | pr_warn("alloc maps for object failed\n"); |
bf829271 AN |
1340 | return ERR_PTR(-ENOMEM); |
1341 | } | |
1342 | ||
1343 | obj->maps_cap = new_cap; | |
1344 | obj->maps = new_maps; | |
1345 | ||
1346 | /* zero out new maps */ | |
1347 | memset(obj->maps + obj->nr_maps, 0, | |
1348 | (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps)); | |
1349 | /* | |
1350 | * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin) | |
1351 | * when failure (zclose won't close negative fd)). | |
1352 | */ | |
1353 | for (i = obj->nr_maps; i < obj->maps_cap; i++) { | |
1354 | obj->maps[i].fd = -1; | |
1355 | obj->maps[i].inner_map_fd = -1; | |
1356 | } | |
1357 | ||
1358 | return &obj->maps[obj->nr_maps++]; | |
d859900c DB |
1359 | } |
1360 | ||
eba9c5f4 AN |
1361 | static size_t bpf_map_mmap_sz(const struct bpf_map *map) |
1362 | { | |
1363 | long page_sz = sysconf(_SC_PAGE_SIZE); | |
1364 | size_t map_sz; | |
1365 | ||
c701917e | 1366 | map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries; |
eba9c5f4 AN |
1367 | map_sz = roundup(map_sz, page_sz); |
1368 | return map_sz; | |
1369 | } | |
1370 | ||
81bfdd08 AN |
1371 | static char *internal_map_name(struct bpf_object *obj, |
1372 | enum libbpf_map_type type) | |
1373 | { | |
113e6b7e | 1374 | char map_name[BPF_OBJ_NAME_LEN], *p; |
81bfdd08 AN |
1375 | const char *sfx = libbpf_type_to_btf_name[type]; |
1376 | int sfx_len = max((size_t)7, strlen(sfx)); | |
1377 | int pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, | |
1378 | strlen(obj->name)); | |
1379 | ||
1380 | snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name, | |
1381 | sfx_len, libbpf_type_to_btf_name[type]); | |
1382 | ||
113e6b7e THJ |
1383 | /* sanitise map name to characters allowed by kernel */ |
1384 | for (p = map_name; *p && p < map_name + sizeof(map_name); p++) | |
1385 | if (!isalnum(*p) && *p != '_' && *p != '.') | |
1386 | *p = '_'; | |
1387 | ||
81bfdd08 AN |
1388 | return strdup(map_name); |
1389 | } | |
1390 | ||
d859900c | 1391 | static int |
bf829271 | 1392 | bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type, |
eba9c5f4 | 1393 | int sec_idx, void *data, size_t data_sz) |
d859900c | 1394 | { |
bf829271 AN |
1395 | struct bpf_map_def *def; |
1396 | struct bpf_map *map; | |
eba9c5f4 | 1397 | int err; |
bf829271 AN |
1398 | |
1399 | map = bpf_object__add_map(obj); | |
1400 | if (IS_ERR(map)) | |
1401 | return PTR_ERR(map); | |
d859900c DB |
1402 | |
1403 | map->libbpf_type = type; | |
db48814b AN |
1404 | map->sec_idx = sec_idx; |
1405 | map->sec_offset = 0; | |
81bfdd08 | 1406 | map->name = internal_map_name(obj, type); |
d859900c | 1407 | if (!map->name) { |
be18010e | 1408 | pr_warn("failed to alloc map name\n"); |
d859900c DB |
1409 | return -ENOMEM; |
1410 | } | |
1411 | ||
bf829271 | 1412 | def = &map->def; |
d859900c DB |
1413 | def->type = BPF_MAP_TYPE_ARRAY; |
1414 | def->key_size = sizeof(int); | |
eba9c5f4 | 1415 | def->value_size = data_sz; |
d859900c | 1416 | def->max_entries = 1; |
81bfdd08 | 1417 | def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG |
166750bc | 1418 | ? BPF_F_RDONLY_PROG : 0; |
0d13bfce | 1419 | def->map_flags |= BPF_F_MMAPABLE; |
7fe74b43 AN |
1420 | |
1421 | pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n", | |
81bfdd08 | 1422 | map->name, map->sec_idx, map->sec_offset, def->map_flags); |
7fe74b43 | 1423 | |
eba9c5f4 AN |
1424 | map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE, |
1425 | MAP_SHARED | MAP_ANONYMOUS, -1, 0); | |
1426 | if (map->mmaped == MAP_FAILED) { | |
1427 | err = -errno; | |
1428 | map->mmaped = NULL; | |
1429 | pr_warn("failed to alloc map '%s' content buffer: %d\n", | |
1430 | map->name, err); | |
1431 | zfree(&map->name); | |
1432 | return err; | |
d859900c DB |
1433 | } |
1434 | ||
166750bc | 1435 | if (data) |
eba9c5f4 AN |
1436 | memcpy(map->mmaped, data, data_sz); |
1437 | ||
e1d1dc46 | 1438 | pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name); |
d859900c DB |
1439 | return 0; |
1440 | } | |
1441 | ||
bf829271 AN |
1442 | static int bpf_object__init_global_data_maps(struct bpf_object *obj) |
1443 | { | |
1444 | int err; | |
1445 | ||
bf829271 AN |
1446 | /* |
1447 | * Populate obj->maps with libbpf internal maps. | |
1448 | */ | |
1449 | if (obj->efile.data_shndx >= 0) { | |
1450 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA, | |
db48814b | 1451 | obj->efile.data_shndx, |
eba9c5f4 AN |
1452 | obj->efile.data->d_buf, |
1453 | obj->efile.data->d_size); | |
bf829271 AN |
1454 | if (err) |
1455 | return err; | |
1456 | } | |
1457 | if (obj->efile.rodata_shndx >= 0) { | |
1458 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA, | |
db48814b | 1459 | obj->efile.rodata_shndx, |
eba9c5f4 AN |
1460 | obj->efile.rodata->d_buf, |
1461 | obj->efile.rodata->d_size); | |
bf829271 AN |
1462 | if (err) |
1463 | return err; | |
5d23328d YZ |
1464 | |
1465 | obj->rodata_map_idx = obj->nr_maps - 1; | |
bf829271 AN |
1466 | } |
1467 | if (obj->efile.bss_shndx >= 0) { | |
1468 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS, | |
db48814b | 1469 | obj->efile.bss_shndx, |
eba9c5f4 AN |
1470 | NULL, |
1471 | obj->efile.bss->d_size); | |
bf829271 AN |
1472 | if (err) |
1473 | return err; | |
1474 | } | |
1475 | return 0; | |
1476 | } | |
1477 | ||
166750bc AN |
1478 | |
1479 | static struct extern_desc *find_extern_by_name(const struct bpf_object *obj, | |
1480 | const void *name) | |
1481 | { | |
1482 | int i; | |
1483 | ||
1484 | for (i = 0; i < obj->nr_extern; i++) { | |
1485 | if (strcmp(obj->externs[i].name, name) == 0) | |
1486 | return &obj->externs[i]; | |
1487 | } | |
1488 | return NULL; | |
1489 | } | |
1490 | ||
2e33efe3 AN |
1491 | static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val, |
1492 | char value) | |
166750bc | 1493 | { |
2e33efe3 AN |
1494 | switch (ext->kcfg.type) { |
1495 | case KCFG_BOOL: | |
166750bc | 1496 | if (value == 'm') { |
2e33efe3 | 1497 | pr_warn("extern (kcfg) %s=%c should be tristate or char\n", |
166750bc AN |
1498 | ext->name, value); |
1499 | return -EINVAL; | |
1500 | } | |
1501 | *(bool *)ext_val = value == 'y' ? true : false; | |
1502 | break; | |
2e33efe3 | 1503 | case KCFG_TRISTATE: |
166750bc AN |
1504 | if (value == 'y') |
1505 | *(enum libbpf_tristate *)ext_val = TRI_YES; | |
1506 | else if (value == 'm') | |
1507 | *(enum libbpf_tristate *)ext_val = TRI_MODULE; | |
1508 | else /* value == 'n' */ | |
1509 | *(enum libbpf_tristate *)ext_val = TRI_NO; | |
1510 | break; | |
2e33efe3 | 1511 | case KCFG_CHAR: |
166750bc AN |
1512 | *(char *)ext_val = value; |
1513 | break; | |
2e33efe3 AN |
1514 | case KCFG_UNKNOWN: |
1515 | case KCFG_INT: | |
1516 | case KCFG_CHAR_ARR: | |
166750bc | 1517 | default: |
2e33efe3 | 1518 | pr_warn("extern (kcfg) %s=%c should be bool, tristate, or char\n", |
166750bc AN |
1519 | ext->name, value); |
1520 | return -EINVAL; | |
1521 | } | |
1522 | ext->is_set = true; | |
1523 | return 0; | |
1524 | } | |
1525 | ||
2e33efe3 AN |
1526 | static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val, |
1527 | const char *value) | |
166750bc AN |
1528 | { |
1529 | size_t len; | |
1530 | ||
2e33efe3 AN |
1531 | if (ext->kcfg.type != KCFG_CHAR_ARR) { |
1532 | pr_warn("extern (kcfg) %s=%s should be char array\n", ext->name, value); | |
166750bc AN |
1533 | return -EINVAL; |
1534 | } | |
1535 | ||
1536 | len = strlen(value); | |
1537 | if (value[len - 1] != '"') { | |
2e33efe3 | 1538 | pr_warn("extern (kcfg) '%s': invalid string config '%s'\n", |
166750bc AN |
1539 | ext->name, value); |
1540 | return -EINVAL; | |
1541 | } | |
1542 | ||
1543 | /* strip quotes */ | |
1544 | len -= 2; | |
2e33efe3 AN |
1545 | if (len >= ext->kcfg.sz) { |
1546 | pr_warn("extern (kcfg) '%s': long string config %s of (%zu bytes) truncated to %d bytes\n", | |
1547 | ext->name, value, len, ext->kcfg.sz - 1); | |
1548 | len = ext->kcfg.sz - 1; | |
166750bc AN |
1549 | } |
1550 | memcpy(ext_val, value + 1, len); | |
1551 | ext_val[len] = '\0'; | |
1552 | ext->is_set = true; | |
1553 | return 0; | |
1554 | } | |
1555 | ||
1556 | static int parse_u64(const char *value, __u64 *res) | |
1557 | { | |
1558 | char *value_end; | |
1559 | int err; | |
1560 | ||
1561 | errno = 0; | |
1562 | *res = strtoull(value, &value_end, 0); | |
1563 | if (errno) { | |
1564 | err = -errno; | |
1565 | pr_warn("failed to parse '%s' as integer: %d\n", value, err); | |
1566 | return err; | |
1567 | } | |
1568 | if (*value_end) { | |
1569 | pr_warn("failed to parse '%s' as integer completely\n", value); | |
1570 | return -EINVAL; | |
1571 | } | |
1572 | return 0; | |
1573 | } | |
1574 | ||
2e33efe3 | 1575 | static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v) |
166750bc | 1576 | { |
2e33efe3 | 1577 | int bit_sz = ext->kcfg.sz * 8; |
166750bc | 1578 | |
2e33efe3 | 1579 | if (ext->kcfg.sz == 8) |
166750bc AN |
1580 | return true; |
1581 | ||
1582 | /* Validate that value stored in u64 fits in integer of `ext->sz` | |
1583 | * bytes size without any loss of information. If the target integer | |
1584 | * is signed, we rely on the following limits of integer type of | |
1585 | * Y bits and subsequent transformation: | |
1586 | * | |
1587 | * -2^(Y-1) <= X <= 2^(Y-1) - 1 | |
1588 | * 0 <= X + 2^(Y-1) <= 2^Y - 1 | |
1589 | * 0 <= X + 2^(Y-1) < 2^Y | |
1590 | * | |
1591 | * For unsigned target integer, check that all the (64 - Y) bits are | |
1592 | * zero. | |
1593 | */ | |
2e33efe3 | 1594 | if (ext->kcfg.is_signed) |
166750bc AN |
1595 | return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz); |
1596 | else | |
1597 | return (v >> bit_sz) == 0; | |
1598 | } | |
1599 | ||
2e33efe3 AN |
1600 | static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val, |
1601 | __u64 value) | |
166750bc | 1602 | { |
2e33efe3 AN |
1603 | if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) { |
1604 | pr_warn("extern (kcfg) %s=%llu should be integer\n", | |
7745ff98 | 1605 | ext->name, (unsigned long long)value); |
166750bc AN |
1606 | return -EINVAL; |
1607 | } | |
2e33efe3 AN |
1608 | if (!is_kcfg_value_in_range(ext, value)) { |
1609 | pr_warn("extern (kcfg) %s=%llu value doesn't fit in %d bytes\n", | |
1610 | ext->name, (unsigned long long)value, ext->kcfg.sz); | |
166750bc AN |
1611 | return -ERANGE; |
1612 | } | |
2e33efe3 | 1613 | switch (ext->kcfg.sz) { |
166750bc AN |
1614 | case 1: *(__u8 *)ext_val = value; break; |
1615 | case 2: *(__u16 *)ext_val = value; break; | |
1616 | case 4: *(__u32 *)ext_val = value; break; | |
1617 | case 8: *(__u64 *)ext_val = value; break; | |
1618 | default: | |
1619 | return -EINVAL; | |
1620 | } | |
1621 | ext->is_set = true; | |
1622 | return 0; | |
1623 | } | |
1624 | ||
8601fd42 AN |
1625 | static int bpf_object__process_kconfig_line(struct bpf_object *obj, |
1626 | char *buf, void *data) | |
166750bc | 1627 | { |
166750bc | 1628 | struct extern_desc *ext; |
8601fd42 | 1629 | char *sep, *value; |
166750bc AN |
1630 | int len, err = 0; |
1631 | void *ext_val; | |
1632 | __u64 num; | |
166750bc | 1633 | |
8601fd42 AN |
1634 | if (strncmp(buf, "CONFIG_", 7)) |
1635 | return 0; | |
166750bc | 1636 | |
8601fd42 AN |
1637 | sep = strchr(buf, '='); |
1638 | if (!sep) { | |
1639 | pr_warn("failed to parse '%s': no separator\n", buf); | |
1640 | return -EINVAL; | |
1641 | } | |
1642 | ||
1643 | /* Trim ending '\n' */ | |
1644 | len = strlen(buf); | |
1645 | if (buf[len - 1] == '\n') | |
1646 | buf[len - 1] = '\0'; | |
1647 | /* Split on '=' and ensure that a value is present. */ | |
1648 | *sep = '\0'; | |
1649 | if (!sep[1]) { | |
1650 | *sep = '='; | |
1651 | pr_warn("failed to parse '%s': no value\n", buf); | |
1652 | return -EINVAL; | |
1653 | } | |
1654 | ||
1655 | ext = find_extern_by_name(obj, buf); | |
1656 | if (!ext || ext->is_set) | |
1657 | return 0; | |
1658 | ||
2e33efe3 | 1659 | ext_val = data + ext->kcfg.data_off; |
8601fd42 AN |
1660 | value = sep + 1; |
1661 | ||
1662 | switch (*value) { | |
1663 | case 'y': case 'n': case 'm': | |
2e33efe3 | 1664 | err = set_kcfg_value_tri(ext, ext_val, *value); |
8601fd42 AN |
1665 | break; |
1666 | case '"': | |
2e33efe3 | 1667 | err = set_kcfg_value_str(ext, ext_val, value); |
8601fd42 AN |
1668 | break; |
1669 | default: | |
1670 | /* assume integer */ | |
1671 | err = parse_u64(value, &num); | |
1672 | if (err) { | |
2e33efe3 | 1673 | pr_warn("extern (kcfg) %s=%s should be integer\n", |
8601fd42 AN |
1674 | ext->name, value); |
1675 | return err; | |
1676 | } | |
2e33efe3 | 1677 | err = set_kcfg_value_num(ext, ext_val, num); |
8601fd42 | 1678 | break; |
166750bc | 1679 | } |
8601fd42 AN |
1680 | if (err) |
1681 | return err; | |
2e33efe3 | 1682 | pr_debug("extern (kcfg) %s=%s\n", ext->name, value); |
8601fd42 AN |
1683 | return 0; |
1684 | } | |
1685 | ||
1686 | static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data) | |
1687 | { | |
1688 | char buf[PATH_MAX]; | |
1689 | struct utsname uts; | |
1690 | int len, err = 0; | |
1691 | gzFile file; | |
1692 | ||
1693 | uname(&uts); | |
1694 | len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release); | |
1695 | if (len < 0) | |
1696 | return -EINVAL; | |
1697 | else if (len >= PATH_MAX) | |
1698 | return -ENAMETOOLONG; | |
1699 | ||
1700 | /* gzopen also accepts uncompressed files. */ | |
1701 | file = gzopen(buf, "r"); | |
1702 | if (!file) | |
1703 | file = gzopen("/proc/config.gz", "r"); | |
1704 | ||
166750bc | 1705 | if (!file) { |
8601fd42 | 1706 | pr_warn("failed to open system Kconfig\n"); |
166750bc AN |
1707 | return -ENOENT; |
1708 | } | |
1709 | ||
1710 | while (gzgets(file, buf, sizeof(buf))) { | |
8601fd42 AN |
1711 | err = bpf_object__process_kconfig_line(obj, buf, data); |
1712 | if (err) { | |
1713 | pr_warn("error parsing system Kconfig line '%s': %d\n", | |
1714 | buf, err); | |
166750bc AN |
1715 | goto out; |
1716 | } | |
8601fd42 | 1717 | } |
166750bc | 1718 | |
8601fd42 AN |
1719 | out: |
1720 | gzclose(file); | |
1721 | return err; | |
1722 | } | |
166750bc | 1723 | |
8601fd42 AN |
1724 | static int bpf_object__read_kconfig_mem(struct bpf_object *obj, |
1725 | const char *config, void *data) | |
1726 | { | |
1727 | char buf[PATH_MAX]; | |
1728 | int err = 0; | |
1729 | FILE *file; | |
166750bc | 1730 | |
8601fd42 AN |
1731 | file = fmemopen((void *)config, strlen(config), "r"); |
1732 | if (!file) { | |
1733 | err = -errno; | |
1734 | pr_warn("failed to open in-memory Kconfig: %d\n", err); | |
1735 | return err; | |
1736 | } | |
1737 | ||
1738 | while (fgets(buf, sizeof(buf), file)) { | |
1739 | err = bpf_object__process_kconfig_line(obj, buf, data); | |
1740 | if (err) { | |
1741 | pr_warn("error parsing in-memory Kconfig line '%s': %d\n", | |
1742 | buf, err); | |
166750bc AN |
1743 | break; |
1744 | } | |
166750bc AN |
1745 | } |
1746 | ||
8601fd42 | 1747 | fclose(file); |
166750bc AN |
1748 | return err; |
1749 | } | |
1750 | ||
81bfdd08 | 1751 | static int bpf_object__init_kconfig_map(struct bpf_object *obj) |
166750bc | 1752 | { |
2e33efe3 | 1753 | struct extern_desc *last_ext = NULL, *ext; |
166750bc | 1754 | size_t map_sz; |
2e33efe3 | 1755 | int i, err; |
166750bc | 1756 | |
2e33efe3 AN |
1757 | for (i = 0; i < obj->nr_extern; i++) { |
1758 | ext = &obj->externs[i]; | |
1759 | if (ext->type == EXT_KCFG) | |
1760 | last_ext = ext; | |
1761 | } | |
166750bc | 1762 | |
2e33efe3 AN |
1763 | if (!last_ext) |
1764 | return 0; | |
166750bc | 1765 | |
2e33efe3 | 1766 | map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz; |
81bfdd08 | 1767 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG, |
166750bc AN |
1768 | obj->efile.symbols_shndx, |
1769 | NULL, map_sz); | |
1770 | if (err) | |
1771 | return err; | |
1772 | ||
81bfdd08 | 1773 | obj->kconfig_map_idx = obj->nr_maps - 1; |
166750bc AN |
1774 | |
1775 | return 0; | |
1776 | } | |
1777 | ||
bf829271 | 1778 | static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict) |
561bbcca | 1779 | { |
561bbcca | 1780 | Elf_Data *symbols = obj->efile.symbols; |
bf829271 | 1781 | int i, map_def_sz = 0, nr_maps = 0, nr_syms; |
d859900c | 1782 | Elf_Data *data = NULL; |
bf829271 AN |
1783 | Elf_Scn *scn; |
1784 | ||
1785 | if (obj->efile.maps_shndx < 0) | |
1786 | return 0; | |
561bbcca | 1787 | |
4708bbda EL |
1788 | if (!symbols) |
1789 | return -EINVAL; | |
1790 | ||
88a82120 AN |
1791 | |
1792 | scn = elf_sec_by_idx(obj, obj->efile.maps_shndx); | |
1793 | data = elf_sec_data(obj, scn); | |
bf829271 | 1794 | if (!scn || !data) { |
88a82120 AN |
1795 | pr_warn("elf: failed to get legacy map definitions for %s\n", |
1796 | obj->path); | |
bf829271 | 1797 | return -EINVAL; |
4708bbda | 1798 | } |
561bbcca | 1799 | |
4708bbda EL |
1800 | /* |
1801 | * Count number of maps. Each map has a name. | |
1802 | * Array of maps is not supported: only the first element is | |
1803 | * considered. | |
1804 | * | |
1805 | * TODO: Detect array of map and report error. | |
1806 | */ | |
bf829271 AN |
1807 | nr_syms = symbols->d_size / sizeof(GElf_Sym); |
1808 | for (i = 0; i < nr_syms; i++) { | |
561bbcca | 1809 | GElf_Sym sym; |
4708bbda EL |
1810 | |
1811 | if (!gelf_getsym(symbols, i, &sym)) | |
1812 | continue; | |
1813 | if (sym.st_shndx != obj->efile.maps_shndx) | |
1814 | continue; | |
1815 | nr_maps++; | |
1816 | } | |
b13c5c14 | 1817 | /* Assume equally sized map definitions */ |
88a82120 AN |
1818 | pr_debug("elf: found %d legacy map definitions (%zd bytes) in %s\n", |
1819 | nr_maps, data->d_size, obj->path); | |
bf829271 | 1820 | |
98e527af | 1821 | if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) { |
88a82120 AN |
1822 | pr_warn("elf: unable to determine legacy map definition size in %s\n", |
1823 | obj->path); | |
bf829271 | 1824 | return -EINVAL; |
addb9fc9 | 1825 | } |
98e527af | 1826 | map_def_sz = data->d_size / nr_maps; |
4708bbda | 1827 | |
bf829271 AN |
1828 | /* Fill obj->maps using data in "maps" section. */ |
1829 | for (i = 0; i < nr_syms; i++) { | |
4708bbda | 1830 | GElf_Sym sym; |
561bbcca | 1831 | const char *map_name; |
4708bbda | 1832 | struct bpf_map_def *def; |
bf829271 | 1833 | struct bpf_map *map; |
561bbcca WN |
1834 | |
1835 | if (!gelf_getsym(symbols, i, &sym)) | |
1836 | continue; | |
666810e8 | 1837 | if (sym.st_shndx != obj->efile.maps_shndx) |
561bbcca WN |
1838 | continue; |
1839 | ||
bf829271 AN |
1840 | map = bpf_object__add_map(obj); |
1841 | if (IS_ERR(map)) | |
1842 | return PTR_ERR(map); | |
1843 | ||
88a82120 | 1844 | map_name = elf_sym_str(obj, sym.st_name); |
c51829bb | 1845 | if (!map_name) { |
be18010e KW |
1846 | pr_warn("failed to get map #%d name sym string for obj %s\n", |
1847 | i, obj->path); | |
c51829bb AN |
1848 | return -LIBBPF_ERRNO__FORMAT; |
1849 | } | |
d859900c | 1850 | |
bf829271 | 1851 | map->libbpf_type = LIBBPF_MAP_UNSPEC; |
db48814b AN |
1852 | map->sec_idx = sym.st_shndx; |
1853 | map->sec_offset = sym.st_value; | |
1854 | pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n", | |
1855 | map_name, map->sec_idx, map->sec_offset); | |
b13c5c14 | 1856 | if (sym.st_value + map_def_sz > data->d_size) { |
be18010e KW |
1857 | pr_warn("corrupted maps section in %s: last map \"%s\" too small\n", |
1858 | obj->path, map_name); | |
4708bbda | 1859 | return -EINVAL; |
561bbcca | 1860 | } |
4708bbda | 1861 | |
bf829271 AN |
1862 | map->name = strdup(map_name); |
1863 | if (!map->name) { | |
be18010e | 1864 | pr_warn("failed to alloc map name\n"); |
973170e6 WN |
1865 | return -ENOMEM; |
1866 | } | |
bf829271 | 1867 | pr_debug("map %d is \"%s\"\n", i, map->name); |
4708bbda | 1868 | def = (struct bpf_map_def *)(data->d_buf + sym.st_value); |
b13c5c14 CG |
1869 | /* |
1870 | * If the definition of the map in the object file fits in | |
1871 | * bpf_map_def, copy it. Any extra fields in our version | |
1872 | * of bpf_map_def will default to zero as a result of the | |
1873 | * calloc above. | |
1874 | */ | |
1875 | if (map_def_sz <= sizeof(struct bpf_map_def)) { | |
bf829271 | 1876 | memcpy(&map->def, def, map_def_sz); |
b13c5c14 CG |
1877 | } else { |
1878 | /* | |
1879 | * Here the map structure being read is bigger than what | |
1880 | * we expect, truncate if the excess bits are all zero. | |
1881 | * If they are not zero, reject this map as | |
1882 | * incompatible. | |
1883 | */ | |
1884 | char *b; | |
8983b731 | 1885 | |
b13c5c14 CG |
1886 | for (b = ((char *)def) + sizeof(struct bpf_map_def); |
1887 | b < ((char *)def) + map_def_sz; b++) { | |
1888 | if (*b != 0) { | |
8983b731 | 1889 | pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n", |
be18010e | 1890 | obj->path, map_name); |
c034a177 JF |
1891 | if (strict) |
1892 | return -EINVAL; | |
b13c5c14 CG |
1893 | } |
1894 | } | |
bf829271 | 1895 | memcpy(&map->def, def, sizeof(struct bpf_map_def)); |
b13c5c14 | 1896 | } |
561bbcca | 1897 | } |
bf829271 AN |
1898 | return 0; |
1899 | } | |
4708bbda | 1900 | |
ddc7c304 AN |
1901 | static const struct btf_type * |
1902 | skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id) | |
abd29c93 AN |
1903 | { |
1904 | const struct btf_type *t = btf__type_by_id(btf, id); | |
8837fe5d | 1905 | |
ddc7c304 AN |
1906 | if (res_id) |
1907 | *res_id = id; | |
1908 | ||
1909 | while (btf_is_mod(t) || btf_is_typedef(t)) { | |
1910 | if (res_id) | |
1911 | *res_id = t->type; | |
1912 | t = btf__type_by_id(btf, t->type); | |
abd29c93 | 1913 | } |
ddc7c304 AN |
1914 | |
1915 | return t; | |
abd29c93 AN |
1916 | } |
1917 | ||
590a0088 MKL |
1918 | static const struct btf_type * |
1919 | resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id) | |
1920 | { | |
1921 | const struct btf_type *t; | |
1922 | ||
1923 | t = skip_mods_and_typedefs(btf, id, NULL); | |
1924 | if (!btf_is_ptr(t)) | |
1925 | return NULL; | |
1926 | ||
1927 | t = skip_mods_and_typedefs(btf, t->type, res_id); | |
1928 | ||
1929 | return btf_is_func_proto(t) ? t : NULL; | |
1930 | } | |
1931 | ||
81ba0889 AN |
1932 | static const char *btf_kind_str(const struct btf_type *t) |
1933 | { | |
1934 | switch (btf_kind(t)) { | |
1935 | case BTF_KIND_UNKN: return "void"; | |
1936 | case BTF_KIND_INT: return "int"; | |
1937 | case BTF_KIND_PTR: return "ptr"; | |
1938 | case BTF_KIND_ARRAY: return "array"; | |
1939 | case BTF_KIND_STRUCT: return "struct"; | |
1940 | case BTF_KIND_UNION: return "union"; | |
1941 | case BTF_KIND_ENUM: return "enum"; | |
1942 | case BTF_KIND_FWD: return "fwd"; | |
1943 | case BTF_KIND_TYPEDEF: return "typedef"; | |
1944 | case BTF_KIND_VOLATILE: return "volatile"; | |
1945 | case BTF_KIND_CONST: return "const"; | |
1946 | case BTF_KIND_RESTRICT: return "restrict"; | |
1947 | case BTF_KIND_FUNC: return "func"; | |
1948 | case BTF_KIND_FUNC_PROTO: return "func_proto"; | |
1949 | case BTF_KIND_VAR: return "var"; | |
1950 | case BTF_KIND_DATASEC: return "datasec"; | |
22541a9e | 1951 | case BTF_KIND_FLOAT: return "float"; |
81ba0889 AN |
1952 | default: return "unknown"; |
1953 | } | |
1954 | } | |
1955 | ||
ef99b02b AN |
1956 | /* |
1957 | * Fetch integer attribute of BTF map definition. Such attributes are | |
1958 | * represented using a pointer to an array, in which dimensionality of array | |
1959 | * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY]; | |
1960 | * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF | |
1961 | * type definition, while using only sizeof(void *) space in ELF data section. | |
1962 | */ | |
1963 | static bool get_map_field_int(const char *map_name, const struct btf *btf, | |
8983b731 AN |
1964 | const struct btf_member *m, __u32 *res) |
1965 | { | |
ddc7c304 | 1966 | const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL); |
abd29c93 | 1967 | const char *name = btf__name_by_offset(btf, m->name_off); |
ef99b02b AN |
1968 | const struct btf_array *arr_info; |
1969 | const struct btf_type *arr_t; | |
abd29c93 | 1970 | |
b03bc685 | 1971 | if (!btf_is_ptr(t)) { |
81ba0889 AN |
1972 | pr_warn("map '%s': attr '%s': expected PTR, got %s.\n", |
1973 | map_name, name, btf_kind_str(t)); | |
abd29c93 AN |
1974 | return false; |
1975 | } | |
ef99b02b AN |
1976 | |
1977 | arr_t = btf__type_by_id(btf, t->type); | |
1978 | if (!arr_t) { | |
be18010e KW |
1979 | pr_warn("map '%s': attr '%s': type [%u] not found.\n", |
1980 | map_name, name, t->type); | |
abd29c93 AN |
1981 | return false; |
1982 | } | |
b03bc685 | 1983 | if (!btf_is_array(arr_t)) { |
81ba0889 AN |
1984 | pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n", |
1985 | map_name, name, btf_kind_str(arr_t)); | |
abd29c93 AN |
1986 | return false; |
1987 | } | |
b03bc685 | 1988 | arr_info = btf_array(arr_t); |
ef99b02b | 1989 | *res = arr_info->nelems; |
abd29c93 AN |
1990 | return true; |
1991 | } | |
1992 | ||
57a00f41 THJ |
1993 | static int build_map_pin_path(struct bpf_map *map, const char *path) |
1994 | { | |
1995 | char buf[PATH_MAX]; | |
6e9cab2e | 1996 | int len; |
57a00f41 THJ |
1997 | |
1998 | if (!path) | |
1999 | path = "/sys/fs/bpf"; | |
2000 | ||
2001 | len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map)); | |
2002 | if (len < 0) | |
2003 | return -EINVAL; | |
2004 | else if (len >= PATH_MAX) | |
2005 | return -ENAMETOOLONG; | |
2006 | ||
6e9cab2e | 2007 | return bpf_map__set_pin_path(map, buf); |
57a00f41 THJ |
2008 | } |
2009 | ||
41017e56 AN |
2010 | |
2011 | static int parse_btf_map_def(struct bpf_object *obj, | |
2012 | struct bpf_map *map, | |
2013 | const struct btf_type *def, | |
646f02ff | 2014 | bool strict, bool is_inner, |
41017e56 | 2015 | const char *pin_root_path) |
abd29c93 | 2016 | { |
41017e56 | 2017 | const struct btf_type *t; |
abd29c93 | 2018 | const struct btf_member *m; |
abd29c93 AN |
2019 | int vlen, i; |
2020 | ||
b03bc685 AN |
2021 | vlen = btf_vlen(def); |
2022 | m = btf_members(def); | |
abd29c93 AN |
2023 | for (i = 0; i < vlen; i++, m++) { |
2024 | const char *name = btf__name_by_offset(obj->btf, m->name_off); | |
2025 | ||
2026 | if (!name) { | |
41017e56 | 2027 | pr_warn("map '%s': invalid field #%d.\n", map->name, i); |
abd29c93 AN |
2028 | return -EINVAL; |
2029 | } | |
2030 | if (strcmp(name, "type") == 0) { | |
41017e56 | 2031 | if (!get_map_field_int(map->name, obj->btf, m, |
ef99b02b | 2032 | &map->def.type)) |
abd29c93 AN |
2033 | return -EINVAL; |
2034 | pr_debug("map '%s': found type = %u.\n", | |
41017e56 | 2035 | map->name, map->def.type); |
abd29c93 | 2036 | } else if (strcmp(name, "max_entries") == 0) { |
41017e56 | 2037 | if (!get_map_field_int(map->name, obj->btf, m, |
ef99b02b | 2038 | &map->def.max_entries)) |
abd29c93 AN |
2039 | return -EINVAL; |
2040 | pr_debug("map '%s': found max_entries = %u.\n", | |
41017e56 | 2041 | map->name, map->def.max_entries); |
abd29c93 | 2042 | } else if (strcmp(name, "map_flags") == 0) { |
41017e56 | 2043 | if (!get_map_field_int(map->name, obj->btf, m, |
ef99b02b | 2044 | &map->def.map_flags)) |
abd29c93 AN |
2045 | return -EINVAL; |
2046 | pr_debug("map '%s': found map_flags = %u.\n", | |
41017e56 | 2047 | map->name, map->def.map_flags); |
1bdb6c9a AN |
2048 | } else if (strcmp(name, "numa_node") == 0) { |
2049 | if (!get_map_field_int(map->name, obj->btf, m, &map->numa_node)) | |
2050 | return -EINVAL; | |
2051 | pr_debug("map '%s': found numa_node = %u.\n", map->name, map->numa_node); | |
abd29c93 AN |
2052 | } else if (strcmp(name, "key_size") == 0) { |
2053 | __u32 sz; | |
2054 | ||
41017e56 | 2055 | if (!get_map_field_int(map->name, obj->btf, m, &sz)) |
abd29c93 AN |
2056 | return -EINVAL; |
2057 | pr_debug("map '%s': found key_size = %u.\n", | |
41017e56 | 2058 | map->name, sz); |
abd29c93 | 2059 | if (map->def.key_size && map->def.key_size != sz) { |
be18010e | 2060 | pr_warn("map '%s': conflicting key size %u != %u.\n", |
41017e56 | 2061 | map->name, map->def.key_size, sz); |
abd29c93 AN |
2062 | return -EINVAL; |
2063 | } | |
2064 | map->def.key_size = sz; | |
2065 | } else if (strcmp(name, "key") == 0) { | |
2066 | __s64 sz; | |
2067 | ||
2068 | t = btf__type_by_id(obj->btf, m->type); | |
2069 | if (!t) { | |
be18010e | 2070 | pr_warn("map '%s': key type [%d] not found.\n", |
41017e56 | 2071 | map->name, m->type); |
abd29c93 AN |
2072 | return -EINVAL; |
2073 | } | |
b03bc685 | 2074 | if (!btf_is_ptr(t)) { |
81ba0889 AN |
2075 | pr_warn("map '%s': key spec is not PTR: %s.\n", |
2076 | map->name, btf_kind_str(t)); | |
abd29c93 AN |
2077 | return -EINVAL; |
2078 | } | |
2079 | sz = btf__resolve_size(obj->btf, t->type); | |
2080 | if (sz < 0) { | |
679152d3 | 2081 | pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n", |
41017e56 | 2082 | map->name, t->type, (ssize_t)sz); |
abd29c93 AN |
2083 | return sz; |
2084 | } | |
679152d3 | 2085 | pr_debug("map '%s': found key [%u], sz = %zd.\n", |
41017e56 | 2086 | map->name, t->type, (ssize_t)sz); |
abd29c93 | 2087 | if (map->def.key_size && map->def.key_size != sz) { |
679152d3 | 2088 | pr_warn("map '%s': conflicting key size %u != %zd.\n", |
41017e56 | 2089 | map->name, map->def.key_size, (ssize_t)sz); |
abd29c93 AN |
2090 | return -EINVAL; |
2091 | } | |
2092 | map->def.key_size = sz; | |
2093 | map->btf_key_type_id = t->type; | |
2094 | } else if (strcmp(name, "value_size") == 0) { | |
2095 | __u32 sz; | |
2096 | ||
41017e56 | 2097 | if (!get_map_field_int(map->name, obj->btf, m, &sz)) |
abd29c93 AN |
2098 | return -EINVAL; |
2099 | pr_debug("map '%s': found value_size = %u.\n", | |
41017e56 | 2100 | map->name, sz); |
abd29c93 | 2101 | if (map->def.value_size && map->def.value_size != sz) { |
be18010e | 2102 | pr_warn("map '%s': conflicting value size %u != %u.\n", |
41017e56 | 2103 | map->name, map->def.value_size, sz); |
abd29c93 AN |
2104 | return -EINVAL; |
2105 | } | |
2106 | map->def.value_size = sz; | |
2107 | } else if (strcmp(name, "value") == 0) { | |
2108 | __s64 sz; | |
2109 | ||
2110 | t = btf__type_by_id(obj->btf, m->type); | |
2111 | if (!t) { | |
be18010e | 2112 | pr_warn("map '%s': value type [%d] not found.\n", |
41017e56 | 2113 | map->name, m->type); |
abd29c93 AN |
2114 | return -EINVAL; |
2115 | } | |
b03bc685 | 2116 | if (!btf_is_ptr(t)) { |
81ba0889 AN |
2117 | pr_warn("map '%s': value spec is not PTR: %s.\n", |
2118 | map->name, btf_kind_str(t)); | |
abd29c93 AN |
2119 | return -EINVAL; |
2120 | } | |
2121 | sz = btf__resolve_size(obj->btf, t->type); | |
2122 | if (sz < 0) { | |
679152d3 | 2123 | pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n", |
41017e56 | 2124 | map->name, t->type, (ssize_t)sz); |
abd29c93 AN |
2125 | return sz; |
2126 | } | |
679152d3 | 2127 | pr_debug("map '%s': found value [%u], sz = %zd.\n", |
41017e56 | 2128 | map->name, t->type, (ssize_t)sz); |
abd29c93 | 2129 | if (map->def.value_size && map->def.value_size != sz) { |
679152d3 | 2130 | pr_warn("map '%s': conflicting value size %u != %zd.\n", |
41017e56 | 2131 | map->name, map->def.value_size, (ssize_t)sz); |
abd29c93 AN |
2132 | return -EINVAL; |
2133 | } | |
2134 | map->def.value_size = sz; | |
2135 | map->btf_value_type_id = t->type; | |
646f02ff AN |
2136 | } |
2137 | else if (strcmp(name, "values") == 0) { | |
2138 | int err; | |
2139 | ||
2140 | if (is_inner) { | |
2141 | pr_warn("map '%s': multi-level inner maps not supported.\n", | |
2142 | map->name); | |
2143 | return -ENOTSUP; | |
2144 | } | |
2145 | if (i != vlen - 1) { | |
2146 | pr_warn("map '%s': '%s' member should be last.\n", | |
2147 | map->name, name); | |
2148 | return -EINVAL; | |
2149 | } | |
2150 | if (!bpf_map_type__is_map_in_map(map->def.type)) { | |
2151 | pr_warn("map '%s': should be map-in-map.\n", | |
2152 | map->name); | |
2153 | return -ENOTSUP; | |
2154 | } | |
2155 | if (map->def.value_size && map->def.value_size != 4) { | |
2156 | pr_warn("map '%s': conflicting value size %u != 4.\n", | |
2157 | map->name, map->def.value_size); | |
2158 | return -EINVAL; | |
2159 | } | |
2160 | map->def.value_size = 4; | |
2161 | t = btf__type_by_id(obj->btf, m->type); | |
2162 | if (!t) { | |
2163 | pr_warn("map '%s': map-in-map inner type [%d] not found.\n", | |
2164 | map->name, m->type); | |
2165 | return -EINVAL; | |
2166 | } | |
2167 | if (!btf_is_array(t) || btf_array(t)->nelems) { | |
2168 | pr_warn("map '%s': map-in-map inner spec is not a zero-sized array.\n", | |
2169 | map->name); | |
2170 | return -EINVAL; | |
2171 | } | |
2172 | t = skip_mods_and_typedefs(obj->btf, btf_array(t)->type, | |
2173 | NULL); | |
2174 | if (!btf_is_ptr(t)) { | |
81ba0889 AN |
2175 | pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n", |
2176 | map->name, btf_kind_str(t)); | |
646f02ff AN |
2177 | return -EINVAL; |
2178 | } | |
2179 | t = skip_mods_and_typedefs(obj->btf, t->type, NULL); | |
2180 | if (!btf_is_struct(t)) { | |
81ba0889 AN |
2181 | pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n", |
2182 | map->name, btf_kind_str(t)); | |
646f02ff AN |
2183 | return -EINVAL; |
2184 | } | |
2185 | ||
2186 | map->inner_map = calloc(1, sizeof(*map->inner_map)); | |
2187 | if (!map->inner_map) | |
2188 | return -ENOMEM; | |
2189 | map->inner_map->sec_idx = obj->efile.btf_maps_shndx; | |
2190 | map->inner_map->name = malloc(strlen(map->name) + | |
2191 | sizeof(".inner") + 1); | |
2192 | if (!map->inner_map->name) | |
2193 | return -ENOMEM; | |
2194 | sprintf(map->inner_map->name, "%s.inner", map->name); | |
2195 | ||
2196 | err = parse_btf_map_def(obj, map->inner_map, t, strict, | |
2197 | true /* is_inner */, NULL); | |
2198 | if (err) | |
2199 | return err; | |
57a00f41 THJ |
2200 | } else if (strcmp(name, "pinning") == 0) { |
2201 | __u32 val; | |
2202 | int err; | |
2203 | ||
646f02ff AN |
2204 | if (is_inner) { |
2205 | pr_debug("map '%s': inner def can't be pinned.\n", | |
2206 | map->name); | |
2207 | return -EINVAL; | |
2208 | } | |
41017e56 | 2209 | if (!get_map_field_int(map->name, obj->btf, m, &val)) |
57a00f41 THJ |
2210 | return -EINVAL; |
2211 | pr_debug("map '%s': found pinning = %u.\n", | |
41017e56 | 2212 | map->name, val); |
57a00f41 THJ |
2213 | |
2214 | if (val != LIBBPF_PIN_NONE && | |
2215 | val != LIBBPF_PIN_BY_NAME) { | |
2216 | pr_warn("map '%s': invalid pinning value %u.\n", | |
41017e56 | 2217 | map->name, val); |
57a00f41 THJ |
2218 | return -EINVAL; |
2219 | } | |
2220 | if (val == LIBBPF_PIN_BY_NAME) { | |
2221 | err = build_map_pin_path(map, pin_root_path); | |
2222 | if (err) { | |
2223 | pr_warn("map '%s': couldn't build pin path.\n", | |
41017e56 | 2224 | map->name); |
57a00f41 THJ |
2225 | return err; |
2226 | } | |
2227 | } | |
abd29c93 AN |
2228 | } else { |
2229 | if (strict) { | |
be18010e | 2230 | pr_warn("map '%s': unknown field '%s'.\n", |
41017e56 | 2231 | map->name, name); |
abd29c93 AN |
2232 | return -ENOTSUP; |
2233 | } | |
2234 | pr_debug("map '%s': ignoring unknown field '%s'.\n", | |
41017e56 | 2235 | map->name, name); |
abd29c93 AN |
2236 | } |
2237 | } | |
2238 | ||
2239 | if (map->def.type == BPF_MAP_TYPE_UNSPEC) { | |
41017e56 | 2240 | pr_warn("map '%s': map type isn't specified.\n", map->name); |
abd29c93 AN |
2241 | return -EINVAL; |
2242 | } | |
2243 | ||
2244 | return 0; | |
2245 | } | |
2246 | ||
41017e56 AN |
2247 | static int bpf_object__init_user_btf_map(struct bpf_object *obj, |
2248 | const struct btf_type *sec, | |
2249 | int var_idx, int sec_idx, | |
2250 | const Elf_Data *data, bool strict, | |
2251 | const char *pin_root_path) | |
2252 | { | |
2253 | const struct btf_type *var, *def; | |
2254 | const struct btf_var_secinfo *vi; | |
2255 | const struct btf_var *var_extra; | |
2256 | const char *map_name; | |
2257 | struct bpf_map *map; | |
2258 | ||
2259 | vi = btf_var_secinfos(sec) + var_idx; | |
2260 | var = btf__type_by_id(obj->btf, vi->type); | |
2261 | var_extra = btf_var(var); | |
2262 | map_name = btf__name_by_offset(obj->btf, var->name_off); | |
2263 | ||
2264 | if (map_name == NULL || map_name[0] == '\0') { | |
2265 | pr_warn("map #%d: empty name.\n", var_idx); | |
2266 | return -EINVAL; | |
2267 | } | |
2268 | if ((__u64)vi->offset + vi->size > data->d_size) { | |
2269 | pr_warn("map '%s' BTF data is corrupted.\n", map_name); | |
2270 | return -EINVAL; | |
2271 | } | |
2272 | if (!btf_is_var(var)) { | |
81ba0889 AN |
2273 | pr_warn("map '%s': unexpected var kind %s.\n", |
2274 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2275 | return -EINVAL; |
2276 | } | |
2277 | if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED && | |
2278 | var_extra->linkage != BTF_VAR_STATIC) { | |
2279 | pr_warn("map '%s': unsupported var linkage %u.\n", | |
2280 | map_name, var_extra->linkage); | |
2281 | return -EOPNOTSUPP; | |
2282 | } | |
2283 | ||
2284 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
2285 | if (!btf_is_struct(def)) { | |
81ba0889 AN |
2286 | pr_warn("map '%s': unexpected def kind %s.\n", |
2287 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2288 | return -EINVAL; |
2289 | } | |
2290 | if (def->size > vi->size) { | |
2291 | pr_warn("map '%s': invalid def size.\n", map_name); | |
2292 | return -EINVAL; | |
2293 | } | |
2294 | ||
2295 | map = bpf_object__add_map(obj); | |
2296 | if (IS_ERR(map)) | |
2297 | return PTR_ERR(map); | |
2298 | map->name = strdup(map_name); | |
2299 | if (!map->name) { | |
2300 | pr_warn("map '%s': failed to alloc map name.\n", map_name); | |
2301 | return -ENOMEM; | |
2302 | } | |
2303 | map->libbpf_type = LIBBPF_MAP_UNSPEC; | |
2304 | map->def.type = BPF_MAP_TYPE_UNSPEC; | |
2305 | map->sec_idx = sec_idx; | |
2306 | map->sec_offset = vi->offset; | |
646f02ff | 2307 | map->btf_var_idx = var_idx; |
41017e56 AN |
2308 | pr_debug("map '%s': at sec_idx %d, offset %zu.\n", |
2309 | map_name, map->sec_idx, map->sec_offset); | |
2310 | ||
646f02ff | 2311 | return parse_btf_map_def(obj, map, def, strict, false, pin_root_path); |
41017e56 AN |
2312 | } |
2313 | ||
57a00f41 THJ |
2314 | static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict, |
2315 | const char *pin_root_path) | |
abd29c93 AN |
2316 | { |
2317 | const struct btf_type *sec = NULL; | |
2318 | int nr_types, i, vlen, err; | |
2319 | const struct btf_type *t; | |
2320 | const char *name; | |
2321 | Elf_Data *data; | |
2322 | Elf_Scn *scn; | |
2323 | ||
2324 | if (obj->efile.btf_maps_shndx < 0) | |
2325 | return 0; | |
2326 | ||
88a82120 AN |
2327 | scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx); |
2328 | data = elf_sec_data(obj, scn); | |
abd29c93 | 2329 | if (!scn || !data) { |
88a82120 AN |
2330 | pr_warn("elf: failed to get %s map definitions for %s\n", |
2331 | MAPS_ELF_SEC, obj->path); | |
abd29c93 AN |
2332 | return -EINVAL; |
2333 | } | |
2334 | ||
2335 | nr_types = btf__get_nr_types(obj->btf); | |
2336 | for (i = 1; i <= nr_types; i++) { | |
2337 | t = btf__type_by_id(obj->btf, i); | |
b03bc685 | 2338 | if (!btf_is_datasec(t)) |
abd29c93 AN |
2339 | continue; |
2340 | name = btf__name_by_offset(obj->btf, t->name_off); | |
2341 | if (strcmp(name, MAPS_ELF_SEC) == 0) { | |
2342 | sec = t; | |
646f02ff | 2343 | obj->efile.btf_maps_sec_btf_id = i; |
abd29c93 AN |
2344 | break; |
2345 | } | |
2346 | } | |
2347 | ||
2348 | if (!sec) { | |
be18010e | 2349 | pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC); |
abd29c93 AN |
2350 | return -ENOENT; |
2351 | } | |
2352 | ||
b03bc685 | 2353 | vlen = btf_vlen(sec); |
abd29c93 AN |
2354 | for (i = 0; i < vlen; i++) { |
2355 | err = bpf_object__init_user_btf_map(obj, sec, i, | |
2356 | obj->efile.btf_maps_shndx, | |
8983b731 AN |
2357 | data, strict, |
2358 | pin_root_path); | |
abd29c93 AN |
2359 | if (err) |
2360 | return err; | |
2361 | } | |
2362 | ||
2363 | return 0; | |
2364 | } | |
2365 | ||
0d13bfce | 2366 | static int bpf_object__init_maps(struct bpf_object *obj, |
01af3bf0 | 2367 | const struct bpf_object_open_opts *opts) |
bf829271 | 2368 | { |
166750bc AN |
2369 | const char *pin_root_path; |
2370 | bool strict; | |
bf829271 | 2371 | int err; |
8837fe5d | 2372 | |
166750bc AN |
2373 | strict = !OPTS_GET(opts, relaxed_maps, false); |
2374 | pin_root_path = OPTS_GET(opts, pin_root_path, NULL); | |
bf829271 | 2375 | |
166750bc AN |
2376 | err = bpf_object__init_user_maps(obj, strict); |
2377 | err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path); | |
2378 | err = err ?: bpf_object__init_global_data_maps(obj); | |
81bfdd08 | 2379 | err = err ?: bpf_object__init_kconfig_map(obj); |
590a0088 | 2380 | err = err ?: bpf_object__init_struct_ops_maps(obj); |
bf829271 AN |
2381 | if (err) |
2382 | return err; | |
2383 | ||
bf829271 | 2384 | return 0; |
561bbcca WN |
2385 | } |
2386 | ||
e3d91b0c JDB |
2387 | static bool section_have_execinstr(struct bpf_object *obj, int idx) |
2388 | { | |
e3d91b0c JDB |
2389 | GElf_Shdr sh; |
2390 | ||
88a82120 | 2391 | if (elf_sec_hdr(obj, elf_sec_by_idx(obj, idx), &sh)) |
e3d91b0c JDB |
2392 | return false; |
2393 | ||
88a82120 | 2394 | return sh.sh_flags & SHF_EXECINSTR; |
e3d91b0c JDB |
2395 | } |
2396 | ||
0f0e55d8 AN |
2397 | static bool btf_needs_sanitization(struct bpf_object *obj) |
2398 | { | |
47b6cb4d AN |
2399 | bool has_func_global = kernel_supports(FEAT_BTF_GLOBAL_FUNC); |
2400 | bool has_datasec = kernel_supports(FEAT_BTF_DATASEC); | |
22541a9e | 2401 | bool has_float = kernel_supports(FEAT_BTF_FLOAT); |
47b6cb4d | 2402 | bool has_func = kernel_supports(FEAT_BTF_FUNC); |
0f0e55d8 | 2403 | |
22541a9e | 2404 | return !has_func || !has_datasec || !has_func_global || !has_float; |
0f0e55d8 AN |
2405 | } |
2406 | ||
2407 | static void bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf) | |
d7c4b398 | 2408 | { |
47b6cb4d AN |
2409 | bool has_func_global = kernel_supports(FEAT_BTF_GLOBAL_FUNC); |
2410 | bool has_datasec = kernel_supports(FEAT_BTF_DATASEC); | |
22541a9e | 2411 | bool has_float = kernel_supports(FEAT_BTF_FLOAT); |
47b6cb4d | 2412 | bool has_func = kernel_supports(FEAT_BTF_FUNC); |
d7c4b398 AN |
2413 | struct btf_type *t; |
2414 | int i, j, vlen; | |
d7c4b398 | 2415 | |
d7c4b398 AN |
2416 | for (i = 1; i <= btf__get_nr_types(btf); i++) { |
2417 | t = (struct btf_type *)btf__type_by_id(btf, i); | |
d7c4b398 | 2418 | |
b03bc685 | 2419 | if (!has_datasec && btf_is_var(t)) { |
d7c4b398 AN |
2420 | /* replace VAR with INT */ |
2421 | t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0); | |
1d4126c4 AN |
2422 | /* |
2423 | * using size = 1 is the safest choice, 4 will be too | |
2424 | * big and cause kernel BTF validation failure if | |
2425 | * original variable took less than 4 bytes | |
2426 | */ | |
2427 | t->size = 1; | |
708852dc | 2428 | *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8); |
b03bc685 | 2429 | } else if (!has_datasec && btf_is_datasec(t)) { |
d7c4b398 | 2430 | /* replace DATASEC with STRUCT */ |
b03bc685 AN |
2431 | const struct btf_var_secinfo *v = btf_var_secinfos(t); |
2432 | struct btf_member *m = btf_members(t); | |
d7c4b398 AN |
2433 | struct btf_type *vt; |
2434 | char *name; | |
2435 | ||
2436 | name = (char *)btf__name_by_offset(btf, t->name_off); | |
2437 | while (*name) { | |
2438 | if (*name == '.') | |
2439 | *name = '_'; | |
2440 | name++; | |
2441 | } | |
2442 | ||
b03bc685 | 2443 | vlen = btf_vlen(t); |
d7c4b398 AN |
2444 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen); |
2445 | for (j = 0; j < vlen; j++, v++, m++) { | |
2446 | /* order of field assignments is important */ | |
2447 | m->offset = v->offset * 8; | |
2448 | m->type = v->type; | |
2449 | /* preserve variable name as member name */ | |
2450 | vt = (void *)btf__type_by_id(btf, v->type); | |
2451 | m->name_off = vt->name_off; | |
2452 | } | |
b03bc685 | 2453 | } else if (!has_func && btf_is_func_proto(t)) { |
d7c4b398 | 2454 | /* replace FUNC_PROTO with ENUM */ |
b03bc685 | 2455 | vlen = btf_vlen(t); |
d7c4b398 AN |
2456 | t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen); |
2457 | t->size = sizeof(__u32); /* kernel enforced */ | |
b03bc685 | 2458 | } else if (!has_func && btf_is_func(t)) { |
d7c4b398 AN |
2459 | /* replace FUNC with TYPEDEF */ |
2460 | t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0); | |
2d3eb67f AS |
2461 | } else if (!has_func_global && btf_is_func(t)) { |
2462 | /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */ | |
2463 | t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0); | |
22541a9e IL |
2464 | } else if (!has_float && btf_is_float(t)) { |
2465 | /* replace FLOAT with an equally-sized empty STRUCT; | |
2466 | * since C compilers do not accept e.g. "float" as a | |
2467 | * valid struct name, make it anonymous | |
2468 | */ | |
2469 | t->name_off = 0; | |
2470 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0); | |
d7c4b398 AN |
2471 | } |
2472 | } | |
2473 | } | |
2474 | ||
b35f14f4 | 2475 | static bool libbpf_needs_btf(const struct bpf_object *obj) |
abd29c93 | 2476 | { |
b35f14f4 AN |
2477 | return obj->efile.btf_maps_shndx >= 0 || |
2478 | obj->efile.st_ops_shndx >= 0 || | |
2479 | obj->nr_extern > 0; | |
2480 | } | |
2481 | ||
2482 | static bool kernel_needs_btf(const struct bpf_object *obj) | |
2483 | { | |
2484 | return obj->efile.st_ops_shndx >= 0; | |
abd29c93 AN |
2485 | } |
2486 | ||
063183bf | 2487 | static int bpf_object__init_btf(struct bpf_object *obj, |
9c6660d0 AN |
2488 | Elf_Data *btf_data, |
2489 | Elf_Data *btf_ext_data) | |
2490 | { | |
b7d7f3e1 | 2491 | int err = -ENOENT; |
9c6660d0 AN |
2492 | |
2493 | if (btf_data) { | |
2494 | obj->btf = btf__new(btf_data->d_buf, btf_data->d_size); | |
2495 | if (IS_ERR(obj->btf)) { | |
b7d7f3e1 AN |
2496 | err = PTR_ERR(obj->btf); |
2497 | obj->btf = NULL; | |
be18010e KW |
2498 | pr_warn("Error loading ELF section %s: %d.\n", |
2499 | BTF_ELF_SEC, err); | |
9c6660d0 AN |
2500 | goto out; |
2501 | } | |
4c01925f AN |
2502 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
2503 | btf__set_pointer_size(obj->btf, 8); | |
b7d7f3e1 | 2504 | err = 0; |
9c6660d0 AN |
2505 | } |
2506 | if (btf_ext_data) { | |
2507 | if (!obj->btf) { | |
2508 | pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n", | |
2509 | BTF_EXT_ELF_SEC, BTF_ELF_SEC); | |
2510 | goto out; | |
2511 | } | |
2512 | obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, | |
2513 | btf_ext_data->d_size); | |
2514 | if (IS_ERR(obj->btf_ext)) { | |
be18010e KW |
2515 | pr_warn("Error loading ELF section %s: %ld. Ignored and continue.\n", |
2516 | BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext)); | |
9c6660d0 AN |
2517 | obj->btf_ext = NULL; |
2518 | goto out; | |
2519 | } | |
9c6660d0 AN |
2520 | } |
2521 | out: | |
b35f14f4 | 2522 | if (err && libbpf_needs_btf(obj)) { |
be18010e | 2523 | pr_warn("BTF is required, but is missing or corrupted.\n"); |
b7d7f3e1 | 2524 | return err; |
abd29c93 | 2525 | } |
9c6660d0 AN |
2526 | return 0; |
2527 | } | |
2528 | ||
166750bc AN |
2529 | static int bpf_object__finalize_btf(struct bpf_object *obj) |
2530 | { | |
2531 | int err; | |
2532 | ||
2533 | if (!obj->btf) | |
2534 | return 0; | |
2535 | ||
2536 | err = btf__finalize_data(obj, obj->btf); | |
bfc96656 AN |
2537 | if (err) { |
2538 | pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err); | |
2539 | return err; | |
166750bc | 2540 | } |
bfc96656 | 2541 | |
166750bc AN |
2542 | return 0; |
2543 | } | |
2544 | ||
fe62de31 | 2545 | static bool prog_needs_vmlinux_btf(struct bpf_program *prog) |
a6ed02ca | 2546 | { |
1e092a03 KS |
2547 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS || |
2548 | prog->type == BPF_PROG_TYPE_LSM) | |
a6ed02ca KS |
2549 | return true; |
2550 | ||
2551 | /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs | |
2552 | * also need vmlinux BTF | |
2553 | */ | |
2554 | if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd) | |
2555 | return true; | |
2556 | ||
2557 | return false; | |
2558 | } | |
2559 | ||
fe62de31 | 2560 | static bool obj_needs_vmlinux_btf(const struct bpf_object *obj) |
a6ed02ca KS |
2561 | { |
2562 | struct bpf_program *prog; | |
fe62de31 | 2563 | int i; |
a6ed02ca | 2564 | |
192b6638 | 2565 | /* CO-RE relocations need kernel BTF */ |
28b93c64 | 2566 | if (obj->btf_ext && obj->btf_ext->core_relo_info.len) |
fe62de31 | 2567 | return true; |
192b6638 | 2568 | |
d370bbe1 HL |
2569 | /* Support for typed ksyms needs kernel BTF */ |
2570 | for (i = 0; i < obj->nr_extern; i++) { | |
2571 | const struct extern_desc *ext; | |
2572 | ||
2573 | ext = &obj->externs[i]; | |
fe62de31 AN |
2574 | if (ext->type == EXT_KSYM && ext->ksym.type_id) |
2575 | return true; | |
d370bbe1 HL |
2576 | } |
2577 | ||
a6ed02ca | 2578 | bpf_object__for_each_program(prog, obj) { |
d9297581 AN |
2579 | if (!prog->load) |
2580 | continue; | |
fe62de31 AN |
2581 | if (prog_needs_vmlinux_btf(prog)) |
2582 | return true; | |
a6ed02ca KS |
2583 | } |
2584 | ||
fe62de31 AN |
2585 | return false; |
2586 | } | |
2587 | ||
2588 | static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force) | |
2589 | { | |
2590 | int err; | |
2591 | ||
2592 | /* btf_vmlinux could be loaded earlier */ | |
2593 | if (obj->btf_vmlinux) | |
2594 | return 0; | |
2595 | ||
2596 | if (!force && !obj_needs_vmlinux_btf(obj)) | |
192b6638 AN |
2597 | return 0; |
2598 | ||
2599 | obj->btf_vmlinux = libbpf_find_kernel_btf(); | |
2600 | if (IS_ERR(obj->btf_vmlinux)) { | |
2601 | err = PTR_ERR(obj->btf_vmlinux); | |
2602 | pr_warn("Error loading vmlinux BTF: %d\n", err); | |
2603 | obj->btf_vmlinux = NULL; | |
2604 | return err; | |
2605 | } | |
a6ed02ca KS |
2606 | return 0; |
2607 | } | |
2608 | ||
063183bf AN |
2609 | static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj) |
2610 | { | |
0f0e55d8 AN |
2611 | struct btf *kern_btf = obj->btf; |
2612 | bool btf_mandatory, sanitize; | |
063183bf AN |
2613 | int err = 0; |
2614 | ||
2615 | if (!obj->btf) | |
2616 | return 0; | |
2617 | ||
68b08647 AN |
2618 | if (!kernel_supports(FEAT_BTF)) { |
2619 | if (kernel_needs_btf(obj)) { | |
2620 | err = -EOPNOTSUPP; | |
2621 | goto report; | |
2622 | } | |
2623 | pr_debug("Kernel doesn't support BTF, skipping uploading it.\n"); | |
2624 | return 0; | |
2625 | } | |
2626 | ||
0f0e55d8 AN |
2627 | sanitize = btf_needs_sanitization(obj); |
2628 | if (sanitize) { | |
5c3320d7 | 2629 | const void *raw_data; |
0f0e55d8 | 2630 | __u32 sz; |
063183bf | 2631 | |
0f0e55d8 | 2632 | /* clone BTF to sanitize a copy and leave the original intact */ |
5c3320d7 AN |
2633 | raw_data = btf__get_raw_data(obj->btf, &sz); |
2634 | kern_btf = btf__new(raw_data, sz); | |
0f0e55d8 AN |
2635 | if (IS_ERR(kern_btf)) |
2636 | return PTR_ERR(kern_btf); | |
04efe591 | 2637 | |
4c01925f AN |
2638 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
2639 | btf__set_pointer_size(obj->btf, 8); | |
0f0e55d8 | 2640 | bpf_object__sanitize_btf(obj, kern_btf); |
063183bf | 2641 | } |
0f0e55d8 AN |
2642 | |
2643 | err = btf__load(kern_btf); | |
2644 | if (sanitize) { | |
2645 | if (!err) { | |
2646 | /* move fd to libbpf's BTF */ | |
2647 | btf__set_fd(obj->btf, btf__fd(kern_btf)); | |
2648 | btf__set_fd(kern_btf, -1); | |
2649 | } | |
2650 | btf__free(kern_btf); | |
2651 | } | |
68b08647 | 2652 | report: |
0f0e55d8 AN |
2653 | if (err) { |
2654 | btf_mandatory = kernel_needs_btf(obj); | |
2655 | pr_warn("Error loading .BTF into kernel: %d. %s\n", err, | |
2656 | btf_mandatory ? "BTF is mandatory, can't proceed." | |
2657 | : "BTF is optional, ignoring."); | |
2658 | if (!btf_mandatory) | |
2659 | err = 0; | |
2660 | } | |
2661 | return err; | |
063183bf AN |
2662 | } |
2663 | ||
88a82120 AN |
2664 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off) |
2665 | { | |
2666 | const char *name; | |
2667 | ||
2668 | name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off); | |
2669 | if (!name) { | |
2670 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
2671 | off, obj->path, elf_errmsg(-1)); | |
2672 | return NULL; | |
2673 | } | |
2674 | ||
2675 | return name; | |
2676 | } | |
2677 | ||
2678 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off) | |
2679 | { | |
2680 | const char *name; | |
2681 | ||
2682 | name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off); | |
2683 | if (!name) { | |
2684 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
2685 | off, obj->path, elf_errmsg(-1)); | |
2686 | return NULL; | |
2687 | } | |
2688 | ||
2689 | return name; | |
2690 | } | |
2691 | ||
2692 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx) | |
2693 | { | |
2694 | Elf_Scn *scn; | |
2695 | ||
2696 | scn = elf_getscn(obj->efile.elf, idx); | |
2697 | if (!scn) { | |
2698 | pr_warn("elf: failed to get section(%zu) from %s: %s\n", | |
2699 | idx, obj->path, elf_errmsg(-1)); | |
2700 | return NULL; | |
2701 | } | |
2702 | return scn; | |
2703 | } | |
2704 | ||
2705 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name) | |
2706 | { | |
2707 | Elf_Scn *scn = NULL; | |
2708 | Elf *elf = obj->efile.elf; | |
2709 | const char *sec_name; | |
2710 | ||
2711 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
2712 | sec_name = elf_sec_name(obj, scn); | |
2713 | if (!sec_name) | |
2714 | return NULL; | |
2715 | ||
2716 | if (strcmp(sec_name, name) != 0) | |
2717 | continue; | |
2718 | ||
2719 | return scn; | |
2720 | } | |
2721 | return NULL; | |
2722 | } | |
2723 | ||
2724 | static int elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn, GElf_Shdr *hdr) | |
2725 | { | |
2726 | if (!scn) | |
2727 | return -EINVAL; | |
2728 | ||
2729 | if (gelf_getshdr(scn, hdr) != hdr) { | |
2730 | pr_warn("elf: failed to get section(%zu) header from %s: %s\n", | |
2731 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
2732 | return -EINVAL; | |
2733 | } | |
2734 | ||
2735 | return 0; | |
2736 | } | |
2737 | ||
2738 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn) | |
2739 | { | |
2740 | const char *name; | |
2741 | GElf_Shdr sh; | |
2742 | ||
2743 | if (!scn) | |
2744 | return NULL; | |
2745 | ||
2746 | if (elf_sec_hdr(obj, scn, &sh)) | |
2747 | return NULL; | |
2748 | ||
2749 | name = elf_sec_str(obj, sh.sh_name); | |
2750 | if (!name) { | |
2751 | pr_warn("elf: failed to get section(%zu) name from %s: %s\n", | |
2752 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
2753 | return NULL; | |
2754 | } | |
2755 | ||
2756 | return name; | |
2757 | } | |
2758 | ||
2759 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn) | |
2760 | { | |
2761 | Elf_Data *data; | |
2762 | ||
2763 | if (!scn) | |
2764 | return NULL; | |
2765 | ||
2766 | data = elf_getdata(scn, 0); | |
2767 | if (!data) { | |
2768 | pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n", | |
2769 | elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>", | |
2770 | obj->path, elf_errmsg(-1)); | |
2771 | return NULL; | |
2772 | } | |
2773 | ||
2774 | return data; | |
2775 | } | |
2776 | ||
c1122392 AN |
2777 | static int elf_sym_by_sec_off(const struct bpf_object *obj, size_t sec_idx, |
2778 | size_t off, __u32 sym_type, GElf_Sym *sym) | |
2779 | { | |
2780 | Elf_Data *symbols = obj->efile.symbols; | |
2781 | size_t n = symbols->d_size / sizeof(GElf_Sym); | |
2782 | int i; | |
2783 | ||
2784 | for (i = 0; i < n; i++) { | |
2785 | if (!gelf_getsym(symbols, i, sym)) | |
2786 | continue; | |
2787 | if (sym->st_shndx != sec_idx || sym->st_value != off) | |
2788 | continue; | |
2789 | if (GELF_ST_TYPE(sym->st_info) != sym_type) | |
2790 | continue; | |
2791 | return 0; | |
2792 | } | |
2793 | ||
2794 | return -ENOENT; | |
2795 | } | |
2796 | ||
50e09460 AN |
2797 | static bool is_sec_name_dwarf(const char *name) |
2798 | { | |
2799 | /* approximation, but the actual list is too long */ | |
2800 | return strncmp(name, ".debug_", sizeof(".debug_") - 1) == 0; | |
2801 | } | |
2802 | ||
2803 | static bool ignore_elf_section(GElf_Shdr *hdr, const char *name) | |
2804 | { | |
2805 | /* no special handling of .strtab */ | |
2806 | if (hdr->sh_type == SHT_STRTAB) | |
2807 | return true; | |
2808 | ||
2809 | /* ignore .llvm_addrsig section as well */ | |
2810 | if (hdr->sh_type == 0x6FFF4C03 /* SHT_LLVM_ADDRSIG */) | |
2811 | return true; | |
2812 | ||
2813 | /* no subprograms will lead to an empty .text section, ignore it */ | |
2814 | if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 && | |
2815 | strcmp(name, ".text") == 0) | |
2816 | return true; | |
2817 | ||
2818 | /* DWARF sections */ | |
2819 | if (is_sec_name_dwarf(name)) | |
2820 | return true; | |
2821 | ||
2822 | if (strncmp(name, ".rel", sizeof(".rel") - 1) == 0) { | |
2823 | name += sizeof(".rel") - 1; | |
2824 | /* DWARF section relocations */ | |
2825 | if (is_sec_name_dwarf(name)) | |
2826 | return true; | |
2827 | ||
2828 | /* .BTF and .BTF.ext don't need relocations */ | |
2829 | if (strcmp(name, BTF_ELF_SEC) == 0 || | |
2830 | strcmp(name, BTF_EXT_ELF_SEC) == 0) | |
2831 | return true; | |
2832 | } | |
2833 | ||
2834 | return false; | |
2835 | } | |
2836 | ||
db2b8b06 AN |
2837 | static int cmp_progs(const void *_a, const void *_b) |
2838 | { | |
2839 | const struct bpf_program *a = _a; | |
2840 | const struct bpf_program *b = _b; | |
2841 | ||
2842 | if (a->sec_idx != b->sec_idx) | |
2843 | return a->sec_idx < b->sec_idx ? -1 : 1; | |
2844 | ||
2845 | /* sec_insn_off can't be the same within the section */ | |
2846 | return a->sec_insn_off < b->sec_insn_off ? -1 : 1; | |
2847 | } | |
2848 | ||
0d13bfce | 2849 | static int bpf_object__elf_collect(struct bpf_object *obj) |
29603665 WN |
2850 | { |
2851 | Elf *elf = obj->efile.elf; | |
f0187f0b | 2852 | Elf_Data *btf_ext_data = NULL; |
1713d68b | 2853 | Elf_Data *btf_data = NULL; |
666810e8 | 2854 | int idx = 0, err = 0; |
0201c575 AN |
2855 | const char *name; |
2856 | Elf_Data *data; | |
2857 | Elf_Scn *scn; | |
2858 | GElf_Shdr sh; | |
29603665 | 2859 | |
0201c575 AN |
2860 | /* a bunch of ELF parsing functionality depends on processing symbols, |
2861 | * so do the first pass and find the symbol table | |
2862 | */ | |
2863 | scn = NULL; | |
29603665 | 2864 | while ((scn = elf_nextscn(elf, scn)) != NULL) { |
0201c575 AN |
2865 | if (elf_sec_hdr(obj, scn, &sh)) |
2866 | return -LIBBPF_ERRNO__FORMAT; | |
2867 | ||
2868 | if (sh.sh_type == SHT_SYMTAB) { | |
2869 | if (obj->efile.symbols) { | |
2870 | pr_warn("elf: multiple symbol tables in %s\n", obj->path); | |
2871 | return -LIBBPF_ERRNO__FORMAT; | |
2872 | } | |
29603665 | 2873 | |
0201c575 AN |
2874 | data = elf_sec_data(obj, scn); |
2875 | if (!data) | |
2876 | return -LIBBPF_ERRNO__FORMAT; | |
2877 | ||
2878 | obj->efile.symbols = data; | |
2879 | obj->efile.symbols_shndx = elf_ndxscn(scn); | |
2880 | obj->efile.strtabidx = sh.sh_link; | |
2881 | } | |
2882 | } | |
2883 | ||
2884 | scn = NULL; | |
2885 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
29603665 | 2886 | idx++; |
88a82120 AN |
2887 | |
2888 | if (elf_sec_hdr(obj, scn, &sh)) | |
01b29d1d | 2889 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 2890 | |
88a82120 AN |
2891 | name = elf_sec_str(obj, sh.sh_name); |
2892 | if (!name) | |
01b29d1d | 2893 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 2894 | |
50e09460 AN |
2895 | if (ignore_elf_section(&sh, name)) |
2896 | continue; | |
2897 | ||
88a82120 AN |
2898 | data = elf_sec_data(obj, scn); |
2899 | if (!data) | |
01b29d1d | 2900 | return -LIBBPF_ERRNO__FORMAT; |
88a82120 AN |
2901 | |
2902 | pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n", | |
077c066a | 2903 | idx, name, (unsigned long)data->d_size, |
29603665 WN |
2904 | (int)sh.sh_link, (unsigned long)sh.sh_flags, |
2905 | (int)sh.sh_type); | |
cb1e5e96 | 2906 | |
1713d68b | 2907 | if (strcmp(name, "license") == 0) { |
88a82120 | 2908 | err = bpf_object__init_license(obj, data->d_buf, data->d_size); |
01b29d1d AN |
2909 | if (err) |
2910 | return err; | |
1713d68b | 2911 | } else if (strcmp(name, "version") == 0) { |
88a82120 | 2912 | err = bpf_object__init_kversion(obj, data->d_buf, data->d_size); |
54b8625c JF |
2913 | if (err) |
2914 | return err; | |
1713d68b | 2915 | } else if (strcmp(name, "maps") == 0) { |
666810e8 | 2916 | obj->efile.maps_shndx = idx; |
abd29c93 AN |
2917 | } else if (strcmp(name, MAPS_ELF_SEC) == 0) { |
2918 | obj->efile.btf_maps_shndx = idx; | |
1713d68b DB |
2919 | } else if (strcmp(name, BTF_ELF_SEC) == 0) { |
2920 | btf_data = data; | |
2993e051 | 2921 | } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) { |
f0187f0b | 2922 | btf_ext_data = data; |
8a138aed | 2923 | } else if (sh.sh_type == SHT_SYMTAB) { |
0201c575 | 2924 | /* already processed during the first pass above */ |
f8c7a4d4 JS |
2925 | } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) { |
2926 | if (sh.sh_flags & SHF_EXECINSTR) { | |
2927 | if (strcmp(name, ".text") == 0) | |
2928 | obj->efile.text_shndx = idx; | |
c1122392 | 2929 | err = bpf_object__add_programs(obj, data, name, idx); |
88a82120 | 2930 | if (err) |
01b29d1d | 2931 | return err; |
ac9d1389 | 2932 | } else if (strcmp(name, DATA_SEC) == 0) { |
d859900c DB |
2933 | obj->efile.data = data; |
2934 | obj->efile.data_shndx = idx; | |
ac9d1389 | 2935 | } else if (strcmp(name, RODATA_SEC) == 0) { |
d859900c DB |
2936 | obj->efile.rodata = data; |
2937 | obj->efile.rodata_shndx = idx; | |
590a0088 MKL |
2938 | } else if (strcmp(name, STRUCT_OPS_SEC) == 0) { |
2939 | obj->efile.st_ops_data = data; | |
2940 | obj->efile.st_ops_shndx = idx; | |
d859900c | 2941 | } else { |
50e09460 AN |
2942 | pr_info("elf: skipping unrecognized data section(%d) %s\n", |
2943 | idx, name); | |
a5b8bd47 | 2944 | } |
b62f06e8 | 2945 | } else if (sh.sh_type == SHT_REL) { |
1f8e2bcb AN |
2946 | int nr_sects = obj->efile.nr_reloc_sects; |
2947 | void *sects = obj->efile.reloc_sects; | |
e3d91b0c JDB |
2948 | int sec = sh.sh_info; /* points to other section */ |
2949 | ||
2950 | /* Only do relo for section with exec instructions */ | |
590a0088 | 2951 | if (!section_have_execinstr(obj, sec) && |
646f02ff AN |
2952 | strcmp(name, ".rel" STRUCT_OPS_SEC) && |
2953 | strcmp(name, ".rel" MAPS_ELF_SEC)) { | |
50e09460 AN |
2954 | pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n", |
2955 | idx, name, sec, | |
2956 | elf_sec_name(obj, elf_sec_by_idx(obj, sec)) ?: "<?>"); | |
e3d91b0c JDB |
2957 | continue; |
2958 | } | |
b62f06e8 | 2959 | |
029258d7 AN |
2960 | sects = libbpf_reallocarray(sects, nr_sects + 1, |
2961 | sizeof(*obj->efile.reloc_sects)); | |
88a82120 | 2962 | if (!sects) |
01b29d1d | 2963 | return -ENOMEM; |
b62f06e8 | 2964 | |
1f8e2bcb AN |
2965 | obj->efile.reloc_sects = sects; |
2966 | obj->efile.nr_reloc_sects++; | |
b62f06e8 | 2967 | |
1f8e2bcb AN |
2968 | obj->efile.reloc_sects[nr_sects].shdr = sh; |
2969 | obj->efile.reloc_sects[nr_sects].data = data; | |
88a82120 | 2970 | } else if (sh.sh_type == SHT_NOBITS && strcmp(name, BSS_SEC) == 0) { |
d859900c DB |
2971 | obj->efile.bss = data; |
2972 | obj->efile.bss_shndx = idx; | |
077c066a | 2973 | } else { |
2e80be60 AN |
2974 | pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name, |
2975 | (size_t)sh.sh_size); | |
bec7d68c | 2976 | } |
29603665 | 2977 | } |
561bbcca | 2978 | |
d3a3aa0c | 2979 | if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) { |
88a82120 | 2980 | pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path); |
f102154d | 2981 | return -LIBBPF_ERRNO__FORMAT; |
77ba9a5b | 2982 | } |
db2b8b06 AN |
2983 | |
2984 | /* sort BPF programs by section name and in-section instruction offset | |
2985 | * for faster search */ | |
2986 | qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs); | |
2987 | ||
0d13bfce | 2988 | return bpf_object__init_btf(obj, btf_data, btf_ext_data); |
29603665 WN |
2989 | } |
2990 | ||
166750bc AN |
2991 | static bool sym_is_extern(const GElf_Sym *sym) |
2992 | { | |
2993 | int bind = GELF_ST_BIND(sym->st_info); | |
2994 | /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */ | |
2995 | return sym->st_shndx == SHN_UNDEF && | |
2996 | (bind == STB_GLOBAL || bind == STB_WEAK) && | |
2997 | GELF_ST_TYPE(sym->st_info) == STT_NOTYPE; | |
2998 | } | |
2999 | ||
53eddb5e YS |
3000 | static bool sym_is_subprog(const GElf_Sym *sym, int text_shndx) |
3001 | { | |
3002 | int bind = GELF_ST_BIND(sym->st_info); | |
3003 | int type = GELF_ST_TYPE(sym->st_info); | |
3004 | ||
3005 | /* in .text section */ | |
3006 | if (sym->st_shndx != text_shndx) | |
3007 | return false; | |
3008 | ||
3009 | /* local function */ | |
3010 | if (bind == STB_LOCAL && type == STT_SECTION) | |
3011 | return true; | |
3012 | ||
3013 | /* global function */ | |
3014 | return bind == STB_GLOBAL && type == STT_FUNC; | |
3015 | } | |
3016 | ||
166750bc AN |
3017 | static int find_extern_btf_id(const struct btf *btf, const char *ext_name) |
3018 | { | |
3019 | const struct btf_type *t; | |
3020 | const char *var_name; | |
3021 | int i, n; | |
3022 | ||
3023 | if (!btf) | |
3024 | return -ESRCH; | |
3025 | ||
3026 | n = btf__get_nr_types(btf); | |
3027 | for (i = 1; i <= n; i++) { | |
3028 | t = btf__type_by_id(btf, i); | |
3029 | ||
3030 | if (!btf_is_var(t)) | |
3031 | continue; | |
3032 | ||
3033 | var_name = btf__name_by_offset(btf, t->name_off); | |
3034 | if (strcmp(var_name, ext_name)) | |
3035 | continue; | |
3036 | ||
3037 | if (btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN) | |
3038 | return -EINVAL; | |
3039 | ||
3040 | return i; | |
3041 | } | |
3042 | ||
3043 | return -ENOENT; | |
3044 | } | |
3045 | ||
2e33efe3 AN |
3046 | static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) { |
3047 | const struct btf_var_secinfo *vs; | |
3048 | const struct btf_type *t; | |
3049 | int i, j, n; | |
3050 | ||
3051 | if (!btf) | |
3052 | return -ESRCH; | |
3053 | ||
3054 | n = btf__get_nr_types(btf); | |
3055 | for (i = 1; i <= n; i++) { | |
3056 | t = btf__type_by_id(btf, i); | |
3057 | ||
3058 | if (!btf_is_datasec(t)) | |
3059 | continue; | |
3060 | ||
3061 | vs = btf_var_secinfos(t); | |
3062 | for (j = 0; j < btf_vlen(t); j++, vs++) { | |
3063 | if (vs->type == ext_btf_id) | |
3064 | return i; | |
3065 | } | |
3066 | } | |
3067 | ||
3068 | return -ENOENT; | |
3069 | } | |
3070 | ||
3071 | static enum kcfg_type find_kcfg_type(const struct btf *btf, int id, | |
3072 | bool *is_signed) | |
166750bc AN |
3073 | { |
3074 | const struct btf_type *t; | |
3075 | const char *name; | |
3076 | ||
3077 | t = skip_mods_and_typedefs(btf, id, NULL); | |
3078 | name = btf__name_by_offset(btf, t->name_off); | |
3079 | ||
3080 | if (is_signed) | |
3081 | *is_signed = false; | |
3082 | switch (btf_kind(t)) { | |
3083 | case BTF_KIND_INT: { | |
3084 | int enc = btf_int_encoding(t); | |
3085 | ||
3086 | if (enc & BTF_INT_BOOL) | |
2e33efe3 | 3087 | return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN; |
166750bc AN |
3088 | if (is_signed) |
3089 | *is_signed = enc & BTF_INT_SIGNED; | |
3090 | if (t->size == 1) | |
2e33efe3 | 3091 | return KCFG_CHAR; |
166750bc | 3092 | if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1))) |
2e33efe3 AN |
3093 | return KCFG_UNKNOWN; |
3094 | return KCFG_INT; | |
166750bc AN |
3095 | } |
3096 | case BTF_KIND_ENUM: | |
3097 | if (t->size != 4) | |
2e33efe3 | 3098 | return KCFG_UNKNOWN; |
166750bc | 3099 | if (strcmp(name, "libbpf_tristate")) |
2e33efe3 AN |
3100 | return KCFG_UNKNOWN; |
3101 | return KCFG_TRISTATE; | |
166750bc AN |
3102 | case BTF_KIND_ARRAY: |
3103 | if (btf_array(t)->nelems == 0) | |
2e33efe3 AN |
3104 | return KCFG_UNKNOWN; |
3105 | if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR) | |
3106 | return KCFG_UNKNOWN; | |
3107 | return KCFG_CHAR_ARR; | |
166750bc | 3108 | default: |
2e33efe3 | 3109 | return KCFG_UNKNOWN; |
166750bc AN |
3110 | } |
3111 | } | |
3112 | ||
3113 | static int cmp_externs(const void *_a, const void *_b) | |
3114 | { | |
3115 | const struct extern_desc *a = _a; | |
3116 | const struct extern_desc *b = _b; | |
3117 | ||
2e33efe3 AN |
3118 | if (a->type != b->type) |
3119 | return a->type < b->type ? -1 : 1; | |
3120 | ||
3121 | if (a->type == EXT_KCFG) { | |
3122 | /* descending order by alignment requirements */ | |
3123 | if (a->kcfg.align != b->kcfg.align) | |
3124 | return a->kcfg.align > b->kcfg.align ? -1 : 1; | |
3125 | /* ascending order by size, within same alignment class */ | |
3126 | if (a->kcfg.sz != b->kcfg.sz) | |
3127 | return a->kcfg.sz < b->kcfg.sz ? -1 : 1; | |
3128 | } | |
3129 | ||
166750bc AN |
3130 | /* resolve ties by name */ |
3131 | return strcmp(a->name, b->name); | |
3132 | } | |
3133 | ||
1c0c7074 AN |
3134 | static int find_int_btf_id(const struct btf *btf) |
3135 | { | |
3136 | const struct btf_type *t; | |
3137 | int i, n; | |
3138 | ||
3139 | n = btf__get_nr_types(btf); | |
3140 | for (i = 1; i <= n; i++) { | |
3141 | t = btf__type_by_id(btf, i); | |
3142 | ||
3143 | if (btf_is_int(t) && btf_int_bits(t) == 32) | |
3144 | return i; | |
3145 | } | |
3146 | ||
3147 | return 0; | |
3148 | } | |
3149 | ||
166750bc AN |
3150 | static int bpf_object__collect_externs(struct bpf_object *obj) |
3151 | { | |
1c0c7074 | 3152 | struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL; |
166750bc AN |
3153 | const struct btf_type *t; |
3154 | struct extern_desc *ext; | |
2e33efe3 AN |
3155 | int i, n, off; |
3156 | const char *ext_name, *sec_name; | |
166750bc AN |
3157 | Elf_Scn *scn; |
3158 | GElf_Shdr sh; | |
3159 | ||
3160 | if (!obj->efile.symbols) | |
3161 | return 0; | |
3162 | ||
88a82120 AN |
3163 | scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx); |
3164 | if (elf_sec_hdr(obj, scn, &sh)) | |
166750bc | 3165 | return -LIBBPF_ERRNO__FORMAT; |
166750bc | 3166 | |
88a82120 | 3167 | n = sh.sh_size / sh.sh_entsize; |
166750bc | 3168 | pr_debug("looking for externs among %d symbols...\n", n); |
88a82120 | 3169 | |
166750bc AN |
3170 | for (i = 0; i < n; i++) { |
3171 | GElf_Sym sym; | |
3172 | ||
3173 | if (!gelf_getsym(obj->efile.symbols, i, &sym)) | |
3174 | return -LIBBPF_ERRNO__FORMAT; | |
3175 | if (!sym_is_extern(&sym)) | |
3176 | continue; | |
88a82120 | 3177 | ext_name = elf_sym_str(obj, sym.st_name); |
166750bc AN |
3178 | if (!ext_name || !ext_name[0]) |
3179 | continue; | |
3180 | ||
3181 | ext = obj->externs; | |
029258d7 | 3182 | ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext)); |
166750bc AN |
3183 | if (!ext) |
3184 | return -ENOMEM; | |
3185 | obj->externs = ext; | |
3186 | ext = &ext[obj->nr_extern]; | |
3187 | memset(ext, 0, sizeof(*ext)); | |
3188 | obj->nr_extern++; | |
3189 | ||
3190 | ext->btf_id = find_extern_btf_id(obj->btf, ext_name); | |
3191 | if (ext->btf_id <= 0) { | |
3192 | pr_warn("failed to find BTF for extern '%s': %d\n", | |
3193 | ext_name, ext->btf_id); | |
3194 | return ext->btf_id; | |
3195 | } | |
3196 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
3197 | ext->name = btf__name_by_offset(obj->btf, t->name_off); | |
3198 | ext->sym_idx = i; | |
3199 | ext->is_weak = GELF_ST_BIND(sym.st_info) == STB_WEAK; | |
2e33efe3 AN |
3200 | |
3201 | ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id); | |
3202 | if (ext->sec_btf_id <= 0) { | |
3203 | pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n", | |
3204 | ext_name, ext->btf_id, ext->sec_btf_id); | |
3205 | return ext->sec_btf_id; | |
166750bc | 3206 | } |
2e33efe3 AN |
3207 | sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id); |
3208 | sec_name = btf__name_by_offset(obj->btf, sec->name_off); | |
3209 | ||
3210 | if (strcmp(sec_name, KCONFIG_SEC) == 0) { | |
3211 | kcfg_sec = sec; | |
3212 | ext->type = EXT_KCFG; | |
3213 | ext->kcfg.sz = btf__resolve_size(obj->btf, t->type); | |
3214 | if (ext->kcfg.sz <= 0) { | |
3215 | pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n", | |
3216 | ext_name, ext->kcfg.sz); | |
3217 | return ext->kcfg.sz; | |
3218 | } | |
3219 | ext->kcfg.align = btf__align_of(obj->btf, t->type); | |
3220 | if (ext->kcfg.align <= 0) { | |
3221 | pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n", | |
3222 | ext_name, ext->kcfg.align); | |
3223 | return -EINVAL; | |
3224 | } | |
3225 | ext->kcfg.type = find_kcfg_type(obj->btf, t->type, | |
3226 | &ext->kcfg.is_signed); | |
3227 | if (ext->kcfg.type == KCFG_UNKNOWN) { | |
3228 | pr_warn("extern (kcfg) '%s' type is unsupported\n", ext_name); | |
3229 | return -ENOTSUP; | |
3230 | } | |
1c0c7074 | 3231 | } else if (strcmp(sec_name, KSYMS_SEC) == 0) { |
1c0c7074 AN |
3232 | ksym_sec = sec; |
3233 | ext->type = EXT_KSYM; | |
d370bbe1 HL |
3234 | skip_mods_and_typedefs(obj->btf, t->type, |
3235 | &ext->ksym.type_id); | |
2e33efe3 AN |
3236 | } else { |
3237 | pr_warn("unrecognized extern section '%s'\n", sec_name); | |
166750bc AN |
3238 | return -ENOTSUP; |
3239 | } | |
3240 | } | |
3241 | pr_debug("collected %d externs total\n", obj->nr_extern); | |
3242 | ||
3243 | if (!obj->nr_extern) | |
3244 | return 0; | |
3245 | ||
2e33efe3 | 3246 | /* sort externs by type, for kcfg ones also by (align, size, name) */ |
166750bc | 3247 | qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs); |
166750bc | 3248 | |
1c0c7074 AN |
3249 | /* for .ksyms section, we need to turn all externs into allocated |
3250 | * variables in BTF to pass kernel verification; we do this by | |
3251 | * pretending that each extern is a 8-byte variable | |
3252 | */ | |
3253 | if (ksym_sec) { | |
3254 | /* find existing 4-byte integer type in BTF to use for fake | |
3255 | * extern variables in DATASEC | |
3256 | */ | |
3257 | int int_btf_id = find_int_btf_id(obj->btf); | |
3258 | ||
3259 | for (i = 0; i < obj->nr_extern; i++) { | |
3260 | ext = &obj->externs[i]; | |
3261 | if (ext->type != EXT_KSYM) | |
3262 | continue; | |
3263 | pr_debug("extern (ksym) #%d: symbol %d, name %s\n", | |
3264 | i, ext->sym_idx, ext->name); | |
3265 | } | |
3266 | ||
3267 | sec = ksym_sec; | |
3268 | n = btf_vlen(sec); | |
3269 | for (i = 0, off = 0; i < n; i++, off += sizeof(int)) { | |
3270 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
3271 | struct btf_type *vt; | |
3272 | ||
3273 | vt = (void *)btf__type_by_id(obj->btf, vs->type); | |
3274 | ext_name = btf__name_by_offset(obj->btf, vt->name_off); | |
3275 | ext = find_extern_by_name(obj, ext_name); | |
3276 | if (!ext) { | |
3277 | pr_warn("failed to find extern definition for BTF var '%s'\n", | |
3278 | ext_name); | |
3279 | return -ESRCH; | |
3280 | } | |
3281 | btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
3282 | vt->type = int_btf_id; | |
3283 | vs->offset = off; | |
3284 | vs->size = sizeof(int); | |
3285 | } | |
3286 | sec->size = off; | |
3287 | } | |
3288 | ||
2e33efe3 AN |
3289 | if (kcfg_sec) { |
3290 | sec = kcfg_sec; | |
3291 | /* for kcfg externs calculate their offsets within a .kconfig map */ | |
3292 | off = 0; | |
3293 | for (i = 0; i < obj->nr_extern; i++) { | |
3294 | ext = &obj->externs[i]; | |
3295 | if (ext->type != EXT_KCFG) | |
3296 | continue; | |
166750bc | 3297 | |
2e33efe3 AN |
3298 | ext->kcfg.data_off = roundup(off, ext->kcfg.align); |
3299 | off = ext->kcfg.data_off + ext->kcfg.sz; | |
1c0c7074 | 3300 | pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n", |
2e33efe3 AN |
3301 | i, ext->sym_idx, ext->kcfg.data_off, ext->name); |
3302 | } | |
3303 | sec->size = off; | |
3304 | n = btf_vlen(sec); | |
3305 | for (i = 0; i < n; i++) { | |
3306 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
3307 | ||
3308 | t = btf__type_by_id(obj->btf, vs->type); | |
3309 | ext_name = btf__name_by_offset(obj->btf, t->name_off); | |
3310 | ext = find_extern_by_name(obj, ext_name); | |
3311 | if (!ext) { | |
3312 | pr_warn("failed to find extern definition for BTF var '%s'\n", | |
3313 | ext_name); | |
3314 | return -ESRCH; | |
3315 | } | |
3316 | btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
3317 | vs->offset = ext->kcfg.data_off; | |
166750bc | 3318 | } |
166750bc | 3319 | } |
166750bc AN |
3320 | return 0; |
3321 | } | |
3322 | ||
6d4b198b | 3323 | struct bpf_program * |
a324aae3 AN |
3324 | bpf_object__find_program_by_title(const struct bpf_object *obj, |
3325 | const char *title) | |
6d4b198b JK |
3326 | { |
3327 | struct bpf_program *pos; | |
3328 | ||
3329 | bpf_object__for_each_program(pos, obj) { | |
52109584 | 3330 | if (pos->sec_name && !strcmp(pos->sec_name, title)) |
6d4b198b JK |
3331 | return pos; |
3332 | } | |
3333 | return NULL; | |
3334 | } | |
3335 | ||
c3c55696 AN |
3336 | static bool prog_is_subprog(const struct bpf_object *obj, |
3337 | const struct bpf_program *prog) | |
3338 | { | |
197afc63 AN |
3339 | /* For legacy reasons, libbpf supports an entry-point BPF programs |
3340 | * without SEC() attribute, i.e., those in the .text section. But if | |
3341 | * there are 2 or more such programs in the .text section, they all | |
3342 | * must be subprograms called from entry-point BPF programs in | |
3343 | * designated SEC()'tions, otherwise there is no way to distinguish | |
3344 | * which of those programs should be loaded vs which are a subprogram. | |
3345 | * Similarly, if there is a function/program in .text and at least one | |
3346 | * other BPF program with custom SEC() attribute, then we just assume | |
3347 | * .text programs are subprograms (even if they are not called from | |
3348 | * other programs), because libbpf never explicitly supported mixing | |
3349 | * SEC()-designated BPF programs and .text entry-point BPF programs. | |
3350 | */ | |
3351 | return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1; | |
c3c55696 AN |
3352 | } |
3353 | ||
01af3bf0 AN |
3354 | struct bpf_program * |
3355 | bpf_object__find_program_by_name(const struct bpf_object *obj, | |
3356 | const char *name) | |
3357 | { | |
3358 | struct bpf_program *prog; | |
3359 | ||
3360 | bpf_object__for_each_program(prog, obj) { | |
c3c55696 AN |
3361 | if (prog_is_subprog(obj, prog)) |
3362 | continue; | |
01af3bf0 AN |
3363 | if (!strcmp(prog->name, name)) |
3364 | return prog; | |
3365 | } | |
3366 | return NULL; | |
3367 | } | |
3368 | ||
d859900c DB |
3369 | static bool bpf_object__shndx_is_data(const struct bpf_object *obj, |
3370 | int shndx) | |
3371 | { | |
3372 | return shndx == obj->efile.data_shndx || | |
3373 | shndx == obj->efile.bss_shndx || | |
3374 | shndx == obj->efile.rodata_shndx; | |
3375 | } | |
3376 | ||
3377 | static bool bpf_object__shndx_is_maps(const struct bpf_object *obj, | |
3378 | int shndx) | |
3379 | { | |
abd29c93 AN |
3380 | return shndx == obj->efile.maps_shndx || |
3381 | shndx == obj->efile.btf_maps_shndx; | |
d859900c DB |
3382 | } |
3383 | ||
d859900c DB |
3384 | static enum libbpf_map_type |
3385 | bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx) | |
3386 | { | |
3387 | if (shndx == obj->efile.data_shndx) | |
3388 | return LIBBPF_MAP_DATA; | |
3389 | else if (shndx == obj->efile.bss_shndx) | |
3390 | return LIBBPF_MAP_BSS; | |
3391 | else if (shndx == obj->efile.rodata_shndx) | |
3392 | return LIBBPF_MAP_RODATA; | |
166750bc | 3393 | else if (shndx == obj->efile.symbols_shndx) |
81bfdd08 | 3394 | return LIBBPF_MAP_KCONFIG; |
d859900c DB |
3395 | else |
3396 | return LIBBPF_MAP_UNSPEC; | |
3397 | } | |
3398 | ||
1f8e2bcb AN |
3399 | static int bpf_program__record_reloc(struct bpf_program *prog, |
3400 | struct reloc_desc *reloc_desc, | |
9c0f8cbd | 3401 | __u32 insn_idx, const char *sym_name, |
1f8e2bcb AN |
3402 | const GElf_Sym *sym, const GElf_Rel *rel) |
3403 | { | |
3404 | struct bpf_insn *insn = &prog->insns[insn_idx]; | |
3405 | size_t map_idx, nr_maps = prog->obj->nr_maps; | |
3406 | struct bpf_object *obj = prog->obj; | |
3407 | __u32 shdr_idx = sym->st_shndx; | |
3408 | enum libbpf_map_type type; | |
9c0f8cbd | 3409 | const char *sym_sec_name; |
1f8e2bcb AN |
3410 | struct bpf_map *map; |
3411 | ||
c3c55696 AN |
3412 | reloc_desc->processed = false; |
3413 | ||
1f8e2bcb AN |
3414 | /* sub-program call relocation */ |
3415 | if (insn->code == (BPF_JMP | BPF_CALL)) { | |
3416 | if (insn->src_reg != BPF_PSEUDO_CALL) { | |
9c0f8cbd | 3417 | pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name); |
1f8e2bcb AN |
3418 | return -LIBBPF_ERRNO__RELOC; |
3419 | } | |
3420 | /* text_shndx can be 0, if no default "main" program exists */ | |
3421 | if (!shdr_idx || shdr_idx != obj->efile.text_shndx) { | |
9c0f8cbd AN |
3422 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); |
3423 | pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n", | |
3424 | prog->name, sym_name, sym_sec_name); | |
1f8e2bcb AN |
3425 | return -LIBBPF_ERRNO__RELOC; |
3426 | } | |
9c0f8cbd AN |
3427 | if (sym->st_value % BPF_INSN_SZ) { |
3428 | pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n", | |
3429 | prog->name, sym_name, (size_t)sym->st_value); | |
1f8e2bcb AN |
3430 | return -LIBBPF_ERRNO__RELOC; |
3431 | } | |
3432 | reloc_desc->type = RELO_CALL; | |
3433 | reloc_desc->insn_idx = insn_idx; | |
53f8dd43 | 3434 | reloc_desc->sym_off = sym->st_value; |
1f8e2bcb AN |
3435 | return 0; |
3436 | } | |
3437 | ||
b8f871fa | 3438 | if (!is_ldimm64(insn)) { |
9c0f8cbd AN |
3439 | pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n", |
3440 | prog->name, sym_name, insn_idx, insn->code); | |
1f8e2bcb AN |
3441 | return -LIBBPF_ERRNO__RELOC; |
3442 | } | |
166750bc AN |
3443 | |
3444 | if (sym_is_extern(sym)) { | |
3445 | int sym_idx = GELF_R_SYM(rel->r_info); | |
3446 | int i, n = obj->nr_extern; | |
3447 | struct extern_desc *ext; | |
3448 | ||
3449 | for (i = 0; i < n; i++) { | |
3450 | ext = &obj->externs[i]; | |
3451 | if (ext->sym_idx == sym_idx) | |
3452 | break; | |
3453 | } | |
3454 | if (i >= n) { | |
9c0f8cbd AN |
3455 | pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n", |
3456 | prog->name, sym_name, sym_idx); | |
166750bc AN |
3457 | return -LIBBPF_ERRNO__RELOC; |
3458 | } | |
9c0f8cbd AN |
3459 | pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n", |
3460 | prog->name, i, ext->name, ext->sym_idx, insn_idx); | |
166750bc AN |
3461 | reloc_desc->type = RELO_EXTERN; |
3462 | reloc_desc->insn_idx = insn_idx; | |
2e33efe3 | 3463 | reloc_desc->sym_off = i; /* sym_off stores extern index */ |
166750bc AN |
3464 | return 0; |
3465 | } | |
3466 | ||
1f8e2bcb | 3467 | if (!shdr_idx || shdr_idx >= SHN_LORESERVE) { |
9c0f8cbd AN |
3468 | pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n", |
3469 | prog->name, sym_name, shdr_idx); | |
1f8e2bcb AN |
3470 | return -LIBBPF_ERRNO__RELOC; |
3471 | } | |
3472 | ||
53eddb5e YS |
3473 | /* loading subprog addresses */ |
3474 | if (sym_is_subprog(sym, obj->efile.text_shndx)) { | |
3475 | /* global_func: sym->st_value = offset in the section, insn->imm = 0. | |
3476 | * local_func: sym->st_value = 0, insn->imm = offset in the section. | |
3477 | */ | |
3478 | if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) { | |
3479 | pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n", | |
3480 | prog->name, sym_name, (size_t)sym->st_value, insn->imm); | |
3481 | return -LIBBPF_ERRNO__RELOC; | |
3482 | } | |
3483 | ||
3484 | reloc_desc->type = RELO_SUBPROG_ADDR; | |
3485 | reloc_desc->insn_idx = insn_idx; | |
3486 | reloc_desc->sym_off = sym->st_value; | |
3487 | return 0; | |
3488 | } | |
3489 | ||
1f8e2bcb | 3490 | type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx); |
9c0f8cbd | 3491 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); |
1f8e2bcb AN |
3492 | |
3493 | /* generic map reference relocation */ | |
3494 | if (type == LIBBPF_MAP_UNSPEC) { | |
3495 | if (!bpf_object__shndx_is_maps(obj, shdr_idx)) { | |
9c0f8cbd AN |
3496 | pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n", |
3497 | prog->name, sym_name, sym_sec_name); | |
1f8e2bcb AN |
3498 | return -LIBBPF_ERRNO__RELOC; |
3499 | } | |
3500 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { | |
3501 | map = &obj->maps[map_idx]; | |
3502 | if (map->libbpf_type != type || | |
3503 | map->sec_idx != sym->st_shndx || | |
3504 | map->sec_offset != sym->st_value) | |
3505 | continue; | |
9c0f8cbd AN |
3506 | pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n", |
3507 | prog->name, map_idx, map->name, map->sec_idx, | |
1f8e2bcb AN |
3508 | map->sec_offset, insn_idx); |
3509 | break; | |
3510 | } | |
3511 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
3512 | pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n", |
3513 | prog->name, sym_sec_name, (size_t)sym->st_value); | |
1f8e2bcb AN |
3514 | return -LIBBPF_ERRNO__RELOC; |
3515 | } | |
3516 | reloc_desc->type = RELO_LD64; | |
3517 | reloc_desc->insn_idx = insn_idx; | |
3518 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 3519 | reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */ |
1f8e2bcb AN |
3520 | return 0; |
3521 | } | |
3522 | ||
3523 | /* global data map relocation */ | |
3524 | if (!bpf_object__shndx_is_data(obj, shdr_idx)) { | |
9c0f8cbd AN |
3525 | pr_warn("prog '%s': bad data relo against section '%s'\n", |
3526 | prog->name, sym_sec_name); | |
1f8e2bcb | 3527 | return -LIBBPF_ERRNO__RELOC; |
1f8e2bcb | 3528 | } |
1f8e2bcb AN |
3529 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { |
3530 | map = &obj->maps[map_idx]; | |
3531 | if (map->libbpf_type != type) | |
3532 | continue; | |
9c0f8cbd AN |
3533 | pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n", |
3534 | prog->name, map_idx, map->name, map->sec_idx, | |
3535 | map->sec_offset, insn_idx); | |
1f8e2bcb AN |
3536 | break; |
3537 | } | |
3538 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
3539 | pr_warn("prog '%s': data relo failed to find map for section '%s'\n", |
3540 | prog->name, sym_sec_name); | |
1f8e2bcb AN |
3541 | return -LIBBPF_ERRNO__RELOC; |
3542 | } | |
3543 | ||
3544 | reloc_desc->type = RELO_DATA; | |
3545 | reloc_desc->insn_idx = insn_idx; | |
3546 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 3547 | reloc_desc->sym_off = sym->st_value; |
1f8e2bcb AN |
3548 | return 0; |
3549 | } | |
3550 | ||
db2b8b06 AN |
3551 | static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx) |
3552 | { | |
3553 | return insn_idx >= prog->sec_insn_off && | |
3554 | insn_idx < prog->sec_insn_off + prog->sec_insn_cnt; | |
3555 | } | |
3556 | ||
3557 | static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj, | |
3558 | size_t sec_idx, size_t insn_idx) | |
3559 | { | |
3560 | int l = 0, r = obj->nr_programs - 1, m; | |
3561 | struct bpf_program *prog; | |
3562 | ||
3563 | while (l < r) { | |
3564 | m = l + (r - l + 1) / 2; | |
3565 | prog = &obj->programs[m]; | |
3566 | ||
3567 | if (prog->sec_idx < sec_idx || | |
3568 | (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx)) | |
3569 | l = m; | |
3570 | else | |
3571 | r = m - 1; | |
3572 | } | |
3573 | /* matching program could be at index l, but it still might be the | |
3574 | * wrong one, so we need to double check conditions for the last time | |
3575 | */ | |
3576 | prog = &obj->programs[l]; | |
3577 | if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx)) | |
3578 | return prog; | |
3579 | return NULL; | |
3580 | } | |
3581 | ||
34090915 | 3582 | static int |
c3c55696 | 3583 | bpf_object__collect_prog_relos(struct bpf_object *obj, GElf_Shdr *shdr, Elf_Data *data) |
34090915 | 3584 | { |
48cca7e4 | 3585 | Elf_Data *symbols = obj->efile.symbols; |
9c0f8cbd AN |
3586 | const char *relo_sec_name, *sec_name; |
3587 | size_t sec_idx = shdr->sh_info; | |
c3c55696 AN |
3588 | struct bpf_program *prog; |
3589 | struct reloc_desc *relos; | |
1f8e2bcb | 3590 | int err, i, nrels; |
c3c55696 AN |
3591 | const char *sym_name; |
3592 | __u32 insn_idx; | |
3593 | GElf_Sym sym; | |
3594 | GElf_Rel rel; | |
34090915 | 3595 | |
9c0f8cbd AN |
3596 | relo_sec_name = elf_sec_str(obj, shdr->sh_name); |
3597 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); | |
3598 | if (!relo_sec_name || !sec_name) | |
3599 | return -EINVAL; | |
3600 | ||
3601 | pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n", | |
3602 | relo_sec_name, sec_idx, sec_name); | |
34090915 WN |
3603 | nrels = shdr->sh_size / shdr->sh_entsize; |
3604 | ||
34090915 | 3605 | for (i = 0; i < nrels; i++) { |
34090915 | 3606 | if (!gelf_getrel(data, i, &rel)) { |
9c0f8cbd | 3607 | pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i); |
6371ca3b | 3608 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 3609 | } |
399dc65e | 3610 | if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) { |
9c0f8cbd AN |
3611 | pr_warn("sec '%s': symbol 0x%zx not found for relo #%d\n", |
3612 | relo_sec_name, (size_t)GELF_R_SYM(rel.r_info), i); | |
6371ca3b | 3613 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 3614 | } |
9c0f8cbd AN |
3615 | if (rel.r_offset % BPF_INSN_SZ) { |
3616 | pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n", | |
3617 | relo_sec_name, (size_t)GELF_R_SYM(rel.r_info), i); | |
1f8e2bcb | 3618 | return -LIBBPF_ERRNO__FORMAT; |
9c0f8cbd | 3619 | } |
d859900c | 3620 | |
9c0f8cbd | 3621 | insn_idx = rel.r_offset / BPF_INSN_SZ; |
c3c55696 AN |
3622 | /* relocations against static functions are recorded as |
3623 | * relocations against the section that contains a function; | |
3624 | * in such case, symbol will be STT_SECTION and sym.st_name | |
3625 | * will point to empty string (0), so fetch section name | |
3626 | * instead | |
3627 | */ | |
3628 | if (GELF_ST_TYPE(sym.st_info) == STT_SECTION && sym.st_name == 0) | |
3629 | sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym.st_shndx)); | |
3630 | else | |
3631 | sym_name = elf_sym_str(obj, sym.st_name); | |
3632 | sym_name = sym_name ?: "<?"; | |
d859900c | 3633 | |
9c0f8cbd AN |
3634 | pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n", |
3635 | relo_sec_name, i, insn_idx, sym_name); | |
666810e8 | 3636 | |
c3c55696 AN |
3637 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); |
3638 | if (!prog) { | |
3639 | pr_warn("sec '%s': relo #%d: program not found in section '%s' for insn #%u\n", | |
3640 | relo_sec_name, i, sec_name, insn_idx); | |
3641 | return -LIBBPF_ERRNO__RELOC; | |
3642 | } | |
3643 | ||
3644 | relos = libbpf_reallocarray(prog->reloc_desc, | |
3645 | prog->nr_reloc + 1, sizeof(*relos)); | |
3646 | if (!relos) | |
3647 | return -ENOMEM; | |
3648 | prog->reloc_desc = relos; | |
3649 | ||
3650 | /* adjust insn_idx to local BPF program frame of reference */ | |
3651 | insn_idx -= prog->sec_insn_off; | |
3652 | err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc], | |
9c0f8cbd | 3653 | insn_idx, sym_name, &sym, &rel); |
1f8e2bcb AN |
3654 | if (err) |
3655 | return err; | |
c3c55696 AN |
3656 | |
3657 | prog->nr_reloc++; | |
34090915 WN |
3658 | } |
3659 | return 0; | |
3660 | } | |
3661 | ||
abd29c93 | 3662 | static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map) |
8a138aed MKL |
3663 | { |
3664 | struct bpf_map_def *def = &map->def; | |
d859900c | 3665 | __u32 key_type_id = 0, value_type_id = 0; |
96408c43 | 3666 | int ret; |
8a138aed | 3667 | |
590a0088 MKL |
3668 | /* if it's BTF-defined map, we don't need to search for type IDs. |
3669 | * For struct_ops map, it does not need btf_key_type_id and | |
3670 | * btf_value_type_id. | |
3671 | */ | |
3672 | if (map->sec_idx == obj->efile.btf_maps_shndx || | |
3673 | bpf_map__is_struct_ops(map)) | |
abd29c93 AN |
3674 | return 0; |
3675 | ||
d859900c | 3676 | if (!bpf_map__is_internal(map)) { |
abd29c93 | 3677 | ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size, |
d859900c DB |
3678 | def->value_size, &key_type_id, |
3679 | &value_type_id); | |
3680 | } else { | |
3681 | /* | |
3682 | * LLVM annotates global data differently in BTF, that is, | |
3683 | * only as '.data', '.bss' or '.rodata'. | |
3684 | */ | |
abd29c93 | 3685 | ret = btf__find_by_name(obj->btf, |
d859900c DB |
3686 | libbpf_type_to_btf_name[map->libbpf_type]); |
3687 | } | |
3688 | if (ret < 0) | |
96408c43 | 3689 | return ret; |
8a138aed | 3690 | |
96408c43 | 3691 | map->btf_key_type_id = key_type_id; |
d859900c DB |
3692 | map->btf_value_type_id = bpf_map__is_internal(map) ? |
3693 | ret : value_type_id; | |
8a138aed MKL |
3694 | return 0; |
3695 | } | |
3696 | ||
26736eb9 JK |
3697 | int bpf_map__reuse_fd(struct bpf_map *map, int fd) |
3698 | { | |
3699 | struct bpf_map_info info = {}; | |
3700 | __u32 len = sizeof(info); | |
3701 | int new_fd, err; | |
3702 | char *new_name; | |
3703 | ||
3704 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
3705 | if (err) | |
3706 | return err; | |
3707 | ||
3708 | new_name = strdup(info.name); | |
3709 | if (!new_name) | |
3710 | return -errno; | |
3711 | ||
3712 | new_fd = open("/", O_RDONLY | O_CLOEXEC); | |
d1b4574a THJ |
3713 | if (new_fd < 0) { |
3714 | err = -errno; | |
26736eb9 | 3715 | goto err_free_new_name; |
d1b4574a | 3716 | } |
26736eb9 JK |
3717 | |
3718 | new_fd = dup3(fd, new_fd, O_CLOEXEC); | |
d1b4574a THJ |
3719 | if (new_fd < 0) { |
3720 | err = -errno; | |
26736eb9 | 3721 | goto err_close_new_fd; |
d1b4574a | 3722 | } |
26736eb9 JK |
3723 | |
3724 | err = zclose(map->fd); | |
d1b4574a THJ |
3725 | if (err) { |
3726 | err = -errno; | |
26736eb9 | 3727 | goto err_close_new_fd; |
d1b4574a | 3728 | } |
26736eb9 JK |
3729 | free(map->name); |
3730 | ||
3731 | map->fd = new_fd; | |
3732 | map->name = new_name; | |
3733 | map->def.type = info.type; | |
3734 | map->def.key_size = info.key_size; | |
3735 | map->def.value_size = info.value_size; | |
3736 | map->def.max_entries = info.max_entries; | |
3737 | map->def.map_flags = info.map_flags; | |
3738 | map->btf_key_type_id = info.btf_key_type_id; | |
3739 | map->btf_value_type_id = info.btf_value_type_id; | |
ec6d5f47 | 3740 | map->reused = true; |
26736eb9 JK |
3741 | |
3742 | return 0; | |
3743 | ||
3744 | err_close_new_fd: | |
3745 | close(new_fd); | |
3746 | err_free_new_name: | |
3747 | free(new_name); | |
d1b4574a | 3748 | return err; |
26736eb9 JK |
3749 | } |
3750 | ||
1bdb6c9a | 3751 | __u32 bpf_map__max_entries(const struct bpf_map *map) |
1a11a4c7 | 3752 | { |
1bdb6c9a AN |
3753 | return map->def.max_entries; |
3754 | } | |
1a11a4c7 | 3755 | |
1bdb6c9a AN |
3756 | int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries) |
3757 | { | |
1a11a4c7 AI |
3758 | if (map->fd >= 0) |
3759 | return -EBUSY; | |
1a11a4c7 | 3760 | map->def.max_entries = max_entries; |
1a11a4c7 AI |
3761 | return 0; |
3762 | } | |
3763 | ||
1bdb6c9a AN |
3764 | int bpf_map__resize(struct bpf_map *map, __u32 max_entries) |
3765 | { | |
3766 | if (!map || !max_entries) | |
3767 | return -EINVAL; | |
3768 | ||
3769 | return bpf_map__set_max_entries(map, max_entries); | |
3770 | } | |
3771 | ||
47eff617 | 3772 | static int |
fd9eef1a | 3773 | bpf_object__probe_loading(struct bpf_object *obj) |
47eff617 SF |
3774 | { |
3775 | struct bpf_load_program_attr attr; | |
3776 | char *cp, errmsg[STRERR_BUFSIZE]; | |
3777 | struct bpf_insn insns[] = { | |
3778 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
3779 | BPF_EXIT_INSN(), | |
3780 | }; | |
3781 | int ret; | |
3782 | ||
3783 | /* make sure basic loading works */ | |
3784 | ||
3785 | memset(&attr, 0, sizeof(attr)); | |
3786 | attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; | |
3787 | attr.insns = insns; | |
3788 | attr.insns_cnt = ARRAY_SIZE(insns); | |
3789 | attr.license = "GPL"; | |
3790 | ||
3791 | ret = bpf_load_program_xattr(&attr, NULL, 0); | |
3792 | if (ret < 0) { | |
fd9eef1a EC |
3793 | ret = errno; |
3794 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
3795 | pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF " | |
3796 | "program. Make sure your kernel supports BPF " | |
3797 | "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is " | |
3798 | "set to big enough value.\n", __func__, cp, ret); | |
3799 | return -ret; | |
47eff617 SF |
3800 | } |
3801 | close(ret); | |
3802 | ||
fd9eef1a EC |
3803 | return 0; |
3804 | } | |
3805 | ||
bb180fb2 AN |
3806 | static int probe_fd(int fd) |
3807 | { | |
3808 | if (fd >= 0) | |
3809 | close(fd); | |
3810 | return fd >= 0; | |
3811 | } | |
3812 | ||
47b6cb4d | 3813 | static int probe_kern_prog_name(void) |
fd9eef1a EC |
3814 | { |
3815 | struct bpf_load_program_attr attr; | |
3816 | struct bpf_insn insns[] = { | |
3817 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
3818 | BPF_EXIT_INSN(), | |
3819 | }; | |
3820 | int ret; | |
3821 | ||
3822 | /* make sure loading with name works */ | |
47eff617 | 3823 | |
fd9eef1a EC |
3824 | memset(&attr, 0, sizeof(attr)); |
3825 | attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; | |
3826 | attr.insns = insns; | |
3827 | attr.insns_cnt = ARRAY_SIZE(insns); | |
3828 | attr.license = "GPL"; | |
47eff617 SF |
3829 | attr.name = "test"; |
3830 | ret = bpf_load_program_xattr(&attr, NULL, 0); | |
bb180fb2 | 3831 | return probe_fd(ret); |
47eff617 SF |
3832 | } |
3833 | ||
47b6cb4d | 3834 | static int probe_kern_global_data(void) |
8837fe5d DB |
3835 | { |
3836 | struct bpf_load_program_attr prg_attr; | |
3837 | struct bpf_create_map_attr map_attr; | |
3838 | char *cp, errmsg[STRERR_BUFSIZE]; | |
3839 | struct bpf_insn insns[] = { | |
3840 | BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16), | |
3841 | BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42), | |
3842 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
3843 | BPF_EXIT_INSN(), | |
3844 | }; | |
3845 | int ret, map; | |
3846 | ||
3847 | memset(&map_attr, 0, sizeof(map_attr)); | |
3848 | map_attr.map_type = BPF_MAP_TYPE_ARRAY; | |
3849 | map_attr.key_size = sizeof(int); | |
3850 | map_attr.value_size = 32; | |
3851 | map_attr.max_entries = 1; | |
3852 | ||
3853 | map = bpf_create_map_xattr(&map_attr); | |
3854 | if (map < 0) { | |
23ab656b THJ |
3855 | ret = -errno; |
3856 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
be18010e | 3857 | pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n", |
23ab656b THJ |
3858 | __func__, cp, -ret); |
3859 | return ret; | |
8837fe5d DB |
3860 | } |
3861 | ||
3862 | insns[0].imm = map; | |
3863 | ||
3864 | memset(&prg_attr, 0, sizeof(prg_attr)); | |
3865 | prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; | |
3866 | prg_attr.insns = insns; | |
3867 | prg_attr.insns_cnt = ARRAY_SIZE(insns); | |
3868 | prg_attr.license = "GPL"; | |
3869 | ||
3870 | ret = bpf_load_program_xattr(&prg_attr, NULL, 0); | |
47b6cb4d | 3871 | close(map); |
bb180fb2 | 3872 | return probe_fd(ret); |
8837fe5d DB |
3873 | } |
3874 | ||
68b08647 AN |
3875 | static int probe_kern_btf(void) |
3876 | { | |
3877 | static const char strs[] = "\0int"; | |
3878 | __u32 types[] = { | |
3879 | /* int */ | |
3880 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), | |
3881 | }; | |
3882 | ||
3883 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
3884 | strs, sizeof(strs))); | |
3885 | } | |
3886 | ||
47b6cb4d | 3887 | static int probe_kern_btf_func(void) |
d7c4b398 | 3888 | { |
8983b731 | 3889 | static const char strs[] = "\0int\0x\0a"; |
d7c4b398 AN |
3890 | /* void x(int a) {} */ |
3891 | __u32 types[] = { | |
3892 | /* int */ | |
3893 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
3894 | /* FUNC_PROTO */ /* [2] */ | |
3895 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0), | |
3896 | BTF_PARAM_ENC(7, 1), | |
3897 | /* FUNC x */ /* [3] */ | |
3898 | BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2), | |
3899 | }; | |
d7c4b398 | 3900 | |
bb180fb2 AN |
3901 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
3902 | strs, sizeof(strs))); | |
d7c4b398 AN |
3903 | } |
3904 | ||
47b6cb4d | 3905 | static int probe_kern_btf_func_global(void) |
2d3eb67f AS |
3906 | { |
3907 | static const char strs[] = "\0int\0x\0a"; | |
3908 | /* static void x(int a) {} */ | |
3909 | __u32 types[] = { | |
3910 | /* int */ | |
3911 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
3912 | /* FUNC_PROTO */ /* [2] */ | |
3913 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0), | |
3914 | BTF_PARAM_ENC(7, 1), | |
3915 | /* FUNC x BTF_FUNC_GLOBAL */ /* [3] */ | |
3916 | BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2), | |
3917 | }; | |
2d3eb67f | 3918 | |
bb180fb2 AN |
3919 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
3920 | strs, sizeof(strs))); | |
2d3eb67f AS |
3921 | } |
3922 | ||
47b6cb4d | 3923 | static int probe_kern_btf_datasec(void) |
d7c4b398 | 3924 | { |
8983b731 | 3925 | static const char strs[] = "\0x\0.data"; |
d7c4b398 AN |
3926 | /* static int a; */ |
3927 | __u32 types[] = { | |
3928 | /* int */ | |
3929 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
3930 | /* VAR x */ /* [2] */ | |
3931 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1), | |
3932 | BTF_VAR_STATIC, | |
3933 | /* DATASEC val */ /* [3] */ | |
3934 | BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4), | |
3935 | BTF_VAR_SECINFO_ENC(2, 0, 4), | |
3936 | }; | |
cfd49210 | 3937 | |
bb180fb2 AN |
3938 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
3939 | strs, sizeof(strs))); | |
d7c4b398 AN |
3940 | } |
3941 | ||
22541a9e IL |
3942 | static int probe_kern_btf_float(void) |
3943 | { | |
3944 | static const char strs[] = "\0float"; | |
3945 | __u32 types[] = { | |
3946 | /* float */ | |
3947 | BTF_TYPE_FLOAT_ENC(1, 4), | |
3948 | }; | |
3949 | ||
3950 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
3951 | strs, sizeof(strs))); | |
3952 | } | |
3953 | ||
47b6cb4d | 3954 | static int probe_kern_array_mmap(void) |
7fe74b43 AN |
3955 | { |
3956 | struct bpf_create_map_attr attr = { | |
3957 | .map_type = BPF_MAP_TYPE_ARRAY, | |
3958 | .map_flags = BPF_F_MMAPABLE, | |
3959 | .key_size = sizeof(int), | |
3960 | .value_size = sizeof(int), | |
3961 | .max_entries = 1, | |
3962 | }; | |
7fe74b43 | 3963 | |
bb180fb2 | 3964 | return probe_fd(bpf_create_map_xattr(&attr)); |
7fe74b43 AN |
3965 | } |
3966 | ||
47b6cb4d | 3967 | static int probe_kern_exp_attach_type(void) |
25498a19 AN |
3968 | { |
3969 | struct bpf_load_program_attr attr; | |
3970 | struct bpf_insn insns[] = { | |
3971 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
3972 | BPF_EXIT_INSN(), | |
3973 | }; | |
25498a19 AN |
3974 | |
3975 | memset(&attr, 0, sizeof(attr)); | |
3976 | /* use any valid combination of program type and (optional) | |
3977 | * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS) | |
3978 | * to see if kernel supports expected_attach_type field for | |
3979 | * BPF_PROG_LOAD command | |
3980 | */ | |
3981 | attr.prog_type = BPF_PROG_TYPE_CGROUP_SOCK; | |
3982 | attr.expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE; | |
3983 | attr.insns = insns; | |
3984 | attr.insns_cnt = ARRAY_SIZE(insns); | |
3985 | attr.license = "GPL"; | |
3986 | ||
bb180fb2 | 3987 | return probe_fd(bpf_load_program_xattr(&attr, NULL, 0)); |
25498a19 AN |
3988 | } |
3989 | ||
109cea5a AN |
3990 | static int probe_kern_probe_read_kernel(void) |
3991 | { | |
3992 | struct bpf_load_program_attr attr; | |
3993 | struct bpf_insn insns[] = { | |
3994 | BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), /* r1 = r10 (fp) */ | |
3995 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8), /* r1 += -8 */ | |
3996 | BPF_MOV64_IMM(BPF_REG_2, 8), /* r2 = 8 */ | |
3997 | BPF_MOV64_IMM(BPF_REG_3, 0), /* r3 = 0 */ | |
3998 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel), | |
3999 | BPF_EXIT_INSN(), | |
4000 | }; | |
4001 | ||
4002 | memset(&attr, 0, sizeof(attr)); | |
4003 | attr.prog_type = BPF_PROG_TYPE_KPROBE; | |
4004 | attr.insns = insns; | |
4005 | attr.insns_cnt = ARRAY_SIZE(insns); | |
4006 | attr.license = "GPL"; | |
4007 | ||
4008 | return probe_fd(bpf_load_program_xattr(&attr, NULL, 0)); | |
4009 | } | |
4010 | ||
5d23328d YZ |
4011 | static int probe_prog_bind_map(void) |
4012 | { | |
4013 | struct bpf_load_program_attr prg_attr; | |
4014 | struct bpf_create_map_attr map_attr; | |
4015 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4016 | struct bpf_insn insns[] = { | |
4017 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4018 | BPF_EXIT_INSN(), | |
4019 | }; | |
4020 | int ret, map, prog; | |
4021 | ||
4022 | memset(&map_attr, 0, sizeof(map_attr)); | |
4023 | map_attr.map_type = BPF_MAP_TYPE_ARRAY; | |
4024 | map_attr.key_size = sizeof(int); | |
4025 | map_attr.value_size = 32; | |
4026 | map_attr.max_entries = 1; | |
4027 | ||
4028 | map = bpf_create_map_xattr(&map_attr); | |
4029 | if (map < 0) { | |
4030 | ret = -errno; | |
4031 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4032 | pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n", | |
4033 | __func__, cp, -ret); | |
4034 | return ret; | |
4035 | } | |
4036 | ||
4037 | memset(&prg_attr, 0, sizeof(prg_attr)); | |
4038 | prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; | |
4039 | prg_attr.insns = insns; | |
4040 | prg_attr.insns_cnt = ARRAY_SIZE(insns); | |
4041 | prg_attr.license = "GPL"; | |
4042 | ||
4043 | prog = bpf_load_program_xattr(&prg_attr, NULL, 0); | |
4044 | if (prog < 0) { | |
4045 | close(map); | |
4046 | return 0; | |
4047 | } | |
4048 | ||
4049 | ret = bpf_prog_bind_map(prog, map, NULL); | |
4050 | ||
4051 | close(map); | |
4052 | close(prog); | |
4053 | ||
4054 | return ret >= 0; | |
4055 | } | |
4056 | ||
4f33a53d AN |
4057 | static int probe_module_btf(void) |
4058 | { | |
4059 | static const char strs[] = "\0int"; | |
4060 | __u32 types[] = { | |
4061 | /* int */ | |
4062 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), | |
4063 | }; | |
4064 | struct bpf_btf_info info; | |
4065 | __u32 len = sizeof(info); | |
4066 | char name[16]; | |
4067 | int fd, err; | |
4068 | ||
4069 | fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs)); | |
4070 | if (fd < 0) | |
4071 | return 0; /* BTF not supported at all */ | |
4072 | ||
4073 | memset(&info, 0, sizeof(info)); | |
4074 | info.name = ptr_to_u64(name); | |
4075 | info.name_len = sizeof(name); | |
4076 | ||
4077 | /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer; | |
4078 | * kernel's module BTF support coincides with support for | |
4079 | * name/name_len fields in struct bpf_btf_info. | |
4080 | */ | |
4081 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
4082 | close(fd); | |
4083 | return !err; | |
4084 | } | |
4085 | ||
47b6cb4d AN |
4086 | enum kern_feature_result { |
4087 | FEAT_UNKNOWN = 0, | |
4088 | FEAT_SUPPORTED = 1, | |
4089 | FEAT_MISSING = 2, | |
4090 | }; | |
4091 | ||
4092 | typedef int (*feature_probe_fn)(void); | |
4093 | ||
4094 | static struct kern_feature_desc { | |
4095 | const char *desc; | |
4096 | feature_probe_fn probe; | |
4097 | enum kern_feature_result res; | |
4098 | } feature_probes[__FEAT_CNT] = { | |
4099 | [FEAT_PROG_NAME] = { | |
4100 | "BPF program name", probe_kern_prog_name, | |
4101 | }, | |
4102 | [FEAT_GLOBAL_DATA] = { | |
4103 | "global variables", probe_kern_global_data, | |
4104 | }, | |
68b08647 AN |
4105 | [FEAT_BTF] = { |
4106 | "minimal BTF", probe_kern_btf, | |
4107 | }, | |
47b6cb4d AN |
4108 | [FEAT_BTF_FUNC] = { |
4109 | "BTF functions", probe_kern_btf_func, | |
4110 | }, | |
4111 | [FEAT_BTF_GLOBAL_FUNC] = { | |
4112 | "BTF global function", probe_kern_btf_func_global, | |
4113 | }, | |
4114 | [FEAT_BTF_DATASEC] = { | |
4115 | "BTF data section and variable", probe_kern_btf_datasec, | |
4116 | }, | |
4117 | [FEAT_ARRAY_MMAP] = { | |
4118 | "ARRAY map mmap()", probe_kern_array_mmap, | |
4119 | }, | |
4120 | [FEAT_EXP_ATTACH_TYPE] = { | |
4121 | "BPF_PROG_LOAD expected_attach_type attribute", | |
4122 | probe_kern_exp_attach_type, | |
4123 | }, | |
109cea5a AN |
4124 | [FEAT_PROBE_READ_KERN] = { |
4125 | "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel, | |
5d23328d YZ |
4126 | }, |
4127 | [FEAT_PROG_BIND_MAP] = { | |
4128 | "BPF_PROG_BIND_MAP support", probe_prog_bind_map, | |
4f33a53d AN |
4129 | }, |
4130 | [FEAT_MODULE_BTF] = { | |
4131 | "module BTF support", probe_module_btf, | |
4132 | }, | |
22541a9e IL |
4133 | [FEAT_BTF_FLOAT] = { |
4134 | "BTF_KIND_FLOAT support", probe_kern_btf_float, | |
4135 | }, | |
47b6cb4d | 4136 | }; |
8837fe5d | 4137 | |
47b6cb4d AN |
4138 | static bool kernel_supports(enum kern_feature_id feat_id) |
4139 | { | |
4140 | struct kern_feature_desc *feat = &feature_probes[feat_id]; | |
4141 | int ret; | |
4142 | ||
4143 | if (READ_ONCE(feat->res) == FEAT_UNKNOWN) { | |
4144 | ret = feat->probe(); | |
4145 | if (ret > 0) { | |
4146 | WRITE_ONCE(feat->res, FEAT_SUPPORTED); | |
4147 | } else if (ret == 0) { | |
4148 | WRITE_ONCE(feat->res, FEAT_MISSING); | |
4149 | } else { | |
4150 | pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret); | |
4151 | WRITE_ONCE(feat->res, FEAT_MISSING); | |
4152 | } | |
8837fe5d DB |
4153 | } |
4154 | ||
47b6cb4d | 4155 | return READ_ONCE(feat->res) == FEAT_SUPPORTED; |
47eff617 SF |
4156 | } |
4157 | ||
57a00f41 THJ |
4158 | static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd) |
4159 | { | |
4160 | struct bpf_map_info map_info = {}; | |
4161 | char msg[STRERR_BUFSIZE]; | |
4162 | __u32 map_info_len; | |
4163 | ||
4164 | map_info_len = sizeof(map_info); | |
4165 | ||
4166 | if (bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len)) { | |
4167 | pr_warn("failed to get map info for map FD %d: %s\n", | |
4168 | map_fd, libbpf_strerror_r(errno, msg, sizeof(msg))); | |
4169 | return false; | |
4170 | } | |
4171 | ||
4172 | return (map_info.type == map->def.type && | |
4173 | map_info.key_size == map->def.key_size && | |
4174 | map_info.value_size == map->def.value_size && | |
4175 | map_info.max_entries == map->def.max_entries && | |
4176 | map_info.map_flags == map->def.map_flags); | |
4177 | } | |
4178 | ||
4179 | static int | |
4180 | bpf_object__reuse_map(struct bpf_map *map) | |
4181 | { | |
4182 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4183 | int err, pin_fd; | |
4184 | ||
4185 | pin_fd = bpf_obj_get(map->pin_path); | |
4186 | if (pin_fd < 0) { | |
4187 | err = -errno; | |
4188 | if (err == -ENOENT) { | |
4189 | pr_debug("found no pinned map to reuse at '%s'\n", | |
4190 | map->pin_path); | |
4191 | return 0; | |
4192 | } | |
4193 | ||
4194 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
4195 | pr_warn("couldn't retrieve pinned map '%s': %s\n", | |
4196 | map->pin_path, cp); | |
4197 | return err; | |
4198 | } | |
4199 | ||
4200 | if (!map_is_reuse_compat(map, pin_fd)) { | |
4201 | pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n", | |
4202 | map->pin_path); | |
4203 | close(pin_fd); | |
4204 | return -EINVAL; | |
4205 | } | |
4206 | ||
4207 | err = bpf_map__reuse_fd(map, pin_fd); | |
4208 | if (err) { | |
4209 | close(pin_fd); | |
4210 | return err; | |
4211 | } | |
4212 | map->pinned = true; | |
4213 | pr_debug("reused pinned map at '%s'\n", map->pin_path); | |
4214 | ||
4215 | return 0; | |
4216 | } | |
4217 | ||
d859900c DB |
4218 | static int |
4219 | bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map) | |
4220 | { | |
166750bc | 4221 | enum libbpf_map_type map_type = map->libbpf_type; |
d859900c DB |
4222 | char *cp, errmsg[STRERR_BUFSIZE]; |
4223 | int err, zero = 0; | |
d859900c | 4224 | |
eba9c5f4 AN |
4225 | err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0); |
4226 | if (err) { | |
4227 | err = -errno; | |
4228 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
4229 | pr_warn("Error setting initial map(%s) contents: %s\n", | |
4230 | map->name, cp); | |
4231 | return err; | |
4232 | } | |
d859900c | 4233 | |
81bfdd08 AN |
4234 | /* Freeze .rodata and .kconfig map as read-only from syscall side. */ |
4235 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) { | |
d859900c DB |
4236 | err = bpf_map_freeze(map->fd); |
4237 | if (err) { | |
eba9c5f4 AN |
4238 | err = -errno; |
4239 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e KW |
4240 | pr_warn("Error freezing map(%s) as read-only: %s\n", |
4241 | map->name, cp); | |
eba9c5f4 | 4242 | return err; |
d859900c DB |
4243 | } |
4244 | } | |
eba9c5f4 | 4245 | return 0; |
d859900c DB |
4246 | } |
4247 | ||
2d39d7c5 AN |
4248 | static void bpf_map__destroy(struct bpf_map *map); |
4249 | ||
4250 | static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map) | |
4251 | { | |
4252 | struct bpf_create_map_attr create_attr; | |
4253 | struct bpf_map_def *def = &map->def; | |
4254 | ||
4255 | memset(&create_attr, 0, sizeof(create_attr)); | |
4256 | ||
47b6cb4d | 4257 | if (kernel_supports(FEAT_PROG_NAME)) |
2d39d7c5 AN |
4258 | create_attr.name = map->name; |
4259 | create_attr.map_ifindex = map->map_ifindex; | |
4260 | create_attr.map_type = def->type; | |
4261 | create_attr.map_flags = def->map_flags; | |
4262 | create_attr.key_size = def->key_size; | |
4263 | create_attr.value_size = def->value_size; | |
1bdb6c9a | 4264 | create_attr.numa_node = map->numa_node; |
2d39d7c5 AN |
4265 | |
4266 | if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !def->max_entries) { | |
4267 | int nr_cpus; | |
4268 | ||
4269 | nr_cpus = libbpf_num_possible_cpus(); | |
4270 | if (nr_cpus < 0) { | |
4271 | pr_warn("map '%s': failed to determine number of system CPUs: %d\n", | |
4272 | map->name, nr_cpus); | |
4273 | return nr_cpus; | |
4274 | } | |
4275 | pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus); | |
4276 | create_attr.max_entries = nr_cpus; | |
4277 | } else { | |
4278 | create_attr.max_entries = def->max_entries; | |
4279 | } | |
4280 | ||
4281 | if (bpf_map__is_struct_ops(map)) | |
4282 | create_attr.btf_vmlinux_value_type_id = | |
4283 | map->btf_vmlinux_value_type_id; | |
4284 | ||
4285 | create_attr.btf_fd = 0; | |
4286 | create_attr.btf_key_type_id = 0; | |
4287 | create_attr.btf_value_type_id = 0; | |
0f0e55d8 | 4288 | if (obj->btf && btf__fd(obj->btf) >= 0 && !bpf_map_find_btf_info(obj, map)) { |
2d39d7c5 AN |
4289 | create_attr.btf_fd = btf__fd(obj->btf); |
4290 | create_attr.btf_key_type_id = map->btf_key_type_id; | |
4291 | create_attr.btf_value_type_id = map->btf_value_type_id; | |
4292 | } | |
4293 | ||
646f02ff AN |
4294 | if (bpf_map_type__is_map_in_map(def->type)) { |
4295 | if (map->inner_map) { | |
4296 | int err; | |
4297 | ||
4298 | err = bpf_object__create_map(obj, map->inner_map); | |
4299 | if (err) { | |
4300 | pr_warn("map '%s': failed to create inner map: %d\n", | |
4301 | map->name, err); | |
4302 | return err; | |
4303 | } | |
4304 | map->inner_map_fd = bpf_map__fd(map->inner_map); | |
4305 | } | |
4306 | if (map->inner_map_fd >= 0) | |
4307 | create_attr.inner_map_fd = map->inner_map_fd; | |
4308 | } | |
4309 | ||
2d39d7c5 AN |
4310 | map->fd = bpf_create_map_xattr(&create_attr); |
4311 | if (map->fd < 0 && (create_attr.btf_key_type_id || | |
4312 | create_attr.btf_value_type_id)) { | |
4313 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4314 | int err = -errno; | |
4315 | ||
4316 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
4317 | pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n", | |
4318 | map->name, cp, err); | |
4319 | create_attr.btf_fd = 0; | |
4320 | create_attr.btf_key_type_id = 0; | |
4321 | create_attr.btf_value_type_id = 0; | |
4322 | map->btf_key_type_id = 0; | |
4323 | map->btf_value_type_id = 0; | |
4324 | map->fd = bpf_create_map_xattr(&create_attr); | |
4325 | } | |
4326 | ||
4327 | if (map->fd < 0) | |
4328 | return -errno; | |
4329 | ||
646f02ff AN |
4330 | if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) { |
4331 | bpf_map__destroy(map->inner_map); | |
4332 | zfree(&map->inner_map); | |
4333 | } | |
4334 | ||
2d39d7c5 AN |
4335 | return 0; |
4336 | } | |
4337 | ||
a0f2b7ac HL |
4338 | static int init_map_slots(struct bpf_map *map) |
4339 | { | |
4340 | const struct bpf_map *targ_map; | |
4341 | unsigned int i; | |
4342 | int fd, err; | |
4343 | ||
4344 | for (i = 0; i < map->init_slots_sz; i++) { | |
4345 | if (!map->init_slots[i]) | |
4346 | continue; | |
4347 | ||
4348 | targ_map = map->init_slots[i]; | |
4349 | fd = bpf_map__fd(targ_map); | |
4350 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
4351 | if (err) { | |
4352 | err = -errno; | |
4353 | pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n", | |
4354 | map->name, i, targ_map->name, | |
4355 | fd, err); | |
4356 | return err; | |
4357 | } | |
4358 | pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n", | |
4359 | map->name, i, targ_map->name, fd); | |
4360 | } | |
4361 | ||
4362 | zfree(&map->init_slots); | |
4363 | map->init_slots_sz = 0; | |
4364 | ||
4365 | return 0; | |
4366 | } | |
4367 | ||
52d3352e WN |
4368 | static int |
4369 | bpf_object__create_maps(struct bpf_object *obj) | |
4370 | { | |
2d39d7c5 AN |
4371 | struct bpf_map *map; |
4372 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4373 | unsigned int i, j; | |
8a138aed | 4374 | int err; |
52d3352e | 4375 | |
9d759a9b | 4376 | for (i = 0; i < obj->nr_maps; i++) { |
2d39d7c5 | 4377 | map = &obj->maps[i]; |
8a138aed | 4378 | |
57a00f41 THJ |
4379 | if (map->pin_path) { |
4380 | err = bpf_object__reuse_map(map); | |
4381 | if (err) { | |
2d39d7c5 | 4382 | pr_warn("map '%s': error reusing pinned map\n", |
57a00f41 | 4383 | map->name); |
2d39d7c5 | 4384 | goto err_out; |
57a00f41 THJ |
4385 | } |
4386 | } | |
4387 | ||
26736eb9 | 4388 | if (map->fd >= 0) { |
2d39d7c5 | 4389 | pr_debug("map '%s': skipping creation (preset fd=%d)\n", |
26736eb9 | 4390 | map->name, map->fd); |
2c193d32 HL |
4391 | } else { |
4392 | err = bpf_object__create_map(obj, map); | |
4393 | if (err) | |
d859900c | 4394 | goto err_out; |
d859900c | 4395 | |
2c193d32 HL |
4396 | pr_debug("map '%s': created successfully, fd=%d\n", |
4397 | map->name, map->fd); | |
646f02ff | 4398 | |
2c193d32 HL |
4399 | if (bpf_map__is_internal(map)) { |
4400 | err = bpf_object__populate_internal_map(obj, map); | |
4401 | if (err < 0) { | |
4402 | zclose(map->fd); | |
4403 | goto err_out; | |
4404 | } | |
d859900c | 4405 | } |
646f02ff | 4406 | |
2c193d32 HL |
4407 | if (map->init_slots_sz) { |
4408 | err = init_map_slots(map); | |
4409 | if (err < 0) { | |
4410 | zclose(map->fd); | |
646f02ff AN |
4411 | goto err_out; |
4412 | } | |
646f02ff | 4413 | } |
646f02ff AN |
4414 | } |
4415 | ||
57a00f41 THJ |
4416 | if (map->pin_path && !map->pinned) { |
4417 | err = bpf_map__pin(map, NULL); | |
4418 | if (err) { | |
2d39d7c5 AN |
4419 | pr_warn("map '%s': failed to auto-pin at '%s': %d\n", |
4420 | map->name, map->pin_path, err); | |
4421 | zclose(map->fd); | |
4422 | goto err_out; | |
57a00f41 THJ |
4423 | } |
4424 | } | |
52d3352e WN |
4425 | } |
4426 | ||
52d3352e | 4427 | return 0; |
2d39d7c5 AN |
4428 | |
4429 | err_out: | |
4430 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
4431 | pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err); | |
4432 | pr_perm_msg(err); | |
4433 | for (j = 0; j < i; j++) | |
4434 | zclose(obj->maps[j].fd); | |
4435 | return err; | |
52d3352e WN |
4436 | } |
4437 | ||
ddc7c304 AN |
4438 | #define BPF_CORE_SPEC_MAX_LEN 64 |
4439 | ||
4440 | /* represents BPF CO-RE field or array element accessor */ | |
4441 | struct bpf_core_accessor { | |
4442 | __u32 type_id; /* struct/union type or array element type */ | |
4443 | __u32 idx; /* field index or array index */ | |
4444 | const char *name; /* field name or NULL for array accessor */ | |
4445 | }; | |
4446 | ||
4447 | struct bpf_core_spec { | |
4448 | const struct btf *btf; | |
4449 | /* high-level spec: named fields and array indices only */ | |
4450 | struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN]; | |
28b93c64 AN |
4451 | /* original unresolved (no skip_mods_or_typedefs) root type ID */ |
4452 | __u32 root_type_id; | |
4453 | /* CO-RE relocation kind */ | |
4454 | enum bpf_core_relo_kind relo_kind; | |
ddc7c304 AN |
4455 | /* high-level spec length */ |
4456 | int len; | |
4457 | /* raw, low-level spec: 1-to-1 with accessor spec string */ | |
4458 | int raw_spec[BPF_CORE_SPEC_MAX_LEN]; | |
4459 | /* raw spec length */ | |
4460 | int raw_len; | |
ee26dade AN |
4461 | /* field bit offset represented by spec */ |
4462 | __u32 bit_offset; | |
ddc7c304 AN |
4463 | }; |
4464 | ||
4465 | static bool str_is_empty(const char *s) | |
4466 | { | |
4467 | return !s || !s[0]; | |
4468 | } | |
4469 | ||
1b484b30 AN |
4470 | static bool is_flex_arr(const struct btf *btf, |
4471 | const struct bpf_core_accessor *acc, | |
4472 | const struct btf_array *arr) | |
4473 | { | |
4474 | const struct btf_type *t; | |
4475 | ||
4476 | /* not a flexible array, if not inside a struct or has non-zero size */ | |
4477 | if (!acc->name || arr->nelems > 0) | |
4478 | return false; | |
4479 | ||
4480 | /* has to be the last member of enclosing struct */ | |
4481 | t = btf__type_by_id(btf, acc->type_id); | |
4482 | return acc->idx == btf_vlen(t) - 1; | |
4483 | } | |
4484 | ||
28b93c64 AN |
4485 | static const char *core_relo_kind_str(enum bpf_core_relo_kind kind) |
4486 | { | |
4487 | switch (kind) { | |
4488 | case BPF_FIELD_BYTE_OFFSET: return "byte_off"; | |
4489 | case BPF_FIELD_BYTE_SIZE: return "byte_sz"; | |
4490 | case BPF_FIELD_EXISTS: return "field_exists"; | |
4491 | case BPF_FIELD_SIGNED: return "signed"; | |
4492 | case BPF_FIELD_LSHIFT_U64: return "lshift_u64"; | |
4493 | case BPF_FIELD_RSHIFT_U64: return "rshift_u64"; | |
3fc32f40 AN |
4494 | case BPF_TYPE_ID_LOCAL: return "local_type_id"; |
4495 | case BPF_TYPE_ID_TARGET: return "target_type_id"; | |
4496 | case BPF_TYPE_EXISTS: return "type_exists"; | |
4497 | case BPF_TYPE_SIZE: return "type_size"; | |
eacaaed7 AN |
4498 | case BPF_ENUMVAL_EXISTS: return "enumval_exists"; |
4499 | case BPF_ENUMVAL_VALUE: return "enumval_value"; | |
28b93c64 AN |
4500 | default: return "unknown"; |
4501 | } | |
4502 | } | |
4503 | ||
4504 | static bool core_relo_is_field_based(enum bpf_core_relo_kind kind) | |
4505 | { | |
4506 | switch (kind) { | |
4507 | case BPF_FIELD_BYTE_OFFSET: | |
4508 | case BPF_FIELD_BYTE_SIZE: | |
4509 | case BPF_FIELD_EXISTS: | |
4510 | case BPF_FIELD_SIGNED: | |
4511 | case BPF_FIELD_LSHIFT_U64: | |
4512 | case BPF_FIELD_RSHIFT_U64: | |
4513 | return true; | |
4514 | default: | |
4515 | return false; | |
4516 | } | |
4517 | } | |
4518 | ||
3fc32f40 AN |
4519 | static bool core_relo_is_type_based(enum bpf_core_relo_kind kind) |
4520 | { | |
4521 | switch (kind) { | |
4522 | case BPF_TYPE_ID_LOCAL: | |
4523 | case BPF_TYPE_ID_TARGET: | |
4524 | case BPF_TYPE_EXISTS: | |
4525 | case BPF_TYPE_SIZE: | |
4526 | return true; | |
4527 | default: | |
4528 | return false; | |
4529 | } | |
4530 | } | |
4531 | ||
eacaaed7 AN |
4532 | static bool core_relo_is_enumval_based(enum bpf_core_relo_kind kind) |
4533 | { | |
4534 | switch (kind) { | |
4535 | case BPF_ENUMVAL_EXISTS: | |
4536 | case BPF_ENUMVAL_VALUE: | |
4537 | return true; | |
4538 | default: | |
4539 | return false; | |
4540 | } | |
4541 | } | |
4542 | ||
ddc7c304 | 4543 | /* |
28b93c64 | 4544 | * Turn bpf_core_relo into a low- and high-level spec representation, |
ddc7c304 | 4545 | * validating correctness along the way, as well as calculating resulting |
ee26dade AN |
4546 | * field bit offset, specified by accessor string. Low-level spec captures |
4547 | * every single level of nestedness, including traversing anonymous | |
ddc7c304 AN |
4548 | * struct/union members. High-level one only captures semantically meaningful |
4549 | * "turning points": named fields and array indicies. | |
4550 | * E.g., for this case: | |
4551 | * | |
4552 | * struct sample { | |
4553 | * int __unimportant; | |
4554 | * struct { | |
4555 | * int __1; | |
4556 | * int __2; | |
4557 | * int a[7]; | |
4558 | * }; | |
4559 | * }; | |
4560 | * | |
4561 | * struct sample *s = ...; | |
4562 | * | |
4563 | * int x = &s->a[3]; // access string = '0:1:2:3' | |
4564 | * | |
4565 | * Low-level spec has 1:1 mapping with each element of access string (it's | |
4566 | * just a parsed access string representation): [0, 1, 2, 3]. | |
4567 | * | |
4568 | * High-level spec will capture only 3 points: | |
4569 | * - intial zero-index access by pointer (&s->... is the same as &s[0]...); | |
4570 | * - field 'a' access (corresponds to '2' in low-level spec); | |
4571 | * - array element #3 access (corresponds to '3' in low-level spec). | |
4572 | * | |
3fc32f40 AN |
4573 | * Type-based relocations (TYPE_EXISTS/TYPE_SIZE, |
4574 | * TYPE_ID_LOCAL/TYPE_ID_TARGET) don't capture any field information. Their | |
4575 | * spec and raw_spec are kept empty. | |
eacaaed7 AN |
4576 | * |
4577 | * Enum value-based relocations (ENUMVAL_EXISTS/ENUMVAL_VALUE) use access | |
4578 | * string to specify enumerator's value index that need to be relocated. | |
ddc7c304 | 4579 | */ |
28b93c64 | 4580 | static int bpf_core_parse_spec(const struct btf *btf, |
ddc7c304 AN |
4581 | __u32 type_id, |
4582 | const char *spec_str, | |
28b93c64 | 4583 | enum bpf_core_relo_kind relo_kind, |
ddc7c304 AN |
4584 | struct bpf_core_spec *spec) |
4585 | { | |
4586 | int access_idx, parsed_len, i; | |
1b484b30 | 4587 | struct bpf_core_accessor *acc; |
ddc7c304 AN |
4588 | const struct btf_type *t; |
4589 | const char *name; | |
4590 | __u32 id; | |
4591 | __s64 sz; | |
4592 | ||
4593 | if (str_is_empty(spec_str) || *spec_str == ':') | |
4594 | return -EINVAL; | |
4595 | ||
4596 | memset(spec, 0, sizeof(*spec)); | |
4597 | spec->btf = btf; | |
28b93c64 AN |
4598 | spec->root_type_id = type_id; |
4599 | spec->relo_kind = relo_kind; | |
ddc7c304 | 4600 | |
3fc32f40 AN |
4601 | /* type-based relocations don't have a field access string */ |
4602 | if (core_relo_is_type_based(relo_kind)) { | |
4603 | if (strcmp(spec_str, "0")) | |
4604 | return -EINVAL; | |
4605 | return 0; | |
4606 | } | |
4607 | ||
ddc7c304 AN |
4608 | /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */ |
4609 | while (*spec_str) { | |
4610 | if (*spec_str == ':') | |
4611 | ++spec_str; | |
4612 | if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1) | |
4613 | return -EINVAL; | |
4614 | if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN) | |
4615 | return -E2BIG; | |
4616 | spec_str += parsed_len; | |
4617 | spec->raw_spec[spec->raw_len++] = access_idx; | |
4618 | } | |
4619 | ||
4620 | if (spec->raw_len == 0) | |
4621 | return -EINVAL; | |
4622 | ||
ddc7c304 AN |
4623 | t = skip_mods_and_typedefs(btf, type_id, &id); |
4624 | if (!t) | |
4625 | return -EINVAL; | |
4626 | ||
4627 | access_idx = spec->raw_spec[0]; | |
eacaaed7 AN |
4628 | acc = &spec->spec[0]; |
4629 | acc->type_id = id; | |
4630 | acc->idx = access_idx; | |
ddc7c304 AN |
4631 | spec->len++; |
4632 | ||
eacaaed7 AN |
4633 | if (core_relo_is_enumval_based(relo_kind)) { |
4634 | if (!btf_is_enum(t) || spec->raw_len > 1 || access_idx >= btf_vlen(t)) | |
4635 | return -EINVAL; | |
4636 | ||
4637 | /* record enumerator name in a first accessor */ | |
4638 | acc->name = btf__name_by_offset(btf, btf_enum(t)[access_idx].name_off); | |
4639 | return 0; | |
4640 | } | |
4641 | ||
28b93c64 AN |
4642 | if (!core_relo_is_field_based(relo_kind)) |
4643 | return -EINVAL; | |
4644 | ||
ddc7c304 AN |
4645 | sz = btf__resolve_size(btf, id); |
4646 | if (sz < 0) | |
4647 | return sz; | |
ee26dade | 4648 | spec->bit_offset = access_idx * sz * 8; |
ddc7c304 AN |
4649 | |
4650 | for (i = 1; i < spec->raw_len; i++) { | |
4651 | t = skip_mods_and_typedefs(btf, id, &id); | |
4652 | if (!t) | |
4653 | return -EINVAL; | |
4654 | ||
4655 | access_idx = spec->raw_spec[i]; | |
1b484b30 | 4656 | acc = &spec->spec[spec->len]; |
ddc7c304 AN |
4657 | |
4658 | if (btf_is_composite(t)) { | |
4659 | const struct btf_member *m; | |
ee26dade | 4660 | __u32 bit_offset; |
ddc7c304 AN |
4661 | |
4662 | if (access_idx >= btf_vlen(t)) | |
4663 | return -EINVAL; | |
ddc7c304 | 4664 | |
ee26dade AN |
4665 | bit_offset = btf_member_bit_offset(t, access_idx); |
4666 | spec->bit_offset += bit_offset; | |
ddc7c304 AN |
4667 | |
4668 | m = btf_members(t) + access_idx; | |
4669 | if (m->name_off) { | |
4670 | name = btf__name_by_offset(btf, m->name_off); | |
4671 | if (str_is_empty(name)) | |
4672 | return -EINVAL; | |
4673 | ||
1b484b30 AN |
4674 | acc->type_id = id; |
4675 | acc->idx = access_idx; | |
4676 | acc->name = name; | |
ddc7c304 AN |
4677 | spec->len++; |
4678 | } | |
4679 | ||
4680 | id = m->type; | |
4681 | } else if (btf_is_array(t)) { | |
4682 | const struct btf_array *a = btf_array(t); | |
1b484b30 | 4683 | bool flex; |
ddc7c304 AN |
4684 | |
4685 | t = skip_mods_and_typedefs(btf, a->type, &id); | |
1b484b30 AN |
4686 | if (!t) |
4687 | return -EINVAL; | |
4688 | ||
4689 | flex = is_flex_arr(btf, acc - 1, a); | |
4690 | if (!flex && access_idx >= a->nelems) | |
ddc7c304 AN |
4691 | return -EINVAL; |
4692 | ||
4693 | spec->spec[spec->len].type_id = id; | |
4694 | spec->spec[spec->len].idx = access_idx; | |
4695 | spec->len++; | |
4696 | ||
4697 | sz = btf__resolve_size(btf, id); | |
4698 | if (sz < 0) | |
4699 | return sz; | |
ee26dade | 4700 | spec->bit_offset += access_idx * sz * 8; |
ddc7c304 | 4701 | } else { |
81ba0889 AN |
4702 | pr_warn("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %s\n", |
4703 | type_id, spec_str, i, id, btf_kind_str(t)); | |
ddc7c304 AN |
4704 | return -EINVAL; |
4705 | } | |
4706 | } | |
4707 | ||
4708 | return 0; | |
4709 | } | |
4710 | ||
4711 | static bool bpf_core_is_flavor_sep(const char *s) | |
4712 | { | |
4713 | /* check X___Y name pattern, where X and Y are not underscores */ | |
4714 | return s[0] != '_' && /* X */ | |
4715 | s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */ | |
4716 | s[4] != '_'; /* Y */ | |
4717 | } | |
4718 | ||
4719 | /* Given 'some_struct_name___with_flavor' return the length of a name prefix | |
4720 | * before last triple underscore. Struct name part after last triple | |
4721 | * underscore is ignored by BPF CO-RE relocation during relocation matching. | |
4722 | */ | |
4723 | static size_t bpf_core_essential_name_len(const char *name) | |
4724 | { | |
4725 | size_t n = strlen(name); | |
4726 | int i; | |
4727 | ||
4728 | for (i = n - 5; i >= 0; i--) { | |
4729 | if (bpf_core_is_flavor_sep(name + i)) | |
4730 | return i + 1; | |
4731 | } | |
4732 | return n; | |
4733 | } | |
4734 | ||
0f7515ca AN |
4735 | struct core_cand |
4736 | { | |
4737 | const struct btf *btf; | |
4738 | const struct btf_type *t; | |
4739 | const char *name; | |
4740 | __u32 id; | |
4741 | }; | |
4742 | ||
4743 | /* dynamically sized list of type IDs and its associated struct btf */ | |
4744 | struct core_cand_list { | |
4745 | struct core_cand *cands; | |
ddc7c304 AN |
4746 | int len; |
4747 | }; | |
4748 | ||
0f7515ca | 4749 | static void bpf_core_free_cands(struct core_cand_list *cands) |
ddc7c304 | 4750 | { |
0f7515ca AN |
4751 | free(cands->cands); |
4752 | free(cands); | |
ddc7c304 AN |
4753 | } |
4754 | ||
0f7515ca AN |
4755 | static int bpf_core_add_cands(struct core_cand *local_cand, |
4756 | size_t local_essent_len, | |
4757 | const struct btf *targ_btf, | |
4758 | const char *targ_btf_name, | |
4759 | int targ_start_id, | |
4760 | struct core_cand_list *cands) | |
ddc7c304 | 4761 | { |
0f7515ca AN |
4762 | struct core_cand *new_cands, *cand; |
4763 | const struct btf_type *t; | |
4764 | const char *targ_name; | |
4765 | size_t targ_essent_len; | |
4766 | int n, i; | |
ddc7c304 AN |
4767 | |
4768 | n = btf__get_nr_types(targ_btf); | |
0f7515ca | 4769 | for (i = targ_start_id; i <= n; i++) { |
ddc7c304 | 4770 | t = btf__type_by_id(targ_btf, i); |
0f7515ca | 4771 | if (btf_kind(t) != btf_kind(local_cand->t)) |
ddc7c304 AN |
4772 | continue; |
4773 | ||
3fc32f40 AN |
4774 | targ_name = btf__name_by_offset(targ_btf, t->name_off); |
4775 | if (str_is_empty(targ_name)) | |
d121e1d3 AN |
4776 | continue; |
4777 | ||
ddc7c304 AN |
4778 | targ_essent_len = bpf_core_essential_name_len(targ_name); |
4779 | if (targ_essent_len != local_essent_len) | |
4780 | continue; | |
4781 | ||
0f7515ca AN |
4782 | if (strncmp(local_cand->name, targ_name, local_essent_len) != 0) |
4783 | continue; | |
4784 | ||
4785 | pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n", | |
4786 | local_cand->id, btf_kind_str(local_cand->t), | |
4787 | local_cand->name, i, btf_kind_str(t), targ_name, | |
4788 | targ_btf_name); | |
4789 | new_cands = libbpf_reallocarray(cands->cands, cands->len + 1, | |
4790 | sizeof(*cands->cands)); | |
4791 | if (!new_cands) | |
4792 | return -ENOMEM; | |
4793 | ||
4794 | cand = &new_cands[cands->len]; | |
4795 | cand->btf = targ_btf; | |
4796 | cand->t = t; | |
4797 | cand->name = targ_name; | |
4798 | cand->id = i; | |
4799 | ||
4800 | cands->cands = new_cands; | |
4801 | cands->len++; | |
ddc7c304 | 4802 | } |
0f7515ca AN |
4803 | return 0; |
4804 | } | |
4805 | ||
4f33a53d AN |
4806 | static int load_module_btfs(struct bpf_object *obj) |
4807 | { | |
4808 | struct bpf_btf_info info; | |
4809 | struct module_btf *mod_btf; | |
4810 | struct btf *btf; | |
4811 | char name[64]; | |
4812 | __u32 id = 0, len; | |
4813 | int err, fd; | |
4814 | ||
4815 | if (obj->btf_modules_loaded) | |
4816 | return 0; | |
4817 | ||
4818 | /* don't do this again, even if we find no module BTFs */ | |
4819 | obj->btf_modules_loaded = true; | |
4820 | ||
4821 | /* kernel too old to support module BTFs */ | |
4822 | if (!kernel_supports(FEAT_MODULE_BTF)) | |
4823 | return 0; | |
4824 | ||
4825 | while (true) { | |
4826 | err = bpf_btf_get_next_id(id, &id); | |
4827 | if (err && errno == ENOENT) | |
4828 | return 0; | |
4829 | if (err) { | |
4830 | err = -errno; | |
4831 | pr_warn("failed to iterate BTF objects: %d\n", err); | |
4832 | return err; | |
ddc7c304 | 4833 | } |
4f33a53d AN |
4834 | |
4835 | fd = bpf_btf_get_fd_by_id(id); | |
4836 | if (fd < 0) { | |
4837 | if (errno == ENOENT) | |
4838 | continue; /* expected race: BTF was unloaded */ | |
4839 | err = -errno; | |
4840 | pr_warn("failed to get BTF object #%d FD: %d\n", id, err); | |
4841 | return err; | |
4842 | } | |
4843 | ||
4844 | len = sizeof(info); | |
4845 | memset(&info, 0, sizeof(info)); | |
4846 | info.name = ptr_to_u64(name); | |
4847 | info.name_len = sizeof(name); | |
4848 | ||
4849 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
4850 | if (err) { | |
4851 | err = -errno; | |
4852 | pr_warn("failed to get BTF object #%d info: %d\n", id, err); | |
91abb4a6 | 4853 | goto err_out; |
4f33a53d AN |
4854 | } |
4855 | ||
4856 | /* ignore non-module BTFs */ | |
4857 | if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) { | |
4858 | close(fd); | |
4859 | continue; | |
4860 | } | |
4861 | ||
4862 | btf = btf_get_from_fd(fd, obj->btf_vmlinux); | |
4f33a53d AN |
4863 | if (IS_ERR(btf)) { |
4864 | pr_warn("failed to load module [%s]'s BTF object #%d: %ld\n", | |
4865 | name, id, PTR_ERR(btf)); | |
91abb4a6 AN |
4866 | err = PTR_ERR(btf); |
4867 | goto err_out; | |
4f33a53d AN |
4868 | } |
4869 | ||
3b029e06 AN |
4870 | err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap, |
4871 | sizeof(*obj->btf_modules), obj->btf_module_cnt + 1); | |
4f33a53d | 4872 | if (err) |
91abb4a6 | 4873 | goto err_out; |
4f33a53d AN |
4874 | |
4875 | mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; | |
4876 | ||
4877 | mod_btf->btf = btf; | |
4878 | mod_btf->id = id; | |
91abb4a6 | 4879 | mod_btf->fd = fd; |
4f33a53d | 4880 | mod_btf->name = strdup(name); |
91abb4a6 AN |
4881 | if (!mod_btf->name) { |
4882 | err = -ENOMEM; | |
4883 | goto err_out; | |
4884 | } | |
4885 | continue; | |
4886 | ||
4887 | err_out: | |
4888 | close(fd); | |
4889 | return err; | |
ddc7c304 | 4890 | } |
4f33a53d AN |
4891 | |
4892 | return 0; | |
4893 | } | |
4894 | ||
0f7515ca AN |
4895 | static struct core_cand_list * |
4896 | bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id) | |
4897 | { | |
4898 | struct core_cand local_cand = {}; | |
4899 | struct core_cand_list *cands; | |
4f33a53d | 4900 | const struct btf *main_btf; |
0f7515ca | 4901 | size_t local_essent_len; |
4f33a53d | 4902 | int err, i; |
0f7515ca AN |
4903 | |
4904 | local_cand.btf = local_btf; | |
4905 | local_cand.t = btf__type_by_id(local_btf, local_type_id); | |
4906 | if (!local_cand.t) | |
4907 | return ERR_PTR(-EINVAL); | |
4908 | ||
4909 | local_cand.name = btf__name_by_offset(local_btf, local_cand.t->name_off); | |
4910 | if (str_is_empty(local_cand.name)) | |
4911 | return ERR_PTR(-EINVAL); | |
4912 | local_essent_len = bpf_core_essential_name_len(local_cand.name); | |
4913 | ||
4914 | cands = calloc(1, sizeof(*cands)); | |
4915 | if (!cands) | |
4916 | return ERR_PTR(-ENOMEM); | |
4917 | ||
4918 | /* Attempt to find target candidates in vmlinux BTF first */ | |
4f33a53d AN |
4919 | main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux; |
4920 | err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands); | |
4921 | if (err) | |
4922 | goto err_out; | |
4923 | ||
4924 | /* if vmlinux BTF has any candidate, don't got for module BTFs */ | |
4925 | if (cands->len) | |
4926 | return cands; | |
4927 | ||
4928 | /* if vmlinux BTF was overridden, don't attempt to load module BTFs */ | |
4929 | if (obj->btf_vmlinux_override) | |
4930 | return cands; | |
4931 | ||
4932 | /* now look through module BTFs, trying to still find candidates */ | |
4933 | err = load_module_btfs(obj); | |
4934 | if (err) | |
4935 | goto err_out; | |
4936 | ||
4937 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
4938 | err = bpf_core_add_cands(&local_cand, local_essent_len, | |
4939 | obj->btf_modules[i].btf, | |
4940 | obj->btf_modules[i].name, | |
4941 | btf__get_nr_types(obj->btf_vmlinux) + 1, | |
4942 | cands); | |
4943 | if (err) | |
4944 | goto err_out; | |
0f7515ca AN |
4945 | } |
4946 | ||
4947 | return cands; | |
ddc7c304 | 4948 | err_out: |
4f33a53d | 4949 | bpf_core_free_cands(cands); |
ddc7c304 AN |
4950 | return ERR_PTR(err); |
4951 | } | |
4952 | ||
3fc32f40 AN |
4953 | /* Check two types for compatibility for the purpose of field access |
4954 | * relocation. const/volatile/restrict and typedefs are skipped to ensure we | |
4955 | * are relocating semantically compatible entities: | |
ddc7c304 | 4956 | * - any two STRUCTs/UNIONs are compatible and can be mixed; |
94f060e9 | 4957 | * - any two FWDs are compatible, if their names match (modulo flavor suffix); |
ddc7c304 | 4958 | * - any two PTRs are always compatible; |
94f060e9 AN |
4959 | * - for ENUMs, names should be the same (ignoring flavor suffix) or at |
4960 | * least one of enums should be anonymous; | |
ddc7c304 | 4961 | * - for ENUMs, check sizes, names are ignored; |
ee26dade | 4962 | * - for INT, size and signedness are ignored; |
ddc7c304 AN |
4963 | * - for ARRAY, dimensionality is ignored, element types are checked for |
4964 | * compatibility recursively; | |
4965 | * - everything else shouldn't be ever a target of relocation. | |
4966 | * These rules are not set in stone and probably will be adjusted as we get | |
4967 | * more experience with using BPF CO-RE relocations. | |
4968 | */ | |
4969 | static int bpf_core_fields_are_compat(const struct btf *local_btf, | |
4970 | __u32 local_id, | |
4971 | const struct btf *targ_btf, | |
4972 | __u32 targ_id) | |
4973 | { | |
4974 | const struct btf_type *local_type, *targ_type; | |
4975 | ||
4976 | recur: | |
4977 | local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id); | |
4978 | targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id); | |
4979 | if (!local_type || !targ_type) | |
4980 | return -EINVAL; | |
4981 | ||
4982 | if (btf_is_composite(local_type) && btf_is_composite(targ_type)) | |
4983 | return 1; | |
4984 | if (btf_kind(local_type) != btf_kind(targ_type)) | |
4985 | return 0; | |
4986 | ||
4987 | switch (btf_kind(local_type)) { | |
ddc7c304 AN |
4988 | case BTF_KIND_PTR: |
4989 | return 1; | |
94f060e9 AN |
4990 | case BTF_KIND_FWD: |
4991 | case BTF_KIND_ENUM: { | |
4992 | const char *local_name, *targ_name; | |
4993 | size_t local_len, targ_len; | |
4994 | ||
4995 | local_name = btf__name_by_offset(local_btf, | |
4996 | local_type->name_off); | |
4997 | targ_name = btf__name_by_offset(targ_btf, targ_type->name_off); | |
4998 | local_len = bpf_core_essential_name_len(local_name); | |
4999 | targ_len = bpf_core_essential_name_len(targ_name); | |
5000 | /* one of them is anonymous or both w/ same flavor-less names */ | |
5001 | return local_len == 0 || targ_len == 0 || | |
5002 | (local_len == targ_len && | |
5003 | strncmp(local_name, targ_name, local_len) == 0); | |
5004 | } | |
ddc7c304 | 5005 | case BTF_KIND_INT: |
ee26dade AN |
5006 | /* just reject deprecated bitfield-like integers; all other |
5007 | * integers are by default compatible between each other | |
5008 | */ | |
ddc7c304 | 5009 | return btf_int_offset(local_type) == 0 && |
ee26dade | 5010 | btf_int_offset(targ_type) == 0; |
ddc7c304 AN |
5011 | case BTF_KIND_ARRAY: |
5012 | local_id = btf_array(local_type)->type; | |
5013 | targ_id = btf_array(targ_type)->type; | |
5014 | goto recur; | |
5015 | default: | |
be18010e KW |
5016 | pr_warn("unexpected kind %d relocated, local [%d], target [%d]\n", |
5017 | btf_kind(local_type), local_id, targ_id); | |
ddc7c304 AN |
5018 | return 0; |
5019 | } | |
5020 | } | |
5021 | ||
5022 | /* | |
5023 | * Given single high-level named field accessor in local type, find | |
5024 | * corresponding high-level accessor for a target type. Along the way, | |
5025 | * maintain low-level spec for target as well. Also keep updating target | |
ee26dade | 5026 | * bit offset. |
ddc7c304 AN |
5027 | * |
5028 | * Searching is performed through recursive exhaustive enumeration of all | |
5029 | * fields of a struct/union. If there are any anonymous (embedded) | |
5030 | * structs/unions, they are recursively searched as well. If field with | |
5031 | * desired name is found, check compatibility between local and target types, | |
5032 | * before returning result. | |
5033 | * | |
5034 | * 1 is returned, if field is found. | |
5035 | * 0 is returned if no compatible field is found. | |
5036 | * <0 is returned on error. | |
5037 | */ | |
5038 | static int bpf_core_match_member(const struct btf *local_btf, | |
5039 | const struct bpf_core_accessor *local_acc, | |
5040 | const struct btf *targ_btf, | |
5041 | __u32 targ_id, | |
5042 | struct bpf_core_spec *spec, | |
5043 | __u32 *next_targ_id) | |
5044 | { | |
5045 | const struct btf_type *local_type, *targ_type; | |
5046 | const struct btf_member *local_member, *m; | |
5047 | const char *local_name, *targ_name; | |
5048 | __u32 local_id; | |
5049 | int i, n, found; | |
5050 | ||
5051 | targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id); | |
5052 | if (!targ_type) | |
5053 | return -EINVAL; | |
5054 | if (!btf_is_composite(targ_type)) | |
5055 | return 0; | |
5056 | ||
5057 | local_id = local_acc->type_id; | |
5058 | local_type = btf__type_by_id(local_btf, local_id); | |
5059 | local_member = btf_members(local_type) + local_acc->idx; | |
5060 | local_name = btf__name_by_offset(local_btf, local_member->name_off); | |
5061 | ||
5062 | n = btf_vlen(targ_type); | |
5063 | m = btf_members(targ_type); | |
5064 | for (i = 0; i < n; i++, m++) { | |
ee26dade | 5065 | __u32 bit_offset; |
ddc7c304 | 5066 | |
ee26dade | 5067 | bit_offset = btf_member_bit_offset(targ_type, i); |
ddc7c304 AN |
5068 | |
5069 | /* too deep struct/union/array nesting */ | |
5070 | if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN) | |
5071 | return -E2BIG; | |
5072 | ||
5073 | /* speculate this member will be the good one */ | |
ee26dade | 5074 | spec->bit_offset += bit_offset; |
ddc7c304 AN |
5075 | spec->raw_spec[spec->raw_len++] = i; |
5076 | ||
5077 | targ_name = btf__name_by_offset(targ_btf, m->name_off); | |
5078 | if (str_is_empty(targ_name)) { | |
5079 | /* embedded struct/union, we need to go deeper */ | |
5080 | found = bpf_core_match_member(local_btf, local_acc, | |
5081 | targ_btf, m->type, | |
5082 | spec, next_targ_id); | |
5083 | if (found) /* either found or error */ | |
5084 | return found; | |
5085 | } else if (strcmp(local_name, targ_name) == 0) { | |
5086 | /* matching named field */ | |
5087 | struct bpf_core_accessor *targ_acc; | |
5088 | ||
5089 | targ_acc = &spec->spec[spec->len++]; | |
5090 | targ_acc->type_id = targ_id; | |
5091 | targ_acc->idx = i; | |
5092 | targ_acc->name = targ_name; | |
5093 | ||
5094 | *next_targ_id = m->type; | |
5095 | found = bpf_core_fields_are_compat(local_btf, | |
5096 | local_member->type, | |
5097 | targ_btf, m->type); | |
5098 | if (!found) | |
5099 | spec->len--; /* pop accessor */ | |
5100 | return found; | |
5101 | } | |
5102 | /* member turned out not to be what we looked for */ | |
ee26dade | 5103 | spec->bit_offset -= bit_offset; |
ddc7c304 AN |
5104 | spec->raw_len--; |
5105 | } | |
5106 | ||
5107 | return 0; | |
5108 | } | |
5109 | ||
3fc32f40 AN |
5110 | /* Check local and target types for compatibility. This check is used for |
5111 | * type-based CO-RE relocations and follow slightly different rules than | |
5112 | * field-based relocations. This function assumes that root types were already | |
5113 | * checked for name match. Beyond that initial root-level name check, names | |
5114 | * are completely ignored. Compatibility rules are as follows: | |
5115 | * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but | |
5116 | * kind should match for local and target types (i.e., STRUCT is not | |
5117 | * compatible with UNION); | |
5118 | * - for ENUMs, the size is ignored; | |
5119 | * - for INT, size and signedness are ignored; | |
5120 | * - for ARRAY, dimensionality is ignored, element types are checked for | |
5121 | * compatibility recursively; | |
5122 | * - CONST/VOLATILE/RESTRICT modifiers are ignored; | |
5123 | * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible; | |
5124 | * - FUNC_PROTOs are compatible if they have compatible signature: same | |
5125 | * number of input args and compatible return and argument types. | |
5126 | * These rules are not set in stone and probably will be adjusted as we get | |
5127 | * more experience with using BPF CO-RE relocations. | |
5128 | */ | |
5129 | static int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id, | |
5130 | const struct btf *targ_btf, __u32 targ_id) | |
5131 | { | |
5132 | const struct btf_type *local_type, *targ_type; | |
5133 | int depth = 32; /* max recursion depth */ | |
5134 | ||
5135 | /* caller made sure that names match (ignoring flavor suffix) */ | |
5136 | local_type = btf__type_by_id(local_btf, local_id); | |
f872e4bc | 5137 | targ_type = btf__type_by_id(targ_btf, targ_id); |
3fc32f40 AN |
5138 | if (btf_kind(local_type) != btf_kind(targ_type)) |
5139 | return 0; | |
5140 | ||
5141 | recur: | |
5142 | depth--; | |
5143 | if (depth < 0) | |
5144 | return -EINVAL; | |
5145 | ||
5146 | local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id); | |
5147 | targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id); | |
5148 | if (!local_type || !targ_type) | |
5149 | return -EINVAL; | |
5150 | ||
5151 | if (btf_kind(local_type) != btf_kind(targ_type)) | |
5152 | return 0; | |
5153 | ||
5154 | switch (btf_kind(local_type)) { | |
5155 | case BTF_KIND_UNKN: | |
5156 | case BTF_KIND_STRUCT: | |
5157 | case BTF_KIND_UNION: | |
5158 | case BTF_KIND_ENUM: | |
5159 | case BTF_KIND_FWD: | |
5160 | return 1; | |
5161 | case BTF_KIND_INT: | |
5162 | /* just reject deprecated bitfield-like integers; all other | |
5163 | * integers are by default compatible between each other | |
5164 | */ | |
5165 | return btf_int_offset(local_type) == 0 && btf_int_offset(targ_type) == 0; | |
5166 | case BTF_KIND_PTR: | |
5167 | local_id = local_type->type; | |
5168 | targ_id = targ_type->type; | |
5169 | goto recur; | |
5170 | case BTF_KIND_ARRAY: | |
5171 | local_id = btf_array(local_type)->type; | |
5172 | targ_id = btf_array(targ_type)->type; | |
5173 | goto recur; | |
5174 | case BTF_KIND_FUNC_PROTO: { | |
5175 | struct btf_param *local_p = btf_params(local_type); | |
5176 | struct btf_param *targ_p = btf_params(targ_type); | |
5177 | __u16 local_vlen = btf_vlen(local_type); | |
5178 | __u16 targ_vlen = btf_vlen(targ_type); | |
5179 | int i, err; | |
5180 | ||
5181 | if (local_vlen != targ_vlen) | |
5182 | return 0; | |
5183 | ||
5184 | for (i = 0; i < local_vlen; i++, local_p++, targ_p++) { | |
5185 | skip_mods_and_typedefs(local_btf, local_p->type, &local_id); | |
5186 | skip_mods_and_typedefs(targ_btf, targ_p->type, &targ_id); | |
5187 | err = bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id); | |
5188 | if (err <= 0) | |
5189 | return err; | |
5190 | } | |
5191 | ||
5192 | /* tail recurse for return type check */ | |
5193 | skip_mods_and_typedefs(local_btf, local_type->type, &local_id); | |
5194 | skip_mods_and_typedefs(targ_btf, targ_type->type, &targ_id); | |
5195 | goto recur; | |
5196 | } | |
5197 | default: | |
5198 | pr_warn("unexpected kind %s relocated, local [%d], target [%d]\n", | |
5199 | btf_kind_str(local_type), local_id, targ_id); | |
5200 | return 0; | |
5201 | } | |
5202 | } | |
5203 | ||
ddc7c304 AN |
5204 | /* |
5205 | * Try to match local spec to a target type and, if successful, produce full | |
ee26dade | 5206 | * target spec (high-level, low-level + bit offset). |
ddc7c304 AN |
5207 | */ |
5208 | static int bpf_core_spec_match(struct bpf_core_spec *local_spec, | |
5209 | const struct btf *targ_btf, __u32 targ_id, | |
5210 | struct bpf_core_spec *targ_spec) | |
5211 | { | |
5212 | const struct btf_type *targ_type; | |
5213 | const struct bpf_core_accessor *local_acc; | |
5214 | struct bpf_core_accessor *targ_acc; | |
5215 | int i, sz, matched; | |
5216 | ||
5217 | memset(targ_spec, 0, sizeof(*targ_spec)); | |
5218 | targ_spec->btf = targ_btf; | |
28b93c64 AN |
5219 | targ_spec->root_type_id = targ_id; |
5220 | targ_spec->relo_kind = local_spec->relo_kind; | |
ddc7c304 | 5221 | |
3fc32f40 AN |
5222 | if (core_relo_is_type_based(local_spec->relo_kind)) { |
5223 | return bpf_core_types_are_compat(local_spec->btf, | |
5224 | local_spec->root_type_id, | |
5225 | targ_btf, targ_id); | |
5226 | } | |
5227 | ||
ddc7c304 AN |
5228 | local_acc = &local_spec->spec[0]; |
5229 | targ_acc = &targ_spec->spec[0]; | |
5230 | ||
eacaaed7 AN |
5231 | if (core_relo_is_enumval_based(local_spec->relo_kind)) { |
5232 | size_t local_essent_len, targ_essent_len; | |
5233 | const struct btf_enum *e; | |
5234 | const char *targ_name; | |
5235 | ||
5236 | /* has to resolve to an enum */ | |
5237 | targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id, &targ_id); | |
5238 | if (!btf_is_enum(targ_type)) | |
5239 | return 0; | |
5240 | ||
5241 | local_essent_len = bpf_core_essential_name_len(local_acc->name); | |
5242 | ||
5243 | for (i = 0, e = btf_enum(targ_type); i < btf_vlen(targ_type); i++, e++) { | |
5244 | targ_name = btf__name_by_offset(targ_spec->btf, e->name_off); | |
5245 | targ_essent_len = bpf_core_essential_name_len(targ_name); | |
5246 | if (targ_essent_len != local_essent_len) | |
5247 | continue; | |
5248 | if (strncmp(local_acc->name, targ_name, local_essent_len) == 0) { | |
5249 | targ_acc->type_id = targ_id; | |
5250 | targ_acc->idx = i; | |
5251 | targ_acc->name = targ_name; | |
5252 | targ_spec->len++; | |
5253 | targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx; | |
5254 | targ_spec->raw_len++; | |
5255 | return 1; | |
5256 | } | |
5257 | } | |
5258 | return 0; | |
5259 | } | |
5260 | ||
5261 | if (!core_relo_is_field_based(local_spec->relo_kind)) | |
5262 | return -EINVAL; | |
5263 | ||
ddc7c304 AN |
5264 | for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) { |
5265 | targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id, | |
5266 | &targ_id); | |
5267 | if (!targ_type) | |
5268 | return -EINVAL; | |
5269 | ||
5270 | if (local_acc->name) { | |
5271 | matched = bpf_core_match_member(local_spec->btf, | |
5272 | local_acc, | |
5273 | targ_btf, targ_id, | |
5274 | targ_spec, &targ_id); | |
5275 | if (matched <= 0) | |
5276 | return matched; | |
5277 | } else { | |
5278 | /* for i=0, targ_id is already treated as array element | |
5279 | * type (because it's the original struct), for others | |
5280 | * we should find array element type first | |
5281 | */ | |
5282 | if (i > 0) { | |
5283 | const struct btf_array *a; | |
1b484b30 | 5284 | bool flex; |
ddc7c304 AN |
5285 | |
5286 | if (!btf_is_array(targ_type)) | |
5287 | return 0; | |
5288 | ||
5289 | a = btf_array(targ_type); | |
1b484b30 AN |
5290 | flex = is_flex_arr(targ_btf, targ_acc - 1, a); |
5291 | if (!flex && local_acc->idx >= a->nelems) | |
ddc7c304 AN |
5292 | return 0; |
5293 | if (!skip_mods_and_typedefs(targ_btf, a->type, | |
5294 | &targ_id)) | |
5295 | return -EINVAL; | |
5296 | } | |
5297 | ||
5298 | /* too deep struct/union/array nesting */ | |
5299 | if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN) | |
5300 | return -E2BIG; | |
5301 | ||
5302 | targ_acc->type_id = targ_id; | |
5303 | targ_acc->idx = local_acc->idx; | |
5304 | targ_acc->name = NULL; | |
5305 | targ_spec->len++; | |
5306 | targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx; | |
5307 | targ_spec->raw_len++; | |
5308 | ||
5309 | sz = btf__resolve_size(targ_btf, targ_id); | |
5310 | if (sz < 0) | |
5311 | return sz; | |
ee26dade | 5312 | targ_spec->bit_offset += local_acc->idx * sz * 8; |
ddc7c304 AN |
5313 | } |
5314 | } | |
5315 | ||
5316 | return 1; | |
5317 | } | |
5318 | ||
ee26dade | 5319 | static int bpf_core_calc_field_relo(const struct bpf_program *prog, |
28b93c64 | 5320 | const struct bpf_core_relo *relo, |
ee26dade | 5321 | const struct bpf_core_spec *spec, |
a66345bc AN |
5322 | __u32 *val, __u32 *field_sz, __u32 *type_id, |
5323 | bool *validate) | |
ee26dade | 5324 | { |
353c788c AN |
5325 | const struct bpf_core_accessor *acc; |
5326 | const struct btf_type *t; | |
a66345bc | 5327 | __u32 byte_off, byte_sz, bit_off, bit_sz, field_type_id; |
ee26dade AN |
5328 | const struct btf_member *m; |
5329 | const struct btf_type *mt; | |
5330 | bool bitfield; | |
94f060e9 | 5331 | __s64 sz; |
ee26dade | 5332 | |
a66345bc AN |
5333 | *field_sz = 0; |
5334 | ||
353c788c AN |
5335 | if (relo->kind == BPF_FIELD_EXISTS) { |
5336 | *val = spec ? 1 : 0; | |
5337 | return 0; | |
5338 | } | |
5339 | ||
5340 | if (!spec) | |
5341 | return -EUCLEAN; /* request instruction poisoning */ | |
5342 | ||
5343 | acc = &spec->spec[spec->len - 1]; | |
5344 | t = btf__type_by_id(spec->btf, acc->type_id); | |
5345 | ||
ee26dade AN |
5346 | /* a[n] accessor needs special handling */ |
5347 | if (!acc->name) { | |
94f060e9 AN |
5348 | if (relo->kind == BPF_FIELD_BYTE_OFFSET) { |
5349 | *val = spec->bit_offset / 8; | |
a66345bc AN |
5350 | /* remember field size for load/store mem size */ |
5351 | sz = btf__resolve_size(spec->btf, acc->type_id); | |
5352 | if (sz < 0) | |
5353 | return -EINVAL; | |
5354 | *field_sz = sz; | |
5355 | *type_id = acc->type_id; | |
94f060e9 AN |
5356 | } else if (relo->kind == BPF_FIELD_BYTE_SIZE) { |
5357 | sz = btf__resolve_size(spec->btf, acc->type_id); | |
5358 | if (sz < 0) | |
5359 | return -EINVAL; | |
5360 | *val = sz; | |
5361 | } else { | |
5362 | pr_warn("prog '%s': relo %d at insn #%d can't be applied to array access\n", | |
52109584 | 5363 | prog->name, relo->kind, relo->insn_off / 8); |
ee26dade AN |
5364 | return -EINVAL; |
5365 | } | |
ee26dade AN |
5366 | if (validate) |
5367 | *validate = true; | |
5368 | return 0; | |
5369 | } | |
5370 | ||
5371 | m = btf_members(t) + acc->idx; | |
a66345bc | 5372 | mt = skip_mods_and_typedefs(spec->btf, m->type, &field_type_id); |
ee26dade AN |
5373 | bit_off = spec->bit_offset; |
5374 | bit_sz = btf_member_bitfield_size(t, acc->idx); | |
5375 | ||
5376 | bitfield = bit_sz > 0; | |
5377 | if (bitfield) { | |
5378 | byte_sz = mt->size; | |
5379 | byte_off = bit_off / 8 / byte_sz * byte_sz; | |
5380 | /* figure out smallest int size necessary for bitfield load */ | |
5381 | while (bit_off + bit_sz - byte_off * 8 > byte_sz * 8) { | |
5382 | if (byte_sz >= 8) { | |
5383 | /* bitfield can't be read with 64-bit read */ | |
5384 | pr_warn("prog '%s': relo %d at insn #%d can't be satisfied for bitfield\n", | |
52109584 | 5385 | prog->name, relo->kind, relo->insn_off / 8); |
ee26dade AN |
5386 | return -E2BIG; |
5387 | } | |
5388 | byte_sz *= 2; | |
5389 | byte_off = bit_off / 8 / byte_sz * byte_sz; | |
5390 | } | |
5391 | } else { | |
a66345bc | 5392 | sz = btf__resolve_size(spec->btf, field_type_id); |
94f060e9 AN |
5393 | if (sz < 0) |
5394 | return -EINVAL; | |
5395 | byte_sz = sz; | |
ee26dade AN |
5396 | byte_off = spec->bit_offset / 8; |
5397 | bit_sz = byte_sz * 8; | |
5398 | } | |
5399 | ||
5400 | /* for bitfields, all the relocatable aspects are ambiguous and we | |
5401 | * might disagree with compiler, so turn off validation of expected | |
5402 | * value, except for signedness | |
5403 | */ | |
5404 | if (validate) | |
5405 | *validate = !bitfield; | |
5406 | ||
5407 | switch (relo->kind) { | |
5408 | case BPF_FIELD_BYTE_OFFSET: | |
5409 | *val = byte_off; | |
a66345bc AN |
5410 | if (!bitfield) { |
5411 | *field_sz = byte_sz; | |
5412 | *type_id = field_type_id; | |
5413 | } | |
ee26dade AN |
5414 | break; |
5415 | case BPF_FIELD_BYTE_SIZE: | |
5416 | *val = byte_sz; | |
5417 | break; | |
5418 | case BPF_FIELD_SIGNED: | |
5419 | /* enums will be assumed unsigned */ | |
5420 | *val = btf_is_enum(mt) || | |
5421 | (btf_int_encoding(mt) & BTF_INT_SIGNED); | |
5422 | if (validate) | |
5423 | *validate = true; /* signedness is never ambiguous */ | |
5424 | break; | |
5425 | case BPF_FIELD_LSHIFT_U64: | |
5426 | #if __BYTE_ORDER == __LITTLE_ENDIAN | |
5427 | *val = 64 - (bit_off + bit_sz - byte_off * 8); | |
5428 | #else | |
5429 | *val = (8 - byte_sz) * 8 + (bit_off - byte_off * 8); | |
5430 | #endif | |
5431 | break; | |
5432 | case BPF_FIELD_RSHIFT_U64: | |
5433 | *val = 64 - bit_sz; | |
5434 | if (validate) | |
5435 | *validate = true; /* right shift is never ambiguous */ | |
5436 | break; | |
5437 | case BPF_FIELD_EXISTS: | |
5438 | default: | |
353c788c | 5439 | return -EOPNOTSUPP; |
ee26dade AN |
5440 | } |
5441 | ||
5442 | return 0; | |
5443 | } | |
5444 | ||
3fc32f40 AN |
5445 | static int bpf_core_calc_type_relo(const struct bpf_core_relo *relo, |
5446 | const struct bpf_core_spec *spec, | |
5447 | __u32 *val) | |
5448 | { | |
5449 | __s64 sz; | |
5450 | ||
5451 | /* type-based relos return zero when target type is not found */ | |
5452 | if (!spec) { | |
5453 | *val = 0; | |
5454 | return 0; | |
5455 | } | |
5456 | ||
5457 | switch (relo->kind) { | |
5458 | case BPF_TYPE_ID_TARGET: | |
5459 | *val = spec->root_type_id; | |
5460 | break; | |
5461 | case BPF_TYPE_EXISTS: | |
5462 | *val = 1; | |
5463 | break; | |
5464 | case BPF_TYPE_SIZE: | |
5465 | sz = btf__resolve_size(spec->btf, spec->root_type_id); | |
5466 | if (sz < 0) | |
5467 | return -EINVAL; | |
5468 | *val = sz; | |
5469 | break; | |
5470 | case BPF_TYPE_ID_LOCAL: | |
5471 | /* BPF_TYPE_ID_LOCAL is handled specially and shouldn't get here */ | |
5472 | default: | |
5473 | return -EOPNOTSUPP; | |
5474 | } | |
5475 | ||
5476 | return 0; | |
5477 | } | |
5478 | ||
eacaaed7 AN |
5479 | static int bpf_core_calc_enumval_relo(const struct bpf_core_relo *relo, |
5480 | const struct bpf_core_spec *spec, | |
5481 | __u32 *val) | |
5482 | { | |
5483 | const struct btf_type *t; | |
5484 | const struct btf_enum *e; | |
5485 | ||
5486 | switch (relo->kind) { | |
5487 | case BPF_ENUMVAL_EXISTS: | |
5488 | *val = spec ? 1 : 0; | |
5489 | break; | |
5490 | case BPF_ENUMVAL_VALUE: | |
5491 | if (!spec) | |
5492 | return -EUCLEAN; /* request instruction poisoning */ | |
5493 | t = btf__type_by_id(spec->btf, spec->spec[0].type_id); | |
5494 | e = btf_enum(t) + spec->spec[0].idx; | |
5495 | *val = e->val; | |
5496 | break; | |
5497 | default: | |
5498 | return -EOPNOTSUPP; | |
5499 | } | |
5500 | ||
5501 | return 0; | |
5502 | } | |
5503 | ||
353c788c AN |
5504 | struct bpf_core_relo_res |
5505 | { | |
5506 | /* expected value in the instruction, unless validate == false */ | |
5507 | __u32 orig_val; | |
5508 | /* new value that needs to be patched up to */ | |
5509 | __u32 new_val; | |
5510 | /* relocation unsuccessful, poison instruction, but don't fail load */ | |
5511 | bool poison; | |
5512 | /* some relocations can't be validated against orig_val */ | |
5513 | bool validate; | |
a66345bc AN |
5514 | /* for field byte offset relocations or the forms: |
5515 | * *(T *)(rX + <off>) = rY | |
5516 | * rX = *(T *)(rY + <off>), | |
5517 | * we remember original and resolved field size to adjust direct | |
5518 | * memory loads of pointers and integers; this is necessary for 32-bit | |
5519 | * host kernel architectures, but also allows to automatically | |
5520 | * relocate fields that were resized from, e.g., u32 to u64, etc. | |
5521 | */ | |
5522 | bool fail_memsz_adjust; | |
5523 | __u32 orig_sz; | |
5524 | __u32 orig_type_id; | |
5525 | __u32 new_sz; | |
5526 | __u32 new_type_id; | |
353c788c AN |
5527 | }; |
5528 | ||
5529 | /* Calculate original and target relocation values, given local and target | |
5530 | * specs and relocation kind. These values are calculated for each candidate. | |
5531 | * If there are multiple candidates, resulting values should all be consistent | |
5532 | * with each other. Otherwise, libbpf will refuse to proceed due to ambiguity. | |
5533 | * If instruction has to be poisoned, *poison will be set to true. | |
5534 | */ | |
5535 | static int bpf_core_calc_relo(const struct bpf_program *prog, | |
5536 | const struct bpf_core_relo *relo, | |
5537 | int relo_idx, | |
5538 | const struct bpf_core_spec *local_spec, | |
5539 | const struct bpf_core_spec *targ_spec, | |
5540 | struct bpf_core_relo_res *res) | |
5541 | { | |
5542 | int err = -EOPNOTSUPP; | |
5543 | ||
5544 | res->orig_val = 0; | |
5545 | res->new_val = 0; | |
5546 | res->poison = false; | |
5547 | res->validate = true; | |
a66345bc AN |
5548 | res->fail_memsz_adjust = false; |
5549 | res->orig_sz = res->new_sz = 0; | |
5550 | res->orig_type_id = res->new_type_id = 0; | |
353c788c AN |
5551 | |
5552 | if (core_relo_is_field_based(relo->kind)) { | |
a66345bc AN |
5553 | err = bpf_core_calc_field_relo(prog, relo, local_spec, |
5554 | &res->orig_val, &res->orig_sz, | |
5555 | &res->orig_type_id, &res->validate); | |
5556 | err = err ?: bpf_core_calc_field_relo(prog, relo, targ_spec, | |
5557 | &res->new_val, &res->new_sz, | |
5558 | &res->new_type_id, NULL); | |
5559 | if (err) | |
5560 | goto done; | |
5561 | /* Validate if it's safe to adjust load/store memory size. | |
5562 | * Adjustments are performed only if original and new memory | |
5563 | * sizes differ. | |
5564 | */ | |
5565 | res->fail_memsz_adjust = false; | |
5566 | if (res->orig_sz != res->new_sz) { | |
5567 | const struct btf_type *orig_t, *new_t; | |
5568 | ||
5569 | orig_t = btf__type_by_id(local_spec->btf, res->orig_type_id); | |
5570 | new_t = btf__type_by_id(targ_spec->btf, res->new_type_id); | |
5571 | ||
5572 | /* There are two use cases in which it's safe to | |
5573 | * adjust load/store's mem size: | |
5574 | * - reading a 32-bit kernel pointer, while on BPF | |
5575 | * size pointers are always 64-bit; in this case | |
5576 | * it's safe to "downsize" instruction size due to | |
5577 | * pointer being treated as unsigned integer with | |
5578 | * zero-extended upper 32-bits; | |
5579 | * - reading unsigned integers, again due to | |
5580 | * zero-extension is preserving the value correctly. | |
5581 | * | |
5582 | * In all other cases it's incorrect to attempt to | |
5583 | * load/store field because read value will be | |
5584 | * incorrect, so we poison relocated instruction. | |
5585 | */ | |
5586 | if (btf_is_ptr(orig_t) && btf_is_ptr(new_t)) | |
5587 | goto done; | |
5588 | if (btf_is_int(orig_t) && btf_is_int(new_t) && | |
5589 | btf_int_encoding(orig_t) != BTF_INT_SIGNED && | |
5590 | btf_int_encoding(new_t) != BTF_INT_SIGNED) | |
5591 | goto done; | |
5592 | ||
5593 | /* mark as invalid mem size adjustment, but this will | |
5594 | * only be checked for LDX/STX/ST insns | |
5595 | */ | |
5596 | res->fail_memsz_adjust = true; | |
5597 | } | |
3fc32f40 AN |
5598 | } else if (core_relo_is_type_based(relo->kind)) { |
5599 | err = bpf_core_calc_type_relo(relo, local_spec, &res->orig_val); | |
5600 | err = err ?: bpf_core_calc_type_relo(relo, targ_spec, &res->new_val); | |
eacaaed7 AN |
5601 | } else if (core_relo_is_enumval_based(relo->kind)) { |
5602 | err = bpf_core_calc_enumval_relo(relo, local_spec, &res->orig_val); | |
5603 | err = err ?: bpf_core_calc_enumval_relo(relo, targ_spec, &res->new_val); | |
353c788c AN |
5604 | } |
5605 | ||
a66345bc | 5606 | done: |
353c788c AN |
5607 | if (err == -EUCLEAN) { |
5608 | /* EUCLEAN is used to signal instruction poisoning request */ | |
5609 | res->poison = true; | |
5610 | err = 0; | |
5611 | } else if (err == -EOPNOTSUPP) { | |
5612 | /* EOPNOTSUPP means unknown/unsupported relocation */ | |
5613 | pr_warn("prog '%s': relo #%d: unrecognized CO-RE relocation %s (%d) at insn #%d\n", | |
52109584 AN |
5614 | prog->name, relo_idx, core_relo_kind_str(relo->kind), |
5615 | relo->kind, relo->insn_off / 8); | |
353c788c AN |
5616 | } |
5617 | ||
5618 | return err; | |
5619 | } | |
5620 | ||
5621 | /* | |
5622 | * Turn instruction for which CO_RE relocation failed into invalid one with | |
5623 | * distinct signature. | |
5624 | */ | |
5625 | static void bpf_core_poison_insn(struct bpf_program *prog, int relo_idx, | |
5626 | int insn_idx, struct bpf_insn *insn) | |
5627 | { | |
5628 | pr_debug("prog '%s': relo #%d: substituting insn #%d w/ invalid insn\n", | |
52109584 | 5629 | prog->name, relo_idx, insn_idx); |
353c788c AN |
5630 | insn->code = BPF_JMP | BPF_CALL; |
5631 | insn->dst_reg = 0; | |
5632 | insn->src_reg = 0; | |
5633 | insn->off = 0; | |
5634 | /* if this instruction is reachable (not a dead code), | |
5635 | * verifier will complain with the following message: | |
5636 | * invalid func unknown#195896080 | |
5637 | */ | |
5638 | insn->imm = 195896080; /* => 0xbad2310 => "bad relo" */ | |
5639 | } | |
5640 | ||
a66345bc AN |
5641 | static int insn_bpf_size_to_bytes(struct bpf_insn *insn) |
5642 | { | |
5643 | switch (BPF_SIZE(insn->code)) { | |
5644 | case BPF_DW: return 8; | |
5645 | case BPF_W: return 4; | |
5646 | case BPF_H: return 2; | |
5647 | case BPF_B: return 1; | |
5648 | default: return -1; | |
5649 | } | |
5650 | } | |
5651 | ||
5652 | static int insn_bytes_to_bpf_size(__u32 sz) | |
5653 | { | |
5654 | switch (sz) { | |
5655 | case 8: return BPF_DW; | |
5656 | case 4: return BPF_W; | |
5657 | case 2: return BPF_H; | |
5658 | case 1: return BPF_B; | |
5659 | default: return -1; | |
5660 | } | |
5661 | } | |
5662 | ||
ddc7c304 AN |
5663 | /* |
5664 | * Patch relocatable BPF instruction. | |
62561eb4 AN |
5665 | * |
5666 | * Patched value is determined by relocation kind and target specification. | |
353c788c | 5667 | * For existence relocations target spec will be NULL if field/type is not found. |
62561eb4 AN |
5668 | * Expected insn->imm value is determined using relocation kind and local |
5669 | * spec, and is checked before patching instruction. If actual insn->imm value | |
5670 | * is wrong, bail out with error. | |
ddc7c304 | 5671 | * |
a66345bc | 5672 | * Currently supported classes of BPF instruction are: |
ddc7c304 AN |
5673 | * 1. rX = <imm> (assignment with immediate operand); |
5674 | * 2. rX += <imm> (arithmetic operations with immediate operand); | |
a66345bc AN |
5675 | * 3. rX = <imm64> (load with 64-bit immediate value); |
5676 | * 4. rX = *(T *)(rY + <off>), where T is one of {u8, u16, u32, u64}; | |
5677 | * 5. *(T *)(rX + <off>) = rY, where T is one of {u8, u16, u32, u64}; | |
5678 | * 6. *(T *)(rX + <off>) = <imm>, where T is one of {u8, u16, u32, u64}. | |
ddc7c304 | 5679 | */ |
353c788c | 5680 | static int bpf_core_patch_insn(struct bpf_program *prog, |
28b93c64 | 5681 | const struct bpf_core_relo *relo, |
d7a25270 | 5682 | int relo_idx, |
353c788c | 5683 | const struct bpf_core_relo_res *res) |
ddc7c304 | 5684 | { |
62561eb4 | 5685 | __u32 orig_val, new_val; |
ddc7c304 | 5686 | struct bpf_insn *insn; |
353c788c | 5687 | int insn_idx; |
ddc7c304 AN |
5688 | __u8 class; |
5689 | ||
9c0f8cbd | 5690 | if (relo->insn_off % BPF_INSN_SZ) |
62561eb4 | 5691 | return -EINVAL; |
9c0f8cbd | 5692 | insn_idx = relo->insn_off / BPF_INSN_SZ; |
db2b8b06 AN |
5693 | /* adjust insn_idx from section frame of reference to the local |
5694 | * program's frame of reference; (sub-)program code is not yet | |
5695 | * relocated, so it's enough to just subtract in-section offset | |
5696 | */ | |
5697 | insn_idx = insn_idx - prog->sec_insn_off; | |
d7a25270 AN |
5698 | insn = &prog->insns[insn_idx]; |
5699 | class = BPF_CLASS(insn->code); | |
62561eb4 | 5700 | |
353c788c | 5701 | if (res->poison) { |
a66345bc | 5702 | poison: |
eacaaed7 AN |
5703 | /* poison second part of ldimm64 to avoid confusing error from |
5704 | * verifier about "unknown opcode 00" | |
5705 | */ | |
5706 | if (is_ldimm64(insn)) | |
5707 | bpf_core_poison_insn(prog, relo_idx, insn_idx + 1, insn + 1); | |
353c788c | 5708 | bpf_core_poison_insn(prog, relo_idx, insn_idx, insn); |
d7a25270 | 5709 | return 0; |
62561eb4 | 5710 | } |
ddc7c304 | 5711 | |
353c788c AN |
5712 | orig_val = res->orig_val; |
5713 | new_val = res->new_val; | |
5714 | ||
8ab9da57 AN |
5715 | switch (class) { |
5716 | case BPF_ALU: | |
5717 | case BPF_ALU64: | |
ddc7c304 AN |
5718 | if (BPF_SRC(insn->code) != BPF_K) |
5719 | return -EINVAL; | |
353c788c | 5720 | if (res->validate && insn->imm != orig_val) { |
d7a25270 | 5721 | pr_warn("prog '%s': relo #%d: unexpected insn #%d (ALU/ALU64) value: got %u, exp %u -> %u\n", |
52109584 | 5722 | prog->name, relo_idx, |
d7a25270 | 5723 | insn_idx, insn->imm, orig_val, new_val); |
ddc7c304 | 5724 | return -EINVAL; |
ee26dade AN |
5725 | } |
5726 | orig_val = insn->imm; | |
62561eb4 | 5727 | insn->imm = new_val; |
d7a25270 | 5728 | pr_debug("prog '%s': relo #%d: patched insn #%d (ALU/ALU64) imm %u -> %u\n", |
52109584 | 5729 | prog->name, relo_idx, insn_idx, |
d7a25270 | 5730 | orig_val, new_val); |
8ab9da57 AN |
5731 | break; |
5732 | case BPF_LDX: | |
5733 | case BPF_ST: | |
5734 | case BPF_STX: | |
353c788c | 5735 | if (res->validate && insn->off != orig_val) { |
eacaaed7 | 5736 | pr_warn("prog '%s': relo #%d: unexpected insn #%d (LDX/ST/STX) value: got %u, exp %u -> %u\n", |
52109584 | 5737 | prog->name, relo_idx, insn_idx, insn->off, orig_val, new_val); |
8ab9da57 AN |
5738 | return -EINVAL; |
5739 | } | |
5740 | if (new_val > SHRT_MAX) { | |
d7a25270 | 5741 | pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) value too big: %u\n", |
52109584 | 5742 | prog->name, relo_idx, insn_idx, new_val); |
8ab9da57 AN |
5743 | return -ERANGE; |
5744 | } | |
a66345bc AN |
5745 | if (res->fail_memsz_adjust) { |
5746 | pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) accesses field incorrectly. " | |
5747 | "Make sure you are accessing pointers, unsigned integers, or fields of matching type and size.\n", | |
5748 | prog->name, relo_idx, insn_idx); | |
5749 | goto poison; | |
5750 | } | |
5751 | ||
8ab9da57 AN |
5752 | orig_val = insn->off; |
5753 | insn->off = new_val; | |
d7a25270 | 5754 | pr_debug("prog '%s': relo #%d: patched insn #%d (LDX/ST/STX) off %u -> %u\n", |
52109584 | 5755 | prog->name, relo_idx, insn_idx, orig_val, new_val); |
a66345bc AN |
5756 | |
5757 | if (res->new_sz != res->orig_sz) { | |
5758 | int insn_bytes_sz, insn_bpf_sz; | |
5759 | ||
5760 | insn_bytes_sz = insn_bpf_size_to_bytes(insn); | |
5761 | if (insn_bytes_sz != res->orig_sz) { | |
5762 | pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) unexpected mem size: got %d, exp %u\n", | |
5763 | prog->name, relo_idx, insn_idx, insn_bytes_sz, res->orig_sz); | |
5764 | return -EINVAL; | |
5765 | } | |
5766 | ||
5767 | insn_bpf_sz = insn_bytes_to_bpf_size(res->new_sz); | |
5768 | if (insn_bpf_sz < 0) { | |
5769 | pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) invalid new mem size: %u\n", | |
5770 | prog->name, relo_idx, insn_idx, res->new_sz); | |
5771 | return -EINVAL; | |
5772 | } | |
5773 | ||
5774 | insn->code = BPF_MODE(insn->code) | insn_bpf_sz | BPF_CLASS(insn->code); | |
5775 | pr_debug("prog '%s': relo #%d: patched insn #%d (LDX/ST/STX) mem_sz %u -> %u\n", | |
5776 | prog->name, relo_idx, insn_idx, res->orig_sz, res->new_sz); | |
5777 | } | |
8ab9da57 | 5778 | break; |
eacaaed7 AN |
5779 | case BPF_LD: { |
5780 | __u64 imm; | |
5781 | ||
5782 | if (!is_ldimm64(insn) || | |
5783 | insn[0].src_reg != 0 || insn[0].off != 0 || | |
5784 | insn_idx + 1 >= prog->insns_cnt || | |
5785 | insn[1].code != 0 || insn[1].dst_reg != 0 || | |
5786 | insn[1].src_reg != 0 || insn[1].off != 0) { | |
5787 | pr_warn("prog '%s': relo #%d: insn #%d (LDIMM64) has unexpected form\n", | |
52109584 | 5788 | prog->name, relo_idx, insn_idx); |
eacaaed7 AN |
5789 | return -EINVAL; |
5790 | } | |
5791 | ||
5792 | imm = insn[0].imm + ((__u64)insn[1].imm << 32); | |
5793 | if (res->validate && imm != orig_val) { | |
5794 | pr_warn("prog '%s': relo #%d: unexpected insn #%d (LDIMM64) value: got %llu, exp %u -> %u\n", | |
52109584 | 5795 | prog->name, relo_idx, |
2e80be60 AN |
5796 | insn_idx, (unsigned long long)imm, |
5797 | orig_val, new_val); | |
eacaaed7 AN |
5798 | return -EINVAL; |
5799 | } | |
5800 | ||
5801 | insn[0].imm = new_val; | |
5802 | insn[1].imm = 0; /* currently only 32-bit values are supported */ | |
5803 | pr_debug("prog '%s': relo #%d: patched insn #%d (LDIMM64) imm64 %llu -> %u\n", | |
52109584 | 5804 | prog->name, relo_idx, insn_idx, |
2e80be60 | 5805 | (unsigned long long)imm, new_val); |
eacaaed7 AN |
5806 | break; |
5807 | } | |
8ab9da57 | 5808 | default: |
eacaaed7 | 5809 | pr_warn("prog '%s': relo #%d: trying to relocate unrecognized insn #%d, code:0x%x, src:0x%x, dst:0x%x, off:0x%x, imm:0x%x\n", |
52109584 AN |
5810 | prog->name, relo_idx, insn_idx, insn->code, |
5811 | insn->src_reg, insn->dst_reg, insn->off, insn->imm); | |
ddc7c304 AN |
5812 | return -EINVAL; |
5813 | } | |
62561eb4 | 5814 | |
ddc7c304 AN |
5815 | return 0; |
5816 | } | |
5817 | ||
ddc7c304 AN |
5818 | /* Output spec definition in the format: |
5819 | * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>, | |
5820 | * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b | |
5821 | */ | |
5822 | static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec) | |
5823 | { | |
5824 | const struct btf_type *t; | |
eacaaed7 | 5825 | const struct btf_enum *e; |
ddc7c304 AN |
5826 | const char *s; |
5827 | __u32 type_id; | |
5828 | int i; | |
5829 | ||
28b93c64 | 5830 | type_id = spec->root_type_id; |
ddc7c304 AN |
5831 | t = btf__type_by_id(spec->btf, type_id); |
5832 | s = btf__name_by_offset(spec->btf, t->name_off); | |
ddc7c304 | 5833 | |
28b93c64 AN |
5834 | libbpf_print(level, "[%u] %s %s", type_id, btf_kind_str(t), str_is_empty(s) ? "<anon>" : s); |
5835 | ||
3fc32f40 AN |
5836 | if (core_relo_is_type_based(spec->relo_kind)) |
5837 | return; | |
5838 | ||
eacaaed7 AN |
5839 | if (core_relo_is_enumval_based(spec->relo_kind)) { |
5840 | t = skip_mods_and_typedefs(spec->btf, type_id, NULL); | |
5841 | e = btf_enum(t) + spec->raw_spec[0]; | |
5842 | s = btf__name_by_offset(spec->btf, e->name_off); | |
5843 | ||
5844 | libbpf_print(level, "::%s = %u", s, e->val); | |
5845 | return; | |
5846 | } | |
5847 | ||
28b93c64 AN |
5848 | if (core_relo_is_field_based(spec->relo_kind)) { |
5849 | for (i = 0; i < spec->len; i++) { | |
5850 | if (spec->spec[i].name) | |
5851 | libbpf_print(level, ".%s", spec->spec[i].name); | |
5852 | else if (i > 0 || spec->spec[i].idx > 0) | |
5853 | libbpf_print(level, "[%u]", spec->spec[i].idx); | |
5854 | } | |
ddc7c304 | 5855 | |
28b93c64 AN |
5856 | libbpf_print(level, " ("); |
5857 | for (i = 0; i < spec->raw_len; i++) | |
5858 | libbpf_print(level, "%s%d", i == 0 ? "" : ":", spec->raw_spec[i]); | |
ddc7c304 | 5859 | |
28b93c64 AN |
5860 | if (spec->bit_offset % 8) |
5861 | libbpf_print(level, " @ offset %u.%u)", | |
5862 | spec->bit_offset / 8, spec->bit_offset % 8); | |
ddc7c304 | 5863 | else |
28b93c64 | 5864 | libbpf_print(level, " @ offset %u)", spec->bit_offset / 8); |
3fc32f40 | 5865 | return; |
ddc7c304 | 5866 | } |
ddc7c304 AN |
5867 | } |
5868 | ||
5869 | static size_t bpf_core_hash_fn(const void *key, void *ctx) | |
5870 | { | |
5871 | return (size_t)key; | |
5872 | } | |
5873 | ||
5874 | static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx) | |
5875 | { | |
5876 | return k1 == k2; | |
5877 | } | |
5878 | ||
5879 | static void *u32_as_hash_key(__u32 x) | |
5880 | { | |
5881 | return (void *)(uintptr_t)x; | |
5882 | } | |
5883 | ||
5884 | /* | |
5885 | * CO-RE relocate single instruction. | |
5886 | * | |
5887 | * The outline and important points of the algorithm: | |
5888 | * 1. For given local type, find corresponding candidate target types. | |
5889 | * Candidate type is a type with the same "essential" name, ignoring | |
5890 | * everything after last triple underscore (___). E.g., `sample`, | |
5891 | * `sample___flavor_one`, `sample___flavor_another_one`, are all candidates | |
5892 | * for each other. Names with triple underscore are referred to as | |
5893 | * "flavors" and are useful, among other things, to allow to | |
5894 | * specify/support incompatible variations of the same kernel struct, which | |
5895 | * might differ between different kernel versions and/or build | |
5896 | * configurations. | |
5897 | * | |
5898 | * N.B. Struct "flavors" could be generated by bpftool's BTF-to-C | |
5899 | * converter, when deduplicated BTF of a kernel still contains more than | |
5900 | * one different types with the same name. In that case, ___2, ___3, etc | |
5901 | * are appended starting from second name conflict. But start flavors are | |
5902 | * also useful to be defined "locally", in BPF program, to extract same | |
5903 | * data from incompatible changes between different kernel | |
5904 | * versions/configurations. For instance, to handle field renames between | |
5905 | * kernel versions, one can use two flavors of the struct name with the | |
5906 | * same common name and use conditional relocations to extract that field, | |
5907 | * depending on target kernel version. | |
5908 | * 2. For each candidate type, try to match local specification to this | |
5909 | * candidate target type. Matching involves finding corresponding | |
5910 | * high-level spec accessors, meaning that all named fields should match, | |
5911 | * as well as all array accesses should be within the actual bounds. Also, | |
5912 | * types should be compatible (see bpf_core_fields_are_compat for details). | |
5913 | * 3. It is supported and expected that there might be multiple flavors | |
5914 | * matching the spec. As long as all the specs resolve to the same set of | |
511bb008 | 5915 | * offsets across all candidates, there is no error. If there is any |
ddc7c304 AN |
5916 | * ambiguity, CO-RE relocation will fail. This is necessary to accomodate |
5917 | * imprefection of BTF deduplication, which can cause slight duplication of | |
5918 | * the same BTF type, if some directly or indirectly referenced (by | |
5919 | * pointer) type gets resolved to different actual types in different | |
5920 | * object files. If such situation occurs, deduplicated BTF will end up | |
5921 | * with two (or more) structurally identical types, which differ only in | |
5922 | * types they refer to through pointer. This should be OK in most cases and | |
5923 | * is not an error. | |
5924 | * 4. Candidate types search is performed by linearly scanning through all | |
5925 | * types in target BTF. It is anticipated that this is overall more | |
5926 | * efficient memory-wise and not significantly worse (if not better) | |
5927 | * CPU-wise compared to prebuilding a map from all local type names to | |
5928 | * a list of candidate type names. It's also sped up by caching resolved | |
5929 | * list of matching candidates per each local "root" type ID, that has at | |
28b93c64 | 5930 | * least one bpf_core_relo associated with it. This list is shared |
ddc7c304 AN |
5931 | * between multiple relocations for the same type ID and is updated as some |
5932 | * of the candidates are pruned due to structural incompatibility. | |
5933 | */ | |
3fc32f40 AN |
5934 | static int bpf_core_apply_relo(struct bpf_program *prog, |
5935 | const struct bpf_core_relo *relo, | |
5936 | int relo_idx, | |
5937 | const struct btf *local_btf, | |
3fc32f40 | 5938 | struct hashmap *cand_cache) |
ddc7c304 | 5939 | { |
3418c56d | 5940 | struct bpf_core_spec local_spec, cand_spec, targ_spec = {}; |
ddc7c304 | 5941 | const void *type_key = u32_as_hash_key(relo->type_id); |
353c788c | 5942 | struct bpf_core_relo_res cand_res, targ_res; |
28b93c64 AN |
5943 | const struct btf_type *local_type; |
5944 | const char *local_name; | |
0f7515ca AN |
5945 | struct core_cand_list *cands = NULL; |
5946 | __u32 local_id; | |
ddc7c304 AN |
5947 | const char *spec_str; |
5948 | int i, j, err; | |
5949 | ||
5950 | local_id = relo->type_id; | |
5951 | local_type = btf__type_by_id(local_btf, local_id); | |
5952 | if (!local_type) | |
5953 | return -EINVAL; | |
5954 | ||
5955 | local_name = btf__name_by_offset(local_btf, local_type->name_off); | |
3fc32f40 | 5956 | if (!local_name) |
ddc7c304 AN |
5957 | return -EINVAL; |
5958 | ||
5959 | spec_str = btf__name_by_offset(local_btf, relo->access_str_off); | |
5960 | if (str_is_empty(spec_str)) | |
5961 | return -EINVAL; | |
5962 | ||
28b93c64 | 5963 | err = bpf_core_parse_spec(local_btf, local_id, spec_str, relo->kind, &local_spec); |
ddc7c304 | 5964 | if (err) { |
28b93c64 | 5965 | pr_warn("prog '%s': relo #%d: parsing [%d] %s %s + %s failed: %d\n", |
52109584 | 5966 | prog->name, relo_idx, local_id, btf_kind_str(local_type), |
3fc32f40 AN |
5967 | str_is_empty(local_name) ? "<anon>" : local_name, |
5968 | spec_str, err); | |
ddc7c304 AN |
5969 | return -EINVAL; |
5970 | } | |
5971 | ||
52109584 | 5972 | pr_debug("prog '%s': relo #%d: kind <%s> (%d), spec is ", prog->name, |
28b93c64 | 5973 | relo_idx, core_relo_kind_str(relo->kind), relo->kind); |
ddc7c304 AN |
5974 | bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec); |
5975 | libbpf_print(LIBBPF_DEBUG, "\n"); | |
5976 | ||
3fc32f40 AN |
5977 | /* TYPE_ID_LOCAL relo is special and doesn't need candidate search */ |
5978 | if (relo->kind == BPF_TYPE_ID_LOCAL) { | |
5979 | targ_res.validate = true; | |
5980 | targ_res.poison = false; | |
5981 | targ_res.orig_val = local_spec.root_type_id; | |
5982 | targ_res.new_val = local_spec.root_type_id; | |
5983 | goto patch_insn; | |
5984 | } | |
5985 | ||
5986 | /* libbpf doesn't support candidate search for anonymous types */ | |
5987 | if (str_is_empty(spec_str)) { | |
5988 | pr_warn("prog '%s': relo #%d: <%s> (%d) relocation doesn't support anonymous types\n", | |
52109584 | 5989 | prog->name, relo_idx, core_relo_kind_str(relo->kind), relo->kind); |
3fc32f40 AN |
5990 | return -EOPNOTSUPP; |
5991 | } | |
5992 | ||
0f7515ca AN |
5993 | if (!hashmap__find(cand_cache, type_key, (void **)&cands)) { |
5994 | cands = bpf_core_find_cands(prog->obj, local_btf, local_id); | |
5995 | if (IS_ERR(cands)) { | |
4f33a53d | 5996 | pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n", |
52109584 | 5997 | prog->name, relo_idx, local_id, btf_kind_str(local_type), |
0f7515ca AN |
5998 | local_name, PTR_ERR(cands)); |
5999 | return PTR_ERR(cands); | |
ddc7c304 | 6000 | } |
0f7515ca | 6001 | err = hashmap__set(cand_cache, type_key, cands, NULL, NULL); |
ddc7c304 | 6002 | if (err) { |
0f7515ca | 6003 | bpf_core_free_cands(cands); |
ddc7c304 AN |
6004 | return err; |
6005 | } | |
6006 | } | |
6007 | ||
0f7515ca AN |
6008 | for (i = 0, j = 0; i < cands->len; i++) { |
6009 | err = bpf_core_spec_match(&local_spec, cands->cands[i].btf, | |
6010 | cands->cands[i].id, &cand_spec); | |
ddc7c304 | 6011 | if (err < 0) { |
28b93c64 | 6012 | pr_warn("prog '%s': relo #%d: error matching candidate #%d ", |
52109584 | 6013 | prog->name, relo_idx, i); |
28b93c64 AN |
6014 | bpf_core_dump_spec(LIBBPF_WARN, &cand_spec); |
6015 | libbpf_print(LIBBPF_WARN, ": %d\n", err); | |
ddc7c304 AN |
6016 | return err; |
6017 | } | |
28b93c64 | 6018 | |
52109584 | 6019 | pr_debug("prog '%s': relo #%d: %s candidate #%d ", prog->name, |
28b93c64 AN |
6020 | relo_idx, err == 0 ? "non-matching" : "matching", i); |
6021 | bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec); | |
6022 | libbpf_print(LIBBPF_DEBUG, "\n"); | |
6023 | ||
ddc7c304 AN |
6024 | if (err == 0) |
6025 | continue; | |
6026 | ||
353c788c AN |
6027 | err = bpf_core_calc_relo(prog, relo, relo_idx, &local_spec, &cand_spec, &cand_res); |
6028 | if (err) | |
6029 | return err; | |
6030 | ||
ddc7c304 | 6031 | if (j == 0) { |
353c788c | 6032 | targ_res = cand_res; |
ddc7c304 | 6033 | targ_spec = cand_spec; |
ee26dade | 6034 | } else if (cand_spec.bit_offset != targ_spec.bit_offset) { |
353c788c AN |
6035 | /* if there are many field relo candidates, they |
6036 | * should all resolve to the same bit offset | |
ddc7c304 | 6037 | */ |
353c788c | 6038 | pr_warn("prog '%s': relo #%d: field offset ambiguity: %u != %u\n", |
52109584 | 6039 | prog->name, relo_idx, cand_spec.bit_offset, |
ee26dade | 6040 | targ_spec.bit_offset); |
ddc7c304 | 6041 | return -EINVAL; |
353c788c AN |
6042 | } else if (cand_res.poison != targ_res.poison || cand_res.new_val != targ_res.new_val) { |
6043 | /* all candidates should result in the same relocation | |
6044 | * decision and value, otherwise it's dangerous to | |
6045 | * proceed due to ambiguity | |
6046 | */ | |
6047 | pr_warn("prog '%s': relo #%d: relocation decision ambiguity: %s %u != %s %u\n", | |
52109584 | 6048 | prog->name, relo_idx, |
353c788c AN |
6049 | cand_res.poison ? "failure" : "success", cand_res.new_val, |
6050 | targ_res.poison ? "failure" : "success", targ_res.new_val); | |
6051 | return -EINVAL; | |
ddc7c304 AN |
6052 | } |
6053 | ||
0f7515ca | 6054 | cands->cands[j++] = cands->cands[i]; |
ddc7c304 AN |
6055 | } |
6056 | ||
62561eb4 | 6057 | /* |
d7a25270 AN |
6058 | * For BPF_FIELD_EXISTS relo or when used BPF program has field |
6059 | * existence checks or kernel version/config checks, it's expected | |
6060 | * that we might not find any candidates. In this case, if field | |
6061 | * wasn't found in any candidate, the list of candidates shouldn't | |
6062 | * change at all, we'll just handle relocating appropriately, | |
6063 | * depending on relo's kind. | |
62561eb4 AN |
6064 | */ |
6065 | if (j > 0) | |
0f7515ca | 6066 | cands->len = j; |
62561eb4 | 6067 | |
d7a25270 AN |
6068 | /* |
6069 | * If no candidates were found, it might be both a programmer error, | |
6070 | * as well as expected case, depending whether instruction w/ | |
6071 | * relocation is guarded in some way that makes it unreachable (dead | |
6072 | * code) if relocation can't be resolved. This is handled in | |
3fc32f40 | 6073 | * bpf_core_patch_insn() uniformly by replacing that instruction with |
d7a25270 AN |
6074 | * BPF helper call insn (using invalid helper ID). If that instruction |
6075 | * is indeed unreachable, then it will be ignored and eliminated by | |
6076 | * verifier. If it was an error, then verifier will complain and point | |
6077 | * to a specific instruction number in its log. | |
6078 | */ | |
353c788c | 6079 | if (j == 0) { |
28b93c64 | 6080 | pr_debug("prog '%s': relo #%d: no matching targets found\n", |
52109584 | 6081 | prog->name, relo_idx); |
ddc7c304 | 6082 | |
353c788c AN |
6083 | /* calculate single target relo result explicitly */ |
6084 | err = bpf_core_calc_relo(prog, relo, relo_idx, &local_spec, NULL, &targ_res); | |
6085 | if (err) | |
6086 | return err; | |
6087 | } | |
6088 | ||
3fc32f40 | 6089 | patch_insn: |
353c788c AN |
6090 | /* bpf_core_patch_insn() should know how to handle missing targ_spec */ |
6091 | err = bpf_core_patch_insn(prog, relo, relo_idx, &targ_res); | |
ddc7c304 | 6092 | if (err) { |
be18010e | 6093 | pr_warn("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n", |
52109584 | 6094 | prog->name, relo_idx, relo->insn_off, err); |
ddc7c304 AN |
6095 | return -EINVAL; |
6096 | } | |
6097 | ||
6098 | return 0; | |
6099 | } | |
6100 | ||
6101 | static int | |
28b93c64 | 6102 | bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) |
ddc7c304 AN |
6103 | { |
6104 | const struct btf_ext_info_sec *sec; | |
28b93c64 | 6105 | const struct bpf_core_relo *rec; |
ddc7c304 AN |
6106 | const struct btf_ext_info *seg; |
6107 | struct hashmap_entry *entry; | |
6108 | struct hashmap *cand_cache = NULL; | |
6109 | struct bpf_program *prog; | |
ddc7c304 | 6110 | const char *sec_name; |
db2b8b06 | 6111 | int i, err = 0, insn_idx, sec_idx; |
ddc7c304 | 6112 | |
28b93c64 AN |
6113 | if (obj->btf_ext->core_relo_info.len == 0) |
6114 | return 0; | |
6115 | ||
0f7515ca AN |
6116 | if (targ_btf_path) { |
6117 | obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL); | |
6118 | if (IS_ERR_OR_NULL(obj->btf_vmlinux_override)) { | |
6119 | err = PTR_ERR(obj->btf_vmlinux_override); | |
6120 | pr_warn("failed to parse target BTF: %d\n", err); | |
6121 | return err; | |
6122 | } | |
ddc7c304 AN |
6123 | } |
6124 | ||
6125 | cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL); | |
6126 | if (IS_ERR(cand_cache)) { | |
6127 | err = PTR_ERR(cand_cache); | |
6128 | goto out; | |
6129 | } | |
6130 | ||
28b93c64 | 6131 | seg = &obj->btf_ext->core_relo_info; |
ddc7c304 AN |
6132 | for_each_btf_ext_sec(seg, sec) { |
6133 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); | |
6134 | if (str_is_empty(sec_name)) { | |
6135 | err = -EINVAL; | |
6136 | goto out; | |
6137 | } | |
db2b8b06 AN |
6138 | /* bpf_object's ELF is gone by now so it's not easy to find |
6139 | * section index by section name, but we can find *any* | |
6140 | * bpf_program within desired section name and use it's | |
6141 | * prog->sec_idx to do a proper search by section index and | |
6142 | * instruction offset | |
6143 | */ | |
9c82a63c AN |
6144 | prog = NULL; |
6145 | for (i = 0; i < obj->nr_programs; i++) { | |
db2b8b06 | 6146 | prog = &obj->programs[i]; |
52109584 | 6147 | if (strcmp(prog->sec_name, sec_name) == 0) |
9c82a63c | 6148 | break; |
9c82a63c | 6149 | } |
ddc7c304 | 6150 | if (!prog) { |
db2b8b06 AN |
6151 | pr_warn("sec '%s': failed to find a BPF program\n", sec_name); |
6152 | return -ENOENT; | |
ddc7c304 | 6153 | } |
db2b8b06 | 6154 | sec_idx = prog->sec_idx; |
ddc7c304 | 6155 | |
9c0f8cbd | 6156 | pr_debug("sec '%s': found %d CO-RE relocations\n", |
ddc7c304 AN |
6157 | sec_name, sec->num_info); |
6158 | ||
6159 | for_each_btf_ext_rec(seg, sec, i, rec) { | |
db2b8b06 AN |
6160 | insn_idx = rec->insn_off / BPF_INSN_SZ; |
6161 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); | |
6162 | if (!prog) { | |
6163 | pr_warn("sec '%s': failed to find program at insn #%d for CO-RE offset relocation #%d\n", | |
6164 | sec_name, insn_idx, i); | |
6165 | err = -EINVAL; | |
6166 | goto out; | |
6167 | } | |
47f7cf63 AN |
6168 | /* no need to apply CO-RE relocation if the program is |
6169 | * not going to be loaded | |
6170 | */ | |
6171 | if (!prog->load) | |
6172 | continue; | |
db2b8b06 | 6173 | |
0f7515ca | 6174 | err = bpf_core_apply_relo(prog, rec, i, obj->btf, cand_cache); |
ddc7c304 | 6175 | if (err) { |
be18010e | 6176 | pr_warn("prog '%s': relo #%d: failed to relocate: %d\n", |
9c0f8cbd | 6177 | prog->name, i, err); |
ddc7c304 AN |
6178 | goto out; |
6179 | } | |
6180 | } | |
6181 | } | |
6182 | ||
6183 | out: | |
4f33a53d | 6184 | /* obj->btf_vmlinux and module BTFs are freed after object load */ |
0f7515ca AN |
6185 | btf__free(obj->btf_vmlinux_override); |
6186 | obj->btf_vmlinux_override = NULL; | |
6187 | ||
ddc7c304 AN |
6188 | if (!IS_ERR_OR_NULL(cand_cache)) { |
6189 | hashmap__for_each_entry(cand_cache, entry, i) { | |
6190 | bpf_core_free_cands(entry->value); | |
6191 | } | |
6192 | hashmap__free(cand_cache); | |
6193 | } | |
6194 | return err; | |
6195 | } | |
6196 | ||
c3c55696 AN |
6197 | /* Relocate data references within program code: |
6198 | * - map references; | |
6199 | * - global variable references; | |
6200 | * - extern references. | |
6201 | */ | |
48cca7e4 | 6202 | static int |
c3c55696 | 6203 | bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog) |
8a47a6c5 | 6204 | { |
c3c55696 | 6205 | int i; |
8a47a6c5 WN |
6206 | |
6207 | for (i = 0; i < prog->nr_reloc; i++) { | |
53f8dd43 | 6208 | struct reloc_desc *relo = &prog->reloc_desc[i]; |
166750bc | 6209 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; |
2e33efe3 | 6210 | struct extern_desc *ext; |
8a47a6c5 | 6211 | |
166750bc AN |
6212 | switch (relo->type) { |
6213 | case RELO_LD64: | |
6214 | insn[0].src_reg = BPF_PSEUDO_MAP_FD; | |
6215 | insn[0].imm = obj->maps[relo->map_idx].fd; | |
c3c55696 | 6216 | relo->processed = true; |
166750bc AN |
6217 | break; |
6218 | case RELO_DATA: | |
6219 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; | |
6220 | insn[1].imm = insn[0].imm + relo->sym_off; | |
53f8dd43 | 6221 | insn[0].imm = obj->maps[relo->map_idx].fd; |
c3c55696 | 6222 | relo->processed = true; |
166750bc AN |
6223 | break; |
6224 | case RELO_EXTERN: | |
2e33efe3 | 6225 | ext = &obj->externs[relo->sym_off]; |
1c0c7074 AN |
6226 | if (ext->type == EXT_KCFG) { |
6227 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; | |
6228 | insn[0].imm = obj->maps[obj->kconfig_map_idx].fd; | |
6229 | insn[1].imm = ext->kcfg.data_off; | |
6230 | } else /* EXT_KSYM */ { | |
d370bbe1 HL |
6231 | if (ext->ksym.type_id) { /* typed ksyms */ |
6232 | insn[0].src_reg = BPF_PSEUDO_BTF_ID; | |
284d2587 AN |
6233 | insn[0].imm = ext->ksym.kernel_btf_id; |
6234 | insn[1].imm = ext->ksym.kernel_btf_obj_fd; | |
d370bbe1 HL |
6235 | } else { /* typeless ksyms */ |
6236 | insn[0].imm = (__u32)ext->ksym.addr; | |
6237 | insn[1].imm = ext->ksym.addr >> 32; | |
6238 | } | |
1c0c7074 | 6239 | } |
c3c55696 | 6240 | relo->processed = true; |
166750bc | 6241 | break; |
53eddb5e YS |
6242 | case RELO_SUBPROG_ADDR: |
6243 | insn[0].src_reg = BPF_PSEUDO_FUNC; | |
6244 | /* will be handled as a follow up pass */ | |
6245 | break; | |
166750bc | 6246 | case RELO_CALL: |
c3c55696 | 6247 | /* will be handled as a follow up pass */ |
166750bc AN |
6248 | break; |
6249 | default: | |
9c0f8cbd AN |
6250 | pr_warn("prog '%s': relo #%d: bad relo type %d\n", |
6251 | prog->name, i, relo->type); | |
166750bc | 6252 | return -EINVAL; |
8a47a6c5 | 6253 | } |
8a47a6c5 WN |
6254 | } |
6255 | ||
c3c55696 AN |
6256 | return 0; |
6257 | } | |
6258 | ||
8505e870 AN |
6259 | static int adjust_prog_btf_ext_info(const struct bpf_object *obj, |
6260 | const struct bpf_program *prog, | |
6261 | const struct btf_ext_info *ext_info, | |
6262 | void **prog_info, __u32 *prog_rec_cnt, | |
6263 | __u32 *prog_rec_sz) | |
6264 | { | |
6265 | void *copy_start = NULL, *copy_end = NULL; | |
6266 | void *rec, *rec_end, *new_prog_info; | |
6267 | const struct btf_ext_info_sec *sec; | |
6268 | size_t old_sz, new_sz; | |
6269 | const char *sec_name; | |
6270 | int i, off_adj; | |
6271 | ||
6272 | for_each_btf_ext_sec(ext_info, sec) { | |
6273 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); | |
6274 | if (!sec_name) | |
6275 | return -EINVAL; | |
52109584 | 6276 | if (strcmp(sec_name, prog->sec_name) != 0) |
8505e870 AN |
6277 | continue; |
6278 | ||
6279 | for_each_btf_ext_rec(ext_info, sec, i, rec) { | |
6280 | __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ; | |
6281 | ||
6282 | if (insn_off < prog->sec_insn_off) | |
6283 | continue; | |
6284 | if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt) | |
6285 | break; | |
6286 | ||
6287 | if (!copy_start) | |
6288 | copy_start = rec; | |
6289 | copy_end = rec + ext_info->rec_size; | |
6290 | } | |
6291 | ||
6292 | if (!copy_start) | |
6293 | return -ENOENT; | |
6294 | ||
6295 | /* append func/line info of a given (sub-)program to the main | |
6296 | * program func/line info | |
6297 | */ | |
8eb62958 | 6298 | old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size; |
8505e870 AN |
6299 | new_sz = old_sz + (copy_end - copy_start); |
6300 | new_prog_info = realloc(*prog_info, new_sz); | |
6301 | if (!new_prog_info) | |
6302 | return -ENOMEM; | |
6303 | *prog_info = new_prog_info; | |
6304 | *prog_rec_cnt = new_sz / ext_info->rec_size; | |
6305 | memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start); | |
6306 | ||
6307 | /* Kernel instruction offsets are in units of 8-byte | |
6308 | * instructions, while .BTF.ext instruction offsets generated | |
6309 | * by Clang are in units of bytes. So convert Clang offsets | |
6310 | * into kernel offsets and adjust offset according to program | |
6311 | * relocated position. | |
6312 | */ | |
6313 | off_adj = prog->sub_insn_off - prog->sec_insn_off; | |
6314 | rec = new_prog_info + old_sz; | |
6315 | rec_end = new_prog_info + new_sz; | |
6316 | for (; rec < rec_end; rec += ext_info->rec_size) { | |
6317 | __u32 *insn_off = rec; | |
6318 | ||
6319 | *insn_off = *insn_off / BPF_INSN_SZ + off_adj; | |
6320 | } | |
6321 | *prog_rec_sz = ext_info->rec_size; | |
6322 | return 0; | |
6323 | } | |
6324 | ||
6325 | return -ENOENT; | |
6326 | } | |
6327 | ||
6328 | static int | |
6329 | reloc_prog_func_and_line_info(const struct bpf_object *obj, | |
6330 | struct bpf_program *main_prog, | |
6331 | const struct bpf_program *prog) | |
6332 | { | |
6333 | int err; | |
6334 | ||
6335 | /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't | |
6336 | * supprot func/line info | |
6337 | */ | |
6338 | if (!obj->btf_ext || !kernel_supports(FEAT_BTF_FUNC)) | |
6339 | return 0; | |
6340 | ||
6341 | /* only attempt func info relocation if main program's func_info | |
6342 | * relocation was successful | |
6343 | */ | |
6344 | if (main_prog != prog && !main_prog->func_info) | |
6345 | goto line_info; | |
6346 | ||
6347 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info, | |
6348 | &main_prog->func_info, | |
6349 | &main_prog->func_info_cnt, | |
6350 | &main_prog->func_info_rec_size); | |
6351 | if (err) { | |
6352 | if (err != -ENOENT) { | |
6353 | pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n", | |
6354 | prog->name, err); | |
6355 | return err; | |
6356 | } | |
6357 | if (main_prog->func_info) { | |
6358 | /* | |
6359 | * Some info has already been found but has problem | |
6360 | * in the last btf_ext reloc. Must have to error out. | |
6361 | */ | |
6362 | pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name); | |
6363 | return err; | |
6364 | } | |
6365 | /* Have problem loading the very first info. Ignore the rest. */ | |
6366 | pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n", | |
6367 | prog->name); | |
6368 | } | |
6369 | ||
6370 | line_info: | |
6371 | /* don't relocate line info if main program's relocation failed */ | |
6372 | if (main_prog != prog && !main_prog->line_info) | |
6373 | return 0; | |
6374 | ||
6375 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info, | |
6376 | &main_prog->line_info, | |
6377 | &main_prog->line_info_cnt, | |
6378 | &main_prog->line_info_rec_size); | |
6379 | if (err) { | |
6380 | if (err != -ENOENT) { | |
6381 | pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n", | |
6382 | prog->name, err); | |
6383 | return err; | |
6384 | } | |
6385 | if (main_prog->line_info) { | |
6386 | /* | |
6387 | * Some info has already been found but has problem | |
6388 | * in the last btf_ext reloc. Must have to error out. | |
6389 | */ | |
6390 | pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name); | |
6391 | return err; | |
6392 | } | |
6393 | /* Have problem loading the very first info. Ignore the rest. */ | |
6394 | pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n", | |
6395 | prog->name); | |
6396 | } | |
6397 | return 0; | |
6398 | } | |
6399 | ||
c3c55696 AN |
6400 | static int cmp_relo_by_insn_idx(const void *key, const void *elem) |
6401 | { | |
6402 | size_t insn_idx = *(const size_t *)key; | |
6403 | const struct reloc_desc *relo = elem; | |
6404 | ||
6405 | if (insn_idx == relo->insn_idx) | |
6406 | return 0; | |
6407 | return insn_idx < relo->insn_idx ? -1 : 1; | |
6408 | } | |
6409 | ||
6410 | static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx) | |
6411 | { | |
6412 | return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc, | |
6413 | sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx); | |
6414 | } | |
6415 | ||
6416 | static int | |
6417 | bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog, | |
6418 | struct bpf_program *prog) | |
6419 | { | |
6420 | size_t sub_insn_idx, insn_idx, new_cnt; | |
6421 | struct bpf_program *subprog; | |
6422 | struct bpf_insn *insns, *insn; | |
6423 | struct reloc_desc *relo; | |
6424 | int err; | |
6425 | ||
6426 | err = reloc_prog_func_and_line_info(obj, main_prog, prog); | |
6427 | if (err) | |
6428 | return err; | |
6429 | ||
6430 | for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) { | |
6431 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
53eddb5e | 6432 | if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn)) |
c3c55696 AN |
6433 | continue; |
6434 | ||
6435 | relo = find_prog_insn_relo(prog, insn_idx); | |
53eddb5e | 6436 | if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) { |
c3c55696 AN |
6437 | pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n", |
6438 | prog->name, insn_idx, relo->type); | |
6439 | return -LIBBPF_ERRNO__RELOC; | |
6440 | } | |
6441 | if (relo) { | |
6442 | /* sub-program instruction index is a combination of | |
6443 | * an offset of a symbol pointed to by relocation and | |
6444 | * call instruction's imm field; for global functions, | |
6445 | * call always has imm = -1, but for static functions | |
6446 | * relocation is against STT_SECTION and insn->imm | |
6447 | * points to a start of a static function | |
53eddb5e YS |
6448 | * |
6449 | * for subprog addr relocation, the relo->sym_off + insn->imm is | |
6450 | * the byte offset in the corresponding section. | |
c3c55696 | 6451 | */ |
53eddb5e YS |
6452 | if (relo->type == RELO_CALL) |
6453 | sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1; | |
6454 | else | |
6455 | sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ; | |
6456 | } else if (insn_is_pseudo_func(insn)) { | |
6457 | /* | |
6458 | * RELO_SUBPROG_ADDR relo is always emitted even if both | |
6459 | * functions are in the same section, so it shouldn't reach here. | |
6460 | */ | |
6461 | pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n", | |
6462 | prog->name, insn_idx); | |
6463 | return -LIBBPF_ERRNO__RELOC; | |
c3c55696 AN |
6464 | } else { |
6465 | /* if subprogram call is to a static function within | |
6466 | * the same ELF section, there won't be any relocation | |
6467 | * emitted, but it also means there is no additional | |
6468 | * offset necessary, insns->imm is relative to | |
6469 | * instruction's original position within the section | |
6470 | */ | |
6471 | sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1; | |
6472 | } | |
6473 | ||
6474 | /* we enforce that sub-programs should be in .text section */ | |
6475 | subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx); | |
6476 | if (!subprog) { | |
6477 | pr_warn("prog '%s': no .text section found yet sub-program call exists\n", | |
6478 | prog->name); | |
6479 | return -LIBBPF_ERRNO__RELOC; | |
6480 | } | |
6481 | ||
6482 | /* if it's the first call instruction calling into this | |
6483 | * subprogram (meaning this subprog hasn't been processed | |
6484 | * yet) within the context of current main program: | |
6485 | * - append it at the end of main program's instructions blog; | |
6486 | * - process is recursively, while current program is put on hold; | |
6487 | * - if that subprogram calls some other not yet processes | |
6488 | * subprogram, same thing will happen recursively until | |
6489 | * there are no more unprocesses subprograms left to append | |
6490 | * and relocate. | |
6491 | */ | |
6492 | if (subprog->sub_insn_off == 0) { | |
6493 | subprog->sub_insn_off = main_prog->insns_cnt; | |
6494 | ||
6495 | new_cnt = main_prog->insns_cnt + subprog->insns_cnt; | |
6496 | insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns)); | |
6497 | if (!insns) { | |
6498 | pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name); | |
6499 | return -ENOMEM; | |
6500 | } | |
6501 | main_prog->insns = insns; | |
6502 | main_prog->insns_cnt = new_cnt; | |
6503 | ||
6504 | memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns, | |
6505 | subprog->insns_cnt * sizeof(*insns)); | |
6506 | ||
6507 | pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n", | |
6508 | main_prog->name, subprog->insns_cnt, subprog->name); | |
6509 | ||
6510 | err = bpf_object__reloc_code(obj, main_prog, subprog); | |
6511 | if (err) | |
6512 | return err; | |
6513 | } | |
6514 | ||
6515 | /* main_prog->insns memory could have been re-allocated, so | |
6516 | * calculate pointer again | |
6517 | */ | |
6518 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
6519 | /* calculate correct instruction position within current main | |
6520 | * prog; each main prog can have a different set of | |
6521 | * subprograms appended (potentially in different order as | |
6522 | * well), so position of any subprog can be different for | |
6523 | * different main programs */ | |
6524 | insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1; | |
6525 | ||
6526 | if (relo) | |
6527 | relo->processed = true; | |
6528 | ||
6529 | pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n", | |
6530 | prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off); | |
6531 | } | |
6532 | ||
6533 | return 0; | |
6534 | } | |
6535 | ||
6536 | /* | |
6537 | * Relocate sub-program calls. | |
6538 | * | |
6539 | * Algorithm operates as follows. Each entry-point BPF program (referred to as | |
6540 | * main prog) is processed separately. For each subprog (non-entry functions, | |
6541 | * that can be called from either entry progs or other subprogs) gets their | |
6542 | * sub_insn_off reset to zero. This serves as indicator that this subprogram | |
6543 | * hasn't been yet appended and relocated within current main prog. Once its | |
6544 | * relocated, sub_insn_off will point at the position within current main prog | |
6545 | * where given subprog was appended. This will further be used to relocate all | |
6546 | * the call instructions jumping into this subprog. | |
6547 | * | |
6548 | * We start with main program and process all call instructions. If the call | |
6549 | * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off | |
6550 | * is zero), subprog instructions are appended at the end of main program's | |
6551 | * instruction array. Then main program is "put on hold" while we recursively | |
6552 | * process newly appended subprogram. If that subprogram calls into another | |
6553 | * subprogram that hasn't been appended, new subprogram is appended again to | |
6554 | * the *main* prog's instructions (subprog's instructions are always left | |
6555 | * untouched, as they need to be in unmodified state for subsequent main progs | |
6556 | * and subprog instructions are always sent only as part of a main prog) and | |
6557 | * the process continues recursively. Once all the subprogs called from a main | |
6558 | * prog or any of its subprogs are appended (and relocated), all their | |
6559 | * positions within finalized instructions array are known, so it's easy to | |
6560 | * rewrite call instructions with correct relative offsets, corresponding to | |
6561 | * desired target subprog. | |
6562 | * | |
6563 | * Its important to realize that some subprogs might not be called from some | |
6564 | * main prog and any of its called/used subprogs. Those will keep their | |
6565 | * subprog->sub_insn_off as zero at all times and won't be appended to current | |
6566 | * main prog and won't be relocated within the context of current main prog. | |
6567 | * They might still be used from other main progs later. | |
6568 | * | |
6569 | * Visually this process can be shown as below. Suppose we have two main | |
6570 | * programs mainA and mainB and BPF object contains three subprogs: subA, | |
6571 | * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and | |
6572 | * subC both call subB: | |
6573 | * | |
6574 | * +--------+ +-------+ | |
6575 | * | v v | | |
6576 | * +--+---+ +--+-+-+ +---+--+ | |
6577 | * | subA | | subB | | subC | | |
6578 | * +--+---+ +------+ +---+--+ | |
6579 | * ^ ^ | |
6580 | * | | | |
6581 | * +---+-------+ +------+----+ | |
6582 | * | mainA | | mainB | | |
6583 | * +-----------+ +-----------+ | |
6584 | * | |
6585 | * We'll start relocating mainA, will find subA, append it and start | |
6586 | * processing sub A recursively: | |
6587 | * | |
6588 | * +-----------+------+ | |
6589 | * | mainA | subA | | |
6590 | * +-----------+------+ | |
6591 | * | |
6592 | * At this point we notice that subB is used from subA, so we append it and | |
6593 | * relocate (there are no further subcalls from subB): | |
6594 | * | |
6595 | * +-----------+------+------+ | |
6596 | * | mainA | subA | subB | | |
6597 | * +-----------+------+------+ | |
6598 | * | |
6599 | * At this point, we relocate subA calls, then go one level up and finish with | |
6600 | * relocatin mainA calls. mainA is done. | |
6601 | * | |
6602 | * For mainB process is similar but results in different order. We start with | |
6603 | * mainB and skip subA and subB, as mainB never calls them (at least | |
6604 | * directly), but we see subC is needed, so we append and start processing it: | |
6605 | * | |
6606 | * +-----------+------+ | |
6607 | * | mainB | subC | | |
6608 | * +-----------+------+ | |
6609 | * Now we see subC needs subB, so we go back to it, append and relocate it: | |
6610 | * | |
6611 | * +-----------+------+------+ | |
6612 | * | mainB | subC | subB | | |
6613 | * +-----------+------+------+ | |
6614 | * | |
6615 | * At this point we unwind recursion, relocate calls in subC, then in mainB. | |
6616 | */ | |
6617 | static int | |
6618 | bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog) | |
6619 | { | |
6620 | struct bpf_program *subprog; | |
6621 | int i, j, err; | |
6622 | ||
c3c55696 AN |
6623 | /* mark all subprogs as not relocated (yet) within the context of |
6624 | * current main program | |
6625 | */ | |
6626 | for (i = 0; i < obj->nr_programs; i++) { | |
6627 | subprog = &obj->programs[i]; | |
6628 | if (!prog_is_subprog(obj, subprog)) | |
6629 | continue; | |
6630 | ||
6631 | subprog->sub_insn_off = 0; | |
6632 | for (j = 0; j < subprog->nr_reloc; j++) | |
6633 | if (subprog->reloc_desc[j].type == RELO_CALL) | |
6634 | subprog->reloc_desc[j].processed = false; | |
6635 | } | |
6636 | ||
6637 | err = bpf_object__reloc_code(obj, prog, prog); | |
6638 | if (err) | |
6639 | return err; | |
6640 | ||
6641 | ||
8a47a6c5 WN |
6642 | return 0; |
6643 | } | |
6644 | ||
8a47a6c5 | 6645 | static int |
ddc7c304 | 6646 | bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path) |
8a47a6c5 WN |
6647 | { |
6648 | struct bpf_program *prog; | |
6649 | size_t i; | |
6650 | int err; | |
6651 | ||
ddc7c304 AN |
6652 | if (obj->btf_ext) { |
6653 | err = bpf_object__relocate_core(obj, targ_btf_path); | |
6654 | if (err) { | |
be18010e KW |
6655 | pr_warn("failed to perform CO-RE relocations: %d\n", |
6656 | err); | |
ddc7c304 AN |
6657 | return err; |
6658 | } | |
6659 | } | |
c3c55696 AN |
6660 | /* relocate data references first for all programs and sub-programs, |
6661 | * as they don't change relative to code locations, so subsequent | |
6662 | * subprogram processing won't need to re-calculate any of them | |
9173cac3 AN |
6663 | */ |
6664 | for (i = 0; i < obj->nr_programs; i++) { | |
6665 | prog = &obj->programs[i]; | |
c3c55696 | 6666 | err = bpf_object__relocate_data(obj, prog); |
9173cac3 | 6667 | if (err) { |
9c0f8cbd AN |
6668 | pr_warn("prog '%s': failed to relocate data references: %d\n", |
6669 | prog->name, err); | |
9173cac3 AN |
6670 | return err; |
6671 | } | |
9173cac3 | 6672 | } |
c3c55696 AN |
6673 | /* now relocate subprogram calls and append used subprograms to main |
6674 | * programs; each copy of subprogram code needs to be relocated | |
6675 | * differently for each main program, because its code location might | |
6676 | * have changed | |
9173cac3 | 6677 | */ |
8a47a6c5 WN |
6678 | for (i = 0; i < obj->nr_programs; i++) { |
6679 | prog = &obj->programs[i]; | |
c3c55696 AN |
6680 | /* sub-program's sub-calls are relocated within the context of |
6681 | * its main program only | |
6682 | */ | |
6683 | if (prog_is_subprog(obj, prog)) | |
9173cac3 | 6684 | continue; |
8a47a6c5 | 6685 | |
c3c55696 | 6686 | err = bpf_object__relocate_calls(obj, prog); |
8a47a6c5 | 6687 | if (err) { |
9c0f8cbd AN |
6688 | pr_warn("prog '%s': failed to relocate calls: %d\n", |
6689 | prog->name, err); | |
8a47a6c5 WN |
6690 | return err; |
6691 | } | |
6692 | } | |
c3c55696 AN |
6693 | /* free up relocation descriptors */ |
6694 | for (i = 0; i < obj->nr_programs; i++) { | |
6695 | prog = &obj->programs[i]; | |
6696 | zfree(&prog->reloc_desc); | |
6697 | prog->nr_reloc = 0; | |
6698 | } | |
8a47a6c5 WN |
6699 | return 0; |
6700 | } | |
6701 | ||
646f02ff AN |
6702 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
6703 | GElf_Shdr *shdr, Elf_Data *data); | |
6704 | ||
6705 | static int bpf_object__collect_map_relos(struct bpf_object *obj, | |
6706 | GElf_Shdr *shdr, Elf_Data *data) | |
6707 | { | |
15728ad3 AN |
6708 | const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *); |
6709 | int i, j, nrels, new_sz; | |
063e6881 | 6710 | const struct btf_var_secinfo *vi = NULL; |
646f02ff | 6711 | const struct btf_type *sec, *var, *def; |
3168c158 | 6712 | struct bpf_map *map = NULL, *targ_map; |
646f02ff | 6713 | const struct btf_member *member; |
646f02ff AN |
6714 | const char *name, *mname; |
6715 | Elf_Data *symbols; | |
6716 | unsigned int moff; | |
6717 | GElf_Sym sym; | |
6718 | GElf_Rel rel; | |
6719 | void *tmp; | |
6720 | ||
6721 | if (!obj->efile.btf_maps_sec_btf_id || !obj->btf) | |
6722 | return -EINVAL; | |
6723 | sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id); | |
6724 | if (!sec) | |
6725 | return -EINVAL; | |
6726 | ||
6727 | symbols = obj->efile.symbols; | |
6728 | nrels = shdr->sh_size / shdr->sh_entsize; | |
6729 | for (i = 0; i < nrels; i++) { | |
6730 | if (!gelf_getrel(data, i, &rel)) { | |
6731 | pr_warn(".maps relo #%d: failed to get ELF relo\n", i); | |
6732 | return -LIBBPF_ERRNO__FORMAT; | |
6733 | } | |
6734 | if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) { | |
6735 | pr_warn(".maps relo #%d: symbol %zx not found\n", | |
6736 | i, (size_t)GELF_R_SYM(rel.r_info)); | |
6737 | return -LIBBPF_ERRNO__FORMAT; | |
6738 | } | |
88a82120 | 6739 | name = elf_sym_str(obj, sym.st_name) ?: "<?>"; |
646f02ff AN |
6740 | if (sym.st_shndx != obj->efile.btf_maps_shndx) { |
6741 | pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n", | |
6742 | i, name); | |
6743 | return -LIBBPF_ERRNO__RELOC; | |
6744 | } | |
6745 | ||
6746 | pr_debug(".maps relo #%d: for %zd value %zd rel.r_offset %zu name %d ('%s')\n", | |
6747 | i, (ssize_t)(rel.r_info >> 32), (size_t)sym.st_value, | |
6748 | (size_t)rel.r_offset, sym.st_name, name); | |
6749 | ||
6750 | for (j = 0; j < obj->nr_maps; j++) { | |
6751 | map = &obj->maps[j]; | |
6752 | if (map->sec_idx != obj->efile.btf_maps_shndx) | |
6753 | continue; | |
6754 | ||
6755 | vi = btf_var_secinfos(sec) + map->btf_var_idx; | |
6756 | if (vi->offset <= rel.r_offset && | |
15728ad3 | 6757 | rel.r_offset + bpf_ptr_sz <= vi->offset + vi->size) |
646f02ff AN |
6758 | break; |
6759 | } | |
6760 | if (j == obj->nr_maps) { | |
6761 | pr_warn(".maps relo #%d: cannot find map '%s' at rel.r_offset %zu\n", | |
6762 | i, name, (size_t)rel.r_offset); | |
6763 | return -EINVAL; | |
6764 | } | |
6765 | ||
6766 | if (!bpf_map_type__is_map_in_map(map->def.type)) | |
6767 | return -EINVAL; | |
6768 | if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS && | |
6769 | map->def.key_size != sizeof(int)) { | |
6770 | pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n", | |
6771 | i, map->name, sizeof(int)); | |
6772 | return -EINVAL; | |
6773 | } | |
6774 | ||
6775 | targ_map = bpf_object__find_map_by_name(obj, name); | |
6776 | if (!targ_map) | |
6777 | return -ESRCH; | |
6778 | ||
6779 | var = btf__type_by_id(obj->btf, vi->type); | |
6780 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
6781 | if (btf_vlen(def) == 0) | |
6782 | return -EINVAL; | |
6783 | member = btf_members(def) + btf_vlen(def) - 1; | |
6784 | mname = btf__name_by_offset(obj->btf, member->name_off); | |
6785 | if (strcmp(mname, "values")) | |
6786 | return -EINVAL; | |
6787 | ||
6788 | moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8; | |
6789 | if (rel.r_offset - vi->offset < moff) | |
6790 | return -EINVAL; | |
6791 | ||
6792 | moff = rel.r_offset - vi->offset - moff; | |
15728ad3 AN |
6793 | /* here we use BPF pointer size, which is always 64 bit, as we |
6794 | * are parsing ELF that was built for BPF target | |
6795 | */ | |
6796 | if (moff % bpf_ptr_sz) | |
646f02ff | 6797 | return -EINVAL; |
15728ad3 | 6798 | moff /= bpf_ptr_sz; |
646f02ff AN |
6799 | if (moff >= map->init_slots_sz) { |
6800 | new_sz = moff + 1; | |
029258d7 | 6801 | tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz); |
646f02ff AN |
6802 | if (!tmp) |
6803 | return -ENOMEM; | |
6804 | map->init_slots = tmp; | |
6805 | memset(map->init_slots + map->init_slots_sz, 0, | |
15728ad3 | 6806 | (new_sz - map->init_slots_sz) * host_ptr_sz); |
646f02ff AN |
6807 | map->init_slots_sz = new_sz; |
6808 | } | |
6809 | map->init_slots[moff] = targ_map; | |
6810 | ||
6811 | pr_debug(".maps relo #%d: map '%s' slot [%d] points to map '%s'\n", | |
6812 | i, map->name, moff, name); | |
6813 | } | |
6814 | ||
6815 | return 0; | |
6816 | } | |
590a0088 | 6817 | |
c3c55696 | 6818 | static int cmp_relocs(const void *_a, const void *_b) |
34090915 | 6819 | { |
c3c55696 AN |
6820 | const struct reloc_desc *a = _a; |
6821 | const struct reloc_desc *b = _b; | |
34090915 | 6822 | |
c3c55696 AN |
6823 | if (a->insn_idx != b->insn_idx) |
6824 | return a->insn_idx < b->insn_idx ? -1 : 1; | |
6825 | ||
6826 | /* no two relocations should have the same insn_idx, but ... */ | |
6827 | if (a->type != b->type) | |
6828 | return a->type < b->type ? -1 : 1; | |
6829 | ||
6830 | return 0; | |
6831 | } | |
6832 | ||
6833 | static int bpf_object__collect_relos(struct bpf_object *obj) | |
6834 | { | |
6835 | int i, err; | |
34090915 | 6836 | |
1f8e2bcb AN |
6837 | for (i = 0; i < obj->efile.nr_reloc_sects; i++) { |
6838 | GElf_Shdr *shdr = &obj->efile.reloc_sects[i].shdr; | |
6839 | Elf_Data *data = obj->efile.reloc_sects[i].data; | |
34090915 | 6840 | int idx = shdr->sh_info; |
34090915 WN |
6841 | |
6842 | if (shdr->sh_type != SHT_REL) { | |
be18010e | 6843 | pr_warn("internal error at %d\n", __LINE__); |
6371ca3b | 6844 | return -LIBBPF_ERRNO__INTERNAL; |
34090915 WN |
6845 | } |
6846 | ||
c3c55696 | 6847 | if (idx == obj->efile.st_ops_shndx) |
646f02ff | 6848 | err = bpf_object__collect_st_ops_relos(obj, shdr, data); |
c3c55696 | 6849 | else if (idx == obj->efile.btf_maps_shndx) |
646f02ff | 6850 | err = bpf_object__collect_map_relos(obj, shdr, data); |
c3c55696 AN |
6851 | else |
6852 | err = bpf_object__collect_prog_relos(obj, shdr, data); | |
34090915 | 6853 | if (err) |
6371ca3b | 6854 | return err; |
34090915 | 6855 | } |
c3c55696 AN |
6856 | |
6857 | for (i = 0; i < obj->nr_programs; i++) { | |
6858 | struct bpf_program *p = &obj->programs[i]; | |
6859 | ||
6860 | if (!p->nr_reloc) | |
6861 | continue; | |
6862 | ||
6863 | qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs); | |
6864 | } | |
34090915 WN |
6865 | return 0; |
6866 | } | |
6867 | ||
109cea5a AN |
6868 | static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id) |
6869 | { | |
9b2f6fec | 6870 | if (BPF_CLASS(insn->code) == BPF_JMP && |
109cea5a AN |
6871 | BPF_OP(insn->code) == BPF_CALL && |
6872 | BPF_SRC(insn->code) == BPF_K && | |
9b2f6fec AN |
6873 | insn->src_reg == 0 && |
6874 | insn->dst_reg == 0) { | |
6875 | *func_id = insn->imm; | |
109cea5a AN |
6876 | return true; |
6877 | } | |
6878 | return false; | |
6879 | } | |
6880 | ||
6881 | static int bpf_object__sanitize_prog(struct bpf_object* obj, struct bpf_program *prog) | |
6882 | { | |
6883 | struct bpf_insn *insn = prog->insns; | |
6884 | enum bpf_func_id func_id; | |
6885 | int i; | |
6886 | ||
6887 | for (i = 0; i < prog->insns_cnt; i++, insn++) { | |
6888 | if (!insn_is_helper_call(insn, &func_id)) | |
6889 | continue; | |
6890 | ||
6891 | /* on kernels that don't yet support | |
6892 | * bpf_probe_read_{kernel,user}[_str] helpers, fall back | |
6893 | * to bpf_probe_read() which works well for old kernels | |
6894 | */ | |
6895 | switch (func_id) { | |
6896 | case BPF_FUNC_probe_read_kernel: | |
6897 | case BPF_FUNC_probe_read_user: | |
6898 | if (!kernel_supports(FEAT_PROBE_READ_KERN)) | |
6899 | insn->imm = BPF_FUNC_probe_read; | |
6900 | break; | |
6901 | case BPF_FUNC_probe_read_kernel_str: | |
6902 | case BPF_FUNC_probe_read_user_str: | |
6903 | if (!kernel_supports(FEAT_PROBE_READ_KERN)) | |
6904 | insn->imm = BPF_FUNC_probe_read_str; | |
6905 | break; | |
6906 | default: | |
6907 | break; | |
6908 | } | |
6909 | } | |
6910 | return 0; | |
6911 | } | |
6912 | ||
55cffde2 | 6913 | static int |
2993e051 | 6914 | load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt, |
f0187f0b | 6915 | char *license, __u32 kern_version, int *pfd) |
55cffde2 | 6916 | { |
6aef10a4 | 6917 | struct bpf_prog_load_params load_attr = {}; |
1ce6a9fc | 6918 | char *cp, errmsg[STRERR_BUFSIZE]; |
8395f320 SF |
6919 | size_t log_buf_size = 0; |
6920 | char *log_buf = NULL; | |
5d01ab7b | 6921 | int btf_fd, ret; |
55cffde2 | 6922 | |
80b2b5c3 AM |
6923 | if (prog->type == BPF_PROG_TYPE_UNSPEC) { |
6924 | /* | |
6925 | * The program type must be set. Most likely we couldn't find a proper | |
6926 | * section definition at load time, and thus we didn't infer the type. | |
6927 | */ | |
6928 | pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n", | |
6929 | prog->name, prog->sec_name); | |
6930 | return -EINVAL; | |
6931 | } | |
6932 | ||
fba01a06 AN |
6933 | if (!insns || !insns_cnt) |
6934 | return -EINVAL; | |
6935 | ||
2993e051 | 6936 | load_attr.prog_type = prog->type; |
25498a19 | 6937 | /* old kernels might not support specifying expected_attach_type */ |
47b6cb4d | 6938 | if (!kernel_supports(FEAT_EXP_ATTACH_TYPE) && prog->sec_def && |
25498a19 AN |
6939 | prog->sec_def->is_exp_attach_type_optional) |
6940 | load_attr.expected_attach_type = 0; | |
6941 | else | |
6942 | load_attr.expected_attach_type = prog->expected_attach_type; | |
47b6cb4d | 6943 | if (kernel_supports(FEAT_PROG_NAME)) |
5b32a23e | 6944 | load_attr.name = prog->name; |
d7be143b | 6945 | load_attr.insns = insns; |
6aef10a4 | 6946 | load_attr.insn_cnt = insns_cnt; |
d7be143b | 6947 | load_attr.license = license; |
6aef10a4 | 6948 | load_attr.attach_btf_id = prog->attach_btf_id; |
91abb4a6 | 6949 | if (prog->attach_prog_fd) |
e7bf94db | 6950 | load_attr.attach_prog_fd = prog->attach_prog_fd; |
91abb4a6 AN |
6951 | else |
6952 | load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd; | |
6aef10a4 AN |
6953 | load_attr.attach_btf_id = prog->attach_btf_id; |
6954 | load_attr.kern_version = kern_version; | |
6955 | load_attr.prog_ifindex = prog->prog_ifindex; | |
6956 | ||
0f0e55d8 AN |
6957 | /* specify func_info/line_info only if kernel supports them */ |
6958 | btf_fd = bpf_object__btf_fd(prog->obj); | |
47b6cb4d | 6959 | if (btf_fd >= 0 && kernel_supports(FEAT_BTF_FUNC)) { |
0f0e55d8 AN |
6960 | load_attr.prog_btf_fd = btf_fd; |
6961 | load_attr.func_info = prog->func_info; | |
6962 | load_attr.func_info_rec_size = prog->func_info_rec_size; | |
6963 | load_attr.func_info_cnt = prog->func_info_cnt; | |
6964 | load_attr.line_info = prog->line_info; | |
6965 | load_attr.line_info_rec_size = prog->line_info_rec_size; | |
6966 | load_attr.line_info_cnt = prog->line_info_cnt; | |
6967 | } | |
da11b417 | 6968 | load_attr.log_level = prog->log_level; |
04656198 | 6969 | load_attr.prog_flags = prog->prog_flags; |
55cffde2 | 6970 | |
da11b417 | 6971 | retry_load: |
8395f320 SF |
6972 | if (log_buf_size) { |
6973 | log_buf = malloc(log_buf_size); | |
6974 | if (!log_buf) | |
6975 | return -ENOMEM; | |
6976 | ||
6977 | *log_buf = 0; | |
6978 | } | |
55cffde2 | 6979 | |
6aef10a4 AN |
6980 | load_attr.log_buf = log_buf; |
6981 | load_attr.log_buf_sz = log_buf_size; | |
6982 | ret = libbpf__bpf_prog_load(&load_attr); | |
55cffde2 WN |
6983 | |
6984 | if (ret >= 0) { | |
8395f320 | 6985 | if (log_buf && load_attr.log_level) |
da11b417 | 6986 | pr_debug("verifier log:\n%s", log_buf); |
5d23328d YZ |
6987 | |
6988 | if (prog->obj->rodata_map_idx >= 0 && | |
6989 | kernel_supports(FEAT_PROG_BIND_MAP)) { | |
6990 | struct bpf_map *rodata_map = | |
6991 | &prog->obj->maps[prog->obj->rodata_map_idx]; | |
6992 | ||
6993 | if (bpf_prog_bind_map(ret, bpf_map__fd(rodata_map), NULL)) { | |
6994 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); | |
6995 | pr_warn("prog '%s': failed to bind .rodata map: %s\n", | |
6996 | prog->name, cp); | |
6997 | /* Don't fail hard if can't bind rodata. */ | |
6998 | } | |
6999 | } | |
7000 | ||
55cffde2 WN |
7001 | *pfd = ret; |
7002 | ret = 0; | |
7003 | goto out; | |
7004 | } | |
7005 | ||
8395f320 SF |
7006 | if (!log_buf || errno == ENOSPC) { |
7007 | log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, | |
7008 | log_buf_size << 1); | |
7009 | ||
da11b417 AS |
7010 | free(log_buf); |
7011 | goto retry_load; | |
7012 | } | |
ef05afa6 | 7013 | ret = errno ? -errno : -LIBBPF_ERRNO__LOAD; |
24d6a808 | 7014 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
be18010e | 7015 | pr_warn("load bpf program failed: %s\n", cp); |
dc3a2d25 | 7016 | pr_perm_msg(ret); |
55cffde2 | 7017 | |
6371ca3b WN |
7018 | if (log_buf && log_buf[0] != '\0') { |
7019 | ret = -LIBBPF_ERRNO__VERIFY; | |
be18010e KW |
7020 | pr_warn("-- BEGIN DUMP LOG ---\n"); |
7021 | pr_warn("\n%s\n", log_buf); | |
7022 | pr_warn("-- END LOG --\n"); | |
6aef10a4 | 7023 | } else if (load_attr.insn_cnt >= BPF_MAXINSNS) { |
be18010e | 7024 | pr_warn("Program too large (%zu insns), at most %d insns\n", |
6aef10a4 | 7025 | load_attr.insn_cnt, BPF_MAXINSNS); |
705fa219 | 7026 | ret = -LIBBPF_ERRNO__PROG2BIG; |
4f33ddb4 | 7027 | } else if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) { |
705fa219 | 7028 | /* Wrong program type? */ |
4f33ddb4 | 7029 | int fd; |
705fa219 | 7030 | |
4f33ddb4 THJ |
7031 | load_attr.prog_type = BPF_PROG_TYPE_KPROBE; |
7032 | load_attr.expected_attach_type = 0; | |
6aef10a4 AN |
7033 | load_attr.log_buf = NULL; |
7034 | load_attr.log_buf_sz = 0; | |
7035 | fd = libbpf__bpf_prog_load(&load_attr); | |
4f33ddb4 THJ |
7036 | if (fd >= 0) { |
7037 | close(fd); | |
7038 | ret = -LIBBPF_ERRNO__PROGTYPE; | |
7039 | goto out; | |
7040 | } | |
55cffde2 WN |
7041 | } |
7042 | ||
7043 | out: | |
7044 | free(log_buf); | |
7045 | return ret; | |
7046 | } | |
7047 | ||
91abb4a6 | 7048 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, int *btf_obj_fd, int *btf_type_id); |
13acb508 AN |
7049 | |
7050 | int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver) | |
55cffde2 | 7051 | { |
91abb4a6 | 7052 | int err = 0, fd, i; |
13acb508 | 7053 | |
d9297581 | 7054 | if (prog->obj->loaded) { |
52109584 | 7055 | pr_warn("prog '%s': can't load after object was loaded\n", prog->name); |
d9297581 AN |
7056 | return -EINVAL; |
7057 | } | |
7058 | ||
ff26ce5c | 7059 | if ((prog->type == BPF_PROG_TYPE_TRACING || |
1e092a03 | 7060 | prog->type == BPF_PROG_TYPE_LSM || |
ff26ce5c | 7061 | prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) { |
91abb4a6 AN |
7062 | int btf_obj_fd = 0, btf_type_id = 0; |
7063 | ||
7064 | err = libbpf_find_attach_btf_id(prog, &btf_obj_fd, &btf_type_id); | |
7065 | if (err) | |
7066 | return err; | |
7067 | ||
7068 | prog->attach_btf_obj_fd = btf_obj_fd; | |
7069 | prog->attach_btf_id = btf_type_id; | |
13acb508 | 7070 | } |
55cffde2 | 7071 | |
b580563e WN |
7072 | if (prog->instances.nr < 0 || !prog->instances.fds) { |
7073 | if (prog->preprocessor) { | |
be18010e | 7074 | pr_warn("Internal error: can't load program '%s'\n", |
52109584 | 7075 | prog->name); |
b580563e WN |
7076 | return -LIBBPF_ERRNO__INTERNAL; |
7077 | } | |
55cffde2 | 7078 | |
b580563e WN |
7079 | prog->instances.fds = malloc(sizeof(int)); |
7080 | if (!prog->instances.fds) { | |
be18010e | 7081 | pr_warn("Not enough memory for BPF fds\n"); |
b580563e WN |
7082 | return -ENOMEM; |
7083 | } | |
7084 | prog->instances.nr = 1; | |
7085 | prog->instances.fds[0] = -1; | |
7086 | } | |
7087 | ||
7088 | if (!prog->preprocessor) { | |
7089 | if (prog->instances.nr != 1) { | |
52109584 AN |
7090 | pr_warn("prog '%s': inconsistent nr(%d) != 1\n", |
7091 | prog->name, prog->instances.nr); | |
b580563e | 7092 | } |
2993e051 | 7093 | err = load_program(prog, prog->insns, prog->insns_cnt, |
13acb508 | 7094 | license, kern_ver, &fd); |
b580563e WN |
7095 | if (!err) |
7096 | prog->instances.fds[0] = fd; | |
7097 | goto out; | |
7098 | } | |
7099 | ||
7100 | for (i = 0; i < prog->instances.nr; i++) { | |
7101 | struct bpf_prog_prep_result result; | |
7102 | bpf_program_prep_t preprocessor = prog->preprocessor; | |
7103 | ||
1ad9cbb8 | 7104 | memset(&result, 0, sizeof(result)); |
b580563e WN |
7105 | err = preprocessor(prog, i, prog->insns, |
7106 | prog->insns_cnt, &result); | |
7107 | if (err) { | |
be18010e | 7108 | pr_warn("Preprocessing the %dth instance of program '%s' failed\n", |
52109584 | 7109 | i, prog->name); |
b580563e WN |
7110 | goto out; |
7111 | } | |
7112 | ||
7113 | if (!result.new_insn_ptr || !result.new_insn_cnt) { | |
7114 | pr_debug("Skip loading the %dth instance of program '%s'\n", | |
52109584 | 7115 | i, prog->name); |
b580563e WN |
7116 | prog->instances.fds[i] = -1; |
7117 | if (result.pfd) | |
7118 | *result.pfd = -1; | |
7119 | continue; | |
7120 | } | |
7121 | ||
2993e051 | 7122 | err = load_program(prog, result.new_insn_ptr, |
13acb508 | 7123 | result.new_insn_cnt, license, kern_ver, &fd); |
b580563e | 7124 | if (err) { |
be18010e | 7125 | pr_warn("Loading the %dth instance of program '%s' failed\n", |
52109584 | 7126 | i, prog->name); |
b580563e WN |
7127 | goto out; |
7128 | } | |
7129 | ||
7130 | if (result.pfd) | |
7131 | *result.pfd = fd; | |
7132 | prog->instances.fds[i] = fd; | |
7133 | } | |
7134 | out: | |
55cffde2 | 7135 | if (err) |
52109584 | 7136 | pr_warn("failed to load program '%s'\n", prog->name); |
55cffde2 WN |
7137 | zfree(&prog->insns); |
7138 | prog->insns_cnt = 0; | |
7139 | return err; | |
7140 | } | |
7141 | ||
7142 | static int | |
60276f98 | 7143 | bpf_object__load_progs(struct bpf_object *obj, int log_level) |
55cffde2 | 7144 | { |
d9297581 | 7145 | struct bpf_program *prog; |
55cffde2 WN |
7146 | size_t i; |
7147 | int err; | |
7148 | ||
109cea5a AN |
7149 | for (i = 0; i < obj->nr_programs; i++) { |
7150 | prog = &obj->programs[i]; | |
7151 | err = bpf_object__sanitize_prog(obj, prog); | |
7152 | if (err) | |
7153 | return err; | |
7154 | } | |
7155 | ||
55cffde2 | 7156 | for (i = 0; i < obj->nr_programs; i++) { |
d9297581 | 7157 | prog = &obj->programs[i]; |
c3c55696 | 7158 | if (prog_is_subprog(obj, prog)) |
48cca7e4 | 7159 | continue; |
d9297581 | 7160 | if (!prog->load) { |
9c0f8cbd | 7161 | pr_debug("prog '%s': skipped loading\n", prog->name); |
d9297581 AN |
7162 | continue; |
7163 | } | |
7164 | prog->log_level |= log_level; | |
7165 | err = bpf_program__load(prog, obj->license, obj->kern_version); | |
55cffde2 WN |
7166 | if (err) |
7167 | return err; | |
7168 | } | |
7169 | return 0; | |
7170 | } | |
7171 | ||
25498a19 AN |
7172 | static const struct bpf_sec_def *find_sec_def(const char *sec_name); |
7173 | ||
1a5e3fb1 | 7174 | static struct bpf_object * |
5e61f270 | 7175 | __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz, |
01af3bf0 | 7176 | const struct bpf_object_open_opts *opts) |
1a5e3fb1 | 7177 | { |
8601fd42 | 7178 | const char *obj_name, *kconfig; |
dd4436bb | 7179 | struct bpf_program *prog; |
1a5e3fb1 | 7180 | struct bpf_object *obj; |
291ee02b | 7181 | char tmp_name[64]; |
6371ca3b | 7182 | int err; |
1a5e3fb1 WN |
7183 | |
7184 | if (elf_version(EV_CURRENT) == EV_NONE) { | |
be18010e KW |
7185 | pr_warn("failed to init libelf for %s\n", |
7186 | path ? : "(mem buf)"); | |
6371ca3b | 7187 | return ERR_PTR(-LIBBPF_ERRNO__LIBELF); |
1a5e3fb1 WN |
7188 | } |
7189 | ||
291ee02b AN |
7190 | if (!OPTS_VALID(opts, bpf_object_open_opts)) |
7191 | return ERR_PTR(-EINVAL); | |
7192 | ||
1aace10f | 7193 | obj_name = OPTS_GET(opts, object_name, NULL); |
291ee02b AN |
7194 | if (obj_buf) { |
7195 | if (!obj_name) { | |
7196 | snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx", | |
7197 | (unsigned long)obj_buf, | |
7198 | (unsigned long)obj_buf_sz); | |
7199 | obj_name = tmp_name; | |
7200 | } | |
7201 | path = obj_name; | |
7202 | pr_debug("loading object '%s' from buffer\n", obj_name); | |
7203 | } | |
7204 | ||
2ce8450e | 7205 | obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name); |
6371ca3b WN |
7206 | if (IS_ERR(obj)) |
7207 | return obj; | |
1a5e3fb1 | 7208 | |
8601fd42 AN |
7209 | kconfig = OPTS_GET(opts, kconfig, NULL); |
7210 | if (kconfig) { | |
7211 | obj->kconfig = strdup(kconfig); | |
7212 | if (!obj->kconfig) | |
166750bc AN |
7213 | return ERR_PTR(-ENOMEM); |
7214 | } | |
291ee02b | 7215 | |
0d13bfce AN |
7216 | err = bpf_object__elf_init(obj); |
7217 | err = err ? : bpf_object__check_endianness(obj); | |
7218 | err = err ? : bpf_object__elf_collect(obj); | |
166750bc AN |
7219 | err = err ? : bpf_object__collect_externs(obj); |
7220 | err = err ? : bpf_object__finalize_btf(obj); | |
0d13bfce | 7221 | err = err ? : bpf_object__init_maps(obj, opts); |
c3c55696 | 7222 | err = err ? : bpf_object__collect_relos(obj); |
0d13bfce AN |
7223 | if (err) |
7224 | goto out; | |
1a5e3fb1 | 7225 | bpf_object__elf_finish(obj); |
dd4436bb AN |
7226 | |
7227 | bpf_object__for_each_program(prog, obj) { | |
52109584 | 7228 | prog->sec_def = find_sec_def(prog->sec_name); |
80b2b5c3 | 7229 | if (!prog->sec_def) { |
dd4436bb | 7230 | /* couldn't guess, but user might manually specify */ |
80b2b5c3 AM |
7231 | pr_debug("prog '%s': unrecognized ELF section name '%s'\n", |
7232 | prog->name, prog->sec_name); | |
dd4436bb | 7233 | continue; |
80b2b5c3 | 7234 | } |
dd4436bb | 7235 | |
2b288740 AS |
7236 | if (prog->sec_def->is_sleepable) |
7237 | prog->prog_flags |= BPF_F_SLEEPABLE; | |
25498a19 AN |
7238 | bpf_program__set_type(prog, prog->sec_def->prog_type); |
7239 | bpf_program__set_expected_attach_type(prog, | |
7240 | prog->sec_def->expected_attach_type); | |
7241 | ||
7242 | if (prog->sec_def->prog_type == BPF_PROG_TYPE_TRACING || | |
7243 | prog->sec_def->prog_type == BPF_PROG_TYPE_EXT) | |
166750bc | 7244 | prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0); |
dd4436bb AN |
7245 | } |
7246 | ||
1a5e3fb1 WN |
7247 | return obj; |
7248 | out: | |
7249 | bpf_object__close(obj); | |
6371ca3b | 7250 | return ERR_PTR(err); |
1a5e3fb1 WN |
7251 | } |
7252 | ||
5e61f270 AN |
7253 | static struct bpf_object * |
7254 | __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags) | |
1a5e3fb1 | 7255 | { |
e00aca65 | 7256 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts, |
291ee02b AN |
7257 | .relaxed_maps = flags & MAPS_RELAX_COMPAT, |
7258 | ); | |
7259 | ||
1a5e3fb1 | 7260 | /* param validation */ |
07f2d4ea | 7261 | if (!attr->file) |
1a5e3fb1 WN |
7262 | return NULL; |
7263 | ||
07f2d4ea | 7264 | pr_debug("loading %s\n", attr->file); |
291ee02b | 7265 | return __bpf_object__open(attr->file, NULL, 0, &opts); |
c034a177 JF |
7266 | } |
7267 | ||
7268 | struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr) | |
7269 | { | |
7270 | return __bpf_object__open_xattr(attr, 0); | |
07f2d4ea JK |
7271 | } |
7272 | ||
7273 | struct bpf_object *bpf_object__open(const char *path) | |
7274 | { | |
7275 | struct bpf_object_open_attr attr = { | |
7276 | .file = path, | |
7277 | .prog_type = BPF_PROG_TYPE_UNSPEC, | |
7278 | }; | |
1a5e3fb1 | 7279 | |
07f2d4ea | 7280 | return bpf_object__open_xattr(&attr); |
6c956392 WN |
7281 | } |
7282 | ||
2ce8450e | 7283 | struct bpf_object * |
01af3bf0 | 7284 | bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts) |
2ce8450e | 7285 | { |
2ce8450e AN |
7286 | if (!path) |
7287 | return ERR_PTR(-EINVAL); | |
7288 | ||
7289 | pr_debug("loading %s\n", path); | |
7290 | ||
291ee02b | 7291 | return __bpf_object__open(path, NULL, 0, opts); |
2ce8450e AN |
7292 | } |
7293 | ||
7294 | struct bpf_object * | |
7295 | bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz, | |
01af3bf0 | 7296 | const struct bpf_object_open_opts *opts) |
6c956392 | 7297 | { |
2ce8450e AN |
7298 | if (!obj_buf || obj_buf_sz == 0) |
7299 | return ERR_PTR(-EINVAL); | |
6c956392 | 7300 | |
291ee02b | 7301 | return __bpf_object__open(NULL, obj_buf, obj_buf_sz, opts); |
2ce8450e AN |
7302 | } |
7303 | ||
7304 | struct bpf_object * | |
7305 | bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz, | |
7306 | const char *name) | |
7307 | { | |
e00aca65 | 7308 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts, |
2ce8450e AN |
7309 | .object_name = name, |
7310 | /* wrong default, but backwards-compatible */ | |
7311 | .relaxed_maps = true, | |
7312 | ); | |
7313 | ||
7314 | /* returning NULL is wrong, but backwards-compatible */ | |
7315 | if (!obj_buf || obj_buf_sz == 0) | |
7316 | return NULL; | |
6c956392 | 7317 | |
2ce8450e | 7318 | return bpf_object__open_mem(obj_buf, obj_buf_sz, &opts); |
1a5e3fb1 WN |
7319 | } |
7320 | ||
52d3352e WN |
7321 | int bpf_object__unload(struct bpf_object *obj) |
7322 | { | |
7323 | size_t i; | |
7324 | ||
7325 | if (!obj) | |
7326 | return -EINVAL; | |
7327 | ||
590a0088 | 7328 | for (i = 0; i < obj->nr_maps; i++) { |
9d759a9b | 7329 | zclose(obj->maps[i].fd); |
590a0088 MKL |
7330 | if (obj->maps[i].st_ops) |
7331 | zfree(&obj->maps[i].st_ops->kern_vdata); | |
7332 | } | |
52d3352e | 7333 | |
55cffde2 WN |
7334 | for (i = 0; i < obj->nr_programs; i++) |
7335 | bpf_program__unload(&obj->programs[i]); | |
7336 | ||
52d3352e WN |
7337 | return 0; |
7338 | } | |
7339 | ||
0d13bfce AN |
7340 | static int bpf_object__sanitize_maps(struct bpf_object *obj) |
7341 | { | |
7342 | struct bpf_map *m; | |
7343 | ||
7344 | bpf_object__for_each_map(m, obj) { | |
7345 | if (!bpf_map__is_internal(m)) | |
7346 | continue; | |
47b6cb4d | 7347 | if (!kernel_supports(FEAT_GLOBAL_DATA)) { |
0d13bfce AN |
7348 | pr_warn("kernel doesn't support global data\n"); |
7349 | return -ENOTSUP; | |
7350 | } | |
47b6cb4d | 7351 | if (!kernel_supports(FEAT_ARRAY_MMAP)) |
0d13bfce AN |
7352 | m->def.map_flags ^= BPF_F_MMAPABLE; |
7353 | } | |
7354 | ||
7355 | return 0; | |
7356 | } | |
7357 | ||
1c0c7074 AN |
7358 | static int bpf_object__read_kallsyms_file(struct bpf_object *obj) |
7359 | { | |
7360 | char sym_type, sym_name[500]; | |
7361 | unsigned long long sym_addr; | |
7362 | struct extern_desc *ext; | |
7363 | int ret, err = 0; | |
7364 | FILE *f; | |
7365 | ||
7366 | f = fopen("/proc/kallsyms", "r"); | |
7367 | if (!f) { | |
7368 | err = -errno; | |
7369 | pr_warn("failed to open /proc/kallsyms: %d\n", err); | |
7370 | return err; | |
7371 | } | |
7372 | ||
7373 | while (true) { | |
7374 | ret = fscanf(f, "%llx %c %499s%*[^\n]\n", | |
7375 | &sym_addr, &sym_type, sym_name); | |
7376 | if (ret == EOF && feof(f)) | |
7377 | break; | |
7378 | if (ret != 3) { | |
135c783f | 7379 | pr_warn("failed to read kallsyms entry: %d\n", ret); |
1c0c7074 AN |
7380 | err = -EINVAL; |
7381 | goto out; | |
7382 | } | |
7383 | ||
7384 | ext = find_extern_by_name(obj, sym_name); | |
7385 | if (!ext || ext->type != EXT_KSYM) | |
7386 | continue; | |
7387 | ||
7388 | if (ext->is_set && ext->ksym.addr != sym_addr) { | |
7389 | pr_warn("extern (ksym) '%s' resolution is ambiguous: 0x%llx or 0x%llx\n", | |
7390 | sym_name, ext->ksym.addr, sym_addr); | |
7391 | err = -EINVAL; | |
7392 | goto out; | |
7393 | } | |
7394 | if (!ext->is_set) { | |
7395 | ext->is_set = true; | |
7396 | ext->ksym.addr = sym_addr; | |
7397 | pr_debug("extern (ksym) %s=0x%llx\n", sym_name, sym_addr); | |
7398 | } | |
7399 | } | |
7400 | ||
7401 | out: | |
7402 | fclose(f); | |
7403 | return err; | |
7404 | } | |
7405 | ||
d370bbe1 HL |
7406 | static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj) |
7407 | { | |
7408 | struct extern_desc *ext; | |
284d2587 AN |
7409 | struct btf *btf; |
7410 | int i, j, id, btf_fd, err; | |
d370bbe1 HL |
7411 | |
7412 | for (i = 0; i < obj->nr_extern; i++) { | |
7413 | const struct btf_type *targ_var, *targ_type; | |
7414 | __u32 targ_type_id, local_type_id; | |
7415 | const char *targ_var_name; | |
7416 | int ret; | |
7417 | ||
7418 | ext = &obj->externs[i]; | |
7419 | if (ext->type != EXT_KSYM || !ext->ksym.type_id) | |
7420 | continue; | |
7421 | ||
284d2587 AN |
7422 | btf = obj->btf_vmlinux; |
7423 | btf_fd = 0; | |
7424 | id = btf__find_by_name_kind(btf, ext->name, BTF_KIND_VAR); | |
7425 | if (id == -ENOENT) { | |
7426 | err = load_module_btfs(obj); | |
7427 | if (err) | |
7428 | return err; | |
7429 | ||
7430 | for (j = 0; j < obj->btf_module_cnt; j++) { | |
7431 | btf = obj->btf_modules[j].btf; | |
7432 | /* we assume module BTF FD is always >0 */ | |
7433 | btf_fd = obj->btf_modules[j].fd; | |
7434 | id = btf__find_by_name_kind(btf, ext->name, BTF_KIND_VAR); | |
7435 | if (id != -ENOENT) | |
7436 | break; | |
7437 | } | |
7438 | } | |
d370bbe1 | 7439 | if (id <= 0) { |
284d2587 | 7440 | pr_warn("extern (ksym) '%s': failed to find BTF ID in kernel BTF(s).\n", |
d370bbe1 HL |
7441 | ext->name); |
7442 | return -ESRCH; | |
7443 | } | |
7444 | ||
7445 | /* find local type_id */ | |
7446 | local_type_id = ext->ksym.type_id; | |
7447 | ||
7448 | /* find target type_id */ | |
284d2587 AN |
7449 | targ_var = btf__type_by_id(btf, id); |
7450 | targ_var_name = btf__name_by_offset(btf, targ_var->name_off); | |
7451 | targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id); | |
d370bbe1 HL |
7452 | |
7453 | ret = bpf_core_types_are_compat(obj->btf, local_type_id, | |
284d2587 | 7454 | btf, targ_type_id); |
d370bbe1 HL |
7455 | if (ret <= 0) { |
7456 | const struct btf_type *local_type; | |
7457 | const char *targ_name, *local_name; | |
7458 | ||
7459 | local_type = btf__type_by_id(obj->btf, local_type_id); | |
284d2587 AN |
7460 | local_name = btf__name_by_offset(obj->btf, local_type->name_off); |
7461 | targ_name = btf__name_by_offset(btf, targ_type->name_off); | |
d370bbe1 HL |
7462 | |
7463 | pr_warn("extern (ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n", | |
7464 | ext->name, local_type_id, | |
7465 | btf_kind_str(local_type), local_name, targ_type_id, | |
7466 | btf_kind_str(targ_type), targ_name); | |
7467 | return -EINVAL; | |
7468 | } | |
7469 | ||
7470 | ext->is_set = true; | |
284d2587 AN |
7471 | ext->ksym.kernel_btf_obj_fd = btf_fd; |
7472 | ext->ksym.kernel_btf_id = id; | |
d370bbe1 HL |
7473 | pr_debug("extern (ksym) '%s': resolved to [%d] %s %s\n", |
7474 | ext->name, id, btf_kind_str(targ_var), targ_var_name); | |
7475 | } | |
7476 | return 0; | |
7477 | } | |
7478 | ||
166750bc | 7479 | static int bpf_object__resolve_externs(struct bpf_object *obj, |
8601fd42 | 7480 | const char *extra_kconfig) |
166750bc | 7481 | { |
1c0c7074 | 7482 | bool need_config = false, need_kallsyms = false; |
d370bbe1 | 7483 | bool need_vmlinux_btf = false; |
166750bc | 7484 | struct extern_desc *ext; |
2e33efe3 | 7485 | void *kcfg_data = NULL; |
166750bc | 7486 | int err, i; |
166750bc AN |
7487 | |
7488 | if (obj->nr_extern == 0) | |
7489 | return 0; | |
7490 | ||
2e33efe3 AN |
7491 | if (obj->kconfig_map_idx >= 0) |
7492 | kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped; | |
166750bc AN |
7493 | |
7494 | for (i = 0; i < obj->nr_extern; i++) { | |
7495 | ext = &obj->externs[i]; | |
7496 | ||
2e33efe3 AN |
7497 | if (ext->type == EXT_KCFG && |
7498 | strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) { | |
7499 | void *ext_val = kcfg_data + ext->kcfg.data_off; | |
166750bc AN |
7500 | __u32 kver = get_kernel_version(); |
7501 | ||
7502 | if (!kver) { | |
7503 | pr_warn("failed to get kernel version\n"); | |
7504 | return -EINVAL; | |
7505 | } | |
2e33efe3 | 7506 | err = set_kcfg_value_num(ext, ext_val, kver); |
166750bc AN |
7507 | if (err) |
7508 | return err; | |
2e33efe3 AN |
7509 | pr_debug("extern (kcfg) %s=0x%x\n", ext->name, kver); |
7510 | } else if (ext->type == EXT_KCFG && | |
7511 | strncmp(ext->name, "CONFIG_", 7) == 0) { | |
166750bc | 7512 | need_config = true; |
1c0c7074 | 7513 | } else if (ext->type == EXT_KSYM) { |
d370bbe1 HL |
7514 | if (ext->ksym.type_id) |
7515 | need_vmlinux_btf = true; | |
7516 | else | |
7517 | need_kallsyms = true; | |
166750bc AN |
7518 | } else { |
7519 | pr_warn("unrecognized extern '%s'\n", ext->name); | |
7520 | return -EINVAL; | |
7521 | } | |
7522 | } | |
8601fd42 | 7523 | if (need_config && extra_kconfig) { |
2e33efe3 | 7524 | err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data); |
8601fd42 AN |
7525 | if (err) |
7526 | return -EINVAL; | |
7527 | need_config = false; | |
7528 | for (i = 0; i < obj->nr_extern; i++) { | |
7529 | ext = &obj->externs[i]; | |
2e33efe3 | 7530 | if (ext->type == EXT_KCFG && !ext->is_set) { |
8601fd42 AN |
7531 | need_config = true; |
7532 | break; | |
7533 | } | |
7534 | } | |
7535 | } | |
166750bc | 7536 | if (need_config) { |
2e33efe3 | 7537 | err = bpf_object__read_kconfig_file(obj, kcfg_data); |
166750bc AN |
7538 | if (err) |
7539 | return -EINVAL; | |
7540 | } | |
1c0c7074 AN |
7541 | if (need_kallsyms) { |
7542 | err = bpf_object__read_kallsyms_file(obj); | |
7543 | if (err) | |
7544 | return -EINVAL; | |
7545 | } | |
d370bbe1 HL |
7546 | if (need_vmlinux_btf) { |
7547 | err = bpf_object__resolve_ksyms_btf_id(obj); | |
7548 | if (err) | |
7549 | return -EINVAL; | |
7550 | } | |
166750bc AN |
7551 | for (i = 0; i < obj->nr_extern; i++) { |
7552 | ext = &obj->externs[i]; | |
7553 | ||
7554 | if (!ext->is_set && !ext->is_weak) { | |
7555 | pr_warn("extern %s (strong) not resolved\n", ext->name); | |
7556 | return -ESRCH; | |
7557 | } else if (!ext->is_set) { | |
7558 | pr_debug("extern %s (weak) not resolved, defaulting to zero\n", | |
7559 | ext->name); | |
7560 | } | |
7561 | } | |
7562 | ||
7563 | return 0; | |
7564 | } | |
7565 | ||
60276f98 | 7566 | int bpf_object__load_xattr(struct bpf_object_load_attr *attr) |
52d3352e | 7567 | { |
60276f98 | 7568 | struct bpf_object *obj; |
ec6d5f47 | 7569 | int err, i; |
6371ca3b | 7570 | |
60276f98 QM |
7571 | if (!attr) |
7572 | return -EINVAL; | |
7573 | obj = attr->obj; | |
52d3352e WN |
7574 | if (!obj) |
7575 | return -EINVAL; | |
7576 | ||
7577 | if (obj->loaded) { | |
d9297581 | 7578 | pr_warn("object '%s': load can't be attempted twice\n", obj->name); |
52d3352e WN |
7579 | return -EINVAL; |
7580 | } | |
7581 | ||
fd9eef1a | 7582 | err = bpf_object__probe_loading(obj); |
fe62de31 | 7583 | err = err ? : bpf_object__load_vmlinux_btf(obj, false); |
8601fd42 | 7584 | err = err ? : bpf_object__resolve_externs(obj, obj->kconfig); |
0d13bfce AN |
7585 | err = err ? : bpf_object__sanitize_and_load_btf(obj); |
7586 | err = err ? : bpf_object__sanitize_maps(obj); | |
590a0088 | 7587 | err = err ? : bpf_object__init_kern_struct_ops_maps(obj); |
0d13bfce AN |
7588 | err = err ? : bpf_object__create_maps(obj); |
7589 | err = err ? : bpf_object__relocate(obj, attr->target_btf_path); | |
7590 | err = err ? : bpf_object__load_progs(obj, attr->log_level); | |
a6ed02ca | 7591 | |
4f33a53d AN |
7592 | /* clean up module BTFs */ |
7593 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
91abb4a6 | 7594 | close(obj->btf_modules[i].fd); |
4f33a53d AN |
7595 | btf__free(obj->btf_modules[i].btf); |
7596 | free(obj->btf_modules[i].name); | |
7597 | } | |
7598 | free(obj->btf_modules); | |
7599 | ||
7600 | /* clean up vmlinux BTF */ | |
a6ed02ca KS |
7601 | btf__free(obj->btf_vmlinux); |
7602 | obj->btf_vmlinux = NULL; | |
7603 | ||
d9297581 AN |
7604 | obj->loaded = true; /* doesn't matter if successfully or not */ |
7605 | ||
0d13bfce AN |
7606 | if (err) |
7607 | goto out; | |
52d3352e WN |
7608 | |
7609 | return 0; | |
7610 | out: | |
ec6d5f47 THJ |
7611 | /* unpin any maps that were auto-pinned during load */ |
7612 | for (i = 0; i < obj->nr_maps; i++) | |
7613 | if (obj->maps[i].pinned && !obj->maps[i].reused) | |
7614 | bpf_map__unpin(&obj->maps[i], NULL); | |
7615 | ||
52d3352e | 7616 | bpf_object__unload(obj); |
be18010e | 7617 | pr_warn("failed to load object '%s'\n", obj->path); |
6371ca3b | 7618 | return err; |
52d3352e WN |
7619 | } |
7620 | ||
60276f98 QM |
7621 | int bpf_object__load(struct bpf_object *obj) |
7622 | { | |
7623 | struct bpf_object_load_attr attr = { | |
7624 | .obj = obj, | |
7625 | }; | |
7626 | ||
7627 | return bpf_object__load_xattr(&attr); | |
7628 | } | |
7629 | ||
196f8487 THJ |
7630 | static int make_parent_dir(const char *path) |
7631 | { | |
7632 | char *cp, errmsg[STRERR_BUFSIZE]; | |
7633 | char *dname, *dir; | |
7634 | int err = 0; | |
7635 | ||
7636 | dname = strdup(path); | |
7637 | if (dname == NULL) | |
7638 | return -ENOMEM; | |
7639 | ||
7640 | dir = dirname(dname); | |
7641 | if (mkdir(dir, 0700) && errno != EEXIST) | |
7642 | err = -errno; | |
7643 | ||
7644 | free(dname); | |
7645 | if (err) { | |
7646 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
7647 | pr_warn("failed to mkdir %s: %s\n", path, cp); | |
7648 | } | |
7649 | return err; | |
7650 | } | |
7651 | ||
f367540c JS |
7652 | static int check_path(const char *path) |
7653 | { | |
1ce6a9fc | 7654 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
7655 | struct statfs st_fs; |
7656 | char *dname, *dir; | |
7657 | int err = 0; | |
7658 | ||
7659 | if (path == NULL) | |
7660 | return -EINVAL; | |
7661 | ||
7662 | dname = strdup(path); | |
7663 | if (dname == NULL) | |
7664 | return -ENOMEM; | |
7665 | ||
7666 | dir = dirname(dname); | |
7667 | if (statfs(dir, &st_fs)) { | |
24d6a808 | 7668 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
be18010e | 7669 | pr_warn("failed to statfs %s: %s\n", dir, cp); |
f367540c JS |
7670 | err = -errno; |
7671 | } | |
7672 | free(dname); | |
7673 | ||
7674 | if (!err && st_fs.f_type != BPF_FS_MAGIC) { | |
be18010e | 7675 | pr_warn("specified path %s is not on BPF FS\n", path); |
f367540c JS |
7676 | err = -EINVAL; |
7677 | } | |
7678 | ||
7679 | return err; | |
7680 | } | |
7681 | ||
7682 | int bpf_program__pin_instance(struct bpf_program *prog, const char *path, | |
7683 | int instance) | |
7684 | { | |
1ce6a9fc | 7685 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
7686 | int err; |
7687 | ||
196f8487 THJ |
7688 | err = make_parent_dir(path); |
7689 | if (err) | |
7690 | return err; | |
7691 | ||
f367540c JS |
7692 | err = check_path(path); |
7693 | if (err) | |
7694 | return err; | |
7695 | ||
7696 | if (prog == NULL) { | |
be18010e | 7697 | pr_warn("invalid program pointer\n"); |
f367540c JS |
7698 | return -EINVAL; |
7699 | } | |
7700 | ||
7701 | if (instance < 0 || instance >= prog->instances.nr) { | |
be18010e | 7702 | pr_warn("invalid prog instance %d of prog %s (max %d)\n", |
52109584 | 7703 | instance, prog->name, prog->instances.nr); |
f367540c JS |
7704 | return -EINVAL; |
7705 | } | |
7706 | ||
7707 | if (bpf_obj_pin(prog->instances.fds[instance], path)) { | |
23ab656b THJ |
7708 | err = -errno; |
7709 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e | 7710 | pr_warn("failed to pin program: %s\n", cp); |
23ab656b | 7711 | return err; |
f367540c JS |
7712 | } |
7713 | pr_debug("pinned program '%s'\n", path); | |
7714 | ||
7715 | return 0; | |
7716 | } | |
7717 | ||
0c19a9fb SF |
7718 | int bpf_program__unpin_instance(struct bpf_program *prog, const char *path, |
7719 | int instance) | |
7720 | { | |
7721 | int err; | |
7722 | ||
7723 | err = check_path(path); | |
7724 | if (err) | |
7725 | return err; | |
7726 | ||
7727 | if (prog == NULL) { | |
be18010e | 7728 | pr_warn("invalid program pointer\n"); |
0c19a9fb SF |
7729 | return -EINVAL; |
7730 | } | |
7731 | ||
7732 | if (instance < 0 || instance >= prog->instances.nr) { | |
be18010e | 7733 | pr_warn("invalid prog instance %d of prog %s (max %d)\n", |
52109584 | 7734 | instance, prog->name, prog->instances.nr); |
0c19a9fb SF |
7735 | return -EINVAL; |
7736 | } | |
7737 | ||
7738 | err = unlink(path); | |
7739 | if (err != 0) | |
7740 | return -errno; | |
7741 | pr_debug("unpinned program '%s'\n", path); | |
7742 | ||
7743 | return 0; | |
7744 | } | |
7745 | ||
f367540c JS |
7746 | int bpf_program__pin(struct bpf_program *prog, const char *path) |
7747 | { | |
7748 | int i, err; | |
7749 | ||
196f8487 THJ |
7750 | err = make_parent_dir(path); |
7751 | if (err) | |
7752 | return err; | |
7753 | ||
f367540c JS |
7754 | err = check_path(path); |
7755 | if (err) | |
7756 | return err; | |
7757 | ||
7758 | if (prog == NULL) { | |
be18010e | 7759 | pr_warn("invalid program pointer\n"); |
f367540c JS |
7760 | return -EINVAL; |
7761 | } | |
7762 | ||
7763 | if (prog->instances.nr <= 0) { | |
52109584 | 7764 | pr_warn("no instances of prog %s to pin\n", prog->name); |
f367540c JS |
7765 | return -EINVAL; |
7766 | } | |
7767 | ||
fd734c5c SF |
7768 | if (prog->instances.nr == 1) { |
7769 | /* don't create subdirs when pinning single instance */ | |
7770 | return bpf_program__pin_instance(prog, path, 0); | |
7771 | } | |
7772 | ||
0c19a9fb SF |
7773 | for (i = 0; i < prog->instances.nr; i++) { |
7774 | char buf[PATH_MAX]; | |
7775 | int len; | |
7776 | ||
7777 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7778 | if (len < 0) { | |
7779 | err = -EINVAL; | |
7780 | goto err_unpin; | |
7781 | } else if (len >= PATH_MAX) { | |
7782 | err = -ENAMETOOLONG; | |
7783 | goto err_unpin; | |
7784 | } | |
7785 | ||
7786 | err = bpf_program__pin_instance(prog, buf, i); | |
7787 | if (err) | |
7788 | goto err_unpin; | |
7789 | } | |
7790 | ||
7791 | return 0; | |
7792 | ||
7793 | err_unpin: | |
7794 | for (i = i - 1; i >= 0; i--) { | |
7795 | char buf[PATH_MAX]; | |
7796 | int len; | |
7797 | ||
7798 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7799 | if (len < 0) | |
7800 | continue; | |
7801 | else if (len >= PATH_MAX) | |
7802 | continue; | |
7803 | ||
7804 | bpf_program__unpin_instance(prog, buf, i); | |
7805 | } | |
7806 | ||
7807 | rmdir(path); | |
7808 | ||
7809 | return err; | |
7810 | } | |
7811 | ||
7812 | int bpf_program__unpin(struct bpf_program *prog, const char *path) | |
7813 | { | |
7814 | int i, err; | |
7815 | ||
7816 | err = check_path(path); | |
7817 | if (err) | |
7818 | return err; | |
7819 | ||
7820 | if (prog == NULL) { | |
be18010e | 7821 | pr_warn("invalid program pointer\n"); |
0c19a9fb SF |
7822 | return -EINVAL; |
7823 | } | |
7824 | ||
7825 | if (prog->instances.nr <= 0) { | |
52109584 | 7826 | pr_warn("no instances of prog %s to pin\n", prog->name); |
0c19a9fb | 7827 | return -EINVAL; |
fd734c5c SF |
7828 | } |
7829 | ||
7830 | if (prog->instances.nr == 1) { | |
7831 | /* don't create subdirs when pinning single instance */ | |
7832 | return bpf_program__unpin_instance(prog, path, 0); | |
0c19a9fb SF |
7833 | } |
7834 | ||
f367540c JS |
7835 | for (i = 0; i < prog->instances.nr; i++) { |
7836 | char buf[PATH_MAX]; | |
7837 | int len; | |
7838 | ||
7839 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7840 | if (len < 0) | |
7841 | return -EINVAL; | |
7842 | else if (len >= PATH_MAX) | |
7843 | return -ENAMETOOLONG; | |
7844 | ||
0c19a9fb | 7845 | err = bpf_program__unpin_instance(prog, buf, i); |
f367540c JS |
7846 | if (err) |
7847 | return err; | |
7848 | } | |
7849 | ||
0c19a9fb SF |
7850 | err = rmdir(path); |
7851 | if (err) | |
7852 | return -errno; | |
7853 | ||
f367540c JS |
7854 | return 0; |
7855 | } | |
7856 | ||
b6989f35 JS |
7857 | int bpf_map__pin(struct bpf_map *map, const char *path) |
7858 | { | |
1ce6a9fc | 7859 | char *cp, errmsg[STRERR_BUFSIZE]; |
b6989f35 JS |
7860 | int err; |
7861 | ||
b6989f35 | 7862 | if (map == NULL) { |
be18010e | 7863 | pr_warn("invalid map pointer\n"); |
b6989f35 JS |
7864 | return -EINVAL; |
7865 | } | |
7866 | ||
4580b25f THJ |
7867 | if (map->pin_path) { |
7868 | if (path && strcmp(path, map->pin_path)) { | |
7869 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
7870 | bpf_map__name(map), map->pin_path, path); | |
7871 | return -EINVAL; | |
7872 | } else if (map->pinned) { | |
7873 | pr_debug("map '%s' already pinned at '%s'; not re-pinning\n", | |
7874 | bpf_map__name(map), map->pin_path); | |
7875 | return 0; | |
7876 | } | |
7877 | } else { | |
7878 | if (!path) { | |
7879 | pr_warn("missing a path to pin map '%s' at\n", | |
7880 | bpf_map__name(map)); | |
7881 | return -EINVAL; | |
7882 | } else if (map->pinned) { | |
7883 | pr_warn("map '%s' already pinned\n", bpf_map__name(map)); | |
7884 | return -EEXIST; | |
7885 | } | |
7886 | ||
7887 | map->pin_path = strdup(path); | |
7888 | if (!map->pin_path) { | |
7889 | err = -errno; | |
7890 | goto out_err; | |
7891 | } | |
b6989f35 JS |
7892 | } |
7893 | ||
196f8487 THJ |
7894 | err = make_parent_dir(map->pin_path); |
7895 | if (err) | |
7896 | return err; | |
7897 | ||
4580b25f THJ |
7898 | err = check_path(map->pin_path); |
7899 | if (err) | |
7900 | return err; | |
7901 | ||
7902 | if (bpf_obj_pin(map->fd, map->pin_path)) { | |
7903 | err = -errno; | |
7904 | goto out_err; | |
7905 | } | |
7906 | ||
7907 | map->pinned = true; | |
7908 | pr_debug("pinned map '%s'\n", map->pin_path); | |
0c19a9fb | 7909 | |
b6989f35 | 7910 | return 0; |
4580b25f THJ |
7911 | |
7912 | out_err: | |
7913 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
7914 | pr_warn("failed to pin map: %s\n", cp); | |
7915 | return err; | |
b6989f35 JS |
7916 | } |
7917 | ||
0c19a9fb SF |
7918 | int bpf_map__unpin(struct bpf_map *map, const char *path) |
7919 | { | |
7920 | int err; | |
7921 | ||
0c19a9fb | 7922 | if (map == NULL) { |
be18010e | 7923 | pr_warn("invalid map pointer\n"); |
0c19a9fb SF |
7924 | return -EINVAL; |
7925 | } | |
7926 | ||
4580b25f THJ |
7927 | if (map->pin_path) { |
7928 | if (path && strcmp(path, map->pin_path)) { | |
7929 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
7930 | bpf_map__name(map), map->pin_path, path); | |
7931 | return -EINVAL; | |
7932 | } | |
7933 | path = map->pin_path; | |
7934 | } else if (!path) { | |
7935 | pr_warn("no path to unpin map '%s' from\n", | |
7936 | bpf_map__name(map)); | |
7937 | return -EINVAL; | |
7938 | } | |
7939 | ||
7940 | err = check_path(path); | |
7941 | if (err) | |
7942 | return err; | |
7943 | ||
0c19a9fb SF |
7944 | err = unlink(path); |
7945 | if (err != 0) | |
7946 | return -errno; | |
4580b25f THJ |
7947 | |
7948 | map->pinned = false; | |
7949 | pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path); | |
0c19a9fb SF |
7950 | |
7951 | return 0; | |
7952 | } | |
7953 | ||
4580b25f THJ |
7954 | int bpf_map__set_pin_path(struct bpf_map *map, const char *path) |
7955 | { | |
7956 | char *new = NULL; | |
7957 | ||
7958 | if (path) { | |
7959 | new = strdup(path); | |
7960 | if (!new) | |
7961 | return -errno; | |
7962 | } | |
7963 | ||
7964 | free(map->pin_path); | |
7965 | map->pin_path = new; | |
7966 | return 0; | |
7967 | } | |
7968 | ||
7969 | const char *bpf_map__get_pin_path(const struct bpf_map *map) | |
7970 | { | |
7971 | return map->pin_path; | |
7972 | } | |
7973 | ||
7974 | bool bpf_map__is_pinned(const struct bpf_map *map) | |
7975 | { | |
7976 | return map->pinned; | |
7977 | } | |
7978 | ||
9cf309c5 THJ |
7979 | static void sanitize_pin_path(char *s) |
7980 | { | |
7981 | /* bpffs disallows periods in path names */ | |
7982 | while (*s) { | |
7983 | if (*s == '.') | |
7984 | *s = '_'; | |
7985 | s++; | |
7986 | } | |
7987 | } | |
7988 | ||
0c19a9fb | 7989 | int bpf_object__pin_maps(struct bpf_object *obj, const char *path) |
d5148d85 | 7990 | { |
d5148d85 JS |
7991 | struct bpf_map *map; |
7992 | int err; | |
7993 | ||
7994 | if (!obj) | |
7995 | return -ENOENT; | |
7996 | ||
7997 | if (!obj->loaded) { | |
be18010e | 7998 | pr_warn("object not yet loaded; load it first\n"); |
d5148d85 JS |
7999 | return -ENOENT; |
8000 | } | |
8001 | ||
f74a53d9 | 8002 | bpf_object__for_each_map(map, obj) { |
4580b25f | 8003 | char *pin_path = NULL; |
0c19a9fb | 8004 | char buf[PATH_MAX]; |
0c19a9fb | 8005 | |
4580b25f THJ |
8006 | if (path) { |
8007 | int len; | |
8008 | ||
8009 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
8010 | bpf_map__name(map)); | |
8011 | if (len < 0) { | |
8012 | err = -EINVAL; | |
8013 | goto err_unpin_maps; | |
8014 | } else if (len >= PATH_MAX) { | |
8015 | err = -ENAMETOOLONG; | |
8016 | goto err_unpin_maps; | |
8017 | } | |
9cf309c5 | 8018 | sanitize_pin_path(buf); |
4580b25f THJ |
8019 | pin_path = buf; |
8020 | } else if (!map->pin_path) { | |
8021 | continue; | |
0c19a9fb SF |
8022 | } |
8023 | ||
4580b25f | 8024 | err = bpf_map__pin(map, pin_path); |
0c19a9fb SF |
8025 | if (err) |
8026 | goto err_unpin_maps; | |
8027 | } | |
8028 | ||
8029 | return 0; | |
8030 | ||
8031 | err_unpin_maps: | |
8032 | while ((map = bpf_map__prev(map, obj))) { | |
4580b25f | 8033 | if (!map->pin_path) |
0c19a9fb SF |
8034 | continue; |
8035 | ||
4580b25f | 8036 | bpf_map__unpin(map, NULL); |
0c19a9fb SF |
8037 | } |
8038 | ||
8039 | return err; | |
8040 | } | |
8041 | ||
8042 | int bpf_object__unpin_maps(struct bpf_object *obj, const char *path) | |
8043 | { | |
8044 | struct bpf_map *map; | |
8045 | int err; | |
8046 | ||
8047 | if (!obj) | |
8048 | return -ENOENT; | |
8049 | ||
f74a53d9 | 8050 | bpf_object__for_each_map(map, obj) { |
4580b25f | 8051 | char *pin_path = NULL; |
d5148d85 | 8052 | char buf[PATH_MAX]; |
d5148d85 | 8053 | |
4580b25f THJ |
8054 | if (path) { |
8055 | int len; | |
8056 | ||
8057 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
8058 | bpf_map__name(map)); | |
8059 | if (len < 0) | |
8060 | return -EINVAL; | |
8061 | else if (len >= PATH_MAX) | |
8062 | return -ENAMETOOLONG; | |
9cf309c5 | 8063 | sanitize_pin_path(buf); |
4580b25f THJ |
8064 | pin_path = buf; |
8065 | } else if (!map->pin_path) { | |
8066 | continue; | |
8067 | } | |
d5148d85 | 8068 | |
4580b25f | 8069 | err = bpf_map__unpin(map, pin_path); |
d5148d85 JS |
8070 | if (err) |
8071 | return err; | |
8072 | } | |
8073 | ||
0c19a9fb SF |
8074 | return 0; |
8075 | } | |
8076 | ||
8077 | int bpf_object__pin_programs(struct bpf_object *obj, const char *path) | |
8078 | { | |
8079 | struct bpf_program *prog; | |
8080 | int err; | |
8081 | ||
8082 | if (!obj) | |
8083 | return -ENOENT; | |
8084 | ||
8085 | if (!obj->loaded) { | |
be18010e | 8086 | pr_warn("object not yet loaded; load it first\n"); |
0c19a9fb SF |
8087 | return -ENOENT; |
8088 | } | |
8089 | ||
0c19a9fb SF |
8090 | bpf_object__for_each_program(prog, obj) { |
8091 | char buf[PATH_MAX]; | |
8092 | int len; | |
8093 | ||
8094 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 8095 | prog->pin_name); |
0c19a9fb SF |
8096 | if (len < 0) { |
8097 | err = -EINVAL; | |
8098 | goto err_unpin_programs; | |
8099 | } else if (len >= PATH_MAX) { | |
8100 | err = -ENAMETOOLONG; | |
8101 | goto err_unpin_programs; | |
8102 | } | |
8103 | ||
8104 | err = bpf_program__pin(prog, buf); | |
8105 | if (err) | |
8106 | goto err_unpin_programs; | |
8107 | } | |
8108 | ||
8109 | return 0; | |
8110 | ||
8111 | err_unpin_programs: | |
8112 | while ((prog = bpf_program__prev(prog, obj))) { | |
8113 | char buf[PATH_MAX]; | |
8114 | int len; | |
8115 | ||
8116 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 8117 | prog->pin_name); |
0c19a9fb SF |
8118 | if (len < 0) |
8119 | continue; | |
8120 | else if (len >= PATH_MAX) | |
8121 | continue; | |
8122 | ||
8123 | bpf_program__unpin(prog, buf); | |
8124 | } | |
8125 | ||
8126 | return err; | |
8127 | } | |
8128 | ||
8129 | int bpf_object__unpin_programs(struct bpf_object *obj, const char *path) | |
8130 | { | |
8131 | struct bpf_program *prog; | |
8132 | int err; | |
8133 | ||
8134 | if (!obj) | |
8135 | return -ENOENT; | |
8136 | ||
d5148d85 JS |
8137 | bpf_object__for_each_program(prog, obj) { |
8138 | char buf[PATH_MAX]; | |
8139 | int len; | |
8140 | ||
8141 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 8142 | prog->pin_name); |
d5148d85 JS |
8143 | if (len < 0) |
8144 | return -EINVAL; | |
8145 | else if (len >= PATH_MAX) | |
8146 | return -ENAMETOOLONG; | |
8147 | ||
0c19a9fb | 8148 | err = bpf_program__unpin(prog, buf); |
d5148d85 JS |
8149 | if (err) |
8150 | return err; | |
8151 | } | |
8152 | ||
8153 | return 0; | |
8154 | } | |
8155 | ||
0c19a9fb SF |
8156 | int bpf_object__pin(struct bpf_object *obj, const char *path) |
8157 | { | |
8158 | int err; | |
8159 | ||
8160 | err = bpf_object__pin_maps(obj, path); | |
8161 | if (err) | |
8162 | return err; | |
8163 | ||
8164 | err = bpf_object__pin_programs(obj, path); | |
8165 | if (err) { | |
8166 | bpf_object__unpin_maps(obj, path); | |
8167 | return err; | |
8168 | } | |
8169 | ||
8170 | return 0; | |
8171 | } | |
8172 | ||
2d39d7c5 AN |
8173 | static void bpf_map__destroy(struct bpf_map *map) |
8174 | { | |
8175 | if (map->clear_priv) | |
8176 | map->clear_priv(map, map->priv); | |
8177 | map->priv = NULL; | |
8178 | map->clear_priv = NULL; | |
8179 | ||
646f02ff AN |
8180 | if (map->inner_map) { |
8181 | bpf_map__destroy(map->inner_map); | |
8182 | zfree(&map->inner_map); | |
8183 | } | |
8184 | ||
8185 | zfree(&map->init_slots); | |
8186 | map->init_slots_sz = 0; | |
8187 | ||
2d39d7c5 AN |
8188 | if (map->mmaped) { |
8189 | munmap(map->mmaped, bpf_map_mmap_sz(map)); | |
8190 | map->mmaped = NULL; | |
8191 | } | |
8192 | ||
8193 | if (map->st_ops) { | |
8194 | zfree(&map->st_ops->data); | |
8195 | zfree(&map->st_ops->progs); | |
8196 | zfree(&map->st_ops->kern_func_off); | |
8197 | zfree(&map->st_ops); | |
8198 | } | |
8199 | ||
8200 | zfree(&map->name); | |
8201 | zfree(&map->pin_path); | |
8202 | ||
8203 | if (map->fd >= 0) | |
8204 | zclose(map->fd); | |
8205 | } | |
8206 | ||
1a5e3fb1 WN |
8207 | void bpf_object__close(struct bpf_object *obj) |
8208 | { | |
a5b8bd47 WN |
8209 | size_t i; |
8210 | ||
50450fc7 | 8211 | if (IS_ERR_OR_NULL(obj)) |
1a5e3fb1 WN |
8212 | return; |
8213 | ||
10931d24 WN |
8214 | if (obj->clear_priv) |
8215 | obj->clear_priv(obj, obj->priv); | |
8216 | ||
1a5e3fb1 | 8217 | bpf_object__elf_finish(obj); |
52d3352e | 8218 | bpf_object__unload(obj); |
8a138aed | 8219 | btf__free(obj->btf); |
2993e051 | 8220 | btf_ext__free(obj->btf_ext); |
1a5e3fb1 | 8221 | |
2d39d7c5 AN |
8222 | for (i = 0; i < obj->nr_maps; i++) |
8223 | bpf_map__destroy(&obj->maps[i]); | |
d859900c | 8224 | |
8601fd42 | 8225 | zfree(&obj->kconfig); |
166750bc AN |
8226 | zfree(&obj->externs); |
8227 | obj->nr_extern = 0; | |
8228 | ||
9d759a9b WN |
8229 | zfree(&obj->maps); |
8230 | obj->nr_maps = 0; | |
a5b8bd47 WN |
8231 | |
8232 | if (obj->programs && obj->nr_programs) { | |
8233 | for (i = 0; i < obj->nr_programs; i++) | |
8234 | bpf_program__exit(&obj->programs[i]); | |
8235 | } | |
8236 | zfree(&obj->programs); | |
8237 | ||
9a208eff | 8238 | list_del(&obj->list); |
1a5e3fb1 WN |
8239 | free(obj); |
8240 | } | |
aa9b1ac3 | 8241 | |
9a208eff WN |
8242 | struct bpf_object * |
8243 | bpf_object__next(struct bpf_object *prev) | |
8244 | { | |
8245 | struct bpf_object *next; | |
8246 | ||
8247 | if (!prev) | |
8248 | next = list_first_entry(&bpf_objects_list, | |
8249 | struct bpf_object, | |
8250 | list); | |
8251 | else | |
8252 | next = list_next_entry(prev, list); | |
8253 | ||
8254 | /* Empty list is noticed here so don't need checking on entry. */ | |
8255 | if (&next->list == &bpf_objects_list) | |
8256 | return NULL; | |
8257 | ||
8258 | return next; | |
8259 | } | |
8260 | ||
a324aae3 | 8261 | const char *bpf_object__name(const struct bpf_object *obj) |
acf860ae | 8262 | { |
c9e4c301 | 8263 | return obj ? obj->name : ERR_PTR(-EINVAL); |
acf860ae WN |
8264 | } |
8265 | ||
a324aae3 | 8266 | unsigned int bpf_object__kversion(const struct bpf_object *obj) |
45825d8a | 8267 | { |
a7fe0450 | 8268 | return obj ? obj->kern_version : 0; |
45825d8a WN |
8269 | } |
8270 | ||
a324aae3 | 8271 | struct btf *bpf_object__btf(const struct bpf_object *obj) |
789f6bab AI |
8272 | { |
8273 | return obj ? obj->btf : NULL; | |
8274 | } | |
8275 | ||
8a138aed MKL |
8276 | int bpf_object__btf_fd(const struct bpf_object *obj) |
8277 | { | |
8278 | return obj->btf ? btf__fd(obj->btf) : -1; | |
8279 | } | |
8280 | ||
10931d24 WN |
8281 | int bpf_object__set_priv(struct bpf_object *obj, void *priv, |
8282 | bpf_object_clear_priv_t clear_priv) | |
8283 | { | |
8284 | if (obj->priv && obj->clear_priv) | |
8285 | obj->clear_priv(obj, obj->priv); | |
8286 | ||
8287 | obj->priv = priv; | |
8288 | obj->clear_priv = clear_priv; | |
8289 | return 0; | |
8290 | } | |
8291 | ||
a324aae3 | 8292 | void *bpf_object__priv(const struct bpf_object *obj) |
10931d24 WN |
8293 | { |
8294 | return obj ? obj->priv : ERR_PTR(-EINVAL); | |
8295 | } | |
8296 | ||
eac7d845 | 8297 | static struct bpf_program * |
a324aae3 AN |
8298 | __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj, |
8299 | bool forward) | |
aa9b1ac3 | 8300 | { |
a83d6e76 | 8301 | size_t nr_programs = obj->nr_programs; |
0c19a9fb | 8302 | ssize_t idx; |
aa9b1ac3 | 8303 | |
a83d6e76 | 8304 | if (!nr_programs) |
aa9b1ac3 | 8305 | return NULL; |
aa9b1ac3 | 8306 | |
a83d6e76 MKL |
8307 | if (!p) |
8308 | /* Iter from the beginning */ | |
8309 | return forward ? &obj->programs[0] : | |
8310 | &obj->programs[nr_programs - 1]; | |
8311 | ||
0c19a9fb | 8312 | if (p->obj != obj) { |
be18010e | 8313 | pr_warn("error: program handler doesn't match object\n"); |
aa9b1ac3 WN |
8314 | return NULL; |
8315 | } | |
8316 | ||
a83d6e76 | 8317 | idx = (p - obj->programs) + (forward ? 1 : -1); |
0c19a9fb | 8318 | if (idx >= obj->nr_programs || idx < 0) |
aa9b1ac3 WN |
8319 | return NULL; |
8320 | return &obj->programs[idx]; | |
8321 | } | |
8322 | ||
eac7d845 | 8323 | struct bpf_program * |
a324aae3 | 8324 | bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj) |
eac7d845 JK |
8325 | { |
8326 | struct bpf_program *prog = prev; | |
8327 | ||
8328 | do { | |
a83d6e76 | 8329 | prog = __bpf_program__iter(prog, obj, true); |
c3c55696 | 8330 | } while (prog && prog_is_subprog(obj, prog)); |
0c19a9fb SF |
8331 | |
8332 | return prog; | |
8333 | } | |
8334 | ||
8335 | struct bpf_program * | |
a324aae3 | 8336 | bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj) |
0c19a9fb SF |
8337 | { |
8338 | struct bpf_program *prog = next; | |
8339 | ||
0c19a9fb | 8340 | do { |
a83d6e76 | 8341 | prog = __bpf_program__iter(prog, obj, false); |
c3c55696 | 8342 | } while (prog && prog_is_subprog(obj, prog)); |
eac7d845 JK |
8343 | |
8344 | return prog; | |
8345 | } | |
8346 | ||
edb13ed4 ACM |
8347 | int bpf_program__set_priv(struct bpf_program *prog, void *priv, |
8348 | bpf_program_clear_priv_t clear_priv) | |
aa9b1ac3 WN |
8349 | { |
8350 | if (prog->priv && prog->clear_priv) | |
8351 | prog->clear_priv(prog, prog->priv); | |
8352 | ||
8353 | prog->priv = priv; | |
8354 | prog->clear_priv = clear_priv; | |
8355 | return 0; | |
8356 | } | |
8357 | ||
a324aae3 | 8358 | void *bpf_program__priv(const struct bpf_program *prog) |
aa9b1ac3 | 8359 | { |
be834ffb | 8360 | return prog ? prog->priv : ERR_PTR(-EINVAL); |
aa9b1ac3 WN |
8361 | } |
8362 | ||
9aba3613 JK |
8363 | void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex) |
8364 | { | |
8365 | prog->prog_ifindex = ifindex; | |
8366 | } | |
8367 | ||
01af3bf0 AN |
8368 | const char *bpf_program__name(const struct bpf_program *prog) |
8369 | { | |
8370 | return prog->name; | |
8371 | } | |
8372 | ||
52109584 AN |
8373 | const char *bpf_program__section_name(const struct bpf_program *prog) |
8374 | { | |
8375 | return prog->sec_name; | |
8376 | } | |
8377 | ||
a324aae3 | 8378 | const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy) |
aa9b1ac3 WN |
8379 | { |
8380 | const char *title; | |
8381 | ||
52109584 | 8382 | title = prog->sec_name; |
715f8db9 | 8383 | if (needs_copy) { |
aa9b1ac3 WN |
8384 | title = strdup(title); |
8385 | if (!title) { | |
be18010e | 8386 | pr_warn("failed to strdup program title\n"); |
6371ca3b | 8387 | return ERR_PTR(-ENOMEM); |
aa9b1ac3 WN |
8388 | } |
8389 | } | |
8390 | ||
8391 | return title; | |
8392 | } | |
8393 | ||
d9297581 AN |
8394 | bool bpf_program__autoload(const struct bpf_program *prog) |
8395 | { | |
8396 | return prog->load; | |
8397 | } | |
8398 | ||
8399 | int bpf_program__set_autoload(struct bpf_program *prog, bool autoload) | |
8400 | { | |
8401 | if (prog->obj->loaded) | |
8402 | return -EINVAL; | |
8403 | ||
8404 | prog->load = autoload; | |
8405 | return 0; | |
8406 | } | |
8407 | ||
a324aae3 | 8408 | int bpf_program__fd(const struct bpf_program *prog) |
aa9b1ac3 | 8409 | { |
b580563e WN |
8410 | return bpf_program__nth_fd(prog, 0); |
8411 | } | |
8412 | ||
1a734efe THJ |
8413 | size_t bpf_program__size(const struct bpf_program *prog) |
8414 | { | |
9c0f8cbd | 8415 | return prog->insns_cnt * BPF_INSN_SZ; |
1a734efe THJ |
8416 | } |
8417 | ||
b580563e WN |
8418 | int bpf_program__set_prep(struct bpf_program *prog, int nr_instances, |
8419 | bpf_program_prep_t prep) | |
8420 | { | |
8421 | int *instances_fds; | |
8422 | ||
8423 | if (nr_instances <= 0 || !prep) | |
8424 | return -EINVAL; | |
8425 | ||
8426 | if (prog->instances.nr > 0 || prog->instances.fds) { | |
be18010e | 8427 | pr_warn("Can't set pre-processor after loading\n"); |
b580563e WN |
8428 | return -EINVAL; |
8429 | } | |
8430 | ||
8431 | instances_fds = malloc(sizeof(int) * nr_instances); | |
8432 | if (!instances_fds) { | |
be18010e | 8433 | pr_warn("alloc memory failed for fds\n"); |
b580563e WN |
8434 | return -ENOMEM; |
8435 | } | |
8436 | ||
8437 | /* fill all fd with -1 */ | |
8438 | memset(instances_fds, -1, sizeof(int) * nr_instances); | |
8439 | ||
8440 | prog->instances.nr = nr_instances; | |
8441 | prog->instances.fds = instances_fds; | |
8442 | prog->preprocessor = prep; | |
8443 | return 0; | |
8444 | } | |
8445 | ||
a324aae3 | 8446 | int bpf_program__nth_fd(const struct bpf_program *prog, int n) |
b580563e WN |
8447 | { |
8448 | int fd; | |
8449 | ||
1e960043 JK |
8450 | if (!prog) |
8451 | return -EINVAL; | |
8452 | ||
b580563e | 8453 | if (n >= prog->instances.nr || n < 0) { |
be18010e | 8454 | pr_warn("Can't get the %dth fd from program %s: only %d instances\n", |
52109584 | 8455 | n, prog->name, prog->instances.nr); |
b580563e WN |
8456 | return -EINVAL; |
8457 | } | |
8458 | ||
8459 | fd = prog->instances.fds[n]; | |
8460 | if (fd < 0) { | |
be18010e | 8461 | pr_warn("%dth instance of program '%s' is invalid\n", |
52109584 | 8462 | n, prog->name); |
b580563e WN |
8463 | return -ENOENT; |
8464 | } | |
8465 | ||
8466 | return fd; | |
aa9b1ac3 | 8467 | } |
9d759a9b | 8468 | |
f1eead9e AN |
8469 | enum bpf_prog_type bpf_program__get_type(struct bpf_program *prog) |
8470 | { | |
8471 | return prog->type; | |
8472 | } | |
8473 | ||
dd26b7f5 | 8474 | void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type) |
5f44e4c8 WN |
8475 | { |
8476 | prog->type = type; | |
8477 | } | |
8478 | ||
a324aae3 | 8479 | static bool bpf_program__is_type(const struct bpf_program *prog, |
5f44e4c8 WN |
8480 | enum bpf_prog_type type) |
8481 | { | |
8482 | return prog ? (prog->type == type) : false; | |
8483 | } | |
8484 | ||
a324aae3 AN |
8485 | #define BPF_PROG_TYPE_FNS(NAME, TYPE) \ |
8486 | int bpf_program__set_##NAME(struct bpf_program *prog) \ | |
8487 | { \ | |
8488 | if (!prog) \ | |
8489 | return -EINVAL; \ | |
8490 | bpf_program__set_type(prog, TYPE); \ | |
8491 | return 0; \ | |
8492 | } \ | |
8493 | \ | |
8494 | bool bpf_program__is_##NAME(const struct bpf_program *prog) \ | |
8495 | { \ | |
8496 | return bpf_program__is_type(prog, TYPE); \ | |
8497 | } \ | |
ed794073 | 8498 | |
7803ba73 | 8499 | BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER); |
1e092a03 | 8500 | BPF_PROG_TYPE_FNS(lsm, BPF_PROG_TYPE_LSM); |
ed794073 | 8501 | BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE); |
7803ba73 JS |
8502 | BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS); |
8503 | BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT); | |
ed794073 | 8504 | BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT); |
e14c93fd | 8505 | BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT); |
7803ba73 JS |
8506 | BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP); |
8507 | BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT); | |
12a8654b | 8508 | BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING); |
590a0088 | 8509 | BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS); |
2db6eab1 | 8510 | BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT); |
499dd29d | 8511 | BPF_PROG_TYPE_FNS(sk_lookup, BPF_PROG_TYPE_SK_LOOKUP); |
5f44e4c8 | 8512 | |
f1eead9e AN |
8513 | enum bpf_attach_type |
8514 | bpf_program__get_expected_attach_type(struct bpf_program *prog) | |
8515 | { | |
8516 | return prog->expected_attach_type; | |
8517 | } | |
8518 | ||
16962b24 JF |
8519 | void bpf_program__set_expected_attach_type(struct bpf_program *prog, |
8520 | enum bpf_attach_type type) | |
d7be143b AI |
8521 | { |
8522 | prog->expected_attach_type = type; | |
8523 | } | |
8524 | ||
25498a19 AN |
8525 | #define BPF_PROG_SEC_IMPL(string, ptype, eatype, eatype_optional, \ |
8526 | attachable, attach_btf) \ | |
8527 | { \ | |
8528 | .sec = string, \ | |
8529 | .len = sizeof(string) - 1, \ | |
8530 | .prog_type = ptype, \ | |
8531 | .expected_attach_type = eatype, \ | |
8532 | .is_exp_attach_type_optional = eatype_optional, \ | |
8533 | .is_attachable = attachable, \ | |
8534 | .is_attach_btf = attach_btf, \ | |
8535 | } | |
d7be143b | 8536 | |
956b620f | 8537 | /* Programs that can NOT be attached. */ |
f75a697e | 8538 | #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0, 0) |
d7be143b | 8539 | |
956b620f AI |
8540 | /* Programs that can be attached. */ |
8541 | #define BPF_APROG_SEC(string, ptype, atype) \ | |
25498a19 | 8542 | BPF_PROG_SEC_IMPL(string, ptype, atype, true, 1, 0) |
81efee75 | 8543 | |
956b620f AI |
8544 | /* Programs that must specify expected attach type at load time. */ |
8545 | #define BPF_EAPROG_SEC(string, ptype, eatype) \ | |
25498a19 | 8546 | BPF_PROG_SEC_IMPL(string, ptype, eatype, false, 1, 0) |
f75a697e AS |
8547 | |
8548 | /* Programs that use BTF to identify attach point */ | |
12a8654b | 8549 | #define BPF_PROG_BTF(string, ptype, eatype) \ |
25498a19 | 8550 | BPF_PROG_SEC_IMPL(string, ptype, eatype, false, 0, 1) |
956b620f AI |
8551 | |
8552 | /* Programs that can be attached but attach type can't be identified by section | |
8553 | * name. Kept for backward compatibility. | |
8554 | */ | |
8555 | #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype) | |
e50b0a6f | 8556 | |
d7a18ea7 AN |
8557 | #define SEC_DEF(sec_pfx, ptype, ...) { \ |
8558 | .sec = sec_pfx, \ | |
8559 | .len = sizeof(sec_pfx) - 1, \ | |
8560 | .prog_type = BPF_PROG_TYPE_##ptype, \ | |
8561 | __VA_ARGS__ \ | |
8562 | } | |
8563 | ||
d7a18ea7 AN |
8564 | static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec, |
8565 | struct bpf_program *prog); | |
8566 | static struct bpf_link *attach_tp(const struct bpf_sec_def *sec, | |
8567 | struct bpf_program *prog); | |
8568 | static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec, | |
8569 | struct bpf_program *prog); | |
8570 | static struct bpf_link *attach_trace(const struct bpf_sec_def *sec, | |
8571 | struct bpf_program *prog); | |
1e092a03 KS |
8572 | static struct bpf_link *attach_lsm(const struct bpf_sec_def *sec, |
8573 | struct bpf_program *prog); | |
c09add2f YS |
8574 | static struct bpf_link *attach_iter(const struct bpf_sec_def *sec, |
8575 | struct bpf_program *prog); | |
d7a18ea7 | 8576 | |
d7a18ea7 | 8577 | static const struct bpf_sec_def section_defs[] = { |
956b620f | 8578 | BPF_PROG_SEC("socket", BPF_PROG_TYPE_SOCKET_FILTER), |
67d69ccd | 8579 | BPF_PROG_SEC("sk_reuseport", BPF_PROG_TYPE_SK_REUSEPORT), |
d7a18ea7 AN |
8580 | SEC_DEF("kprobe/", KPROBE, |
8581 | .attach_fn = attach_kprobe), | |
32dff6db | 8582 | BPF_PROG_SEC("uprobe/", BPF_PROG_TYPE_KPROBE), |
d7a18ea7 AN |
8583 | SEC_DEF("kretprobe/", KPROBE, |
8584 | .attach_fn = attach_kprobe), | |
32dff6db | 8585 | BPF_PROG_SEC("uretprobe/", BPF_PROG_TYPE_KPROBE), |
956b620f AI |
8586 | BPF_PROG_SEC("classifier", BPF_PROG_TYPE_SCHED_CLS), |
8587 | BPF_PROG_SEC("action", BPF_PROG_TYPE_SCHED_ACT), | |
d7a18ea7 AN |
8588 | SEC_DEF("tracepoint/", TRACEPOINT, |
8589 | .attach_fn = attach_tp), | |
8590 | SEC_DEF("tp/", TRACEPOINT, | |
8591 | .attach_fn = attach_tp), | |
8592 | SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT, | |
8593 | .attach_fn = attach_raw_tp), | |
8594 | SEC_DEF("raw_tp/", RAW_TRACEPOINT, | |
8595 | .attach_fn = attach_raw_tp), | |
8596 | SEC_DEF("tp_btf/", TRACING, | |
8597 | .expected_attach_type = BPF_TRACE_RAW_TP, | |
8598 | .is_attach_btf = true, | |
8599 | .attach_fn = attach_trace), | |
8600 | SEC_DEF("fentry/", TRACING, | |
8601 | .expected_attach_type = BPF_TRACE_FENTRY, | |
8602 | .is_attach_btf = true, | |
8603 | .attach_fn = attach_trace), | |
aca228cd KS |
8604 | SEC_DEF("fmod_ret/", TRACING, |
8605 | .expected_attach_type = BPF_MODIFY_RETURN, | |
8606 | .is_attach_btf = true, | |
8607 | .attach_fn = attach_trace), | |
d7a18ea7 AN |
8608 | SEC_DEF("fexit/", TRACING, |
8609 | .expected_attach_type = BPF_TRACE_FEXIT, | |
8610 | .is_attach_btf = true, | |
8611 | .attach_fn = attach_trace), | |
2b288740 AS |
8612 | SEC_DEF("fentry.s/", TRACING, |
8613 | .expected_attach_type = BPF_TRACE_FENTRY, | |
8614 | .is_attach_btf = true, | |
8615 | .is_sleepable = true, | |
8616 | .attach_fn = attach_trace), | |
8617 | SEC_DEF("fmod_ret.s/", TRACING, | |
8618 | .expected_attach_type = BPF_MODIFY_RETURN, | |
8619 | .is_attach_btf = true, | |
8620 | .is_sleepable = true, | |
8621 | .attach_fn = attach_trace), | |
8622 | SEC_DEF("fexit.s/", TRACING, | |
8623 | .expected_attach_type = BPF_TRACE_FEXIT, | |
8624 | .is_attach_btf = true, | |
8625 | .is_sleepable = true, | |
8626 | .attach_fn = attach_trace), | |
2db6eab1 AS |
8627 | SEC_DEF("freplace/", EXT, |
8628 | .is_attach_btf = true, | |
8629 | .attach_fn = attach_trace), | |
1e092a03 KS |
8630 | SEC_DEF("lsm/", LSM, |
8631 | .is_attach_btf = true, | |
8632 | .expected_attach_type = BPF_LSM_MAC, | |
8633 | .attach_fn = attach_lsm), | |
2b288740 AS |
8634 | SEC_DEF("lsm.s/", LSM, |
8635 | .is_attach_btf = true, | |
8636 | .is_sleepable = true, | |
8637 | .expected_attach_type = BPF_LSM_MAC, | |
8638 | .attach_fn = attach_lsm), | |
c09add2f YS |
8639 | SEC_DEF("iter/", TRACING, |
8640 | .expected_attach_type = BPF_TRACE_ITER, | |
8641 | .is_attach_btf = true, | |
8642 | .attach_fn = attach_iter), | |
7a64135f | 8643 | BPF_EAPROG_SEC("xdp_devmap/", BPF_PROG_TYPE_XDP, |
27787970 | 8644 | BPF_XDP_DEVMAP), |
4be556cf LB |
8645 | BPF_EAPROG_SEC("xdp_cpumap/", BPF_PROG_TYPE_XDP, |
8646 | BPF_XDP_CPUMAP), | |
87f92ac4 | 8647 | BPF_APROG_SEC("xdp", BPF_PROG_TYPE_XDP, |
dc8698ca | 8648 | BPF_XDP), |
956b620f AI |
8649 | BPF_PROG_SEC("perf_event", BPF_PROG_TYPE_PERF_EVENT), |
8650 | BPF_PROG_SEC("lwt_in", BPF_PROG_TYPE_LWT_IN), | |
8651 | BPF_PROG_SEC("lwt_out", BPF_PROG_TYPE_LWT_OUT), | |
8652 | BPF_PROG_SEC("lwt_xmit", BPF_PROG_TYPE_LWT_XMIT), | |
8653 | BPF_PROG_SEC("lwt_seg6local", BPF_PROG_TYPE_LWT_SEG6LOCAL), | |
bafa7afe AI |
8654 | BPF_APROG_SEC("cgroup_skb/ingress", BPF_PROG_TYPE_CGROUP_SKB, |
8655 | BPF_CGROUP_INET_INGRESS), | |
8656 | BPF_APROG_SEC("cgroup_skb/egress", BPF_PROG_TYPE_CGROUP_SKB, | |
8657 | BPF_CGROUP_INET_EGRESS), | |
956b620f | 8658 | BPF_APROG_COMPAT("cgroup/skb", BPF_PROG_TYPE_CGROUP_SKB), |
e8b012e9 SF |
8659 | BPF_EAPROG_SEC("cgroup/sock_create", BPF_PROG_TYPE_CGROUP_SOCK, |
8660 | BPF_CGROUP_INET_SOCK_CREATE), | |
8661 | BPF_EAPROG_SEC("cgroup/sock_release", BPF_PROG_TYPE_CGROUP_SOCK, | |
8662 | BPF_CGROUP_INET_SOCK_RELEASE), | |
956b620f AI |
8663 | BPF_APROG_SEC("cgroup/sock", BPF_PROG_TYPE_CGROUP_SOCK, |
8664 | BPF_CGROUP_INET_SOCK_CREATE), | |
8665 | BPF_EAPROG_SEC("cgroup/post_bind4", BPF_PROG_TYPE_CGROUP_SOCK, | |
8666 | BPF_CGROUP_INET4_POST_BIND), | |
8667 | BPF_EAPROG_SEC("cgroup/post_bind6", BPF_PROG_TYPE_CGROUP_SOCK, | |
8668 | BPF_CGROUP_INET6_POST_BIND), | |
8669 | BPF_APROG_SEC("cgroup/dev", BPF_PROG_TYPE_CGROUP_DEVICE, | |
8670 | BPF_CGROUP_DEVICE), | |
8671 | BPF_APROG_SEC("sockops", BPF_PROG_TYPE_SOCK_OPS, | |
8672 | BPF_CGROUP_SOCK_OPS), | |
c6f6851b AI |
8673 | BPF_APROG_SEC("sk_skb/stream_parser", BPF_PROG_TYPE_SK_SKB, |
8674 | BPF_SK_SKB_STREAM_PARSER), | |
8675 | BPF_APROG_SEC("sk_skb/stream_verdict", BPF_PROG_TYPE_SK_SKB, | |
8676 | BPF_SK_SKB_STREAM_VERDICT), | |
956b620f AI |
8677 | BPF_APROG_COMPAT("sk_skb", BPF_PROG_TYPE_SK_SKB), |
8678 | BPF_APROG_SEC("sk_msg", BPF_PROG_TYPE_SK_MSG, | |
8679 | BPF_SK_MSG_VERDICT), | |
8680 | BPF_APROG_SEC("lirc_mode2", BPF_PROG_TYPE_LIRC_MODE2, | |
8681 | BPF_LIRC_MODE2), | |
8682 | BPF_APROG_SEC("flow_dissector", BPF_PROG_TYPE_FLOW_DISSECTOR, | |
8683 | BPF_FLOW_DISSECTOR), | |
8684 | BPF_EAPROG_SEC("cgroup/bind4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, | |
8685 | BPF_CGROUP_INET4_BIND), | |
8686 | BPF_EAPROG_SEC("cgroup/bind6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, | |
8687 | BPF_CGROUP_INET6_BIND), | |
8688 | BPF_EAPROG_SEC("cgroup/connect4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, | |
8689 | BPF_CGROUP_INET4_CONNECT), | |
8690 | BPF_EAPROG_SEC("cgroup/connect6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, | |
8691 | BPF_CGROUP_INET6_CONNECT), | |
8692 | BPF_EAPROG_SEC("cgroup/sendmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, | |
8693 | BPF_CGROUP_UDP4_SENDMSG), | |
8694 | BPF_EAPROG_SEC("cgroup/sendmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, | |
8695 | BPF_CGROUP_UDP6_SENDMSG), | |
9bb59ac1 DB |
8696 | BPF_EAPROG_SEC("cgroup/recvmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, |
8697 | BPF_CGROUP_UDP4_RECVMSG), | |
8698 | BPF_EAPROG_SEC("cgroup/recvmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, | |
8699 | BPF_CGROUP_UDP6_RECVMSG), | |
f15ed018 DB |
8700 | BPF_EAPROG_SEC("cgroup/getpeername4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, |
8701 | BPF_CGROUP_INET4_GETPEERNAME), | |
8702 | BPF_EAPROG_SEC("cgroup/getpeername6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, | |
8703 | BPF_CGROUP_INET6_GETPEERNAME), | |
8704 | BPF_EAPROG_SEC("cgroup/getsockname4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, | |
8705 | BPF_CGROUP_INET4_GETSOCKNAME), | |
8706 | BPF_EAPROG_SEC("cgroup/getsockname6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR, | |
8707 | BPF_CGROUP_INET6_GETSOCKNAME), | |
063cc9f0 AI |
8708 | BPF_EAPROG_SEC("cgroup/sysctl", BPF_PROG_TYPE_CGROUP_SYSCTL, |
8709 | BPF_CGROUP_SYSCTL), | |
4cdbfb59 SF |
8710 | BPF_EAPROG_SEC("cgroup/getsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT, |
8711 | BPF_CGROUP_GETSOCKOPT), | |
8712 | BPF_EAPROG_SEC("cgroup/setsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT, | |
8713 | BPF_CGROUP_SETSOCKOPT), | |
590a0088 | 8714 | BPF_PROG_SEC("struct_ops", BPF_PROG_TYPE_STRUCT_OPS), |
499dd29d JS |
8715 | BPF_EAPROG_SEC("sk_lookup/", BPF_PROG_TYPE_SK_LOOKUP, |
8716 | BPF_SK_LOOKUP), | |
583c9009 | 8717 | }; |
d7be143b | 8718 | |
956b620f | 8719 | #undef BPF_PROG_SEC_IMPL |
583c9009 | 8720 | #undef BPF_PROG_SEC |
956b620f AI |
8721 | #undef BPF_APROG_SEC |
8722 | #undef BPF_EAPROG_SEC | |
8723 | #undef BPF_APROG_COMPAT | |
d7a18ea7 | 8724 | #undef SEC_DEF |
583c9009 | 8725 | |
c76e4c22 TS |
8726 | #define MAX_TYPE_NAME_SIZE 32 |
8727 | ||
d7a18ea7 AN |
8728 | static const struct bpf_sec_def *find_sec_def(const char *sec_name) |
8729 | { | |
8730 | int i, n = ARRAY_SIZE(section_defs); | |
8731 | ||
8732 | for (i = 0; i < n; i++) { | |
8733 | if (strncmp(sec_name, | |
8734 | section_defs[i].sec, section_defs[i].len)) | |
8735 | continue; | |
8736 | return §ion_defs[i]; | |
8737 | } | |
8738 | return NULL; | |
8739 | } | |
8740 | ||
c76e4c22 TS |
8741 | static char *libbpf_get_type_names(bool attach_type) |
8742 | { | |
d7a18ea7 | 8743 | int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE; |
c76e4c22 TS |
8744 | char *buf; |
8745 | ||
8746 | buf = malloc(len); | |
8747 | if (!buf) | |
8748 | return NULL; | |
8749 | ||
8750 | buf[0] = '\0'; | |
8751 | /* Forge string buf with all available names */ | |
d7a18ea7 AN |
8752 | for (i = 0; i < ARRAY_SIZE(section_defs); i++) { |
8753 | if (attach_type && !section_defs[i].is_attachable) | |
c76e4c22 TS |
8754 | continue; |
8755 | ||
d7a18ea7 | 8756 | if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) { |
c76e4c22 TS |
8757 | free(buf); |
8758 | return NULL; | |
8759 | } | |
8760 | strcat(buf, " "); | |
d7a18ea7 | 8761 | strcat(buf, section_defs[i].sec); |
c76e4c22 TS |
8762 | } |
8763 | ||
8764 | return buf; | |
8765 | } | |
8766 | ||
b60df2a0 JK |
8767 | int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type, |
8768 | enum bpf_attach_type *expected_attach_type) | |
583c9009 | 8769 | { |
d7a18ea7 | 8770 | const struct bpf_sec_def *sec_def; |
c76e4c22 | 8771 | char *type_names; |
583c9009 | 8772 | |
b60df2a0 JK |
8773 | if (!name) |
8774 | return -EINVAL; | |
583c9009 | 8775 | |
d7a18ea7 AN |
8776 | sec_def = find_sec_def(name); |
8777 | if (sec_def) { | |
8778 | *prog_type = sec_def->prog_type; | |
8779 | *expected_attach_type = sec_def->expected_attach_type; | |
b60df2a0 JK |
8780 | return 0; |
8781 | } | |
d7a18ea7 | 8782 | |
4a3d6c6a | 8783 | pr_debug("failed to guess program type from ELF section '%s'\n", name); |
c76e4c22 TS |
8784 | type_names = libbpf_get_type_names(false); |
8785 | if (type_names != NULL) { | |
3f519353 | 8786 | pr_debug("supported section(type) names are:%s\n", type_names); |
c76e4c22 TS |
8787 | free(type_names); |
8788 | } | |
8789 | ||
dd4436bb | 8790 | return -ESRCH; |
b60df2a0 | 8791 | } |
583c9009 | 8792 | |
590a0088 MKL |
8793 | static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj, |
8794 | size_t offset) | |
8795 | { | |
8796 | struct bpf_map *map; | |
8797 | size_t i; | |
8798 | ||
8799 | for (i = 0; i < obj->nr_maps; i++) { | |
8800 | map = &obj->maps[i]; | |
8801 | if (!bpf_map__is_struct_ops(map)) | |
8802 | continue; | |
8803 | if (map->sec_offset <= offset && | |
8804 | offset - map->sec_offset < map->def.value_size) | |
8805 | return map; | |
8806 | } | |
8807 | ||
8808 | return NULL; | |
8809 | } | |
8810 | ||
8811 | /* Collect the reloc from ELF and populate the st_ops->progs[] */ | |
646f02ff AN |
8812 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
8813 | GElf_Shdr *shdr, Elf_Data *data) | |
590a0088 MKL |
8814 | { |
8815 | const struct btf_member *member; | |
8816 | struct bpf_struct_ops *st_ops; | |
8817 | struct bpf_program *prog; | |
8818 | unsigned int shdr_idx; | |
8819 | const struct btf *btf; | |
8820 | struct bpf_map *map; | |
8821 | Elf_Data *symbols; | |
7e06aad5 | 8822 | unsigned int moff, insn_idx; |
590a0088 | 8823 | const char *name; |
1d1a3bcf | 8824 | __u32 member_idx; |
590a0088 MKL |
8825 | GElf_Sym sym; |
8826 | GElf_Rel rel; | |
8827 | int i, nrels; | |
8828 | ||
8829 | symbols = obj->efile.symbols; | |
8830 | btf = obj->btf; | |
8831 | nrels = shdr->sh_size / shdr->sh_entsize; | |
8832 | for (i = 0; i < nrels; i++) { | |
8833 | if (!gelf_getrel(data, i, &rel)) { | |
8834 | pr_warn("struct_ops reloc: failed to get %d reloc\n", i); | |
8835 | return -LIBBPF_ERRNO__FORMAT; | |
8836 | } | |
8837 | ||
8838 | if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) { | |
8839 | pr_warn("struct_ops reloc: symbol %zx not found\n", | |
8840 | (size_t)GELF_R_SYM(rel.r_info)); | |
8841 | return -LIBBPF_ERRNO__FORMAT; | |
8842 | } | |
8843 | ||
88a82120 | 8844 | name = elf_sym_str(obj, sym.st_name) ?: "<?>"; |
590a0088 MKL |
8845 | map = find_struct_ops_map_by_offset(obj, rel.r_offset); |
8846 | if (!map) { | |
8847 | pr_warn("struct_ops reloc: cannot find map at rel.r_offset %zu\n", | |
8848 | (size_t)rel.r_offset); | |
8849 | return -EINVAL; | |
8850 | } | |
8851 | ||
8852 | moff = rel.r_offset - map->sec_offset; | |
8853 | shdr_idx = sym.st_shndx; | |
8854 | st_ops = map->st_ops; | |
8855 | 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", | |
8856 | map->name, | |
8857 | (long long)(rel.r_info >> 32), | |
8858 | (long long)sym.st_value, | |
8859 | shdr_idx, (size_t)rel.r_offset, | |
8860 | map->sec_offset, sym.st_name, name); | |
8861 | ||
8862 | if (shdr_idx >= SHN_LORESERVE) { | |
8863 | pr_warn("struct_ops reloc %s: rel.r_offset %zu shdr_idx %u unsupported non-static function\n", | |
8864 | map->name, (size_t)rel.r_offset, shdr_idx); | |
8865 | return -LIBBPF_ERRNO__RELOC; | |
8866 | } | |
7e06aad5 AN |
8867 | if (sym.st_value % BPF_INSN_SZ) { |
8868 | pr_warn("struct_ops reloc %s: invalid target program offset %llu\n", | |
17e54b09 | 8869 | map->name, (unsigned long long)sym.st_value); |
7e06aad5 AN |
8870 | return -LIBBPF_ERRNO__FORMAT; |
8871 | } | |
8872 | insn_idx = sym.st_value / BPF_INSN_SZ; | |
590a0088 MKL |
8873 | |
8874 | member = find_member_by_offset(st_ops->type, moff * 8); | |
8875 | if (!member) { | |
8876 | pr_warn("struct_ops reloc %s: cannot find member at moff %u\n", | |
8877 | map->name, moff); | |
8878 | return -EINVAL; | |
8879 | } | |
8880 | member_idx = member - btf_members(st_ops->type); | |
8881 | name = btf__name_by_offset(btf, member->name_off); | |
8882 | ||
8883 | if (!resolve_func_ptr(btf, member->type, NULL)) { | |
8884 | pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n", | |
8885 | map->name, name); | |
8886 | return -EINVAL; | |
8887 | } | |
8888 | ||
7e06aad5 | 8889 | prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx); |
590a0088 MKL |
8890 | if (!prog) { |
8891 | pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n", | |
8892 | map->name, shdr_idx, name); | |
8893 | return -EINVAL; | |
8894 | } | |
8895 | ||
8896 | if (prog->type == BPF_PROG_TYPE_UNSPEC) { | |
8897 | const struct bpf_sec_def *sec_def; | |
8898 | ||
52109584 | 8899 | sec_def = find_sec_def(prog->sec_name); |
590a0088 MKL |
8900 | if (sec_def && |
8901 | sec_def->prog_type != BPF_PROG_TYPE_STRUCT_OPS) { | |
8902 | /* for pr_warn */ | |
8903 | prog->type = sec_def->prog_type; | |
8904 | goto invalid_prog; | |
8905 | } | |
8906 | ||
8907 | prog->type = BPF_PROG_TYPE_STRUCT_OPS; | |
8908 | prog->attach_btf_id = st_ops->type_id; | |
8909 | prog->expected_attach_type = member_idx; | |
8910 | } else if (prog->type != BPF_PROG_TYPE_STRUCT_OPS || | |
8911 | prog->attach_btf_id != st_ops->type_id || | |
8912 | prog->expected_attach_type != member_idx) { | |
8913 | goto invalid_prog; | |
8914 | } | |
8915 | st_ops->progs[member_idx] = prog; | |
8916 | } | |
8917 | ||
8918 | return 0; | |
8919 | ||
8920 | invalid_prog: | |
8921 | 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", | |
52109584 | 8922 | map->name, prog->name, prog->sec_name, prog->type, |
590a0088 MKL |
8923 | prog->attach_btf_id, prog->expected_attach_type, name); |
8924 | return -EINVAL; | |
8925 | } | |
8926 | ||
a6ed02ca | 8927 | #define BTF_TRACE_PREFIX "btf_trace_" |
1e092a03 | 8928 | #define BTF_LSM_PREFIX "bpf_lsm_" |
21aef70e | 8929 | #define BTF_ITER_PREFIX "bpf_iter_" |
a6ed02ca KS |
8930 | #define BTF_MAX_NAME_SIZE 128 |
8931 | ||
8932 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, | |
8933 | const char *name, __u32 kind) | |
8934 | { | |
8935 | char btf_type_name[BTF_MAX_NAME_SIZE]; | |
8936 | int ret; | |
8937 | ||
8938 | ret = snprintf(btf_type_name, sizeof(btf_type_name), | |
8939 | "%s%s", prefix, name); | |
8940 | /* snprintf returns the number of characters written excluding the | |
8941 | * the terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it | |
8942 | * indicates truncation. | |
8943 | */ | |
8944 | if (ret < 0 || ret >= sizeof(btf_type_name)) | |
8945 | return -ENAMETOOLONG; | |
8946 | return btf__find_by_name_kind(btf, btf_type_name, kind); | |
8947 | } | |
8948 | ||
91abb4a6 AN |
8949 | static inline int find_attach_btf_id(struct btf *btf, const char *name, |
8950 | enum bpf_attach_type attach_type) | |
a6ed02ca KS |
8951 | { |
8952 | int err; | |
8953 | ||
8954 | if (attach_type == BPF_TRACE_RAW_TP) | |
8955 | err = find_btf_by_prefix_kind(btf, BTF_TRACE_PREFIX, name, | |
8956 | BTF_KIND_TYPEDEF); | |
1e092a03 KS |
8957 | else if (attach_type == BPF_LSM_MAC) |
8958 | err = find_btf_by_prefix_kind(btf, BTF_LSM_PREFIX, name, | |
8959 | BTF_KIND_FUNC); | |
c09add2f YS |
8960 | else if (attach_type == BPF_TRACE_ITER) |
8961 | err = find_btf_by_prefix_kind(btf, BTF_ITER_PREFIX, name, | |
8962 | BTF_KIND_FUNC); | |
a6ed02ca KS |
8963 | else |
8964 | err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC); | |
8965 | ||
8966 | return err; | |
8967 | } | |
8968 | ||
b8c54ea4 AS |
8969 | int libbpf_find_vmlinux_btf_id(const char *name, |
8970 | enum bpf_attach_type attach_type) | |
12a8654b | 8971 | { |
a6ed02ca | 8972 | struct btf *btf; |
3521ffa2 | 8973 | int err; |
12a8654b | 8974 | |
a6ed02ca | 8975 | btf = libbpf_find_kernel_btf(); |
12a8654b AS |
8976 | if (IS_ERR(btf)) { |
8977 | pr_warn("vmlinux BTF is not found\n"); | |
8978 | return -EINVAL; | |
8979 | } | |
8980 | ||
91abb4a6 AN |
8981 | err = find_attach_btf_id(btf, name, attach_type); |
8982 | if (err <= 0) | |
8983 | pr_warn("%s is not found in vmlinux BTF\n", name); | |
8984 | ||
3521ffa2 AN |
8985 | btf__free(btf); |
8986 | return err; | |
b8c54ea4 AS |
8987 | } |
8988 | ||
e7bf94db AS |
8989 | static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd) |
8990 | { | |
8991 | struct bpf_prog_info_linear *info_linear; | |
8992 | struct bpf_prog_info *info; | |
8993 | struct btf *btf = NULL; | |
8994 | int err = -EINVAL; | |
8995 | ||
8996 | info_linear = bpf_program__get_prog_info_linear(attach_prog_fd, 0); | |
8997 | if (IS_ERR_OR_NULL(info_linear)) { | |
8998 | pr_warn("failed get_prog_info_linear for FD %d\n", | |
8999 | attach_prog_fd); | |
9000 | return -EINVAL; | |
9001 | } | |
9002 | info = &info_linear->info; | |
9003 | if (!info->btf_id) { | |
9004 | pr_warn("The target program doesn't have BTF\n"); | |
9005 | goto out; | |
9006 | } | |
9007 | if (btf__get_from_id(info->btf_id, &btf)) { | |
9008 | pr_warn("Failed to get BTF of the program\n"); | |
9009 | goto out; | |
9010 | } | |
9011 | err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC); | |
9012 | btf__free(btf); | |
9013 | if (err <= 0) { | |
9014 | pr_warn("%s is not found in prog's BTF\n", name); | |
9015 | goto out; | |
9016 | } | |
9017 | out: | |
9018 | free(info_linear); | |
9019 | return err; | |
9020 | } | |
9021 | ||
91abb4a6 AN |
9022 | static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name, |
9023 | enum bpf_attach_type attach_type, | |
9024 | int *btf_obj_fd, int *btf_type_id) | |
9025 | { | |
9026 | int ret, i; | |
9027 | ||
9028 | ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type); | |
9029 | if (ret > 0) { | |
9030 | *btf_obj_fd = 0; /* vmlinux BTF */ | |
9031 | *btf_type_id = ret; | |
9032 | return 0; | |
9033 | } | |
9034 | if (ret != -ENOENT) | |
9035 | return ret; | |
9036 | ||
9037 | ret = load_module_btfs(obj); | |
9038 | if (ret) | |
9039 | return ret; | |
9040 | ||
9041 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
9042 | const struct module_btf *mod = &obj->btf_modules[i]; | |
9043 | ||
9044 | ret = find_attach_btf_id(mod->btf, attach_name, attach_type); | |
9045 | if (ret > 0) { | |
9046 | *btf_obj_fd = mod->fd; | |
9047 | *btf_type_id = ret; | |
9048 | return 0; | |
9049 | } | |
9050 | if (ret == -ENOENT) | |
9051 | continue; | |
9052 | ||
9053 | return ret; | |
9054 | } | |
9055 | ||
9056 | return -ESRCH; | |
9057 | } | |
9058 | ||
9059 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, int *btf_obj_fd, int *btf_type_id) | |
b8c54ea4 | 9060 | { |
a6ed02ca KS |
9061 | enum bpf_attach_type attach_type = prog->expected_attach_type; |
9062 | __u32 attach_prog_fd = prog->attach_prog_fd; | |
91abb4a6 AN |
9063 | const char *name = prog->sec_name, *attach_name; |
9064 | const struct bpf_sec_def *sec = NULL; | |
b8c54ea4 AS |
9065 | int i, err; |
9066 | ||
12a8654b | 9067 | if (!name) |
b8c54ea4 | 9068 | return -EINVAL; |
12a8654b | 9069 | |
d7a18ea7 AN |
9070 | for (i = 0; i < ARRAY_SIZE(section_defs); i++) { |
9071 | if (!section_defs[i].is_attach_btf) | |
12a8654b | 9072 | continue; |
d7a18ea7 | 9073 | if (strncmp(name, section_defs[i].sec, section_defs[i].len)) |
12a8654b | 9074 | continue; |
91abb4a6 AN |
9075 | |
9076 | sec = §ion_defs[i]; | |
9077 | break; | |
9078 | } | |
9079 | ||
9080 | if (!sec) { | |
9081 | pr_warn("failed to identify BTF ID based on ELF section name '%s'\n", name); | |
9082 | return -ESRCH; | |
9083 | } | |
9084 | attach_name = name + sec->len; | |
9085 | ||
9086 | /* BPF program's BTF ID */ | |
9087 | if (attach_prog_fd) { | |
9088 | err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd); | |
9089 | if (err < 0) { | |
9090 | pr_warn("failed to find BPF program (FD %d) BTF ID for '%s': %d\n", | |
9091 | attach_prog_fd, attach_name, err); | |
9092 | return err; | |
9093 | } | |
9094 | *btf_obj_fd = 0; | |
9095 | *btf_type_id = err; | |
9096 | return 0; | |
9097 | } | |
9098 | ||
9099 | /* kernel/module BTF ID */ | |
9100 | err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id); | |
9101 | if (err) { | |
9102 | pr_warn("failed to find kernel BTF type ID of '%s': %d\n", attach_name, err); | |
b8c54ea4 | 9103 | return err; |
12a8654b | 9104 | } |
91abb4a6 | 9105 | return 0; |
12a8654b AS |
9106 | } |
9107 | ||
956b620f AI |
9108 | int libbpf_attach_type_by_name(const char *name, |
9109 | enum bpf_attach_type *attach_type) | |
9110 | { | |
c76e4c22 | 9111 | char *type_names; |
956b620f AI |
9112 | int i; |
9113 | ||
9114 | if (!name) | |
9115 | return -EINVAL; | |
9116 | ||
d7a18ea7 AN |
9117 | for (i = 0; i < ARRAY_SIZE(section_defs); i++) { |
9118 | if (strncmp(name, section_defs[i].sec, section_defs[i].len)) | |
956b620f | 9119 | continue; |
d7a18ea7 | 9120 | if (!section_defs[i].is_attachable) |
956b620f | 9121 | return -EINVAL; |
25498a19 | 9122 | *attach_type = section_defs[i].expected_attach_type; |
956b620f AI |
9123 | return 0; |
9124 | } | |
4a3d6c6a | 9125 | pr_debug("failed to guess attach type based on ELF section name '%s'\n", name); |
c76e4c22 TS |
9126 | type_names = libbpf_get_type_names(true); |
9127 | if (type_names != NULL) { | |
4a3d6c6a | 9128 | pr_debug("attachable section(type) names are:%s\n", type_names); |
c76e4c22 TS |
9129 | free(type_names); |
9130 | } | |
9131 | ||
956b620f AI |
9132 | return -EINVAL; |
9133 | } | |
9134 | ||
a324aae3 | 9135 | int bpf_map__fd(const struct bpf_map *map) |
9d759a9b | 9136 | { |
6e009e65 | 9137 | return map ? map->fd : -EINVAL; |
9d759a9b WN |
9138 | } |
9139 | ||
a324aae3 | 9140 | const struct bpf_map_def *bpf_map__def(const struct bpf_map *map) |
9d759a9b | 9141 | { |
53897a78 | 9142 | return map ? &map->def : ERR_PTR(-EINVAL); |
9d759a9b WN |
9143 | } |
9144 | ||
a324aae3 | 9145 | const char *bpf_map__name(const struct bpf_map *map) |
561bbcca | 9146 | { |
009ad5d5 | 9147 | return map ? map->name : NULL; |
561bbcca WN |
9148 | } |
9149 | ||
1bdb6c9a AN |
9150 | enum bpf_map_type bpf_map__type(const struct bpf_map *map) |
9151 | { | |
9152 | return map->def.type; | |
9153 | } | |
9154 | ||
9155 | int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type) | |
9156 | { | |
9157 | if (map->fd >= 0) | |
9158 | return -EBUSY; | |
9159 | map->def.type = type; | |
9160 | return 0; | |
9161 | } | |
9162 | ||
9163 | __u32 bpf_map__map_flags(const struct bpf_map *map) | |
9164 | { | |
9165 | return map->def.map_flags; | |
9166 | } | |
9167 | ||
9168 | int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags) | |
9169 | { | |
9170 | if (map->fd >= 0) | |
9171 | return -EBUSY; | |
9172 | map->def.map_flags = flags; | |
9173 | return 0; | |
9174 | } | |
9175 | ||
9176 | __u32 bpf_map__numa_node(const struct bpf_map *map) | |
9177 | { | |
9178 | return map->numa_node; | |
9179 | } | |
9180 | ||
9181 | int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node) | |
9182 | { | |
9183 | if (map->fd >= 0) | |
9184 | return -EBUSY; | |
9185 | map->numa_node = numa_node; | |
9186 | return 0; | |
9187 | } | |
9188 | ||
9189 | __u32 bpf_map__key_size(const struct bpf_map *map) | |
9190 | { | |
9191 | return map->def.key_size; | |
9192 | } | |
9193 | ||
9194 | int bpf_map__set_key_size(struct bpf_map *map, __u32 size) | |
9195 | { | |
9196 | if (map->fd >= 0) | |
9197 | return -EBUSY; | |
9198 | map->def.key_size = size; | |
9199 | return 0; | |
9200 | } | |
9201 | ||
9202 | __u32 bpf_map__value_size(const struct bpf_map *map) | |
9203 | { | |
9204 | return map->def.value_size; | |
9205 | } | |
9206 | ||
9207 | int bpf_map__set_value_size(struct bpf_map *map, __u32 size) | |
9208 | { | |
9209 | if (map->fd >= 0) | |
9210 | return -EBUSY; | |
9211 | map->def.value_size = size; | |
9212 | return 0; | |
9213 | } | |
9214 | ||
5b891af7 | 9215 | __u32 bpf_map__btf_key_type_id(const struct bpf_map *map) |
8a138aed | 9216 | { |
61746dbe | 9217 | return map ? map->btf_key_type_id : 0; |
8a138aed MKL |
9218 | } |
9219 | ||
5b891af7 | 9220 | __u32 bpf_map__btf_value_type_id(const struct bpf_map *map) |
8a138aed | 9221 | { |
61746dbe | 9222 | return map ? map->btf_value_type_id : 0; |
8a138aed MKL |
9223 | } |
9224 | ||
edb13ed4 ACM |
9225 | int bpf_map__set_priv(struct bpf_map *map, void *priv, |
9226 | bpf_map_clear_priv_t clear_priv) | |
9d759a9b WN |
9227 | { |
9228 | if (!map) | |
9229 | return -EINVAL; | |
9230 | ||
9231 | if (map->priv) { | |
9232 | if (map->clear_priv) | |
9233 | map->clear_priv(map, map->priv); | |
9234 | } | |
9235 | ||
9236 | map->priv = priv; | |
9237 | map->clear_priv = clear_priv; | |
9238 | return 0; | |
9239 | } | |
9240 | ||
a324aae3 | 9241 | void *bpf_map__priv(const struct bpf_map *map) |
9d759a9b | 9242 | { |
b4cbfa56 | 9243 | return map ? map->priv : ERR_PTR(-EINVAL); |
9d759a9b WN |
9244 | } |
9245 | ||
e2842be5 THJ |
9246 | int bpf_map__set_initial_value(struct bpf_map *map, |
9247 | const void *data, size_t size) | |
9248 | { | |
9249 | if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG || | |
9250 | size != map->def.value_size || map->fd >= 0) | |
9251 | return -EINVAL; | |
9252 | ||
9253 | memcpy(map->mmaped, data, size); | |
9254 | return 0; | |
9255 | } | |
9256 | ||
a324aae3 | 9257 | bool bpf_map__is_offload_neutral(const struct bpf_map *map) |
f83fb22c JK |
9258 | { |
9259 | return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; | |
9260 | } | |
9261 | ||
a324aae3 | 9262 | bool bpf_map__is_internal(const struct bpf_map *map) |
d859900c DB |
9263 | { |
9264 | return map->libbpf_type != LIBBPF_MAP_UNSPEC; | |
9265 | } | |
9266 | ||
1bdb6c9a AN |
9267 | __u32 bpf_map__ifindex(const struct bpf_map *map) |
9268 | { | |
9269 | return map->map_ifindex; | |
9270 | } | |
9271 | ||
9272 | int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex) | |
9aba3613 | 9273 | { |
1bdb6c9a AN |
9274 | if (map->fd >= 0) |
9275 | return -EBUSY; | |
9aba3613 | 9276 | map->map_ifindex = ifindex; |
1bdb6c9a | 9277 | return 0; |
9aba3613 JK |
9278 | } |
9279 | ||
addb9fc9 NS |
9280 | int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd) |
9281 | { | |
9282 | if (!bpf_map_type__is_map_in_map(map->def.type)) { | |
be18010e | 9283 | pr_warn("error: unsupported map type\n"); |
addb9fc9 NS |
9284 | return -EINVAL; |
9285 | } | |
9286 | if (map->inner_map_fd != -1) { | |
be18010e | 9287 | pr_warn("error: inner_map_fd already specified\n"); |
addb9fc9 NS |
9288 | return -EINVAL; |
9289 | } | |
9290 | map->inner_map_fd = fd; | |
9291 | return 0; | |
9292 | } | |
9293 | ||
0c19a9fb | 9294 | static struct bpf_map * |
a324aae3 | 9295 | __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i) |
9d759a9b | 9296 | { |
0c19a9fb | 9297 | ssize_t idx; |
9d759a9b WN |
9298 | struct bpf_map *s, *e; |
9299 | ||
9300 | if (!obj || !obj->maps) | |
9301 | return NULL; | |
9302 | ||
9303 | s = obj->maps; | |
9304 | e = obj->maps + obj->nr_maps; | |
9305 | ||
0c19a9fb | 9306 | if ((m < s) || (m >= e)) { |
be18010e KW |
9307 | pr_warn("error in %s: map handler doesn't belong to object\n", |
9308 | __func__); | |
9d759a9b WN |
9309 | return NULL; |
9310 | } | |
9311 | ||
0c19a9fb SF |
9312 | idx = (m - obj->maps) + i; |
9313 | if (idx >= obj->nr_maps || idx < 0) | |
9d759a9b WN |
9314 | return NULL; |
9315 | return &obj->maps[idx]; | |
9316 | } | |
561bbcca | 9317 | |
0c19a9fb | 9318 | struct bpf_map * |
a324aae3 | 9319 | bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj) |
0c19a9fb SF |
9320 | { |
9321 | if (prev == NULL) | |
9322 | return obj->maps; | |
9323 | ||
9324 | return __bpf_map__iter(prev, obj, 1); | |
9325 | } | |
9326 | ||
9327 | struct bpf_map * | |
a324aae3 | 9328 | bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj) |
0c19a9fb SF |
9329 | { |
9330 | if (next == NULL) { | |
9331 | if (!obj->nr_maps) | |
9332 | return NULL; | |
9333 | return obj->maps + obj->nr_maps - 1; | |
9334 | } | |
9335 | ||
9336 | return __bpf_map__iter(next, obj, -1); | |
9337 | } | |
9338 | ||
561bbcca | 9339 | struct bpf_map * |
a324aae3 | 9340 | bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name) |
561bbcca WN |
9341 | { |
9342 | struct bpf_map *pos; | |
9343 | ||
f74a53d9 | 9344 | bpf_object__for_each_map(pos, obj) { |
973170e6 | 9345 | if (pos->name && !strcmp(pos->name, name)) |
561bbcca WN |
9346 | return pos; |
9347 | } | |
9348 | return NULL; | |
9349 | } | |
5a6acad1 | 9350 | |
f3cea32d | 9351 | int |
a324aae3 | 9352 | bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name) |
f3cea32d MF |
9353 | { |
9354 | return bpf_map__fd(bpf_object__find_map_by_name(obj, name)); | |
9355 | } | |
9356 | ||
5a6acad1 WN |
9357 | struct bpf_map * |
9358 | bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset) | |
9359 | { | |
db48814b | 9360 | return ERR_PTR(-ENOTSUP); |
5a6acad1 | 9361 | } |
e28ff1a8 JS |
9362 | |
9363 | long libbpf_get_error(const void *ptr) | |
9364 | { | |
d98363b5 | 9365 | return PTR_ERR_OR_ZERO(ptr); |
e28ff1a8 | 9366 | } |
6f6d33f3 JF |
9367 | |
9368 | int bpf_prog_load(const char *file, enum bpf_prog_type type, | |
9369 | struct bpf_object **pobj, int *prog_fd) | |
d7be143b AI |
9370 | { |
9371 | struct bpf_prog_load_attr attr; | |
9372 | ||
9373 | memset(&attr, 0, sizeof(struct bpf_prog_load_attr)); | |
9374 | attr.file = file; | |
9375 | attr.prog_type = type; | |
9376 | attr.expected_attach_type = 0; | |
9377 | ||
9378 | return bpf_prog_load_xattr(&attr, pobj, prog_fd); | |
9379 | } | |
9380 | ||
9381 | int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr, | |
9382 | struct bpf_object **pobj, int *prog_fd) | |
6f6d33f3 | 9383 | { |
33bae185 | 9384 | struct bpf_object_open_attr open_attr = {}; |
48cca7e4 | 9385 | struct bpf_program *prog, *first_prog = NULL; |
6f6d33f3 | 9386 | struct bpf_object *obj; |
f0307a7e | 9387 | struct bpf_map *map; |
6f6d33f3 JF |
9388 | int err; |
9389 | ||
d7be143b AI |
9390 | if (!attr) |
9391 | return -EINVAL; | |
17387dd5 JK |
9392 | if (!attr->file) |
9393 | return -EINVAL; | |
d7be143b | 9394 | |
33bae185 LY |
9395 | open_attr.file = attr->file; |
9396 | open_attr.prog_type = attr->prog_type; | |
9397 | ||
07f2d4ea | 9398 | obj = bpf_object__open_xattr(&open_attr); |
3597683c | 9399 | if (IS_ERR_OR_NULL(obj)) |
6f6d33f3 JF |
9400 | return -ENOENT; |
9401 | ||
48cca7e4 | 9402 | bpf_object__for_each_program(prog, obj) { |
dd4436bb | 9403 | enum bpf_attach_type attach_type = attr->expected_attach_type; |
48cca7e4 | 9404 | /* |
dd4436bb AN |
9405 | * to preserve backwards compatibility, bpf_prog_load treats |
9406 | * attr->prog_type, if specified, as an override to whatever | |
9407 | * bpf_object__open guessed | |
48cca7e4 | 9408 | */ |
dd4436bb AN |
9409 | if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) { |
9410 | bpf_program__set_type(prog, attr->prog_type); | |
9411 | bpf_program__set_expected_attach_type(prog, | |
9412 | attach_type); | |
9413 | } | |
9414 | if (bpf_program__get_type(prog) == BPF_PROG_TYPE_UNSPEC) { | |
9415 | /* | |
9416 | * we haven't guessed from section name and user | |
9417 | * didn't provide a fallback type, too bad... | |
9418 | */ | |
9419 | bpf_object__close(obj); | |
9420 | return -EINVAL; | |
583c9009 | 9421 | } |
48cca7e4 | 9422 | |
dd4436bb | 9423 | prog->prog_ifindex = attr->ifindex; |
da11b417 | 9424 | prog->log_level = attr->log_level; |
2b288740 | 9425 | prog->prog_flags |= attr->prog_flags; |
69495d2a | 9426 | if (!first_prog) |
48cca7e4 AS |
9427 | first_prog = prog; |
9428 | } | |
9429 | ||
f74a53d9 | 9430 | bpf_object__for_each_map(map, obj) { |
f83fb22c JK |
9431 | if (!bpf_map__is_offload_neutral(map)) |
9432 | map->map_ifindex = attr->ifindex; | |
f0307a7e DB |
9433 | } |
9434 | ||
48cca7e4 | 9435 | if (!first_prog) { |
be18010e | 9436 | pr_warn("object file doesn't contain bpf program\n"); |
48cca7e4 AS |
9437 | bpf_object__close(obj); |
9438 | return -ENOENT; | |
583c9009 RG |
9439 | } |
9440 | ||
6f6d33f3 JF |
9441 | err = bpf_object__load(obj); |
9442 | if (err) { | |
9443 | bpf_object__close(obj); | |
e411eb25 | 9444 | return err; |
6f6d33f3 JF |
9445 | } |
9446 | ||
9447 | *pobj = obj; | |
48cca7e4 | 9448 | *prog_fd = bpf_program__fd(first_prog); |
6f6d33f3 JF |
9449 | return 0; |
9450 | } | |
d0cabbb0 | 9451 | |
1c2e9efc | 9452 | struct bpf_link { |
d6958706 | 9453 | int (*detach)(struct bpf_link *link); |
1c2e9efc | 9454 | int (*destroy)(struct bpf_link *link); |
c016b68e AN |
9455 | char *pin_path; /* NULL, if not pinned */ |
9456 | int fd; /* hook FD, -1 if not applicable */ | |
d6958706 | 9457 | bool disconnected; |
1c2e9efc AN |
9458 | }; |
9459 | ||
cc4f864b AN |
9460 | /* Replace link's underlying BPF program with the new one */ |
9461 | int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog) | |
9462 | { | |
9463 | return bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL); | |
9464 | } | |
9465 | ||
d6958706 AN |
9466 | /* Release "ownership" of underlying BPF resource (typically, BPF program |
9467 | * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected | |
9468 | * link, when destructed through bpf_link__destroy() call won't attempt to | |
9469 | * detach/unregisted that BPF resource. This is useful in situations where, | |
9470 | * say, attached BPF program has to outlive userspace program that attached it | |
9471 | * in the system. Depending on type of BPF program, though, there might be | |
9472 | * additional steps (like pinning BPF program in BPF FS) necessary to ensure | |
9473 | * exit of userspace program doesn't trigger automatic detachment and clean up | |
9474 | * inside the kernel. | |
9475 | */ | |
9476 | void bpf_link__disconnect(struct bpf_link *link) | |
9477 | { | |
9478 | link->disconnected = true; | |
9479 | } | |
9480 | ||
1c2e9efc AN |
9481 | int bpf_link__destroy(struct bpf_link *link) |
9482 | { | |
d6958706 | 9483 | int err = 0; |
1c2e9efc | 9484 | |
50450fc7 | 9485 | if (IS_ERR_OR_NULL(link)) |
1c2e9efc AN |
9486 | return 0; |
9487 | ||
d6958706 AN |
9488 | if (!link->disconnected && link->detach) |
9489 | err = link->detach(link); | |
9490 | if (link->destroy) | |
9491 | link->destroy(link); | |
c016b68e AN |
9492 | if (link->pin_path) |
9493 | free(link->pin_path); | |
1c2e9efc AN |
9494 | free(link); |
9495 | ||
9496 | return err; | |
9497 | } | |
9498 | ||
c016b68e AN |
9499 | int bpf_link__fd(const struct bpf_link *link) |
9500 | { | |
9501 | return link->fd; | |
9502 | } | |
9503 | ||
9504 | const char *bpf_link__pin_path(const struct bpf_link *link) | |
9505 | { | |
9506 | return link->pin_path; | |
9507 | } | |
9508 | ||
9509 | static int bpf_link__detach_fd(struct bpf_link *link) | |
9510 | { | |
9511 | return close(link->fd); | |
9512 | } | |
9513 | ||
9514 | struct bpf_link *bpf_link__open(const char *path) | |
9515 | { | |
9516 | struct bpf_link *link; | |
9517 | int fd; | |
9518 | ||
9519 | fd = bpf_obj_get(path); | |
9520 | if (fd < 0) { | |
9521 | fd = -errno; | |
9522 | pr_warn("failed to open link at %s: %d\n", path, fd); | |
9523 | return ERR_PTR(fd); | |
9524 | } | |
9525 | ||
9526 | link = calloc(1, sizeof(*link)); | |
9527 | if (!link) { | |
9528 | close(fd); | |
9529 | return ERR_PTR(-ENOMEM); | |
9530 | } | |
9531 | link->detach = &bpf_link__detach_fd; | |
9532 | link->fd = fd; | |
9533 | ||
9534 | link->pin_path = strdup(path); | |
9535 | if (!link->pin_path) { | |
9536 | bpf_link__destroy(link); | |
9537 | return ERR_PTR(-ENOMEM); | |
9538 | } | |
9539 | ||
9540 | return link; | |
9541 | } | |
9542 | ||
2e49527e AN |
9543 | int bpf_link__detach(struct bpf_link *link) |
9544 | { | |
9545 | return bpf_link_detach(link->fd) ? -errno : 0; | |
9546 | } | |
9547 | ||
c016b68e AN |
9548 | int bpf_link__pin(struct bpf_link *link, const char *path) |
9549 | { | |
9550 | int err; | |
9551 | ||
9552 | if (link->pin_path) | |
9553 | return -EBUSY; | |
9554 | err = make_parent_dir(path); | |
9555 | if (err) | |
9556 | return err; | |
9557 | err = check_path(path); | |
9558 | if (err) | |
9559 | return err; | |
9560 | ||
9561 | link->pin_path = strdup(path); | |
9562 | if (!link->pin_path) | |
9563 | return -ENOMEM; | |
9564 | ||
9565 | if (bpf_obj_pin(link->fd, link->pin_path)) { | |
9566 | err = -errno; | |
9567 | zfree(&link->pin_path); | |
9568 | return err; | |
9569 | } | |
9570 | ||
9571 | pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path); | |
9572 | return 0; | |
9573 | } | |
9574 | ||
9575 | int bpf_link__unpin(struct bpf_link *link) | |
9576 | { | |
9577 | int err; | |
9578 | ||
9579 | if (!link->pin_path) | |
9580 | return -EINVAL; | |
9581 | ||
9582 | err = unlink(link->pin_path); | |
9583 | if (err != 0) | |
9584 | return -errno; | |
9585 | ||
9586 | pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path); | |
9587 | zfree(&link->pin_path); | |
9588 | return 0; | |
9589 | } | |
63f2f5ee | 9590 | |
d6958706 | 9591 | static int bpf_link__detach_perf_event(struct bpf_link *link) |
63f2f5ee | 9592 | { |
63f2f5ee AN |
9593 | int err; |
9594 | ||
c016b68e | 9595 | err = ioctl(link->fd, PERF_EVENT_IOC_DISABLE, 0); |
63f2f5ee AN |
9596 | if (err) |
9597 | err = -errno; | |
9598 | ||
c016b68e | 9599 | close(link->fd); |
63f2f5ee AN |
9600 | return err; |
9601 | } | |
9602 | ||
9603 | struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog, | |
9604 | int pfd) | |
9605 | { | |
9606 | char errmsg[STRERR_BUFSIZE]; | |
c016b68e | 9607 | struct bpf_link *link; |
63f2f5ee AN |
9608 | int prog_fd, err; |
9609 | ||
9610 | if (pfd < 0) { | |
52109584 AN |
9611 | pr_warn("prog '%s': invalid perf event FD %d\n", |
9612 | prog->name, pfd); | |
63f2f5ee AN |
9613 | return ERR_PTR(-EINVAL); |
9614 | } | |
9615 | prog_fd = bpf_program__fd(prog); | |
9616 | if (prog_fd < 0) { | |
52109584 AN |
9617 | pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n", |
9618 | prog->name); | |
63f2f5ee AN |
9619 | return ERR_PTR(-EINVAL); |
9620 | } | |
9621 | ||
d6958706 | 9622 | link = calloc(1, sizeof(*link)); |
63f2f5ee AN |
9623 | if (!link) |
9624 | return ERR_PTR(-ENOMEM); | |
c016b68e | 9625 | link->detach = &bpf_link__detach_perf_event; |
63f2f5ee AN |
9626 | link->fd = pfd; |
9627 | ||
9628 | if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) { | |
9629 | err = -errno; | |
9630 | free(link); | |
52109584 AN |
9631 | pr_warn("prog '%s': failed to attach to pfd %d: %s\n", |
9632 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
d4b4dd6c | 9633 | if (err == -EPROTO) |
52109584 AN |
9634 | pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n", |
9635 | prog->name, pfd); | |
63f2f5ee AN |
9636 | return ERR_PTR(err); |
9637 | } | |
9638 | if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
9639 | err = -errno; | |
9640 | free(link); | |
52109584 AN |
9641 | pr_warn("prog '%s': failed to enable pfd %d: %s\n", |
9642 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
63f2f5ee AN |
9643 | return ERR_PTR(err); |
9644 | } | |
c016b68e | 9645 | return link; |
63f2f5ee AN |
9646 | } |
9647 | ||
b2650027 AN |
9648 | /* |
9649 | * this function is expected to parse integer in the range of [0, 2^31-1] from | |
9650 | * given file using scanf format string fmt. If actual parsed value is | |
9651 | * negative, the result might be indistinguishable from error | |
9652 | */ | |
9653 | static int parse_uint_from_file(const char *file, const char *fmt) | |
9654 | { | |
9655 | char buf[STRERR_BUFSIZE]; | |
9656 | int err, ret; | |
9657 | FILE *f; | |
9658 | ||
9659 | f = fopen(file, "r"); | |
9660 | if (!f) { | |
9661 | err = -errno; | |
9662 | pr_debug("failed to open '%s': %s\n", file, | |
9663 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
9664 | return err; | |
9665 | } | |
9666 | err = fscanf(f, fmt, &ret); | |
9667 | if (err != 1) { | |
9668 | err = err == EOF ? -EIO : -errno; | |
9669 | pr_debug("failed to parse '%s': %s\n", file, | |
9670 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
9671 | fclose(f); | |
9672 | return err; | |
9673 | } | |
9674 | fclose(f); | |
9675 | return ret; | |
9676 | } | |
9677 | ||
9678 | static int determine_kprobe_perf_type(void) | |
9679 | { | |
9680 | const char *file = "/sys/bus/event_source/devices/kprobe/type"; | |
9681 | ||
9682 | return parse_uint_from_file(file, "%d\n"); | |
9683 | } | |
9684 | ||
9685 | static int determine_uprobe_perf_type(void) | |
9686 | { | |
9687 | const char *file = "/sys/bus/event_source/devices/uprobe/type"; | |
9688 | ||
9689 | return parse_uint_from_file(file, "%d\n"); | |
9690 | } | |
9691 | ||
9692 | static int determine_kprobe_retprobe_bit(void) | |
9693 | { | |
9694 | const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe"; | |
9695 | ||
9696 | return parse_uint_from_file(file, "config:%d\n"); | |
9697 | } | |
9698 | ||
9699 | static int determine_uprobe_retprobe_bit(void) | |
9700 | { | |
9701 | const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe"; | |
9702 | ||
9703 | return parse_uint_from_file(file, "config:%d\n"); | |
9704 | } | |
9705 | ||
9706 | static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name, | |
9707 | uint64_t offset, int pid) | |
9708 | { | |
9709 | struct perf_event_attr attr = {}; | |
9710 | char errmsg[STRERR_BUFSIZE]; | |
9711 | int type, pfd, err; | |
9712 | ||
9713 | type = uprobe ? determine_uprobe_perf_type() | |
9714 | : determine_kprobe_perf_type(); | |
9715 | if (type < 0) { | |
be18010e KW |
9716 | pr_warn("failed to determine %s perf type: %s\n", |
9717 | uprobe ? "uprobe" : "kprobe", | |
9718 | libbpf_strerror_r(type, errmsg, sizeof(errmsg))); | |
b2650027 AN |
9719 | return type; |
9720 | } | |
9721 | if (retprobe) { | |
9722 | int bit = uprobe ? determine_uprobe_retprobe_bit() | |
9723 | : determine_kprobe_retprobe_bit(); | |
9724 | ||
9725 | if (bit < 0) { | |
be18010e KW |
9726 | pr_warn("failed to determine %s retprobe bit: %s\n", |
9727 | uprobe ? "uprobe" : "kprobe", | |
9728 | libbpf_strerror_r(bit, errmsg, sizeof(errmsg))); | |
b2650027 AN |
9729 | return bit; |
9730 | } | |
9731 | attr.config |= 1 << bit; | |
9732 | } | |
9733 | attr.size = sizeof(attr); | |
9734 | attr.type = type; | |
36db2a94 AN |
9735 | attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */ |
9736 | attr.config2 = offset; /* kprobe_addr or probe_offset */ | |
b2650027 AN |
9737 | |
9738 | /* pid filter is meaningful only for uprobes */ | |
9739 | pfd = syscall(__NR_perf_event_open, &attr, | |
9740 | pid < 0 ? -1 : pid /* pid */, | |
9741 | pid == -1 ? 0 : -1 /* cpu */, | |
9742 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
9743 | if (pfd < 0) { | |
9744 | err = -errno; | |
be18010e KW |
9745 | pr_warn("%s perf_event_open() failed: %s\n", |
9746 | uprobe ? "uprobe" : "kprobe", | |
9747 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
b2650027 AN |
9748 | return err; |
9749 | } | |
9750 | return pfd; | |
9751 | } | |
9752 | ||
9753 | struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog, | |
9754 | bool retprobe, | |
9755 | const char *func_name) | |
9756 | { | |
9757 | char errmsg[STRERR_BUFSIZE]; | |
9758 | struct bpf_link *link; | |
9759 | int pfd, err; | |
9760 | ||
9761 | pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name, | |
9762 | 0 /* offset */, -1 /* pid */); | |
9763 | if (pfd < 0) { | |
52109584 AN |
9764 | pr_warn("prog '%s': failed to create %s '%s' perf event: %s\n", |
9765 | prog->name, retprobe ? "kretprobe" : "kprobe", func_name, | |
be18010e | 9766 | libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); |
b2650027 AN |
9767 | return ERR_PTR(pfd); |
9768 | } | |
9769 | link = bpf_program__attach_perf_event(prog, pfd); | |
9770 | if (IS_ERR(link)) { | |
9771 | close(pfd); | |
9772 | err = PTR_ERR(link); | |
52109584 AN |
9773 | pr_warn("prog '%s': failed to attach to %s '%s': %s\n", |
9774 | prog->name, retprobe ? "kretprobe" : "kprobe", func_name, | |
be18010e | 9775 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
b2650027 AN |
9776 | return link; |
9777 | } | |
9778 | return link; | |
9779 | } | |
9780 | ||
d7a18ea7 AN |
9781 | static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec, |
9782 | struct bpf_program *prog) | |
9783 | { | |
9784 | const char *func_name; | |
9785 | bool retprobe; | |
9786 | ||
52109584 | 9787 | func_name = prog->sec_name + sec->len; |
d7a18ea7 AN |
9788 | retprobe = strcmp(sec->sec, "kretprobe/") == 0; |
9789 | ||
9790 | return bpf_program__attach_kprobe(prog, retprobe, func_name); | |
9791 | } | |
9792 | ||
b2650027 AN |
9793 | struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog, |
9794 | bool retprobe, pid_t pid, | |
9795 | const char *binary_path, | |
9796 | size_t func_offset) | |
9797 | { | |
9798 | char errmsg[STRERR_BUFSIZE]; | |
9799 | struct bpf_link *link; | |
9800 | int pfd, err; | |
9801 | ||
9802 | pfd = perf_event_open_probe(true /* uprobe */, retprobe, | |
9803 | binary_path, func_offset, pid); | |
9804 | if (pfd < 0) { | |
52109584 AN |
9805 | pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n", |
9806 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e KW |
9807 | binary_path, func_offset, |
9808 | libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
b2650027 AN |
9809 | return ERR_PTR(pfd); |
9810 | } | |
9811 | link = bpf_program__attach_perf_event(prog, pfd); | |
9812 | if (IS_ERR(link)) { | |
9813 | close(pfd); | |
9814 | err = PTR_ERR(link); | |
52109584 AN |
9815 | pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n", |
9816 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e KW |
9817 | binary_path, func_offset, |
9818 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
b2650027 AN |
9819 | return link; |
9820 | } | |
9821 | return link; | |
9822 | } | |
9823 | ||
f6de59c1 AN |
9824 | static int determine_tracepoint_id(const char *tp_category, |
9825 | const char *tp_name) | |
9826 | { | |
9827 | char file[PATH_MAX]; | |
9828 | int ret; | |
9829 | ||
9830 | ret = snprintf(file, sizeof(file), | |
9831 | "/sys/kernel/debug/tracing/events/%s/%s/id", | |
9832 | tp_category, tp_name); | |
9833 | if (ret < 0) | |
9834 | return -errno; | |
9835 | if (ret >= sizeof(file)) { | |
9836 | pr_debug("tracepoint %s/%s path is too long\n", | |
9837 | tp_category, tp_name); | |
9838 | return -E2BIG; | |
9839 | } | |
9840 | return parse_uint_from_file(file, "%d\n"); | |
9841 | } | |
9842 | ||
9843 | static int perf_event_open_tracepoint(const char *tp_category, | |
9844 | const char *tp_name) | |
9845 | { | |
9846 | struct perf_event_attr attr = {}; | |
9847 | char errmsg[STRERR_BUFSIZE]; | |
9848 | int tp_id, pfd, err; | |
9849 | ||
9850 | tp_id = determine_tracepoint_id(tp_category, tp_name); | |
9851 | if (tp_id < 0) { | |
be18010e KW |
9852 | pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n", |
9853 | tp_category, tp_name, | |
9854 | libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
9855 | return tp_id; |
9856 | } | |
9857 | ||
9858 | attr.type = PERF_TYPE_TRACEPOINT; | |
9859 | attr.size = sizeof(attr); | |
9860 | attr.config = tp_id; | |
9861 | ||
9862 | pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */, | |
9863 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
9864 | if (pfd < 0) { | |
9865 | err = -errno; | |
be18010e KW |
9866 | pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n", |
9867 | tp_category, tp_name, | |
9868 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
9869 | return err; |
9870 | } | |
9871 | return pfd; | |
9872 | } | |
9873 | ||
9874 | struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog, | |
9875 | const char *tp_category, | |
9876 | const char *tp_name) | |
9877 | { | |
9878 | char errmsg[STRERR_BUFSIZE]; | |
9879 | struct bpf_link *link; | |
9880 | int pfd, err; | |
9881 | ||
9882 | pfd = perf_event_open_tracepoint(tp_category, tp_name); | |
9883 | if (pfd < 0) { | |
52109584 AN |
9884 | pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n", |
9885 | prog->name, tp_category, tp_name, | |
be18010e | 9886 | libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); |
f6de59c1 AN |
9887 | return ERR_PTR(pfd); |
9888 | } | |
9889 | link = bpf_program__attach_perf_event(prog, pfd); | |
9890 | if (IS_ERR(link)) { | |
9891 | close(pfd); | |
9892 | err = PTR_ERR(link); | |
52109584 AN |
9893 | pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n", |
9894 | prog->name, tp_category, tp_name, | |
be18010e | 9895 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
f6de59c1 AN |
9896 | return link; |
9897 | } | |
9898 | return link; | |
9899 | } | |
9900 | ||
d7a18ea7 AN |
9901 | static struct bpf_link *attach_tp(const struct bpf_sec_def *sec, |
9902 | struct bpf_program *prog) | |
9903 | { | |
9904 | char *sec_name, *tp_cat, *tp_name; | |
9905 | struct bpf_link *link; | |
9906 | ||
52109584 | 9907 | sec_name = strdup(prog->sec_name); |
d7a18ea7 AN |
9908 | if (!sec_name) |
9909 | return ERR_PTR(-ENOMEM); | |
9910 | ||
9911 | /* extract "tp/<category>/<name>" */ | |
9912 | tp_cat = sec_name + sec->len; | |
9913 | tp_name = strchr(tp_cat, '/'); | |
9914 | if (!tp_name) { | |
9915 | link = ERR_PTR(-EINVAL); | |
9916 | goto out; | |
9917 | } | |
9918 | *tp_name = '\0'; | |
9919 | tp_name++; | |
9920 | ||
9921 | link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name); | |
9922 | out: | |
9923 | free(sec_name); | |
9924 | return link; | |
9925 | } | |
9926 | ||
84bf5e1f AN |
9927 | struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog, |
9928 | const char *tp_name) | |
9929 | { | |
9930 | char errmsg[STRERR_BUFSIZE]; | |
c016b68e | 9931 | struct bpf_link *link; |
84bf5e1f AN |
9932 | int prog_fd, pfd; |
9933 | ||
9934 | prog_fd = bpf_program__fd(prog); | |
9935 | if (prog_fd < 0) { | |
52109584 | 9936 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
84bf5e1f AN |
9937 | return ERR_PTR(-EINVAL); |
9938 | } | |
9939 | ||
d6958706 | 9940 | link = calloc(1, sizeof(*link)); |
84bf5e1f AN |
9941 | if (!link) |
9942 | return ERR_PTR(-ENOMEM); | |
c016b68e | 9943 | link->detach = &bpf_link__detach_fd; |
84bf5e1f AN |
9944 | |
9945 | pfd = bpf_raw_tracepoint_open(tp_name, prog_fd); | |
9946 | if (pfd < 0) { | |
9947 | pfd = -errno; | |
9948 | free(link); | |
52109584 AN |
9949 | pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n", |
9950 | prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
84bf5e1f AN |
9951 | return ERR_PTR(pfd); |
9952 | } | |
9953 | link->fd = pfd; | |
c016b68e | 9954 | return link; |
84bf5e1f AN |
9955 | } |
9956 | ||
d7a18ea7 AN |
9957 | static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec, |
9958 | struct bpf_program *prog) | |
9959 | { | |
52109584 | 9960 | const char *tp_name = prog->sec_name + sec->len; |
d7a18ea7 AN |
9961 | |
9962 | return bpf_program__attach_raw_tracepoint(prog, tp_name); | |
9963 | } | |
9964 | ||
1e092a03 KS |
9965 | /* Common logic for all BPF program types that attach to a btf_id */ |
9966 | static struct bpf_link *bpf_program__attach_btf_id(struct bpf_program *prog) | |
b8c54ea4 AS |
9967 | { |
9968 | char errmsg[STRERR_BUFSIZE]; | |
c016b68e | 9969 | struct bpf_link *link; |
b8c54ea4 AS |
9970 | int prog_fd, pfd; |
9971 | ||
9972 | prog_fd = bpf_program__fd(prog); | |
9973 | if (prog_fd < 0) { | |
52109584 | 9974 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
b8c54ea4 AS |
9975 | return ERR_PTR(-EINVAL); |
9976 | } | |
9977 | ||
d6958706 | 9978 | link = calloc(1, sizeof(*link)); |
b8c54ea4 AS |
9979 | if (!link) |
9980 | return ERR_PTR(-ENOMEM); | |
c016b68e | 9981 | link->detach = &bpf_link__detach_fd; |
b8c54ea4 AS |
9982 | |
9983 | pfd = bpf_raw_tracepoint_open(NULL, prog_fd); | |
9984 | if (pfd < 0) { | |
9985 | pfd = -errno; | |
9986 | free(link); | |
52109584 AN |
9987 | pr_warn("prog '%s': failed to attach: %s\n", |
9988 | prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
b8c54ea4 AS |
9989 | return ERR_PTR(pfd); |
9990 | } | |
9991 | link->fd = pfd; | |
9992 | return (struct bpf_link *)link; | |
9993 | } | |
9994 | ||
1e092a03 KS |
9995 | struct bpf_link *bpf_program__attach_trace(struct bpf_program *prog) |
9996 | { | |
9997 | return bpf_program__attach_btf_id(prog); | |
9998 | } | |
9999 | ||
10000 | struct bpf_link *bpf_program__attach_lsm(struct bpf_program *prog) | |
10001 | { | |
10002 | return bpf_program__attach_btf_id(prog); | |
10003 | } | |
10004 | ||
d7a18ea7 AN |
10005 | static struct bpf_link *attach_trace(const struct bpf_sec_def *sec, |
10006 | struct bpf_program *prog) | |
10007 | { | |
10008 | return bpf_program__attach_trace(prog); | |
10009 | } | |
10010 | ||
1e092a03 KS |
10011 | static struct bpf_link *attach_lsm(const struct bpf_sec_def *sec, |
10012 | struct bpf_program *prog) | |
10013 | { | |
10014 | return bpf_program__attach_lsm(prog); | |
10015 | } | |
10016 | ||
c09add2f YS |
10017 | static struct bpf_link *attach_iter(const struct bpf_sec_def *sec, |
10018 | struct bpf_program *prog) | |
10019 | { | |
10020 | return bpf_program__attach_iter(prog, NULL); | |
10021 | } | |
10022 | ||
d60d81ac | 10023 | static struct bpf_link * |
a5359091 | 10024 | bpf_program__attach_fd(struct bpf_program *prog, int target_fd, int btf_id, |
d60d81ac | 10025 | const char *target_name) |
cc4f864b | 10026 | { |
a5359091 THJ |
10027 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts, |
10028 | .target_btf_id = btf_id); | |
cc4f864b AN |
10029 | enum bpf_attach_type attach_type; |
10030 | char errmsg[STRERR_BUFSIZE]; | |
10031 | struct bpf_link *link; | |
10032 | int prog_fd, link_fd; | |
10033 | ||
10034 | prog_fd = bpf_program__fd(prog); | |
10035 | if (prog_fd < 0) { | |
52109584 | 10036 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
cc4f864b AN |
10037 | return ERR_PTR(-EINVAL); |
10038 | } | |
10039 | ||
10040 | link = calloc(1, sizeof(*link)); | |
10041 | if (!link) | |
10042 | return ERR_PTR(-ENOMEM); | |
10043 | link->detach = &bpf_link__detach_fd; | |
10044 | ||
10045 | attach_type = bpf_program__get_expected_attach_type(prog); | |
a5359091 | 10046 | link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts); |
cc4f864b AN |
10047 | if (link_fd < 0) { |
10048 | link_fd = -errno; | |
10049 | free(link); | |
52109584 AN |
10050 | pr_warn("prog '%s': failed to attach to %s: %s\n", |
10051 | prog->name, target_name, | |
cc4f864b AN |
10052 | libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); |
10053 | return ERR_PTR(link_fd); | |
10054 | } | |
10055 | link->fd = link_fd; | |
10056 | return link; | |
10057 | } | |
10058 | ||
d60d81ac JS |
10059 | struct bpf_link * |
10060 | bpf_program__attach_cgroup(struct bpf_program *prog, int cgroup_fd) | |
10061 | { | |
a5359091 | 10062 | return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup"); |
d60d81ac JS |
10063 | } |
10064 | ||
10065 | struct bpf_link * | |
10066 | bpf_program__attach_netns(struct bpf_program *prog, int netns_fd) | |
10067 | { | |
a5359091 | 10068 | return bpf_program__attach_fd(prog, netns_fd, 0, "netns"); |
d60d81ac JS |
10069 | } |
10070 | ||
dc8698ca AN |
10071 | struct bpf_link *bpf_program__attach_xdp(struct bpf_program *prog, int ifindex) |
10072 | { | |
10073 | /* target_fd/target_ifindex use the same field in LINK_CREATE */ | |
a5359091 THJ |
10074 | return bpf_program__attach_fd(prog, ifindex, 0, "xdp"); |
10075 | } | |
10076 | ||
10077 | struct bpf_link *bpf_program__attach_freplace(struct bpf_program *prog, | |
10078 | int target_fd, | |
10079 | const char *attach_func_name) | |
10080 | { | |
10081 | int btf_id; | |
10082 | ||
10083 | if (!!target_fd != !!attach_func_name) { | |
10084 | pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n", | |
10085 | prog->name); | |
10086 | return ERR_PTR(-EINVAL); | |
10087 | } | |
10088 | ||
10089 | if (prog->type != BPF_PROG_TYPE_EXT) { | |
10090 | pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace", | |
10091 | prog->name); | |
10092 | return ERR_PTR(-EINVAL); | |
10093 | } | |
10094 | ||
10095 | if (target_fd) { | |
10096 | btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd); | |
10097 | if (btf_id < 0) | |
10098 | return ERR_PTR(btf_id); | |
10099 | ||
10100 | return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace"); | |
10101 | } else { | |
10102 | /* no target, so use raw_tracepoint_open for compatibility | |
10103 | * with old kernels | |
10104 | */ | |
10105 | return bpf_program__attach_trace(prog); | |
10106 | } | |
dc8698ca AN |
10107 | } |
10108 | ||
c09add2f YS |
10109 | struct bpf_link * |
10110 | bpf_program__attach_iter(struct bpf_program *prog, | |
10111 | const struct bpf_iter_attach_opts *opts) | |
10112 | { | |
cd31039a | 10113 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); |
c09add2f YS |
10114 | char errmsg[STRERR_BUFSIZE]; |
10115 | struct bpf_link *link; | |
10116 | int prog_fd, link_fd; | |
cd31039a | 10117 | __u32 target_fd = 0; |
c09add2f YS |
10118 | |
10119 | if (!OPTS_VALID(opts, bpf_iter_attach_opts)) | |
10120 | return ERR_PTR(-EINVAL); | |
10121 | ||
74fc097d YS |
10122 | link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0); |
10123 | link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0); | |
cd31039a | 10124 | |
c09add2f YS |
10125 | prog_fd = bpf_program__fd(prog); |
10126 | if (prog_fd < 0) { | |
52109584 | 10127 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
c09add2f YS |
10128 | return ERR_PTR(-EINVAL); |
10129 | } | |
10130 | ||
10131 | link = calloc(1, sizeof(*link)); | |
10132 | if (!link) | |
10133 | return ERR_PTR(-ENOMEM); | |
10134 | link->detach = &bpf_link__detach_fd; | |
10135 | ||
cd31039a YS |
10136 | link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER, |
10137 | &link_create_opts); | |
c09add2f YS |
10138 | if (link_fd < 0) { |
10139 | link_fd = -errno; | |
10140 | free(link); | |
52109584 AN |
10141 | pr_warn("prog '%s': failed to attach to iterator: %s\n", |
10142 | prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); | |
c09add2f YS |
10143 | return ERR_PTR(link_fd); |
10144 | } | |
10145 | link->fd = link_fd; | |
10146 | return link; | |
10147 | } | |
10148 | ||
d7a18ea7 AN |
10149 | struct bpf_link *bpf_program__attach(struct bpf_program *prog) |
10150 | { | |
10151 | const struct bpf_sec_def *sec_def; | |
10152 | ||
52109584 | 10153 | sec_def = find_sec_def(prog->sec_name); |
d7a18ea7 AN |
10154 | if (!sec_def || !sec_def->attach_fn) |
10155 | return ERR_PTR(-ESRCH); | |
10156 | ||
10157 | return sec_def->attach_fn(sec_def, prog); | |
10158 | } | |
10159 | ||
590a0088 MKL |
10160 | static int bpf_link__detach_struct_ops(struct bpf_link *link) |
10161 | { | |
590a0088 MKL |
10162 | __u32 zero = 0; |
10163 | ||
c016b68e | 10164 | if (bpf_map_delete_elem(link->fd, &zero)) |
590a0088 MKL |
10165 | return -errno; |
10166 | ||
10167 | return 0; | |
10168 | } | |
10169 | ||
10170 | struct bpf_link *bpf_map__attach_struct_ops(struct bpf_map *map) | |
10171 | { | |
10172 | struct bpf_struct_ops *st_ops; | |
c016b68e | 10173 | struct bpf_link *link; |
590a0088 MKL |
10174 | __u32 i, zero = 0; |
10175 | int err; | |
10176 | ||
10177 | if (!bpf_map__is_struct_ops(map) || map->fd == -1) | |
10178 | return ERR_PTR(-EINVAL); | |
10179 | ||
10180 | link = calloc(1, sizeof(*link)); | |
10181 | if (!link) | |
10182 | return ERR_PTR(-EINVAL); | |
10183 | ||
10184 | st_ops = map->st_ops; | |
10185 | for (i = 0; i < btf_vlen(st_ops->type); i++) { | |
10186 | struct bpf_program *prog = st_ops->progs[i]; | |
10187 | void *kern_data; | |
10188 | int prog_fd; | |
10189 | ||
10190 | if (!prog) | |
10191 | continue; | |
10192 | ||
10193 | prog_fd = bpf_program__fd(prog); | |
10194 | kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i]; | |
10195 | *(unsigned long *)kern_data = prog_fd; | |
10196 | } | |
10197 | ||
10198 | err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0); | |
10199 | if (err) { | |
10200 | err = -errno; | |
10201 | free(link); | |
10202 | return ERR_PTR(err); | |
10203 | } | |
10204 | ||
c016b68e | 10205 | link->detach = bpf_link__detach_struct_ops; |
590a0088 MKL |
10206 | link->fd = map->fd; |
10207 | ||
c016b68e | 10208 | return link; |
590a0088 MKL |
10209 | } |
10210 | ||
d0cabbb0 | 10211 | enum bpf_perf_event_ret |
3dca2115 DB |
10212 | bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size, |
10213 | void **copy_mem, size_t *copy_size, | |
10214 | bpf_perf_event_print_t fn, void *private_data) | |
d0cabbb0 | 10215 | { |
3dca2115 | 10216 | struct perf_event_mmap_page *header = mmap_mem; |
a64af0ef | 10217 | __u64 data_head = ring_buffer_read_head(header); |
d0cabbb0 | 10218 | __u64 data_tail = header->data_tail; |
3dca2115 DB |
10219 | void *base = ((__u8 *)header) + page_size; |
10220 | int ret = LIBBPF_PERF_EVENT_CONT; | |
10221 | struct perf_event_header *ehdr; | |
10222 | size_t ehdr_size; | |
10223 | ||
10224 | while (data_head != data_tail) { | |
10225 | ehdr = base + (data_tail & (mmap_size - 1)); | |
10226 | ehdr_size = ehdr->size; | |
10227 | ||
10228 | if (((void *)ehdr) + ehdr_size > base + mmap_size) { | |
10229 | void *copy_start = ehdr; | |
10230 | size_t len_first = base + mmap_size - copy_start; | |
10231 | size_t len_secnd = ehdr_size - len_first; | |
10232 | ||
10233 | if (*copy_size < ehdr_size) { | |
10234 | free(*copy_mem); | |
10235 | *copy_mem = malloc(ehdr_size); | |
10236 | if (!*copy_mem) { | |
10237 | *copy_size = 0; | |
d0cabbb0 JK |
10238 | ret = LIBBPF_PERF_EVENT_ERROR; |
10239 | break; | |
10240 | } | |
3dca2115 | 10241 | *copy_size = ehdr_size; |
d0cabbb0 JK |
10242 | } |
10243 | ||
3dca2115 DB |
10244 | memcpy(*copy_mem, copy_start, len_first); |
10245 | memcpy(*copy_mem + len_first, base, len_secnd); | |
10246 | ehdr = *copy_mem; | |
d0cabbb0 JK |
10247 | } |
10248 | ||
3dca2115 DB |
10249 | ret = fn(ehdr, private_data); |
10250 | data_tail += ehdr_size; | |
d0cabbb0 JK |
10251 | if (ret != LIBBPF_PERF_EVENT_CONT) |
10252 | break; | |
d0cabbb0 JK |
10253 | } |
10254 | ||
a64af0ef | 10255 | ring_buffer_write_tail(header, data_tail); |
d0cabbb0 JK |
10256 | return ret; |
10257 | } | |
34be1646 | 10258 | |
fb84b822 AN |
10259 | struct perf_buffer; |
10260 | ||
10261 | struct perf_buffer_params { | |
10262 | struct perf_event_attr *attr; | |
10263 | /* if event_cb is specified, it takes precendence */ | |
10264 | perf_buffer_event_fn event_cb; | |
10265 | /* sample_cb and lost_cb are higher-level common-case callbacks */ | |
10266 | perf_buffer_sample_fn sample_cb; | |
10267 | perf_buffer_lost_fn lost_cb; | |
10268 | void *ctx; | |
10269 | int cpu_cnt; | |
10270 | int *cpus; | |
10271 | int *map_keys; | |
10272 | }; | |
10273 | ||
10274 | struct perf_cpu_buf { | |
10275 | struct perf_buffer *pb; | |
10276 | void *base; /* mmap()'ed memory */ | |
10277 | void *buf; /* for reconstructing segmented data */ | |
10278 | size_t buf_size; | |
10279 | int fd; | |
10280 | int cpu; | |
10281 | int map_key; | |
10282 | }; | |
10283 | ||
10284 | struct perf_buffer { | |
10285 | perf_buffer_event_fn event_cb; | |
10286 | perf_buffer_sample_fn sample_cb; | |
10287 | perf_buffer_lost_fn lost_cb; | |
10288 | void *ctx; /* passed into callbacks */ | |
10289 | ||
10290 | size_t page_size; | |
10291 | size_t mmap_size; | |
10292 | struct perf_cpu_buf **cpu_bufs; | |
10293 | struct epoll_event *events; | |
783b8f01 | 10294 | int cpu_cnt; /* number of allocated CPU buffers */ |
fb84b822 AN |
10295 | int epoll_fd; /* perf event FD */ |
10296 | int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */ | |
10297 | }; | |
10298 | ||
10299 | static void perf_buffer__free_cpu_buf(struct perf_buffer *pb, | |
10300 | struct perf_cpu_buf *cpu_buf) | |
10301 | { | |
10302 | if (!cpu_buf) | |
10303 | return; | |
10304 | if (cpu_buf->base && | |
10305 | munmap(cpu_buf->base, pb->mmap_size + pb->page_size)) | |
be18010e | 10306 | pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu); |
fb84b822 AN |
10307 | if (cpu_buf->fd >= 0) { |
10308 | ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0); | |
10309 | close(cpu_buf->fd); | |
10310 | } | |
10311 | free(cpu_buf->buf); | |
10312 | free(cpu_buf); | |
10313 | } | |
10314 | ||
10315 | void perf_buffer__free(struct perf_buffer *pb) | |
10316 | { | |
10317 | int i; | |
10318 | ||
50450fc7 | 10319 | if (IS_ERR_OR_NULL(pb)) |
fb84b822 AN |
10320 | return; |
10321 | if (pb->cpu_bufs) { | |
601b05ca | 10322 | for (i = 0; i < pb->cpu_cnt; i++) { |
fb84b822 AN |
10323 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; |
10324 | ||
601b05ca EC |
10325 | if (!cpu_buf) |
10326 | continue; | |
10327 | ||
fb84b822 AN |
10328 | bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key); |
10329 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
10330 | } | |
10331 | free(pb->cpu_bufs); | |
10332 | } | |
10333 | if (pb->epoll_fd >= 0) | |
10334 | close(pb->epoll_fd); | |
10335 | free(pb->events); | |
10336 | free(pb); | |
10337 | } | |
10338 | ||
10339 | static struct perf_cpu_buf * | |
10340 | perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr, | |
10341 | int cpu, int map_key) | |
10342 | { | |
10343 | struct perf_cpu_buf *cpu_buf; | |
10344 | char msg[STRERR_BUFSIZE]; | |
10345 | int err; | |
10346 | ||
10347 | cpu_buf = calloc(1, sizeof(*cpu_buf)); | |
10348 | if (!cpu_buf) | |
10349 | return ERR_PTR(-ENOMEM); | |
10350 | ||
10351 | cpu_buf->pb = pb; | |
10352 | cpu_buf->cpu = cpu; | |
10353 | cpu_buf->map_key = map_key; | |
10354 | ||
10355 | cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu, | |
10356 | -1, PERF_FLAG_FD_CLOEXEC); | |
10357 | if (cpu_buf->fd < 0) { | |
10358 | err = -errno; | |
be18010e KW |
10359 | pr_warn("failed to open perf buffer event on cpu #%d: %s\n", |
10360 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10361 | goto error; |
10362 | } | |
10363 | ||
10364 | cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size, | |
10365 | PROT_READ | PROT_WRITE, MAP_SHARED, | |
10366 | cpu_buf->fd, 0); | |
10367 | if (cpu_buf->base == MAP_FAILED) { | |
10368 | cpu_buf->base = NULL; | |
10369 | err = -errno; | |
be18010e KW |
10370 | pr_warn("failed to mmap perf buffer on cpu #%d: %s\n", |
10371 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10372 | goto error; |
10373 | } | |
10374 | ||
10375 | if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
10376 | err = -errno; | |
be18010e KW |
10377 | pr_warn("failed to enable perf buffer event on cpu #%d: %s\n", |
10378 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10379 | goto error; |
10380 | } | |
10381 | ||
10382 | return cpu_buf; | |
10383 | ||
10384 | error: | |
10385 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
10386 | return (struct perf_cpu_buf *)ERR_PTR(err); | |
10387 | } | |
10388 | ||
10389 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
10390 | struct perf_buffer_params *p); | |
10391 | ||
10392 | struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt, | |
10393 | const struct perf_buffer_opts *opts) | |
10394 | { | |
10395 | struct perf_buffer_params p = {}; | |
4be6e05c ACM |
10396 | struct perf_event_attr attr = { 0, }; |
10397 | ||
65bb2e0f | 10398 | attr.config = PERF_COUNT_SW_BPF_OUTPUT; |
4be6e05c ACM |
10399 | attr.type = PERF_TYPE_SOFTWARE; |
10400 | attr.sample_type = PERF_SAMPLE_RAW; | |
10401 | attr.sample_period = 1; | |
10402 | attr.wakeup_events = 1; | |
fb84b822 AN |
10403 | |
10404 | p.attr = &attr; | |
10405 | p.sample_cb = opts ? opts->sample_cb : NULL; | |
10406 | p.lost_cb = opts ? opts->lost_cb : NULL; | |
10407 | p.ctx = opts ? opts->ctx : NULL; | |
10408 | ||
10409 | return __perf_buffer__new(map_fd, page_cnt, &p); | |
10410 | } | |
10411 | ||
10412 | struct perf_buffer * | |
10413 | perf_buffer__new_raw(int map_fd, size_t page_cnt, | |
10414 | const struct perf_buffer_raw_opts *opts) | |
10415 | { | |
10416 | struct perf_buffer_params p = {}; | |
10417 | ||
10418 | p.attr = opts->attr; | |
10419 | p.event_cb = opts->event_cb; | |
10420 | p.ctx = opts->ctx; | |
10421 | p.cpu_cnt = opts->cpu_cnt; | |
10422 | p.cpus = opts->cpus; | |
10423 | p.map_keys = opts->map_keys; | |
10424 | ||
10425 | return __perf_buffer__new(map_fd, page_cnt, &p); | |
10426 | } | |
10427 | ||
10428 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
10429 | struct perf_buffer_params *p) | |
10430 | { | |
783b8f01 | 10431 | const char *online_cpus_file = "/sys/devices/system/cpu/online"; |
0e289487 | 10432 | struct bpf_map_info map; |
fb84b822 AN |
10433 | char msg[STRERR_BUFSIZE]; |
10434 | struct perf_buffer *pb; | |
783b8f01 | 10435 | bool *online = NULL; |
fb84b822 | 10436 | __u32 map_info_len; |
783b8f01 | 10437 | int err, i, j, n; |
fb84b822 AN |
10438 | |
10439 | if (page_cnt & (page_cnt - 1)) { | |
be18010e KW |
10440 | pr_warn("page count should be power of two, but is %zu\n", |
10441 | page_cnt); | |
fb84b822 AN |
10442 | return ERR_PTR(-EINVAL); |
10443 | } | |
10444 | ||
0e289487 AN |
10445 | /* best-effort sanity checks */ |
10446 | memset(&map, 0, sizeof(map)); | |
fb84b822 AN |
10447 | map_info_len = sizeof(map); |
10448 | err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len); | |
10449 | if (err) { | |
10450 | err = -errno; | |
0e289487 AN |
10451 | /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return |
10452 | * -EBADFD, -EFAULT, or -E2BIG on real error | |
10453 | */ | |
10454 | if (err != -EINVAL) { | |
10455 | pr_warn("failed to get map info for map FD %d: %s\n", | |
10456 | map_fd, libbpf_strerror_r(err, msg, sizeof(msg))); | |
10457 | return ERR_PTR(err); | |
10458 | } | |
10459 | pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n", | |
10460 | map_fd); | |
10461 | } else { | |
10462 | if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) { | |
10463 | pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n", | |
10464 | map.name); | |
10465 | return ERR_PTR(-EINVAL); | |
10466 | } | |
fb84b822 AN |
10467 | } |
10468 | ||
10469 | pb = calloc(1, sizeof(*pb)); | |
10470 | if (!pb) | |
10471 | return ERR_PTR(-ENOMEM); | |
10472 | ||
10473 | pb->event_cb = p->event_cb; | |
10474 | pb->sample_cb = p->sample_cb; | |
10475 | pb->lost_cb = p->lost_cb; | |
10476 | pb->ctx = p->ctx; | |
10477 | ||
10478 | pb->page_size = getpagesize(); | |
10479 | pb->mmap_size = pb->page_size * page_cnt; | |
10480 | pb->map_fd = map_fd; | |
10481 | ||
10482 | pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
10483 | if (pb->epoll_fd < 0) { | |
10484 | err = -errno; | |
be18010e KW |
10485 | pr_warn("failed to create epoll instance: %s\n", |
10486 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10487 | goto error; |
10488 | } | |
10489 | ||
10490 | if (p->cpu_cnt > 0) { | |
10491 | pb->cpu_cnt = p->cpu_cnt; | |
10492 | } else { | |
10493 | pb->cpu_cnt = libbpf_num_possible_cpus(); | |
10494 | if (pb->cpu_cnt < 0) { | |
10495 | err = pb->cpu_cnt; | |
10496 | goto error; | |
10497 | } | |
0e289487 | 10498 | if (map.max_entries && map.max_entries < pb->cpu_cnt) |
fb84b822 AN |
10499 | pb->cpu_cnt = map.max_entries; |
10500 | } | |
10501 | ||
10502 | pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events)); | |
10503 | if (!pb->events) { | |
10504 | err = -ENOMEM; | |
be18010e | 10505 | pr_warn("failed to allocate events: out of memory\n"); |
fb84b822 AN |
10506 | goto error; |
10507 | } | |
10508 | pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs)); | |
10509 | if (!pb->cpu_bufs) { | |
10510 | err = -ENOMEM; | |
be18010e | 10511 | pr_warn("failed to allocate buffers: out of memory\n"); |
fb84b822 AN |
10512 | goto error; |
10513 | } | |
10514 | ||
783b8f01 AN |
10515 | err = parse_cpu_mask_file(online_cpus_file, &online, &n); |
10516 | if (err) { | |
10517 | pr_warn("failed to get online CPU mask: %d\n", err); | |
10518 | goto error; | |
10519 | } | |
10520 | ||
10521 | for (i = 0, j = 0; i < pb->cpu_cnt; i++) { | |
fb84b822 AN |
10522 | struct perf_cpu_buf *cpu_buf; |
10523 | int cpu, map_key; | |
10524 | ||
10525 | cpu = p->cpu_cnt > 0 ? p->cpus[i] : i; | |
10526 | map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i; | |
10527 | ||
783b8f01 AN |
10528 | /* in case user didn't explicitly requested particular CPUs to |
10529 | * be attached to, skip offline/not present CPUs | |
10530 | */ | |
10531 | if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu])) | |
10532 | continue; | |
10533 | ||
fb84b822 AN |
10534 | cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key); |
10535 | if (IS_ERR(cpu_buf)) { | |
10536 | err = PTR_ERR(cpu_buf); | |
10537 | goto error; | |
10538 | } | |
10539 | ||
783b8f01 | 10540 | pb->cpu_bufs[j] = cpu_buf; |
fb84b822 AN |
10541 | |
10542 | err = bpf_map_update_elem(pb->map_fd, &map_key, | |
10543 | &cpu_buf->fd, 0); | |
10544 | if (err) { | |
10545 | err = -errno; | |
be18010e KW |
10546 | pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n", |
10547 | cpu, map_key, cpu_buf->fd, | |
10548 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10549 | goto error; |
10550 | } | |
10551 | ||
783b8f01 AN |
10552 | pb->events[j].events = EPOLLIN; |
10553 | pb->events[j].data.ptr = cpu_buf; | |
fb84b822 | 10554 | if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd, |
783b8f01 | 10555 | &pb->events[j]) < 0) { |
fb84b822 | 10556 | err = -errno; |
be18010e KW |
10557 | pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n", |
10558 | cpu, cpu_buf->fd, | |
10559 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10560 | goto error; |
10561 | } | |
783b8f01 | 10562 | j++; |
fb84b822 | 10563 | } |
783b8f01 AN |
10564 | pb->cpu_cnt = j; |
10565 | free(online); | |
fb84b822 AN |
10566 | |
10567 | return pb; | |
10568 | ||
10569 | error: | |
783b8f01 | 10570 | free(online); |
fb84b822 AN |
10571 | if (pb) |
10572 | perf_buffer__free(pb); | |
10573 | return ERR_PTR(err); | |
10574 | } | |
10575 | ||
10576 | struct perf_sample_raw { | |
10577 | struct perf_event_header header; | |
10578 | uint32_t size; | |
385bbf7b | 10579 | char data[]; |
fb84b822 AN |
10580 | }; |
10581 | ||
10582 | struct perf_sample_lost { | |
10583 | struct perf_event_header header; | |
10584 | uint64_t id; | |
10585 | uint64_t lost; | |
10586 | uint64_t sample_id; | |
10587 | }; | |
10588 | ||
10589 | static enum bpf_perf_event_ret | |
10590 | perf_buffer__process_record(struct perf_event_header *e, void *ctx) | |
10591 | { | |
10592 | struct perf_cpu_buf *cpu_buf = ctx; | |
10593 | struct perf_buffer *pb = cpu_buf->pb; | |
10594 | void *data = e; | |
10595 | ||
10596 | /* user wants full control over parsing perf event */ | |
10597 | if (pb->event_cb) | |
10598 | return pb->event_cb(pb->ctx, cpu_buf->cpu, e); | |
10599 | ||
10600 | switch (e->type) { | |
10601 | case PERF_RECORD_SAMPLE: { | |
10602 | struct perf_sample_raw *s = data; | |
10603 | ||
10604 | if (pb->sample_cb) | |
10605 | pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size); | |
10606 | break; | |
10607 | } | |
10608 | case PERF_RECORD_LOST: { | |
10609 | struct perf_sample_lost *s = data; | |
10610 | ||
10611 | if (pb->lost_cb) | |
10612 | pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost); | |
10613 | break; | |
10614 | } | |
10615 | default: | |
be18010e | 10616 | pr_warn("unknown perf sample type %d\n", e->type); |
fb84b822 AN |
10617 | return LIBBPF_PERF_EVENT_ERROR; |
10618 | } | |
10619 | return LIBBPF_PERF_EVENT_CONT; | |
10620 | } | |
10621 | ||
10622 | static int perf_buffer__process_records(struct perf_buffer *pb, | |
10623 | struct perf_cpu_buf *cpu_buf) | |
10624 | { | |
10625 | enum bpf_perf_event_ret ret; | |
10626 | ||
10627 | ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size, | |
10628 | pb->page_size, &cpu_buf->buf, | |
10629 | &cpu_buf->buf_size, | |
10630 | perf_buffer__process_record, cpu_buf); | |
10631 | if (ret != LIBBPF_PERF_EVENT_CONT) | |
10632 | return ret; | |
10633 | return 0; | |
10634 | } | |
10635 | ||
dca5612f AN |
10636 | int perf_buffer__epoll_fd(const struct perf_buffer *pb) |
10637 | { | |
10638 | return pb->epoll_fd; | |
10639 | } | |
10640 | ||
fb84b822 AN |
10641 | int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms) |
10642 | { | |
10643 | int i, cnt, err; | |
10644 | ||
10645 | cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms); | |
10646 | for (i = 0; i < cnt; i++) { | |
10647 | struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr; | |
10648 | ||
10649 | err = perf_buffer__process_records(pb, cpu_buf); | |
10650 | if (err) { | |
be18010e | 10651 | pr_warn("error while processing records: %d\n", err); |
fb84b822 AN |
10652 | return err; |
10653 | } | |
10654 | } | |
10655 | return cnt < 0 ? -errno : cnt; | |
10656 | } | |
10657 | ||
dca5612f AN |
10658 | /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer |
10659 | * manager. | |
10660 | */ | |
10661 | size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb) | |
10662 | { | |
10663 | return pb->cpu_cnt; | |
10664 | } | |
10665 | ||
10666 | /* | |
10667 | * Return perf_event FD of a ring buffer in *buf_idx* slot of | |
10668 | * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using | |
10669 | * select()/poll()/epoll() Linux syscalls. | |
10670 | */ | |
10671 | int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx) | |
10672 | { | |
10673 | struct perf_cpu_buf *cpu_buf; | |
10674 | ||
10675 | if (buf_idx >= pb->cpu_cnt) | |
10676 | return -EINVAL; | |
10677 | ||
10678 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
10679 | if (!cpu_buf) | |
10680 | return -ENOENT; | |
10681 | ||
10682 | return cpu_buf->fd; | |
10683 | } | |
10684 | ||
10685 | /* | |
10686 | * Consume data from perf ring buffer corresponding to slot *buf_idx* in | |
10687 | * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to | |
10688 | * consume, do nothing and return success. | |
10689 | * Returns: | |
10690 | * - 0 on success; | |
10691 | * - <0 on failure. | |
10692 | */ | |
10693 | int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx) | |
10694 | { | |
10695 | struct perf_cpu_buf *cpu_buf; | |
10696 | ||
10697 | if (buf_idx >= pb->cpu_cnt) | |
10698 | return -EINVAL; | |
10699 | ||
10700 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
10701 | if (!cpu_buf) | |
10702 | return -ENOENT; | |
10703 | ||
10704 | return perf_buffer__process_records(pb, cpu_buf); | |
10705 | } | |
10706 | ||
272d51af EC |
10707 | int perf_buffer__consume(struct perf_buffer *pb) |
10708 | { | |
10709 | int i, err; | |
10710 | ||
10711 | for (i = 0; i < pb->cpu_cnt; i++) { | |
10712 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; | |
10713 | ||
10714 | if (!cpu_buf) | |
10715 | continue; | |
10716 | ||
10717 | err = perf_buffer__process_records(pb, cpu_buf); | |
10718 | if (err) { | |
dca5612f | 10719 | pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err); |
272d51af EC |
10720 | return err; |
10721 | } | |
10722 | } | |
10723 | return 0; | |
10724 | } | |
10725 | ||
34be1646 SL |
10726 | struct bpf_prog_info_array_desc { |
10727 | int array_offset; /* e.g. offset of jited_prog_insns */ | |
10728 | int count_offset; /* e.g. offset of jited_prog_len */ | |
10729 | int size_offset; /* > 0: offset of rec size, | |
10730 | * < 0: fix size of -size_offset | |
10731 | */ | |
10732 | }; | |
10733 | ||
10734 | static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = { | |
10735 | [BPF_PROG_INFO_JITED_INSNS] = { | |
10736 | offsetof(struct bpf_prog_info, jited_prog_insns), | |
10737 | offsetof(struct bpf_prog_info, jited_prog_len), | |
10738 | -1, | |
10739 | }, | |
10740 | [BPF_PROG_INFO_XLATED_INSNS] = { | |
10741 | offsetof(struct bpf_prog_info, xlated_prog_insns), | |
10742 | offsetof(struct bpf_prog_info, xlated_prog_len), | |
10743 | -1, | |
10744 | }, | |
10745 | [BPF_PROG_INFO_MAP_IDS] = { | |
10746 | offsetof(struct bpf_prog_info, map_ids), | |
10747 | offsetof(struct bpf_prog_info, nr_map_ids), | |
10748 | -(int)sizeof(__u32), | |
10749 | }, | |
10750 | [BPF_PROG_INFO_JITED_KSYMS] = { | |
10751 | offsetof(struct bpf_prog_info, jited_ksyms), | |
10752 | offsetof(struct bpf_prog_info, nr_jited_ksyms), | |
10753 | -(int)sizeof(__u64), | |
10754 | }, | |
10755 | [BPF_PROG_INFO_JITED_FUNC_LENS] = { | |
10756 | offsetof(struct bpf_prog_info, jited_func_lens), | |
10757 | offsetof(struct bpf_prog_info, nr_jited_func_lens), | |
10758 | -(int)sizeof(__u32), | |
10759 | }, | |
10760 | [BPF_PROG_INFO_FUNC_INFO] = { | |
10761 | offsetof(struct bpf_prog_info, func_info), | |
10762 | offsetof(struct bpf_prog_info, nr_func_info), | |
10763 | offsetof(struct bpf_prog_info, func_info_rec_size), | |
10764 | }, | |
10765 | [BPF_PROG_INFO_LINE_INFO] = { | |
10766 | offsetof(struct bpf_prog_info, line_info), | |
10767 | offsetof(struct bpf_prog_info, nr_line_info), | |
10768 | offsetof(struct bpf_prog_info, line_info_rec_size), | |
10769 | }, | |
10770 | [BPF_PROG_INFO_JITED_LINE_INFO] = { | |
10771 | offsetof(struct bpf_prog_info, jited_line_info), | |
10772 | offsetof(struct bpf_prog_info, nr_jited_line_info), | |
10773 | offsetof(struct bpf_prog_info, jited_line_info_rec_size), | |
10774 | }, | |
10775 | [BPF_PROG_INFO_PROG_TAGS] = { | |
10776 | offsetof(struct bpf_prog_info, prog_tags), | |
10777 | offsetof(struct bpf_prog_info, nr_prog_tags), | |
10778 | -(int)sizeof(__u8) * BPF_TAG_SIZE, | |
10779 | }, | |
10780 | ||
10781 | }; | |
10782 | ||
8983b731 AN |
10783 | static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, |
10784 | int offset) | |
34be1646 SL |
10785 | { |
10786 | __u32 *array = (__u32 *)info; | |
10787 | ||
10788 | if (offset >= 0) | |
10789 | return array[offset / sizeof(__u32)]; | |
10790 | return -(int)offset; | |
10791 | } | |
10792 | ||
8983b731 AN |
10793 | static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, |
10794 | int offset) | |
34be1646 SL |
10795 | { |
10796 | __u64 *array = (__u64 *)info; | |
10797 | ||
10798 | if (offset >= 0) | |
10799 | return array[offset / sizeof(__u64)]; | |
10800 | return -(int)offset; | |
10801 | } | |
10802 | ||
10803 | static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset, | |
10804 | __u32 val) | |
10805 | { | |
10806 | __u32 *array = (__u32 *)info; | |
10807 | ||
10808 | if (offset >= 0) | |
10809 | array[offset / sizeof(__u32)] = val; | |
10810 | } | |
10811 | ||
10812 | static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset, | |
10813 | __u64 val) | |
10814 | { | |
10815 | __u64 *array = (__u64 *)info; | |
10816 | ||
10817 | if (offset >= 0) | |
10818 | array[offset / sizeof(__u64)] = val; | |
10819 | } | |
10820 | ||
10821 | struct bpf_prog_info_linear * | |
10822 | bpf_program__get_prog_info_linear(int fd, __u64 arrays) | |
10823 | { | |
10824 | struct bpf_prog_info_linear *info_linear; | |
10825 | struct bpf_prog_info info = {}; | |
10826 | __u32 info_len = sizeof(info); | |
10827 | __u32 data_len = 0; | |
10828 | int i, err; | |
10829 | void *ptr; | |
10830 | ||
10831 | if (arrays >> BPF_PROG_INFO_LAST_ARRAY) | |
10832 | return ERR_PTR(-EINVAL); | |
10833 | ||
10834 | /* step 1: get array dimensions */ | |
10835 | err = bpf_obj_get_info_by_fd(fd, &info, &info_len); | |
10836 | if (err) { | |
10837 | pr_debug("can't get prog info: %s", strerror(errno)); | |
10838 | return ERR_PTR(-EFAULT); | |
10839 | } | |
10840 | ||
10841 | /* step 2: calculate total size of all arrays */ | |
10842 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
10843 | bool include_array = (arrays & (1UL << i)) > 0; | |
10844 | struct bpf_prog_info_array_desc *desc; | |
10845 | __u32 count, size; | |
10846 | ||
10847 | desc = bpf_prog_info_array_desc + i; | |
10848 | ||
10849 | /* kernel is too old to support this field */ | |
10850 | if (info_len < desc->array_offset + sizeof(__u32) || | |
10851 | info_len < desc->count_offset + sizeof(__u32) || | |
10852 | (desc->size_offset > 0 && info_len < desc->size_offset)) | |
10853 | include_array = false; | |
10854 | ||
10855 | if (!include_array) { | |
10856 | arrays &= ~(1UL << i); /* clear the bit */ | |
10857 | continue; | |
10858 | } | |
10859 | ||
10860 | count = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
10861 | size = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
10862 | ||
10863 | data_len += count * size; | |
10864 | } | |
10865 | ||
10866 | /* step 3: allocate continuous memory */ | |
10867 | data_len = roundup(data_len, sizeof(__u64)); | |
10868 | info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len); | |
10869 | if (!info_linear) | |
10870 | return ERR_PTR(-ENOMEM); | |
10871 | ||
10872 | /* step 4: fill data to info_linear->info */ | |
10873 | info_linear->arrays = arrays; | |
10874 | memset(&info_linear->info, 0, sizeof(info)); | |
10875 | ptr = info_linear->data; | |
10876 | ||
10877 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
10878 | struct bpf_prog_info_array_desc *desc; | |
10879 | __u32 count, size; | |
10880 | ||
10881 | if ((arrays & (1UL << i)) == 0) | |
10882 | continue; | |
10883 | ||
10884 | desc = bpf_prog_info_array_desc + i; | |
10885 | count = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
10886 | size = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
10887 | bpf_prog_info_set_offset_u32(&info_linear->info, | |
10888 | desc->count_offset, count); | |
10889 | bpf_prog_info_set_offset_u32(&info_linear->info, | |
10890 | desc->size_offset, size); | |
10891 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
10892 | desc->array_offset, | |
10893 | ptr_to_u64(ptr)); | |
10894 | ptr += count * size; | |
10895 | } | |
10896 | ||
10897 | /* step 5: call syscall again to get required arrays */ | |
10898 | err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len); | |
10899 | if (err) { | |
10900 | pr_debug("can't get prog info: %s", strerror(errno)); | |
10901 | free(info_linear); | |
10902 | return ERR_PTR(-EFAULT); | |
10903 | } | |
10904 | ||
10905 | /* step 6: verify the data */ | |
10906 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
10907 | struct bpf_prog_info_array_desc *desc; | |
10908 | __u32 v1, v2; | |
10909 | ||
10910 | if ((arrays & (1UL << i)) == 0) | |
10911 | continue; | |
10912 | ||
10913 | desc = bpf_prog_info_array_desc + i; | |
10914 | v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
10915 | v2 = bpf_prog_info_read_offset_u32(&info_linear->info, | |
10916 | desc->count_offset); | |
10917 | if (v1 != v2) | |
be18010e | 10918 | pr_warn("%s: mismatch in element count\n", __func__); |
34be1646 SL |
10919 | |
10920 | v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
10921 | v2 = bpf_prog_info_read_offset_u32(&info_linear->info, | |
10922 | desc->size_offset); | |
10923 | if (v1 != v2) | |
be18010e | 10924 | pr_warn("%s: mismatch in rec size\n", __func__); |
34be1646 SL |
10925 | } |
10926 | ||
10927 | /* step 7: update info_len and data_len */ | |
10928 | info_linear->info_len = sizeof(struct bpf_prog_info); | |
10929 | info_linear->data_len = data_len; | |
10930 | ||
10931 | return info_linear; | |
10932 | } | |
10933 | ||
10934 | void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear) | |
10935 | { | |
10936 | int i; | |
10937 | ||
10938 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
10939 | struct bpf_prog_info_array_desc *desc; | |
10940 | __u64 addr, offs; | |
10941 | ||
10942 | if ((info_linear->arrays & (1UL << i)) == 0) | |
10943 | continue; | |
10944 | ||
10945 | desc = bpf_prog_info_array_desc + i; | |
10946 | addr = bpf_prog_info_read_offset_u64(&info_linear->info, | |
10947 | desc->array_offset); | |
10948 | offs = addr - ptr_to_u64(info_linear->data); | |
10949 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
10950 | desc->array_offset, offs); | |
10951 | } | |
10952 | } | |
10953 | ||
10954 | void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear) | |
10955 | { | |
10956 | int i; | |
10957 | ||
10958 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
10959 | struct bpf_prog_info_array_desc *desc; | |
10960 | __u64 addr, offs; | |
10961 | ||
10962 | if ((info_linear->arrays & (1UL << i)) == 0) | |
10963 | continue; | |
10964 | ||
10965 | desc = bpf_prog_info_array_desc + i; | |
10966 | offs = bpf_prog_info_read_offset_u64(&info_linear->info, | |
10967 | desc->array_offset); | |
10968 | addr = offs + ptr_to_u64(info_linear->data); | |
10969 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
10970 | desc->array_offset, addr); | |
10971 | } | |
10972 | } | |
6446b315 | 10973 | |
ff26ce5c EC |
10974 | int bpf_program__set_attach_target(struct bpf_program *prog, |
10975 | int attach_prog_fd, | |
10976 | const char *attach_func_name) | |
10977 | { | |
fe62de31 | 10978 | int btf_obj_fd = 0, btf_id = 0, err; |
ff26ce5c EC |
10979 | |
10980 | if (!prog || attach_prog_fd < 0 || !attach_func_name) | |
10981 | return -EINVAL; | |
10982 | ||
fe62de31 AN |
10983 | if (prog->obj->loaded) |
10984 | return -EINVAL; | |
10985 | ||
10986 | if (attach_prog_fd) { | |
ff26ce5c EC |
10987 | btf_id = libbpf_find_prog_btf_id(attach_func_name, |
10988 | attach_prog_fd); | |
fe62de31 AN |
10989 | if (btf_id < 0) |
10990 | return btf_id; | |
10991 | } else { | |
10992 | /* load btf_vmlinux, if not yet */ | |
10993 | err = bpf_object__load_vmlinux_btf(prog->obj, true); | |
10994 | if (err) | |
10995 | return err; | |
10996 | err = find_kernel_btf_id(prog->obj, attach_func_name, | |
10997 | prog->expected_attach_type, | |
10998 | &btf_obj_fd, &btf_id); | |
10999 | if (err) | |
11000 | return err; | |
11001 | } | |
ff26ce5c EC |
11002 | |
11003 | prog->attach_btf_id = btf_id; | |
fe62de31 | 11004 | prog->attach_btf_obj_fd = btf_obj_fd; |
ff26ce5c EC |
11005 | prog->attach_prog_fd = attach_prog_fd; |
11006 | return 0; | |
11007 | } | |
11008 | ||
6803ee25 | 11009 | int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz) |
6446b315 | 11010 | { |
6803ee25 AN |
11011 | int err = 0, n, len, start, end = -1; |
11012 | bool *tmp; | |
6446b315 | 11013 | |
6803ee25 AN |
11014 | *mask = NULL; |
11015 | *mask_sz = 0; | |
11016 | ||
11017 | /* Each sub string separated by ',' has format \d+-\d+ or \d+ */ | |
11018 | while (*s) { | |
11019 | if (*s == ',' || *s == '\n') { | |
11020 | s++; | |
11021 | continue; | |
11022 | } | |
11023 | n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len); | |
11024 | if (n <= 0 || n > 2) { | |
11025 | pr_warn("Failed to get CPU range %s: %d\n", s, n); | |
11026 | err = -EINVAL; | |
11027 | goto cleanup; | |
11028 | } else if (n == 1) { | |
11029 | end = start; | |
11030 | } | |
11031 | if (start < 0 || start > end) { | |
11032 | pr_warn("Invalid CPU range [%d,%d] in %s\n", | |
11033 | start, end, s); | |
11034 | err = -EINVAL; | |
11035 | goto cleanup; | |
11036 | } | |
11037 | tmp = realloc(*mask, end + 1); | |
11038 | if (!tmp) { | |
11039 | err = -ENOMEM; | |
11040 | goto cleanup; | |
11041 | } | |
11042 | *mask = tmp; | |
11043 | memset(tmp + *mask_sz, 0, start - *mask_sz); | |
11044 | memset(tmp + start, 1, end - start + 1); | |
11045 | *mask_sz = end + 1; | |
11046 | s += len; | |
11047 | } | |
11048 | if (!*mask_sz) { | |
11049 | pr_warn("Empty CPU range\n"); | |
11050 | return -EINVAL; | |
11051 | } | |
11052 | return 0; | |
11053 | cleanup: | |
11054 | free(*mask); | |
11055 | *mask = NULL; | |
11056 | return err; | |
11057 | } | |
11058 | ||
11059 | int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz) | |
11060 | { | |
11061 | int fd, err = 0, len; | |
11062 | char buf[128]; | |
6446b315 HL |
11063 | |
11064 | fd = open(fcpu, O_RDONLY); | |
11065 | if (fd < 0) { | |
6803ee25 AN |
11066 | err = -errno; |
11067 | pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err); | |
11068 | return err; | |
6446b315 HL |
11069 | } |
11070 | len = read(fd, buf, sizeof(buf)); | |
11071 | close(fd); | |
11072 | if (len <= 0) { | |
6803ee25 AN |
11073 | err = len ? -errno : -EINVAL; |
11074 | pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err); | |
11075 | return err; | |
6446b315 | 11076 | } |
6803ee25 AN |
11077 | if (len >= sizeof(buf)) { |
11078 | pr_warn("CPU mask is too big in file %s\n", fcpu); | |
11079 | return -E2BIG; | |
6446b315 HL |
11080 | } |
11081 | buf[len] = '\0'; | |
11082 | ||
6803ee25 AN |
11083 | return parse_cpu_mask_str(buf, mask, mask_sz); |
11084 | } | |
11085 | ||
11086 | int libbpf_num_possible_cpus(void) | |
11087 | { | |
11088 | static const char *fcpu = "/sys/devices/system/cpu/possible"; | |
11089 | static int cpus; | |
11090 | int err, n, i, tmp_cpus; | |
11091 | bool *mask; | |
11092 | ||
11093 | tmp_cpus = READ_ONCE(cpus); | |
11094 | if (tmp_cpus > 0) | |
11095 | return tmp_cpus; | |
11096 | ||
11097 | err = parse_cpu_mask_file(fcpu, &mask, &n); | |
11098 | if (err) | |
11099 | return err; | |
11100 | ||
11101 | tmp_cpus = 0; | |
11102 | for (i = 0; i < n; i++) { | |
11103 | if (mask[i]) | |
11104 | tmp_cpus++; | |
6446b315 | 11105 | } |
6803ee25 | 11106 | free(mask); |
56fbc241 TC |
11107 | |
11108 | WRITE_ONCE(cpus, tmp_cpus); | |
11109 | return tmp_cpus; | |
6446b315 | 11110 | } |
d66562fb AN |
11111 | |
11112 | int bpf_object__open_skeleton(struct bpf_object_skeleton *s, | |
11113 | const struct bpf_object_open_opts *opts) | |
11114 | { | |
11115 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts, | |
11116 | .object_name = s->name, | |
11117 | ); | |
11118 | struct bpf_object *obj; | |
11119 | int i; | |
11120 | ||
11121 | /* Attempt to preserve opts->object_name, unless overriden by user | |
11122 | * explicitly. Overwriting object name for skeletons is discouraged, | |
11123 | * as it breaks global data maps, because they contain object name | |
11124 | * prefix as their own map name prefix. When skeleton is generated, | |
11125 | * bpftool is making an assumption that this name will stay the same. | |
11126 | */ | |
11127 | if (opts) { | |
11128 | memcpy(&skel_opts, opts, sizeof(*opts)); | |
11129 | if (!opts->object_name) | |
11130 | skel_opts.object_name = s->name; | |
11131 | } | |
11132 | ||
11133 | obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts); | |
11134 | if (IS_ERR(obj)) { | |
11135 | pr_warn("failed to initialize skeleton BPF object '%s': %ld\n", | |
11136 | s->name, PTR_ERR(obj)); | |
11137 | return PTR_ERR(obj); | |
11138 | } | |
11139 | ||
11140 | *s->obj = obj; | |
11141 | ||
11142 | for (i = 0; i < s->map_cnt; i++) { | |
11143 | struct bpf_map **map = s->maps[i].map; | |
11144 | const char *name = s->maps[i].name; | |
11145 | void **mmaped = s->maps[i].mmaped; | |
11146 | ||
11147 | *map = bpf_object__find_map_by_name(obj, name); | |
11148 | if (!*map) { | |
11149 | pr_warn("failed to find skeleton map '%s'\n", name); | |
11150 | return -ESRCH; | |
11151 | } | |
11152 | ||
2ad97d47 | 11153 | /* externs shouldn't be pre-setup from user code */ |
81bfdd08 | 11154 | if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG) |
d66562fb AN |
11155 | *mmaped = (*map)->mmaped; |
11156 | } | |
11157 | ||
11158 | for (i = 0; i < s->prog_cnt; i++) { | |
11159 | struct bpf_program **prog = s->progs[i].prog; | |
11160 | const char *name = s->progs[i].name; | |
11161 | ||
11162 | *prog = bpf_object__find_program_by_name(obj, name); | |
11163 | if (!*prog) { | |
11164 | pr_warn("failed to find skeleton program '%s'\n", name); | |
11165 | return -ESRCH; | |
11166 | } | |
11167 | } | |
11168 | ||
11169 | return 0; | |
11170 | } | |
11171 | ||
11172 | int bpf_object__load_skeleton(struct bpf_object_skeleton *s) | |
11173 | { | |
11174 | int i, err; | |
11175 | ||
11176 | err = bpf_object__load(*s->obj); | |
11177 | if (err) { | |
11178 | pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err); | |
11179 | return err; | |
11180 | } | |
11181 | ||
11182 | for (i = 0; i < s->map_cnt; i++) { | |
11183 | struct bpf_map *map = *s->maps[i].map; | |
11184 | size_t mmap_sz = bpf_map_mmap_sz(map); | |
11185 | int prot, map_fd = bpf_map__fd(map); | |
11186 | void **mmaped = s->maps[i].mmaped; | |
d66562fb AN |
11187 | |
11188 | if (!mmaped) | |
11189 | continue; | |
11190 | ||
11191 | if (!(map->def.map_flags & BPF_F_MMAPABLE)) { | |
11192 | *mmaped = NULL; | |
11193 | continue; | |
11194 | } | |
11195 | ||
11196 | if (map->def.map_flags & BPF_F_RDONLY_PROG) | |
11197 | prot = PROT_READ; | |
11198 | else | |
11199 | prot = PROT_READ | PROT_WRITE; | |
11200 | ||
11201 | /* Remap anonymous mmap()-ed "map initialization image" as | |
11202 | * a BPF map-backed mmap()-ed memory, but preserving the same | |
11203 | * memory address. This will cause kernel to change process' | |
11204 | * page table to point to a different piece of kernel memory, | |
11205 | * but from userspace point of view memory address (and its | |
11206 | * contents, being identical at this point) will stay the | |
11207 | * same. This mapping will be released by bpf_object__close() | |
11208 | * as per normal clean up procedure, so we don't need to worry | |
11209 | * about it from skeleton's clean up perspective. | |
11210 | */ | |
2ad97d47 AN |
11211 | *mmaped = mmap(map->mmaped, mmap_sz, prot, |
11212 | MAP_SHARED | MAP_FIXED, map_fd, 0); | |
11213 | if (*mmaped == MAP_FAILED) { | |
d66562fb AN |
11214 | err = -errno; |
11215 | *mmaped = NULL; | |
11216 | pr_warn("failed to re-mmap() map '%s': %d\n", | |
11217 | bpf_map__name(map), err); | |
11218 | return err; | |
11219 | } | |
11220 | } | |
11221 | ||
11222 | return 0; | |
11223 | } | |
11224 | ||
11225 | int bpf_object__attach_skeleton(struct bpf_object_skeleton *s) | |
11226 | { | |
11227 | int i; | |
11228 | ||
11229 | for (i = 0; i < s->prog_cnt; i++) { | |
11230 | struct bpf_program *prog = *s->progs[i].prog; | |
11231 | struct bpf_link **link = s->progs[i].link; | |
11232 | const struct bpf_sec_def *sec_def; | |
d66562fb | 11233 | |
d9297581 AN |
11234 | if (!prog->load) |
11235 | continue; | |
11236 | ||
52109584 | 11237 | sec_def = find_sec_def(prog->sec_name); |
d66562fb AN |
11238 | if (!sec_def || !sec_def->attach_fn) |
11239 | continue; | |
11240 | ||
11241 | *link = sec_def->attach_fn(sec_def, prog); | |
11242 | if (IS_ERR(*link)) { | |
11243 | pr_warn("failed to auto-attach program '%s': %ld\n", | |
11244 | bpf_program__name(prog), PTR_ERR(*link)); | |
11245 | return PTR_ERR(*link); | |
11246 | } | |
11247 | } | |
11248 | ||
11249 | return 0; | |
11250 | } | |
11251 | ||
11252 | void bpf_object__detach_skeleton(struct bpf_object_skeleton *s) | |
11253 | { | |
11254 | int i; | |
11255 | ||
11256 | for (i = 0; i < s->prog_cnt; i++) { | |
11257 | struct bpf_link **link = s->progs[i].link; | |
11258 | ||
50450fc7 | 11259 | bpf_link__destroy(*link); |
d66562fb AN |
11260 | *link = NULL; |
11261 | } | |
11262 | } | |
11263 | ||
11264 | void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s) | |
11265 | { | |
11266 | if (s->progs) | |
11267 | bpf_object__detach_skeleton(s); | |
11268 | if (s->obj) | |
11269 | bpf_object__close(*s->obj); | |
11270 | free(s->maps); | |
11271 | free(s->progs); | |
11272 | free(s); | |
11273 | } |