Commit | Line | Data |
---|---|---|
1bc38b8f | 1 | // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) |
6061a3d6 | 2 | |
1b76c13e WN |
3 | /* |
4 | * Common eBPF ELF object loading operations. | |
5 | * | |
6 | * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org> | |
7 | * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com> | |
8 | * Copyright (C) 2015 Huawei Inc. | |
f367540c | 9 | * Copyright (C) 2017 Nicira, Inc. |
d859900c | 10 | * Copyright (C) 2019 Isovalent, Inc. |
1b76c13e WN |
11 | */ |
12 | ||
b4269954 | 13 | #ifndef _GNU_SOURCE |
531b014e | 14 | #define _GNU_SOURCE |
b4269954 | 15 | #endif |
1b76c13e | 16 | #include <stdlib.h> |
b3f59d66 WN |
17 | #include <stdio.h> |
18 | #include <stdarg.h> | |
f367540c | 19 | #include <libgen.h> |
34090915 | 20 | #include <inttypes.h> |
8ab9da57 | 21 | #include <limits.h> |
b3f59d66 | 22 | #include <string.h> |
1b76c13e | 23 | #include <unistd.h> |
cdb2f920 | 24 | #include <endian.h> |
1a5e3fb1 WN |
25 | #include <fcntl.h> |
26 | #include <errno.h> | |
113e6b7e | 27 | #include <ctype.h> |
1b76c13e | 28 | #include <asm/unistd.h> |
e28ff1a8 | 29 | #include <linux/err.h> |
cb1e5e96 | 30 | #include <linux/kernel.h> |
1b76c13e | 31 | #include <linux/bpf.h> |
38d5d3b3 | 32 | #include <linux/btf.h> |
47eff617 | 33 | #include <linux/filter.h> |
9a208eff | 34 | #include <linux/list.h> |
f367540c | 35 | #include <linux/limits.h> |
438363c0 | 36 | #include <linux/perf_event.h> |
a64af0ef | 37 | #include <linux/ring_buffer.h> |
5e61f270 | 38 | #include <linux/version.h> |
fb84b822 | 39 | #include <sys/epoll.h> |
63f2f5ee | 40 | #include <sys/ioctl.h> |
fb84b822 | 41 | #include <sys/mman.h> |
f367540c JS |
42 | #include <sys/stat.h> |
43 | #include <sys/types.h> | |
44 | #include <sys/vfs.h> | |
ddc7c304 | 45 | #include <sys/utsname.h> |
dc3a2d25 | 46 | #include <sys/resource.h> |
1a5e3fb1 WN |
47 | #include <libelf.h> |
48 | #include <gelf.h> | |
166750bc | 49 | #include <zlib.h> |
1b76c13e WN |
50 | |
51 | #include "libbpf.h" | |
52d3352e | 52 | #include "bpf.h" |
8a138aed | 53 | #include "btf.h" |
6d41907c | 54 | #include "str_error.h" |
d7c4b398 | 55 | #include "libbpf_internal.h" |
ddc7c304 | 56 | #include "hashmap.h" |
67234743 | 57 | #include "bpf_gen_internal.h" |
b3f59d66 | 58 | |
f367540c JS |
59 | #ifndef BPF_FS_MAGIC |
60 | #define BPF_FS_MAGIC 0xcafe4a11 | |
61 | #endif | |
62 | ||
9c0f8cbd AN |
63 | #define BPF_INSN_SZ (sizeof(struct bpf_insn)) |
64 | ||
ff466b58 AI |
65 | /* vsprintf() in __base_pr() uses nonliteral format string. It may break |
66 | * compilation if user enables corresponding warning. Disable it explicitly. | |
67 | */ | |
68 | #pragma GCC diagnostic ignored "-Wformat-nonliteral" | |
69 | ||
b3f59d66 WN |
70 | #define __printf(a, b) __attribute__((format(printf, a, b))) |
71 | ||
590a0088 | 72 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj); |
aea28a60 | 73 | static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog); |
590a0088 | 74 | |
a8a1f7d0 SF |
75 | static int __base_pr(enum libbpf_print_level level, const char *format, |
76 | va_list args) | |
b3f59d66 | 77 | { |
6f1ae8b6 YS |
78 | if (level == LIBBPF_DEBUG) |
79 | return 0; | |
80 | ||
a8a1f7d0 | 81 | return vfprintf(stderr, format, args); |
b3f59d66 WN |
82 | } |
83 | ||
a8a1f7d0 | 84 | static libbpf_print_fn_t __libbpf_pr = __base_pr; |
b3f59d66 | 85 | |
e87fd8ba | 86 | libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn) |
b3f59d66 | 87 | { |
e87fd8ba AN |
88 | libbpf_print_fn_t old_print_fn = __libbpf_pr; |
89 | ||
6f1ae8b6 | 90 | __libbpf_pr = fn; |
e87fd8ba | 91 | return old_print_fn; |
b3f59d66 | 92 | } |
1a5e3fb1 | 93 | |
8461ef8b YS |
94 | __printf(2, 3) |
95 | void libbpf_print(enum libbpf_print_level level, const char *format, ...) | |
96 | { | |
97 | va_list args; | |
98 | ||
6f1ae8b6 YS |
99 | if (!__libbpf_pr) |
100 | return; | |
101 | ||
8461ef8b | 102 | va_start(args, format); |
6f1ae8b6 | 103 | __libbpf_pr(level, format, args); |
8461ef8b YS |
104 | va_end(args); |
105 | } | |
106 | ||
dc3a2d25 THJ |
107 | static void pr_perm_msg(int err) |
108 | { | |
109 | struct rlimit limit; | |
110 | char buf[100]; | |
111 | ||
112 | if (err != -EPERM || geteuid() != 0) | |
113 | return; | |
114 | ||
115 | err = getrlimit(RLIMIT_MEMLOCK, &limit); | |
116 | if (err) | |
117 | return; | |
118 | ||
119 | if (limit.rlim_cur == RLIM_INFINITY) | |
120 | return; | |
121 | ||
122 | if (limit.rlim_cur < 1024) | |
b5c7d0d0 | 123 | snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur); |
dc3a2d25 THJ |
124 | else if (limit.rlim_cur < 1024*1024) |
125 | snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024); | |
126 | else | |
127 | snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024)); | |
128 | ||
129 | pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n", | |
130 | buf); | |
131 | } | |
132 | ||
6371ca3b WN |
133 | #define STRERR_BUFSIZE 128 |
134 | ||
1a5e3fb1 WN |
135 | /* Copied from tools/perf/util/util.h */ |
136 | #ifndef zfree | |
137 | # define zfree(ptr) ({ free(*ptr); *ptr = NULL; }) | |
138 | #endif | |
139 | ||
140 | #ifndef zclose | |
141 | # define zclose(fd) ({ \ | |
142 | int ___err = 0; \ | |
143 | if ((fd) >= 0) \ | |
144 | ___err = close((fd)); \ | |
145 | fd = -1; \ | |
146 | ___err; }) | |
147 | #endif | |
148 | ||
34be1646 SL |
149 | static inline __u64 ptr_to_u64(const void *ptr) |
150 | { | |
151 | return (__u64) (unsigned long) ptr; | |
152 | } | |
153 | ||
5981881d AN |
154 | /* this goes away in libbpf 1.0 */ |
155 | enum libbpf_strict_mode libbpf_mode = LIBBPF_STRICT_NONE; | |
156 | ||
157 | int libbpf_set_strict_mode(enum libbpf_strict_mode mode) | |
158 | { | |
159 | /* __LIBBPF_STRICT_LAST is the last power-of-2 value used + 1, so to | |
160 | * get all possible values we compensate last +1, and then (2*x - 1) | |
161 | * to get the bit mask | |
162 | */ | |
163 | if (mode != LIBBPF_STRICT_ALL | |
164 | && (mode & ~((__LIBBPF_STRICT_LAST - 1) * 2 - 1))) | |
165 | return errno = EINVAL, -EINVAL; | |
166 | ||
167 | libbpf_mode = mode; | |
168 | return 0; | |
169 | } | |
170 | ||
47b6cb4d | 171 | enum kern_feature_id { |
47eff617 | 172 | /* v4.14: kernel support for program & map names. */ |
47b6cb4d | 173 | FEAT_PROG_NAME, |
8837fe5d | 174 | /* v5.2: kernel support for global data sections. */ |
47b6cb4d | 175 | FEAT_GLOBAL_DATA, |
68b08647 AN |
176 | /* BTF support */ |
177 | FEAT_BTF, | |
d7c4b398 | 178 | /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */ |
47b6cb4d | 179 | FEAT_BTF_FUNC, |
d7c4b398 | 180 | /* BTF_KIND_VAR and BTF_KIND_DATASEC support */ |
47b6cb4d | 181 | FEAT_BTF_DATASEC, |
2d3eb67f | 182 | /* BTF_FUNC_GLOBAL is supported */ |
47b6cb4d AN |
183 | FEAT_BTF_GLOBAL_FUNC, |
184 | /* BPF_F_MMAPABLE is supported for arrays */ | |
185 | FEAT_ARRAY_MMAP, | |
25498a19 | 186 | /* kernel support for expected_attach_type in BPF_PROG_LOAD */ |
47b6cb4d | 187 | FEAT_EXP_ATTACH_TYPE, |
109cea5a AN |
188 | /* bpf_probe_read_{kernel,user}[_str] helpers */ |
189 | FEAT_PROBE_READ_KERN, | |
5d23328d YZ |
190 | /* BPF_PROG_BIND_MAP is supported */ |
191 | FEAT_PROG_BIND_MAP, | |
4f33a53d AN |
192 | /* Kernel support for module BTFs */ |
193 | FEAT_MODULE_BTF, | |
22541a9e IL |
194 | /* BTF_KIND_FLOAT support */ |
195 | FEAT_BTF_FLOAT, | |
668ace0e AN |
196 | /* BPF perf link support */ |
197 | FEAT_PERF_LINK, | |
223f903e YS |
198 | /* BTF_KIND_DECL_TAG support */ |
199 | FEAT_BTF_DECL_TAG, | |
47b6cb4d | 200 | __FEAT_CNT, |
47eff617 SF |
201 | }; |
202 | ||
9ca1f56a | 203 | static bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id); |
47b6cb4d | 204 | |
166750bc AN |
205 | enum reloc_type { |
206 | RELO_LD64, | |
207 | RELO_CALL, | |
208 | RELO_DATA, | |
0c091e5c | 209 | RELO_EXTERN_VAR, |
5bd022ec | 210 | RELO_EXTERN_FUNC, |
53eddb5e | 211 | RELO_SUBPROG_ADDR, |
166750bc AN |
212 | }; |
213 | ||
214 | struct reloc_desc { | |
215 | enum reloc_type type; | |
216 | int insn_idx; | |
217 | int map_idx; | |
218 | int sym_off; | |
219 | }; | |
220 | ||
25498a19 AN |
221 | struct bpf_sec_def; |
222 | ||
12d9466d AN |
223 | typedef int (*init_fn_t)(struct bpf_program *prog, long cookie); |
224 | typedef int (*preload_fn_t)(struct bpf_program *prog, struct bpf_prog_load_params *attr, long cookie); | |
225 | typedef struct bpf_link *(*attach_fn_t)(const struct bpf_program *prog, long cookie); | |
25498a19 | 226 | |
15ea31fa AN |
227 | /* stored as sec_def->cookie for all libbpf-supported SEC()s */ |
228 | enum sec_def_flags { | |
229 | SEC_NONE = 0, | |
230 | /* expected_attach_type is optional, if kernel doesn't support that */ | |
231 | SEC_EXP_ATTACH_OPT = 1, | |
232 | /* legacy, only used by libbpf_get_type_names() and | |
233 | * libbpf_attach_type_by_name(), not used by libbpf itself at all. | |
234 | * This used to be associated with cgroup (and few other) BPF programs | |
235 | * that were attachable through BPF_PROG_ATTACH command. Pretty | |
236 | * meaningless nowadays, though. | |
237 | */ | |
238 | SEC_ATTACHABLE = 2, | |
239 | SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT, | |
240 | /* attachment target is specified through BTF ID in either kernel or | |
241 | * other BPF program's BTF object */ | |
242 | SEC_ATTACH_BTF = 4, | |
243 | /* BPF program type allows sleeping/blocking in kernel */ | |
244 | SEC_SLEEPABLE = 8, | |
dd94d45c AN |
245 | /* allow non-strict prefix matching */ |
246 | SEC_SLOPPY_PFX = 16, | |
15ea31fa AN |
247 | }; |
248 | ||
25498a19 AN |
249 | struct bpf_sec_def { |
250 | const char *sec; | |
25498a19 AN |
251 | enum bpf_prog_type prog_type; |
252 | enum bpf_attach_type expected_attach_type; | |
15ea31fa | 253 | long cookie; |
12d9466d AN |
254 | |
255 | init_fn_t init_fn; | |
256 | preload_fn_t preload_fn; | |
25498a19 AN |
257 | attach_fn_t attach_fn; |
258 | }; | |
259 | ||
a5b8bd47 WN |
260 | /* |
261 | * bpf_prog should be a better name but it has been used in | |
262 | * linux/filter.h. | |
263 | */ | |
264 | struct bpf_program { | |
25498a19 | 265 | const struct bpf_sec_def *sec_def; |
52109584 | 266 | char *sec_name; |
c1122392 AN |
267 | size_t sec_idx; |
268 | /* this program's instruction offset (in number of instructions) | |
269 | * within its containing ELF section | |
270 | */ | |
271 | size_t sec_insn_off; | |
272 | /* number of original instructions in ELF section belonging to this | |
273 | * program, not taking into account subprogram instructions possible | |
274 | * appended later during relocation | |
275 | */ | |
276 | size_t sec_insn_cnt; | |
277 | /* Offset (in number of instructions) of the start of instruction | |
278 | * belonging to this BPF program within its containing main BPF | |
279 | * program. For the entry-point (main) BPF program, this is always | |
280 | * zero. For a sub-program, this gets reset before each of main BPF | |
281 | * programs are processed and relocated and is used to determined | |
282 | * whether sub-program was already appended to the main program, and | |
283 | * if yes, at which instruction offset. | |
284 | */ | |
285 | size_t sub_insn_off; | |
286 | ||
287 | char *name; | |
a77f879b | 288 | /* name with / replaced by _; makes recursive pinning |
33a2c75c SF |
289 | * in bpf_object__pin_programs easier |
290 | */ | |
291 | char *pin_name; | |
c1122392 AN |
292 | |
293 | /* instructions that belong to BPF program; insns[0] is located at | |
294 | * sec_insn_off instruction within its ELF section in ELF file, so | |
295 | * when mapping ELF file instruction index to the local instruction, | |
296 | * one needs to subtract sec_insn_off; and vice versa. | |
297 | */ | |
a5b8bd47 | 298 | struct bpf_insn *insns; |
c1122392 AN |
299 | /* actual number of instruction in this BPF program's image; for |
300 | * entry-point BPF programs this includes the size of main program | |
301 | * itself plus all the used sub-programs, appended at the end | |
302 | */ | |
c3c55696 | 303 | size_t insns_cnt; |
34090915 | 304 | |
166750bc | 305 | struct reloc_desc *reloc_desc; |
34090915 | 306 | int nr_reloc; |
da11b417 | 307 | int log_level; |
55cffde2 | 308 | |
b580563e WN |
309 | struct { |
310 | int nr; | |
311 | int *fds; | |
312 | } instances; | |
313 | bpf_program_prep_t preprocessor; | |
aa9b1ac3 WN |
314 | |
315 | struct bpf_object *obj; | |
316 | void *priv; | |
317 | bpf_program_clear_priv_t clear_priv; | |
d7be143b | 318 | |
c1122392 | 319 | bool load; |
aea28a60 | 320 | bool mark_btf_static; |
c1122392 | 321 | enum bpf_prog_type type; |
d7be143b | 322 | enum bpf_attach_type expected_attach_type; |
c1122392 | 323 | int prog_ifindex; |
91abb4a6 | 324 | __u32 attach_btf_obj_fd; |
12a8654b | 325 | __u32 attach_btf_id; |
e7bf94db | 326 | __u32 attach_prog_fd; |
2993e051 YS |
327 | void *func_info; |
328 | __u32 func_info_rec_size; | |
f0187f0b | 329 | __u32 func_info_cnt; |
47eff617 | 330 | |
3d650141 MKL |
331 | void *line_info; |
332 | __u32 line_info_rec_size; | |
333 | __u32 line_info_cnt; | |
04656198 | 334 | __u32 prog_flags; |
a5b8bd47 WN |
335 | }; |
336 | ||
590a0088 MKL |
337 | struct bpf_struct_ops { |
338 | const char *tname; | |
339 | const struct btf_type *type; | |
340 | struct bpf_program **progs; | |
341 | __u32 *kern_func_off; | |
342 | /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */ | |
343 | void *data; | |
344 | /* e.g. struct bpf_struct_ops_tcp_congestion_ops in | |
345 | * btf_vmlinux's format. | |
346 | * struct bpf_struct_ops_tcp_congestion_ops { | |
347 | * [... some other kernel fields ...] | |
348 | * struct tcp_congestion_ops data; | |
349 | * } | |
350 | * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops) | |
351 | * bpf_map__init_kern_struct_ops() will populate the "kern_vdata" | |
352 | * from "data". | |
353 | */ | |
354 | void *kern_vdata; | |
355 | __u32 type_id; | |
356 | }; | |
357 | ||
ac9d1389 AN |
358 | #define DATA_SEC ".data" |
359 | #define BSS_SEC ".bss" | |
360 | #define RODATA_SEC ".rodata" | |
81bfdd08 | 361 | #define KCONFIG_SEC ".kconfig" |
1c0c7074 | 362 | #define KSYMS_SEC ".ksyms" |
590a0088 | 363 | #define STRUCT_OPS_SEC ".struct_ops" |
ac9d1389 | 364 | |
d859900c DB |
365 | enum libbpf_map_type { |
366 | LIBBPF_MAP_UNSPEC, | |
367 | LIBBPF_MAP_DATA, | |
368 | LIBBPF_MAP_BSS, | |
369 | LIBBPF_MAP_RODATA, | |
81bfdd08 | 370 | LIBBPF_MAP_KCONFIG, |
d859900c DB |
371 | }; |
372 | ||
9d759a9b | 373 | struct bpf_map { |
561bbcca | 374 | char *name; |
aed65917 AN |
375 | /* real_name is defined for special internal maps (.rodata*, |
376 | * .data*, .bss, .kconfig) and preserves their original ELF section | |
377 | * name. This is important to be be able to find corresponding BTF | |
378 | * DATASEC information. | |
379 | */ | |
380 | char *real_name; | |
01af3bf0 | 381 | int fd; |
db48814b AN |
382 | int sec_idx; |
383 | size_t sec_offset; | |
f0307a7e | 384 | int map_ifindex; |
addb9fc9 | 385 | int inner_map_fd; |
9d759a9b | 386 | struct bpf_map_def def; |
1bdb6c9a | 387 | __u32 numa_node; |
646f02ff | 388 | __u32 btf_var_idx; |
5b891af7 MKL |
389 | __u32 btf_key_type_id; |
390 | __u32 btf_value_type_id; | |
590a0088 | 391 | __u32 btf_vmlinux_value_type_id; |
9d759a9b WN |
392 | void *priv; |
393 | bpf_map_clear_priv_t clear_priv; | |
d859900c | 394 | enum libbpf_map_type libbpf_type; |
eba9c5f4 | 395 | void *mmaped; |
590a0088 | 396 | struct bpf_struct_ops *st_ops; |
646f02ff AN |
397 | struct bpf_map *inner_map; |
398 | void **init_slots; | |
399 | int init_slots_sz; | |
4580b25f THJ |
400 | char *pin_path; |
401 | bool pinned; | |
ec6d5f47 | 402 | bool reused; |
d859900c DB |
403 | }; |
404 | ||
166750bc AN |
405 | enum extern_type { |
406 | EXT_UNKNOWN, | |
2e33efe3 | 407 | EXT_KCFG, |
1c0c7074 | 408 | EXT_KSYM, |
2e33efe3 AN |
409 | }; |
410 | ||
411 | enum kcfg_type { | |
412 | KCFG_UNKNOWN, | |
413 | KCFG_CHAR, | |
414 | KCFG_BOOL, | |
415 | KCFG_INT, | |
416 | KCFG_TRISTATE, | |
417 | KCFG_CHAR_ARR, | |
166750bc AN |
418 | }; |
419 | ||
420 | struct extern_desc { | |
2e33efe3 | 421 | enum extern_type type; |
166750bc AN |
422 | int sym_idx; |
423 | int btf_id; | |
2e33efe3 AN |
424 | int sec_btf_id; |
425 | const char *name; | |
166750bc | 426 | bool is_set; |
2e33efe3 AN |
427 | bool is_weak; |
428 | union { | |
429 | struct { | |
430 | enum kcfg_type type; | |
431 | int sz; | |
432 | int align; | |
433 | int data_off; | |
434 | bool is_signed; | |
435 | } kcfg; | |
1c0c7074 AN |
436 | struct { |
437 | unsigned long long addr; | |
d370bbe1 HL |
438 | |
439 | /* target btf_id of the corresponding kernel var. */ | |
284d2587 AN |
440 | int kernel_btf_obj_fd; |
441 | int kernel_btf_id; | |
d370bbe1 HL |
442 | |
443 | /* local btf_id of the ksym extern's type. */ | |
444 | __u32 type_id; | |
9dbe6015 KKD |
445 | /* BTF fd index to be patched in for insn->off, this is |
446 | * 0 for vmlinux BTF, index in obj->fd_array for module | |
447 | * BTF | |
448 | */ | |
449 | __s16 btf_fd_idx; | |
1c0c7074 | 450 | } ksym; |
2e33efe3 | 451 | }; |
166750bc AN |
452 | }; |
453 | ||
9a208eff WN |
454 | static LIST_HEAD(bpf_objects_list); |
455 | ||
4f33a53d AN |
456 | struct module_btf { |
457 | struct btf *btf; | |
458 | char *name; | |
459 | __u32 id; | |
91abb4a6 | 460 | int fd; |
9dbe6015 | 461 | int fd_array_idx; |
4f33a53d AN |
462 | }; |
463 | ||
25bbbd7a AN |
464 | enum sec_type { |
465 | SEC_UNUSED = 0, | |
466 | SEC_RELO, | |
467 | SEC_BSS, | |
468 | SEC_DATA, | |
469 | SEC_RODATA, | |
470 | }; | |
471 | ||
472 | struct elf_sec_desc { | |
473 | enum sec_type sec_type; | |
474 | Elf64_Shdr *shdr; | |
475 | Elf_Data *data; | |
476 | }; | |
477 | ||
29a30ff5 AN |
478 | struct elf_state { |
479 | int fd; | |
480 | const void *obj_buf; | |
481 | size_t obj_buf_sz; | |
482 | Elf *elf; | |
ad23b723 | 483 | Elf64_Ehdr *ehdr; |
29a30ff5 | 484 | Elf_Data *symbols; |
29a30ff5 AN |
485 | Elf_Data *st_ops_data; |
486 | size_t shstrndx; /* section index for section name strings */ | |
487 | size_t strtabidx; | |
25bbbd7a AN |
488 | struct elf_sec_desc *secs; |
489 | int sec_cnt; | |
29a30ff5 AN |
490 | int maps_shndx; |
491 | int btf_maps_shndx; | |
492 | __u32 btf_maps_sec_btf_id; | |
493 | int text_shndx; | |
494 | int symbols_shndx; | |
29a30ff5 AN |
495 | int st_ops_shndx; |
496 | }; | |
497 | ||
1a5e3fb1 | 498 | struct bpf_object { |
d859900c | 499 | char name[BPF_OBJ_NAME_LEN]; |
cb1e5e96 | 500 | char license[64]; |
438363c0 | 501 | __u32 kern_version; |
0b3d1efa | 502 | |
a5b8bd47 WN |
503 | struct bpf_program *programs; |
504 | size_t nr_programs; | |
9d759a9b WN |
505 | struct bpf_map *maps; |
506 | size_t nr_maps; | |
bf829271 | 507 | size_t maps_cap; |
9d759a9b | 508 | |
8601fd42 | 509 | char *kconfig; |
166750bc AN |
510 | struct extern_desc *externs; |
511 | int nr_extern; | |
81bfdd08 | 512 | int kconfig_map_idx; |
166750bc | 513 | |
52d3352e | 514 | bool loaded; |
c3c55696 | 515 | bool has_subcalls; |
25bbbd7a | 516 | bool has_rodata; |
a5b8bd47 | 517 | |
e2fa0156 AS |
518 | struct bpf_gen *gen_loader; |
519 | ||
29a30ff5 AN |
520 | /* Information when doing ELF related work. Only valid if efile.elf is not NULL */ |
521 | struct elf_state efile; | |
1a5e3fb1 | 522 | /* |
29a30ff5 | 523 | * All loaded bpf_object are linked in a list, which is |
9a208eff WN |
524 | * hidden to caller. bpf_objects__<func> handlers deal with |
525 | * all objects. | |
526 | */ | |
527 | struct list_head list; | |
10931d24 | 528 | |
8a138aed | 529 | struct btf *btf; |
0f7515ca AN |
530 | struct btf_ext *btf_ext; |
531 | ||
a6ed02ca KS |
532 | /* Parse and load BTF vmlinux if any of the programs in the object need |
533 | * it at load time. | |
534 | */ | |
535 | struct btf *btf_vmlinux; | |
1373ff59 SC |
536 | /* Path to the custom BTF to be used for BPF CO-RE relocations as an |
537 | * override for vmlinux BTF. | |
538 | */ | |
539 | char *btf_custom_path; | |
0f7515ca AN |
540 | /* vmlinux BTF override for CO-RE relocations */ |
541 | struct btf *btf_vmlinux_override; | |
4f33a53d AN |
542 | /* Lazily initialized kernel module BTFs */ |
543 | struct module_btf *btf_modules; | |
544 | bool btf_modules_loaded; | |
545 | size_t btf_module_cnt; | |
546 | size_t btf_module_cap; | |
8a138aed | 547 | |
10931d24 WN |
548 | void *priv; |
549 | bpf_object_clear_priv_t clear_priv; | |
550 | ||
9dbe6015 KKD |
551 | int *fd_array; |
552 | size_t fd_array_cap; | |
553 | size_t fd_array_cnt; | |
554 | ||
1a5e3fb1 WN |
555 | char path[]; |
556 | }; | |
1a5e3fb1 | 557 | |
88a82120 AN |
558 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off); |
559 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off); | |
560 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx); | |
561 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name); | |
ad23b723 | 562 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn); |
88a82120 AN |
563 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn); |
564 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn); | |
ad23b723 AN |
565 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx); |
566 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx); | |
88a82120 | 567 | |
29cd77f4 | 568 | void bpf_program__unload(struct bpf_program *prog) |
55cffde2 | 569 | { |
b580563e WN |
570 | int i; |
571 | ||
55cffde2 WN |
572 | if (!prog) |
573 | return; | |
574 | ||
b580563e WN |
575 | /* |
576 | * If the object is opened but the program was never loaded, | |
577 | * it is possible that prog->instances.nr == -1. | |
578 | */ | |
579 | if (prog->instances.nr > 0) { | |
580 | for (i = 0; i < prog->instances.nr; i++) | |
581 | zclose(prog->instances.fds[i]); | |
582 | } else if (prog->instances.nr != -1) { | |
be18010e KW |
583 | pr_warn("Internal error: instances.nr is %d\n", |
584 | prog->instances.nr); | |
b580563e WN |
585 | } |
586 | ||
587 | prog->instances.nr = -1; | |
588 | zfree(&prog->instances.fds); | |
2993e051 | 589 | |
2993e051 | 590 | zfree(&prog->func_info); |
07a09d1b | 591 | zfree(&prog->line_info); |
55cffde2 WN |
592 | } |
593 | ||
a5b8bd47 WN |
594 | static void bpf_program__exit(struct bpf_program *prog) |
595 | { | |
596 | if (!prog) | |
597 | return; | |
598 | ||
aa9b1ac3 WN |
599 | if (prog->clear_priv) |
600 | prog->clear_priv(prog, prog->priv); | |
601 | ||
602 | prog->priv = NULL; | |
603 | prog->clear_priv = NULL; | |
604 | ||
55cffde2 | 605 | bpf_program__unload(prog); |
88cda1c9 | 606 | zfree(&prog->name); |
52109584 | 607 | zfree(&prog->sec_name); |
33a2c75c | 608 | zfree(&prog->pin_name); |
a5b8bd47 | 609 | zfree(&prog->insns); |
34090915 WN |
610 | zfree(&prog->reloc_desc); |
611 | ||
612 | prog->nr_reloc = 0; | |
a5b8bd47 | 613 | prog->insns_cnt = 0; |
c1122392 | 614 | prog->sec_idx = -1; |
a5b8bd47 WN |
615 | } |
616 | ||
33a2c75c SF |
617 | static char *__bpf_program__pin_name(struct bpf_program *prog) |
618 | { | |
619 | char *name, *p; | |
620 | ||
a77f879b SF |
621 | if (libbpf_mode & LIBBPF_STRICT_SEC_NAME) |
622 | name = strdup(prog->name); | |
623 | else | |
624 | name = strdup(prog->sec_name); | |
625 | ||
626 | if (!name) | |
627 | return NULL; | |
628 | ||
629 | p = name; | |
630 | ||
33a2c75c SF |
631 | while ((p = strchr(p, '/'))) |
632 | *p = '_'; | |
633 | ||
634 | return name; | |
635 | } | |
636 | ||
c3c55696 AN |
637 | static bool insn_is_subprog_call(const struct bpf_insn *insn) |
638 | { | |
639 | return BPF_CLASS(insn->code) == BPF_JMP && | |
640 | BPF_OP(insn->code) == BPF_CALL && | |
641 | BPF_SRC(insn->code) == BPF_K && | |
642 | insn->src_reg == BPF_PSEUDO_CALL && | |
643 | insn->dst_reg == 0 && | |
644 | insn->off == 0; | |
645 | } | |
646 | ||
aa0b8d43 MKL |
647 | static bool is_call_insn(const struct bpf_insn *insn) |
648 | { | |
649 | return insn->code == (BPF_JMP | BPF_CALL); | |
650 | } | |
651 | ||
53eddb5e YS |
652 | static bool insn_is_pseudo_func(struct bpf_insn *insn) |
653 | { | |
aa0b8d43 | 654 | return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC; |
53eddb5e YS |
655 | } |
656 | ||
a5b8bd47 | 657 | static int |
c3c55696 AN |
658 | bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog, |
659 | const char *name, size_t sec_idx, const char *sec_name, | |
660 | size_t sec_off, void *insn_data, size_t insn_data_sz) | |
a5b8bd47 | 661 | { |
c1122392 AN |
662 | if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) { |
663 | pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n", | |
664 | sec_name, name, sec_off, insn_data_sz); | |
a5b8bd47 WN |
665 | return -EINVAL; |
666 | } | |
667 | ||
1ad9cbb8 | 668 | memset(prog, 0, sizeof(*prog)); |
c3c55696 AN |
669 | prog->obj = obj; |
670 | ||
c1122392 AN |
671 | prog->sec_idx = sec_idx; |
672 | prog->sec_insn_off = sec_off / BPF_INSN_SZ; | |
673 | prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ; | |
674 | /* insns_cnt can later be increased by appending used subprograms */ | |
675 | prog->insns_cnt = prog->sec_insn_cnt; | |
a5b8bd47 | 676 | |
c1122392 AN |
677 | prog->type = BPF_PROG_TYPE_UNSPEC; |
678 | prog->load = true; | |
a5b8bd47 | 679 | |
c1122392 AN |
680 | prog->instances.fds = NULL; |
681 | prog->instances.nr = -1; | |
682 | ||
52109584 AN |
683 | prog->sec_name = strdup(sec_name); |
684 | if (!prog->sec_name) | |
c1122392 AN |
685 | goto errout; |
686 | ||
687 | prog->name = strdup(name); | |
688 | if (!prog->name) | |
a5b8bd47 | 689 | goto errout; |
a5b8bd47 | 690 | |
33a2c75c | 691 | prog->pin_name = __bpf_program__pin_name(prog); |
c1122392 | 692 | if (!prog->pin_name) |
33a2c75c | 693 | goto errout; |
33a2c75c | 694 | |
c1122392 AN |
695 | prog->insns = malloc(insn_data_sz); |
696 | if (!prog->insns) | |
a5b8bd47 | 697 | goto errout; |
c1122392 | 698 | memcpy(prog->insns, insn_data, insn_data_sz); |
a5b8bd47 WN |
699 | |
700 | return 0; | |
701 | errout: | |
c1122392 | 702 | pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name); |
a5b8bd47 WN |
703 | bpf_program__exit(prog); |
704 | return -ENOMEM; | |
705 | } | |
706 | ||
707 | static int | |
c1122392 AN |
708 | bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data, |
709 | const char *sec_name, int sec_idx) | |
a5b8bd47 | 710 | { |
6245947c | 711 | Elf_Data *symbols = obj->efile.symbols; |
c1122392 AN |
712 | struct bpf_program *prog, *progs; |
713 | void *data = sec_data->d_buf; | |
6245947c AN |
714 | size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms; |
715 | int nr_progs, err, i; | |
c1122392 | 716 | const char *name; |
ad23b723 | 717 | Elf64_Sym *sym; |
a5b8bd47 WN |
718 | |
719 | progs = obj->programs; | |
720 | nr_progs = obj->nr_programs; | |
ad23b723 | 721 | nr_syms = symbols->d_size / sizeof(Elf64_Sym); |
c1122392 | 722 | sec_off = 0; |
a5b8bd47 | 723 | |
6245947c | 724 | for (i = 0; i < nr_syms; i++) { |
ad23b723 AN |
725 | sym = elf_sym_by_idx(obj, i); |
726 | ||
727 | if (sym->st_shndx != sec_idx) | |
6245947c | 728 | continue; |
ad23b723 | 729 | if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC) |
6245947c | 730 | continue; |
88cda1c9 | 731 | |
ad23b723 AN |
732 | prog_sz = sym->st_size; |
733 | sec_off = sym->st_value; | |
88cda1c9 | 734 | |
ad23b723 | 735 | name = elf_sym_str(obj, sym->st_name); |
c1122392 AN |
736 | if (!name) { |
737 | pr_warn("sec '%s': failed to get symbol name for offset %zu\n", | |
738 | sec_name, sec_off); | |
739 | return -LIBBPF_ERRNO__FORMAT; | |
740 | } | |
88cda1c9 | 741 | |
c1122392 AN |
742 | if (sec_off + prog_sz > sec_sz) { |
743 | pr_warn("sec '%s': program at offset %zu crosses section boundary\n", | |
744 | sec_name, sec_off); | |
745 | return -LIBBPF_ERRNO__FORMAT; | |
746 | } | |
88cda1c9 | 747 | |
ad23b723 | 748 | if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) { |
513f485c AN |
749 | pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name); |
750 | return -ENOTSUP; | |
751 | } | |
752 | ||
c3c55696 AN |
753 | pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n", |
754 | sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz); | |
88cda1c9 | 755 | |
c3c55696 | 756 | progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs)); |
c1122392 AN |
757 | if (!progs) { |
758 | /* | |
759 | * In this case the original obj->programs | |
760 | * is still valid, so don't need special treat for | |
761 | * bpf_close_object(). | |
762 | */ | |
763 | pr_warn("sec '%s': failed to alloc memory for new program '%s'\n", | |
764 | sec_name, name); | |
765 | return -ENOMEM; | |
88cda1c9 | 766 | } |
c1122392 | 767 | obj->programs = progs; |
88cda1c9 | 768 | |
c1122392 | 769 | prog = &progs[nr_progs]; |
9a94f277 | 770 | |
c3c55696 AN |
771 | err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name, |
772 | sec_off, data + sec_off, prog_sz); | |
c1122392 AN |
773 | if (err) |
774 | return err; | |
9a94f277 | 775 | |
e5670fa0 AN |
776 | /* if function is a global/weak symbol, but has restricted |
777 | * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC | |
778 | * as static to enable more permissive BPF verification mode | |
779 | * with more outside context available to BPF verifier | |
aea28a60 | 780 | */ |
ad23b723 AN |
781 | if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL |
782 | && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN | |
783 | || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL)) | |
aea28a60 AN |
784 | prog->mark_btf_static = true; |
785 | ||
c1122392 AN |
786 | nr_progs++; |
787 | obj->nr_programs = nr_progs; | |
88cda1c9 MKL |
788 | } |
789 | ||
790 | return 0; | |
791 | } | |
792 | ||
5e61f270 AN |
793 | static __u32 get_kernel_version(void) |
794 | { | |
795 | __u32 major, minor, patch; | |
796 | struct utsname info; | |
797 | ||
798 | uname(&info); | |
799 | if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3) | |
800 | return 0; | |
801 | return KERNEL_VERSION(major, minor, patch); | |
802 | } | |
803 | ||
590a0088 MKL |
804 | static const struct btf_member * |
805 | find_member_by_offset(const struct btf_type *t, __u32 bit_offset) | |
806 | { | |
807 | struct btf_member *m; | |
808 | int i; | |
809 | ||
810 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
811 | if (btf_member_bit_offset(t, i) == bit_offset) | |
812 | return m; | |
813 | } | |
814 | ||
815 | return NULL; | |
816 | } | |
817 | ||
818 | static const struct btf_member * | |
819 | find_member_by_name(const struct btf *btf, const struct btf_type *t, | |
820 | const char *name) | |
821 | { | |
822 | struct btf_member *m; | |
823 | int i; | |
824 | ||
825 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
826 | if (!strcmp(btf__name_by_offset(btf, m->name_off), name)) | |
827 | return m; | |
828 | } | |
829 | ||
830 | return NULL; | |
831 | } | |
832 | ||
833 | #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_" | |
a6ed02ca KS |
834 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
835 | const char *name, __u32 kind); | |
590a0088 MKL |
836 | |
837 | static int | |
838 | find_struct_ops_kern_types(const struct btf *btf, const char *tname, | |
839 | const struct btf_type **type, __u32 *type_id, | |
840 | const struct btf_type **vtype, __u32 *vtype_id, | |
841 | const struct btf_member **data_member) | |
842 | { | |
843 | const struct btf_type *kern_type, *kern_vtype; | |
844 | const struct btf_member *kern_data_member; | |
845 | __s32 kern_vtype_id, kern_type_id; | |
590a0088 MKL |
846 | __u32 i; |
847 | ||
848 | kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT); | |
849 | if (kern_type_id < 0) { | |
850 | pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n", | |
851 | tname); | |
852 | return kern_type_id; | |
853 | } | |
854 | kern_type = btf__type_by_id(btf, kern_type_id); | |
855 | ||
856 | /* Find the corresponding "map_value" type that will be used | |
857 | * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example, | |
858 | * find "struct bpf_struct_ops_tcp_congestion_ops" from the | |
859 | * btf_vmlinux. | |
860 | */ | |
a6ed02ca KS |
861 | kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX, |
862 | tname, BTF_KIND_STRUCT); | |
590a0088 | 863 | if (kern_vtype_id < 0) { |
a6ed02ca KS |
864 | pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n", |
865 | STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
866 | return kern_vtype_id; |
867 | } | |
868 | kern_vtype = btf__type_by_id(btf, kern_vtype_id); | |
869 | ||
870 | /* Find "struct tcp_congestion_ops" from | |
871 | * struct bpf_struct_ops_tcp_congestion_ops { | |
872 | * [ ... ] | |
873 | * struct tcp_congestion_ops data; | |
874 | * } | |
875 | */ | |
876 | kern_data_member = btf_members(kern_vtype); | |
877 | for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) { | |
878 | if (kern_data_member->type == kern_type_id) | |
879 | break; | |
880 | } | |
881 | if (i == btf_vlen(kern_vtype)) { | |
a6ed02ca KS |
882 | pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n", |
883 | tname, STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
884 | return -EINVAL; |
885 | } | |
886 | ||
887 | *type = kern_type; | |
888 | *type_id = kern_type_id; | |
889 | *vtype = kern_vtype; | |
890 | *vtype_id = kern_vtype_id; | |
891 | *data_member = kern_data_member; | |
892 | ||
893 | return 0; | |
894 | } | |
895 | ||
896 | static bool bpf_map__is_struct_ops(const struct bpf_map *map) | |
897 | { | |
898 | return map->def.type == BPF_MAP_TYPE_STRUCT_OPS; | |
899 | } | |
900 | ||
901 | /* Init the map's fields that depend on kern_btf */ | |
902 | static int bpf_map__init_kern_struct_ops(struct bpf_map *map, | |
903 | const struct btf *btf, | |
904 | const struct btf *kern_btf) | |
905 | { | |
906 | const struct btf_member *member, *kern_member, *kern_data_member; | |
907 | const struct btf_type *type, *kern_type, *kern_vtype; | |
908 | __u32 i, kern_type_id, kern_vtype_id, kern_data_off; | |
909 | struct bpf_struct_ops *st_ops; | |
910 | void *data, *kern_data; | |
911 | const char *tname; | |
912 | int err; | |
913 | ||
914 | st_ops = map->st_ops; | |
915 | type = st_ops->type; | |
916 | tname = st_ops->tname; | |
917 | err = find_struct_ops_kern_types(kern_btf, tname, | |
918 | &kern_type, &kern_type_id, | |
919 | &kern_vtype, &kern_vtype_id, | |
920 | &kern_data_member); | |
921 | if (err) | |
922 | return err; | |
923 | ||
924 | pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n", | |
925 | map->name, st_ops->type_id, kern_type_id, kern_vtype_id); | |
926 | ||
927 | map->def.value_size = kern_vtype->size; | |
928 | map->btf_vmlinux_value_type_id = kern_vtype_id; | |
929 | ||
930 | st_ops->kern_vdata = calloc(1, kern_vtype->size); | |
931 | if (!st_ops->kern_vdata) | |
932 | return -ENOMEM; | |
933 | ||
934 | data = st_ops->data; | |
935 | kern_data_off = kern_data_member->offset / 8; | |
936 | kern_data = st_ops->kern_vdata + kern_data_off; | |
937 | ||
938 | member = btf_members(type); | |
939 | for (i = 0; i < btf_vlen(type); i++, member++) { | |
940 | const struct btf_type *mtype, *kern_mtype; | |
941 | __u32 mtype_id, kern_mtype_id; | |
942 | void *mdata, *kern_mdata; | |
943 | __s64 msize, kern_msize; | |
944 | __u32 moff, kern_moff; | |
945 | __u32 kern_member_idx; | |
946 | const char *mname; | |
947 | ||
948 | mname = btf__name_by_offset(btf, member->name_off); | |
949 | kern_member = find_member_by_name(kern_btf, kern_type, mname); | |
950 | if (!kern_member) { | |
951 | pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n", | |
952 | map->name, mname); | |
953 | return -ENOTSUP; | |
954 | } | |
955 | ||
956 | kern_member_idx = kern_member - btf_members(kern_type); | |
957 | if (btf_member_bitfield_size(type, i) || | |
958 | btf_member_bitfield_size(kern_type, kern_member_idx)) { | |
959 | pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n", | |
960 | map->name, mname); | |
961 | return -ENOTSUP; | |
962 | } | |
963 | ||
964 | moff = member->offset / 8; | |
965 | kern_moff = kern_member->offset / 8; | |
966 | ||
967 | mdata = data + moff; | |
968 | kern_mdata = kern_data + kern_moff; | |
969 | ||
970 | mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id); | |
971 | kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type, | |
972 | &kern_mtype_id); | |
973 | if (BTF_INFO_KIND(mtype->info) != | |
974 | BTF_INFO_KIND(kern_mtype->info)) { | |
975 | pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n", | |
976 | map->name, mname, BTF_INFO_KIND(mtype->info), | |
977 | BTF_INFO_KIND(kern_mtype->info)); | |
978 | return -ENOTSUP; | |
979 | } | |
980 | ||
981 | if (btf_is_ptr(mtype)) { | |
982 | struct bpf_program *prog; | |
983 | ||
d2836ddd MKL |
984 | prog = st_ops->progs[i]; |
985 | if (!prog) | |
986 | continue; | |
987 | ||
590a0088 MKL |
988 | kern_mtype = skip_mods_and_typedefs(kern_btf, |
989 | kern_mtype->type, | |
990 | &kern_mtype_id); | |
d2836ddd MKL |
991 | |
992 | /* mtype->type must be a func_proto which was | |
993 | * guaranteed in bpf_object__collect_st_ops_relos(), | |
994 | * so only check kern_mtype for func_proto here. | |
995 | */ | |
996 | if (!btf_is_func_proto(kern_mtype)) { | |
997 | pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n", | |
590a0088 MKL |
998 | map->name, mname); |
999 | return -ENOTSUP; | |
1000 | } | |
1001 | ||
590a0088 MKL |
1002 | prog->attach_btf_id = kern_type_id; |
1003 | prog->expected_attach_type = kern_member_idx; | |
1004 | ||
1005 | st_ops->kern_func_off[i] = kern_data_off + kern_moff; | |
1006 | ||
1007 | pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n", | |
1008 | map->name, mname, prog->name, moff, | |
1009 | kern_moff); | |
1010 | ||
1011 | continue; | |
1012 | } | |
1013 | ||
1014 | msize = btf__resolve_size(btf, mtype_id); | |
1015 | kern_msize = btf__resolve_size(kern_btf, kern_mtype_id); | |
1016 | if (msize < 0 || kern_msize < 0 || msize != kern_msize) { | |
1017 | pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n", | |
1018 | map->name, mname, (ssize_t)msize, | |
1019 | (ssize_t)kern_msize); | |
1020 | return -ENOTSUP; | |
1021 | } | |
1022 | ||
1023 | pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n", | |
1024 | map->name, mname, (unsigned int)msize, | |
1025 | moff, kern_moff); | |
1026 | memcpy(kern_mdata, mdata, msize); | |
1027 | } | |
1028 | ||
1029 | return 0; | |
1030 | } | |
1031 | ||
1032 | static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj) | |
1033 | { | |
590a0088 MKL |
1034 | struct bpf_map *map; |
1035 | size_t i; | |
1036 | int err; | |
1037 | ||
1038 | for (i = 0; i < obj->nr_maps; i++) { | |
1039 | map = &obj->maps[i]; | |
1040 | ||
1041 | if (!bpf_map__is_struct_ops(map)) | |
1042 | continue; | |
1043 | ||
a6ed02ca KS |
1044 | err = bpf_map__init_kern_struct_ops(map, obj->btf, |
1045 | obj->btf_vmlinux); | |
1046 | if (err) | |
590a0088 | 1047 | return err; |
590a0088 MKL |
1048 | } |
1049 | ||
590a0088 MKL |
1050 | return 0; |
1051 | } | |
1052 | ||
1053 | static int bpf_object__init_struct_ops_maps(struct bpf_object *obj) | |
1054 | { | |
1055 | const struct btf_type *type, *datasec; | |
1056 | const struct btf_var_secinfo *vsi; | |
1057 | struct bpf_struct_ops *st_ops; | |
1058 | const char *tname, *var_name; | |
1059 | __s32 type_id, datasec_id; | |
1060 | const struct btf *btf; | |
1061 | struct bpf_map *map; | |
1062 | __u32 i; | |
1063 | ||
1064 | if (obj->efile.st_ops_shndx == -1) | |
1065 | return 0; | |
1066 | ||
1067 | btf = obj->btf; | |
1068 | datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC, | |
1069 | BTF_KIND_DATASEC); | |
1070 | if (datasec_id < 0) { | |
1071 | pr_warn("struct_ops init: DATASEC %s not found\n", | |
1072 | STRUCT_OPS_SEC); | |
1073 | return -EINVAL; | |
1074 | } | |
1075 | ||
1076 | datasec = btf__type_by_id(btf, datasec_id); | |
1077 | vsi = btf_var_secinfos(datasec); | |
1078 | for (i = 0; i < btf_vlen(datasec); i++, vsi++) { | |
1079 | type = btf__type_by_id(obj->btf, vsi->type); | |
1080 | var_name = btf__name_by_offset(obj->btf, type->name_off); | |
1081 | ||
1082 | type_id = btf__resolve_type(obj->btf, vsi->type); | |
1083 | if (type_id < 0) { | |
1084 | pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n", | |
1085 | vsi->type, STRUCT_OPS_SEC); | |
1086 | return -EINVAL; | |
1087 | } | |
1088 | ||
1089 | type = btf__type_by_id(obj->btf, type_id); | |
1090 | tname = btf__name_by_offset(obj->btf, type->name_off); | |
1091 | if (!tname[0]) { | |
1092 | pr_warn("struct_ops init: anonymous type is not supported\n"); | |
1093 | return -ENOTSUP; | |
1094 | } | |
1095 | if (!btf_is_struct(type)) { | |
1096 | pr_warn("struct_ops init: %s is not a struct\n", tname); | |
1097 | return -EINVAL; | |
1098 | } | |
1099 | ||
1100 | map = bpf_object__add_map(obj); | |
1101 | if (IS_ERR(map)) | |
1102 | return PTR_ERR(map); | |
1103 | ||
1104 | map->sec_idx = obj->efile.st_ops_shndx; | |
1105 | map->sec_offset = vsi->offset; | |
1106 | map->name = strdup(var_name); | |
1107 | if (!map->name) | |
1108 | return -ENOMEM; | |
1109 | ||
1110 | map->def.type = BPF_MAP_TYPE_STRUCT_OPS; | |
1111 | map->def.key_size = sizeof(int); | |
1112 | map->def.value_size = type->size; | |
1113 | map->def.max_entries = 1; | |
1114 | ||
1115 | map->st_ops = calloc(1, sizeof(*map->st_ops)); | |
1116 | if (!map->st_ops) | |
1117 | return -ENOMEM; | |
1118 | st_ops = map->st_ops; | |
1119 | st_ops->data = malloc(type->size); | |
1120 | st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs)); | |
1121 | st_ops->kern_func_off = malloc(btf_vlen(type) * | |
1122 | sizeof(*st_ops->kern_func_off)); | |
1123 | if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off) | |
1124 | return -ENOMEM; | |
1125 | ||
1126 | if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) { | |
1127 | pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n", | |
1128 | var_name, STRUCT_OPS_SEC); | |
1129 | return -EINVAL; | |
1130 | } | |
1131 | ||
1132 | memcpy(st_ops->data, | |
1133 | obj->efile.st_ops_data->d_buf + vsi->offset, | |
1134 | type->size); | |
1135 | st_ops->tname = tname; | |
1136 | st_ops->type = type; | |
1137 | st_ops->type_id = type_id; | |
1138 | ||
1139 | pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n", | |
1140 | tname, type_id, var_name, vsi->offset); | |
1141 | } | |
1142 | ||
1143 | return 0; | |
1144 | } | |
1145 | ||
6c956392 | 1146 | static struct bpf_object *bpf_object__new(const char *path, |
5e61f270 | 1147 | const void *obj_buf, |
2ce8450e AN |
1148 | size_t obj_buf_sz, |
1149 | const char *obj_name) | |
1a5e3fb1 | 1150 | { |
689624f0 | 1151 | bool strict = (libbpf_mode & LIBBPF_STRICT_NO_OBJECT_LIST); |
1a5e3fb1 | 1152 | struct bpf_object *obj; |
d859900c | 1153 | char *end; |
1a5e3fb1 WN |
1154 | |
1155 | obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1); | |
1156 | if (!obj) { | |
be18010e | 1157 | pr_warn("alloc memory failed for %s\n", path); |
6371ca3b | 1158 | return ERR_PTR(-ENOMEM); |
1a5e3fb1 WN |
1159 | } |
1160 | ||
1161 | strcpy(obj->path, path); | |
2ce8450e AN |
1162 | if (obj_name) { |
1163 | strncpy(obj->name, obj_name, sizeof(obj->name) - 1); | |
1164 | obj->name[sizeof(obj->name) - 1] = 0; | |
1165 | } else { | |
1166 | /* Using basename() GNU version which doesn't modify arg. */ | |
1167 | strncpy(obj->name, basename((void *)path), | |
1168 | sizeof(obj->name) - 1); | |
1169 | end = strchr(obj->name, '.'); | |
1170 | if (end) | |
1171 | *end = 0; | |
1172 | } | |
6c956392 | 1173 | |
d859900c | 1174 | obj->efile.fd = -1; |
6c956392 | 1175 | /* |
76e1022b | 1176 | * Caller of this function should also call |
6c956392 WN |
1177 | * bpf_object__elf_finish() after data collection to return |
1178 | * obj_buf to user. If not, we should duplicate the buffer to | |
1179 | * avoid user freeing them before elf finish. | |
1180 | */ | |
1181 | obj->efile.obj_buf = obj_buf; | |
1182 | obj->efile.obj_buf_sz = obj_buf_sz; | |
666810e8 | 1183 | obj->efile.maps_shndx = -1; |
abd29c93 | 1184 | obj->efile.btf_maps_shndx = -1; |
590a0088 | 1185 | obj->efile.st_ops_shndx = -1; |
81bfdd08 | 1186 | obj->kconfig_map_idx = -1; |
6c956392 | 1187 | |
5e61f270 | 1188 | obj->kern_version = get_kernel_version(); |
52d3352e | 1189 | obj->loaded = false; |
9a208eff WN |
1190 | |
1191 | INIT_LIST_HEAD(&obj->list); | |
689624f0 JB |
1192 | if (!strict) |
1193 | list_add(&obj->list, &bpf_objects_list); | |
1a5e3fb1 WN |
1194 | return obj; |
1195 | } | |
1196 | ||
1197 | static void bpf_object__elf_finish(struct bpf_object *obj) | |
1198 | { | |
29a30ff5 | 1199 | if (!obj->efile.elf) |
1a5e3fb1 WN |
1200 | return; |
1201 | ||
1202 | if (obj->efile.elf) { | |
1203 | elf_end(obj->efile.elf); | |
1204 | obj->efile.elf = NULL; | |
1205 | } | |
bec7d68c | 1206 | obj->efile.symbols = NULL; |
590a0088 | 1207 | obj->efile.st_ops_data = NULL; |
b62f06e8 | 1208 | |
25bbbd7a AN |
1209 | zfree(&obj->efile.secs); |
1210 | obj->efile.sec_cnt = 0; | |
1a5e3fb1 | 1211 | zclose(obj->efile.fd); |
6c956392 WN |
1212 | obj->efile.obj_buf = NULL; |
1213 | obj->efile.obj_buf_sz = 0; | |
1a5e3fb1 WN |
1214 | } |
1215 | ||
1216 | static int bpf_object__elf_init(struct bpf_object *obj) | |
1217 | { | |
ad23b723 | 1218 | Elf64_Ehdr *ehdr; |
1a5e3fb1 | 1219 | int err = 0; |
ad23b723 | 1220 | Elf *elf; |
1a5e3fb1 | 1221 | |
29a30ff5 | 1222 | if (obj->efile.elf) { |
88a82120 | 1223 | pr_warn("elf: init internal error\n"); |
6371ca3b | 1224 | return -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1225 | } |
1226 | ||
6c956392 WN |
1227 | if (obj->efile.obj_buf_sz > 0) { |
1228 | /* | |
1229 | * obj_buf should have been validated by | |
1230 | * bpf_object__open_buffer(). | |
1231 | */ | |
ad23b723 | 1232 | elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz); |
6c956392 WN |
1233 | } else { |
1234 | obj->efile.fd = open(obj->path, O_RDONLY); | |
1235 | if (obj->efile.fd < 0) { | |
be5c5d4e | 1236 | char errmsg[STRERR_BUFSIZE], *cp; |
1ce6a9fc | 1237 | |
be5c5d4e AN |
1238 | err = -errno; |
1239 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
88a82120 | 1240 | pr_warn("elf: failed to open %s: %s\n", obj->path, cp); |
be5c5d4e | 1241 | return err; |
6c956392 WN |
1242 | } |
1243 | ||
ad23b723 | 1244 | elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL); |
1a5e3fb1 WN |
1245 | } |
1246 | ||
ad23b723 | 1247 | if (!elf) { |
88a82120 | 1248 | pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1249 | err = -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1250 | goto errout; |
1251 | } | |
1252 | ||
ad23b723 AN |
1253 | obj->efile.elf = elf; |
1254 | ||
1255 | if (elf_kind(elf) != ELF_K_ELF) { | |
1256 | err = -LIBBPF_ERRNO__FORMAT; | |
1257 | pr_warn("elf: '%s' is not a proper ELF object\n", obj->path); | |
1258 | goto errout; | |
1259 | } | |
1260 | ||
1261 | if (gelf_getclass(elf) != ELFCLASS64) { | |
1262 | err = -LIBBPF_ERRNO__FORMAT; | |
1263 | pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path); | |
1264 | goto errout; | |
1265 | } | |
1266 | ||
1267 | obj->efile.ehdr = ehdr = elf64_getehdr(elf); | |
1268 | if (!obj->efile.ehdr) { | |
88a82120 | 1269 | pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1270 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1271 | goto errout; |
1272 | } | |
1a5e3fb1 | 1273 | |
ad23b723 | 1274 | if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) { |
88a82120 AN |
1275 | pr_warn("elf: failed to get section names section index for %s: %s\n", |
1276 | obj->path, elf_errmsg(-1)); | |
1277 | err = -LIBBPF_ERRNO__FORMAT; | |
1278 | goto errout; | |
1279 | } | |
1280 | ||
1281 | /* Elf is corrupted/truncated, avoid calling elf_strptr. */ | |
ad23b723 | 1282 | if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) { |
88a82120 AN |
1283 | pr_warn("elf: failed to get section names strings from %s: %s\n", |
1284 | obj->path, elf_errmsg(-1)); | |
8f3f5792 NK |
1285 | err = -LIBBPF_ERRNO__FORMAT; |
1286 | goto errout; | |
88a82120 AN |
1287 | } |
1288 | ||
9b16137a | 1289 | /* Old LLVM set e_machine to EM_NONE */ |
ad23b723 | 1290 | if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) { |
88a82120 | 1291 | pr_warn("elf: %s is not a valid eBPF object file\n", obj->path); |
6371ca3b | 1292 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1293 | goto errout; |
1294 | } | |
1295 | ||
1296 | return 0; | |
1297 | errout: | |
1298 | bpf_object__elf_finish(obj); | |
1299 | return err; | |
1300 | } | |
1301 | ||
12ef5634 | 1302 | static int bpf_object__check_endianness(struct bpf_object *obj) |
cc4228d5 | 1303 | { |
3930198d | 1304 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
ad23b723 | 1305 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB) |
12ef5634 | 1306 | return 0; |
3930198d | 1307 | #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
ad23b723 | 1308 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB) |
12ef5634 AN |
1309 | return 0; |
1310 | #else | |
1311 | # error "Unrecognized __BYTE_ORDER__" | |
1312 | #endif | |
88a82120 | 1313 | pr_warn("elf: endianness mismatch in %s.\n", obj->path); |
6371ca3b | 1314 | return -LIBBPF_ERRNO__ENDIAN; |
cc4228d5 WN |
1315 | } |
1316 | ||
cb1e5e96 | 1317 | static int |
399dc65e | 1318 | bpf_object__init_license(struct bpf_object *obj, void *data, size_t size) |
cb1e5e96 | 1319 | { |
399dc65e | 1320 | memcpy(obj->license, data, min(size, sizeof(obj->license) - 1)); |
cb1e5e96 WN |
1321 | pr_debug("license of %s is %s\n", obj->path, obj->license); |
1322 | return 0; | |
1323 | } | |
1324 | ||
54b8625c JF |
1325 | static int |
1326 | bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size) | |
1327 | { | |
1328 | __u32 kver; | |
1329 | ||
1330 | if (size != sizeof(kver)) { | |
be18010e | 1331 | pr_warn("invalid kver section in %s\n", obj->path); |
54b8625c JF |
1332 | return -LIBBPF_ERRNO__FORMAT; |
1333 | } | |
1334 | memcpy(&kver, data, sizeof(kver)); | |
1335 | obj->kern_version = kver; | |
1336 | pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version); | |
1337 | return 0; | |
1338 | } | |
1339 | ||
addb9fc9 NS |
1340 | static bool bpf_map_type__is_map_in_map(enum bpf_map_type type) |
1341 | { | |
1342 | if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1343 | type == BPF_MAP_TYPE_HASH_OF_MAPS) | |
1344 | return true; | |
1345 | return false; | |
1346 | } | |
1347 | ||
b96c07f3 | 1348 | static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size) |
1713d68b DB |
1349 | { |
1350 | int ret = -ENOENT; | |
25bbbd7a AN |
1351 | Elf_Data *data; |
1352 | Elf_Scn *scn; | |
1713d68b DB |
1353 | |
1354 | *size = 0; | |
25bbbd7a | 1355 | if (!name) |
1713d68b | 1356 | return -EINVAL; |
88a82120 | 1357 | |
25bbbd7a AN |
1358 | scn = elf_sec_by_name(obj, name); |
1359 | data = elf_sec_data(obj, scn); | |
1360 | if (data) { | |
1361 | ret = 0; /* found it */ | |
1362 | *size = data->d_size; | |
1713d68b DB |
1363 | } |
1364 | ||
1365 | return *size ? 0 : ret; | |
1366 | } | |
1367 | ||
b96c07f3 | 1368 | static int find_elf_var_offset(const struct bpf_object *obj, const char *name, __u32 *off) |
1713d68b DB |
1369 | { |
1370 | Elf_Data *symbols = obj->efile.symbols; | |
1371 | const char *sname; | |
1372 | size_t si; | |
1373 | ||
1374 | if (!name || !off) | |
1375 | return -EINVAL; | |
1376 | ||
ad23b723 AN |
1377 | for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) { |
1378 | Elf64_Sym *sym = elf_sym_by_idx(obj, si); | |
1713d68b | 1379 | |
ad23b723 AN |
1380 | if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL || |
1381 | ELF64_ST_TYPE(sym->st_info) != STT_OBJECT) | |
1713d68b DB |
1382 | continue; |
1383 | ||
ad23b723 | 1384 | sname = elf_sym_str(obj, sym->st_name); |
1713d68b | 1385 | if (!sname) { |
ad23b723 | 1386 | pr_warn("failed to get sym name string for var %s\n", name); |
1713d68b DB |
1387 | return -EIO; |
1388 | } | |
1389 | if (strcmp(name, sname) == 0) { | |
ad23b723 | 1390 | *off = sym->st_value; |
1713d68b DB |
1391 | return 0; |
1392 | } | |
1393 | } | |
1394 | ||
1395 | return -ENOENT; | |
1396 | } | |
1397 | ||
bf829271 | 1398 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj) |
d859900c | 1399 | { |
bf829271 AN |
1400 | struct bpf_map *new_maps; |
1401 | size_t new_cap; | |
1402 | int i; | |
1403 | ||
1404 | if (obj->nr_maps < obj->maps_cap) | |
1405 | return &obj->maps[obj->nr_maps++]; | |
1406 | ||
95064979 | 1407 | new_cap = max((size_t)4, obj->maps_cap * 3 / 2); |
029258d7 | 1408 | new_maps = libbpf_reallocarray(obj->maps, new_cap, sizeof(*obj->maps)); |
bf829271 | 1409 | if (!new_maps) { |
be18010e | 1410 | pr_warn("alloc maps for object failed\n"); |
bf829271 AN |
1411 | return ERR_PTR(-ENOMEM); |
1412 | } | |
1413 | ||
1414 | obj->maps_cap = new_cap; | |
1415 | obj->maps = new_maps; | |
1416 | ||
1417 | /* zero out new maps */ | |
1418 | memset(obj->maps + obj->nr_maps, 0, | |
1419 | (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps)); | |
1420 | /* | |
1421 | * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin) | |
1422 | * when failure (zclose won't close negative fd)). | |
1423 | */ | |
1424 | for (i = obj->nr_maps; i < obj->maps_cap; i++) { | |
1425 | obj->maps[i].fd = -1; | |
1426 | obj->maps[i].inner_map_fd = -1; | |
1427 | } | |
1428 | ||
1429 | return &obj->maps[obj->nr_maps++]; | |
d859900c DB |
1430 | } |
1431 | ||
eba9c5f4 AN |
1432 | static size_t bpf_map_mmap_sz(const struct bpf_map *map) |
1433 | { | |
1434 | long page_sz = sysconf(_SC_PAGE_SIZE); | |
1435 | size_t map_sz; | |
1436 | ||
c701917e | 1437 | map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries; |
eba9c5f4 AN |
1438 | map_sz = roundup(map_sz, page_sz); |
1439 | return map_sz; | |
1440 | } | |
1441 | ||
aed65917 | 1442 | static char *internal_map_name(struct bpf_object *obj, const char *real_name) |
81bfdd08 | 1443 | { |
113e6b7e | 1444 | char map_name[BPF_OBJ_NAME_LEN], *p; |
aed65917 AN |
1445 | int pfx_len, sfx_len = max((size_t)7, strlen(real_name)); |
1446 | ||
1447 | /* This is one of the more confusing parts of libbpf for various | |
1448 | * reasons, some of which are historical. The original idea for naming | |
1449 | * internal names was to include as much of BPF object name prefix as | |
1450 | * possible, so that it can be distinguished from similar internal | |
1451 | * maps of a different BPF object. | |
1452 | * As an example, let's say we have bpf_object named 'my_object_name' | |
1453 | * and internal map corresponding to '.rodata' ELF section. The final | |
1454 | * map name advertised to user and to the kernel will be | |
1455 | * 'my_objec.rodata', taking first 8 characters of object name and | |
1456 | * entire 7 characters of '.rodata'. | |
1457 | * Somewhat confusingly, if internal map ELF section name is shorter | |
1458 | * than 7 characters, e.g., '.bss', we still reserve 7 characters | |
1459 | * for the suffix, even though we only have 4 actual characters, and | |
1460 | * resulting map will be called 'my_objec.bss', not even using all 15 | |
1461 | * characters allowed by the kernel. Oh well, at least the truncated | |
1462 | * object name is somewhat consistent in this case. But if the map | |
1463 | * name is '.kconfig', we'll still have entirety of '.kconfig' added | |
1464 | * (8 chars) and thus will be left with only first 7 characters of the | |
1465 | * object name ('my_obje'). Happy guessing, user, that the final map | |
1466 | * name will be "my_obje.kconfig". | |
1467 | * Now, with libbpf starting to support arbitrarily named .rodata.* | |
1468 | * and .data.* data sections, it's possible that ELF section name is | |
1469 | * longer than allowed 15 chars, so we now need to be careful to take | |
1470 | * only up to 15 first characters of ELF name, taking no BPF object | |
1471 | * name characters at all. So '.rodata.abracadabra' will result in | |
1472 | * '.rodata.abracad' kernel and user-visible name. | |
1473 | * We need to keep this convoluted logic intact for .data, .bss and | |
1474 | * .rodata maps, but for new custom .data.custom and .rodata.custom | |
1475 | * maps we use their ELF names as is, not prepending bpf_object name | |
1476 | * in front. We still need to truncate them to 15 characters for the | |
1477 | * kernel. Full name can be recovered for such maps by using DATASEC | |
1478 | * BTF type associated with such map's value type, though. | |
1479 | */ | |
1480 | if (sfx_len >= BPF_OBJ_NAME_LEN) | |
1481 | sfx_len = BPF_OBJ_NAME_LEN - 1; | |
1482 | ||
1483 | /* if there are two or more dots in map name, it's a custom dot map */ | |
1484 | if (strchr(real_name + 1, '.') != NULL) | |
1485 | pfx_len = 0; | |
1486 | else | |
1487 | pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name)); | |
81bfdd08 AN |
1488 | |
1489 | snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name, | |
aed65917 | 1490 | sfx_len, real_name); |
81bfdd08 | 1491 | |
113e6b7e THJ |
1492 | /* sanitise map name to characters allowed by kernel */ |
1493 | for (p = map_name; *p && p < map_name + sizeof(map_name); p++) | |
1494 | if (!isalnum(*p) && *p != '_' && *p != '.') | |
1495 | *p = '_'; | |
1496 | ||
81bfdd08 AN |
1497 | return strdup(map_name); |
1498 | } | |
1499 | ||
d859900c | 1500 | static int |
bf829271 | 1501 | bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type, |
aed65917 | 1502 | const char *real_name, int sec_idx, void *data, size_t data_sz) |
d859900c | 1503 | { |
bf829271 AN |
1504 | struct bpf_map_def *def; |
1505 | struct bpf_map *map; | |
eba9c5f4 | 1506 | int err; |
bf829271 AN |
1507 | |
1508 | map = bpf_object__add_map(obj); | |
1509 | if (IS_ERR(map)) | |
1510 | return PTR_ERR(map); | |
d859900c DB |
1511 | |
1512 | map->libbpf_type = type; | |
db48814b AN |
1513 | map->sec_idx = sec_idx; |
1514 | map->sec_offset = 0; | |
aed65917 AN |
1515 | map->real_name = strdup(real_name); |
1516 | map->name = internal_map_name(obj, real_name); | |
1517 | if (!map->real_name || !map->name) { | |
1518 | zfree(&map->real_name); | |
1519 | zfree(&map->name); | |
d859900c DB |
1520 | return -ENOMEM; |
1521 | } | |
1522 | ||
bf829271 | 1523 | def = &map->def; |
d859900c DB |
1524 | def->type = BPF_MAP_TYPE_ARRAY; |
1525 | def->key_size = sizeof(int); | |
eba9c5f4 | 1526 | def->value_size = data_sz; |
d859900c | 1527 | def->max_entries = 1; |
81bfdd08 | 1528 | def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG |
166750bc | 1529 | ? BPF_F_RDONLY_PROG : 0; |
0d13bfce | 1530 | def->map_flags |= BPF_F_MMAPABLE; |
7fe74b43 AN |
1531 | |
1532 | pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n", | |
81bfdd08 | 1533 | map->name, map->sec_idx, map->sec_offset, def->map_flags); |
7fe74b43 | 1534 | |
eba9c5f4 AN |
1535 | map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE, |
1536 | MAP_SHARED | MAP_ANONYMOUS, -1, 0); | |
1537 | if (map->mmaped == MAP_FAILED) { | |
1538 | err = -errno; | |
1539 | map->mmaped = NULL; | |
1540 | pr_warn("failed to alloc map '%s' content buffer: %d\n", | |
1541 | map->name, err); | |
aed65917 | 1542 | zfree(&map->real_name); |
eba9c5f4 AN |
1543 | zfree(&map->name); |
1544 | return err; | |
d859900c DB |
1545 | } |
1546 | ||
166750bc | 1547 | if (data) |
eba9c5f4 AN |
1548 | memcpy(map->mmaped, data, data_sz); |
1549 | ||
e1d1dc46 | 1550 | pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name); |
d859900c DB |
1551 | return 0; |
1552 | } | |
1553 | ||
bf829271 AN |
1554 | static int bpf_object__init_global_data_maps(struct bpf_object *obj) |
1555 | { | |
25bbbd7a | 1556 | struct elf_sec_desc *sec_desc; |
aed65917 | 1557 | const char *sec_name; |
25bbbd7a | 1558 | int err = 0, sec_idx; |
bf829271 | 1559 | |
bf829271 AN |
1560 | /* |
1561 | * Populate obj->maps with libbpf internal maps. | |
1562 | */ | |
25bbbd7a AN |
1563 | for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) { |
1564 | sec_desc = &obj->efile.secs[sec_idx]; | |
1565 | ||
1566 | switch (sec_desc->sec_type) { | |
1567 | case SEC_DATA: | |
aed65917 | 1568 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1569 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA, |
aed65917 | 1570 | sec_name, sec_idx, |
25bbbd7a AN |
1571 | sec_desc->data->d_buf, |
1572 | sec_desc->data->d_size); | |
1573 | break; | |
1574 | case SEC_RODATA: | |
1575 | obj->has_rodata = true; | |
aed65917 | 1576 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1577 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA, |
aed65917 | 1578 | sec_name, sec_idx, |
25bbbd7a AN |
1579 | sec_desc->data->d_buf, |
1580 | sec_desc->data->d_size); | |
1581 | break; | |
1582 | case SEC_BSS: | |
aed65917 | 1583 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1584 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS, |
aed65917 | 1585 | sec_name, sec_idx, |
25bbbd7a AN |
1586 | NULL, |
1587 | sec_desc->data->d_size); | |
1588 | break; | |
1589 | default: | |
1590 | /* skip */ | |
1591 | break; | |
1592 | } | |
bf829271 AN |
1593 | if (err) |
1594 | return err; | |
1595 | } | |
1596 | return 0; | |
1597 | } | |
1598 | ||
166750bc AN |
1599 | |
1600 | static struct extern_desc *find_extern_by_name(const struct bpf_object *obj, | |
1601 | const void *name) | |
1602 | { | |
1603 | int i; | |
1604 | ||
1605 | for (i = 0; i < obj->nr_extern; i++) { | |
1606 | if (strcmp(obj->externs[i].name, name) == 0) | |
1607 | return &obj->externs[i]; | |
1608 | } | |
1609 | return NULL; | |
1610 | } | |
1611 | ||
2e33efe3 AN |
1612 | static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val, |
1613 | char value) | |
166750bc | 1614 | { |
2e33efe3 AN |
1615 | switch (ext->kcfg.type) { |
1616 | case KCFG_BOOL: | |
166750bc | 1617 | if (value == 'm') { |
2e33efe3 | 1618 | pr_warn("extern (kcfg) %s=%c should be tristate or char\n", |
166750bc AN |
1619 | ext->name, value); |
1620 | return -EINVAL; | |
1621 | } | |
1622 | *(bool *)ext_val = value == 'y' ? true : false; | |
1623 | break; | |
2e33efe3 | 1624 | case KCFG_TRISTATE: |
166750bc AN |
1625 | if (value == 'y') |
1626 | *(enum libbpf_tristate *)ext_val = TRI_YES; | |
1627 | else if (value == 'm') | |
1628 | *(enum libbpf_tristate *)ext_val = TRI_MODULE; | |
1629 | else /* value == 'n' */ | |
1630 | *(enum libbpf_tristate *)ext_val = TRI_NO; | |
1631 | break; | |
2e33efe3 | 1632 | case KCFG_CHAR: |
166750bc AN |
1633 | *(char *)ext_val = value; |
1634 | break; | |
2e33efe3 AN |
1635 | case KCFG_UNKNOWN: |
1636 | case KCFG_INT: | |
1637 | case KCFG_CHAR_ARR: | |
166750bc | 1638 | default: |
2e33efe3 | 1639 | pr_warn("extern (kcfg) %s=%c should be bool, tristate, or char\n", |
166750bc AN |
1640 | ext->name, value); |
1641 | return -EINVAL; | |
1642 | } | |
1643 | ext->is_set = true; | |
1644 | return 0; | |
1645 | } | |
1646 | ||
2e33efe3 AN |
1647 | static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val, |
1648 | const char *value) | |
166750bc AN |
1649 | { |
1650 | size_t len; | |
1651 | ||
2e33efe3 AN |
1652 | if (ext->kcfg.type != KCFG_CHAR_ARR) { |
1653 | pr_warn("extern (kcfg) %s=%s should be char array\n", ext->name, value); | |
166750bc AN |
1654 | return -EINVAL; |
1655 | } | |
1656 | ||
1657 | len = strlen(value); | |
1658 | if (value[len - 1] != '"') { | |
2e33efe3 | 1659 | pr_warn("extern (kcfg) '%s': invalid string config '%s'\n", |
166750bc AN |
1660 | ext->name, value); |
1661 | return -EINVAL; | |
1662 | } | |
1663 | ||
1664 | /* strip quotes */ | |
1665 | len -= 2; | |
2e33efe3 AN |
1666 | if (len >= ext->kcfg.sz) { |
1667 | pr_warn("extern (kcfg) '%s': long string config %s of (%zu bytes) truncated to %d bytes\n", | |
1668 | ext->name, value, len, ext->kcfg.sz - 1); | |
1669 | len = ext->kcfg.sz - 1; | |
166750bc AN |
1670 | } |
1671 | memcpy(ext_val, value + 1, len); | |
1672 | ext_val[len] = '\0'; | |
1673 | ext->is_set = true; | |
1674 | return 0; | |
1675 | } | |
1676 | ||
1677 | static int parse_u64(const char *value, __u64 *res) | |
1678 | { | |
1679 | char *value_end; | |
1680 | int err; | |
1681 | ||
1682 | errno = 0; | |
1683 | *res = strtoull(value, &value_end, 0); | |
1684 | if (errno) { | |
1685 | err = -errno; | |
1686 | pr_warn("failed to parse '%s' as integer: %d\n", value, err); | |
1687 | return err; | |
1688 | } | |
1689 | if (*value_end) { | |
1690 | pr_warn("failed to parse '%s' as integer completely\n", value); | |
1691 | return -EINVAL; | |
1692 | } | |
1693 | return 0; | |
1694 | } | |
1695 | ||
2e33efe3 | 1696 | static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v) |
166750bc | 1697 | { |
2e33efe3 | 1698 | int bit_sz = ext->kcfg.sz * 8; |
166750bc | 1699 | |
2e33efe3 | 1700 | if (ext->kcfg.sz == 8) |
166750bc AN |
1701 | return true; |
1702 | ||
1703 | /* Validate that value stored in u64 fits in integer of `ext->sz` | |
1704 | * bytes size without any loss of information. If the target integer | |
1705 | * is signed, we rely on the following limits of integer type of | |
1706 | * Y bits and subsequent transformation: | |
1707 | * | |
1708 | * -2^(Y-1) <= X <= 2^(Y-1) - 1 | |
1709 | * 0 <= X + 2^(Y-1) <= 2^Y - 1 | |
1710 | * 0 <= X + 2^(Y-1) < 2^Y | |
1711 | * | |
1712 | * For unsigned target integer, check that all the (64 - Y) bits are | |
1713 | * zero. | |
1714 | */ | |
2e33efe3 | 1715 | if (ext->kcfg.is_signed) |
166750bc AN |
1716 | return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz); |
1717 | else | |
1718 | return (v >> bit_sz) == 0; | |
1719 | } | |
1720 | ||
2e33efe3 AN |
1721 | static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val, |
1722 | __u64 value) | |
166750bc | 1723 | { |
2e33efe3 AN |
1724 | if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) { |
1725 | pr_warn("extern (kcfg) %s=%llu should be integer\n", | |
7745ff98 | 1726 | ext->name, (unsigned long long)value); |
166750bc AN |
1727 | return -EINVAL; |
1728 | } | |
2e33efe3 AN |
1729 | if (!is_kcfg_value_in_range(ext, value)) { |
1730 | pr_warn("extern (kcfg) %s=%llu value doesn't fit in %d bytes\n", | |
1731 | ext->name, (unsigned long long)value, ext->kcfg.sz); | |
166750bc AN |
1732 | return -ERANGE; |
1733 | } | |
2e33efe3 | 1734 | switch (ext->kcfg.sz) { |
166750bc AN |
1735 | case 1: *(__u8 *)ext_val = value; break; |
1736 | case 2: *(__u16 *)ext_val = value; break; | |
1737 | case 4: *(__u32 *)ext_val = value; break; | |
1738 | case 8: *(__u64 *)ext_val = value; break; | |
1739 | default: | |
1740 | return -EINVAL; | |
1741 | } | |
1742 | ext->is_set = true; | |
1743 | return 0; | |
1744 | } | |
1745 | ||
8601fd42 AN |
1746 | static int bpf_object__process_kconfig_line(struct bpf_object *obj, |
1747 | char *buf, void *data) | |
166750bc | 1748 | { |
166750bc | 1749 | struct extern_desc *ext; |
8601fd42 | 1750 | char *sep, *value; |
166750bc AN |
1751 | int len, err = 0; |
1752 | void *ext_val; | |
1753 | __u64 num; | |
166750bc | 1754 | |
13d35a0c | 1755 | if (!str_has_pfx(buf, "CONFIG_")) |
8601fd42 | 1756 | return 0; |
166750bc | 1757 | |
8601fd42 AN |
1758 | sep = strchr(buf, '='); |
1759 | if (!sep) { | |
1760 | pr_warn("failed to parse '%s': no separator\n", buf); | |
1761 | return -EINVAL; | |
1762 | } | |
1763 | ||
1764 | /* Trim ending '\n' */ | |
1765 | len = strlen(buf); | |
1766 | if (buf[len - 1] == '\n') | |
1767 | buf[len - 1] = '\0'; | |
1768 | /* Split on '=' and ensure that a value is present. */ | |
1769 | *sep = '\0'; | |
1770 | if (!sep[1]) { | |
1771 | *sep = '='; | |
1772 | pr_warn("failed to parse '%s': no value\n", buf); | |
1773 | return -EINVAL; | |
1774 | } | |
1775 | ||
1776 | ext = find_extern_by_name(obj, buf); | |
1777 | if (!ext || ext->is_set) | |
1778 | return 0; | |
1779 | ||
2e33efe3 | 1780 | ext_val = data + ext->kcfg.data_off; |
8601fd42 AN |
1781 | value = sep + 1; |
1782 | ||
1783 | switch (*value) { | |
1784 | case 'y': case 'n': case 'm': | |
2e33efe3 | 1785 | err = set_kcfg_value_tri(ext, ext_val, *value); |
8601fd42 AN |
1786 | break; |
1787 | case '"': | |
2e33efe3 | 1788 | err = set_kcfg_value_str(ext, ext_val, value); |
8601fd42 AN |
1789 | break; |
1790 | default: | |
1791 | /* assume integer */ | |
1792 | err = parse_u64(value, &num); | |
1793 | if (err) { | |
2e33efe3 | 1794 | pr_warn("extern (kcfg) %s=%s should be integer\n", |
8601fd42 AN |
1795 | ext->name, value); |
1796 | return err; | |
1797 | } | |
2e33efe3 | 1798 | err = set_kcfg_value_num(ext, ext_val, num); |
8601fd42 | 1799 | break; |
166750bc | 1800 | } |
8601fd42 AN |
1801 | if (err) |
1802 | return err; | |
2e33efe3 | 1803 | pr_debug("extern (kcfg) %s=%s\n", ext->name, value); |
8601fd42 AN |
1804 | return 0; |
1805 | } | |
1806 | ||
1807 | static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data) | |
1808 | { | |
1809 | char buf[PATH_MAX]; | |
1810 | struct utsname uts; | |
1811 | int len, err = 0; | |
1812 | gzFile file; | |
1813 | ||
1814 | uname(&uts); | |
1815 | len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release); | |
1816 | if (len < 0) | |
1817 | return -EINVAL; | |
1818 | else if (len >= PATH_MAX) | |
1819 | return -ENAMETOOLONG; | |
1820 | ||
1821 | /* gzopen also accepts uncompressed files. */ | |
1822 | file = gzopen(buf, "r"); | |
1823 | if (!file) | |
1824 | file = gzopen("/proc/config.gz", "r"); | |
1825 | ||
166750bc | 1826 | if (!file) { |
8601fd42 | 1827 | pr_warn("failed to open system Kconfig\n"); |
166750bc AN |
1828 | return -ENOENT; |
1829 | } | |
1830 | ||
1831 | while (gzgets(file, buf, sizeof(buf))) { | |
8601fd42 AN |
1832 | err = bpf_object__process_kconfig_line(obj, buf, data); |
1833 | if (err) { | |
1834 | pr_warn("error parsing system Kconfig line '%s': %d\n", | |
1835 | buf, err); | |
166750bc AN |
1836 | goto out; |
1837 | } | |
8601fd42 | 1838 | } |
166750bc | 1839 | |
8601fd42 AN |
1840 | out: |
1841 | gzclose(file); | |
1842 | return err; | |
1843 | } | |
166750bc | 1844 | |
8601fd42 AN |
1845 | static int bpf_object__read_kconfig_mem(struct bpf_object *obj, |
1846 | const char *config, void *data) | |
1847 | { | |
1848 | char buf[PATH_MAX]; | |
1849 | int err = 0; | |
1850 | FILE *file; | |
166750bc | 1851 | |
8601fd42 AN |
1852 | file = fmemopen((void *)config, strlen(config), "r"); |
1853 | if (!file) { | |
1854 | err = -errno; | |
1855 | pr_warn("failed to open in-memory Kconfig: %d\n", err); | |
1856 | return err; | |
1857 | } | |
1858 | ||
1859 | while (fgets(buf, sizeof(buf), file)) { | |
1860 | err = bpf_object__process_kconfig_line(obj, buf, data); | |
1861 | if (err) { | |
1862 | pr_warn("error parsing in-memory Kconfig line '%s': %d\n", | |
1863 | buf, err); | |
166750bc AN |
1864 | break; |
1865 | } | |
166750bc AN |
1866 | } |
1867 | ||
8601fd42 | 1868 | fclose(file); |
166750bc AN |
1869 | return err; |
1870 | } | |
1871 | ||
81bfdd08 | 1872 | static int bpf_object__init_kconfig_map(struct bpf_object *obj) |
166750bc | 1873 | { |
2e33efe3 | 1874 | struct extern_desc *last_ext = NULL, *ext; |
166750bc | 1875 | size_t map_sz; |
2e33efe3 | 1876 | int i, err; |
166750bc | 1877 | |
2e33efe3 AN |
1878 | for (i = 0; i < obj->nr_extern; i++) { |
1879 | ext = &obj->externs[i]; | |
1880 | if (ext->type == EXT_KCFG) | |
1881 | last_ext = ext; | |
1882 | } | |
166750bc | 1883 | |
2e33efe3 AN |
1884 | if (!last_ext) |
1885 | return 0; | |
166750bc | 1886 | |
2e33efe3 | 1887 | map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz; |
81bfdd08 | 1888 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG, |
aed65917 | 1889 | ".kconfig", obj->efile.symbols_shndx, |
166750bc AN |
1890 | NULL, map_sz); |
1891 | if (err) | |
1892 | return err; | |
1893 | ||
81bfdd08 | 1894 | obj->kconfig_map_idx = obj->nr_maps - 1; |
166750bc AN |
1895 | |
1896 | return 0; | |
1897 | } | |
1898 | ||
bf829271 | 1899 | static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict) |
561bbcca | 1900 | { |
561bbcca | 1901 | Elf_Data *symbols = obj->efile.symbols; |
bf829271 | 1902 | int i, map_def_sz = 0, nr_maps = 0, nr_syms; |
d859900c | 1903 | Elf_Data *data = NULL; |
bf829271 AN |
1904 | Elf_Scn *scn; |
1905 | ||
1906 | if (obj->efile.maps_shndx < 0) | |
1907 | return 0; | |
561bbcca | 1908 | |
4708bbda EL |
1909 | if (!symbols) |
1910 | return -EINVAL; | |
1911 | ||
88a82120 AN |
1912 | scn = elf_sec_by_idx(obj, obj->efile.maps_shndx); |
1913 | data = elf_sec_data(obj, scn); | |
bf829271 | 1914 | if (!scn || !data) { |
88a82120 AN |
1915 | pr_warn("elf: failed to get legacy map definitions for %s\n", |
1916 | obj->path); | |
bf829271 | 1917 | return -EINVAL; |
4708bbda | 1918 | } |
561bbcca | 1919 | |
4708bbda EL |
1920 | /* |
1921 | * Count number of maps. Each map has a name. | |
1922 | * Array of maps is not supported: only the first element is | |
1923 | * considered. | |
1924 | * | |
1925 | * TODO: Detect array of map and report error. | |
1926 | */ | |
ad23b723 | 1927 | nr_syms = symbols->d_size / sizeof(Elf64_Sym); |
bf829271 | 1928 | for (i = 0; i < nr_syms; i++) { |
ad23b723 | 1929 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
4708bbda | 1930 | |
ad23b723 | 1931 | if (sym->st_shndx != obj->efile.maps_shndx) |
4708bbda | 1932 | continue; |
ad23b723 | 1933 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION) |
161ecd53 | 1934 | continue; |
4708bbda EL |
1935 | nr_maps++; |
1936 | } | |
b13c5c14 | 1937 | /* Assume equally sized map definitions */ |
88a82120 AN |
1938 | pr_debug("elf: found %d legacy map definitions (%zd bytes) in %s\n", |
1939 | nr_maps, data->d_size, obj->path); | |
bf829271 | 1940 | |
98e527af | 1941 | if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) { |
88a82120 AN |
1942 | pr_warn("elf: unable to determine legacy map definition size in %s\n", |
1943 | obj->path); | |
bf829271 | 1944 | return -EINVAL; |
addb9fc9 | 1945 | } |
98e527af | 1946 | map_def_sz = data->d_size / nr_maps; |
4708bbda | 1947 | |
bf829271 AN |
1948 | /* Fill obj->maps using data in "maps" section. */ |
1949 | for (i = 0; i < nr_syms; i++) { | |
ad23b723 | 1950 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
561bbcca | 1951 | const char *map_name; |
4708bbda | 1952 | struct bpf_map_def *def; |
bf829271 | 1953 | struct bpf_map *map; |
561bbcca | 1954 | |
ad23b723 | 1955 | if (sym->st_shndx != obj->efile.maps_shndx) |
561bbcca | 1956 | continue; |
ad23b723 | 1957 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION) |
c3e8c44a | 1958 | continue; |
561bbcca | 1959 | |
bf829271 AN |
1960 | map = bpf_object__add_map(obj); |
1961 | if (IS_ERR(map)) | |
1962 | return PTR_ERR(map); | |
1963 | ||
ad23b723 | 1964 | map_name = elf_sym_str(obj, sym->st_name); |
c51829bb | 1965 | if (!map_name) { |
be18010e KW |
1966 | pr_warn("failed to get map #%d name sym string for obj %s\n", |
1967 | i, obj->path); | |
c51829bb AN |
1968 | return -LIBBPF_ERRNO__FORMAT; |
1969 | } | |
d859900c | 1970 | |
ad23b723 | 1971 | if (ELF64_ST_BIND(sym->st_info) == STB_LOCAL) { |
c1cccec9 AN |
1972 | pr_warn("map '%s' (legacy): static maps are not supported\n", map_name); |
1973 | return -ENOTSUP; | |
1974 | } | |
1975 | ||
bf829271 | 1976 | map->libbpf_type = LIBBPF_MAP_UNSPEC; |
ad23b723 AN |
1977 | map->sec_idx = sym->st_shndx; |
1978 | map->sec_offset = sym->st_value; | |
db48814b AN |
1979 | pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n", |
1980 | map_name, map->sec_idx, map->sec_offset); | |
ad23b723 | 1981 | if (sym->st_value + map_def_sz > data->d_size) { |
be18010e KW |
1982 | pr_warn("corrupted maps section in %s: last map \"%s\" too small\n", |
1983 | obj->path, map_name); | |
4708bbda | 1984 | return -EINVAL; |
561bbcca | 1985 | } |
4708bbda | 1986 | |
bf829271 AN |
1987 | map->name = strdup(map_name); |
1988 | if (!map->name) { | |
aed65917 | 1989 | pr_warn("map '%s': failed to alloc map name\n", map_name); |
973170e6 WN |
1990 | return -ENOMEM; |
1991 | } | |
bf829271 | 1992 | pr_debug("map %d is \"%s\"\n", i, map->name); |
ad23b723 | 1993 | def = (struct bpf_map_def *)(data->d_buf + sym->st_value); |
b13c5c14 CG |
1994 | /* |
1995 | * If the definition of the map in the object file fits in | |
1996 | * bpf_map_def, copy it. Any extra fields in our version | |
1997 | * of bpf_map_def will default to zero as a result of the | |
1998 | * calloc above. | |
1999 | */ | |
2000 | if (map_def_sz <= sizeof(struct bpf_map_def)) { | |
bf829271 | 2001 | memcpy(&map->def, def, map_def_sz); |
b13c5c14 CG |
2002 | } else { |
2003 | /* | |
2004 | * Here the map structure being read is bigger than what | |
2005 | * we expect, truncate if the excess bits are all zero. | |
2006 | * If they are not zero, reject this map as | |
2007 | * incompatible. | |
2008 | */ | |
2009 | char *b; | |
8983b731 | 2010 | |
b13c5c14 CG |
2011 | for (b = ((char *)def) + sizeof(struct bpf_map_def); |
2012 | b < ((char *)def) + map_def_sz; b++) { | |
2013 | if (*b != 0) { | |
8983b731 | 2014 | pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n", |
be18010e | 2015 | obj->path, map_name); |
c034a177 JF |
2016 | if (strict) |
2017 | return -EINVAL; | |
b13c5c14 CG |
2018 | } |
2019 | } | |
bf829271 | 2020 | memcpy(&map->def, def, sizeof(struct bpf_map_def)); |
b13c5c14 | 2021 | } |
561bbcca | 2022 | } |
bf829271 AN |
2023 | return 0; |
2024 | } | |
4708bbda | 2025 | |
42869d28 | 2026 | const struct btf_type * |
ddc7c304 | 2027 | skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id) |
abd29c93 AN |
2028 | { |
2029 | const struct btf_type *t = btf__type_by_id(btf, id); | |
8837fe5d | 2030 | |
ddc7c304 AN |
2031 | if (res_id) |
2032 | *res_id = id; | |
2033 | ||
2034 | while (btf_is_mod(t) || btf_is_typedef(t)) { | |
2035 | if (res_id) | |
2036 | *res_id = t->type; | |
2037 | t = btf__type_by_id(btf, t->type); | |
abd29c93 | 2038 | } |
ddc7c304 AN |
2039 | |
2040 | return t; | |
abd29c93 AN |
2041 | } |
2042 | ||
590a0088 MKL |
2043 | static const struct btf_type * |
2044 | resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id) | |
2045 | { | |
2046 | const struct btf_type *t; | |
2047 | ||
2048 | t = skip_mods_and_typedefs(btf, id, NULL); | |
2049 | if (!btf_is_ptr(t)) | |
2050 | return NULL; | |
2051 | ||
2052 | t = skip_mods_and_typedefs(btf, t->type, res_id); | |
2053 | ||
2054 | return btf_is_func_proto(t) ? t : NULL; | |
2055 | } | |
2056 | ||
774e132e | 2057 | static const char *__btf_kind_str(__u16 kind) |
81ba0889 | 2058 | { |
774e132e | 2059 | switch (kind) { |
81ba0889 AN |
2060 | case BTF_KIND_UNKN: return "void"; |
2061 | case BTF_KIND_INT: return "int"; | |
2062 | case BTF_KIND_PTR: return "ptr"; | |
2063 | case BTF_KIND_ARRAY: return "array"; | |
2064 | case BTF_KIND_STRUCT: return "struct"; | |
2065 | case BTF_KIND_UNION: return "union"; | |
2066 | case BTF_KIND_ENUM: return "enum"; | |
2067 | case BTF_KIND_FWD: return "fwd"; | |
2068 | case BTF_KIND_TYPEDEF: return "typedef"; | |
2069 | case BTF_KIND_VOLATILE: return "volatile"; | |
2070 | case BTF_KIND_CONST: return "const"; | |
2071 | case BTF_KIND_RESTRICT: return "restrict"; | |
2072 | case BTF_KIND_FUNC: return "func"; | |
2073 | case BTF_KIND_FUNC_PROTO: return "func_proto"; | |
2074 | case BTF_KIND_VAR: return "var"; | |
2075 | case BTF_KIND_DATASEC: return "datasec"; | |
22541a9e | 2076 | case BTF_KIND_FLOAT: return "float"; |
223f903e | 2077 | case BTF_KIND_DECL_TAG: return "decl_tag"; |
81ba0889 AN |
2078 | default: return "unknown"; |
2079 | } | |
2080 | } | |
2081 | ||
42869d28 | 2082 | const char *btf_kind_str(const struct btf_type *t) |
774e132e MKL |
2083 | { |
2084 | return __btf_kind_str(btf_kind(t)); | |
2085 | } | |
2086 | ||
ef99b02b AN |
2087 | /* |
2088 | * Fetch integer attribute of BTF map definition. Such attributes are | |
2089 | * represented using a pointer to an array, in which dimensionality of array | |
2090 | * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY]; | |
2091 | * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF | |
2092 | * type definition, while using only sizeof(void *) space in ELF data section. | |
2093 | */ | |
2094 | static bool get_map_field_int(const char *map_name, const struct btf *btf, | |
8983b731 AN |
2095 | const struct btf_member *m, __u32 *res) |
2096 | { | |
ddc7c304 | 2097 | const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL); |
abd29c93 | 2098 | const char *name = btf__name_by_offset(btf, m->name_off); |
ef99b02b AN |
2099 | const struct btf_array *arr_info; |
2100 | const struct btf_type *arr_t; | |
abd29c93 | 2101 | |
b03bc685 | 2102 | if (!btf_is_ptr(t)) { |
81ba0889 AN |
2103 | pr_warn("map '%s': attr '%s': expected PTR, got %s.\n", |
2104 | map_name, name, btf_kind_str(t)); | |
abd29c93 AN |
2105 | return false; |
2106 | } | |
ef99b02b AN |
2107 | |
2108 | arr_t = btf__type_by_id(btf, t->type); | |
2109 | if (!arr_t) { | |
be18010e KW |
2110 | pr_warn("map '%s': attr '%s': type [%u] not found.\n", |
2111 | map_name, name, t->type); | |
abd29c93 AN |
2112 | return false; |
2113 | } | |
b03bc685 | 2114 | if (!btf_is_array(arr_t)) { |
81ba0889 AN |
2115 | pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n", |
2116 | map_name, name, btf_kind_str(arr_t)); | |
abd29c93 AN |
2117 | return false; |
2118 | } | |
b03bc685 | 2119 | arr_info = btf_array(arr_t); |
ef99b02b | 2120 | *res = arr_info->nelems; |
abd29c93 AN |
2121 | return true; |
2122 | } | |
2123 | ||
57a00f41 THJ |
2124 | static int build_map_pin_path(struct bpf_map *map, const char *path) |
2125 | { | |
2126 | char buf[PATH_MAX]; | |
6e9cab2e | 2127 | int len; |
57a00f41 THJ |
2128 | |
2129 | if (!path) | |
2130 | path = "/sys/fs/bpf"; | |
2131 | ||
2132 | len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map)); | |
2133 | if (len < 0) | |
2134 | return -EINVAL; | |
2135 | else if (len >= PATH_MAX) | |
2136 | return -ENAMETOOLONG; | |
2137 | ||
6e9cab2e | 2138 | return bpf_map__set_pin_path(map, buf); |
57a00f41 THJ |
2139 | } |
2140 | ||
c7ef5ec9 AN |
2141 | int parse_btf_map_def(const char *map_name, struct btf *btf, |
2142 | const struct btf_type *def_t, bool strict, | |
2143 | struct btf_map_def *map_def, struct btf_map_def *inner_def) | |
abd29c93 | 2144 | { |
41017e56 | 2145 | const struct btf_type *t; |
abd29c93 | 2146 | const struct btf_member *m; |
c7ef5ec9 | 2147 | bool is_inner = inner_def == NULL; |
abd29c93 AN |
2148 | int vlen, i; |
2149 | ||
c7ef5ec9 AN |
2150 | vlen = btf_vlen(def_t); |
2151 | m = btf_members(def_t); | |
abd29c93 | 2152 | for (i = 0; i < vlen; i++, m++) { |
c7ef5ec9 | 2153 | const char *name = btf__name_by_offset(btf, m->name_off); |
abd29c93 AN |
2154 | |
2155 | if (!name) { | |
c7ef5ec9 | 2156 | pr_warn("map '%s': invalid field #%d.\n", map_name, i); |
abd29c93 AN |
2157 | return -EINVAL; |
2158 | } | |
2159 | if (strcmp(name, "type") == 0) { | |
c7ef5ec9 | 2160 | if (!get_map_field_int(map_name, btf, m, &map_def->map_type)) |
abd29c93 | 2161 | return -EINVAL; |
c7ef5ec9 | 2162 | map_def->parts |= MAP_DEF_MAP_TYPE; |
abd29c93 | 2163 | } else if (strcmp(name, "max_entries") == 0) { |
c7ef5ec9 | 2164 | if (!get_map_field_int(map_name, btf, m, &map_def->max_entries)) |
abd29c93 | 2165 | return -EINVAL; |
c7ef5ec9 | 2166 | map_def->parts |= MAP_DEF_MAX_ENTRIES; |
abd29c93 | 2167 | } else if (strcmp(name, "map_flags") == 0) { |
c7ef5ec9 | 2168 | if (!get_map_field_int(map_name, btf, m, &map_def->map_flags)) |
abd29c93 | 2169 | return -EINVAL; |
c7ef5ec9 | 2170 | map_def->parts |= MAP_DEF_MAP_FLAGS; |
1bdb6c9a | 2171 | } else if (strcmp(name, "numa_node") == 0) { |
c7ef5ec9 | 2172 | if (!get_map_field_int(map_name, btf, m, &map_def->numa_node)) |
1bdb6c9a | 2173 | return -EINVAL; |
c7ef5ec9 | 2174 | map_def->parts |= MAP_DEF_NUMA_NODE; |
abd29c93 AN |
2175 | } else if (strcmp(name, "key_size") == 0) { |
2176 | __u32 sz; | |
2177 | ||
c7ef5ec9 | 2178 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2179 | return -EINVAL; |
c7ef5ec9 | 2180 | if (map_def->key_size && map_def->key_size != sz) { |
be18010e | 2181 | pr_warn("map '%s': conflicting key size %u != %u.\n", |
c7ef5ec9 | 2182 | map_name, map_def->key_size, sz); |
abd29c93 AN |
2183 | return -EINVAL; |
2184 | } | |
c7ef5ec9 AN |
2185 | map_def->key_size = sz; |
2186 | map_def->parts |= MAP_DEF_KEY_SIZE; | |
abd29c93 AN |
2187 | } else if (strcmp(name, "key") == 0) { |
2188 | __s64 sz; | |
2189 | ||
c7ef5ec9 | 2190 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2191 | if (!t) { |
be18010e | 2192 | pr_warn("map '%s': key type [%d] not found.\n", |
c7ef5ec9 | 2193 | map_name, m->type); |
abd29c93 AN |
2194 | return -EINVAL; |
2195 | } | |
b03bc685 | 2196 | if (!btf_is_ptr(t)) { |
81ba0889 | 2197 | pr_warn("map '%s': key spec is not PTR: %s.\n", |
c7ef5ec9 | 2198 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2199 | return -EINVAL; |
2200 | } | |
c7ef5ec9 | 2201 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2202 | if (sz < 0) { |
679152d3 | 2203 | pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n", |
c7ef5ec9 | 2204 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2205 | return sz; |
2206 | } | |
c7ef5ec9 | 2207 | if (map_def->key_size && map_def->key_size != sz) { |
679152d3 | 2208 | pr_warn("map '%s': conflicting key size %u != %zd.\n", |
c7ef5ec9 | 2209 | map_name, map_def->key_size, (ssize_t)sz); |
abd29c93 AN |
2210 | return -EINVAL; |
2211 | } | |
c7ef5ec9 AN |
2212 | map_def->key_size = sz; |
2213 | map_def->key_type_id = t->type; | |
2214 | map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE; | |
abd29c93 AN |
2215 | } else if (strcmp(name, "value_size") == 0) { |
2216 | __u32 sz; | |
2217 | ||
c7ef5ec9 | 2218 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2219 | return -EINVAL; |
c7ef5ec9 | 2220 | if (map_def->value_size && map_def->value_size != sz) { |
be18010e | 2221 | pr_warn("map '%s': conflicting value size %u != %u.\n", |
c7ef5ec9 | 2222 | map_name, map_def->value_size, sz); |
abd29c93 AN |
2223 | return -EINVAL; |
2224 | } | |
c7ef5ec9 AN |
2225 | map_def->value_size = sz; |
2226 | map_def->parts |= MAP_DEF_VALUE_SIZE; | |
abd29c93 AN |
2227 | } else if (strcmp(name, "value") == 0) { |
2228 | __s64 sz; | |
2229 | ||
c7ef5ec9 | 2230 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2231 | if (!t) { |
be18010e | 2232 | pr_warn("map '%s': value type [%d] not found.\n", |
c7ef5ec9 | 2233 | map_name, m->type); |
abd29c93 AN |
2234 | return -EINVAL; |
2235 | } | |
b03bc685 | 2236 | if (!btf_is_ptr(t)) { |
81ba0889 | 2237 | pr_warn("map '%s': value spec is not PTR: %s.\n", |
c7ef5ec9 | 2238 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2239 | return -EINVAL; |
2240 | } | |
c7ef5ec9 | 2241 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2242 | if (sz < 0) { |
679152d3 | 2243 | pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n", |
c7ef5ec9 | 2244 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2245 | return sz; |
2246 | } | |
c7ef5ec9 | 2247 | if (map_def->value_size && map_def->value_size != sz) { |
679152d3 | 2248 | pr_warn("map '%s': conflicting value size %u != %zd.\n", |
c7ef5ec9 | 2249 | map_name, map_def->value_size, (ssize_t)sz); |
abd29c93 AN |
2250 | return -EINVAL; |
2251 | } | |
c7ef5ec9 AN |
2252 | map_def->value_size = sz; |
2253 | map_def->value_type_id = t->type; | |
2254 | map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE; | |
646f02ff AN |
2255 | } |
2256 | else if (strcmp(name, "values") == 0) { | |
c7ef5ec9 | 2257 | char inner_map_name[128]; |
646f02ff AN |
2258 | int err; |
2259 | ||
2260 | if (is_inner) { | |
2261 | pr_warn("map '%s': multi-level inner maps not supported.\n", | |
c7ef5ec9 | 2262 | map_name); |
646f02ff AN |
2263 | return -ENOTSUP; |
2264 | } | |
2265 | if (i != vlen - 1) { | |
2266 | pr_warn("map '%s': '%s' member should be last.\n", | |
c7ef5ec9 | 2267 | map_name, name); |
646f02ff AN |
2268 | return -EINVAL; |
2269 | } | |
c7ef5ec9 | 2270 | if (!bpf_map_type__is_map_in_map(map_def->map_type)) { |
646f02ff | 2271 | pr_warn("map '%s': should be map-in-map.\n", |
c7ef5ec9 | 2272 | map_name); |
646f02ff AN |
2273 | return -ENOTSUP; |
2274 | } | |
c7ef5ec9 | 2275 | if (map_def->value_size && map_def->value_size != 4) { |
646f02ff | 2276 | pr_warn("map '%s': conflicting value size %u != 4.\n", |
c7ef5ec9 | 2277 | map_name, map_def->value_size); |
646f02ff AN |
2278 | return -EINVAL; |
2279 | } | |
c7ef5ec9 AN |
2280 | map_def->value_size = 4; |
2281 | t = btf__type_by_id(btf, m->type); | |
646f02ff AN |
2282 | if (!t) { |
2283 | pr_warn("map '%s': map-in-map inner type [%d] not found.\n", | |
c7ef5ec9 | 2284 | map_name, m->type); |
646f02ff AN |
2285 | return -EINVAL; |
2286 | } | |
2287 | if (!btf_is_array(t) || btf_array(t)->nelems) { | |
2288 | pr_warn("map '%s': map-in-map inner spec is not a zero-sized array.\n", | |
c7ef5ec9 | 2289 | map_name); |
646f02ff AN |
2290 | return -EINVAL; |
2291 | } | |
c7ef5ec9 | 2292 | t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL); |
646f02ff | 2293 | if (!btf_is_ptr(t)) { |
81ba0889 | 2294 | pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n", |
c7ef5ec9 | 2295 | map_name, btf_kind_str(t)); |
646f02ff AN |
2296 | return -EINVAL; |
2297 | } | |
c7ef5ec9 | 2298 | t = skip_mods_and_typedefs(btf, t->type, NULL); |
646f02ff | 2299 | if (!btf_is_struct(t)) { |
81ba0889 | 2300 | pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n", |
c7ef5ec9 | 2301 | map_name, btf_kind_str(t)); |
646f02ff AN |
2302 | return -EINVAL; |
2303 | } | |
2304 | ||
c7ef5ec9 AN |
2305 | snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name); |
2306 | err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL); | |
646f02ff AN |
2307 | if (err) |
2308 | return err; | |
c7ef5ec9 AN |
2309 | |
2310 | map_def->parts |= MAP_DEF_INNER_MAP; | |
57a00f41 THJ |
2311 | } else if (strcmp(name, "pinning") == 0) { |
2312 | __u32 val; | |
57a00f41 | 2313 | |
646f02ff | 2314 | if (is_inner) { |
c7ef5ec9 | 2315 | pr_warn("map '%s': inner def can't be pinned.\n", map_name); |
646f02ff AN |
2316 | return -EINVAL; |
2317 | } | |
c7ef5ec9 | 2318 | if (!get_map_field_int(map_name, btf, m, &val)) |
57a00f41 | 2319 | return -EINVAL; |
c7ef5ec9 | 2320 | if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) { |
57a00f41 | 2321 | pr_warn("map '%s': invalid pinning value %u.\n", |
c7ef5ec9 | 2322 | map_name, val); |
57a00f41 THJ |
2323 | return -EINVAL; |
2324 | } | |
c7ef5ec9 AN |
2325 | map_def->pinning = val; |
2326 | map_def->parts |= MAP_DEF_PINNING; | |
abd29c93 AN |
2327 | } else { |
2328 | if (strict) { | |
c7ef5ec9 | 2329 | pr_warn("map '%s': unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2330 | return -ENOTSUP; |
2331 | } | |
c7ef5ec9 | 2332 | pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2333 | } |
2334 | } | |
2335 | ||
c7ef5ec9 AN |
2336 | if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) { |
2337 | pr_warn("map '%s': map type isn't specified.\n", map_name); | |
abd29c93 AN |
2338 | return -EINVAL; |
2339 | } | |
2340 | ||
2341 | return 0; | |
2342 | } | |
2343 | ||
c7ef5ec9 AN |
2344 | static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def) |
2345 | { | |
2346 | map->def.type = def->map_type; | |
2347 | map->def.key_size = def->key_size; | |
2348 | map->def.value_size = def->value_size; | |
2349 | map->def.max_entries = def->max_entries; | |
2350 | map->def.map_flags = def->map_flags; | |
2351 | ||
2352 | map->numa_node = def->numa_node; | |
2353 | map->btf_key_type_id = def->key_type_id; | |
2354 | map->btf_value_type_id = def->value_type_id; | |
2355 | ||
2356 | if (def->parts & MAP_DEF_MAP_TYPE) | |
2357 | pr_debug("map '%s': found type = %u.\n", map->name, def->map_type); | |
2358 | ||
2359 | if (def->parts & MAP_DEF_KEY_TYPE) | |
2360 | pr_debug("map '%s': found key [%u], sz = %u.\n", | |
2361 | map->name, def->key_type_id, def->key_size); | |
2362 | else if (def->parts & MAP_DEF_KEY_SIZE) | |
2363 | pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size); | |
2364 | ||
2365 | if (def->parts & MAP_DEF_VALUE_TYPE) | |
2366 | pr_debug("map '%s': found value [%u], sz = %u.\n", | |
2367 | map->name, def->value_type_id, def->value_size); | |
2368 | else if (def->parts & MAP_DEF_VALUE_SIZE) | |
2369 | pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size); | |
2370 | ||
2371 | if (def->parts & MAP_DEF_MAX_ENTRIES) | |
2372 | pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries); | |
2373 | if (def->parts & MAP_DEF_MAP_FLAGS) | |
2374 | pr_debug("map '%s': found map_flags = %u.\n", map->name, def->map_flags); | |
2375 | if (def->parts & MAP_DEF_PINNING) | |
2376 | pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning); | |
2377 | if (def->parts & MAP_DEF_NUMA_NODE) | |
2378 | pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node); | |
2379 | ||
2380 | if (def->parts & MAP_DEF_INNER_MAP) | |
2381 | pr_debug("map '%s': found inner map definition.\n", map->name); | |
2382 | } | |
2383 | ||
c1cccec9 AN |
2384 | static const char *btf_var_linkage_str(__u32 linkage) |
2385 | { | |
2386 | switch (linkage) { | |
2387 | case BTF_VAR_STATIC: return "static"; | |
2388 | case BTF_VAR_GLOBAL_ALLOCATED: return "global"; | |
2389 | case BTF_VAR_GLOBAL_EXTERN: return "extern"; | |
2390 | default: return "unknown"; | |
2391 | } | |
2392 | } | |
2393 | ||
41017e56 AN |
2394 | static int bpf_object__init_user_btf_map(struct bpf_object *obj, |
2395 | const struct btf_type *sec, | |
2396 | int var_idx, int sec_idx, | |
2397 | const Elf_Data *data, bool strict, | |
2398 | const char *pin_root_path) | |
2399 | { | |
c7ef5ec9 | 2400 | struct btf_map_def map_def = {}, inner_def = {}; |
41017e56 AN |
2401 | const struct btf_type *var, *def; |
2402 | const struct btf_var_secinfo *vi; | |
2403 | const struct btf_var *var_extra; | |
2404 | const char *map_name; | |
2405 | struct bpf_map *map; | |
c7ef5ec9 | 2406 | int err; |
41017e56 AN |
2407 | |
2408 | vi = btf_var_secinfos(sec) + var_idx; | |
2409 | var = btf__type_by_id(obj->btf, vi->type); | |
2410 | var_extra = btf_var(var); | |
2411 | map_name = btf__name_by_offset(obj->btf, var->name_off); | |
2412 | ||
2413 | if (map_name == NULL || map_name[0] == '\0') { | |
2414 | pr_warn("map #%d: empty name.\n", var_idx); | |
2415 | return -EINVAL; | |
2416 | } | |
2417 | if ((__u64)vi->offset + vi->size > data->d_size) { | |
2418 | pr_warn("map '%s' BTF data is corrupted.\n", map_name); | |
2419 | return -EINVAL; | |
2420 | } | |
2421 | if (!btf_is_var(var)) { | |
81ba0889 AN |
2422 | pr_warn("map '%s': unexpected var kind %s.\n", |
2423 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2424 | return -EINVAL; |
2425 | } | |
c1cccec9 AN |
2426 | if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) { |
2427 | pr_warn("map '%s': unsupported map linkage %s.\n", | |
2428 | map_name, btf_var_linkage_str(var_extra->linkage)); | |
41017e56 AN |
2429 | return -EOPNOTSUPP; |
2430 | } | |
2431 | ||
2432 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
2433 | if (!btf_is_struct(def)) { | |
81ba0889 AN |
2434 | pr_warn("map '%s': unexpected def kind %s.\n", |
2435 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2436 | return -EINVAL; |
2437 | } | |
2438 | if (def->size > vi->size) { | |
2439 | pr_warn("map '%s': invalid def size.\n", map_name); | |
2440 | return -EINVAL; | |
2441 | } | |
2442 | ||
2443 | map = bpf_object__add_map(obj); | |
2444 | if (IS_ERR(map)) | |
2445 | return PTR_ERR(map); | |
2446 | map->name = strdup(map_name); | |
2447 | if (!map->name) { | |
2448 | pr_warn("map '%s': failed to alloc map name.\n", map_name); | |
2449 | return -ENOMEM; | |
2450 | } | |
2451 | map->libbpf_type = LIBBPF_MAP_UNSPEC; | |
2452 | map->def.type = BPF_MAP_TYPE_UNSPEC; | |
2453 | map->sec_idx = sec_idx; | |
2454 | map->sec_offset = vi->offset; | |
646f02ff | 2455 | map->btf_var_idx = var_idx; |
41017e56 AN |
2456 | pr_debug("map '%s': at sec_idx %d, offset %zu.\n", |
2457 | map_name, map->sec_idx, map->sec_offset); | |
2458 | ||
c7ef5ec9 AN |
2459 | err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def); |
2460 | if (err) | |
2461 | return err; | |
2462 | ||
2463 | fill_map_from_def(map, &map_def); | |
2464 | ||
2465 | if (map_def.pinning == LIBBPF_PIN_BY_NAME) { | |
2466 | err = build_map_pin_path(map, pin_root_path); | |
2467 | if (err) { | |
2468 | pr_warn("map '%s': couldn't build pin path.\n", map->name); | |
2469 | return err; | |
2470 | } | |
2471 | } | |
2472 | ||
2473 | if (map_def.parts & MAP_DEF_INNER_MAP) { | |
2474 | map->inner_map = calloc(1, sizeof(*map->inner_map)); | |
2475 | if (!map->inner_map) | |
2476 | return -ENOMEM; | |
2477 | map->inner_map->fd = -1; | |
2478 | map->inner_map->sec_idx = sec_idx; | |
2479 | map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1); | |
2480 | if (!map->inner_map->name) | |
2481 | return -ENOMEM; | |
2482 | sprintf(map->inner_map->name, "%s.inner", map_name); | |
2483 | ||
2484 | fill_map_from_def(map->inner_map, &inner_def); | |
2485 | } | |
2486 | ||
2487 | return 0; | |
41017e56 AN |
2488 | } |
2489 | ||
57a00f41 THJ |
2490 | static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict, |
2491 | const char *pin_root_path) | |
abd29c93 AN |
2492 | { |
2493 | const struct btf_type *sec = NULL; | |
2494 | int nr_types, i, vlen, err; | |
2495 | const struct btf_type *t; | |
2496 | const char *name; | |
2497 | Elf_Data *data; | |
2498 | Elf_Scn *scn; | |
2499 | ||
2500 | if (obj->efile.btf_maps_shndx < 0) | |
2501 | return 0; | |
2502 | ||
88a82120 AN |
2503 | scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx); |
2504 | data = elf_sec_data(obj, scn); | |
abd29c93 | 2505 | if (!scn || !data) { |
88a82120 AN |
2506 | pr_warn("elf: failed to get %s map definitions for %s\n", |
2507 | MAPS_ELF_SEC, obj->path); | |
abd29c93 AN |
2508 | return -EINVAL; |
2509 | } | |
2510 | ||
6a886de0 HC |
2511 | nr_types = btf__type_cnt(obj->btf); |
2512 | for (i = 1; i < nr_types; i++) { | |
abd29c93 | 2513 | t = btf__type_by_id(obj->btf, i); |
b03bc685 | 2514 | if (!btf_is_datasec(t)) |
abd29c93 AN |
2515 | continue; |
2516 | name = btf__name_by_offset(obj->btf, t->name_off); | |
2517 | if (strcmp(name, MAPS_ELF_SEC) == 0) { | |
2518 | sec = t; | |
646f02ff | 2519 | obj->efile.btf_maps_sec_btf_id = i; |
abd29c93 AN |
2520 | break; |
2521 | } | |
2522 | } | |
2523 | ||
2524 | if (!sec) { | |
be18010e | 2525 | pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC); |
abd29c93 AN |
2526 | return -ENOENT; |
2527 | } | |
2528 | ||
b03bc685 | 2529 | vlen = btf_vlen(sec); |
abd29c93 AN |
2530 | for (i = 0; i < vlen; i++) { |
2531 | err = bpf_object__init_user_btf_map(obj, sec, i, | |
2532 | obj->efile.btf_maps_shndx, | |
8983b731 AN |
2533 | data, strict, |
2534 | pin_root_path); | |
abd29c93 AN |
2535 | if (err) |
2536 | return err; | |
2537 | } | |
2538 | ||
2539 | return 0; | |
2540 | } | |
2541 | ||
0d13bfce | 2542 | static int bpf_object__init_maps(struct bpf_object *obj, |
01af3bf0 | 2543 | const struct bpf_object_open_opts *opts) |
bf829271 | 2544 | { |
166750bc AN |
2545 | const char *pin_root_path; |
2546 | bool strict; | |
bf829271 | 2547 | int err; |
8837fe5d | 2548 | |
166750bc AN |
2549 | strict = !OPTS_GET(opts, relaxed_maps, false); |
2550 | pin_root_path = OPTS_GET(opts, pin_root_path, NULL); | |
bf829271 | 2551 | |
166750bc AN |
2552 | err = bpf_object__init_user_maps(obj, strict); |
2553 | err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path); | |
2554 | err = err ?: bpf_object__init_global_data_maps(obj); | |
81bfdd08 | 2555 | err = err ?: bpf_object__init_kconfig_map(obj); |
590a0088 | 2556 | err = err ?: bpf_object__init_struct_ops_maps(obj); |
bf829271 | 2557 | |
3b3af91c | 2558 | return err; |
561bbcca WN |
2559 | } |
2560 | ||
e3d91b0c JDB |
2561 | static bool section_have_execinstr(struct bpf_object *obj, int idx) |
2562 | { | |
ad23b723 | 2563 | Elf64_Shdr *sh; |
e3d91b0c | 2564 | |
ad23b723 AN |
2565 | sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx)); |
2566 | if (!sh) | |
e3d91b0c JDB |
2567 | return false; |
2568 | ||
ad23b723 | 2569 | return sh->sh_flags & SHF_EXECINSTR; |
e3d91b0c JDB |
2570 | } |
2571 | ||
0f0e55d8 AN |
2572 | static bool btf_needs_sanitization(struct bpf_object *obj) |
2573 | { | |
9ca1f56a AS |
2574 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
2575 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
2576 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
2577 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 2578 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
0f0e55d8 | 2579 | |
223f903e | 2580 | return !has_func || !has_datasec || !has_func_global || !has_float || !has_decl_tag; |
0f0e55d8 AN |
2581 | } |
2582 | ||
2583 | static void bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf) | |
d7c4b398 | 2584 | { |
9ca1f56a AS |
2585 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
2586 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
2587 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
2588 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 2589 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
d7c4b398 AN |
2590 | struct btf_type *t; |
2591 | int i, j, vlen; | |
d7c4b398 | 2592 | |
6a886de0 | 2593 | for (i = 1; i < btf__type_cnt(btf); i++) { |
d7c4b398 | 2594 | t = (struct btf_type *)btf__type_by_id(btf, i); |
d7c4b398 | 2595 | |
223f903e YS |
2596 | if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) { |
2597 | /* replace VAR/DECL_TAG with INT */ | |
d7c4b398 | 2598 | t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0); |
1d4126c4 AN |
2599 | /* |
2600 | * using size = 1 is the safest choice, 4 will be too | |
2601 | * big and cause kernel BTF validation failure if | |
2602 | * original variable took less than 4 bytes | |
2603 | */ | |
2604 | t->size = 1; | |
708852dc | 2605 | *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8); |
b03bc685 | 2606 | } else if (!has_datasec && btf_is_datasec(t)) { |
d7c4b398 | 2607 | /* replace DATASEC with STRUCT */ |
b03bc685 AN |
2608 | const struct btf_var_secinfo *v = btf_var_secinfos(t); |
2609 | struct btf_member *m = btf_members(t); | |
d7c4b398 AN |
2610 | struct btf_type *vt; |
2611 | char *name; | |
2612 | ||
2613 | name = (char *)btf__name_by_offset(btf, t->name_off); | |
2614 | while (*name) { | |
2615 | if (*name == '.') | |
2616 | *name = '_'; | |
2617 | name++; | |
2618 | } | |
2619 | ||
b03bc685 | 2620 | vlen = btf_vlen(t); |
d7c4b398 AN |
2621 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen); |
2622 | for (j = 0; j < vlen; j++, v++, m++) { | |
2623 | /* order of field assignments is important */ | |
2624 | m->offset = v->offset * 8; | |
2625 | m->type = v->type; | |
2626 | /* preserve variable name as member name */ | |
2627 | vt = (void *)btf__type_by_id(btf, v->type); | |
2628 | m->name_off = vt->name_off; | |
2629 | } | |
b03bc685 | 2630 | } else if (!has_func && btf_is_func_proto(t)) { |
d7c4b398 | 2631 | /* replace FUNC_PROTO with ENUM */ |
b03bc685 | 2632 | vlen = btf_vlen(t); |
d7c4b398 AN |
2633 | t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen); |
2634 | t->size = sizeof(__u32); /* kernel enforced */ | |
b03bc685 | 2635 | } else if (!has_func && btf_is_func(t)) { |
d7c4b398 AN |
2636 | /* replace FUNC with TYPEDEF */ |
2637 | t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0); | |
2d3eb67f AS |
2638 | } else if (!has_func_global && btf_is_func(t)) { |
2639 | /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */ | |
2640 | t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0); | |
22541a9e IL |
2641 | } else if (!has_float && btf_is_float(t)) { |
2642 | /* replace FLOAT with an equally-sized empty STRUCT; | |
2643 | * since C compilers do not accept e.g. "float" as a | |
2644 | * valid struct name, make it anonymous | |
2645 | */ | |
2646 | t->name_off = 0; | |
2647 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0); | |
d7c4b398 AN |
2648 | } |
2649 | } | |
2650 | } | |
2651 | ||
b35f14f4 | 2652 | static bool libbpf_needs_btf(const struct bpf_object *obj) |
abd29c93 | 2653 | { |
b35f14f4 AN |
2654 | return obj->efile.btf_maps_shndx >= 0 || |
2655 | obj->efile.st_ops_shndx >= 0 || | |
2656 | obj->nr_extern > 0; | |
2657 | } | |
2658 | ||
2659 | static bool kernel_needs_btf(const struct bpf_object *obj) | |
2660 | { | |
2661 | return obj->efile.st_ops_shndx >= 0; | |
abd29c93 AN |
2662 | } |
2663 | ||
063183bf | 2664 | static int bpf_object__init_btf(struct bpf_object *obj, |
9c6660d0 AN |
2665 | Elf_Data *btf_data, |
2666 | Elf_Data *btf_ext_data) | |
2667 | { | |
b7d7f3e1 | 2668 | int err = -ENOENT; |
9c6660d0 AN |
2669 | |
2670 | if (btf_data) { | |
2671 | obj->btf = btf__new(btf_data->d_buf, btf_data->d_size); | |
e9fc3ce9 AN |
2672 | err = libbpf_get_error(obj->btf); |
2673 | if (err) { | |
b7d7f3e1 | 2674 | obj->btf = NULL; |
e9fc3ce9 | 2675 | pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err); |
9c6660d0 AN |
2676 | goto out; |
2677 | } | |
4c01925f AN |
2678 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
2679 | btf__set_pointer_size(obj->btf, 8); | |
9c6660d0 AN |
2680 | } |
2681 | if (btf_ext_data) { | |
2682 | if (!obj->btf) { | |
2683 | pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n", | |
2684 | BTF_EXT_ELF_SEC, BTF_ELF_SEC); | |
2685 | goto out; | |
2686 | } | |
e9fc3ce9 AN |
2687 | obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size); |
2688 | err = libbpf_get_error(obj->btf_ext); | |
2689 | if (err) { | |
2690 | pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n", | |
2691 | BTF_EXT_ELF_SEC, err); | |
9c6660d0 AN |
2692 | obj->btf_ext = NULL; |
2693 | goto out; | |
2694 | } | |
9c6660d0 AN |
2695 | } |
2696 | out: | |
b35f14f4 | 2697 | if (err && libbpf_needs_btf(obj)) { |
be18010e | 2698 | pr_warn("BTF is required, but is missing or corrupted.\n"); |
b7d7f3e1 | 2699 | return err; |
abd29c93 | 2700 | } |
9c6660d0 AN |
2701 | return 0; |
2702 | } | |
2703 | ||
b96c07f3 AN |
2704 | static int compare_vsi_off(const void *_a, const void *_b) |
2705 | { | |
2706 | const struct btf_var_secinfo *a = _a; | |
2707 | const struct btf_var_secinfo *b = _b; | |
2708 | ||
2709 | return a->offset - b->offset; | |
2710 | } | |
2711 | ||
2712 | static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf, | |
2713 | struct btf_type *t) | |
2714 | { | |
2715 | __u32 size = 0, off = 0, i, vars = btf_vlen(t); | |
2716 | const char *name = btf__name_by_offset(btf, t->name_off); | |
2717 | const struct btf_type *t_var; | |
2718 | struct btf_var_secinfo *vsi; | |
2719 | const struct btf_var *var; | |
2720 | int ret; | |
2721 | ||
2722 | if (!name) { | |
2723 | pr_debug("No name found in string section for DATASEC kind.\n"); | |
2724 | return -ENOENT; | |
2725 | } | |
2726 | ||
2727 | /* .extern datasec size and var offsets were set correctly during | |
2728 | * extern collection step, so just skip straight to sorting variables | |
2729 | */ | |
2730 | if (t->size) | |
2731 | goto sort_vars; | |
2732 | ||
2733 | ret = find_elf_sec_sz(obj, name, &size); | |
2734 | if (ret || !size || (t->size && t->size != size)) { | |
2735 | pr_debug("Invalid size for section %s: %u bytes\n", name, size); | |
2736 | return -ENOENT; | |
2737 | } | |
2738 | ||
2739 | t->size = size; | |
2740 | ||
2741 | for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) { | |
2742 | t_var = btf__type_by_id(btf, vsi->type); | |
2743 | var = btf_var(t_var); | |
2744 | ||
2745 | if (!btf_is_var(t_var)) { | |
2746 | pr_debug("Non-VAR type seen in section %s\n", name); | |
2747 | return -EINVAL; | |
2748 | } | |
2749 | ||
2750 | if (var->linkage == BTF_VAR_STATIC) | |
2751 | continue; | |
2752 | ||
2753 | name = btf__name_by_offset(btf, t_var->name_off); | |
2754 | if (!name) { | |
2755 | pr_debug("No name found in string section for VAR kind\n"); | |
2756 | return -ENOENT; | |
2757 | } | |
2758 | ||
2759 | ret = find_elf_var_offset(obj, name, &off); | |
2760 | if (ret) { | |
2761 | pr_debug("No offset found in symbol table for VAR %s\n", | |
2762 | name); | |
2763 | return -ENOENT; | |
2764 | } | |
2765 | ||
2766 | vsi->offset = off; | |
2767 | } | |
2768 | ||
2769 | sort_vars: | |
2770 | qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off); | |
2771 | return 0; | |
2772 | } | |
2773 | ||
2774 | static int btf_finalize_data(struct bpf_object *obj, struct btf *btf) | |
2775 | { | |
2776 | int err = 0; | |
6a886de0 | 2777 | __u32 i, n = btf__type_cnt(btf); |
b96c07f3 | 2778 | |
6a886de0 | 2779 | for (i = 1; i < n; i++) { |
b96c07f3 AN |
2780 | struct btf_type *t = btf_type_by_id(btf, i); |
2781 | ||
2782 | /* Loader needs to fix up some of the things compiler | |
2783 | * couldn't get its hands on while emitting BTF. This | |
2784 | * is section size and global variable offset. We use | |
2785 | * the info from the ELF itself for this purpose. | |
2786 | */ | |
2787 | if (btf_is_datasec(t)) { | |
2788 | err = btf_fixup_datasec(obj, btf, t); | |
2789 | if (err) | |
2790 | break; | |
2791 | } | |
2792 | } | |
2793 | ||
2794 | return libbpf_err(err); | |
2795 | } | |
2796 | ||
2797 | int btf__finalize_data(struct bpf_object *obj, struct btf *btf) | |
2798 | { | |
2799 | return btf_finalize_data(obj, btf); | |
2800 | } | |
2801 | ||
166750bc AN |
2802 | static int bpf_object__finalize_btf(struct bpf_object *obj) |
2803 | { | |
2804 | int err; | |
2805 | ||
2806 | if (!obj->btf) | |
2807 | return 0; | |
2808 | ||
b96c07f3 | 2809 | err = btf_finalize_data(obj, obj->btf); |
bfc96656 AN |
2810 | if (err) { |
2811 | pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err); | |
2812 | return err; | |
166750bc | 2813 | } |
bfc96656 | 2814 | |
166750bc AN |
2815 | return 0; |
2816 | } | |
2817 | ||
fe62de31 | 2818 | static bool prog_needs_vmlinux_btf(struct bpf_program *prog) |
a6ed02ca | 2819 | { |
1e092a03 KS |
2820 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS || |
2821 | prog->type == BPF_PROG_TYPE_LSM) | |
a6ed02ca KS |
2822 | return true; |
2823 | ||
2824 | /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs | |
2825 | * also need vmlinux BTF | |
2826 | */ | |
2827 | if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd) | |
2828 | return true; | |
2829 | ||
2830 | return false; | |
2831 | } | |
2832 | ||
fe62de31 | 2833 | static bool obj_needs_vmlinux_btf(const struct bpf_object *obj) |
a6ed02ca KS |
2834 | { |
2835 | struct bpf_program *prog; | |
fe62de31 | 2836 | int i; |
a6ed02ca | 2837 | |
1373ff59 SC |
2838 | /* CO-RE relocations need kernel BTF, only when btf_custom_path |
2839 | * is not specified | |
2840 | */ | |
2841 | if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path) | |
fe62de31 | 2842 | return true; |
192b6638 | 2843 | |
d370bbe1 HL |
2844 | /* Support for typed ksyms needs kernel BTF */ |
2845 | for (i = 0; i < obj->nr_extern; i++) { | |
2846 | const struct extern_desc *ext; | |
2847 | ||
2848 | ext = &obj->externs[i]; | |
fe62de31 AN |
2849 | if (ext->type == EXT_KSYM && ext->ksym.type_id) |
2850 | return true; | |
d370bbe1 HL |
2851 | } |
2852 | ||
a6ed02ca | 2853 | bpf_object__for_each_program(prog, obj) { |
d9297581 AN |
2854 | if (!prog->load) |
2855 | continue; | |
fe62de31 AN |
2856 | if (prog_needs_vmlinux_btf(prog)) |
2857 | return true; | |
a6ed02ca KS |
2858 | } |
2859 | ||
fe62de31 AN |
2860 | return false; |
2861 | } | |
2862 | ||
2863 | static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force) | |
2864 | { | |
2865 | int err; | |
2866 | ||
2867 | /* btf_vmlinux could be loaded earlier */ | |
67234743 | 2868 | if (obj->btf_vmlinux || obj->gen_loader) |
fe62de31 AN |
2869 | return 0; |
2870 | ||
2871 | if (!force && !obj_needs_vmlinux_btf(obj)) | |
192b6638 AN |
2872 | return 0; |
2873 | ||
a710eed3 | 2874 | obj->btf_vmlinux = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
2875 | err = libbpf_get_error(obj->btf_vmlinux); |
2876 | if (err) { | |
192b6638 AN |
2877 | pr_warn("Error loading vmlinux BTF: %d\n", err); |
2878 | obj->btf_vmlinux = NULL; | |
2879 | return err; | |
2880 | } | |
a6ed02ca KS |
2881 | return 0; |
2882 | } | |
2883 | ||
063183bf AN |
2884 | static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj) |
2885 | { | |
0f0e55d8 AN |
2886 | struct btf *kern_btf = obj->btf; |
2887 | bool btf_mandatory, sanitize; | |
aea28a60 | 2888 | int i, err = 0; |
063183bf AN |
2889 | |
2890 | if (!obj->btf) | |
2891 | return 0; | |
2892 | ||
9ca1f56a | 2893 | if (!kernel_supports(obj, FEAT_BTF)) { |
68b08647 AN |
2894 | if (kernel_needs_btf(obj)) { |
2895 | err = -EOPNOTSUPP; | |
2896 | goto report; | |
2897 | } | |
2898 | pr_debug("Kernel doesn't support BTF, skipping uploading it.\n"); | |
2899 | return 0; | |
2900 | } | |
2901 | ||
aea28a60 AN |
2902 | /* Even though some subprogs are global/weak, user might prefer more |
2903 | * permissive BPF verification process that BPF verifier performs for | |
2904 | * static functions, taking into account more context from the caller | |
2905 | * functions. In such case, they need to mark such subprogs with | |
2906 | * __attribute__((visibility("hidden"))) and libbpf will adjust | |
2907 | * corresponding FUNC BTF type to be marked as static and trigger more | |
2908 | * involved BPF verification process. | |
2909 | */ | |
2910 | for (i = 0; i < obj->nr_programs; i++) { | |
2911 | struct bpf_program *prog = &obj->programs[i]; | |
2912 | struct btf_type *t; | |
2913 | const char *name; | |
2914 | int j, n; | |
2915 | ||
2916 | if (!prog->mark_btf_static || !prog_is_subprog(obj, prog)) | |
2917 | continue; | |
2918 | ||
6a886de0 HC |
2919 | n = btf__type_cnt(obj->btf); |
2920 | for (j = 1; j < n; j++) { | |
aea28a60 AN |
2921 | t = btf_type_by_id(obj->btf, j); |
2922 | if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) | |
2923 | continue; | |
2924 | ||
2925 | name = btf__str_by_offset(obj->btf, t->name_off); | |
2926 | if (strcmp(name, prog->name) != 0) | |
2927 | continue; | |
2928 | ||
2929 | t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0); | |
2930 | break; | |
2931 | } | |
2932 | } | |
2933 | ||
0f0e55d8 AN |
2934 | sanitize = btf_needs_sanitization(obj); |
2935 | if (sanitize) { | |
5c3320d7 | 2936 | const void *raw_data; |
0f0e55d8 | 2937 | __u32 sz; |
063183bf | 2938 | |
0f0e55d8 | 2939 | /* clone BTF to sanitize a copy and leave the original intact */ |
6a886de0 | 2940 | raw_data = btf__raw_data(obj->btf, &sz); |
5c3320d7 | 2941 | kern_btf = btf__new(raw_data, sz); |
e9fc3ce9 AN |
2942 | err = libbpf_get_error(kern_btf); |
2943 | if (err) | |
2944 | return err; | |
04efe591 | 2945 | |
4c01925f AN |
2946 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
2947 | btf__set_pointer_size(obj->btf, 8); | |
0f0e55d8 | 2948 | bpf_object__sanitize_btf(obj, kern_btf); |
063183bf | 2949 | } |
0f0e55d8 | 2950 | |
67234743 AS |
2951 | if (obj->gen_loader) { |
2952 | __u32 raw_size = 0; | |
6a886de0 | 2953 | const void *raw_data = btf__raw_data(kern_btf, &raw_size); |
67234743 AS |
2954 | |
2955 | if (!raw_data) | |
2956 | return -ENOMEM; | |
2957 | bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size); | |
2958 | /* Pretend to have valid FD to pass various fd >= 0 checks. | |
2959 | * This fd == 0 will not be used with any syscall and will be reset to -1 eventually. | |
2960 | */ | |
2961 | btf__set_fd(kern_btf, 0); | |
2962 | } else { | |
3c7e5859 | 2963 | err = btf__load_into_kernel(kern_btf); |
67234743 | 2964 | } |
0f0e55d8 AN |
2965 | if (sanitize) { |
2966 | if (!err) { | |
2967 | /* move fd to libbpf's BTF */ | |
2968 | btf__set_fd(obj->btf, btf__fd(kern_btf)); | |
2969 | btf__set_fd(kern_btf, -1); | |
2970 | } | |
2971 | btf__free(kern_btf); | |
2972 | } | |
68b08647 | 2973 | report: |
0f0e55d8 AN |
2974 | if (err) { |
2975 | btf_mandatory = kernel_needs_btf(obj); | |
2976 | pr_warn("Error loading .BTF into kernel: %d. %s\n", err, | |
2977 | btf_mandatory ? "BTF is mandatory, can't proceed." | |
2978 | : "BTF is optional, ignoring."); | |
2979 | if (!btf_mandatory) | |
2980 | err = 0; | |
2981 | } | |
2982 | return err; | |
063183bf AN |
2983 | } |
2984 | ||
88a82120 AN |
2985 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off) |
2986 | { | |
2987 | const char *name; | |
2988 | ||
2989 | name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off); | |
2990 | if (!name) { | |
2991 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
2992 | off, obj->path, elf_errmsg(-1)); | |
2993 | return NULL; | |
2994 | } | |
2995 | ||
2996 | return name; | |
2997 | } | |
2998 | ||
2999 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off) | |
3000 | { | |
3001 | const char *name; | |
3002 | ||
3003 | name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off); | |
3004 | if (!name) { | |
3005 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3006 | off, obj->path, elf_errmsg(-1)); | |
3007 | return NULL; | |
3008 | } | |
3009 | ||
3010 | return name; | |
3011 | } | |
3012 | ||
3013 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx) | |
3014 | { | |
3015 | Elf_Scn *scn; | |
3016 | ||
3017 | scn = elf_getscn(obj->efile.elf, idx); | |
3018 | if (!scn) { | |
3019 | pr_warn("elf: failed to get section(%zu) from %s: %s\n", | |
3020 | idx, obj->path, elf_errmsg(-1)); | |
3021 | return NULL; | |
3022 | } | |
3023 | return scn; | |
3024 | } | |
3025 | ||
3026 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name) | |
3027 | { | |
3028 | Elf_Scn *scn = NULL; | |
3029 | Elf *elf = obj->efile.elf; | |
3030 | const char *sec_name; | |
3031 | ||
3032 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
3033 | sec_name = elf_sec_name(obj, scn); | |
3034 | if (!sec_name) | |
3035 | return NULL; | |
3036 | ||
3037 | if (strcmp(sec_name, name) != 0) | |
3038 | continue; | |
3039 | ||
3040 | return scn; | |
3041 | } | |
3042 | return NULL; | |
3043 | } | |
3044 | ||
ad23b723 | 3045 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn) |
88a82120 | 3046 | { |
ad23b723 AN |
3047 | Elf64_Shdr *shdr; |
3048 | ||
88a82120 | 3049 | if (!scn) |
ad23b723 | 3050 | return NULL; |
88a82120 | 3051 | |
ad23b723 AN |
3052 | shdr = elf64_getshdr(scn); |
3053 | if (!shdr) { | |
88a82120 AN |
3054 | pr_warn("elf: failed to get section(%zu) header from %s: %s\n", |
3055 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
ad23b723 | 3056 | return NULL; |
88a82120 AN |
3057 | } |
3058 | ||
ad23b723 | 3059 | return shdr; |
88a82120 AN |
3060 | } |
3061 | ||
3062 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn) | |
3063 | { | |
3064 | const char *name; | |
ad23b723 | 3065 | Elf64_Shdr *sh; |
88a82120 AN |
3066 | |
3067 | if (!scn) | |
3068 | return NULL; | |
3069 | ||
ad23b723 AN |
3070 | sh = elf_sec_hdr(obj, scn); |
3071 | if (!sh) | |
88a82120 AN |
3072 | return NULL; |
3073 | ||
ad23b723 | 3074 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 AN |
3075 | if (!name) { |
3076 | pr_warn("elf: failed to get section(%zu) name from %s: %s\n", | |
3077 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
3078 | return NULL; | |
3079 | } | |
3080 | ||
3081 | return name; | |
3082 | } | |
3083 | ||
3084 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn) | |
3085 | { | |
3086 | Elf_Data *data; | |
3087 | ||
3088 | if (!scn) | |
3089 | return NULL; | |
3090 | ||
3091 | data = elf_getdata(scn, 0); | |
3092 | if (!data) { | |
3093 | pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n", | |
3094 | elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>", | |
3095 | obj->path, elf_errmsg(-1)); | |
3096 | return NULL; | |
3097 | } | |
3098 | ||
3099 | return data; | |
3100 | } | |
3101 | ||
ad23b723 AN |
3102 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx) |
3103 | { | |
3104 | if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym)) | |
3105 | return NULL; | |
3106 | ||
3107 | return (Elf64_Sym *)obj->efile.symbols->d_buf + idx; | |
3108 | } | |
3109 | ||
3110 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx) | |
3111 | { | |
3112 | if (idx >= data->d_size / sizeof(Elf64_Rel)) | |
3113 | return NULL; | |
3114 | ||
3115 | return (Elf64_Rel *)data->d_buf + idx; | |
3116 | } | |
3117 | ||
50e09460 AN |
3118 | static bool is_sec_name_dwarf(const char *name) |
3119 | { | |
3120 | /* approximation, but the actual list is too long */ | |
13d35a0c | 3121 | return str_has_pfx(name, ".debug_"); |
50e09460 AN |
3122 | } |
3123 | ||
ad23b723 | 3124 | static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name) |
50e09460 AN |
3125 | { |
3126 | /* no special handling of .strtab */ | |
3127 | if (hdr->sh_type == SHT_STRTAB) | |
3128 | return true; | |
3129 | ||
3130 | /* ignore .llvm_addrsig section as well */ | |
faf6ed32 | 3131 | if (hdr->sh_type == SHT_LLVM_ADDRSIG) |
50e09460 AN |
3132 | return true; |
3133 | ||
3134 | /* no subprograms will lead to an empty .text section, ignore it */ | |
3135 | if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 && | |
3136 | strcmp(name, ".text") == 0) | |
3137 | return true; | |
3138 | ||
3139 | /* DWARF sections */ | |
3140 | if (is_sec_name_dwarf(name)) | |
3141 | return true; | |
3142 | ||
13d35a0c | 3143 | if (str_has_pfx(name, ".rel")) { |
50e09460 AN |
3144 | name += sizeof(".rel") - 1; |
3145 | /* DWARF section relocations */ | |
3146 | if (is_sec_name_dwarf(name)) | |
3147 | return true; | |
3148 | ||
3149 | /* .BTF and .BTF.ext don't need relocations */ | |
3150 | if (strcmp(name, BTF_ELF_SEC) == 0 || | |
3151 | strcmp(name, BTF_EXT_ELF_SEC) == 0) | |
3152 | return true; | |
3153 | } | |
3154 | ||
3155 | return false; | |
3156 | } | |
3157 | ||
db2b8b06 AN |
3158 | static int cmp_progs(const void *_a, const void *_b) |
3159 | { | |
3160 | const struct bpf_program *a = _a; | |
3161 | const struct bpf_program *b = _b; | |
3162 | ||
3163 | if (a->sec_idx != b->sec_idx) | |
3164 | return a->sec_idx < b->sec_idx ? -1 : 1; | |
3165 | ||
3166 | /* sec_insn_off can't be the same within the section */ | |
3167 | return a->sec_insn_off < b->sec_insn_off ? -1 : 1; | |
3168 | } | |
3169 | ||
0d13bfce | 3170 | static int bpf_object__elf_collect(struct bpf_object *obj) |
29603665 | 3171 | { |
25bbbd7a | 3172 | struct elf_sec_desc *sec_desc; |
29603665 | 3173 | Elf *elf = obj->efile.elf; |
f0187f0b | 3174 | Elf_Data *btf_ext_data = NULL; |
1713d68b | 3175 | Elf_Data *btf_data = NULL; |
666810e8 | 3176 | int idx = 0, err = 0; |
0201c575 AN |
3177 | const char *name; |
3178 | Elf_Data *data; | |
3179 | Elf_Scn *scn; | |
ad23b723 | 3180 | Elf64_Shdr *sh; |
29603665 | 3181 | |
25bbbd7a AN |
3182 | /* ELF section indices are 1-based, so allocate +1 element to keep |
3183 | * indexing simple. Also include 0th invalid section into sec_cnt for | |
3184 | * simpler and more traditional iteration logic. | |
3185 | */ | |
3186 | obj->efile.sec_cnt = 1 + obj->efile.ehdr->e_shnum; | |
3187 | obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs)); | |
3188 | if (!obj->efile.secs) | |
3189 | return -ENOMEM; | |
3190 | ||
0201c575 AN |
3191 | /* a bunch of ELF parsing functionality depends on processing symbols, |
3192 | * so do the first pass and find the symbol table | |
3193 | */ | |
3194 | scn = NULL; | |
29603665 | 3195 | while ((scn = elf_nextscn(elf, scn)) != NULL) { |
ad23b723 AN |
3196 | sh = elf_sec_hdr(obj, scn); |
3197 | if (!sh) | |
0201c575 AN |
3198 | return -LIBBPF_ERRNO__FORMAT; |
3199 | ||
ad23b723 | 3200 | if (sh->sh_type == SHT_SYMTAB) { |
0201c575 AN |
3201 | if (obj->efile.symbols) { |
3202 | pr_warn("elf: multiple symbol tables in %s\n", obj->path); | |
3203 | return -LIBBPF_ERRNO__FORMAT; | |
3204 | } | |
29603665 | 3205 | |
0201c575 AN |
3206 | data = elf_sec_data(obj, scn); |
3207 | if (!data) | |
3208 | return -LIBBPF_ERRNO__FORMAT; | |
3209 | ||
25bbbd7a AN |
3210 | idx = elf_ndxscn(scn); |
3211 | ||
0201c575 | 3212 | obj->efile.symbols = data; |
25bbbd7a | 3213 | obj->efile.symbols_shndx = idx; |
ad23b723 | 3214 | obj->efile.strtabidx = sh->sh_link; |
0201c575 AN |
3215 | } |
3216 | } | |
3217 | ||
03e601f4 THJ |
3218 | if (!obj->efile.symbols) { |
3219 | pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n", | |
3220 | obj->path); | |
3221 | return -ENOENT; | |
3222 | } | |
3223 | ||
0201c575 AN |
3224 | scn = NULL; |
3225 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
25bbbd7a AN |
3226 | idx = elf_ndxscn(scn); |
3227 | sec_desc = &obj->efile.secs[idx]; | |
88a82120 | 3228 | |
ad23b723 AN |
3229 | sh = elf_sec_hdr(obj, scn); |
3230 | if (!sh) | |
01b29d1d | 3231 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3232 | |
ad23b723 | 3233 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 | 3234 | if (!name) |
01b29d1d | 3235 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3236 | |
ad23b723 | 3237 | if (ignore_elf_section(sh, name)) |
50e09460 AN |
3238 | continue; |
3239 | ||
88a82120 AN |
3240 | data = elf_sec_data(obj, scn); |
3241 | if (!data) | |
01b29d1d | 3242 | return -LIBBPF_ERRNO__FORMAT; |
88a82120 AN |
3243 | |
3244 | pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n", | |
077c066a | 3245 | idx, name, (unsigned long)data->d_size, |
ad23b723 AN |
3246 | (int)sh->sh_link, (unsigned long)sh->sh_flags, |
3247 | (int)sh->sh_type); | |
cb1e5e96 | 3248 | |
1713d68b | 3249 | if (strcmp(name, "license") == 0) { |
88a82120 | 3250 | err = bpf_object__init_license(obj, data->d_buf, data->d_size); |
01b29d1d AN |
3251 | if (err) |
3252 | return err; | |
1713d68b | 3253 | } else if (strcmp(name, "version") == 0) { |
88a82120 | 3254 | err = bpf_object__init_kversion(obj, data->d_buf, data->d_size); |
54b8625c JF |
3255 | if (err) |
3256 | return err; | |
1713d68b | 3257 | } else if (strcmp(name, "maps") == 0) { |
666810e8 | 3258 | obj->efile.maps_shndx = idx; |
abd29c93 AN |
3259 | } else if (strcmp(name, MAPS_ELF_SEC) == 0) { |
3260 | obj->efile.btf_maps_shndx = idx; | |
1713d68b DB |
3261 | } else if (strcmp(name, BTF_ELF_SEC) == 0) { |
3262 | btf_data = data; | |
2993e051 | 3263 | } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) { |
f0187f0b | 3264 | btf_ext_data = data; |
ad23b723 | 3265 | } else if (sh->sh_type == SHT_SYMTAB) { |
0201c575 | 3266 | /* already processed during the first pass above */ |
ad23b723 AN |
3267 | } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) { |
3268 | if (sh->sh_flags & SHF_EXECINSTR) { | |
f8c7a4d4 JS |
3269 | if (strcmp(name, ".text") == 0) |
3270 | obj->efile.text_shndx = idx; | |
c1122392 | 3271 | err = bpf_object__add_programs(obj, data, name, idx); |
88a82120 | 3272 | if (err) |
01b29d1d | 3273 | return err; |
aed65917 AN |
3274 | } else if (strcmp(name, DATA_SEC) == 0 || |
3275 | str_has_pfx(name, DATA_SEC ".")) { | |
25bbbd7a AN |
3276 | sec_desc->sec_type = SEC_DATA; |
3277 | sec_desc->shdr = sh; | |
3278 | sec_desc->data = data; | |
aed65917 AN |
3279 | } else if (strcmp(name, RODATA_SEC) == 0 || |
3280 | str_has_pfx(name, RODATA_SEC ".")) { | |
25bbbd7a AN |
3281 | sec_desc->sec_type = SEC_RODATA; |
3282 | sec_desc->shdr = sh; | |
3283 | sec_desc->data = data; | |
590a0088 MKL |
3284 | } else if (strcmp(name, STRUCT_OPS_SEC) == 0) { |
3285 | obj->efile.st_ops_data = data; | |
3286 | obj->efile.st_ops_shndx = idx; | |
d859900c | 3287 | } else { |
50e09460 AN |
3288 | pr_info("elf: skipping unrecognized data section(%d) %s\n", |
3289 | idx, name); | |
a5b8bd47 | 3290 | } |
ad23b723 | 3291 | } else if (sh->sh_type == SHT_REL) { |
25bbbd7a | 3292 | int targ_sec_idx = sh->sh_info; /* points to other section */ |
e3d91b0c JDB |
3293 | |
3294 | /* Only do relo for section with exec instructions */ | |
25bbbd7a | 3295 | if (!section_have_execinstr(obj, targ_sec_idx) && |
646f02ff AN |
3296 | strcmp(name, ".rel" STRUCT_OPS_SEC) && |
3297 | strcmp(name, ".rel" MAPS_ELF_SEC)) { | |
50e09460 | 3298 | pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n", |
25bbbd7a AN |
3299 | idx, name, targ_sec_idx, |
3300 | elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>"); | |
e3d91b0c JDB |
3301 | continue; |
3302 | } | |
b62f06e8 | 3303 | |
25bbbd7a AN |
3304 | sec_desc->sec_type = SEC_RELO; |
3305 | sec_desc->shdr = sh; | |
3306 | sec_desc->data = data; | |
ad23b723 | 3307 | } else if (sh->sh_type == SHT_NOBITS && strcmp(name, BSS_SEC) == 0) { |
25bbbd7a AN |
3308 | sec_desc->sec_type = SEC_BSS; |
3309 | sec_desc->shdr = sh; | |
3310 | sec_desc->data = data; | |
077c066a | 3311 | } else { |
2e80be60 | 3312 | pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name, |
ad23b723 | 3313 | (size_t)sh->sh_size); |
bec7d68c | 3314 | } |
29603665 | 3315 | } |
561bbcca | 3316 | |
d3a3aa0c | 3317 | if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) { |
88a82120 | 3318 | pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path); |
f102154d | 3319 | return -LIBBPF_ERRNO__FORMAT; |
77ba9a5b | 3320 | } |
db2b8b06 AN |
3321 | |
3322 | /* sort BPF programs by section name and in-section instruction offset | |
3323 | * for faster search */ | |
3324 | qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs); | |
3325 | ||
0d13bfce | 3326 | return bpf_object__init_btf(obj, btf_data, btf_ext_data); |
29603665 WN |
3327 | } |
3328 | ||
ad23b723 | 3329 | static bool sym_is_extern(const Elf64_Sym *sym) |
166750bc | 3330 | { |
ad23b723 | 3331 | int bind = ELF64_ST_BIND(sym->st_info); |
166750bc AN |
3332 | /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */ |
3333 | return sym->st_shndx == SHN_UNDEF && | |
3334 | (bind == STB_GLOBAL || bind == STB_WEAK) && | |
ad23b723 | 3335 | ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE; |
166750bc AN |
3336 | } |
3337 | ||
ad23b723 | 3338 | static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx) |
53eddb5e | 3339 | { |
ad23b723 AN |
3340 | int bind = ELF64_ST_BIND(sym->st_info); |
3341 | int type = ELF64_ST_TYPE(sym->st_info); | |
53eddb5e YS |
3342 | |
3343 | /* in .text section */ | |
3344 | if (sym->st_shndx != text_shndx) | |
3345 | return false; | |
3346 | ||
3347 | /* local function */ | |
3348 | if (bind == STB_LOCAL && type == STT_SECTION) | |
3349 | return true; | |
3350 | ||
3351 | /* global function */ | |
3352 | return bind == STB_GLOBAL && type == STT_FUNC; | |
3353 | } | |
3354 | ||
166750bc AN |
3355 | static int find_extern_btf_id(const struct btf *btf, const char *ext_name) |
3356 | { | |
3357 | const struct btf_type *t; | |
5bd022ec | 3358 | const char *tname; |
166750bc AN |
3359 | int i, n; |
3360 | ||
3361 | if (!btf) | |
3362 | return -ESRCH; | |
3363 | ||
6a886de0 HC |
3364 | n = btf__type_cnt(btf); |
3365 | for (i = 1; i < n; i++) { | |
166750bc AN |
3366 | t = btf__type_by_id(btf, i); |
3367 | ||
5bd022ec | 3368 | if (!btf_is_var(t) && !btf_is_func(t)) |
166750bc AN |
3369 | continue; |
3370 | ||
5bd022ec MKL |
3371 | tname = btf__name_by_offset(btf, t->name_off); |
3372 | if (strcmp(tname, ext_name)) | |
166750bc AN |
3373 | continue; |
3374 | ||
5bd022ec MKL |
3375 | if (btf_is_var(t) && |
3376 | btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN) | |
3377 | return -EINVAL; | |
3378 | ||
3379 | if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN) | |
166750bc AN |
3380 | return -EINVAL; |
3381 | ||
3382 | return i; | |
3383 | } | |
3384 | ||
3385 | return -ENOENT; | |
3386 | } | |
3387 | ||
2e33efe3 AN |
3388 | static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) { |
3389 | const struct btf_var_secinfo *vs; | |
3390 | const struct btf_type *t; | |
3391 | int i, j, n; | |
3392 | ||
3393 | if (!btf) | |
3394 | return -ESRCH; | |
3395 | ||
6a886de0 HC |
3396 | n = btf__type_cnt(btf); |
3397 | for (i = 1; i < n; i++) { | |
2e33efe3 AN |
3398 | t = btf__type_by_id(btf, i); |
3399 | ||
3400 | if (!btf_is_datasec(t)) | |
3401 | continue; | |
3402 | ||
3403 | vs = btf_var_secinfos(t); | |
3404 | for (j = 0; j < btf_vlen(t); j++, vs++) { | |
3405 | if (vs->type == ext_btf_id) | |
3406 | return i; | |
3407 | } | |
3408 | } | |
3409 | ||
3410 | return -ENOENT; | |
3411 | } | |
3412 | ||
3413 | static enum kcfg_type find_kcfg_type(const struct btf *btf, int id, | |
3414 | bool *is_signed) | |
166750bc AN |
3415 | { |
3416 | const struct btf_type *t; | |
3417 | const char *name; | |
3418 | ||
3419 | t = skip_mods_and_typedefs(btf, id, NULL); | |
3420 | name = btf__name_by_offset(btf, t->name_off); | |
3421 | ||
3422 | if (is_signed) | |
3423 | *is_signed = false; | |
3424 | switch (btf_kind(t)) { | |
3425 | case BTF_KIND_INT: { | |
3426 | int enc = btf_int_encoding(t); | |
3427 | ||
3428 | if (enc & BTF_INT_BOOL) | |
2e33efe3 | 3429 | return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN; |
166750bc AN |
3430 | if (is_signed) |
3431 | *is_signed = enc & BTF_INT_SIGNED; | |
3432 | if (t->size == 1) | |
2e33efe3 | 3433 | return KCFG_CHAR; |
166750bc | 3434 | if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1))) |
2e33efe3 AN |
3435 | return KCFG_UNKNOWN; |
3436 | return KCFG_INT; | |
166750bc AN |
3437 | } |
3438 | case BTF_KIND_ENUM: | |
3439 | if (t->size != 4) | |
2e33efe3 | 3440 | return KCFG_UNKNOWN; |
166750bc | 3441 | if (strcmp(name, "libbpf_tristate")) |
2e33efe3 AN |
3442 | return KCFG_UNKNOWN; |
3443 | return KCFG_TRISTATE; | |
166750bc AN |
3444 | case BTF_KIND_ARRAY: |
3445 | if (btf_array(t)->nelems == 0) | |
2e33efe3 AN |
3446 | return KCFG_UNKNOWN; |
3447 | if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR) | |
3448 | return KCFG_UNKNOWN; | |
3449 | return KCFG_CHAR_ARR; | |
166750bc | 3450 | default: |
2e33efe3 | 3451 | return KCFG_UNKNOWN; |
166750bc AN |
3452 | } |
3453 | } | |
3454 | ||
3455 | static int cmp_externs(const void *_a, const void *_b) | |
3456 | { | |
3457 | const struct extern_desc *a = _a; | |
3458 | const struct extern_desc *b = _b; | |
3459 | ||
2e33efe3 AN |
3460 | if (a->type != b->type) |
3461 | return a->type < b->type ? -1 : 1; | |
3462 | ||
3463 | if (a->type == EXT_KCFG) { | |
3464 | /* descending order by alignment requirements */ | |
3465 | if (a->kcfg.align != b->kcfg.align) | |
3466 | return a->kcfg.align > b->kcfg.align ? -1 : 1; | |
3467 | /* ascending order by size, within same alignment class */ | |
3468 | if (a->kcfg.sz != b->kcfg.sz) | |
3469 | return a->kcfg.sz < b->kcfg.sz ? -1 : 1; | |
3470 | } | |
3471 | ||
166750bc AN |
3472 | /* resolve ties by name */ |
3473 | return strcmp(a->name, b->name); | |
3474 | } | |
3475 | ||
1c0c7074 AN |
3476 | static int find_int_btf_id(const struct btf *btf) |
3477 | { | |
3478 | const struct btf_type *t; | |
3479 | int i, n; | |
3480 | ||
6a886de0 HC |
3481 | n = btf__type_cnt(btf); |
3482 | for (i = 1; i < n; i++) { | |
1c0c7074 AN |
3483 | t = btf__type_by_id(btf, i); |
3484 | ||
3485 | if (btf_is_int(t) && btf_int_bits(t) == 32) | |
3486 | return i; | |
3487 | } | |
3488 | ||
3489 | return 0; | |
3490 | } | |
3491 | ||
5bd022ec MKL |
3492 | static int add_dummy_ksym_var(struct btf *btf) |
3493 | { | |
3494 | int i, int_btf_id, sec_btf_id, dummy_var_btf_id; | |
3495 | const struct btf_var_secinfo *vs; | |
3496 | const struct btf_type *sec; | |
3497 | ||
9683e577 IR |
3498 | if (!btf) |
3499 | return 0; | |
3500 | ||
5bd022ec MKL |
3501 | sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC, |
3502 | BTF_KIND_DATASEC); | |
3503 | if (sec_btf_id < 0) | |
3504 | return 0; | |
3505 | ||
3506 | sec = btf__type_by_id(btf, sec_btf_id); | |
3507 | vs = btf_var_secinfos(sec); | |
3508 | for (i = 0; i < btf_vlen(sec); i++, vs++) { | |
3509 | const struct btf_type *vt; | |
3510 | ||
3511 | vt = btf__type_by_id(btf, vs->type); | |
3512 | if (btf_is_func(vt)) | |
3513 | break; | |
3514 | } | |
3515 | ||
3516 | /* No func in ksyms sec. No need to add dummy var. */ | |
3517 | if (i == btf_vlen(sec)) | |
3518 | return 0; | |
3519 | ||
3520 | int_btf_id = find_int_btf_id(btf); | |
3521 | dummy_var_btf_id = btf__add_var(btf, | |
3522 | "dummy_ksym", | |
3523 | BTF_VAR_GLOBAL_ALLOCATED, | |
3524 | int_btf_id); | |
3525 | if (dummy_var_btf_id < 0) | |
3526 | pr_warn("cannot create a dummy_ksym var\n"); | |
3527 | ||
3528 | return dummy_var_btf_id; | |
3529 | } | |
3530 | ||
166750bc AN |
3531 | static int bpf_object__collect_externs(struct bpf_object *obj) |
3532 | { | |
1c0c7074 | 3533 | struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL; |
166750bc AN |
3534 | const struct btf_type *t; |
3535 | struct extern_desc *ext; | |
5bd022ec | 3536 | int i, n, off, dummy_var_btf_id; |
2e33efe3 | 3537 | const char *ext_name, *sec_name; |
166750bc | 3538 | Elf_Scn *scn; |
ad23b723 | 3539 | Elf64_Shdr *sh; |
166750bc AN |
3540 | |
3541 | if (!obj->efile.symbols) | |
3542 | return 0; | |
3543 | ||
88a82120 | 3544 | scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx); |
ad23b723 AN |
3545 | sh = elf_sec_hdr(obj, scn); |
3546 | if (!sh) | |
166750bc | 3547 | return -LIBBPF_ERRNO__FORMAT; |
166750bc | 3548 | |
5bd022ec MKL |
3549 | dummy_var_btf_id = add_dummy_ksym_var(obj->btf); |
3550 | if (dummy_var_btf_id < 0) | |
3551 | return dummy_var_btf_id; | |
3552 | ||
ad23b723 | 3553 | n = sh->sh_size / sh->sh_entsize; |
166750bc | 3554 | pr_debug("looking for externs among %d symbols...\n", n); |
88a82120 | 3555 | |
166750bc | 3556 | for (i = 0; i < n; i++) { |
ad23b723 | 3557 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
166750bc | 3558 | |
ad23b723 | 3559 | if (!sym) |
166750bc | 3560 | return -LIBBPF_ERRNO__FORMAT; |
ad23b723 | 3561 | if (!sym_is_extern(sym)) |
166750bc | 3562 | continue; |
ad23b723 | 3563 | ext_name = elf_sym_str(obj, sym->st_name); |
166750bc AN |
3564 | if (!ext_name || !ext_name[0]) |
3565 | continue; | |
3566 | ||
3567 | ext = obj->externs; | |
029258d7 | 3568 | ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext)); |
166750bc AN |
3569 | if (!ext) |
3570 | return -ENOMEM; | |
3571 | obj->externs = ext; | |
3572 | ext = &ext[obj->nr_extern]; | |
3573 | memset(ext, 0, sizeof(*ext)); | |
3574 | obj->nr_extern++; | |
3575 | ||
3576 | ext->btf_id = find_extern_btf_id(obj->btf, ext_name); | |
3577 | if (ext->btf_id <= 0) { | |
3578 | pr_warn("failed to find BTF for extern '%s': %d\n", | |
3579 | ext_name, ext->btf_id); | |
3580 | return ext->btf_id; | |
3581 | } | |
3582 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
3583 | ext->name = btf__name_by_offset(obj->btf, t->name_off); | |
3584 | ext->sym_idx = i; | |
ad23b723 | 3585 | ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK; |
2e33efe3 AN |
3586 | |
3587 | ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id); | |
3588 | if (ext->sec_btf_id <= 0) { | |
3589 | pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n", | |
3590 | ext_name, ext->btf_id, ext->sec_btf_id); | |
3591 | return ext->sec_btf_id; | |
166750bc | 3592 | } |
2e33efe3 AN |
3593 | sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id); |
3594 | sec_name = btf__name_by_offset(obj->btf, sec->name_off); | |
3595 | ||
3596 | if (strcmp(sec_name, KCONFIG_SEC) == 0) { | |
5bd022ec MKL |
3597 | if (btf_is_func(t)) { |
3598 | pr_warn("extern function %s is unsupported under %s section\n", | |
3599 | ext->name, KCONFIG_SEC); | |
3600 | return -ENOTSUP; | |
3601 | } | |
2e33efe3 AN |
3602 | kcfg_sec = sec; |
3603 | ext->type = EXT_KCFG; | |
3604 | ext->kcfg.sz = btf__resolve_size(obj->btf, t->type); | |
3605 | if (ext->kcfg.sz <= 0) { | |
3606 | pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n", | |
3607 | ext_name, ext->kcfg.sz); | |
3608 | return ext->kcfg.sz; | |
3609 | } | |
3610 | ext->kcfg.align = btf__align_of(obj->btf, t->type); | |
3611 | if (ext->kcfg.align <= 0) { | |
3612 | pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n", | |
3613 | ext_name, ext->kcfg.align); | |
3614 | return -EINVAL; | |
3615 | } | |
3616 | ext->kcfg.type = find_kcfg_type(obj->btf, t->type, | |
3617 | &ext->kcfg.is_signed); | |
3618 | if (ext->kcfg.type == KCFG_UNKNOWN) { | |
3619 | pr_warn("extern (kcfg) '%s' type is unsupported\n", ext_name); | |
3620 | return -ENOTSUP; | |
3621 | } | |
1c0c7074 | 3622 | } else if (strcmp(sec_name, KSYMS_SEC) == 0) { |
1c0c7074 AN |
3623 | ksym_sec = sec; |
3624 | ext->type = EXT_KSYM; | |
d370bbe1 HL |
3625 | skip_mods_and_typedefs(obj->btf, t->type, |
3626 | &ext->ksym.type_id); | |
2e33efe3 AN |
3627 | } else { |
3628 | pr_warn("unrecognized extern section '%s'\n", sec_name); | |
166750bc AN |
3629 | return -ENOTSUP; |
3630 | } | |
3631 | } | |
3632 | pr_debug("collected %d externs total\n", obj->nr_extern); | |
3633 | ||
3634 | if (!obj->nr_extern) | |
3635 | return 0; | |
3636 | ||
2e33efe3 | 3637 | /* sort externs by type, for kcfg ones also by (align, size, name) */ |
166750bc | 3638 | qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs); |
166750bc | 3639 | |
1c0c7074 AN |
3640 | /* for .ksyms section, we need to turn all externs into allocated |
3641 | * variables in BTF to pass kernel verification; we do this by | |
3642 | * pretending that each extern is a 8-byte variable | |
3643 | */ | |
3644 | if (ksym_sec) { | |
3645 | /* find existing 4-byte integer type in BTF to use for fake | |
3646 | * extern variables in DATASEC | |
3647 | */ | |
3648 | int int_btf_id = find_int_btf_id(obj->btf); | |
5bd022ec MKL |
3649 | /* For extern function, a dummy_var added earlier |
3650 | * will be used to replace the vs->type and | |
3651 | * its name string will be used to refill | |
3652 | * the missing param's name. | |
3653 | */ | |
3654 | const struct btf_type *dummy_var; | |
1c0c7074 | 3655 | |
5bd022ec | 3656 | dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id); |
1c0c7074 AN |
3657 | for (i = 0; i < obj->nr_extern; i++) { |
3658 | ext = &obj->externs[i]; | |
3659 | if (ext->type != EXT_KSYM) | |
3660 | continue; | |
3661 | pr_debug("extern (ksym) #%d: symbol %d, name %s\n", | |
3662 | i, ext->sym_idx, ext->name); | |
3663 | } | |
3664 | ||
3665 | sec = ksym_sec; | |
3666 | n = btf_vlen(sec); | |
3667 | for (i = 0, off = 0; i < n; i++, off += sizeof(int)) { | |
3668 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
3669 | struct btf_type *vt; | |
3670 | ||
3671 | vt = (void *)btf__type_by_id(obj->btf, vs->type); | |
3672 | ext_name = btf__name_by_offset(obj->btf, vt->name_off); | |
3673 | ext = find_extern_by_name(obj, ext_name); | |
3674 | if (!ext) { | |
5bd022ec MKL |
3675 | pr_warn("failed to find extern definition for BTF %s '%s'\n", |
3676 | btf_kind_str(vt), ext_name); | |
1c0c7074 AN |
3677 | return -ESRCH; |
3678 | } | |
5bd022ec MKL |
3679 | if (btf_is_func(vt)) { |
3680 | const struct btf_type *func_proto; | |
3681 | struct btf_param *param; | |
3682 | int j; | |
3683 | ||
3684 | func_proto = btf__type_by_id(obj->btf, | |
3685 | vt->type); | |
3686 | param = btf_params(func_proto); | |
3687 | /* Reuse the dummy_var string if the | |
3688 | * func proto does not have param name. | |
3689 | */ | |
3690 | for (j = 0; j < btf_vlen(func_proto); j++) | |
3691 | if (param[j].type && !param[j].name_off) | |
3692 | param[j].name_off = | |
3693 | dummy_var->name_off; | |
3694 | vs->type = dummy_var_btf_id; | |
3695 | vt->info &= ~0xffff; | |
3696 | vt->info |= BTF_FUNC_GLOBAL; | |
3697 | } else { | |
3698 | btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
3699 | vt->type = int_btf_id; | |
3700 | } | |
1c0c7074 AN |
3701 | vs->offset = off; |
3702 | vs->size = sizeof(int); | |
3703 | } | |
3704 | sec->size = off; | |
3705 | } | |
3706 | ||
2e33efe3 AN |
3707 | if (kcfg_sec) { |
3708 | sec = kcfg_sec; | |
3709 | /* for kcfg externs calculate their offsets within a .kconfig map */ | |
3710 | off = 0; | |
3711 | for (i = 0; i < obj->nr_extern; i++) { | |
3712 | ext = &obj->externs[i]; | |
3713 | if (ext->type != EXT_KCFG) | |
3714 | continue; | |
166750bc | 3715 | |
2e33efe3 AN |
3716 | ext->kcfg.data_off = roundup(off, ext->kcfg.align); |
3717 | off = ext->kcfg.data_off + ext->kcfg.sz; | |
1c0c7074 | 3718 | pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n", |
2e33efe3 AN |
3719 | i, ext->sym_idx, ext->kcfg.data_off, ext->name); |
3720 | } | |
3721 | sec->size = off; | |
3722 | n = btf_vlen(sec); | |
3723 | for (i = 0; i < n; i++) { | |
3724 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
3725 | ||
3726 | t = btf__type_by_id(obj->btf, vs->type); | |
3727 | ext_name = btf__name_by_offset(obj->btf, t->name_off); | |
3728 | ext = find_extern_by_name(obj, ext_name); | |
3729 | if (!ext) { | |
3730 | pr_warn("failed to find extern definition for BTF var '%s'\n", | |
3731 | ext_name); | |
3732 | return -ESRCH; | |
3733 | } | |
3734 | btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
3735 | vs->offset = ext->kcfg.data_off; | |
166750bc | 3736 | } |
166750bc | 3737 | } |
166750bc AN |
3738 | return 0; |
3739 | } | |
3740 | ||
6d4b198b | 3741 | struct bpf_program * |
a324aae3 AN |
3742 | bpf_object__find_program_by_title(const struct bpf_object *obj, |
3743 | const char *title) | |
6d4b198b JK |
3744 | { |
3745 | struct bpf_program *pos; | |
3746 | ||
3747 | bpf_object__for_each_program(pos, obj) { | |
52109584 | 3748 | if (pos->sec_name && !strcmp(pos->sec_name, title)) |
6d4b198b JK |
3749 | return pos; |
3750 | } | |
e9fc3ce9 | 3751 | return errno = ENOENT, NULL; |
6d4b198b JK |
3752 | } |
3753 | ||
c3c55696 AN |
3754 | static bool prog_is_subprog(const struct bpf_object *obj, |
3755 | const struct bpf_program *prog) | |
3756 | { | |
197afc63 AN |
3757 | /* For legacy reasons, libbpf supports an entry-point BPF programs |
3758 | * without SEC() attribute, i.e., those in the .text section. But if | |
3759 | * there are 2 or more such programs in the .text section, they all | |
3760 | * must be subprograms called from entry-point BPF programs in | |
3761 | * designated SEC()'tions, otherwise there is no way to distinguish | |
3762 | * which of those programs should be loaded vs which are a subprogram. | |
3763 | * Similarly, if there is a function/program in .text and at least one | |
3764 | * other BPF program with custom SEC() attribute, then we just assume | |
3765 | * .text programs are subprograms (even if they are not called from | |
3766 | * other programs), because libbpf never explicitly supported mixing | |
3767 | * SEC()-designated BPF programs and .text entry-point BPF programs. | |
3768 | */ | |
3769 | return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1; | |
c3c55696 AN |
3770 | } |
3771 | ||
01af3bf0 AN |
3772 | struct bpf_program * |
3773 | bpf_object__find_program_by_name(const struct bpf_object *obj, | |
3774 | const char *name) | |
3775 | { | |
3776 | struct bpf_program *prog; | |
3777 | ||
3778 | bpf_object__for_each_program(prog, obj) { | |
c3c55696 AN |
3779 | if (prog_is_subprog(obj, prog)) |
3780 | continue; | |
01af3bf0 AN |
3781 | if (!strcmp(prog->name, name)) |
3782 | return prog; | |
3783 | } | |
e9fc3ce9 | 3784 | return errno = ENOENT, NULL; |
01af3bf0 AN |
3785 | } |
3786 | ||
d859900c DB |
3787 | static bool bpf_object__shndx_is_data(const struct bpf_object *obj, |
3788 | int shndx) | |
3789 | { | |
25bbbd7a AN |
3790 | switch (obj->efile.secs[shndx].sec_type) { |
3791 | case SEC_BSS: | |
3792 | case SEC_DATA: | |
3793 | case SEC_RODATA: | |
3794 | return true; | |
3795 | default: | |
3796 | return false; | |
3797 | } | |
d859900c DB |
3798 | } |
3799 | ||
3800 | static bool bpf_object__shndx_is_maps(const struct bpf_object *obj, | |
3801 | int shndx) | |
3802 | { | |
abd29c93 AN |
3803 | return shndx == obj->efile.maps_shndx || |
3804 | shndx == obj->efile.btf_maps_shndx; | |
d859900c DB |
3805 | } |
3806 | ||
d859900c DB |
3807 | static enum libbpf_map_type |
3808 | bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx) | |
3809 | { | |
25bbbd7a AN |
3810 | if (shndx == obj->efile.symbols_shndx) |
3811 | return LIBBPF_MAP_KCONFIG; | |
3812 | ||
3813 | switch (obj->efile.secs[shndx].sec_type) { | |
3814 | case SEC_BSS: | |
d859900c | 3815 | return LIBBPF_MAP_BSS; |
25bbbd7a AN |
3816 | case SEC_DATA: |
3817 | return LIBBPF_MAP_DATA; | |
3818 | case SEC_RODATA: | |
d859900c | 3819 | return LIBBPF_MAP_RODATA; |
25bbbd7a | 3820 | default: |
d859900c | 3821 | return LIBBPF_MAP_UNSPEC; |
25bbbd7a | 3822 | } |
d859900c DB |
3823 | } |
3824 | ||
1f8e2bcb AN |
3825 | static int bpf_program__record_reloc(struct bpf_program *prog, |
3826 | struct reloc_desc *reloc_desc, | |
9c0f8cbd | 3827 | __u32 insn_idx, const char *sym_name, |
ad23b723 | 3828 | const Elf64_Sym *sym, const Elf64_Rel *rel) |
1f8e2bcb AN |
3829 | { |
3830 | struct bpf_insn *insn = &prog->insns[insn_idx]; | |
3831 | size_t map_idx, nr_maps = prog->obj->nr_maps; | |
3832 | struct bpf_object *obj = prog->obj; | |
3833 | __u32 shdr_idx = sym->st_shndx; | |
3834 | enum libbpf_map_type type; | |
9c0f8cbd | 3835 | const char *sym_sec_name; |
1f8e2bcb AN |
3836 | struct bpf_map *map; |
3837 | ||
aa0b8d43 | 3838 | if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) { |
9c0f8cbd AN |
3839 | pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n", |
3840 | prog->name, sym_name, insn_idx, insn->code); | |
1f8e2bcb AN |
3841 | return -LIBBPF_ERRNO__RELOC; |
3842 | } | |
166750bc AN |
3843 | |
3844 | if (sym_is_extern(sym)) { | |
ad23b723 | 3845 | int sym_idx = ELF64_R_SYM(rel->r_info); |
166750bc AN |
3846 | int i, n = obj->nr_extern; |
3847 | struct extern_desc *ext; | |
3848 | ||
3849 | for (i = 0; i < n; i++) { | |
3850 | ext = &obj->externs[i]; | |
3851 | if (ext->sym_idx == sym_idx) | |
3852 | break; | |
3853 | } | |
3854 | if (i >= n) { | |
9c0f8cbd AN |
3855 | pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n", |
3856 | prog->name, sym_name, sym_idx); | |
166750bc AN |
3857 | return -LIBBPF_ERRNO__RELOC; |
3858 | } | |
9c0f8cbd AN |
3859 | pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n", |
3860 | prog->name, i, ext->name, ext->sym_idx, insn_idx); | |
5bd022ec MKL |
3861 | if (insn->code == (BPF_JMP | BPF_CALL)) |
3862 | reloc_desc->type = RELO_EXTERN_FUNC; | |
3863 | else | |
3864 | reloc_desc->type = RELO_EXTERN_VAR; | |
166750bc | 3865 | reloc_desc->insn_idx = insn_idx; |
2e33efe3 | 3866 | reloc_desc->sym_off = i; /* sym_off stores extern index */ |
166750bc AN |
3867 | return 0; |
3868 | } | |
3869 | ||
aa0b8d43 MKL |
3870 | /* sub-program call relocation */ |
3871 | if (is_call_insn(insn)) { | |
3872 | if (insn->src_reg != BPF_PSEUDO_CALL) { | |
3873 | pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name); | |
3874 | return -LIBBPF_ERRNO__RELOC; | |
3875 | } | |
3876 | /* text_shndx can be 0, if no default "main" program exists */ | |
3877 | if (!shdr_idx || shdr_idx != obj->efile.text_shndx) { | |
3878 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); | |
3879 | pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n", | |
3880 | prog->name, sym_name, sym_sec_name); | |
3881 | return -LIBBPF_ERRNO__RELOC; | |
3882 | } | |
3883 | if (sym->st_value % BPF_INSN_SZ) { | |
3884 | pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n", | |
3885 | prog->name, sym_name, (size_t)sym->st_value); | |
3886 | return -LIBBPF_ERRNO__RELOC; | |
3887 | } | |
3888 | reloc_desc->type = RELO_CALL; | |
3889 | reloc_desc->insn_idx = insn_idx; | |
3890 | reloc_desc->sym_off = sym->st_value; | |
3891 | return 0; | |
3892 | } | |
3893 | ||
1f8e2bcb | 3894 | if (!shdr_idx || shdr_idx >= SHN_LORESERVE) { |
9c0f8cbd AN |
3895 | pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n", |
3896 | prog->name, sym_name, shdr_idx); | |
1f8e2bcb AN |
3897 | return -LIBBPF_ERRNO__RELOC; |
3898 | } | |
3899 | ||
53eddb5e YS |
3900 | /* loading subprog addresses */ |
3901 | if (sym_is_subprog(sym, obj->efile.text_shndx)) { | |
3902 | /* global_func: sym->st_value = offset in the section, insn->imm = 0. | |
3903 | * local_func: sym->st_value = 0, insn->imm = offset in the section. | |
3904 | */ | |
3905 | if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) { | |
3906 | pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n", | |
3907 | prog->name, sym_name, (size_t)sym->st_value, insn->imm); | |
3908 | return -LIBBPF_ERRNO__RELOC; | |
3909 | } | |
3910 | ||
3911 | reloc_desc->type = RELO_SUBPROG_ADDR; | |
3912 | reloc_desc->insn_idx = insn_idx; | |
3913 | reloc_desc->sym_off = sym->st_value; | |
3914 | return 0; | |
3915 | } | |
3916 | ||
1f8e2bcb | 3917 | type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx); |
9c0f8cbd | 3918 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); |
1f8e2bcb AN |
3919 | |
3920 | /* generic map reference relocation */ | |
3921 | if (type == LIBBPF_MAP_UNSPEC) { | |
3922 | if (!bpf_object__shndx_is_maps(obj, shdr_idx)) { | |
9c0f8cbd AN |
3923 | pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n", |
3924 | prog->name, sym_name, sym_sec_name); | |
1f8e2bcb AN |
3925 | return -LIBBPF_ERRNO__RELOC; |
3926 | } | |
3927 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { | |
3928 | map = &obj->maps[map_idx]; | |
3929 | if (map->libbpf_type != type || | |
3930 | map->sec_idx != sym->st_shndx || | |
3931 | map->sec_offset != sym->st_value) | |
3932 | continue; | |
9c0f8cbd AN |
3933 | pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n", |
3934 | prog->name, map_idx, map->name, map->sec_idx, | |
1f8e2bcb AN |
3935 | map->sec_offset, insn_idx); |
3936 | break; | |
3937 | } | |
3938 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
3939 | pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n", |
3940 | prog->name, sym_sec_name, (size_t)sym->st_value); | |
1f8e2bcb AN |
3941 | return -LIBBPF_ERRNO__RELOC; |
3942 | } | |
3943 | reloc_desc->type = RELO_LD64; | |
3944 | reloc_desc->insn_idx = insn_idx; | |
3945 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 3946 | reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */ |
1f8e2bcb AN |
3947 | return 0; |
3948 | } | |
3949 | ||
3950 | /* global data map relocation */ | |
3951 | if (!bpf_object__shndx_is_data(obj, shdr_idx)) { | |
9c0f8cbd AN |
3952 | pr_warn("prog '%s': bad data relo against section '%s'\n", |
3953 | prog->name, sym_sec_name); | |
1f8e2bcb | 3954 | return -LIBBPF_ERRNO__RELOC; |
1f8e2bcb | 3955 | } |
1f8e2bcb AN |
3956 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { |
3957 | map = &obj->maps[map_idx]; | |
25bbbd7a | 3958 | if (map->libbpf_type != type || map->sec_idx != sym->st_shndx) |
1f8e2bcb | 3959 | continue; |
9c0f8cbd AN |
3960 | pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n", |
3961 | prog->name, map_idx, map->name, map->sec_idx, | |
3962 | map->sec_offset, insn_idx); | |
1f8e2bcb AN |
3963 | break; |
3964 | } | |
3965 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
3966 | pr_warn("prog '%s': data relo failed to find map for section '%s'\n", |
3967 | prog->name, sym_sec_name); | |
1f8e2bcb AN |
3968 | return -LIBBPF_ERRNO__RELOC; |
3969 | } | |
3970 | ||
3971 | reloc_desc->type = RELO_DATA; | |
3972 | reloc_desc->insn_idx = insn_idx; | |
3973 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 3974 | reloc_desc->sym_off = sym->st_value; |
1f8e2bcb AN |
3975 | return 0; |
3976 | } | |
3977 | ||
db2b8b06 AN |
3978 | static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx) |
3979 | { | |
3980 | return insn_idx >= prog->sec_insn_off && | |
3981 | insn_idx < prog->sec_insn_off + prog->sec_insn_cnt; | |
3982 | } | |
3983 | ||
3984 | static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj, | |
3985 | size_t sec_idx, size_t insn_idx) | |
3986 | { | |
3987 | int l = 0, r = obj->nr_programs - 1, m; | |
3988 | struct bpf_program *prog; | |
3989 | ||
3990 | while (l < r) { | |
3991 | m = l + (r - l + 1) / 2; | |
3992 | prog = &obj->programs[m]; | |
3993 | ||
3994 | if (prog->sec_idx < sec_idx || | |
3995 | (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx)) | |
3996 | l = m; | |
3997 | else | |
3998 | r = m - 1; | |
3999 | } | |
4000 | /* matching program could be at index l, but it still might be the | |
4001 | * wrong one, so we need to double check conditions for the last time | |
4002 | */ | |
4003 | prog = &obj->programs[l]; | |
4004 | if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx)) | |
4005 | return prog; | |
4006 | return NULL; | |
4007 | } | |
4008 | ||
34090915 | 4009 | static int |
ad23b723 | 4010 | bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data) |
34090915 | 4011 | { |
9c0f8cbd AN |
4012 | const char *relo_sec_name, *sec_name; |
4013 | size_t sec_idx = shdr->sh_info; | |
c3c55696 AN |
4014 | struct bpf_program *prog; |
4015 | struct reloc_desc *relos; | |
1f8e2bcb | 4016 | int err, i, nrels; |
c3c55696 AN |
4017 | const char *sym_name; |
4018 | __u32 insn_idx; | |
6245947c AN |
4019 | Elf_Scn *scn; |
4020 | Elf_Data *scn_data; | |
ad23b723 AN |
4021 | Elf64_Sym *sym; |
4022 | Elf64_Rel *rel; | |
34090915 | 4023 | |
6245947c AN |
4024 | scn = elf_sec_by_idx(obj, sec_idx); |
4025 | scn_data = elf_sec_data(obj, scn); | |
4026 | ||
9c0f8cbd | 4027 | relo_sec_name = elf_sec_str(obj, shdr->sh_name); |
6245947c | 4028 | sec_name = elf_sec_name(obj, scn); |
9c0f8cbd AN |
4029 | if (!relo_sec_name || !sec_name) |
4030 | return -EINVAL; | |
4031 | ||
4032 | pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n", | |
4033 | relo_sec_name, sec_idx, sec_name); | |
34090915 WN |
4034 | nrels = shdr->sh_size / shdr->sh_entsize; |
4035 | ||
34090915 | 4036 | for (i = 0; i < nrels; i++) { |
ad23b723 AN |
4037 | rel = elf_rel_by_idx(data, i); |
4038 | if (!rel) { | |
9c0f8cbd | 4039 | pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i); |
6371ca3b | 4040 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4041 | } |
ad23b723 AN |
4042 | |
4043 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); | |
4044 | if (!sym) { | |
9c0f8cbd | 4045 | pr_warn("sec '%s': symbol 0x%zx not found for relo #%d\n", |
ad23b723 | 4046 | relo_sec_name, (size_t)ELF64_R_SYM(rel->r_info), i); |
6371ca3b | 4047 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4048 | } |
6245947c | 4049 | |
ad23b723 | 4050 | if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) { |
9c0f8cbd | 4051 | pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n", |
ad23b723 | 4052 | relo_sec_name, (size_t)ELF64_R_SYM(rel->r_info), i); |
1f8e2bcb | 4053 | return -LIBBPF_ERRNO__FORMAT; |
9c0f8cbd | 4054 | } |
d859900c | 4055 | |
ad23b723 | 4056 | insn_idx = rel->r_offset / BPF_INSN_SZ; |
c3c55696 AN |
4057 | /* relocations against static functions are recorded as |
4058 | * relocations against the section that contains a function; | |
4059 | * in such case, symbol will be STT_SECTION and sym.st_name | |
4060 | * will point to empty string (0), so fetch section name | |
4061 | * instead | |
4062 | */ | |
ad23b723 AN |
4063 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0) |
4064 | sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx)); | |
c3c55696 | 4065 | else |
ad23b723 | 4066 | sym_name = elf_sym_str(obj, sym->st_name); |
c3c55696 | 4067 | sym_name = sym_name ?: "<?"; |
d859900c | 4068 | |
9c0f8cbd AN |
4069 | pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n", |
4070 | relo_sec_name, i, insn_idx, sym_name); | |
666810e8 | 4071 | |
c3c55696 AN |
4072 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); |
4073 | if (!prog) { | |
6245947c | 4074 | pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n", |
c3c55696 | 4075 | relo_sec_name, i, sec_name, insn_idx); |
6245947c | 4076 | continue; |
c3c55696 AN |
4077 | } |
4078 | ||
4079 | relos = libbpf_reallocarray(prog->reloc_desc, | |
4080 | prog->nr_reloc + 1, sizeof(*relos)); | |
4081 | if (!relos) | |
4082 | return -ENOMEM; | |
4083 | prog->reloc_desc = relos; | |
4084 | ||
4085 | /* adjust insn_idx to local BPF program frame of reference */ | |
4086 | insn_idx -= prog->sec_insn_off; | |
4087 | err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc], | |
ad23b723 | 4088 | insn_idx, sym_name, sym, rel); |
1f8e2bcb AN |
4089 | if (err) |
4090 | return err; | |
c3c55696 AN |
4091 | |
4092 | prog->nr_reloc++; | |
34090915 WN |
4093 | } |
4094 | return 0; | |
4095 | } | |
4096 | ||
abd29c93 | 4097 | static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map) |
8a138aed MKL |
4098 | { |
4099 | struct bpf_map_def *def = &map->def; | |
d859900c | 4100 | __u32 key_type_id = 0, value_type_id = 0; |
96408c43 | 4101 | int ret; |
8a138aed | 4102 | |
590a0088 MKL |
4103 | /* if it's BTF-defined map, we don't need to search for type IDs. |
4104 | * For struct_ops map, it does not need btf_key_type_id and | |
4105 | * btf_value_type_id. | |
4106 | */ | |
4107 | if (map->sec_idx == obj->efile.btf_maps_shndx || | |
4108 | bpf_map__is_struct_ops(map)) | |
abd29c93 AN |
4109 | return 0; |
4110 | ||
d859900c | 4111 | if (!bpf_map__is_internal(map)) { |
abd29c93 | 4112 | ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size, |
d859900c DB |
4113 | def->value_size, &key_type_id, |
4114 | &value_type_id); | |
4115 | } else { | |
4116 | /* | |
4117 | * LLVM annotates global data differently in BTF, that is, | |
4118 | * only as '.data', '.bss' or '.rodata'. | |
4119 | */ | |
aed65917 | 4120 | ret = btf__find_by_name(obj->btf, map->real_name); |
d859900c DB |
4121 | } |
4122 | if (ret < 0) | |
96408c43 | 4123 | return ret; |
8a138aed | 4124 | |
96408c43 | 4125 | map->btf_key_type_id = key_type_id; |
d859900c DB |
4126 | map->btf_value_type_id = bpf_map__is_internal(map) ? |
4127 | ret : value_type_id; | |
8a138aed MKL |
4128 | return 0; |
4129 | } | |
4130 | ||
97eb3138 MP |
4131 | static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info) |
4132 | { | |
4133 | char file[PATH_MAX], buff[4096]; | |
4134 | FILE *fp; | |
4135 | __u32 val; | |
4136 | int err; | |
4137 | ||
4138 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
4139 | memset(info, 0, sizeof(*info)); | |
4140 | ||
4141 | fp = fopen(file, "r"); | |
4142 | if (!fp) { | |
4143 | err = -errno; | |
4144 | pr_warn("failed to open %s: %d. No procfs support?\n", file, | |
4145 | err); | |
4146 | return err; | |
4147 | } | |
4148 | ||
4149 | while (fgets(buff, sizeof(buff), fp)) { | |
4150 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
4151 | info->type = val; | |
4152 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) | |
4153 | info->key_size = val; | |
4154 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) | |
4155 | info->value_size = val; | |
4156 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) | |
4157 | info->max_entries = val; | |
4158 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) | |
4159 | info->map_flags = val; | |
4160 | } | |
4161 | ||
4162 | fclose(fp); | |
4163 | ||
4164 | return 0; | |
4165 | } | |
4166 | ||
26736eb9 JK |
4167 | int bpf_map__reuse_fd(struct bpf_map *map, int fd) |
4168 | { | |
4169 | struct bpf_map_info info = {}; | |
4170 | __u32 len = sizeof(info); | |
4171 | int new_fd, err; | |
4172 | char *new_name; | |
4173 | ||
4174 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
97eb3138 MP |
4175 | if (err && errno == EINVAL) |
4176 | err = bpf_get_map_info_from_fdinfo(fd, &info); | |
26736eb9 | 4177 | if (err) |
e9fc3ce9 | 4178 | return libbpf_err(err); |
26736eb9 JK |
4179 | |
4180 | new_name = strdup(info.name); | |
4181 | if (!new_name) | |
e9fc3ce9 | 4182 | return libbpf_err(-errno); |
26736eb9 JK |
4183 | |
4184 | new_fd = open("/", O_RDONLY | O_CLOEXEC); | |
d1b4574a THJ |
4185 | if (new_fd < 0) { |
4186 | err = -errno; | |
26736eb9 | 4187 | goto err_free_new_name; |
d1b4574a | 4188 | } |
26736eb9 JK |
4189 | |
4190 | new_fd = dup3(fd, new_fd, O_CLOEXEC); | |
d1b4574a THJ |
4191 | if (new_fd < 0) { |
4192 | err = -errno; | |
26736eb9 | 4193 | goto err_close_new_fd; |
d1b4574a | 4194 | } |
26736eb9 JK |
4195 | |
4196 | err = zclose(map->fd); | |
d1b4574a THJ |
4197 | if (err) { |
4198 | err = -errno; | |
26736eb9 | 4199 | goto err_close_new_fd; |
d1b4574a | 4200 | } |
26736eb9 JK |
4201 | free(map->name); |
4202 | ||
4203 | map->fd = new_fd; | |
4204 | map->name = new_name; | |
4205 | map->def.type = info.type; | |
4206 | map->def.key_size = info.key_size; | |
4207 | map->def.value_size = info.value_size; | |
4208 | map->def.max_entries = info.max_entries; | |
4209 | map->def.map_flags = info.map_flags; | |
4210 | map->btf_key_type_id = info.btf_key_type_id; | |
4211 | map->btf_value_type_id = info.btf_value_type_id; | |
ec6d5f47 | 4212 | map->reused = true; |
26736eb9 JK |
4213 | |
4214 | return 0; | |
4215 | ||
4216 | err_close_new_fd: | |
4217 | close(new_fd); | |
4218 | err_free_new_name: | |
4219 | free(new_name); | |
e9fc3ce9 | 4220 | return libbpf_err(err); |
26736eb9 JK |
4221 | } |
4222 | ||
1bdb6c9a | 4223 | __u32 bpf_map__max_entries(const struct bpf_map *map) |
1a11a4c7 | 4224 | { |
1bdb6c9a AN |
4225 | return map->def.max_entries; |
4226 | } | |
1a11a4c7 | 4227 | |
b3278099 AN |
4228 | struct bpf_map *bpf_map__inner_map(struct bpf_map *map) |
4229 | { | |
4230 | if (!bpf_map_type__is_map_in_map(map->def.type)) | |
e9fc3ce9 | 4231 | return errno = EINVAL, NULL; |
b3278099 AN |
4232 | |
4233 | return map->inner_map; | |
4234 | } | |
4235 | ||
1bdb6c9a AN |
4236 | int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries) |
4237 | { | |
1a11a4c7 | 4238 | if (map->fd >= 0) |
e9fc3ce9 | 4239 | return libbpf_err(-EBUSY); |
1a11a4c7 | 4240 | map->def.max_entries = max_entries; |
1a11a4c7 AI |
4241 | return 0; |
4242 | } | |
4243 | ||
1bdb6c9a AN |
4244 | int bpf_map__resize(struct bpf_map *map, __u32 max_entries) |
4245 | { | |
4246 | if (!map || !max_entries) | |
e9fc3ce9 | 4247 | return libbpf_err(-EINVAL); |
1bdb6c9a AN |
4248 | |
4249 | return bpf_map__set_max_entries(map, max_entries); | |
4250 | } | |
4251 | ||
47eff617 | 4252 | static int |
fd9eef1a | 4253 | bpf_object__probe_loading(struct bpf_object *obj) |
47eff617 SF |
4254 | { |
4255 | struct bpf_load_program_attr attr; | |
4256 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4257 | struct bpf_insn insns[] = { | |
4258 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4259 | BPF_EXIT_INSN(), | |
4260 | }; | |
4261 | int ret; | |
4262 | ||
f9bceaa5 SF |
4263 | if (obj->gen_loader) |
4264 | return 0; | |
4265 | ||
47eff617 SF |
4266 | /* make sure basic loading works */ |
4267 | ||
4268 | memset(&attr, 0, sizeof(attr)); | |
4269 | attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; | |
4270 | attr.insns = insns; | |
4271 | attr.insns_cnt = ARRAY_SIZE(insns); | |
4272 | attr.license = "GPL"; | |
4273 | ||
4274 | ret = bpf_load_program_xattr(&attr, NULL, 0); | |
5c10a3db JE |
4275 | if (ret < 0) { |
4276 | attr.prog_type = BPF_PROG_TYPE_TRACEPOINT; | |
4277 | ret = bpf_load_program_xattr(&attr, NULL, 0); | |
4278 | } | |
47eff617 | 4279 | if (ret < 0) { |
fd9eef1a EC |
4280 | ret = errno; |
4281 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4282 | pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF " | |
4283 | "program. Make sure your kernel supports BPF " | |
4284 | "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is " | |
4285 | "set to big enough value.\n", __func__, cp, ret); | |
4286 | return -ret; | |
47eff617 SF |
4287 | } |
4288 | close(ret); | |
4289 | ||
fd9eef1a EC |
4290 | return 0; |
4291 | } | |
4292 | ||
bb180fb2 AN |
4293 | static int probe_fd(int fd) |
4294 | { | |
4295 | if (fd >= 0) | |
4296 | close(fd); | |
4297 | return fd >= 0; | |
4298 | } | |
4299 | ||
47b6cb4d | 4300 | static int probe_kern_prog_name(void) |
fd9eef1a EC |
4301 | { |
4302 | struct bpf_load_program_attr attr; | |
4303 | struct bpf_insn insns[] = { | |
4304 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4305 | BPF_EXIT_INSN(), | |
4306 | }; | |
4307 | int ret; | |
4308 | ||
4309 | /* make sure loading with name works */ | |
47eff617 | 4310 | |
fd9eef1a EC |
4311 | memset(&attr, 0, sizeof(attr)); |
4312 | attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; | |
4313 | attr.insns = insns; | |
4314 | attr.insns_cnt = ARRAY_SIZE(insns); | |
4315 | attr.license = "GPL"; | |
47eff617 SF |
4316 | attr.name = "test"; |
4317 | ret = bpf_load_program_xattr(&attr, NULL, 0); | |
bb180fb2 | 4318 | return probe_fd(ret); |
47eff617 SF |
4319 | } |
4320 | ||
47b6cb4d | 4321 | static int probe_kern_global_data(void) |
8837fe5d DB |
4322 | { |
4323 | struct bpf_load_program_attr prg_attr; | |
4324 | struct bpf_create_map_attr map_attr; | |
4325 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4326 | struct bpf_insn insns[] = { | |
4327 | BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16), | |
4328 | BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42), | |
4329 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4330 | BPF_EXIT_INSN(), | |
4331 | }; | |
4332 | int ret, map; | |
4333 | ||
4334 | memset(&map_attr, 0, sizeof(map_attr)); | |
4335 | map_attr.map_type = BPF_MAP_TYPE_ARRAY; | |
4336 | map_attr.key_size = sizeof(int); | |
4337 | map_attr.value_size = 32; | |
4338 | map_attr.max_entries = 1; | |
4339 | ||
4340 | map = bpf_create_map_xattr(&map_attr); | |
4341 | if (map < 0) { | |
23ab656b THJ |
4342 | ret = -errno; |
4343 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
be18010e | 4344 | pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n", |
23ab656b THJ |
4345 | __func__, cp, -ret); |
4346 | return ret; | |
8837fe5d DB |
4347 | } |
4348 | ||
4349 | insns[0].imm = map; | |
4350 | ||
4351 | memset(&prg_attr, 0, sizeof(prg_attr)); | |
4352 | prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; | |
4353 | prg_attr.insns = insns; | |
4354 | prg_attr.insns_cnt = ARRAY_SIZE(insns); | |
4355 | prg_attr.license = "GPL"; | |
4356 | ||
4357 | ret = bpf_load_program_xattr(&prg_attr, NULL, 0); | |
47b6cb4d | 4358 | close(map); |
bb180fb2 | 4359 | return probe_fd(ret); |
8837fe5d DB |
4360 | } |
4361 | ||
68b08647 AN |
4362 | static int probe_kern_btf(void) |
4363 | { | |
4364 | static const char strs[] = "\0int"; | |
4365 | __u32 types[] = { | |
4366 | /* int */ | |
4367 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), | |
4368 | }; | |
4369 | ||
4370 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4371 | strs, sizeof(strs))); | |
4372 | } | |
4373 | ||
47b6cb4d | 4374 | static int probe_kern_btf_func(void) |
d7c4b398 | 4375 | { |
8983b731 | 4376 | static const char strs[] = "\0int\0x\0a"; |
d7c4b398 AN |
4377 | /* void x(int a) {} */ |
4378 | __u32 types[] = { | |
4379 | /* int */ | |
4380 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4381 | /* FUNC_PROTO */ /* [2] */ | |
4382 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0), | |
4383 | BTF_PARAM_ENC(7, 1), | |
4384 | /* FUNC x */ /* [3] */ | |
4385 | BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2), | |
4386 | }; | |
d7c4b398 | 4387 | |
bb180fb2 AN |
4388 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4389 | strs, sizeof(strs))); | |
d7c4b398 AN |
4390 | } |
4391 | ||
47b6cb4d | 4392 | static int probe_kern_btf_func_global(void) |
2d3eb67f AS |
4393 | { |
4394 | static const char strs[] = "\0int\0x\0a"; | |
4395 | /* static void x(int a) {} */ | |
4396 | __u32 types[] = { | |
4397 | /* int */ | |
4398 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4399 | /* FUNC_PROTO */ /* [2] */ | |
4400 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0), | |
4401 | BTF_PARAM_ENC(7, 1), | |
4402 | /* FUNC x BTF_FUNC_GLOBAL */ /* [3] */ | |
4403 | BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2), | |
4404 | }; | |
2d3eb67f | 4405 | |
bb180fb2 AN |
4406 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4407 | strs, sizeof(strs))); | |
2d3eb67f AS |
4408 | } |
4409 | ||
47b6cb4d | 4410 | static int probe_kern_btf_datasec(void) |
d7c4b398 | 4411 | { |
8983b731 | 4412 | static const char strs[] = "\0x\0.data"; |
d7c4b398 AN |
4413 | /* static int a; */ |
4414 | __u32 types[] = { | |
4415 | /* int */ | |
4416 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4417 | /* VAR x */ /* [2] */ | |
4418 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1), | |
4419 | BTF_VAR_STATIC, | |
4420 | /* DATASEC val */ /* [3] */ | |
4421 | BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4), | |
4422 | BTF_VAR_SECINFO_ENC(2, 0, 4), | |
4423 | }; | |
cfd49210 | 4424 | |
bb180fb2 AN |
4425 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4426 | strs, sizeof(strs))); | |
d7c4b398 AN |
4427 | } |
4428 | ||
22541a9e IL |
4429 | static int probe_kern_btf_float(void) |
4430 | { | |
4431 | static const char strs[] = "\0float"; | |
4432 | __u32 types[] = { | |
4433 | /* float */ | |
4434 | BTF_TYPE_FLOAT_ENC(1, 4), | |
4435 | }; | |
4436 | ||
4437 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4438 | strs, sizeof(strs))); | |
4439 | } | |
4440 | ||
223f903e | 4441 | static int probe_kern_btf_decl_tag(void) |
5b84bd10 YS |
4442 | { |
4443 | static const char strs[] = "\0tag"; | |
4444 | __u32 types[] = { | |
4445 | /* int */ | |
4446 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4447 | /* VAR x */ /* [2] */ | |
4448 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1), | |
4449 | BTF_VAR_STATIC, | |
4450 | /* attr */ | |
223f903e | 4451 | BTF_TYPE_DECL_TAG_ENC(1, 2, -1), |
5b84bd10 YS |
4452 | }; |
4453 | ||
4454 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4455 | strs, sizeof(strs))); | |
4456 | } | |
4457 | ||
47b6cb4d | 4458 | static int probe_kern_array_mmap(void) |
7fe74b43 AN |
4459 | { |
4460 | struct bpf_create_map_attr attr = { | |
4461 | .map_type = BPF_MAP_TYPE_ARRAY, | |
4462 | .map_flags = BPF_F_MMAPABLE, | |
4463 | .key_size = sizeof(int), | |
4464 | .value_size = sizeof(int), | |
4465 | .max_entries = 1, | |
4466 | }; | |
7fe74b43 | 4467 | |
bb180fb2 | 4468 | return probe_fd(bpf_create_map_xattr(&attr)); |
7fe74b43 AN |
4469 | } |
4470 | ||
47b6cb4d | 4471 | static int probe_kern_exp_attach_type(void) |
25498a19 AN |
4472 | { |
4473 | struct bpf_load_program_attr attr; | |
4474 | struct bpf_insn insns[] = { | |
4475 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4476 | BPF_EXIT_INSN(), | |
4477 | }; | |
25498a19 AN |
4478 | |
4479 | memset(&attr, 0, sizeof(attr)); | |
4480 | /* use any valid combination of program type and (optional) | |
4481 | * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS) | |
4482 | * to see if kernel supports expected_attach_type field for | |
4483 | * BPF_PROG_LOAD command | |
4484 | */ | |
4485 | attr.prog_type = BPF_PROG_TYPE_CGROUP_SOCK; | |
4486 | attr.expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE; | |
4487 | attr.insns = insns; | |
4488 | attr.insns_cnt = ARRAY_SIZE(insns); | |
4489 | attr.license = "GPL"; | |
4490 | ||
bb180fb2 | 4491 | return probe_fd(bpf_load_program_xattr(&attr, NULL, 0)); |
25498a19 AN |
4492 | } |
4493 | ||
109cea5a AN |
4494 | static int probe_kern_probe_read_kernel(void) |
4495 | { | |
4496 | struct bpf_load_program_attr attr; | |
4497 | struct bpf_insn insns[] = { | |
4498 | BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), /* r1 = r10 (fp) */ | |
4499 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8), /* r1 += -8 */ | |
4500 | BPF_MOV64_IMM(BPF_REG_2, 8), /* r2 = 8 */ | |
4501 | BPF_MOV64_IMM(BPF_REG_3, 0), /* r3 = 0 */ | |
4502 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel), | |
4503 | BPF_EXIT_INSN(), | |
4504 | }; | |
4505 | ||
4506 | memset(&attr, 0, sizeof(attr)); | |
4507 | attr.prog_type = BPF_PROG_TYPE_KPROBE; | |
4508 | attr.insns = insns; | |
4509 | attr.insns_cnt = ARRAY_SIZE(insns); | |
4510 | attr.license = "GPL"; | |
4511 | ||
4512 | return probe_fd(bpf_load_program_xattr(&attr, NULL, 0)); | |
4513 | } | |
4514 | ||
5d23328d YZ |
4515 | static int probe_prog_bind_map(void) |
4516 | { | |
4517 | struct bpf_load_program_attr prg_attr; | |
4518 | struct bpf_create_map_attr map_attr; | |
4519 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4520 | struct bpf_insn insns[] = { | |
4521 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4522 | BPF_EXIT_INSN(), | |
4523 | }; | |
4524 | int ret, map, prog; | |
4525 | ||
4526 | memset(&map_attr, 0, sizeof(map_attr)); | |
4527 | map_attr.map_type = BPF_MAP_TYPE_ARRAY; | |
4528 | map_attr.key_size = sizeof(int); | |
4529 | map_attr.value_size = 32; | |
4530 | map_attr.max_entries = 1; | |
4531 | ||
4532 | map = bpf_create_map_xattr(&map_attr); | |
4533 | if (map < 0) { | |
4534 | ret = -errno; | |
4535 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4536 | pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n", | |
4537 | __func__, cp, -ret); | |
4538 | return ret; | |
4539 | } | |
4540 | ||
4541 | memset(&prg_attr, 0, sizeof(prg_attr)); | |
4542 | prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; | |
4543 | prg_attr.insns = insns; | |
4544 | prg_attr.insns_cnt = ARRAY_SIZE(insns); | |
4545 | prg_attr.license = "GPL"; | |
4546 | ||
4547 | prog = bpf_load_program_xattr(&prg_attr, NULL, 0); | |
4548 | if (prog < 0) { | |
4549 | close(map); | |
4550 | return 0; | |
4551 | } | |
4552 | ||
4553 | ret = bpf_prog_bind_map(prog, map, NULL); | |
4554 | ||
4555 | close(map); | |
4556 | close(prog); | |
4557 | ||
4558 | return ret >= 0; | |
4559 | } | |
4560 | ||
4f33a53d AN |
4561 | static int probe_module_btf(void) |
4562 | { | |
4563 | static const char strs[] = "\0int"; | |
4564 | __u32 types[] = { | |
4565 | /* int */ | |
4566 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), | |
4567 | }; | |
4568 | struct bpf_btf_info info; | |
4569 | __u32 len = sizeof(info); | |
4570 | char name[16]; | |
4571 | int fd, err; | |
4572 | ||
4573 | fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs)); | |
4574 | if (fd < 0) | |
4575 | return 0; /* BTF not supported at all */ | |
4576 | ||
4577 | memset(&info, 0, sizeof(info)); | |
4578 | info.name = ptr_to_u64(name); | |
4579 | info.name_len = sizeof(name); | |
4580 | ||
4581 | /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer; | |
4582 | * kernel's module BTF support coincides with support for | |
4583 | * name/name_len fields in struct bpf_btf_info. | |
4584 | */ | |
4585 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
4586 | close(fd); | |
4587 | return !err; | |
4588 | } | |
4589 | ||
668ace0e AN |
4590 | static int probe_perf_link(void) |
4591 | { | |
4592 | struct bpf_load_program_attr attr; | |
4593 | struct bpf_insn insns[] = { | |
4594 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4595 | BPF_EXIT_INSN(), | |
4596 | }; | |
4597 | int prog_fd, link_fd, err; | |
4598 | ||
4599 | memset(&attr, 0, sizeof(attr)); | |
4600 | attr.prog_type = BPF_PROG_TYPE_TRACEPOINT; | |
4601 | attr.insns = insns; | |
4602 | attr.insns_cnt = ARRAY_SIZE(insns); | |
4603 | attr.license = "GPL"; | |
4604 | prog_fd = bpf_load_program_xattr(&attr, NULL, 0); | |
4605 | if (prog_fd < 0) | |
4606 | return -errno; | |
4607 | ||
4608 | /* use invalid perf_event FD to get EBADF, if link is supported; | |
4609 | * otherwise EINVAL should be returned | |
4610 | */ | |
4611 | link_fd = bpf_link_create(prog_fd, -1, BPF_PERF_EVENT, NULL); | |
4612 | err = -errno; /* close() can clobber errno */ | |
4613 | ||
4614 | if (link_fd >= 0) | |
4615 | close(link_fd); | |
4616 | close(prog_fd); | |
4617 | ||
4618 | return link_fd < 0 && err == -EBADF; | |
4619 | } | |
4620 | ||
47b6cb4d AN |
4621 | enum kern_feature_result { |
4622 | FEAT_UNKNOWN = 0, | |
4623 | FEAT_SUPPORTED = 1, | |
4624 | FEAT_MISSING = 2, | |
4625 | }; | |
4626 | ||
4627 | typedef int (*feature_probe_fn)(void); | |
4628 | ||
4629 | static struct kern_feature_desc { | |
4630 | const char *desc; | |
4631 | feature_probe_fn probe; | |
4632 | enum kern_feature_result res; | |
4633 | } feature_probes[__FEAT_CNT] = { | |
4634 | [FEAT_PROG_NAME] = { | |
4635 | "BPF program name", probe_kern_prog_name, | |
4636 | }, | |
4637 | [FEAT_GLOBAL_DATA] = { | |
4638 | "global variables", probe_kern_global_data, | |
4639 | }, | |
68b08647 AN |
4640 | [FEAT_BTF] = { |
4641 | "minimal BTF", probe_kern_btf, | |
4642 | }, | |
47b6cb4d AN |
4643 | [FEAT_BTF_FUNC] = { |
4644 | "BTF functions", probe_kern_btf_func, | |
4645 | }, | |
4646 | [FEAT_BTF_GLOBAL_FUNC] = { | |
4647 | "BTF global function", probe_kern_btf_func_global, | |
4648 | }, | |
4649 | [FEAT_BTF_DATASEC] = { | |
4650 | "BTF data section and variable", probe_kern_btf_datasec, | |
4651 | }, | |
4652 | [FEAT_ARRAY_MMAP] = { | |
4653 | "ARRAY map mmap()", probe_kern_array_mmap, | |
4654 | }, | |
4655 | [FEAT_EXP_ATTACH_TYPE] = { | |
4656 | "BPF_PROG_LOAD expected_attach_type attribute", | |
4657 | probe_kern_exp_attach_type, | |
4658 | }, | |
109cea5a AN |
4659 | [FEAT_PROBE_READ_KERN] = { |
4660 | "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel, | |
5d23328d YZ |
4661 | }, |
4662 | [FEAT_PROG_BIND_MAP] = { | |
4663 | "BPF_PROG_BIND_MAP support", probe_prog_bind_map, | |
4f33a53d AN |
4664 | }, |
4665 | [FEAT_MODULE_BTF] = { | |
4666 | "module BTF support", probe_module_btf, | |
4667 | }, | |
22541a9e IL |
4668 | [FEAT_BTF_FLOAT] = { |
4669 | "BTF_KIND_FLOAT support", probe_kern_btf_float, | |
4670 | }, | |
668ace0e AN |
4671 | [FEAT_PERF_LINK] = { |
4672 | "BPF perf link support", probe_perf_link, | |
4673 | }, | |
223f903e YS |
4674 | [FEAT_BTF_DECL_TAG] = { |
4675 | "BTF_KIND_DECL_TAG support", probe_kern_btf_decl_tag, | |
5b84bd10 | 4676 | }, |
47b6cb4d | 4677 | }; |
8837fe5d | 4678 | |
9ca1f56a | 4679 | static bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id) |
47b6cb4d AN |
4680 | { |
4681 | struct kern_feature_desc *feat = &feature_probes[feat_id]; | |
4682 | int ret; | |
4683 | ||
67234743 AS |
4684 | if (obj->gen_loader) |
4685 | /* To generate loader program assume the latest kernel | |
4686 | * to avoid doing extra prog_load, map_create syscalls. | |
4687 | */ | |
4688 | return true; | |
4689 | ||
47b6cb4d AN |
4690 | if (READ_ONCE(feat->res) == FEAT_UNKNOWN) { |
4691 | ret = feat->probe(); | |
4692 | if (ret > 0) { | |
4693 | WRITE_ONCE(feat->res, FEAT_SUPPORTED); | |
4694 | } else if (ret == 0) { | |
4695 | WRITE_ONCE(feat->res, FEAT_MISSING); | |
4696 | } else { | |
4697 | pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret); | |
4698 | WRITE_ONCE(feat->res, FEAT_MISSING); | |
4699 | } | |
8837fe5d DB |
4700 | } |
4701 | ||
47b6cb4d | 4702 | return READ_ONCE(feat->res) == FEAT_SUPPORTED; |
47eff617 SF |
4703 | } |
4704 | ||
57a00f41 THJ |
4705 | static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd) |
4706 | { | |
4707 | struct bpf_map_info map_info = {}; | |
4708 | char msg[STRERR_BUFSIZE]; | |
4709 | __u32 map_info_len; | |
97eb3138 | 4710 | int err; |
57a00f41 THJ |
4711 | |
4712 | map_info_len = sizeof(map_info); | |
4713 | ||
97eb3138 MP |
4714 | err = bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len); |
4715 | if (err && errno == EINVAL) | |
4716 | err = bpf_get_map_info_from_fdinfo(map_fd, &map_info); | |
4717 | if (err) { | |
4718 | pr_warn("failed to get map info for map FD %d: %s\n", map_fd, | |
4719 | libbpf_strerror_r(errno, msg, sizeof(msg))); | |
57a00f41 THJ |
4720 | return false; |
4721 | } | |
4722 | ||
4723 | return (map_info.type == map->def.type && | |
4724 | map_info.key_size == map->def.key_size && | |
4725 | map_info.value_size == map->def.value_size && | |
4726 | map_info.max_entries == map->def.max_entries && | |
4727 | map_info.map_flags == map->def.map_flags); | |
4728 | } | |
4729 | ||
4730 | static int | |
4731 | bpf_object__reuse_map(struct bpf_map *map) | |
4732 | { | |
4733 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4734 | int err, pin_fd; | |
4735 | ||
4736 | pin_fd = bpf_obj_get(map->pin_path); | |
4737 | if (pin_fd < 0) { | |
4738 | err = -errno; | |
4739 | if (err == -ENOENT) { | |
4740 | pr_debug("found no pinned map to reuse at '%s'\n", | |
4741 | map->pin_path); | |
4742 | return 0; | |
4743 | } | |
4744 | ||
4745 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
4746 | pr_warn("couldn't retrieve pinned map '%s': %s\n", | |
4747 | map->pin_path, cp); | |
4748 | return err; | |
4749 | } | |
4750 | ||
4751 | if (!map_is_reuse_compat(map, pin_fd)) { | |
4752 | pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n", | |
4753 | map->pin_path); | |
4754 | close(pin_fd); | |
4755 | return -EINVAL; | |
4756 | } | |
4757 | ||
4758 | err = bpf_map__reuse_fd(map, pin_fd); | |
4759 | if (err) { | |
4760 | close(pin_fd); | |
4761 | return err; | |
4762 | } | |
4763 | map->pinned = true; | |
4764 | pr_debug("reused pinned map at '%s'\n", map->pin_path); | |
4765 | ||
4766 | return 0; | |
4767 | } | |
4768 | ||
d859900c DB |
4769 | static int |
4770 | bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map) | |
4771 | { | |
166750bc | 4772 | enum libbpf_map_type map_type = map->libbpf_type; |
d859900c DB |
4773 | char *cp, errmsg[STRERR_BUFSIZE]; |
4774 | int err, zero = 0; | |
d859900c | 4775 | |
67234743 AS |
4776 | if (obj->gen_loader) { |
4777 | bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps, | |
4778 | map->mmaped, map->def.value_size); | |
4779 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) | |
4780 | bpf_gen__map_freeze(obj->gen_loader, map - obj->maps); | |
4781 | return 0; | |
4782 | } | |
eba9c5f4 AN |
4783 | err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0); |
4784 | if (err) { | |
4785 | err = -errno; | |
4786 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
4787 | pr_warn("Error setting initial map(%s) contents: %s\n", | |
4788 | map->name, cp); | |
4789 | return err; | |
4790 | } | |
d859900c | 4791 | |
81bfdd08 AN |
4792 | /* Freeze .rodata and .kconfig map as read-only from syscall side. */ |
4793 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) { | |
d859900c DB |
4794 | err = bpf_map_freeze(map->fd); |
4795 | if (err) { | |
eba9c5f4 AN |
4796 | err = -errno; |
4797 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e KW |
4798 | pr_warn("Error freezing map(%s) as read-only: %s\n", |
4799 | map->name, cp); | |
eba9c5f4 | 4800 | return err; |
d859900c DB |
4801 | } |
4802 | } | |
eba9c5f4 | 4803 | return 0; |
d859900c DB |
4804 | } |
4805 | ||
2d39d7c5 AN |
4806 | static void bpf_map__destroy(struct bpf_map *map); |
4807 | ||
67234743 | 4808 | static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner) |
2d39d7c5 AN |
4809 | { |
4810 | struct bpf_create_map_attr create_attr; | |
4811 | struct bpf_map_def *def = &map->def; | |
a21ab4c5 | 4812 | int err = 0; |
2d39d7c5 AN |
4813 | |
4814 | memset(&create_attr, 0, sizeof(create_attr)); | |
4815 | ||
9ca1f56a | 4816 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
2d39d7c5 AN |
4817 | create_attr.name = map->name; |
4818 | create_attr.map_ifindex = map->map_ifindex; | |
4819 | create_attr.map_type = def->type; | |
4820 | create_attr.map_flags = def->map_flags; | |
4821 | create_attr.key_size = def->key_size; | |
4822 | create_attr.value_size = def->value_size; | |
1bdb6c9a | 4823 | create_attr.numa_node = map->numa_node; |
2d39d7c5 AN |
4824 | |
4825 | if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !def->max_entries) { | |
4826 | int nr_cpus; | |
4827 | ||
4828 | nr_cpus = libbpf_num_possible_cpus(); | |
4829 | if (nr_cpus < 0) { | |
4830 | pr_warn("map '%s': failed to determine number of system CPUs: %d\n", | |
4831 | map->name, nr_cpus); | |
4832 | return nr_cpus; | |
4833 | } | |
4834 | pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus); | |
4835 | create_attr.max_entries = nr_cpus; | |
4836 | } else { | |
4837 | create_attr.max_entries = def->max_entries; | |
4838 | } | |
4839 | ||
4840 | if (bpf_map__is_struct_ops(map)) | |
4841 | create_attr.btf_vmlinux_value_type_id = | |
4842 | map->btf_vmlinux_value_type_id; | |
4843 | ||
4844 | create_attr.btf_fd = 0; | |
4845 | create_attr.btf_key_type_id = 0; | |
4846 | create_attr.btf_value_type_id = 0; | |
0f0e55d8 | 4847 | if (obj->btf && btf__fd(obj->btf) >= 0 && !bpf_map_find_btf_info(obj, map)) { |
2d39d7c5 AN |
4848 | create_attr.btf_fd = btf__fd(obj->btf); |
4849 | create_attr.btf_key_type_id = map->btf_key_type_id; | |
4850 | create_attr.btf_value_type_id = map->btf_value_type_id; | |
4851 | } | |
4852 | ||
646f02ff AN |
4853 | if (bpf_map_type__is_map_in_map(def->type)) { |
4854 | if (map->inner_map) { | |
67234743 | 4855 | err = bpf_object__create_map(obj, map->inner_map, true); |
646f02ff AN |
4856 | if (err) { |
4857 | pr_warn("map '%s': failed to create inner map: %d\n", | |
4858 | map->name, err); | |
4859 | return err; | |
4860 | } | |
4861 | map->inner_map_fd = bpf_map__fd(map->inner_map); | |
4862 | } | |
4863 | if (map->inner_map_fd >= 0) | |
4864 | create_attr.inner_map_fd = map->inner_map_fd; | |
4865 | } | |
4866 | ||
f7310523 HC |
4867 | switch (def->type) { |
4868 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
4869 | case BPF_MAP_TYPE_CGROUP_ARRAY: | |
4870 | case BPF_MAP_TYPE_STACK_TRACE: | |
4871 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
4872 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
4873 | case BPF_MAP_TYPE_DEVMAP: | |
4874 | case BPF_MAP_TYPE_DEVMAP_HASH: | |
4875 | case BPF_MAP_TYPE_CPUMAP: | |
4876 | case BPF_MAP_TYPE_XSKMAP: | |
4877 | case BPF_MAP_TYPE_SOCKMAP: | |
4878 | case BPF_MAP_TYPE_SOCKHASH: | |
4879 | case BPF_MAP_TYPE_QUEUE: | |
4880 | case BPF_MAP_TYPE_STACK: | |
4881 | case BPF_MAP_TYPE_RINGBUF: | |
4882 | create_attr.btf_fd = 0; | |
4883 | create_attr.btf_key_type_id = 0; | |
4884 | create_attr.btf_value_type_id = 0; | |
4885 | map->btf_key_type_id = 0; | |
4886 | map->btf_value_type_id = 0; | |
4887 | default: | |
4888 | break; | |
4889 | } | |
4890 | ||
67234743 AS |
4891 | if (obj->gen_loader) { |
4892 | bpf_gen__map_create(obj->gen_loader, &create_attr, is_inner ? -1 : map - obj->maps); | |
4893 | /* Pretend to have valid FD to pass various fd >= 0 checks. | |
4894 | * This fd == 0 will not be used with any syscall and will be reset to -1 eventually. | |
4895 | */ | |
4896 | map->fd = 0; | |
4897 | } else { | |
4898 | map->fd = bpf_create_map_xattr(&create_attr); | |
4899 | } | |
2d39d7c5 AN |
4900 | if (map->fd < 0 && (create_attr.btf_key_type_id || |
4901 | create_attr.btf_value_type_id)) { | |
4902 | char *cp, errmsg[STRERR_BUFSIZE]; | |
2d39d7c5 | 4903 | |
a21ab4c5 | 4904 | err = -errno; |
2d39d7c5 AN |
4905 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); |
4906 | pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n", | |
4907 | map->name, cp, err); | |
4908 | create_attr.btf_fd = 0; | |
4909 | create_attr.btf_key_type_id = 0; | |
4910 | create_attr.btf_value_type_id = 0; | |
4911 | map->btf_key_type_id = 0; | |
4912 | map->btf_value_type_id = 0; | |
4913 | map->fd = bpf_create_map_xattr(&create_attr); | |
4914 | } | |
4915 | ||
a21ab4c5 | 4916 | err = map->fd < 0 ? -errno : 0; |
2d39d7c5 | 4917 | |
646f02ff | 4918 | if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) { |
67234743 AS |
4919 | if (obj->gen_loader) |
4920 | map->inner_map->fd = -1; | |
646f02ff AN |
4921 | bpf_map__destroy(map->inner_map); |
4922 | zfree(&map->inner_map); | |
4923 | } | |
4924 | ||
a21ab4c5 | 4925 | return err; |
2d39d7c5 AN |
4926 | } |
4927 | ||
67234743 | 4928 | static int init_map_slots(struct bpf_object *obj, struct bpf_map *map) |
a0f2b7ac HL |
4929 | { |
4930 | const struct bpf_map *targ_map; | |
4931 | unsigned int i; | |
67234743 | 4932 | int fd, err = 0; |
a0f2b7ac HL |
4933 | |
4934 | for (i = 0; i < map->init_slots_sz; i++) { | |
4935 | if (!map->init_slots[i]) | |
4936 | continue; | |
4937 | ||
4938 | targ_map = map->init_slots[i]; | |
4939 | fd = bpf_map__fd(targ_map); | |
67234743 | 4940 | if (obj->gen_loader) { |
edc0571c | 4941 | pr_warn("// TODO map_update_elem: idx %td key %d value==map_idx %td\n", |
67234743 AS |
4942 | map - obj->maps, i, targ_map - obj->maps); |
4943 | return -ENOTSUP; | |
4944 | } else { | |
4945 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
4946 | } | |
a0f2b7ac HL |
4947 | if (err) { |
4948 | err = -errno; | |
4949 | pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n", | |
4950 | map->name, i, targ_map->name, | |
4951 | fd, err); | |
4952 | return err; | |
4953 | } | |
4954 | pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n", | |
4955 | map->name, i, targ_map->name, fd); | |
4956 | } | |
4957 | ||
4958 | zfree(&map->init_slots); | |
4959 | map->init_slots_sz = 0; | |
4960 | ||
4961 | return 0; | |
4962 | } | |
4963 | ||
52d3352e WN |
4964 | static int |
4965 | bpf_object__create_maps(struct bpf_object *obj) | |
4966 | { | |
2d39d7c5 AN |
4967 | struct bpf_map *map; |
4968 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4969 | unsigned int i, j; | |
8a138aed | 4970 | int err; |
043c5bb3 | 4971 | bool retried; |
52d3352e | 4972 | |
9d759a9b | 4973 | for (i = 0; i < obj->nr_maps; i++) { |
2d39d7c5 | 4974 | map = &obj->maps[i]; |
8a138aed | 4975 | |
043c5bb3 MP |
4976 | retried = false; |
4977 | retry: | |
57a00f41 THJ |
4978 | if (map->pin_path) { |
4979 | err = bpf_object__reuse_map(map); | |
4980 | if (err) { | |
2d39d7c5 | 4981 | pr_warn("map '%s': error reusing pinned map\n", |
57a00f41 | 4982 | map->name); |
2d39d7c5 | 4983 | goto err_out; |
57a00f41 | 4984 | } |
043c5bb3 MP |
4985 | if (retried && map->fd < 0) { |
4986 | pr_warn("map '%s': cannot find pinned map\n", | |
4987 | map->name); | |
4988 | err = -ENOENT; | |
4989 | goto err_out; | |
4990 | } | |
57a00f41 THJ |
4991 | } |
4992 | ||
26736eb9 | 4993 | if (map->fd >= 0) { |
2d39d7c5 | 4994 | pr_debug("map '%s': skipping creation (preset fd=%d)\n", |
26736eb9 | 4995 | map->name, map->fd); |
2c193d32 | 4996 | } else { |
67234743 | 4997 | err = bpf_object__create_map(obj, map, false); |
2c193d32 | 4998 | if (err) |
d859900c | 4999 | goto err_out; |
d859900c | 5000 | |
2c193d32 HL |
5001 | pr_debug("map '%s': created successfully, fd=%d\n", |
5002 | map->name, map->fd); | |
646f02ff | 5003 | |
2c193d32 HL |
5004 | if (bpf_map__is_internal(map)) { |
5005 | err = bpf_object__populate_internal_map(obj, map); | |
5006 | if (err < 0) { | |
5007 | zclose(map->fd); | |
5008 | goto err_out; | |
5009 | } | |
d859900c | 5010 | } |
646f02ff | 5011 | |
2c193d32 | 5012 | if (map->init_slots_sz) { |
67234743 | 5013 | err = init_map_slots(obj, map); |
2c193d32 HL |
5014 | if (err < 0) { |
5015 | zclose(map->fd); | |
646f02ff AN |
5016 | goto err_out; |
5017 | } | |
646f02ff | 5018 | } |
646f02ff AN |
5019 | } |
5020 | ||
57a00f41 THJ |
5021 | if (map->pin_path && !map->pinned) { |
5022 | err = bpf_map__pin(map, NULL); | |
5023 | if (err) { | |
043c5bb3 MP |
5024 | zclose(map->fd); |
5025 | if (!retried && err == -EEXIST) { | |
5026 | retried = true; | |
5027 | goto retry; | |
5028 | } | |
2d39d7c5 AN |
5029 | pr_warn("map '%s': failed to auto-pin at '%s': %d\n", |
5030 | map->name, map->pin_path, err); | |
2d39d7c5 | 5031 | goto err_out; |
57a00f41 THJ |
5032 | } |
5033 | } | |
52d3352e WN |
5034 | } |
5035 | ||
52d3352e | 5036 | return 0; |
2d39d7c5 AN |
5037 | |
5038 | err_out: | |
5039 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
5040 | pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err); | |
5041 | pr_perm_msg(err); | |
5042 | for (j = 0; j < i; j++) | |
5043 | zclose(obj->maps[j].fd); | |
5044 | return err; | |
52d3352e WN |
5045 | } |
5046 | ||
ddc7c304 AN |
5047 | static bool bpf_core_is_flavor_sep(const char *s) |
5048 | { | |
5049 | /* check X___Y name pattern, where X and Y are not underscores */ | |
5050 | return s[0] != '_' && /* X */ | |
5051 | s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */ | |
5052 | s[4] != '_'; /* Y */ | |
5053 | } | |
5054 | ||
5055 | /* Given 'some_struct_name___with_flavor' return the length of a name prefix | |
5056 | * before last triple underscore. Struct name part after last triple | |
5057 | * underscore is ignored by BPF CO-RE relocation during relocation matching. | |
5058 | */ | |
b0588390 | 5059 | size_t bpf_core_essential_name_len(const char *name) |
ddc7c304 AN |
5060 | { |
5061 | size_t n = strlen(name); | |
5062 | int i; | |
5063 | ||
5064 | for (i = n - 5; i >= 0; i--) { | |
5065 | if (bpf_core_is_flavor_sep(name + i)) | |
5066 | return i + 1; | |
5067 | } | |
5068 | return n; | |
5069 | } | |
5070 | ||
301ba4d7 | 5071 | static void bpf_core_free_cands(struct bpf_core_cand_list *cands) |
ddc7c304 | 5072 | { |
0f7515ca AN |
5073 | free(cands->cands); |
5074 | free(cands); | |
ddc7c304 AN |
5075 | } |
5076 | ||
301ba4d7 | 5077 | static int bpf_core_add_cands(struct bpf_core_cand *local_cand, |
0f7515ca AN |
5078 | size_t local_essent_len, |
5079 | const struct btf *targ_btf, | |
5080 | const char *targ_btf_name, | |
5081 | int targ_start_id, | |
301ba4d7 | 5082 | struct bpf_core_cand_list *cands) |
ddc7c304 | 5083 | { |
301ba4d7 | 5084 | struct bpf_core_cand *new_cands, *cand; |
0f7515ca AN |
5085 | const struct btf_type *t; |
5086 | const char *targ_name; | |
5087 | size_t targ_essent_len; | |
5088 | int n, i; | |
ddc7c304 | 5089 | |
6a886de0 HC |
5090 | n = btf__type_cnt(targ_btf); |
5091 | for (i = targ_start_id; i < n; i++) { | |
ddc7c304 | 5092 | t = btf__type_by_id(targ_btf, i); |
0f7515ca | 5093 | if (btf_kind(t) != btf_kind(local_cand->t)) |
ddc7c304 AN |
5094 | continue; |
5095 | ||
3fc32f40 AN |
5096 | targ_name = btf__name_by_offset(targ_btf, t->name_off); |
5097 | if (str_is_empty(targ_name)) | |
d121e1d3 AN |
5098 | continue; |
5099 | ||
ddc7c304 AN |
5100 | targ_essent_len = bpf_core_essential_name_len(targ_name); |
5101 | if (targ_essent_len != local_essent_len) | |
5102 | continue; | |
5103 | ||
0f7515ca AN |
5104 | if (strncmp(local_cand->name, targ_name, local_essent_len) != 0) |
5105 | continue; | |
5106 | ||
5107 | pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n", | |
5108 | local_cand->id, btf_kind_str(local_cand->t), | |
5109 | local_cand->name, i, btf_kind_str(t), targ_name, | |
5110 | targ_btf_name); | |
5111 | new_cands = libbpf_reallocarray(cands->cands, cands->len + 1, | |
5112 | sizeof(*cands->cands)); | |
5113 | if (!new_cands) | |
5114 | return -ENOMEM; | |
5115 | ||
5116 | cand = &new_cands[cands->len]; | |
5117 | cand->btf = targ_btf; | |
5118 | cand->t = t; | |
5119 | cand->name = targ_name; | |
5120 | cand->id = i; | |
5121 | ||
5122 | cands->cands = new_cands; | |
5123 | cands->len++; | |
ddc7c304 | 5124 | } |
0f7515ca AN |
5125 | return 0; |
5126 | } | |
5127 | ||
4f33a53d AN |
5128 | static int load_module_btfs(struct bpf_object *obj) |
5129 | { | |
5130 | struct bpf_btf_info info; | |
5131 | struct module_btf *mod_btf; | |
5132 | struct btf *btf; | |
5133 | char name[64]; | |
5134 | __u32 id = 0, len; | |
5135 | int err, fd; | |
5136 | ||
5137 | if (obj->btf_modules_loaded) | |
5138 | return 0; | |
5139 | ||
67234743 AS |
5140 | if (obj->gen_loader) |
5141 | return 0; | |
5142 | ||
4f33a53d AN |
5143 | /* don't do this again, even if we find no module BTFs */ |
5144 | obj->btf_modules_loaded = true; | |
5145 | ||
5146 | /* kernel too old to support module BTFs */ | |
9ca1f56a | 5147 | if (!kernel_supports(obj, FEAT_MODULE_BTF)) |
4f33a53d AN |
5148 | return 0; |
5149 | ||
5150 | while (true) { | |
5151 | err = bpf_btf_get_next_id(id, &id); | |
5152 | if (err && errno == ENOENT) | |
5153 | return 0; | |
5154 | if (err) { | |
5155 | err = -errno; | |
5156 | pr_warn("failed to iterate BTF objects: %d\n", err); | |
5157 | return err; | |
ddc7c304 | 5158 | } |
4f33a53d AN |
5159 | |
5160 | fd = bpf_btf_get_fd_by_id(id); | |
5161 | if (fd < 0) { | |
5162 | if (errno == ENOENT) | |
5163 | continue; /* expected race: BTF was unloaded */ | |
5164 | err = -errno; | |
5165 | pr_warn("failed to get BTF object #%d FD: %d\n", id, err); | |
5166 | return err; | |
5167 | } | |
5168 | ||
5169 | len = sizeof(info); | |
5170 | memset(&info, 0, sizeof(info)); | |
5171 | info.name = ptr_to_u64(name); | |
5172 | info.name_len = sizeof(name); | |
5173 | ||
5174 | err = bpf_obj_get_info_by_fd(fd, &info, &len); | |
5175 | if (err) { | |
5176 | err = -errno; | |
5177 | pr_warn("failed to get BTF object #%d info: %d\n", id, err); | |
91abb4a6 | 5178 | goto err_out; |
4f33a53d AN |
5179 | } |
5180 | ||
5181 | /* ignore non-module BTFs */ | |
5182 | if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) { | |
5183 | close(fd); | |
5184 | continue; | |
5185 | } | |
5186 | ||
5187 | btf = btf_get_from_fd(fd, obj->btf_vmlinux); | |
e9fc3ce9 AN |
5188 | err = libbpf_get_error(btf); |
5189 | if (err) { | |
5190 | pr_warn("failed to load module [%s]'s BTF object #%d: %d\n", | |
5191 | name, id, err); | |
91abb4a6 | 5192 | goto err_out; |
4f33a53d AN |
5193 | } |
5194 | ||
3b029e06 AN |
5195 | err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap, |
5196 | sizeof(*obj->btf_modules), obj->btf_module_cnt + 1); | |
4f33a53d | 5197 | if (err) |
91abb4a6 | 5198 | goto err_out; |
4f33a53d AN |
5199 | |
5200 | mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; | |
5201 | ||
5202 | mod_btf->btf = btf; | |
5203 | mod_btf->id = id; | |
91abb4a6 | 5204 | mod_btf->fd = fd; |
4f33a53d | 5205 | mod_btf->name = strdup(name); |
91abb4a6 AN |
5206 | if (!mod_btf->name) { |
5207 | err = -ENOMEM; | |
5208 | goto err_out; | |
5209 | } | |
5210 | continue; | |
5211 | ||
5212 | err_out: | |
5213 | close(fd); | |
5214 | return err; | |
ddc7c304 | 5215 | } |
4f33a53d AN |
5216 | |
5217 | return 0; | |
5218 | } | |
5219 | ||
301ba4d7 | 5220 | static struct bpf_core_cand_list * |
0f7515ca AN |
5221 | bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id) |
5222 | { | |
301ba4d7 AS |
5223 | struct bpf_core_cand local_cand = {}; |
5224 | struct bpf_core_cand_list *cands; | |
4f33a53d | 5225 | const struct btf *main_btf; |
0f7515ca | 5226 | size_t local_essent_len; |
4f33a53d | 5227 | int err, i; |
0f7515ca AN |
5228 | |
5229 | local_cand.btf = local_btf; | |
5230 | local_cand.t = btf__type_by_id(local_btf, local_type_id); | |
5231 | if (!local_cand.t) | |
5232 | return ERR_PTR(-EINVAL); | |
5233 | ||
5234 | local_cand.name = btf__name_by_offset(local_btf, local_cand.t->name_off); | |
5235 | if (str_is_empty(local_cand.name)) | |
5236 | return ERR_PTR(-EINVAL); | |
5237 | local_essent_len = bpf_core_essential_name_len(local_cand.name); | |
5238 | ||
5239 | cands = calloc(1, sizeof(*cands)); | |
5240 | if (!cands) | |
5241 | return ERR_PTR(-ENOMEM); | |
5242 | ||
5243 | /* Attempt to find target candidates in vmlinux BTF first */ | |
4f33a53d AN |
5244 | main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux; |
5245 | err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands); | |
5246 | if (err) | |
5247 | goto err_out; | |
5248 | ||
5249 | /* if vmlinux BTF has any candidate, don't got for module BTFs */ | |
5250 | if (cands->len) | |
5251 | return cands; | |
5252 | ||
5253 | /* if vmlinux BTF was overridden, don't attempt to load module BTFs */ | |
5254 | if (obj->btf_vmlinux_override) | |
5255 | return cands; | |
5256 | ||
5257 | /* now look through module BTFs, trying to still find candidates */ | |
5258 | err = load_module_btfs(obj); | |
5259 | if (err) | |
5260 | goto err_out; | |
5261 | ||
5262 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
5263 | err = bpf_core_add_cands(&local_cand, local_essent_len, | |
5264 | obj->btf_modules[i].btf, | |
5265 | obj->btf_modules[i].name, | |
6a886de0 | 5266 | btf__type_cnt(obj->btf_vmlinux), |
4f33a53d AN |
5267 | cands); |
5268 | if (err) | |
5269 | goto err_out; | |
0f7515ca AN |
5270 | } |
5271 | ||
5272 | return cands; | |
ddc7c304 | 5273 | err_out: |
4f33a53d | 5274 | bpf_core_free_cands(cands); |
ddc7c304 AN |
5275 | return ERR_PTR(err); |
5276 | } | |
5277 | ||
3fc32f40 AN |
5278 | /* Check local and target types for compatibility. This check is used for |
5279 | * type-based CO-RE relocations and follow slightly different rules than | |
5280 | * field-based relocations. This function assumes that root types were already | |
5281 | * checked for name match. Beyond that initial root-level name check, names | |
5282 | * are completely ignored. Compatibility rules are as follows: | |
5283 | * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but | |
5284 | * kind should match for local and target types (i.e., STRUCT is not | |
5285 | * compatible with UNION); | |
5286 | * - for ENUMs, the size is ignored; | |
5287 | * - for INT, size and signedness are ignored; | |
5288 | * - for ARRAY, dimensionality is ignored, element types are checked for | |
5289 | * compatibility recursively; | |
5290 | * - CONST/VOLATILE/RESTRICT modifiers are ignored; | |
5291 | * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible; | |
5292 | * - FUNC_PROTOs are compatible if they have compatible signature: same | |
5293 | * number of input args and compatible return and argument types. | |
5294 | * These rules are not set in stone and probably will be adjusted as we get | |
5295 | * more experience with using BPF CO-RE relocations. | |
5296 | */ | |
b0588390 AS |
5297 | int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id, |
5298 | const struct btf *targ_btf, __u32 targ_id) | |
3fc32f40 AN |
5299 | { |
5300 | const struct btf_type *local_type, *targ_type; | |
5301 | int depth = 32; /* max recursion depth */ | |
5302 | ||
5303 | /* caller made sure that names match (ignoring flavor suffix) */ | |
5304 | local_type = btf__type_by_id(local_btf, local_id); | |
f872e4bc | 5305 | targ_type = btf__type_by_id(targ_btf, targ_id); |
3fc32f40 AN |
5306 | if (btf_kind(local_type) != btf_kind(targ_type)) |
5307 | return 0; | |
5308 | ||
5309 | recur: | |
5310 | depth--; | |
5311 | if (depth < 0) | |
5312 | return -EINVAL; | |
5313 | ||
5314 | local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id); | |
5315 | targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id); | |
5316 | if (!local_type || !targ_type) | |
5317 | return -EINVAL; | |
5318 | ||
5319 | if (btf_kind(local_type) != btf_kind(targ_type)) | |
5320 | return 0; | |
5321 | ||
5322 | switch (btf_kind(local_type)) { | |
5323 | case BTF_KIND_UNKN: | |
5324 | case BTF_KIND_STRUCT: | |
5325 | case BTF_KIND_UNION: | |
5326 | case BTF_KIND_ENUM: | |
5327 | case BTF_KIND_FWD: | |
5328 | return 1; | |
5329 | case BTF_KIND_INT: | |
5330 | /* just reject deprecated bitfield-like integers; all other | |
5331 | * integers are by default compatible between each other | |
5332 | */ | |
5333 | return btf_int_offset(local_type) == 0 && btf_int_offset(targ_type) == 0; | |
5334 | case BTF_KIND_PTR: | |
5335 | local_id = local_type->type; | |
5336 | targ_id = targ_type->type; | |
5337 | goto recur; | |
5338 | case BTF_KIND_ARRAY: | |
5339 | local_id = btf_array(local_type)->type; | |
5340 | targ_id = btf_array(targ_type)->type; | |
5341 | goto recur; | |
5342 | case BTF_KIND_FUNC_PROTO: { | |
5343 | struct btf_param *local_p = btf_params(local_type); | |
5344 | struct btf_param *targ_p = btf_params(targ_type); | |
5345 | __u16 local_vlen = btf_vlen(local_type); | |
5346 | __u16 targ_vlen = btf_vlen(targ_type); | |
5347 | int i, err; | |
5348 | ||
5349 | if (local_vlen != targ_vlen) | |
5350 | return 0; | |
5351 | ||
5352 | for (i = 0; i < local_vlen; i++, local_p++, targ_p++) { | |
5353 | skip_mods_and_typedefs(local_btf, local_p->type, &local_id); | |
5354 | skip_mods_and_typedefs(targ_btf, targ_p->type, &targ_id); | |
5355 | err = bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id); | |
5356 | if (err <= 0) | |
5357 | return err; | |
5358 | } | |
5359 | ||
5360 | /* tail recurse for return type check */ | |
5361 | skip_mods_and_typedefs(local_btf, local_type->type, &local_id); | |
5362 | skip_mods_and_typedefs(targ_btf, targ_type->type, &targ_id); | |
5363 | goto recur; | |
5364 | } | |
5365 | default: | |
5366 | pr_warn("unexpected kind %s relocated, local [%d], target [%d]\n", | |
5367 | btf_kind_str(local_type), local_id, targ_id); | |
5368 | return 0; | |
5369 | } | |
5370 | } | |
5371 | ||
ddc7c304 AN |
5372 | static size_t bpf_core_hash_fn(const void *key, void *ctx) |
5373 | { | |
5374 | return (size_t)key; | |
5375 | } | |
5376 | ||
5377 | static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx) | |
5378 | { | |
5379 | return k1 == k2; | |
5380 | } | |
5381 | ||
5382 | static void *u32_as_hash_key(__u32 x) | |
5383 | { | |
5384 | return (void *)(uintptr_t)x; | |
5385 | } | |
5386 | ||
3ee4f533 AS |
5387 | static int bpf_core_apply_relo(struct bpf_program *prog, |
5388 | const struct bpf_core_relo *relo, | |
5389 | int relo_idx, | |
5390 | const struct btf *local_btf, | |
5391 | struct hashmap *cand_cache) | |
5392 | { | |
5393 | const void *type_key = u32_as_hash_key(relo->type_id); | |
301ba4d7 | 5394 | struct bpf_core_cand_list *cands = NULL; |
3ee4f533 AS |
5395 | const char *prog_name = prog->name; |
5396 | const struct btf_type *local_type; | |
5397 | const char *local_name; | |
5398 | __u32 local_id = relo->type_id; | |
5399 | struct bpf_insn *insn; | |
5400 | int insn_idx, err; | |
5401 | ||
5402 | if (relo->insn_off % BPF_INSN_SZ) | |
5403 | return -EINVAL; | |
5404 | insn_idx = relo->insn_off / BPF_INSN_SZ; | |
5405 | /* adjust insn_idx from section frame of reference to the local | |
5406 | * program's frame of reference; (sub-)program code is not yet | |
5407 | * relocated, so it's enough to just subtract in-section offset | |
5408 | */ | |
5409 | insn_idx = insn_idx - prog->sec_insn_off; | |
de5d0dce | 5410 | if (insn_idx >= prog->insns_cnt) |
3ee4f533 AS |
5411 | return -EINVAL; |
5412 | insn = &prog->insns[insn_idx]; | |
5413 | ||
5414 | local_type = btf__type_by_id(local_btf, local_id); | |
5415 | if (!local_type) | |
5416 | return -EINVAL; | |
5417 | ||
5418 | local_name = btf__name_by_offset(local_btf, local_type->name_off); | |
5419 | if (!local_name) | |
5420 | return -EINVAL; | |
5421 | ||
5422 | if (prog->obj->gen_loader) { | |
5423 | pr_warn("// TODO core_relo: prog %td insn[%d] %s kind %d\n", | |
5424 | prog - prog->obj->programs, relo->insn_off / 8, | |
5425 | local_name, relo->kind); | |
5426 | return -ENOTSUP; | |
5427 | } | |
5428 | ||
5429 | if (relo->kind != BPF_TYPE_ID_LOCAL && | |
5430 | !hashmap__find(cand_cache, type_key, (void **)&cands)) { | |
5431 | cands = bpf_core_find_cands(prog->obj, local_btf, local_id); | |
5432 | if (IS_ERR(cands)) { | |
5433 | pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n", | |
5434 | prog_name, relo_idx, local_id, btf_kind_str(local_type), | |
5435 | local_name, PTR_ERR(cands)); | |
5436 | return PTR_ERR(cands); | |
5437 | } | |
5438 | err = hashmap__set(cand_cache, type_key, cands, NULL, NULL); | |
5439 | if (err) { | |
5440 | bpf_core_free_cands(cands); | |
5441 | return err; | |
5442 | } | |
5443 | } | |
5444 | ||
5445 | return bpf_core_apply_relo_insn(prog_name, insn, insn_idx, relo, relo_idx, local_btf, cands); | |
5446 | } | |
5447 | ||
ddc7c304 | 5448 | static int |
28b93c64 | 5449 | bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) |
ddc7c304 AN |
5450 | { |
5451 | const struct btf_ext_info_sec *sec; | |
28b93c64 | 5452 | const struct bpf_core_relo *rec; |
ddc7c304 AN |
5453 | const struct btf_ext_info *seg; |
5454 | struct hashmap_entry *entry; | |
5455 | struct hashmap *cand_cache = NULL; | |
5456 | struct bpf_program *prog; | |
ddc7c304 | 5457 | const char *sec_name; |
db2b8b06 | 5458 | int i, err = 0, insn_idx, sec_idx; |
ddc7c304 | 5459 | |
28b93c64 AN |
5460 | if (obj->btf_ext->core_relo_info.len == 0) |
5461 | return 0; | |
5462 | ||
0f7515ca AN |
5463 | if (targ_btf_path) { |
5464 | obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL); | |
e9fc3ce9 AN |
5465 | err = libbpf_get_error(obj->btf_vmlinux_override); |
5466 | if (err) { | |
0f7515ca AN |
5467 | pr_warn("failed to parse target BTF: %d\n", err); |
5468 | return err; | |
5469 | } | |
ddc7c304 AN |
5470 | } |
5471 | ||
5472 | cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL); | |
5473 | if (IS_ERR(cand_cache)) { | |
5474 | err = PTR_ERR(cand_cache); | |
5475 | goto out; | |
5476 | } | |
5477 | ||
28b93c64 | 5478 | seg = &obj->btf_ext->core_relo_info; |
ddc7c304 AN |
5479 | for_each_btf_ext_sec(seg, sec) { |
5480 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); | |
5481 | if (str_is_empty(sec_name)) { | |
5482 | err = -EINVAL; | |
5483 | goto out; | |
5484 | } | |
db2b8b06 AN |
5485 | /* bpf_object's ELF is gone by now so it's not easy to find |
5486 | * section index by section name, but we can find *any* | |
5487 | * bpf_program within desired section name and use it's | |
5488 | * prog->sec_idx to do a proper search by section index and | |
5489 | * instruction offset | |
5490 | */ | |
9c82a63c AN |
5491 | prog = NULL; |
5492 | for (i = 0; i < obj->nr_programs; i++) { | |
db2b8b06 | 5493 | prog = &obj->programs[i]; |
52109584 | 5494 | if (strcmp(prog->sec_name, sec_name) == 0) |
9c82a63c | 5495 | break; |
9c82a63c | 5496 | } |
ddc7c304 | 5497 | if (!prog) { |
db2b8b06 AN |
5498 | pr_warn("sec '%s': failed to find a BPF program\n", sec_name); |
5499 | return -ENOENT; | |
ddc7c304 | 5500 | } |
db2b8b06 | 5501 | sec_idx = prog->sec_idx; |
ddc7c304 | 5502 | |
9c0f8cbd | 5503 | pr_debug("sec '%s': found %d CO-RE relocations\n", |
ddc7c304 AN |
5504 | sec_name, sec->num_info); |
5505 | ||
5506 | for_each_btf_ext_rec(seg, sec, i, rec) { | |
db2b8b06 AN |
5507 | insn_idx = rec->insn_off / BPF_INSN_SZ; |
5508 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); | |
5509 | if (!prog) { | |
5510 | pr_warn("sec '%s': failed to find program at insn #%d for CO-RE offset relocation #%d\n", | |
5511 | sec_name, insn_idx, i); | |
5512 | err = -EINVAL; | |
5513 | goto out; | |
5514 | } | |
47f7cf63 AN |
5515 | /* no need to apply CO-RE relocation if the program is |
5516 | * not going to be loaded | |
5517 | */ | |
5518 | if (!prog->load) | |
5519 | continue; | |
db2b8b06 | 5520 | |
0f7515ca | 5521 | err = bpf_core_apply_relo(prog, rec, i, obj->btf, cand_cache); |
ddc7c304 | 5522 | if (err) { |
be18010e | 5523 | pr_warn("prog '%s': relo #%d: failed to relocate: %d\n", |
9c0f8cbd | 5524 | prog->name, i, err); |
ddc7c304 AN |
5525 | goto out; |
5526 | } | |
5527 | } | |
5528 | } | |
5529 | ||
5530 | out: | |
4f33a53d | 5531 | /* obj->btf_vmlinux and module BTFs are freed after object load */ |
0f7515ca AN |
5532 | btf__free(obj->btf_vmlinux_override); |
5533 | obj->btf_vmlinux_override = NULL; | |
5534 | ||
ddc7c304 AN |
5535 | if (!IS_ERR_OR_NULL(cand_cache)) { |
5536 | hashmap__for_each_entry(cand_cache, entry, i) { | |
5537 | bpf_core_free_cands(entry->value); | |
5538 | } | |
5539 | hashmap__free(cand_cache); | |
5540 | } | |
5541 | return err; | |
5542 | } | |
5543 | ||
c3c55696 AN |
5544 | /* Relocate data references within program code: |
5545 | * - map references; | |
5546 | * - global variable references; | |
5547 | * - extern references. | |
5548 | */ | |
48cca7e4 | 5549 | static int |
c3c55696 | 5550 | bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog) |
8a47a6c5 | 5551 | { |
c3c55696 | 5552 | int i; |
8a47a6c5 WN |
5553 | |
5554 | for (i = 0; i < prog->nr_reloc; i++) { | |
53f8dd43 | 5555 | struct reloc_desc *relo = &prog->reloc_desc[i]; |
166750bc | 5556 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; |
2e33efe3 | 5557 | struct extern_desc *ext; |
8a47a6c5 | 5558 | |
166750bc AN |
5559 | switch (relo->type) { |
5560 | case RELO_LD64: | |
e2fa0156 AS |
5561 | if (obj->gen_loader) { |
5562 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX; | |
5563 | insn[0].imm = relo->map_idx; | |
5564 | } else { | |
5565 | insn[0].src_reg = BPF_PSEUDO_MAP_FD; | |
5566 | insn[0].imm = obj->maps[relo->map_idx].fd; | |
5567 | } | |
166750bc AN |
5568 | break; |
5569 | case RELO_DATA: | |
166750bc | 5570 | insn[1].imm = insn[0].imm + relo->sym_off; |
e2fa0156 AS |
5571 | if (obj->gen_loader) { |
5572 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
5573 | insn[0].imm = relo->map_idx; | |
5574 | } else { | |
5575 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; | |
5576 | insn[0].imm = obj->maps[relo->map_idx].fd; | |
5577 | } | |
166750bc | 5578 | break; |
0c091e5c | 5579 | case RELO_EXTERN_VAR: |
2e33efe3 | 5580 | ext = &obj->externs[relo->sym_off]; |
1c0c7074 | 5581 | if (ext->type == EXT_KCFG) { |
e2fa0156 AS |
5582 | if (obj->gen_loader) { |
5583 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
5584 | insn[0].imm = obj->kconfig_map_idx; | |
5585 | } else { | |
5586 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; | |
5587 | insn[0].imm = obj->maps[obj->kconfig_map_idx].fd; | |
5588 | } | |
1c0c7074 AN |
5589 | insn[1].imm = ext->kcfg.data_off; |
5590 | } else /* EXT_KSYM */ { | |
2211c825 | 5591 | if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */ |
d370bbe1 | 5592 | insn[0].src_reg = BPF_PSEUDO_BTF_ID; |
284d2587 AN |
5593 | insn[0].imm = ext->ksym.kernel_btf_id; |
5594 | insn[1].imm = ext->ksym.kernel_btf_obj_fd; | |
2211c825 | 5595 | } else { /* typeless ksyms or unresolved typed ksyms */ |
d370bbe1 HL |
5596 | insn[0].imm = (__u32)ext->ksym.addr; |
5597 | insn[1].imm = ext->ksym.addr >> 32; | |
5598 | } | |
1c0c7074 | 5599 | } |
166750bc | 5600 | break; |
5bd022ec MKL |
5601 | case RELO_EXTERN_FUNC: |
5602 | ext = &obj->externs[relo->sym_off]; | |
5603 | insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL; | |
466b2e13 KKD |
5604 | if (ext->is_set) { |
5605 | insn[0].imm = ext->ksym.kernel_btf_id; | |
5606 | insn[0].off = ext->ksym.btf_fd_idx; | |
5607 | } else { /* unresolved weak kfunc */ | |
5608 | insn[0].imm = 0; | |
5609 | insn[0].off = 0; | |
5610 | } | |
5bd022ec | 5611 | break; |
53eddb5e | 5612 | case RELO_SUBPROG_ADDR: |
b1268826 AS |
5613 | if (insn[0].src_reg != BPF_PSEUDO_FUNC) { |
5614 | pr_warn("prog '%s': relo #%d: bad insn\n", | |
5615 | prog->name, i); | |
5616 | return -EINVAL; | |
5617 | } | |
5618 | /* handled already */ | |
53eddb5e | 5619 | break; |
166750bc | 5620 | case RELO_CALL: |
b1268826 | 5621 | /* handled already */ |
166750bc AN |
5622 | break; |
5623 | default: | |
9c0f8cbd AN |
5624 | pr_warn("prog '%s': relo #%d: bad relo type %d\n", |
5625 | prog->name, i, relo->type); | |
166750bc | 5626 | return -EINVAL; |
8a47a6c5 | 5627 | } |
8a47a6c5 WN |
5628 | } |
5629 | ||
c3c55696 AN |
5630 | return 0; |
5631 | } | |
5632 | ||
8505e870 AN |
5633 | static int adjust_prog_btf_ext_info(const struct bpf_object *obj, |
5634 | const struct bpf_program *prog, | |
5635 | const struct btf_ext_info *ext_info, | |
5636 | void **prog_info, __u32 *prog_rec_cnt, | |
5637 | __u32 *prog_rec_sz) | |
5638 | { | |
5639 | void *copy_start = NULL, *copy_end = NULL; | |
5640 | void *rec, *rec_end, *new_prog_info; | |
5641 | const struct btf_ext_info_sec *sec; | |
5642 | size_t old_sz, new_sz; | |
5643 | const char *sec_name; | |
5644 | int i, off_adj; | |
5645 | ||
5646 | for_each_btf_ext_sec(ext_info, sec) { | |
5647 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); | |
5648 | if (!sec_name) | |
5649 | return -EINVAL; | |
52109584 | 5650 | if (strcmp(sec_name, prog->sec_name) != 0) |
8505e870 AN |
5651 | continue; |
5652 | ||
5653 | for_each_btf_ext_rec(ext_info, sec, i, rec) { | |
5654 | __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ; | |
5655 | ||
5656 | if (insn_off < prog->sec_insn_off) | |
5657 | continue; | |
5658 | if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt) | |
5659 | break; | |
5660 | ||
5661 | if (!copy_start) | |
5662 | copy_start = rec; | |
5663 | copy_end = rec + ext_info->rec_size; | |
5664 | } | |
5665 | ||
5666 | if (!copy_start) | |
5667 | return -ENOENT; | |
5668 | ||
5669 | /* append func/line info of a given (sub-)program to the main | |
5670 | * program func/line info | |
5671 | */ | |
8eb62958 | 5672 | old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size; |
8505e870 AN |
5673 | new_sz = old_sz + (copy_end - copy_start); |
5674 | new_prog_info = realloc(*prog_info, new_sz); | |
5675 | if (!new_prog_info) | |
5676 | return -ENOMEM; | |
5677 | *prog_info = new_prog_info; | |
5678 | *prog_rec_cnt = new_sz / ext_info->rec_size; | |
5679 | memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start); | |
5680 | ||
5681 | /* Kernel instruction offsets are in units of 8-byte | |
5682 | * instructions, while .BTF.ext instruction offsets generated | |
5683 | * by Clang are in units of bytes. So convert Clang offsets | |
5684 | * into kernel offsets and adjust offset according to program | |
5685 | * relocated position. | |
5686 | */ | |
5687 | off_adj = prog->sub_insn_off - prog->sec_insn_off; | |
5688 | rec = new_prog_info + old_sz; | |
5689 | rec_end = new_prog_info + new_sz; | |
5690 | for (; rec < rec_end; rec += ext_info->rec_size) { | |
5691 | __u32 *insn_off = rec; | |
5692 | ||
5693 | *insn_off = *insn_off / BPF_INSN_SZ + off_adj; | |
5694 | } | |
5695 | *prog_rec_sz = ext_info->rec_size; | |
5696 | return 0; | |
5697 | } | |
5698 | ||
5699 | return -ENOENT; | |
5700 | } | |
5701 | ||
5702 | static int | |
5703 | reloc_prog_func_and_line_info(const struct bpf_object *obj, | |
5704 | struct bpf_program *main_prog, | |
5705 | const struct bpf_program *prog) | |
5706 | { | |
5707 | int err; | |
5708 | ||
5709 | /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't | |
5710 | * supprot func/line info | |
5711 | */ | |
9ca1f56a | 5712 | if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC)) |
8505e870 AN |
5713 | return 0; |
5714 | ||
5715 | /* only attempt func info relocation if main program's func_info | |
5716 | * relocation was successful | |
5717 | */ | |
5718 | if (main_prog != prog && !main_prog->func_info) | |
5719 | goto line_info; | |
5720 | ||
5721 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info, | |
5722 | &main_prog->func_info, | |
5723 | &main_prog->func_info_cnt, | |
5724 | &main_prog->func_info_rec_size); | |
5725 | if (err) { | |
5726 | if (err != -ENOENT) { | |
5727 | pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n", | |
5728 | prog->name, err); | |
5729 | return err; | |
5730 | } | |
5731 | if (main_prog->func_info) { | |
5732 | /* | |
5733 | * Some info has already been found but has problem | |
5734 | * in the last btf_ext reloc. Must have to error out. | |
5735 | */ | |
5736 | pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name); | |
5737 | return err; | |
5738 | } | |
5739 | /* Have problem loading the very first info. Ignore the rest. */ | |
5740 | pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n", | |
5741 | prog->name); | |
5742 | } | |
5743 | ||
5744 | line_info: | |
5745 | /* don't relocate line info if main program's relocation failed */ | |
5746 | if (main_prog != prog && !main_prog->line_info) | |
5747 | return 0; | |
5748 | ||
5749 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info, | |
5750 | &main_prog->line_info, | |
5751 | &main_prog->line_info_cnt, | |
5752 | &main_prog->line_info_rec_size); | |
5753 | if (err) { | |
5754 | if (err != -ENOENT) { | |
5755 | pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n", | |
5756 | prog->name, err); | |
5757 | return err; | |
5758 | } | |
5759 | if (main_prog->line_info) { | |
5760 | /* | |
5761 | * Some info has already been found but has problem | |
5762 | * in the last btf_ext reloc. Must have to error out. | |
5763 | */ | |
5764 | pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name); | |
5765 | return err; | |
5766 | } | |
5767 | /* Have problem loading the very first info. Ignore the rest. */ | |
5768 | pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n", | |
5769 | prog->name); | |
5770 | } | |
5771 | return 0; | |
5772 | } | |
5773 | ||
c3c55696 AN |
5774 | static int cmp_relo_by_insn_idx(const void *key, const void *elem) |
5775 | { | |
5776 | size_t insn_idx = *(const size_t *)key; | |
5777 | const struct reloc_desc *relo = elem; | |
5778 | ||
5779 | if (insn_idx == relo->insn_idx) | |
5780 | return 0; | |
5781 | return insn_idx < relo->insn_idx ? -1 : 1; | |
5782 | } | |
5783 | ||
5784 | static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx) | |
5785 | { | |
5786 | return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc, | |
5787 | sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx); | |
5788 | } | |
5789 | ||
b1268826 AS |
5790 | static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog) |
5791 | { | |
5792 | int new_cnt = main_prog->nr_reloc + subprog->nr_reloc; | |
5793 | struct reloc_desc *relos; | |
5794 | int i; | |
5795 | ||
5796 | if (main_prog == subprog) | |
5797 | return 0; | |
5798 | relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos)); | |
5799 | if (!relos) | |
5800 | return -ENOMEM; | |
5801 | memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc, | |
5802 | sizeof(*relos) * subprog->nr_reloc); | |
5803 | ||
5804 | for (i = main_prog->nr_reloc; i < new_cnt; i++) | |
5805 | relos[i].insn_idx += subprog->sub_insn_off; | |
5806 | /* After insn_idx adjustment the 'relos' array is still sorted | |
5807 | * by insn_idx and doesn't break bsearch. | |
5808 | */ | |
5809 | main_prog->reloc_desc = relos; | |
5810 | main_prog->nr_reloc = new_cnt; | |
5811 | return 0; | |
5812 | } | |
5813 | ||
c3c55696 AN |
5814 | static int |
5815 | bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog, | |
5816 | struct bpf_program *prog) | |
5817 | { | |
5818 | size_t sub_insn_idx, insn_idx, new_cnt; | |
5819 | struct bpf_program *subprog; | |
5820 | struct bpf_insn *insns, *insn; | |
5821 | struct reloc_desc *relo; | |
5822 | int err; | |
5823 | ||
5824 | err = reloc_prog_func_and_line_info(obj, main_prog, prog); | |
5825 | if (err) | |
5826 | return err; | |
5827 | ||
5828 | for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) { | |
5829 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
53eddb5e | 5830 | if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn)) |
c3c55696 AN |
5831 | continue; |
5832 | ||
5833 | relo = find_prog_insn_relo(prog, insn_idx); | |
b1268826 AS |
5834 | if (relo && relo->type == RELO_EXTERN_FUNC) |
5835 | /* kfunc relocations will be handled later | |
5836 | * in bpf_object__relocate_data() | |
5837 | */ | |
5838 | continue; | |
53eddb5e | 5839 | if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) { |
c3c55696 AN |
5840 | pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n", |
5841 | prog->name, insn_idx, relo->type); | |
5842 | return -LIBBPF_ERRNO__RELOC; | |
5843 | } | |
5844 | if (relo) { | |
5845 | /* sub-program instruction index is a combination of | |
5846 | * an offset of a symbol pointed to by relocation and | |
5847 | * call instruction's imm field; for global functions, | |
5848 | * call always has imm = -1, but for static functions | |
5849 | * relocation is against STT_SECTION and insn->imm | |
5850 | * points to a start of a static function | |
53eddb5e YS |
5851 | * |
5852 | * for subprog addr relocation, the relo->sym_off + insn->imm is | |
5853 | * the byte offset in the corresponding section. | |
c3c55696 | 5854 | */ |
53eddb5e YS |
5855 | if (relo->type == RELO_CALL) |
5856 | sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1; | |
5857 | else | |
5858 | sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ; | |
5859 | } else if (insn_is_pseudo_func(insn)) { | |
5860 | /* | |
5861 | * RELO_SUBPROG_ADDR relo is always emitted even if both | |
5862 | * functions are in the same section, so it shouldn't reach here. | |
5863 | */ | |
5864 | pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n", | |
5865 | prog->name, insn_idx); | |
5866 | return -LIBBPF_ERRNO__RELOC; | |
c3c55696 AN |
5867 | } else { |
5868 | /* if subprogram call is to a static function within | |
5869 | * the same ELF section, there won't be any relocation | |
5870 | * emitted, but it also means there is no additional | |
5871 | * offset necessary, insns->imm is relative to | |
5872 | * instruction's original position within the section | |
5873 | */ | |
5874 | sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1; | |
5875 | } | |
5876 | ||
5877 | /* we enforce that sub-programs should be in .text section */ | |
5878 | subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx); | |
5879 | if (!subprog) { | |
5880 | pr_warn("prog '%s': no .text section found yet sub-program call exists\n", | |
5881 | prog->name); | |
5882 | return -LIBBPF_ERRNO__RELOC; | |
5883 | } | |
5884 | ||
5885 | /* if it's the first call instruction calling into this | |
5886 | * subprogram (meaning this subprog hasn't been processed | |
5887 | * yet) within the context of current main program: | |
5888 | * - append it at the end of main program's instructions blog; | |
5889 | * - process is recursively, while current program is put on hold; | |
5890 | * - if that subprogram calls some other not yet processes | |
5891 | * subprogram, same thing will happen recursively until | |
5892 | * there are no more unprocesses subprograms left to append | |
5893 | * and relocate. | |
5894 | */ | |
5895 | if (subprog->sub_insn_off == 0) { | |
5896 | subprog->sub_insn_off = main_prog->insns_cnt; | |
5897 | ||
5898 | new_cnt = main_prog->insns_cnt + subprog->insns_cnt; | |
5899 | insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns)); | |
5900 | if (!insns) { | |
5901 | pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name); | |
5902 | return -ENOMEM; | |
5903 | } | |
5904 | main_prog->insns = insns; | |
5905 | main_prog->insns_cnt = new_cnt; | |
5906 | ||
5907 | memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns, | |
5908 | subprog->insns_cnt * sizeof(*insns)); | |
5909 | ||
5910 | pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n", | |
5911 | main_prog->name, subprog->insns_cnt, subprog->name); | |
5912 | ||
b1268826 AS |
5913 | /* The subprog insns are now appended. Append its relos too. */ |
5914 | err = append_subprog_relos(main_prog, subprog); | |
5915 | if (err) | |
5916 | return err; | |
c3c55696 AN |
5917 | err = bpf_object__reloc_code(obj, main_prog, subprog); |
5918 | if (err) | |
5919 | return err; | |
5920 | } | |
5921 | ||
5922 | /* main_prog->insns memory could have been re-allocated, so | |
5923 | * calculate pointer again | |
5924 | */ | |
5925 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
5926 | /* calculate correct instruction position within current main | |
5927 | * prog; each main prog can have a different set of | |
5928 | * subprograms appended (potentially in different order as | |
5929 | * well), so position of any subprog can be different for | |
5930 | * different main programs */ | |
5931 | insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1; | |
5932 | ||
c3c55696 AN |
5933 | pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n", |
5934 | prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off); | |
5935 | } | |
5936 | ||
5937 | return 0; | |
5938 | } | |
5939 | ||
5940 | /* | |
5941 | * Relocate sub-program calls. | |
5942 | * | |
5943 | * Algorithm operates as follows. Each entry-point BPF program (referred to as | |
5944 | * main prog) is processed separately. For each subprog (non-entry functions, | |
5945 | * that can be called from either entry progs or other subprogs) gets their | |
5946 | * sub_insn_off reset to zero. This serves as indicator that this subprogram | |
5947 | * hasn't been yet appended and relocated within current main prog. Once its | |
5948 | * relocated, sub_insn_off will point at the position within current main prog | |
5949 | * where given subprog was appended. This will further be used to relocate all | |
5950 | * the call instructions jumping into this subprog. | |
5951 | * | |
5952 | * We start with main program and process all call instructions. If the call | |
5953 | * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off | |
5954 | * is zero), subprog instructions are appended at the end of main program's | |
5955 | * instruction array. Then main program is "put on hold" while we recursively | |
5956 | * process newly appended subprogram. If that subprogram calls into another | |
5957 | * subprogram that hasn't been appended, new subprogram is appended again to | |
5958 | * the *main* prog's instructions (subprog's instructions are always left | |
5959 | * untouched, as they need to be in unmodified state for subsequent main progs | |
5960 | * and subprog instructions are always sent only as part of a main prog) and | |
5961 | * the process continues recursively. Once all the subprogs called from a main | |
5962 | * prog or any of its subprogs are appended (and relocated), all their | |
5963 | * positions within finalized instructions array are known, so it's easy to | |
5964 | * rewrite call instructions with correct relative offsets, corresponding to | |
5965 | * desired target subprog. | |
5966 | * | |
5967 | * Its important to realize that some subprogs might not be called from some | |
5968 | * main prog and any of its called/used subprogs. Those will keep their | |
5969 | * subprog->sub_insn_off as zero at all times and won't be appended to current | |
5970 | * main prog and won't be relocated within the context of current main prog. | |
5971 | * They might still be used from other main progs later. | |
5972 | * | |
5973 | * Visually this process can be shown as below. Suppose we have two main | |
5974 | * programs mainA and mainB and BPF object contains three subprogs: subA, | |
5975 | * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and | |
5976 | * subC both call subB: | |
5977 | * | |
5978 | * +--------+ +-------+ | |
5979 | * | v v | | |
5980 | * +--+---+ +--+-+-+ +---+--+ | |
5981 | * | subA | | subB | | subC | | |
5982 | * +--+---+ +------+ +---+--+ | |
5983 | * ^ ^ | |
5984 | * | | | |
5985 | * +---+-------+ +------+----+ | |
5986 | * | mainA | | mainB | | |
5987 | * +-----------+ +-----------+ | |
5988 | * | |
5989 | * We'll start relocating mainA, will find subA, append it and start | |
5990 | * processing sub A recursively: | |
5991 | * | |
5992 | * +-----------+------+ | |
5993 | * | mainA | subA | | |
5994 | * +-----------+------+ | |
5995 | * | |
5996 | * At this point we notice that subB is used from subA, so we append it and | |
5997 | * relocate (there are no further subcalls from subB): | |
5998 | * | |
5999 | * +-----------+------+------+ | |
6000 | * | mainA | subA | subB | | |
6001 | * +-----------+------+------+ | |
6002 | * | |
6003 | * At this point, we relocate subA calls, then go one level up and finish with | |
6004 | * relocatin mainA calls. mainA is done. | |
6005 | * | |
6006 | * For mainB process is similar but results in different order. We start with | |
6007 | * mainB and skip subA and subB, as mainB never calls them (at least | |
6008 | * directly), but we see subC is needed, so we append and start processing it: | |
6009 | * | |
6010 | * +-----------+------+ | |
6011 | * | mainB | subC | | |
6012 | * +-----------+------+ | |
6013 | * Now we see subC needs subB, so we go back to it, append and relocate it: | |
6014 | * | |
6015 | * +-----------+------+------+ | |
6016 | * | mainB | subC | subB | | |
6017 | * +-----------+------+------+ | |
6018 | * | |
6019 | * At this point we unwind recursion, relocate calls in subC, then in mainB. | |
6020 | */ | |
6021 | static int | |
6022 | bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog) | |
6023 | { | |
6024 | struct bpf_program *subprog; | |
d3d93e34 | 6025 | int i, err; |
c3c55696 | 6026 | |
c3c55696 AN |
6027 | /* mark all subprogs as not relocated (yet) within the context of |
6028 | * current main program | |
6029 | */ | |
6030 | for (i = 0; i < obj->nr_programs; i++) { | |
6031 | subprog = &obj->programs[i]; | |
6032 | if (!prog_is_subprog(obj, subprog)) | |
6033 | continue; | |
6034 | ||
6035 | subprog->sub_insn_off = 0; | |
c3c55696 AN |
6036 | } |
6037 | ||
6038 | err = bpf_object__reloc_code(obj, prog, prog); | |
6039 | if (err) | |
6040 | return err; | |
6041 | ||
6042 | ||
8a47a6c5 WN |
6043 | return 0; |
6044 | } | |
6045 | ||
67234743 AS |
6046 | static void |
6047 | bpf_object__free_relocs(struct bpf_object *obj) | |
6048 | { | |
6049 | struct bpf_program *prog; | |
6050 | int i; | |
6051 | ||
6052 | /* free up relocation descriptors */ | |
6053 | for (i = 0; i < obj->nr_programs; i++) { | |
6054 | prog = &obj->programs[i]; | |
6055 | zfree(&prog->reloc_desc); | |
6056 | prog->nr_reloc = 0; | |
6057 | } | |
6058 | } | |
6059 | ||
8a47a6c5 | 6060 | static int |
ddc7c304 | 6061 | bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path) |
8a47a6c5 WN |
6062 | { |
6063 | struct bpf_program *prog; | |
b1268826 | 6064 | size_t i, j; |
8a47a6c5 WN |
6065 | int err; |
6066 | ||
ddc7c304 AN |
6067 | if (obj->btf_ext) { |
6068 | err = bpf_object__relocate_core(obj, targ_btf_path); | |
6069 | if (err) { | |
be18010e KW |
6070 | pr_warn("failed to perform CO-RE relocations: %d\n", |
6071 | err); | |
ddc7c304 AN |
6072 | return err; |
6073 | } | |
6074 | } | |
b1268826 AS |
6075 | |
6076 | /* Before relocating calls pre-process relocations and mark | |
6077 | * few ld_imm64 instructions that points to subprogs. | |
6078 | * Otherwise bpf_object__reloc_code() later would have to consider | |
6079 | * all ld_imm64 insns as relocation candidates. That would | |
6080 | * reduce relocation speed, since amount of find_prog_insn_relo() | |
6081 | * would increase and most of them will fail to find a relo. | |
9173cac3 AN |
6082 | */ |
6083 | for (i = 0; i < obj->nr_programs; i++) { | |
6084 | prog = &obj->programs[i]; | |
b1268826 AS |
6085 | for (j = 0; j < prog->nr_reloc; j++) { |
6086 | struct reloc_desc *relo = &prog->reloc_desc[j]; | |
6087 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; | |
6088 | ||
6089 | /* mark the insn, so it's recognized by insn_is_pseudo_func() */ | |
6090 | if (relo->type == RELO_SUBPROG_ADDR) | |
6091 | insn[0].src_reg = BPF_PSEUDO_FUNC; | |
9173cac3 | 6092 | } |
9173cac3 | 6093 | } |
b1268826 AS |
6094 | |
6095 | /* relocate subprogram calls and append used subprograms to main | |
c3c55696 AN |
6096 | * programs; each copy of subprogram code needs to be relocated |
6097 | * differently for each main program, because its code location might | |
b1268826 AS |
6098 | * have changed. |
6099 | * Append subprog relos to main programs to allow data relos to be | |
6100 | * processed after text is completely relocated. | |
9173cac3 | 6101 | */ |
8a47a6c5 WN |
6102 | for (i = 0; i < obj->nr_programs; i++) { |
6103 | prog = &obj->programs[i]; | |
c3c55696 AN |
6104 | /* sub-program's sub-calls are relocated within the context of |
6105 | * its main program only | |
6106 | */ | |
6107 | if (prog_is_subprog(obj, prog)) | |
9173cac3 | 6108 | continue; |
8a47a6c5 | 6109 | |
c3c55696 | 6110 | err = bpf_object__relocate_calls(obj, prog); |
8a47a6c5 | 6111 | if (err) { |
9c0f8cbd AN |
6112 | pr_warn("prog '%s': failed to relocate calls: %d\n", |
6113 | prog->name, err); | |
8a47a6c5 WN |
6114 | return err; |
6115 | } | |
6116 | } | |
b1268826 AS |
6117 | /* Process data relos for main programs */ |
6118 | for (i = 0; i < obj->nr_programs; i++) { | |
6119 | prog = &obj->programs[i]; | |
6120 | if (prog_is_subprog(obj, prog)) | |
6121 | continue; | |
6122 | err = bpf_object__relocate_data(obj, prog); | |
6123 | if (err) { | |
6124 | pr_warn("prog '%s': failed to relocate data references: %d\n", | |
6125 | prog->name, err); | |
6126 | return err; | |
6127 | } | |
6128 | } | |
67234743 AS |
6129 | if (!obj->gen_loader) |
6130 | bpf_object__free_relocs(obj); | |
8a47a6c5 WN |
6131 | return 0; |
6132 | } | |
6133 | ||
646f02ff | 6134 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 6135 | Elf64_Shdr *shdr, Elf_Data *data); |
646f02ff AN |
6136 | |
6137 | static int bpf_object__collect_map_relos(struct bpf_object *obj, | |
ad23b723 | 6138 | Elf64_Shdr *shdr, Elf_Data *data) |
646f02ff | 6139 | { |
15728ad3 AN |
6140 | const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *); |
6141 | int i, j, nrels, new_sz; | |
063e6881 | 6142 | const struct btf_var_secinfo *vi = NULL; |
646f02ff | 6143 | const struct btf_type *sec, *var, *def; |
3168c158 | 6144 | struct bpf_map *map = NULL, *targ_map; |
646f02ff | 6145 | const struct btf_member *member; |
646f02ff | 6146 | const char *name, *mname; |
646f02ff | 6147 | unsigned int moff; |
ad23b723 AN |
6148 | Elf64_Sym *sym; |
6149 | Elf64_Rel *rel; | |
646f02ff AN |
6150 | void *tmp; |
6151 | ||
6152 | if (!obj->efile.btf_maps_sec_btf_id || !obj->btf) | |
6153 | return -EINVAL; | |
6154 | sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id); | |
6155 | if (!sec) | |
6156 | return -EINVAL; | |
6157 | ||
646f02ff AN |
6158 | nrels = shdr->sh_size / shdr->sh_entsize; |
6159 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
6160 | rel = elf_rel_by_idx(data, i); |
6161 | if (!rel) { | |
646f02ff AN |
6162 | pr_warn(".maps relo #%d: failed to get ELF relo\n", i); |
6163 | return -LIBBPF_ERRNO__FORMAT; | |
6164 | } | |
ad23b723 AN |
6165 | |
6166 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); | |
6167 | if (!sym) { | |
646f02ff | 6168 | pr_warn(".maps relo #%d: symbol %zx not found\n", |
ad23b723 | 6169 | i, (size_t)ELF64_R_SYM(rel->r_info)); |
646f02ff AN |
6170 | return -LIBBPF_ERRNO__FORMAT; |
6171 | } | |
ad23b723 AN |
6172 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
6173 | if (sym->st_shndx != obj->efile.btf_maps_shndx) { | |
646f02ff AN |
6174 | pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n", |
6175 | i, name); | |
6176 | return -LIBBPF_ERRNO__RELOC; | |
6177 | } | |
6178 | ||
ad23b723 AN |
6179 | pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n", |
6180 | i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value, | |
6181 | (size_t)rel->r_offset, sym->st_name, name); | |
646f02ff AN |
6182 | |
6183 | for (j = 0; j < obj->nr_maps; j++) { | |
6184 | map = &obj->maps[j]; | |
6185 | if (map->sec_idx != obj->efile.btf_maps_shndx) | |
6186 | continue; | |
6187 | ||
6188 | vi = btf_var_secinfos(sec) + map->btf_var_idx; | |
ad23b723 AN |
6189 | if (vi->offset <= rel->r_offset && |
6190 | rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size) | |
646f02ff AN |
6191 | break; |
6192 | } | |
6193 | if (j == obj->nr_maps) { | |
ad23b723 AN |
6194 | pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n", |
6195 | i, name, (size_t)rel->r_offset); | |
646f02ff AN |
6196 | return -EINVAL; |
6197 | } | |
6198 | ||
6199 | if (!bpf_map_type__is_map_in_map(map->def.type)) | |
6200 | return -EINVAL; | |
6201 | if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS && | |
6202 | map->def.key_size != sizeof(int)) { | |
6203 | pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n", | |
6204 | i, map->name, sizeof(int)); | |
6205 | return -EINVAL; | |
6206 | } | |
6207 | ||
6208 | targ_map = bpf_object__find_map_by_name(obj, name); | |
6209 | if (!targ_map) | |
6210 | return -ESRCH; | |
6211 | ||
6212 | var = btf__type_by_id(obj->btf, vi->type); | |
6213 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
6214 | if (btf_vlen(def) == 0) | |
6215 | return -EINVAL; | |
6216 | member = btf_members(def) + btf_vlen(def) - 1; | |
6217 | mname = btf__name_by_offset(obj->btf, member->name_off); | |
6218 | if (strcmp(mname, "values")) | |
6219 | return -EINVAL; | |
6220 | ||
6221 | moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8; | |
ad23b723 | 6222 | if (rel->r_offset - vi->offset < moff) |
646f02ff AN |
6223 | return -EINVAL; |
6224 | ||
ad23b723 | 6225 | moff = rel->r_offset - vi->offset - moff; |
15728ad3 AN |
6226 | /* here we use BPF pointer size, which is always 64 bit, as we |
6227 | * are parsing ELF that was built for BPF target | |
6228 | */ | |
6229 | if (moff % bpf_ptr_sz) | |
646f02ff | 6230 | return -EINVAL; |
15728ad3 | 6231 | moff /= bpf_ptr_sz; |
646f02ff AN |
6232 | if (moff >= map->init_slots_sz) { |
6233 | new_sz = moff + 1; | |
029258d7 | 6234 | tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz); |
646f02ff AN |
6235 | if (!tmp) |
6236 | return -ENOMEM; | |
6237 | map->init_slots = tmp; | |
6238 | memset(map->init_slots + map->init_slots_sz, 0, | |
15728ad3 | 6239 | (new_sz - map->init_slots_sz) * host_ptr_sz); |
646f02ff AN |
6240 | map->init_slots_sz = new_sz; |
6241 | } | |
6242 | map->init_slots[moff] = targ_map; | |
6243 | ||
6244 | pr_debug(".maps relo #%d: map '%s' slot [%d] points to map '%s'\n", | |
6245 | i, map->name, moff, name); | |
6246 | } | |
6247 | ||
6248 | return 0; | |
6249 | } | |
590a0088 | 6250 | |
c3c55696 | 6251 | static int cmp_relocs(const void *_a, const void *_b) |
34090915 | 6252 | { |
c3c55696 AN |
6253 | const struct reloc_desc *a = _a; |
6254 | const struct reloc_desc *b = _b; | |
34090915 | 6255 | |
c3c55696 AN |
6256 | if (a->insn_idx != b->insn_idx) |
6257 | return a->insn_idx < b->insn_idx ? -1 : 1; | |
6258 | ||
6259 | /* no two relocations should have the same insn_idx, but ... */ | |
6260 | if (a->type != b->type) | |
6261 | return a->type < b->type ? -1 : 1; | |
6262 | ||
6263 | return 0; | |
6264 | } | |
6265 | ||
6266 | static int bpf_object__collect_relos(struct bpf_object *obj) | |
6267 | { | |
6268 | int i, err; | |
34090915 | 6269 | |
25bbbd7a AN |
6270 | for (i = 0; i < obj->efile.sec_cnt; i++) { |
6271 | struct elf_sec_desc *sec_desc = &obj->efile.secs[i]; | |
6272 | Elf64_Shdr *shdr; | |
6273 | Elf_Data *data; | |
6274 | int idx; | |
6275 | ||
6276 | if (sec_desc->sec_type != SEC_RELO) | |
6277 | continue; | |
6278 | ||
6279 | shdr = sec_desc->shdr; | |
6280 | data = sec_desc->data; | |
6281 | idx = shdr->sh_info; | |
34090915 WN |
6282 | |
6283 | if (shdr->sh_type != SHT_REL) { | |
be18010e | 6284 | pr_warn("internal error at %d\n", __LINE__); |
6371ca3b | 6285 | return -LIBBPF_ERRNO__INTERNAL; |
34090915 WN |
6286 | } |
6287 | ||
c3c55696 | 6288 | if (idx == obj->efile.st_ops_shndx) |
646f02ff | 6289 | err = bpf_object__collect_st_ops_relos(obj, shdr, data); |
c3c55696 | 6290 | else if (idx == obj->efile.btf_maps_shndx) |
646f02ff | 6291 | err = bpf_object__collect_map_relos(obj, shdr, data); |
c3c55696 AN |
6292 | else |
6293 | err = bpf_object__collect_prog_relos(obj, shdr, data); | |
34090915 | 6294 | if (err) |
6371ca3b | 6295 | return err; |
34090915 | 6296 | } |
c3c55696 AN |
6297 | |
6298 | for (i = 0; i < obj->nr_programs; i++) { | |
6299 | struct bpf_program *p = &obj->programs[i]; | |
c139e40a | 6300 | |
c3c55696 AN |
6301 | if (!p->nr_reloc) |
6302 | continue; | |
6303 | ||
6304 | qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs); | |
6305 | } | |
34090915 WN |
6306 | return 0; |
6307 | } | |
6308 | ||
109cea5a AN |
6309 | static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id) |
6310 | { | |
9b2f6fec | 6311 | if (BPF_CLASS(insn->code) == BPF_JMP && |
109cea5a AN |
6312 | BPF_OP(insn->code) == BPF_CALL && |
6313 | BPF_SRC(insn->code) == BPF_K && | |
9b2f6fec AN |
6314 | insn->src_reg == 0 && |
6315 | insn->dst_reg == 0) { | |
6316 | *func_id = insn->imm; | |
109cea5a AN |
6317 | return true; |
6318 | } | |
6319 | return false; | |
6320 | } | |
6321 | ||
42869d28 | 6322 | static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog) |
109cea5a AN |
6323 | { |
6324 | struct bpf_insn *insn = prog->insns; | |
6325 | enum bpf_func_id func_id; | |
6326 | int i; | |
6327 | ||
67234743 AS |
6328 | if (obj->gen_loader) |
6329 | return 0; | |
6330 | ||
109cea5a AN |
6331 | for (i = 0; i < prog->insns_cnt; i++, insn++) { |
6332 | if (!insn_is_helper_call(insn, &func_id)) | |
6333 | continue; | |
6334 | ||
6335 | /* on kernels that don't yet support | |
6336 | * bpf_probe_read_{kernel,user}[_str] helpers, fall back | |
6337 | * to bpf_probe_read() which works well for old kernels | |
6338 | */ | |
6339 | switch (func_id) { | |
6340 | case BPF_FUNC_probe_read_kernel: | |
6341 | case BPF_FUNC_probe_read_user: | |
9ca1f56a | 6342 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
6343 | insn->imm = BPF_FUNC_probe_read; |
6344 | break; | |
6345 | case BPF_FUNC_probe_read_kernel_str: | |
6346 | case BPF_FUNC_probe_read_user_str: | |
9ca1f56a | 6347 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
6348 | insn->imm = BPF_FUNC_probe_read_str; |
6349 | break; | |
6350 | default: | |
6351 | break; | |
6352 | } | |
6353 | } | |
6354 | return 0; | |
6355 | } | |
6356 | ||
15ea31fa AN |
6357 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
6358 | int *btf_obj_fd, int *btf_type_id); | |
12d9466d AN |
6359 | |
6360 | /* this is called as prog->sec_def->preload_fn for libbpf-supported sec_defs */ | |
6361 | static int libbpf_preload_prog(struct bpf_program *prog, | |
6362 | struct bpf_prog_load_params *attr, long cookie) | |
6363 | { | |
15ea31fa AN |
6364 | enum sec_def_flags def = cookie; |
6365 | ||
12d9466d | 6366 | /* old kernels might not support specifying expected_attach_type */ |
15ea31fa | 6367 | if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE)) |
12d9466d AN |
6368 | attr->expected_attach_type = 0; |
6369 | ||
15ea31fa | 6370 | if (def & SEC_SLEEPABLE) |
12d9466d AN |
6371 | attr->prog_flags |= BPF_F_SLEEPABLE; |
6372 | ||
6373 | if ((prog->type == BPF_PROG_TYPE_TRACING || | |
6374 | prog->type == BPF_PROG_TYPE_LSM || | |
6375 | prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) { | |
6376 | int btf_obj_fd = 0, btf_type_id = 0, err; | |
15ea31fa | 6377 | const char *attach_name; |
12d9466d | 6378 | |
15ea31fa AN |
6379 | attach_name = strchr(prog->sec_name, '/') + 1; |
6380 | err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id); | |
12d9466d AN |
6381 | if (err) |
6382 | return err; | |
6383 | ||
6384 | /* cache resolved BTF FD and BTF type ID in the prog */ | |
6385 | prog->attach_btf_obj_fd = btf_obj_fd; | |
6386 | prog->attach_btf_id = btf_type_id; | |
6387 | ||
6388 | /* but by now libbpf common logic is not utilizing | |
6389 | * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because | |
6390 | * this callback is called after attrs were populated by | |
6391 | * libbpf, so this callback has to update attr explicitly here | |
6392 | */ | |
6393 | attr->attach_btf_obj_fd = btf_obj_fd; | |
6394 | attr->attach_btf_id = btf_type_id; | |
6395 | } | |
6396 | return 0; | |
6397 | } | |
6398 | ||
55cffde2 | 6399 | static int |
2993e051 | 6400 | load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt, |
f0187f0b | 6401 | char *license, __u32 kern_version, int *pfd) |
55cffde2 | 6402 | { |
6aef10a4 | 6403 | struct bpf_prog_load_params load_attr = {}; |
25bbbd7a | 6404 | struct bpf_object *obj = prog->obj; |
1ce6a9fc | 6405 | char *cp, errmsg[STRERR_BUFSIZE]; |
8395f320 SF |
6406 | size_t log_buf_size = 0; |
6407 | char *log_buf = NULL; | |
12d9466d | 6408 | int btf_fd, ret, err; |
55cffde2 | 6409 | |
80b2b5c3 AM |
6410 | if (prog->type == BPF_PROG_TYPE_UNSPEC) { |
6411 | /* | |
6412 | * The program type must be set. Most likely we couldn't find a proper | |
6413 | * section definition at load time, and thus we didn't infer the type. | |
6414 | */ | |
6415 | pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n", | |
6416 | prog->name, prog->sec_name); | |
6417 | return -EINVAL; | |
6418 | } | |
6419 | ||
fba01a06 AN |
6420 | if (!insns || !insns_cnt) |
6421 | return -EINVAL; | |
6422 | ||
2993e051 | 6423 | load_attr.prog_type = prog->type; |
12d9466d | 6424 | load_attr.expected_attach_type = prog->expected_attach_type; |
25bbbd7a | 6425 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
5b32a23e | 6426 | load_attr.name = prog->name; |
d7be143b | 6427 | load_attr.insns = insns; |
6aef10a4 | 6428 | load_attr.insn_cnt = insns_cnt; |
d7be143b | 6429 | load_attr.license = license; |
6aef10a4 | 6430 | load_attr.attach_btf_id = prog->attach_btf_id; |
12d9466d AN |
6431 | load_attr.attach_prog_fd = prog->attach_prog_fd; |
6432 | load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd; | |
6aef10a4 AN |
6433 | load_attr.attach_btf_id = prog->attach_btf_id; |
6434 | load_attr.kern_version = kern_version; | |
6435 | load_attr.prog_ifindex = prog->prog_ifindex; | |
6436 | ||
0f0e55d8 | 6437 | /* specify func_info/line_info only if kernel supports them */ |
25bbbd7a AN |
6438 | btf_fd = bpf_object__btf_fd(obj); |
6439 | if (btf_fd >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) { | |
0f0e55d8 AN |
6440 | load_attr.prog_btf_fd = btf_fd; |
6441 | load_attr.func_info = prog->func_info; | |
6442 | load_attr.func_info_rec_size = prog->func_info_rec_size; | |
6443 | load_attr.func_info_cnt = prog->func_info_cnt; | |
6444 | load_attr.line_info = prog->line_info; | |
6445 | load_attr.line_info_rec_size = prog->line_info_rec_size; | |
6446 | load_attr.line_info_cnt = prog->line_info_cnt; | |
6447 | } | |
da11b417 | 6448 | load_attr.log_level = prog->log_level; |
04656198 | 6449 | load_attr.prog_flags = prog->prog_flags; |
25bbbd7a | 6450 | load_attr.fd_array = obj->fd_array; |
55cffde2 | 6451 | |
12d9466d AN |
6452 | /* adjust load_attr if sec_def provides custom preload callback */ |
6453 | if (prog->sec_def && prog->sec_def->preload_fn) { | |
6454 | err = prog->sec_def->preload_fn(prog, &load_attr, prog->sec_def->cookie); | |
6455 | if (err < 0) { | |
6456 | pr_warn("prog '%s': failed to prepare load attributes: %d\n", | |
6457 | prog->name, err); | |
6458 | return err; | |
6459 | } | |
6460 | } | |
6461 | ||
25bbbd7a AN |
6462 | if (obj->gen_loader) { |
6463 | bpf_gen__prog_load(obj->gen_loader, &load_attr, | |
6464 | prog - obj->programs); | |
67234743 AS |
6465 | *pfd = -1; |
6466 | return 0; | |
6467 | } | |
da11b417 | 6468 | retry_load: |
8395f320 SF |
6469 | if (log_buf_size) { |
6470 | log_buf = malloc(log_buf_size); | |
6471 | if (!log_buf) | |
6472 | return -ENOMEM; | |
6473 | ||
6474 | *log_buf = 0; | |
6475 | } | |
55cffde2 | 6476 | |
6aef10a4 AN |
6477 | load_attr.log_buf = log_buf; |
6478 | load_attr.log_buf_sz = log_buf_size; | |
6479 | ret = libbpf__bpf_prog_load(&load_attr); | |
55cffde2 WN |
6480 | |
6481 | if (ret >= 0) { | |
8395f320 | 6482 | if (log_buf && load_attr.log_level) |
da11b417 | 6483 | pr_debug("verifier log:\n%s", log_buf); |
5d23328d | 6484 | |
25bbbd7a AN |
6485 | if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) { |
6486 | struct bpf_map *map; | |
6487 | int i; | |
6488 | ||
6489 | for (i = 0; i < obj->nr_maps; i++) { | |
6490 | map = &prog->obj->maps[i]; | |
6491 | if (map->libbpf_type != LIBBPF_MAP_RODATA) | |
6492 | continue; | |
5d23328d | 6493 | |
25bbbd7a AN |
6494 | if (bpf_prog_bind_map(ret, bpf_map__fd(map), NULL)) { |
6495 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); | |
6496 | pr_warn("prog '%s': failed to bind .rodata map: %s\n", | |
6497 | prog->name, cp); | |
6498 | /* Don't fail hard if can't bind rodata. */ | |
6499 | } | |
5d23328d YZ |
6500 | } |
6501 | } | |
6502 | ||
55cffde2 WN |
6503 | *pfd = ret; |
6504 | ret = 0; | |
6505 | goto out; | |
6506 | } | |
6507 | ||
8395f320 SF |
6508 | if (!log_buf || errno == ENOSPC) { |
6509 | log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, | |
6510 | log_buf_size << 1); | |
6511 | ||
da11b417 AS |
6512 | free(log_buf); |
6513 | goto retry_load; | |
6514 | } | |
ef05afa6 | 6515 | ret = errno ? -errno : -LIBBPF_ERRNO__LOAD; |
24d6a808 | 6516 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
be18010e | 6517 | pr_warn("load bpf program failed: %s\n", cp); |
dc3a2d25 | 6518 | pr_perm_msg(ret); |
55cffde2 | 6519 | |
6371ca3b WN |
6520 | if (log_buf && log_buf[0] != '\0') { |
6521 | ret = -LIBBPF_ERRNO__VERIFY; | |
be18010e KW |
6522 | pr_warn("-- BEGIN DUMP LOG ---\n"); |
6523 | pr_warn("\n%s\n", log_buf); | |
6524 | pr_warn("-- END LOG --\n"); | |
6aef10a4 | 6525 | } else if (load_attr.insn_cnt >= BPF_MAXINSNS) { |
be18010e | 6526 | pr_warn("Program too large (%zu insns), at most %d insns\n", |
6aef10a4 | 6527 | load_attr.insn_cnt, BPF_MAXINSNS); |
705fa219 | 6528 | ret = -LIBBPF_ERRNO__PROG2BIG; |
4f33ddb4 | 6529 | } else if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) { |
705fa219 | 6530 | /* Wrong program type? */ |
4f33ddb4 | 6531 | int fd; |
705fa219 | 6532 | |
4f33ddb4 THJ |
6533 | load_attr.prog_type = BPF_PROG_TYPE_KPROBE; |
6534 | load_attr.expected_attach_type = 0; | |
6aef10a4 AN |
6535 | load_attr.log_buf = NULL; |
6536 | load_attr.log_buf_sz = 0; | |
6537 | fd = libbpf__bpf_prog_load(&load_attr); | |
4f33ddb4 THJ |
6538 | if (fd >= 0) { |
6539 | close(fd); | |
6540 | ret = -LIBBPF_ERRNO__PROGTYPE; | |
6541 | goto out; | |
6542 | } | |
55cffde2 WN |
6543 | } |
6544 | ||
6545 | out: | |
6546 | free(log_buf); | |
6547 | return ret; | |
6548 | } | |
6549 | ||
67234743 AS |
6550 | static int bpf_program__record_externs(struct bpf_program *prog) |
6551 | { | |
6552 | struct bpf_object *obj = prog->obj; | |
6553 | int i; | |
6554 | ||
6555 | for (i = 0; i < prog->nr_reloc; i++) { | |
6556 | struct reloc_desc *relo = &prog->reloc_desc[i]; | |
6557 | struct extern_desc *ext = &obj->externs[relo->sym_off]; | |
6558 | ||
6559 | switch (relo->type) { | |
6560 | case RELO_EXTERN_VAR: | |
6561 | if (ext->type != EXT_KSYM) | |
6562 | continue; | |
6563 | if (!ext->ksym.type_id) { | |
6564 | pr_warn("typeless ksym %s is not supported yet\n", | |
6565 | ext->name); | |
6566 | return -ENOTSUP; | |
6567 | } | |
18f4fccb KKD |
6568 | bpf_gen__record_extern(obj->gen_loader, ext->name, ext->is_weak, |
6569 | BTF_KIND_VAR, relo->insn_idx); | |
67234743 AS |
6570 | break; |
6571 | case RELO_EXTERN_FUNC: | |
18f4fccb KKD |
6572 | bpf_gen__record_extern(obj->gen_loader, ext->name, ext->is_weak, |
6573 | BTF_KIND_FUNC, relo->insn_idx); | |
67234743 AS |
6574 | break; |
6575 | default: | |
6576 | continue; | |
6577 | } | |
6578 | } | |
6579 | return 0; | |
6580 | } | |
6581 | ||
13acb508 | 6582 | int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver) |
55cffde2 | 6583 | { |
91abb4a6 | 6584 | int err = 0, fd, i; |
13acb508 | 6585 | |
d9297581 | 6586 | if (prog->obj->loaded) { |
52109584 | 6587 | pr_warn("prog '%s': can't load after object was loaded\n", prog->name); |
e9fc3ce9 | 6588 | return libbpf_err(-EINVAL); |
d9297581 AN |
6589 | } |
6590 | ||
b580563e WN |
6591 | if (prog->instances.nr < 0 || !prog->instances.fds) { |
6592 | if (prog->preprocessor) { | |
be18010e | 6593 | pr_warn("Internal error: can't load program '%s'\n", |
52109584 | 6594 | prog->name); |
e9fc3ce9 | 6595 | return libbpf_err(-LIBBPF_ERRNO__INTERNAL); |
b580563e | 6596 | } |
55cffde2 | 6597 | |
b580563e WN |
6598 | prog->instances.fds = malloc(sizeof(int)); |
6599 | if (!prog->instances.fds) { | |
be18010e | 6600 | pr_warn("Not enough memory for BPF fds\n"); |
e9fc3ce9 | 6601 | return libbpf_err(-ENOMEM); |
b580563e WN |
6602 | } |
6603 | prog->instances.nr = 1; | |
6604 | prog->instances.fds[0] = -1; | |
6605 | } | |
6606 | ||
6607 | if (!prog->preprocessor) { | |
6608 | if (prog->instances.nr != 1) { | |
52109584 AN |
6609 | pr_warn("prog '%s': inconsistent nr(%d) != 1\n", |
6610 | prog->name, prog->instances.nr); | |
b580563e | 6611 | } |
67234743 AS |
6612 | if (prog->obj->gen_loader) |
6613 | bpf_program__record_externs(prog); | |
2993e051 | 6614 | err = load_program(prog, prog->insns, prog->insns_cnt, |
13acb508 | 6615 | license, kern_ver, &fd); |
b580563e WN |
6616 | if (!err) |
6617 | prog->instances.fds[0] = fd; | |
6618 | goto out; | |
6619 | } | |
6620 | ||
6621 | for (i = 0; i < prog->instances.nr; i++) { | |
6622 | struct bpf_prog_prep_result result; | |
6623 | bpf_program_prep_t preprocessor = prog->preprocessor; | |
6624 | ||
1ad9cbb8 | 6625 | memset(&result, 0, sizeof(result)); |
b580563e WN |
6626 | err = preprocessor(prog, i, prog->insns, |
6627 | prog->insns_cnt, &result); | |
6628 | if (err) { | |
be18010e | 6629 | pr_warn("Preprocessing the %dth instance of program '%s' failed\n", |
52109584 | 6630 | i, prog->name); |
b580563e WN |
6631 | goto out; |
6632 | } | |
6633 | ||
6634 | if (!result.new_insn_ptr || !result.new_insn_cnt) { | |
6635 | pr_debug("Skip loading the %dth instance of program '%s'\n", | |
52109584 | 6636 | i, prog->name); |
b580563e WN |
6637 | prog->instances.fds[i] = -1; |
6638 | if (result.pfd) | |
6639 | *result.pfd = -1; | |
6640 | continue; | |
6641 | } | |
6642 | ||
2993e051 | 6643 | err = load_program(prog, result.new_insn_ptr, |
13acb508 | 6644 | result.new_insn_cnt, license, kern_ver, &fd); |
b580563e | 6645 | if (err) { |
be18010e | 6646 | pr_warn("Loading the %dth instance of program '%s' failed\n", |
52109584 | 6647 | i, prog->name); |
b580563e WN |
6648 | goto out; |
6649 | } | |
6650 | ||
6651 | if (result.pfd) | |
6652 | *result.pfd = fd; | |
6653 | prog->instances.fds[i] = fd; | |
6654 | } | |
6655 | out: | |
55cffde2 | 6656 | if (err) |
52109584 | 6657 | pr_warn("failed to load program '%s'\n", prog->name); |
e9fc3ce9 | 6658 | return libbpf_err(err); |
55cffde2 WN |
6659 | } |
6660 | ||
6661 | static int | |
60276f98 | 6662 | bpf_object__load_progs(struct bpf_object *obj, int log_level) |
55cffde2 | 6663 | { |
d9297581 | 6664 | struct bpf_program *prog; |
55cffde2 WN |
6665 | size_t i; |
6666 | int err; | |
6667 | ||
109cea5a AN |
6668 | for (i = 0; i < obj->nr_programs; i++) { |
6669 | prog = &obj->programs[i]; | |
6670 | err = bpf_object__sanitize_prog(obj, prog); | |
6671 | if (err) | |
6672 | return err; | |
6673 | } | |
6674 | ||
55cffde2 | 6675 | for (i = 0; i < obj->nr_programs; i++) { |
d9297581 | 6676 | prog = &obj->programs[i]; |
c3c55696 | 6677 | if (prog_is_subprog(obj, prog)) |
48cca7e4 | 6678 | continue; |
d9297581 | 6679 | if (!prog->load) { |
9c0f8cbd | 6680 | pr_debug("prog '%s': skipped loading\n", prog->name); |
d9297581 AN |
6681 | continue; |
6682 | } | |
6683 | prog->log_level |= log_level; | |
6684 | err = bpf_program__load(prog, obj->license, obj->kern_version); | |
55cffde2 WN |
6685 | if (err) |
6686 | return err; | |
6687 | } | |
67234743 AS |
6688 | if (obj->gen_loader) |
6689 | bpf_object__free_relocs(obj); | |
55cffde2 WN |
6690 | return 0; |
6691 | } | |
6692 | ||
25498a19 AN |
6693 | static const struct bpf_sec_def *find_sec_def(const char *sec_name); |
6694 | ||
91b4d1d1 AN |
6695 | static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts) |
6696 | { | |
6697 | struct bpf_program *prog; | |
12d9466d | 6698 | int err; |
91b4d1d1 AN |
6699 | |
6700 | bpf_object__for_each_program(prog, obj) { | |
6701 | prog->sec_def = find_sec_def(prog->sec_name); | |
6702 | if (!prog->sec_def) { | |
6703 | /* couldn't guess, but user might manually specify */ | |
6704 | pr_debug("prog '%s': unrecognized ELF section name '%s'\n", | |
6705 | prog->name, prog->sec_name); | |
6706 | continue; | |
6707 | } | |
6708 | ||
91b4d1d1 AN |
6709 | bpf_program__set_type(prog, prog->sec_def->prog_type); |
6710 | bpf_program__set_expected_attach_type(prog, prog->sec_def->expected_attach_type); | |
6711 | ||
91b555d7 AN |
6712 | #pragma GCC diagnostic push |
6713 | #pragma GCC diagnostic ignored "-Wdeprecated-declarations" | |
91b4d1d1 AN |
6714 | if (prog->sec_def->prog_type == BPF_PROG_TYPE_TRACING || |
6715 | prog->sec_def->prog_type == BPF_PROG_TYPE_EXT) | |
6716 | prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0); | |
91b555d7 | 6717 | #pragma GCC diagnostic pop |
12d9466d AN |
6718 | |
6719 | /* sec_def can have custom callback which should be called | |
6720 | * after bpf_program is initialized to adjust its properties | |
6721 | */ | |
6722 | if (prog->sec_def->init_fn) { | |
6723 | err = prog->sec_def->init_fn(prog, prog->sec_def->cookie); | |
6724 | if (err < 0) { | |
6725 | pr_warn("prog '%s': failed to initialize: %d\n", | |
6726 | prog->name, err); | |
6727 | return err; | |
6728 | } | |
6729 | } | |
91b4d1d1 AN |
6730 | } |
6731 | ||
6732 | return 0; | |
6733 | } | |
6734 | ||
1a5e3fb1 | 6735 | static struct bpf_object * |
5e61f270 | 6736 | __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz, |
01af3bf0 | 6737 | const struct bpf_object_open_opts *opts) |
1a5e3fb1 | 6738 | { |
1373ff59 | 6739 | const char *obj_name, *kconfig, *btf_tmp_path; |
1a5e3fb1 | 6740 | struct bpf_object *obj; |
291ee02b | 6741 | char tmp_name[64]; |
6371ca3b | 6742 | int err; |
1a5e3fb1 WN |
6743 | |
6744 | if (elf_version(EV_CURRENT) == EV_NONE) { | |
be18010e KW |
6745 | pr_warn("failed to init libelf for %s\n", |
6746 | path ? : "(mem buf)"); | |
6371ca3b | 6747 | return ERR_PTR(-LIBBPF_ERRNO__LIBELF); |
1a5e3fb1 WN |
6748 | } |
6749 | ||
291ee02b AN |
6750 | if (!OPTS_VALID(opts, bpf_object_open_opts)) |
6751 | return ERR_PTR(-EINVAL); | |
6752 | ||
1aace10f | 6753 | obj_name = OPTS_GET(opts, object_name, NULL); |
291ee02b AN |
6754 | if (obj_buf) { |
6755 | if (!obj_name) { | |
6756 | snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx", | |
6757 | (unsigned long)obj_buf, | |
6758 | (unsigned long)obj_buf_sz); | |
6759 | obj_name = tmp_name; | |
6760 | } | |
6761 | path = obj_name; | |
6762 | pr_debug("loading object '%s' from buffer\n", obj_name); | |
6763 | } | |
6764 | ||
2ce8450e | 6765 | obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name); |
6371ca3b WN |
6766 | if (IS_ERR(obj)) |
6767 | return obj; | |
1a5e3fb1 | 6768 | |
1373ff59 SC |
6769 | btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL); |
6770 | if (btf_tmp_path) { | |
6771 | if (strlen(btf_tmp_path) >= PATH_MAX) { | |
6772 | err = -ENAMETOOLONG; | |
6773 | goto out; | |
6774 | } | |
6775 | obj->btf_custom_path = strdup(btf_tmp_path); | |
6776 | if (!obj->btf_custom_path) { | |
6777 | err = -ENOMEM; | |
6778 | goto out; | |
6779 | } | |
6780 | } | |
6781 | ||
8601fd42 AN |
6782 | kconfig = OPTS_GET(opts, kconfig, NULL); |
6783 | if (kconfig) { | |
6784 | obj->kconfig = strdup(kconfig); | |
18353c87 SC |
6785 | if (!obj->kconfig) { |
6786 | err = -ENOMEM; | |
6787 | goto out; | |
6788 | } | |
166750bc | 6789 | } |
291ee02b | 6790 | |
0d13bfce AN |
6791 | err = bpf_object__elf_init(obj); |
6792 | err = err ? : bpf_object__check_endianness(obj); | |
6793 | err = err ? : bpf_object__elf_collect(obj); | |
166750bc AN |
6794 | err = err ? : bpf_object__collect_externs(obj); |
6795 | err = err ? : bpf_object__finalize_btf(obj); | |
0d13bfce | 6796 | err = err ? : bpf_object__init_maps(obj, opts); |
91b4d1d1 | 6797 | err = err ? : bpf_object_init_progs(obj, opts); |
c3c55696 | 6798 | err = err ? : bpf_object__collect_relos(obj); |
0d13bfce AN |
6799 | if (err) |
6800 | goto out; | |
dd4436bb | 6801 | |
91b4d1d1 | 6802 | bpf_object__elf_finish(obj); |
dd4436bb | 6803 | |
1a5e3fb1 WN |
6804 | return obj; |
6805 | out: | |
6806 | bpf_object__close(obj); | |
6371ca3b | 6807 | return ERR_PTR(err); |
1a5e3fb1 WN |
6808 | } |
6809 | ||
5e61f270 AN |
6810 | static struct bpf_object * |
6811 | __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags) | |
1a5e3fb1 | 6812 | { |
e00aca65 | 6813 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts, |
291ee02b AN |
6814 | .relaxed_maps = flags & MAPS_RELAX_COMPAT, |
6815 | ); | |
6816 | ||
1a5e3fb1 | 6817 | /* param validation */ |
07f2d4ea | 6818 | if (!attr->file) |
1a5e3fb1 WN |
6819 | return NULL; |
6820 | ||
07f2d4ea | 6821 | pr_debug("loading %s\n", attr->file); |
291ee02b | 6822 | return __bpf_object__open(attr->file, NULL, 0, &opts); |
c034a177 JF |
6823 | } |
6824 | ||
6825 | struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr) | |
6826 | { | |
e9fc3ce9 | 6827 | return libbpf_ptr(__bpf_object__open_xattr(attr, 0)); |
07f2d4ea JK |
6828 | } |
6829 | ||
6830 | struct bpf_object *bpf_object__open(const char *path) | |
6831 | { | |
6832 | struct bpf_object_open_attr attr = { | |
6833 | .file = path, | |
6834 | .prog_type = BPF_PROG_TYPE_UNSPEC, | |
6835 | }; | |
1a5e3fb1 | 6836 | |
e9fc3ce9 | 6837 | return libbpf_ptr(__bpf_object__open_xattr(&attr, 0)); |
6c956392 WN |
6838 | } |
6839 | ||
2ce8450e | 6840 | struct bpf_object * |
01af3bf0 | 6841 | bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts) |
2ce8450e | 6842 | { |
2ce8450e | 6843 | if (!path) |
e9fc3ce9 | 6844 | return libbpf_err_ptr(-EINVAL); |
2ce8450e AN |
6845 | |
6846 | pr_debug("loading %s\n", path); | |
6847 | ||
e9fc3ce9 | 6848 | return libbpf_ptr(__bpf_object__open(path, NULL, 0, opts)); |
2ce8450e AN |
6849 | } |
6850 | ||
6851 | struct bpf_object * | |
6852 | bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz, | |
01af3bf0 | 6853 | const struct bpf_object_open_opts *opts) |
6c956392 | 6854 | { |
2ce8450e | 6855 | if (!obj_buf || obj_buf_sz == 0) |
e9fc3ce9 | 6856 | return libbpf_err_ptr(-EINVAL); |
6c956392 | 6857 | |
e9fc3ce9 | 6858 | return libbpf_ptr(__bpf_object__open(NULL, obj_buf, obj_buf_sz, opts)); |
2ce8450e AN |
6859 | } |
6860 | ||
6861 | struct bpf_object * | |
6862 | bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz, | |
6863 | const char *name) | |
6864 | { | |
e00aca65 | 6865 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts, |
2ce8450e AN |
6866 | .object_name = name, |
6867 | /* wrong default, but backwards-compatible */ | |
6868 | .relaxed_maps = true, | |
6869 | ); | |
6870 | ||
6871 | /* returning NULL is wrong, but backwards-compatible */ | |
6872 | if (!obj_buf || obj_buf_sz == 0) | |
e9fc3ce9 | 6873 | return errno = EINVAL, NULL; |
6c956392 | 6874 | |
e9fc3ce9 | 6875 | return libbpf_ptr(__bpf_object__open(NULL, obj_buf, obj_buf_sz, &opts)); |
1a5e3fb1 WN |
6876 | } |
6877 | ||
4a404a7e | 6878 | static int bpf_object_unload(struct bpf_object *obj) |
52d3352e WN |
6879 | { |
6880 | size_t i; | |
6881 | ||
6882 | if (!obj) | |
e9fc3ce9 | 6883 | return libbpf_err(-EINVAL); |
52d3352e | 6884 | |
590a0088 | 6885 | for (i = 0; i < obj->nr_maps; i++) { |
9d759a9b | 6886 | zclose(obj->maps[i].fd); |
590a0088 MKL |
6887 | if (obj->maps[i].st_ops) |
6888 | zfree(&obj->maps[i].st_ops->kern_vdata); | |
6889 | } | |
52d3352e | 6890 | |
55cffde2 WN |
6891 | for (i = 0; i < obj->nr_programs; i++) |
6892 | bpf_program__unload(&obj->programs[i]); | |
6893 | ||
52d3352e WN |
6894 | return 0; |
6895 | } | |
6896 | ||
4a404a7e HC |
6897 | int bpf_object__unload(struct bpf_object *obj) __attribute__((alias("bpf_object_unload"))); |
6898 | ||
0d13bfce AN |
6899 | static int bpf_object__sanitize_maps(struct bpf_object *obj) |
6900 | { | |
6901 | struct bpf_map *m; | |
6902 | ||
6903 | bpf_object__for_each_map(m, obj) { | |
6904 | if (!bpf_map__is_internal(m)) | |
6905 | continue; | |
9ca1f56a | 6906 | if (!kernel_supports(obj, FEAT_GLOBAL_DATA)) { |
0d13bfce AN |
6907 | pr_warn("kernel doesn't support global data\n"); |
6908 | return -ENOTSUP; | |
6909 | } | |
9ca1f56a | 6910 | if (!kernel_supports(obj, FEAT_ARRAY_MMAP)) |
0d13bfce AN |
6911 | m->def.map_flags ^= BPF_F_MMAPABLE; |
6912 | } | |
6913 | ||
6914 | return 0; | |
6915 | } | |
6916 | ||
1c0c7074 AN |
6917 | static int bpf_object__read_kallsyms_file(struct bpf_object *obj) |
6918 | { | |
6919 | char sym_type, sym_name[500]; | |
6920 | unsigned long long sym_addr; | |
5bd022ec | 6921 | const struct btf_type *t; |
1c0c7074 AN |
6922 | struct extern_desc *ext; |
6923 | int ret, err = 0; | |
6924 | FILE *f; | |
6925 | ||
6926 | f = fopen("/proc/kallsyms", "r"); | |
6927 | if (!f) { | |
6928 | err = -errno; | |
6929 | pr_warn("failed to open /proc/kallsyms: %d\n", err); | |
6930 | return err; | |
6931 | } | |
6932 | ||
6933 | while (true) { | |
6934 | ret = fscanf(f, "%llx %c %499s%*[^\n]\n", | |
6935 | &sym_addr, &sym_type, sym_name); | |
6936 | if (ret == EOF && feof(f)) | |
6937 | break; | |
6938 | if (ret != 3) { | |
135c783f | 6939 | pr_warn("failed to read kallsyms entry: %d\n", ret); |
1c0c7074 AN |
6940 | err = -EINVAL; |
6941 | goto out; | |
6942 | } | |
6943 | ||
6944 | ext = find_extern_by_name(obj, sym_name); | |
6945 | if (!ext || ext->type != EXT_KSYM) | |
6946 | continue; | |
6947 | ||
5bd022ec MKL |
6948 | t = btf__type_by_id(obj->btf, ext->btf_id); |
6949 | if (!btf_is_var(t)) | |
6950 | continue; | |
6951 | ||
1c0c7074 AN |
6952 | if (ext->is_set && ext->ksym.addr != sym_addr) { |
6953 | pr_warn("extern (ksym) '%s' resolution is ambiguous: 0x%llx or 0x%llx\n", | |
6954 | sym_name, ext->ksym.addr, sym_addr); | |
6955 | err = -EINVAL; | |
6956 | goto out; | |
6957 | } | |
6958 | if (!ext->is_set) { | |
6959 | ext->is_set = true; | |
6960 | ext->ksym.addr = sym_addr; | |
6961 | pr_debug("extern (ksym) %s=0x%llx\n", sym_name, sym_addr); | |
6962 | } | |
6963 | } | |
6964 | ||
6965 | out: | |
6966 | fclose(f); | |
6967 | return err; | |
6968 | } | |
6969 | ||
774e132e MKL |
6970 | static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name, |
6971 | __u16 kind, struct btf **res_btf, | |
9dbe6015 | 6972 | struct module_btf **res_mod_btf) |
d370bbe1 | 6973 | { |
9dbe6015 | 6974 | struct module_btf *mod_btf; |
284d2587 | 6975 | struct btf *btf; |
9dbe6015 | 6976 | int i, id, err; |
d370bbe1 | 6977 | |
933d1aa3 | 6978 | btf = obj->btf_vmlinux; |
9dbe6015 | 6979 | mod_btf = NULL; |
774e132e MKL |
6980 | id = btf__find_by_name_kind(btf, ksym_name, kind); |
6981 | ||
933d1aa3 MKL |
6982 | if (id == -ENOENT) { |
6983 | err = load_module_btfs(obj); | |
6984 | if (err) | |
6985 | return err; | |
d370bbe1 | 6986 | |
933d1aa3 | 6987 | for (i = 0; i < obj->btf_module_cnt; i++) { |
9dbe6015 KKD |
6988 | /* we assume module_btf's BTF FD is always >0 */ |
6989 | mod_btf = &obj->btf_modules[i]; | |
6990 | btf = mod_btf->btf; | |
6991 | id = btf__find_by_name_kind_own(btf, ksym_name, kind); | |
933d1aa3 MKL |
6992 | if (id != -ENOENT) |
6993 | break; | |
6994 | } | |
6995 | } | |
2211c825 | 6996 | if (id <= 0) |
933d1aa3 | 6997 | return -ESRCH; |
d370bbe1 | 6998 | |
774e132e | 6999 | *res_btf = btf; |
9dbe6015 | 7000 | *res_mod_btf = mod_btf; |
774e132e MKL |
7001 | return id; |
7002 | } | |
7003 | ||
7004 | static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj, | |
7005 | struct extern_desc *ext) | |
7006 | { | |
7007 | const struct btf_type *targ_var, *targ_type; | |
7008 | __u32 targ_type_id, local_type_id; | |
9dbe6015 | 7009 | struct module_btf *mod_btf = NULL; |
774e132e | 7010 | const char *targ_var_name; |
774e132e | 7011 | struct btf *btf = NULL; |
9dbe6015 | 7012 | int id, err; |
774e132e | 7013 | |
9dbe6015 | 7014 | id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf); |
466b2e13 KKD |
7015 | if (id < 0) { |
7016 | if (id == -ESRCH && ext->is_weak) | |
7017 | return 0; | |
2211c825 HL |
7018 | pr_warn("extern (var ksym) '%s': not found in kernel BTF\n", |
7019 | ext->name); | |
774e132e | 7020 | return id; |
2211c825 | 7021 | } |
774e132e | 7022 | |
933d1aa3 MKL |
7023 | /* find local type_id */ |
7024 | local_type_id = ext->ksym.type_id; | |
284d2587 | 7025 | |
933d1aa3 MKL |
7026 | /* find target type_id */ |
7027 | targ_var = btf__type_by_id(btf, id); | |
7028 | targ_var_name = btf__name_by_offset(btf, targ_var->name_off); | |
7029 | targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id); | |
d370bbe1 | 7030 | |
933d1aa3 MKL |
7031 | err = bpf_core_types_are_compat(obj->btf, local_type_id, |
7032 | btf, targ_type_id); | |
7033 | if (err <= 0) { | |
7034 | const struct btf_type *local_type; | |
7035 | const char *targ_name, *local_name; | |
d370bbe1 | 7036 | |
933d1aa3 MKL |
7037 | local_type = btf__type_by_id(obj->btf, local_type_id); |
7038 | local_name = btf__name_by_offset(obj->btf, local_type->name_off); | |
7039 | targ_name = btf__name_by_offset(btf, targ_type->name_off); | |
d370bbe1 | 7040 | |
933d1aa3 MKL |
7041 | pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n", |
7042 | ext->name, local_type_id, | |
7043 | btf_kind_str(local_type), local_name, targ_type_id, | |
7044 | btf_kind_str(targ_type), targ_name); | |
7045 | return -EINVAL; | |
7046 | } | |
d370bbe1 | 7047 | |
933d1aa3 | 7048 | ext->is_set = true; |
9dbe6015 | 7049 | ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0; |
933d1aa3 MKL |
7050 | ext->ksym.kernel_btf_id = id; |
7051 | pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n", | |
7052 | ext->name, id, btf_kind_str(targ_var), targ_var_name); | |
d370bbe1 | 7053 | |
933d1aa3 MKL |
7054 | return 0; |
7055 | } | |
d370bbe1 | 7056 | |
5bd022ec MKL |
7057 | static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj, |
7058 | struct extern_desc *ext) | |
7059 | { | |
7060 | int local_func_proto_id, kfunc_proto_id, kfunc_id; | |
9dbe6015 | 7061 | struct module_btf *mod_btf = NULL; |
5bd022ec MKL |
7062 | const struct btf_type *kern_func; |
7063 | struct btf *kern_btf = NULL; | |
9dbe6015 | 7064 | int ret; |
5bd022ec MKL |
7065 | |
7066 | local_func_proto_id = ext->ksym.type_id; | |
7067 | ||
9dbe6015 | 7068 | kfunc_id = find_ksym_btf_id(obj, ext->name, BTF_KIND_FUNC, &kern_btf, &mod_btf); |
5bd022ec | 7069 | if (kfunc_id < 0) { |
466b2e13 KKD |
7070 | if (kfunc_id == -ESRCH && ext->is_weak) |
7071 | return 0; | |
7072 | pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n", | |
5bd022ec MKL |
7073 | ext->name); |
7074 | return kfunc_id; | |
7075 | } | |
7076 | ||
5bd022ec MKL |
7077 | kern_func = btf__type_by_id(kern_btf, kfunc_id); |
7078 | kfunc_proto_id = kern_func->type; | |
7079 | ||
7080 | ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id, | |
7081 | kern_btf, kfunc_proto_id); | |
7082 | if (ret <= 0) { | |
7083 | pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with kernel [%d]\n", | |
7084 | ext->name, local_func_proto_id, kfunc_proto_id); | |
7085 | return -EINVAL; | |
7086 | } | |
7087 | ||
9dbe6015 KKD |
7088 | /* set index for module BTF fd in fd_array, if unset */ |
7089 | if (mod_btf && !mod_btf->fd_array_idx) { | |
7090 | /* insn->off is s16 */ | |
7091 | if (obj->fd_array_cnt == INT16_MAX) { | |
7092 | pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n", | |
7093 | ext->name, mod_btf->fd_array_idx); | |
7094 | return -E2BIG; | |
7095 | } | |
7096 | /* Cannot use index 0 for module BTF fd */ | |
7097 | if (!obj->fd_array_cnt) | |
7098 | obj->fd_array_cnt = 1; | |
7099 | ||
7100 | ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int), | |
7101 | obj->fd_array_cnt + 1); | |
7102 | if (ret) | |
7103 | return ret; | |
7104 | mod_btf->fd_array_idx = obj->fd_array_cnt; | |
7105 | /* we assume module BTF FD is always >0 */ | |
7106 | obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd; | |
7107 | } | |
7108 | ||
5bd022ec | 7109 | ext->is_set = true; |
5bd022ec | 7110 | ext->ksym.kernel_btf_id = kfunc_id; |
9dbe6015 | 7111 | ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0; |
5bd022ec MKL |
7112 | pr_debug("extern (func ksym) '%s': resolved to kernel [%d]\n", |
7113 | ext->name, kfunc_id); | |
7114 | ||
7115 | return 0; | |
7116 | } | |
7117 | ||
933d1aa3 MKL |
7118 | static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj) |
7119 | { | |
5bd022ec | 7120 | const struct btf_type *t; |
933d1aa3 MKL |
7121 | struct extern_desc *ext; |
7122 | int i, err; | |
7123 | ||
7124 | for (i = 0; i < obj->nr_extern; i++) { | |
7125 | ext = &obj->externs[i]; | |
7126 | if (ext->type != EXT_KSYM || !ext->ksym.type_id) | |
7127 | continue; | |
7128 | ||
67234743 AS |
7129 | if (obj->gen_loader) { |
7130 | ext->is_set = true; | |
7131 | ext->ksym.kernel_btf_obj_fd = 0; | |
7132 | ext->ksym.kernel_btf_id = 0; | |
7133 | continue; | |
7134 | } | |
5bd022ec MKL |
7135 | t = btf__type_by_id(obj->btf, ext->btf_id); |
7136 | if (btf_is_var(t)) | |
7137 | err = bpf_object__resolve_ksym_var_btf_id(obj, ext); | |
7138 | else | |
7139 | err = bpf_object__resolve_ksym_func_btf_id(obj, ext); | |
933d1aa3 MKL |
7140 | if (err) |
7141 | return err; | |
d370bbe1 HL |
7142 | } |
7143 | return 0; | |
7144 | } | |
7145 | ||
166750bc | 7146 | static int bpf_object__resolve_externs(struct bpf_object *obj, |
8601fd42 | 7147 | const char *extra_kconfig) |
166750bc | 7148 | { |
1c0c7074 | 7149 | bool need_config = false, need_kallsyms = false; |
d370bbe1 | 7150 | bool need_vmlinux_btf = false; |
166750bc | 7151 | struct extern_desc *ext; |
2e33efe3 | 7152 | void *kcfg_data = NULL; |
166750bc | 7153 | int err, i; |
166750bc AN |
7154 | |
7155 | if (obj->nr_extern == 0) | |
7156 | return 0; | |
7157 | ||
2e33efe3 AN |
7158 | if (obj->kconfig_map_idx >= 0) |
7159 | kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped; | |
166750bc AN |
7160 | |
7161 | for (i = 0; i < obj->nr_extern; i++) { | |
7162 | ext = &obj->externs[i]; | |
7163 | ||
2e33efe3 AN |
7164 | if (ext->type == EXT_KCFG && |
7165 | strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) { | |
7166 | void *ext_val = kcfg_data + ext->kcfg.data_off; | |
166750bc AN |
7167 | __u32 kver = get_kernel_version(); |
7168 | ||
7169 | if (!kver) { | |
7170 | pr_warn("failed to get kernel version\n"); | |
7171 | return -EINVAL; | |
7172 | } | |
2e33efe3 | 7173 | err = set_kcfg_value_num(ext, ext_val, kver); |
166750bc AN |
7174 | if (err) |
7175 | return err; | |
2e33efe3 | 7176 | pr_debug("extern (kcfg) %s=0x%x\n", ext->name, kver); |
13d35a0c | 7177 | } else if (ext->type == EXT_KCFG && str_has_pfx(ext->name, "CONFIG_")) { |
166750bc | 7178 | need_config = true; |
1c0c7074 | 7179 | } else if (ext->type == EXT_KSYM) { |
d370bbe1 HL |
7180 | if (ext->ksym.type_id) |
7181 | need_vmlinux_btf = true; | |
7182 | else | |
7183 | need_kallsyms = true; | |
166750bc AN |
7184 | } else { |
7185 | pr_warn("unrecognized extern '%s'\n", ext->name); | |
7186 | return -EINVAL; | |
7187 | } | |
7188 | } | |
8601fd42 | 7189 | if (need_config && extra_kconfig) { |
2e33efe3 | 7190 | err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data); |
8601fd42 AN |
7191 | if (err) |
7192 | return -EINVAL; | |
7193 | need_config = false; | |
7194 | for (i = 0; i < obj->nr_extern; i++) { | |
7195 | ext = &obj->externs[i]; | |
2e33efe3 | 7196 | if (ext->type == EXT_KCFG && !ext->is_set) { |
8601fd42 AN |
7197 | need_config = true; |
7198 | break; | |
7199 | } | |
7200 | } | |
7201 | } | |
166750bc | 7202 | if (need_config) { |
2e33efe3 | 7203 | err = bpf_object__read_kconfig_file(obj, kcfg_data); |
166750bc AN |
7204 | if (err) |
7205 | return -EINVAL; | |
7206 | } | |
1c0c7074 AN |
7207 | if (need_kallsyms) { |
7208 | err = bpf_object__read_kallsyms_file(obj); | |
7209 | if (err) | |
7210 | return -EINVAL; | |
7211 | } | |
d370bbe1 HL |
7212 | if (need_vmlinux_btf) { |
7213 | err = bpf_object__resolve_ksyms_btf_id(obj); | |
7214 | if (err) | |
7215 | return -EINVAL; | |
7216 | } | |
166750bc AN |
7217 | for (i = 0; i < obj->nr_extern; i++) { |
7218 | ext = &obj->externs[i]; | |
7219 | ||
7220 | if (!ext->is_set && !ext->is_weak) { | |
7221 | pr_warn("extern %s (strong) not resolved\n", ext->name); | |
7222 | return -ESRCH; | |
7223 | } else if (!ext->is_set) { | |
7224 | pr_debug("extern %s (weak) not resolved, defaulting to zero\n", | |
7225 | ext->name); | |
7226 | } | |
7227 | } | |
7228 | ||
7229 | return 0; | |
7230 | } | |
7231 | ||
60276f98 | 7232 | int bpf_object__load_xattr(struct bpf_object_load_attr *attr) |
52d3352e | 7233 | { |
60276f98 | 7234 | struct bpf_object *obj; |
ec6d5f47 | 7235 | int err, i; |
6371ca3b | 7236 | |
60276f98 | 7237 | if (!attr) |
e9fc3ce9 | 7238 | return libbpf_err(-EINVAL); |
60276f98 | 7239 | obj = attr->obj; |
52d3352e | 7240 | if (!obj) |
e9fc3ce9 | 7241 | return libbpf_err(-EINVAL); |
52d3352e WN |
7242 | |
7243 | if (obj->loaded) { | |
d9297581 | 7244 | pr_warn("object '%s': load can't be attempted twice\n", obj->name); |
e9fc3ce9 | 7245 | return libbpf_err(-EINVAL); |
52d3352e WN |
7246 | } |
7247 | ||
67234743 AS |
7248 | if (obj->gen_loader) |
7249 | bpf_gen__init(obj->gen_loader, attr->log_level); | |
7250 | ||
fd9eef1a | 7251 | err = bpf_object__probe_loading(obj); |
fe62de31 | 7252 | err = err ? : bpf_object__load_vmlinux_btf(obj, false); |
8601fd42 | 7253 | err = err ? : bpf_object__resolve_externs(obj, obj->kconfig); |
0d13bfce AN |
7254 | err = err ? : bpf_object__sanitize_and_load_btf(obj); |
7255 | err = err ? : bpf_object__sanitize_maps(obj); | |
590a0088 | 7256 | err = err ? : bpf_object__init_kern_struct_ops_maps(obj); |
0d13bfce | 7257 | err = err ? : bpf_object__create_maps(obj); |
1373ff59 | 7258 | err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : attr->target_btf_path); |
0d13bfce | 7259 | err = err ? : bpf_object__load_progs(obj, attr->log_level); |
a6ed02ca | 7260 | |
67234743 AS |
7261 | if (obj->gen_loader) { |
7262 | /* reset FDs */ | |
7263 | btf__set_fd(obj->btf, -1); | |
7264 | for (i = 0; i < obj->nr_maps; i++) | |
7265 | obj->maps[i].fd = -1; | |
7266 | if (!err) | |
7267 | err = bpf_gen__finish(obj->gen_loader); | |
7268 | } | |
7269 | ||
9dbe6015 KKD |
7270 | /* clean up fd_array */ |
7271 | zfree(&obj->fd_array); | |
7272 | ||
4f33a53d AN |
7273 | /* clean up module BTFs */ |
7274 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
91abb4a6 | 7275 | close(obj->btf_modules[i].fd); |
4f33a53d AN |
7276 | btf__free(obj->btf_modules[i].btf); |
7277 | free(obj->btf_modules[i].name); | |
7278 | } | |
7279 | free(obj->btf_modules); | |
7280 | ||
7281 | /* clean up vmlinux BTF */ | |
a6ed02ca KS |
7282 | btf__free(obj->btf_vmlinux); |
7283 | obj->btf_vmlinux = NULL; | |
7284 | ||
d9297581 AN |
7285 | obj->loaded = true; /* doesn't matter if successfully or not */ |
7286 | ||
0d13bfce AN |
7287 | if (err) |
7288 | goto out; | |
52d3352e WN |
7289 | |
7290 | return 0; | |
7291 | out: | |
ec6d5f47 THJ |
7292 | /* unpin any maps that were auto-pinned during load */ |
7293 | for (i = 0; i < obj->nr_maps; i++) | |
7294 | if (obj->maps[i].pinned && !obj->maps[i].reused) | |
7295 | bpf_map__unpin(&obj->maps[i], NULL); | |
7296 | ||
4a404a7e | 7297 | bpf_object_unload(obj); |
be18010e | 7298 | pr_warn("failed to load object '%s'\n", obj->path); |
e9fc3ce9 | 7299 | return libbpf_err(err); |
52d3352e WN |
7300 | } |
7301 | ||
60276f98 QM |
7302 | int bpf_object__load(struct bpf_object *obj) |
7303 | { | |
7304 | struct bpf_object_load_attr attr = { | |
7305 | .obj = obj, | |
7306 | }; | |
7307 | ||
7308 | return bpf_object__load_xattr(&attr); | |
7309 | } | |
7310 | ||
196f8487 THJ |
7311 | static int make_parent_dir(const char *path) |
7312 | { | |
7313 | char *cp, errmsg[STRERR_BUFSIZE]; | |
7314 | char *dname, *dir; | |
7315 | int err = 0; | |
7316 | ||
7317 | dname = strdup(path); | |
7318 | if (dname == NULL) | |
7319 | return -ENOMEM; | |
7320 | ||
7321 | dir = dirname(dname); | |
7322 | if (mkdir(dir, 0700) && errno != EEXIST) | |
7323 | err = -errno; | |
7324 | ||
7325 | free(dname); | |
7326 | if (err) { | |
7327 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
7328 | pr_warn("failed to mkdir %s: %s\n", path, cp); | |
7329 | } | |
7330 | return err; | |
7331 | } | |
7332 | ||
f367540c JS |
7333 | static int check_path(const char *path) |
7334 | { | |
1ce6a9fc | 7335 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
7336 | struct statfs st_fs; |
7337 | char *dname, *dir; | |
7338 | int err = 0; | |
7339 | ||
7340 | if (path == NULL) | |
7341 | return -EINVAL; | |
7342 | ||
7343 | dname = strdup(path); | |
7344 | if (dname == NULL) | |
7345 | return -ENOMEM; | |
7346 | ||
7347 | dir = dirname(dname); | |
7348 | if (statfs(dir, &st_fs)) { | |
24d6a808 | 7349 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
be18010e | 7350 | pr_warn("failed to statfs %s: %s\n", dir, cp); |
f367540c JS |
7351 | err = -errno; |
7352 | } | |
7353 | free(dname); | |
7354 | ||
7355 | if (!err && st_fs.f_type != BPF_FS_MAGIC) { | |
be18010e | 7356 | pr_warn("specified path %s is not on BPF FS\n", path); |
f367540c JS |
7357 | err = -EINVAL; |
7358 | } | |
7359 | ||
7360 | return err; | |
7361 | } | |
7362 | ||
e21d585c | 7363 | static int bpf_program_pin_instance(struct bpf_program *prog, const char *path, int instance) |
f367540c | 7364 | { |
1ce6a9fc | 7365 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
7366 | int err; |
7367 | ||
196f8487 THJ |
7368 | err = make_parent_dir(path); |
7369 | if (err) | |
e9fc3ce9 | 7370 | return libbpf_err(err); |
196f8487 | 7371 | |
f367540c JS |
7372 | err = check_path(path); |
7373 | if (err) | |
e9fc3ce9 | 7374 | return libbpf_err(err); |
f367540c JS |
7375 | |
7376 | if (prog == NULL) { | |
be18010e | 7377 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7378 | return libbpf_err(-EINVAL); |
f367540c JS |
7379 | } |
7380 | ||
7381 | if (instance < 0 || instance >= prog->instances.nr) { | |
be18010e | 7382 | pr_warn("invalid prog instance %d of prog %s (max %d)\n", |
52109584 | 7383 | instance, prog->name, prog->instances.nr); |
e9fc3ce9 | 7384 | return libbpf_err(-EINVAL); |
f367540c JS |
7385 | } |
7386 | ||
7387 | if (bpf_obj_pin(prog->instances.fds[instance], path)) { | |
23ab656b THJ |
7388 | err = -errno; |
7389 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e | 7390 | pr_warn("failed to pin program: %s\n", cp); |
e9fc3ce9 | 7391 | return libbpf_err(err); |
f367540c JS |
7392 | } |
7393 | pr_debug("pinned program '%s'\n", path); | |
7394 | ||
7395 | return 0; | |
7396 | } | |
7397 | ||
e21d585c | 7398 | static int bpf_program_unpin_instance(struct bpf_program *prog, const char *path, int instance) |
0c19a9fb SF |
7399 | { |
7400 | int err; | |
7401 | ||
7402 | err = check_path(path); | |
7403 | if (err) | |
e9fc3ce9 | 7404 | return libbpf_err(err); |
0c19a9fb SF |
7405 | |
7406 | if (prog == NULL) { | |
be18010e | 7407 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7408 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7409 | } |
7410 | ||
7411 | if (instance < 0 || instance >= prog->instances.nr) { | |
be18010e | 7412 | pr_warn("invalid prog instance %d of prog %s (max %d)\n", |
52109584 | 7413 | instance, prog->name, prog->instances.nr); |
e9fc3ce9 | 7414 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7415 | } |
7416 | ||
7417 | err = unlink(path); | |
7418 | if (err != 0) | |
e9fc3ce9 AN |
7419 | return libbpf_err(-errno); |
7420 | ||
0c19a9fb SF |
7421 | pr_debug("unpinned program '%s'\n", path); |
7422 | ||
7423 | return 0; | |
7424 | } | |
7425 | ||
e21d585c AN |
7426 | __attribute__((alias("bpf_program_pin_instance"))) |
7427 | int bpf_object__pin_instance(struct bpf_program *prog, const char *path, int instance); | |
7428 | ||
7429 | __attribute__((alias("bpf_program_unpin_instance"))) | |
7430 | int bpf_program__unpin_instance(struct bpf_program *prog, const char *path, int instance); | |
7431 | ||
f367540c JS |
7432 | int bpf_program__pin(struct bpf_program *prog, const char *path) |
7433 | { | |
7434 | int i, err; | |
7435 | ||
196f8487 THJ |
7436 | err = make_parent_dir(path); |
7437 | if (err) | |
e9fc3ce9 | 7438 | return libbpf_err(err); |
196f8487 | 7439 | |
f367540c JS |
7440 | err = check_path(path); |
7441 | if (err) | |
e9fc3ce9 | 7442 | return libbpf_err(err); |
f367540c JS |
7443 | |
7444 | if (prog == NULL) { | |
be18010e | 7445 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7446 | return libbpf_err(-EINVAL); |
f367540c JS |
7447 | } |
7448 | ||
7449 | if (prog->instances.nr <= 0) { | |
52109584 | 7450 | pr_warn("no instances of prog %s to pin\n", prog->name); |
e9fc3ce9 | 7451 | return libbpf_err(-EINVAL); |
f367540c JS |
7452 | } |
7453 | ||
fd734c5c SF |
7454 | if (prog->instances.nr == 1) { |
7455 | /* don't create subdirs when pinning single instance */ | |
e21d585c | 7456 | return bpf_program_pin_instance(prog, path, 0); |
fd734c5c SF |
7457 | } |
7458 | ||
0c19a9fb SF |
7459 | for (i = 0; i < prog->instances.nr; i++) { |
7460 | char buf[PATH_MAX]; | |
7461 | int len; | |
7462 | ||
7463 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7464 | if (len < 0) { | |
7465 | err = -EINVAL; | |
7466 | goto err_unpin; | |
7467 | } else if (len >= PATH_MAX) { | |
7468 | err = -ENAMETOOLONG; | |
7469 | goto err_unpin; | |
7470 | } | |
7471 | ||
e21d585c | 7472 | err = bpf_program_pin_instance(prog, buf, i); |
0c19a9fb SF |
7473 | if (err) |
7474 | goto err_unpin; | |
7475 | } | |
7476 | ||
7477 | return 0; | |
7478 | ||
7479 | err_unpin: | |
7480 | for (i = i - 1; i >= 0; i--) { | |
7481 | char buf[PATH_MAX]; | |
7482 | int len; | |
7483 | ||
7484 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7485 | if (len < 0) | |
7486 | continue; | |
7487 | else if (len >= PATH_MAX) | |
7488 | continue; | |
7489 | ||
e21d585c | 7490 | bpf_program_unpin_instance(prog, buf, i); |
0c19a9fb SF |
7491 | } |
7492 | ||
7493 | rmdir(path); | |
7494 | ||
e9fc3ce9 | 7495 | return libbpf_err(err); |
0c19a9fb SF |
7496 | } |
7497 | ||
7498 | int bpf_program__unpin(struct bpf_program *prog, const char *path) | |
7499 | { | |
7500 | int i, err; | |
7501 | ||
7502 | err = check_path(path); | |
7503 | if (err) | |
e9fc3ce9 | 7504 | return libbpf_err(err); |
0c19a9fb SF |
7505 | |
7506 | if (prog == NULL) { | |
be18010e | 7507 | pr_warn("invalid program pointer\n"); |
e9fc3ce9 | 7508 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7509 | } |
7510 | ||
7511 | if (prog->instances.nr <= 0) { | |
52109584 | 7512 | pr_warn("no instances of prog %s to pin\n", prog->name); |
e9fc3ce9 | 7513 | return libbpf_err(-EINVAL); |
fd734c5c SF |
7514 | } |
7515 | ||
7516 | if (prog->instances.nr == 1) { | |
7517 | /* don't create subdirs when pinning single instance */ | |
e21d585c | 7518 | return bpf_program_unpin_instance(prog, path, 0); |
0c19a9fb SF |
7519 | } |
7520 | ||
f367540c JS |
7521 | for (i = 0; i < prog->instances.nr; i++) { |
7522 | char buf[PATH_MAX]; | |
7523 | int len; | |
7524 | ||
7525 | len = snprintf(buf, PATH_MAX, "%s/%d", path, i); | |
7526 | if (len < 0) | |
e9fc3ce9 | 7527 | return libbpf_err(-EINVAL); |
f367540c | 7528 | else if (len >= PATH_MAX) |
e9fc3ce9 | 7529 | return libbpf_err(-ENAMETOOLONG); |
f367540c | 7530 | |
e21d585c | 7531 | err = bpf_program_unpin_instance(prog, buf, i); |
f367540c JS |
7532 | if (err) |
7533 | return err; | |
7534 | } | |
7535 | ||
0c19a9fb SF |
7536 | err = rmdir(path); |
7537 | if (err) | |
e9fc3ce9 | 7538 | return libbpf_err(-errno); |
0c19a9fb | 7539 | |
f367540c JS |
7540 | return 0; |
7541 | } | |
7542 | ||
b6989f35 JS |
7543 | int bpf_map__pin(struct bpf_map *map, const char *path) |
7544 | { | |
1ce6a9fc | 7545 | char *cp, errmsg[STRERR_BUFSIZE]; |
b6989f35 JS |
7546 | int err; |
7547 | ||
b6989f35 | 7548 | if (map == NULL) { |
be18010e | 7549 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 7550 | return libbpf_err(-EINVAL); |
b6989f35 JS |
7551 | } |
7552 | ||
4580b25f THJ |
7553 | if (map->pin_path) { |
7554 | if (path && strcmp(path, map->pin_path)) { | |
7555 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
7556 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 7557 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7558 | } else if (map->pinned) { |
7559 | pr_debug("map '%s' already pinned at '%s'; not re-pinning\n", | |
7560 | bpf_map__name(map), map->pin_path); | |
7561 | return 0; | |
7562 | } | |
7563 | } else { | |
7564 | if (!path) { | |
7565 | pr_warn("missing a path to pin map '%s' at\n", | |
7566 | bpf_map__name(map)); | |
e9fc3ce9 | 7567 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7568 | } else if (map->pinned) { |
7569 | pr_warn("map '%s' already pinned\n", bpf_map__name(map)); | |
e9fc3ce9 | 7570 | return libbpf_err(-EEXIST); |
4580b25f THJ |
7571 | } |
7572 | ||
7573 | map->pin_path = strdup(path); | |
7574 | if (!map->pin_path) { | |
7575 | err = -errno; | |
7576 | goto out_err; | |
7577 | } | |
b6989f35 JS |
7578 | } |
7579 | ||
196f8487 THJ |
7580 | err = make_parent_dir(map->pin_path); |
7581 | if (err) | |
e9fc3ce9 | 7582 | return libbpf_err(err); |
196f8487 | 7583 | |
4580b25f THJ |
7584 | err = check_path(map->pin_path); |
7585 | if (err) | |
e9fc3ce9 | 7586 | return libbpf_err(err); |
4580b25f THJ |
7587 | |
7588 | if (bpf_obj_pin(map->fd, map->pin_path)) { | |
7589 | err = -errno; | |
7590 | goto out_err; | |
7591 | } | |
7592 | ||
7593 | map->pinned = true; | |
7594 | pr_debug("pinned map '%s'\n", map->pin_path); | |
0c19a9fb | 7595 | |
b6989f35 | 7596 | return 0; |
4580b25f THJ |
7597 | |
7598 | out_err: | |
7599 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
7600 | pr_warn("failed to pin map: %s\n", cp); | |
e9fc3ce9 | 7601 | return libbpf_err(err); |
b6989f35 JS |
7602 | } |
7603 | ||
0c19a9fb SF |
7604 | int bpf_map__unpin(struct bpf_map *map, const char *path) |
7605 | { | |
7606 | int err; | |
7607 | ||
0c19a9fb | 7608 | if (map == NULL) { |
be18010e | 7609 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 7610 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
7611 | } |
7612 | ||
4580b25f THJ |
7613 | if (map->pin_path) { |
7614 | if (path && strcmp(path, map->pin_path)) { | |
7615 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
7616 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 7617 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7618 | } |
7619 | path = map->pin_path; | |
7620 | } else if (!path) { | |
7621 | pr_warn("no path to unpin map '%s' from\n", | |
7622 | bpf_map__name(map)); | |
e9fc3ce9 | 7623 | return libbpf_err(-EINVAL); |
4580b25f THJ |
7624 | } |
7625 | ||
7626 | err = check_path(path); | |
7627 | if (err) | |
e9fc3ce9 | 7628 | return libbpf_err(err); |
4580b25f | 7629 | |
0c19a9fb SF |
7630 | err = unlink(path); |
7631 | if (err != 0) | |
e9fc3ce9 | 7632 | return libbpf_err(-errno); |
4580b25f THJ |
7633 | |
7634 | map->pinned = false; | |
7635 | pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path); | |
0c19a9fb SF |
7636 | |
7637 | return 0; | |
7638 | } | |
7639 | ||
4580b25f THJ |
7640 | int bpf_map__set_pin_path(struct bpf_map *map, const char *path) |
7641 | { | |
7642 | char *new = NULL; | |
7643 | ||
7644 | if (path) { | |
7645 | new = strdup(path); | |
7646 | if (!new) | |
e9fc3ce9 | 7647 | return libbpf_err(-errno); |
4580b25f THJ |
7648 | } |
7649 | ||
7650 | free(map->pin_path); | |
7651 | map->pin_path = new; | |
7652 | return 0; | |
7653 | } | |
7654 | ||
7655 | const char *bpf_map__get_pin_path(const struct bpf_map *map) | |
7656 | { | |
7657 | return map->pin_path; | |
7658 | } | |
7659 | ||
e244d34d EL |
7660 | const char *bpf_map__pin_path(const struct bpf_map *map) |
7661 | { | |
7662 | return map->pin_path; | |
7663 | } | |
7664 | ||
4580b25f THJ |
7665 | bool bpf_map__is_pinned(const struct bpf_map *map) |
7666 | { | |
7667 | return map->pinned; | |
7668 | } | |
7669 | ||
9cf309c5 THJ |
7670 | static void sanitize_pin_path(char *s) |
7671 | { | |
7672 | /* bpffs disallows periods in path names */ | |
7673 | while (*s) { | |
7674 | if (*s == '.') | |
7675 | *s = '_'; | |
7676 | s++; | |
7677 | } | |
7678 | } | |
7679 | ||
0c19a9fb | 7680 | int bpf_object__pin_maps(struct bpf_object *obj, const char *path) |
d5148d85 | 7681 | { |
d5148d85 JS |
7682 | struct bpf_map *map; |
7683 | int err; | |
7684 | ||
7685 | if (!obj) | |
e9fc3ce9 | 7686 | return libbpf_err(-ENOENT); |
d5148d85 JS |
7687 | |
7688 | if (!obj->loaded) { | |
be18010e | 7689 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 7690 | return libbpf_err(-ENOENT); |
d5148d85 JS |
7691 | } |
7692 | ||
f74a53d9 | 7693 | bpf_object__for_each_map(map, obj) { |
4580b25f | 7694 | char *pin_path = NULL; |
0c19a9fb | 7695 | char buf[PATH_MAX]; |
0c19a9fb | 7696 | |
4580b25f THJ |
7697 | if (path) { |
7698 | int len; | |
7699 | ||
7700 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
7701 | bpf_map__name(map)); | |
7702 | if (len < 0) { | |
7703 | err = -EINVAL; | |
7704 | goto err_unpin_maps; | |
7705 | } else if (len >= PATH_MAX) { | |
7706 | err = -ENAMETOOLONG; | |
7707 | goto err_unpin_maps; | |
7708 | } | |
9cf309c5 | 7709 | sanitize_pin_path(buf); |
4580b25f THJ |
7710 | pin_path = buf; |
7711 | } else if (!map->pin_path) { | |
7712 | continue; | |
0c19a9fb SF |
7713 | } |
7714 | ||
4580b25f | 7715 | err = bpf_map__pin(map, pin_path); |
0c19a9fb SF |
7716 | if (err) |
7717 | goto err_unpin_maps; | |
7718 | } | |
7719 | ||
7720 | return 0; | |
7721 | ||
7722 | err_unpin_maps: | |
7723 | while ((map = bpf_map__prev(map, obj))) { | |
4580b25f | 7724 | if (!map->pin_path) |
0c19a9fb SF |
7725 | continue; |
7726 | ||
4580b25f | 7727 | bpf_map__unpin(map, NULL); |
0c19a9fb SF |
7728 | } |
7729 | ||
e9fc3ce9 | 7730 | return libbpf_err(err); |
0c19a9fb SF |
7731 | } |
7732 | ||
7733 | int bpf_object__unpin_maps(struct bpf_object *obj, const char *path) | |
7734 | { | |
7735 | struct bpf_map *map; | |
7736 | int err; | |
7737 | ||
7738 | if (!obj) | |
e9fc3ce9 | 7739 | return libbpf_err(-ENOENT); |
0c19a9fb | 7740 | |
f74a53d9 | 7741 | bpf_object__for_each_map(map, obj) { |
4580b25f | 7742 | char *pin_path = NULL; |
d5148d85 | 7743 | char buf[PATH_MAX]; |
d5148d85 | 7744 | |
4580b25f THJ |
7745 | if (path) { |
7746 | int len; | |
7747 | ||
7748 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
7749 | bpf_map__name(map)); | |
7750 | if (len < 0) | |
e9fc3ce9 | 7751 | return libbpf_err(-EINVAL); |
4580b25f | 7752 | else if (len >= PATH_MAX) |
e9fc3ce9 | 7753 | return libbpf_err(-ENAMETOOLONG); |
9cf309c5 | 7754 | sanitize_pin_path(buf); |
4580b25f THJ |
7755 | pin_path = buf; |
7756 | } else if (!map->pin_path) { | |
7757 | continue; | |
7758 | } | |
d5148d85 | 7759 | |
4580b25f | 7760 | err = bpf_map__unpin(map, pin_path); |
d5148d85 | 7761 | if (err) |
e9fc3ce9 | 7762 | return libbpf_err(err); |
d5148d85 JS |
7763 | } |
7764 | ||
0c19a9fb SF |
7765 | return 0; |
7766 | } | |
7767 | ||
7768 | int bpf_object__pin_programs(struct bpf_object *obj, const char *path) | |
7769 | { | |
7770 | struct bpf_program *prog; | |
7771 | int err; | |
7772 | ||
7773 | if (!obj) | |
e9fc3ce9 | 7774 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
7775 | |
7776 | if (!obj->loaded) { | |
be18010e | 7777 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 7778 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
7779 | } |
7780 | ||
0c19a9fb SF |
7781 | bpf_object__for_each_program(prog, obj) { |
7782 | char buf[PATH_MAX]; | |
7783 | int len; | |
7784 | ||
7785 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 7786 | prog->pin_name); |
0c19a9fb SF |
7787 | if (len < 0) { |
7788 | err = -EINVAL; | |
7789 | goto err_unpin_programs; | |
7790 | } else if (len >= PATH_MAX) { | |
7791 | err = -ENAMETOOLONG; | |
7792 | goto err_unpin_programs; | |
7793 | } | |
7794 | ||
7795 | err = bpf_program__pin(prog, buf); | |
7796 | if (err) | |
7797 | goto err_unpin_programs; | |
7798 | } | |
7799 | ||
7800 | return 0; | |
7801 | ||
7802 | err_unpin_programs: | |
7803 | while ((prog = bpf_program__prev(prog, obj))) { | |
7804 | char buf[PATH_MAX]; | |
7805 | int len; | |
7806 | ||
7807 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 7808 | prog->pin_name); |
0c19a9fb SF |
7809 | if (len < 0) |
7810 | continue; | |
7811 | else if (len >= PATH_MAX) | |
7812 | continue; | |
7813 | ||
7814 | bpf_program__unpin(prog, buf); | |
7815 | } | |
7816 | ||
e9fc3ce9 | 7817 | return libbpf_err(err); |
0c19a9fb SF |
7818 | } |
7819 | ||
7820 | int bpf_object__unpin_programs(struct bpf_object *obj, const char *path) | |
7821 | { | |
7822 | struct bpf_program *prog; | |
7823 | int err; | |
7824 | ||
7825 | if (!obj) | |
e9fc3ce9 | 7826 | return libbpf_err(-ENOENT); |
0c19a9fb | 7827 | |
d5148d85 JS |
7828 | bpf_object__for_each_program(prog, obj) { |
7829 | char buf[PATH_MAX]; | |
7830 | int len; | |
7831 | ||
7832 | len = snprintf(buf, PATH_MAX, "%s/%s", path, | |
33a2c75c | 7833 | prog->pin_name); |
d5148d85 | 7834 | if (len < 0) |
e9fc3ce9 | 7835 | return libbpf_err(-EINVAL); |
d5148d85 | 7836 | else if (len >= PATH_MAX) |
e9fc3ce9 | 7837 | return libbpf_err(-ENAMETOOLONG); |
d5148d85 | 7838 | |
0c19a9fb | 7839 | err = bpf_program__unpin(prog, buf); |
d5148d85 | 7840 | if (err) |
e9fc3ce9 | 7841 | return libbpf_err(err); |
d5148d85 JS |
7842 | } |
7843 | ||
7844 | return 0; | |
7845 | } | |
7846 | ||
0c19a9fb SF |
7847 | int bpf_object__pin(struct bpf_object *obj, const char *path) |
7848 | { | |
7849 | int err; | |
7850 | ||
7851 | err = bpf_object__pin_maps(obj, path); | |
7852 | if (err) | |
e9fc3ce9 | 7853 | return libbpf_err(err); |
0c19a9fb SF |
7854 | |
7855 | err = bpf_object__pin_programs(obj, path); | |
7856 | if (err) { | |
7857 | bpf_object__unpin_maps(obj, path); | |
e9fc3ce9 | 7858 | return libbpf_err(err); |
0c19a9fb SF |
7859 | } |
7860 | ||
7861 | return 0; | |
7862 | } | |
7863 | ||
2d39d7c5 AN |
7864 | static void bpf_map__destroy(struct bpf_map *map) |
7865 | { | |
7866 | if (map->clear_priv) | |
7867 | map->clear_priv(map, map->priv); | |
7868 | map->priv = NULL; | |
7869 | map->clear_priv = NULL; | |
7870 | ||
646f02ff AN |
7871 | if (map->inner_map) { |
7872 | bpf_map__destroy(map->inner_map); | |
7873 | zfree(&map->inner_map); | |
7874 | } | |
7875 | ||
7876 | zfree(&map->init_slots); | |
7877 | map->init_slots_sz = 0; | |
7878 | ||
2d39d7c5 AN |
7879 | if (map->mmaped) { |
7880 | munmap(map->mmaped, bpf_map_mmap_sz(map)); | |
7881 | map->mmaped = NULL; | |
7882 | } | |
7883 | ||
7884 | if (map->st_ops) { | |
7885 | zfree(&map->st_ops->data); | |
7886 | zfree(&map->st_ops->progs); | |
7887 | zfree(&map->st_ops->kern_func_off); | |
7888 | zfree(&map->st_ops); | |
7889 | } | |
7890 | ||
7891 | zfree(&map->name); | |
aed65917 | 7892 | zfree(&map->real_name); |
2d39d7c5 AN |
7893 | zfree(&map->pin_path); |
7894 | ||
7895 | if (map->fd >= 0) | |
7896 | zclose(map->fd); | |
7897 | } | |
7898 | ||
1a5e3fb1 WN |
7899 | void bpf_object__close(struct bpf_object *obj) |
7900 | { | |
a5b8bd47 WN |
7901 | size_t i; |
7902 | ||
50450fc7 | 7903 | if (IS_ERR_OR_NULL(obj)) |
1a5e3fb1 WN |
7904 | return; |
7905 | ||
10931d24 WN |
7906 | if (obj->clear_priv) |
7907 | obj->clear_priv(obj, obj->priv); | |
7908 | ||
67234743 | 7909 | bpf_gen__free(obj->gen_loader); |
1a5e3fb1 | 7910 | bpf_object__elf_finish(obj); |
4a404a7e | 7911 | bpf_object_unload(obj); |
8a138aed | 7912 | btf__free(obj->btf); |
2993e051 | 7913 | btf_ext__free(obj->btf_ext); |
1a5e3fb1 | 7914 | |
2d39d7c5 AN |
7915 | for (i = 0; i < obj->nr_maps; i++) |
7916 | bpf_map__destroy(&obj->maps[i]); | |
d859900c | 7917 | |
1373ff59 | 7918 | zfree(&obj->btf_custom_path); |
8601fd42 | 7919 | zfree(&obj->kconfig); |
166750bc AN |
7920 | zfree(&obj->externs); |
7921 | obj->nr_extern = 0; | |
7922 | ||
9d759a9b WN |
7923 | zfree(&obj->maps); |
7924 | obj->nr_maps = 0; | |
a5b8bd47 WN |
7925 | |
7926 | if (obj->programs && obj->nr_programs) { | |
7927 | for (i = 0; i < obj->nr_programs; i++) | |
7928 | bpf_program__exit(&obj->programs[i]); | |
7929 | } | |
7930 | zfree(&obj->programs); | |
7931 | ||
9a208eff | 7932 | list_del(&obj->list); |
1a5e3fb1 WN |
7933 | free(obj); |
7934 | } | |
aa9b1ac3 | 7935 | |
9a208eff WN |
7936 | struct bpf_object * |
7937 | bpf_object__next(struct bpf_object *prev) | |
7938 | { | |
7939 | struct bpf_object *next; | |
689624f0 JB |
7940 | bool strict = (libbpf_mode & LIBBPF_STRICT_NO_OBJECT_LIST); |
7941 | ||
7942 | if (strict) | |
7943 | return NULL; | |
9a208eff WN |
7944 | |
7945 | if (!prev) | |
7946 | next = list_first_entry(&bpf_objects_list, | |
7947 | struct bpf_object, | |
7948 | list); | |
7949 | else | |
7950 | next = list_next_entry(prev, list); | |
7951 | ||
7952 | /* Empty list is noticed here so don't need checking on entry. */ | |
7953 | if (&next->list == &bpf_objects_list) | |
7954 | return NULL; | |
7955 | ||
7956 | return next; | |
7957 | } | |
7958 | ||
a324aae3 | 7959 | const char *bpf_object__name(const struct bpf_object *obj) |
acf860ae | 7960 | { |
e9fc3ce9 | 7961 | return obj ? obj->name : libbpf_err_ptr(-EINVAL); |
acf860ae WN |
7962 | } |
7963 | ||
a324aae3 | 7964 | unsigned int bpf_object__kversion(const struct bpf_object *obj) |
45825d8a | 7965 | { |
a7fe0450 | 7966 | return obj ? obj->kern_version : 0; |
45825d8a WN |
7967 | } |
7968 | ||
a324aae3 | 7969 | struct btf *bpf_object__btf(const struct bpf_object *obj) |
789f6bab AI |
7970 | { |
7971 | return obj ? obj->btf : NULL; | |
7972 | } | |
7973 | ||
8a138aed MKL |
7974 | int bpf_object__btf_fd(const struct bpf_object *obj) |
7975 | { | |
7976 | return obj->btf ? btf__fd(obj->btf) : -1; | |
7977 | } | |
7978 | ||
155f556d RDT |
7979 | int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version) |
7980 | { | |
7981 | if (obj->loaded) | |
e9fc3ce9 | 7982 | return libbpf_err(-EINVAL); |
155f556d RDT |
7983 | |
7984 | obj->kern_version = kern_version; | |
7985 | ||
7986 | return 0; | |
7987 | } | |
7988 | ||
10931d24 WN |
7989 | int bpf_object__set_priv(struct bpf_object *obj, void *priv, |
7990 | bpf_object_clear_priv_t clear_priv) | |
7991 | { | |
7992 | if (obj->priv && obj->clear_priv) | |
7993 | obj->clear_priv(obj, obj->priv); | |
7994 | ||
7995 | obj->priv = priv; | |
7996 | obj->clear_priv = clear_priv; | |
7997 | return 0; | |
7998 | } | |
7999 | ||
a324aae3 | 8000 | void *bpf_object__priv(const struct bpf_object *obj) |
10931d24 | 8001 | { |
e9fc3ce9 | 8002 | return obj ? obj->priv : libbpf_err_ptr(-EINVAL); |
10931d24 WN |
8003 | } |
8004 | ||
67234743 AS |
8005 | int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts) |
8006 | { | |
8007 | struct bpf_gen *gen; | |
8008 | ||
8009 | if (!opts) | |
8010 | return -EFAULT; | |
8011 | if (!OPTS_VALID(opts, gen_loader_opts)) | |
8012 | return -EINVAL; | |
8013 | gen = calloc(sizeof(*gen), 1); | |
8014 | if (!gen) | |
8015 | return -ENOMEM; | |
8016 | gen->opts = opts; | |
8017 | obj->gen_loader = gen; | |
8018 | return 0; | |
8019 | } | |
8020 | ||
eac7d845 | 8021 | static struct bpf_program * |
a324aae3 AN |
8022 | __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj, |
8023 | bool forward) | |
aa9b1ac3 | 8024 | { |
a83d6e76 | 8025 | size_t nr_programs = obj->nr_programs; |
0c19a9fb | 8026 | ssize_t idx; |
aa9b1ac3 | 8027 | |
a83d6e76 | 8028 | if (!nr_programs) |
aa9b1ac3 | 8029 | return NULL; |
aa9b1ac3 | 8030 | |
a83d6e76 MKL |
8031 | if (!p) |
8032 | /* Iter from the beginning */ | |
8033 | return forward ? &obj->programs[0] : | |
8034 | &obj->programs[nr_programs - 1]; | |
8035 | ||
0c19a9fb | 8036 | if (p->obj != obj) { |
be18010e | 8037 | pr_warn("error: program handler doesn't match object\n"); |
e9fc3ce9 | 8038 | return errno = EINVAL, NULL; |
aa9b1ac3 WN |
8039 | } |
8040 | ||
a83d6e76 | 8041 | idx = (p - obj->programs) + (forward ? 1 : -1); |
0c19a9fb | 8042 | if (idx >= obj->nr_programs || idx < 0) |
aa9b1ac3 WN |
8043 | return NULL; |
8044 | return &obj->programs[idx]; | |
8045 | } | |
8046 | ||
eac7d845 | 8047 | struct bpf_program * |
a324aae3 | 8048 | bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj) |
2088a3a7 HC |
8049 | { |
8050 | return bpf_object__next_program(obj, prev); | |
8051 | } | |
8052 | ||
8053 | struct bpf_program * | |
8054 | bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev) | |
eac7d845 JK |
8055 | { |
8056 | struct bpf_program *prog = prev; | |
8057 | ||
8058 | do { | |
a83d6e76 | 8059 | prog = __bpf_program__iter(prog, obj, true); |
c3c55696 | 8060 | } while (prog && prog_is_subprog(obj, prog)); |
0c19a9fb SF |
8061 | |
8062 | return prog; | |
8063 | } | |
8064 | ||
8065 | struct bpf_program * | |
a324aae3 | 8066 | bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj) |
2088a3a7 HC |
8067 | { |
8068 | return bpf_object__prev_program(obj, next); | |
8069 | } | |
8070 | ||
8071 | struct bpf_program * | |
8072 | bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next) | |
0c19a9fb SF |
8073 | { |
8074 | struct bpf_program *prog = next; | |
8075 | ||
0c19a9fb | 8076 | do { |
a83d6e76 | 8077 | prog = __bpf_program__iter(prog, obj, false); |
c3c55696 | 8078 | } while (prog && prog_is_subprog(obj, prog)); |
eac7d845 JK |
8079 | |
8080 | return prog; | |
8081 | } | |
8082 | ||
edb13ed4 ACM |
8083 | int bpf_program__set_priv(struct bpf_program *prog, void *priv, |
8084 | bpf_program_clear_priv_t clear_priv) | |
aa9b1ac3 WN |
8085 | { |
8086 | if (prog->priv && prog->clear_priv) | |
8087 | prog->clear_priv(prog, prog->priv); | |
8088 | ||
8089 | prog->priv = priv; | |
8090 | prog->clear_priv = clear_priv; | |
8091 | return 0; | |
8092 | } | |
8093 | ||
a324aae3 | 8094 | void *bpf_program__priv(const struct bpf_program *prog) |
aa9b1ac3 | 8095 | { |
e9fc3ce9 | 8096 | return prog ? prog->priv : libbpf_err_ptr(-EINVAL); |
aa9b1ac3 WN |
8097 | } |
8098 | ||
9aba3613 JK |
8099 | void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex) |
8100 | { | |
8101 | prog->prog_ifindex = ifindex; | |
8102 | } | |
8103 | ||
01af3bf0 AN |
8104 | const char *bpf_program__name(const struct bpf_program *prog) |
8105 | { | |
8106 | return prog->name; | |
8107 | } | |
8108 | ||
52109584 AN |
8109 | const char *bpf_program__section_name(const struct bpf_program *prog) |
8110 | { | |
8111 | return prog->sec_name; | |
8112 | } | |
8113 | ||
a324aae3 | 8114 | const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy) |
aa9b1ac3 WN |
8115 | { |
8116 | const char *title; | |
8117 | ||
52109584 | 8118 | title = prog->sec_name; |
715f8db9 | 8119 | if (needs_copy) { |
aa9b1ac3 WN |
8120 | title = strdup(title); |
8121 | if (!title) { | |
be18010e | 8122 | pr_warn("failed to strdup program title\n"); |
e9fc3ce9 | 8123 | return libbpf_err_ptr(-ENOMEM); |
aa9b1ac3 WN |
8124 | } |
8125 | } | |
8126 | ||
8127 | return title; | |
8128 | } | |
8129 | ||
d9297581 AN |
8130 | bool bpf_program__autoload(const struct bpf_program *prog) |
8131 | { | |
8132 | return prog->load; | |
8133 | } | |
8134 | ||
8135 | int bpf_program__set_autoload(struct bpf_program *prog, bool autoload) | |
8136 | { | |
8137 | if (prog->obj->loaded) | |
e9fc3ce9 | 8138 | return libbpf_err(-EINVAL); |
d9297581 AN |
8139 | |
8140 | prog->load = autoload; | |
8141 | return 0; | |
8142 | } | |
8143 | ||
a324aae3 | 8144 | int bpf_program__fd(const struct bpf_program *prog) |
aa9b1ac3 | 8145 | { |
b580563e WN |
8146 | return bpf_program__nth_fd(prog, 0); |
8147 | } | |
8148 | ||
1a734efe THJ |
8149 | size_t bpf_program__size(const struct bpf_program *prog) |
8150 | { | |
9c0f8cbd | 8151 | return prog->insns_cnt * BPF_INSN_SZ; |
1a734efe THJ |
8152 | } |
8153 | ||
65a7fa2e AN |
8154 | const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog) |
8155 | { | |
8156 | return prog->insns; | |
8157 | } | |
8158 | ||
8159 | size_t bpf_program__insn_cnt(const struct bpf_program *prog) | |
8160 | { | |
8161 | return prog->insns_cnt; | |
8162 | } | |
8163 | ||
b580563e WN |
8164 | int bpf_program__set_prep(struct bpf_program *prog, int nr_instances, |
8165 | bpf_program_prep_t prep) | |
8166 | { | |
8167 | int *instances_fds; | |
8168 | ||
8169 | if (nr_instances <= 0 || !prep) | |
e9fc3ce9 | 8170 | return libbpf_err(-EINVAL); |
b580563e WN |
8171 | |
8172 | if (prog->instances.nr > 0 || prog->instances.fds) { | |
be18010e | 8173 | pr_warn("Can't set pre-processor after loading\n"); |
e9fc3ce9 | 8174 | return libbpf_err(-EINVAL); |
b580563e WN |
8175 | } |
8176 | ||
8177 | instances_fds = malloc(sizeof(int) * nr_instances); | |
8178 | if (!instances_fds) { | |
be18010e | 8179 | pr_warn("alloc memory failed for fds\n"); |
e9fc3ce9 | 8180 | return libbpf_err(-ENOMEM); |
b580563e WN |
8181 | } |
8182 | ||
8183 | /* fill all fd with -1 */ | |
8184 | memset(instances_fds, -1, sizeof(int) * nr_instances); | |
8185 | ||
8186 | prog->instances.nr = nr_instances; | |
8187 | prog->instances.fds = instances_fds; | |
8188 | prog->preprocessor = prep; | |
8189 | return 0; | |
8190 | } | |
8191 | ||
a324aae3 | 8192 | int bpf_program__nth_fd(const struct bpf_program *prog, int n) |
b580563e WN |
8193 | { |
8194 | int fd; | |
8195 | ||
1e960043 | 8196 | if (!prog) |
e9fc3ce9 | 8197 | return libbpf_err(-EINVAL); |
1e960043 | 8198 | |
b580563e | 8199 | if (n >= prog->instances.nr || n < 0) { |
be18010e | 8200 | pr_warn("Can't get the %dth fd from program %s: only %d instances\n", |
52109584 | 8201 | n, prog->name, prog->instances.nr); |
e9fc3ce9 | 8202 | return libbpf_err(-EINVAL); |
b580563e WN |
8203 | } |
8204 | ||
8205 | fd = prog->instances.fds[n]; | |
8206 | if (fd < 0) { | |
be18010e | 8207 | pr_warn("%dth instance of program '%s' is invalid\n", |
52109584 | 8208 | n, prog->name); |
e9fc3ce9 | 8209 | return libbpf_err(-ENOENT); |
b580563e WN |
8210 | } |
8211 | ||
8212 | return fd; | |
aa9b1ac3 | 8213 | } |
9d759a9b | 8214 | |
a46410d5 | 8215 | enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog) |
f1eead9e AN |
8216 | { |
8217 | return prog->type; | |
8218 | } | |
8219 | ||
dd26b7f5 | 8220 | void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type) |
5f44e4c8 WN |
8221 | { |
8222 | prog->type = type; | |
8223 | } | |
8224 | ||
a324aae3 | 8225 | static bool bpf_program__is_type(const struct bpf_program *prog, |
5f44e4c8 WN |
8226 | enum bpf_prog_type type) |
8227 | { | |
8228 | return prog ? (prog->type == type) : false; | |
8229 | } | |
8230 | ||
a324aae3 AN |
8231 | #define BPF_PROG_TYPE_FNS(NAME, TYPE) \ |
8232 | int bpf_program__set_##NAME(struct bpf_program *prog) \ | |
8233 | { \ | |
8234 | if (!prog) \ | |
e9fc3ce9 | 8235 | return libbpf_err(-EINVAL); \ |
a324aae3 AN |
8236 | bpf_program__set_type(prog, TYPE); \ |
8237 | return 0; \ | |
8238 | } \ | |
8239 | \ | |
8240 | bool bpf_program__is_##NAME(const struct bpf_program *prog) \ | |
8241 | { \ | |
8242 | return bpf_program__is_type(prog, TYPE); \ | |
8243 | } \ | |
ed794073 | 8244 | |
7803ba73 | 8245 | BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER); |
1e092a03 | 8246 | BPF_PROG_TYPE_FNS(lsm, BPF_PROG_TYPE_LSM); |
ed794073 | 8247 | BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE); |
7803ba73 JS |
8248 | BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS); |
8249 | BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT); | |
ed794073 | 8250 | BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT); |
e14c93fd | 8251 | BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT); |
7803ba73 JS |
8252 | BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP); |
8253 | BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT); | |
12a8654b | 8254 | BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING); |
590a0088 | 8255 | BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS); |
2db6eab1 | 8256 | BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT); |
499dd29d | 8257 | BPF_PROG_TYPE_FNS(sk_lookup, BPF_PROG_TYPE_SK_LOOKUP); |
5f44e4c8 | 8258 | |
f1eead9e | 8259 | enum bpf_attach_type |
a46410d5 | 8260 | bpf_program__get_expected_attach_type(const struct bpf_program *prog) |
f1eead9e AN |
8261 | { |
8262 | return prog->expected_attach_type; | |
8263 | } | |
8264 | ||
16962b24 JF |
8265 | void bpf_program__set_expected_attach_type(struct bpf_program *prog, |
8266 | enum bpf_attach_type type) | |
d7be143b AI |
8267 | { |
8268 | prog->expected_attach_type = type; | |
8269 | } | |
8270 | ||
15ea31fa | 8271 | #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \ |
d7a18ea7 | 8272 | .sec = sec_pfx, \ |
d7a18ea7 | 8273 | .prog_type = BPF_PROG_TYPE_##ptype, \ |
15ea31fa AN |
8274 | .expected_attach_type = atype, \ |
8275 | .cookie = (long)(flags), \ | |
12d9466d | 8276 | .preload_fn = libbpf_preload_prog, \ |
d7a18ea7 AN |
8277 | __VA_ARGS__ \ |
8278 | } | |
8279 | ||
12d9466d AN |
8280 | static struct bpf_link *attach_kprobe(const struct bpf_program *prog, long cookie); |
8281 | static struct bpf_link *attach_tp(const struct bpf_program *prog, long cookie); | |
8282 | static struct bpf_link *attach_raw_tp(const struct bpf_program *prog, long cookie); | |
8283 | static struct bpf_link *attach_trace(const struct bpf_program *prog, long cookie); | |
8284 | static struct bpf_link *attach_lsm(const struct bpf_program *prog, long cookie); | |
8285 | static struct bpf_link *attach_iter(const struct bpf_program *prog, long cookie); | |
d7a18ea7 | 8286 | |
d7a18ea7 | 8287 | static const struct bpf_sec_def section_defs[] = { |
dd94d45c AN |
8288 | SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE | SEC_SLOPPY_PFX), |
8289 | SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8290 | SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
15ea31fa AN |
8291 | SEC_DEF("kprobe/", KPROBE, 0, SEC_NONE, attach_kprobe), |
8292 | SEC_DEF("uprobe/", KPROBE, 0, SEC_NONE), | |
8293 | SEC_DEF("kretprobe/", KPROBE, 0, SEC_NONE, attach_kprobe), | |
8294 | SEC_DEF("uretprobe/", KPROBE, 0, SEC_NONE), | |
15ea31fa | 8295 | SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE), |
dd94d45c AN |
8296 | SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE | SEC_SLOPPY_PFX), |
8297 | SEC_DEF("action", SCHED_ACT, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
15ea31fa AN |
8298 | SEC_DEF("tracepoint/", TRACEPOINT, 0, SEC_NONE, attach_tp), |
8299 | SEC_DEF("tp/", TRACEPOINT, 0, SEC_NONE, attach_tp), | |
8300 | SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
8301 | SEC_DEF("raw_tp/", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
ccaf12d6 HT |
8302 | SEC_DEF("raw_tracepoint.w/", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), |
8303 | SEC_DEF("raw_tp.w/", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), | |
15ea31fa AN |
8304 | SEC_DEF("tp_btf/", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace), |
8305 | SEC_DEF("fentry/", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace), | |
8306 | SEC_DEF("fmod_ret/", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace), | |
8307 | SEC_DEF("fexit/", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace), | |
8308 | SEC_DEF("fentry.s/", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8309 | SEC_DEF("fmod_ret.s/", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8310 | SEC_DEF("fexit.s/", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8311 | SEC_DEF("freplace/", EXT, 0, SEC_ATTACH_BTF, attach_trace), | |
8312 | SEC_DEF("lsm/", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm), | |
8313 | SEC_DEF("lsm.s/", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm), | |
8314 | SEC_DEF("iter/", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter), | |
8315 | SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE), | |
d41ea045 AN |
8316 | SEC_DEF("xdp_devmap/", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE), |
8317 | SEC_DEF("xdp_cpumap/", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE), | |
dd94d45c AN |
8318 | SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), |
8319 | SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8320 | SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8321 | SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8322 | SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8323 | SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8324 | SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8325 | SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8326 | SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8327 | SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8328 | SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8329 | SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8330 | SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8331 | SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8332 | SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8333 | SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8334 | SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8335 | SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8336 | SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE | SEC_SLOPPY_PFX), | |
8337 | SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8338 | SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8339 | SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), | |
8340 | SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8341 | SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8342 | SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8343 | SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8344 | SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8345 | SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8346 | SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8347 | SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8348 | SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8349 | SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8350 | SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8351 | SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8352 | SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8353 | SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8354 | SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), | |
8355 | SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), | |
7c80c87a | 8356 | SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX), |
583c9009 | 8357 | }; |
d7be143b | 8358 | |
c76e4c22 TS |
8359 | #define MAX_TYPE_NAME_SIZE 32 |
8360 | ||
d7a18ea7 AN |
8361 | static const struct bpf_sec_def *find_sec_def(const char *sec_name) |
8362 | { | |
dd94d45c AN |
8363 | const struct bpf_sec_def *sec_def; |
8364 | enum sec_def_flags sec_flags; | |
8365 | int i, n = ARRAY_SIZE(section_defs), len; | |
8366 | bool strict = libbpf_mode & LIBBPF_STRICT_SEC_NAME; | |
d7a18ea7 AN |
8367 | |
8368 | for (i = 0; i < n; i++) { | |
dd94d45c AN |
8369 | sec_def = §ion_defs[i]; |
8370 | sec_flags = sec_def->cookie; | |
8371 | len = strlen(sec_def->sec); | |
8372 | ||
8373 | /* "type/" always has to have proper SEC("type/extras") form */ | |
8374 | if (sec_def->sec[len - 1] == '/') { | |
8375 | if (str_has_pfx(sec_name, sec_def->sec)) | |
8376 | return sec_def; | |
8377 | continue; | |
8378 | } | |
8379 | ||
8380 | /* "type+" means it can be either exact SEC("type") or | |
8381 | * well-formed SEC("type/extras") with proper '/' separator | |
8382 | */ | |
8383 | if (sec_def->sec[len - 1] == '+') { | |
8384 | len--; | |
8385 | /* not even a prefix */ | |
8386 | if (strncmp(sec_name, sec_def->sec, len) != 0) | |
8387 | continue; | |
8388 | /* exact match or has '/' separator */ | |
8389 | if (sec_name[len] == '\0' || sec_name[len] == '/') | |
8390 | return sec_def; | |
8391 | continue; | |
8392 | } | |
8393 | ||
8394 | /* SEC_SLOPPY_PFX definitions are allowed to be just prefix | |
8395 | * matches, unless strict section name mode | |
8396 | * (LIBBPF_STRICT_SEC_NAME) is enabled, in which case the | |
8397 | * match has to be exact. | |
8398 | */ | |
8399 | if ((sec_flags & SEC_SLOPPY_PFX) && !strict) { | |
8400 | if (str_has_pfx(sec_name, sec_def->sec)) | |
8401 | return sec_def; | |
8402 | continue; | |
8403 | } | |
8404 | ||
8405 | /* Definitions not marked SEC_SLOPPY_PFX (e.g., | |
8406 | * SEC("syscall")) are exact matches in both modes. | |
8407 | */ | |
8408 | if (strcmp(sec_name, sec_def->sec) == 0) | |
8409 | return sec_def; | |
d7a18ea7 AN |
8410 | } |
8411 | return NULL; | |
8412 | } | |
8413 | ||
c76e4c22 TS |
8414 | static char *libbpf_get_type_names(bool attach_type) |
8415 | { | |
d7a18ea7 | 8416 | int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE; |
c76e4c22 TS |
8417 | char *buf; |
8418 | ||
8419 | buf = malloc(len); | |
8420 | if (!buf) | |
8421 | return NULL; | |
8422 | ||
8423 | buf[0] = '\0'; | |
8424 | /* Forge string buf with all available names */ | |
d7a18ea7 | 8425 | for (i = 0; i < ARRAY_SIZE(section_defs); i++) { |
15ea31fa AN |
8426 | const struct bpf_sec_def *sec_def = §ion_defs[i]; |
8427 | ||
8428 | if (attach_type) { | |
8429 | if (sec_def->preload_fn != libbpf_preload_prog) | |
8430 | continue; | |
8431 | ||
8432 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
8433 | continue; | |
8434 | } | |
c76e4c22 | 8435 | |
d7a18ea7 | 8436 | if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) { |
c76e4c22 TS |
8437 | free(buf); |
8438 | return NULL; | |
8439 | } | |
8440 | strcat(buf, " "); | |
d7a18ea7 | 8441 | strcat(buf, section_defs[i].sec); |
c76e4c22 TS |
8442 | } |
8443 | ||
8444 | return buf; | |
8445 | } | |
8446 | ||
b60df2a0 JK |
8447 | int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type, |
8448 | enum bpf_attach_type *expected_attach_type) | |
583c9009 | 8449 | { |
d7a18ea7 | 8450 | const struct bpf_sec_def *sec_def; |
c76e4c22 | 8451 | char *type_names; |
583c9009 | 8452 | |
b60df2a0 | 8453 | if (!name) |
e9fc3ce9 | 8454 | return libbpf_err(-EINVAL); |
583c9009 | 8455 | |
d7a18ea7 AN |
8456 | sec_def = find_sec_def(name); |
8457 | if (sec_def) { | |
8458 | *prog_type = sec_def->prog_type; | |
8459 | *expected_attach_type = sec_def->expected_attach_type; | |
b60df2a0 JK |
8460 | return 0; |
8461 | } | |
d7a18ea7 | 8462 | |
4a3d6c6a | 8463 | pr_debug("failed to guess program type from ELF section '%s'\n", name); |
c76e4c22 TS |
8464 | type_names = libbpf_get_type_names(false); |
8465 | if (type_names != NULL) { | |
3f519353 | 8466 | pr_debug("supported section(type) names are:%s\n", type_names); |
c76e4c22 TS |
8467 | free(type_names); |
8468 | } | |
8469 | ||
e9fc3ce9 | 8470 | return libbpf_err(-ESRCH); |
b60df2a0 | 8471 | } |
583c9009 | 8472 | |
590a0088 MKL |
8473 | static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj, |
8474 | size_t offset) | |
8475 | { | |
8476 | struct bpf_map *map; | |
8477 | size_t i; | |
8478 | ||
8479 | for (i = 0; i < obj->nr_maps; i++) { | |
8480 | map = &obj->maps[i]; | |
8481 | if (!bpf_map__is_struct_ops(map)) | |
8482 | continue; | |
8483 | if (map->sec_offset <= offset && | |
8484 | offset - map->sec_offset < map->def.value_size) | |
8485 | return map; | |
8486 | } | |
8487 | ||
8488 | return NULL; | |
8489 | } | |
8490 | ||
8491 | /* Collect the reloc from ELF and populate the st_ops->progs[] */ | |
646f02ff | 8492 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 8493 | Elf64_Shdr *shdr, Elf_Data *data) |
590a0088 MKL |
8494 | { |
8495 | const struct btf_member *member; | |
8496 | struct bpf_struct_ops *st_ops; | |
8497 | struct bpf_program *prog; | |
8498 | unsigned int shdr_idx; | |
8499 | const struct btf *btf; | |
8500 | struct bpf_map *map; | |
7e06aad5 | 8501 | unsigned int moff, insn_idx; |
590a0088 | 8502 | const char *name; |
1d1a3bcf | 8503 | __u32 member_idx; |
ad23b723 AN |
8504 | Elf64_Sym *sym; |
8505 | Elf64_Rel *rel; | |
590a0088 MKL |
8506 | int i, nrels; |
8507 | ||
590a0088 MKL |
8508 | btf = obj->btf; |
8509 | nrels = shdr->sh_size / shdr->sh_entsize; | |
8510 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
8511 | rel = elf_rel_by_idx(data, i); |
8512 | if (!rel) { | |
590a0088 MKL |
8513 | pr_warn("struct_ops reloc: failed to get %d reloc\n", i); |
8514 | return -LIBBPF_ERRNO__FORMAT; | |
8515 | } | |
8516 | ||
ad23b723 AN |
8517 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); |
8518 | if (!sym) { | |
590a0088 | 8519 | pr_warn("struct_ops reloc: symbol %zx not found\n", |
ad23b723 | 8520 | (size_t)ELF64_R_SYM(rel->r_info)); |
590a0088 MKL |
8521 | return -LIBBPF_ERRNO__FORMAT; |
8522 | } | |
8523 | ||
ad23b723 AN |
8524 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
8525 | map = find_struct_ops_map_by_offset(obj, rel->r_offset); | |
590a0088 | 8526 | if (!map) { |
ad23b723 AN |
8527 | pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n", |
8528 | (size_t)rel->r_offset); | |
590a0088 MKL |
8529 | return -EINVAL; |
8530 | } | |
8531 | ||
ad23b723 AN |
8532 | moff = rel->r_offset - map->sec_offset; |
8533 | shdr_idx = sym->st_shndx; | |
590a0088 | 8534 | st_ops = map->st_ops; |
ad23b723 | 8535 | pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel->r_offset %zu map->sec_offset %zu name %d (\'%s\')\n", |
590a0088 | 8536 | map->name, |
ad23b723 AN |
8537 | (long long)(rel->r_info >> 32), |
8538 | (long long)sym->st_value, | |
8539 | shdr_idx, (size_t)rel->r_offset, | |
8540 | map->sec_offset, sym->st_name, name); | |
590a0088 MKL |
8541 | |
8542 | if (shdr_idx >= SHN_LORESERVE) { | |
ad23b723 AN |
8543 | pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n", |
8544 | map->name, (size_t)rel->r_offset, shdr_idx); | |
590a0088 MKL |
8545 | return -LIBBPF_ERRNO__RELOC; |
8546 | } | |
ad23b723 | 8547 | if (sym->st_value % BPF_INSN_SZ) { |
7e06aad5 | 8548 | pr_warn("struct_ops reloc %s: invalid target program offset %llu\n", |
ad23b723 | 8549 | map->name, (unsigned long long)sym->st_value); |
7e06aad5 AN |
8550 | return -LIBBPF_ERRNO__FORMAT; |
8551 | } | |
ad23b723 | 8552 | insn_idx = sym->st_value / BPF_INSN_SZ; |
590a0088 MKL |
8553 | |
8554 | member = find_member_by_offset(st_ops->type, moff * 8); | |
8555 | if (!member) { | |
8556 | pr_warn("struct_ops reloc %s: cannot find member at moff %u\n", | |
8557 | map->name, moff); | |
8558 | return -EINVAL; | |
8559 | } | |
8560 | member_idx = member - btf_members(st_ops->type); | |
8561 | name = btf__name_by_offset(btf, member->name_off); | |
8562 | ||
8563 | if (!resolve_func_ptr(btf, member->type, NULL)) { | |
8564 | pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n", | |
8565 | map->name, name); | |
8566 | return -EINVAL; | |
8567 | } | |
8568 | ||
7e06aad5 | 8569 | prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx); |
590a0088 MKL |
8570 | if (!prog) { |
8571 | pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n", | |
8572 | map->name, shdr_idx, name); | |
8573 | return -EINVAL; | |
8574 | } | |
8575 | ||
91b4d1d1 AN |
8576 | /* prevent the use of BPF prog with invalid type */ |
8577 | if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) { | |
8578 | pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n", | |
8579 | map->name, prog->name); | |
8580 | return -EINVAL; | |
8581 | } | |
590a0088 | 8582 | |
91b4d1d1 AN |
8583 | /* if we haven't yet processed this BPF program, record proper |
8584 | * attach_btf_id and member_idx | |
8585 | */ | |
8586 | if (!prog->attach_btf_id) { | |
590a0088 MKL |
8587 | prog->attach_btf_id = st_ops->type_id; |
8588 | prog->expected_attach_type = member_idx; | |
590a0088 | 8589 | } |
91b4d1d1 AN |
8590 | |
8591 | /* struct_ops BPF prog can be re-used between multiple | |
8592 | * .struct_ops as long as it's the same struct_ops struct | |
8593 | * definition and the same function pointer field | |
8594 | */ | |
8595 | if (prog->attach_btf_id != st_ops->type_id || | |
8596 | prog->expected_attach_type != member_idx) { | |
8597 | 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", | |
8598 | map->name, prog->name, prog->sec_name, prog->type, | |
8599 | prog->attach_btf_id, prog->expected_attach_type, name); | |
8600 | return -EINVAL; | |
8601 | } | |
8602 | ||
590a0088 MKL |
8603 | st_ops->progs[member_idx] = prog; |
8604 | } | |
8605 | ||
8606 | return 0; | |
590a0088 MKL |
8607 | } |
8608 | ||
a6ed02ca | 8609 | #define BTF_TRACE_PREFIX "btf_trace_" |
1e092a03 | 8610 | #define BTF_LSM_PREFIX "bpf_lsm_" |
21aef70e | 8611 | #define BTF_ITER_PREFIX "bpf_iter_" |
a6ed02ca KS |
8612 | #define BTF_MAX_NAME_SIZE 128 |
8613 | ||
67234743 AS |
8614 | void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type, |
8615 | const char **prefix, int *kind) | |
8616 | { | |
8617 | switch (attach_type) { | |
8618 | case BPF_TRACE_RAW_TP: | |
8619 | *prefix = BTF_TRACE_PREFIX; | |
8620 | *kind = BTF_KIND_TYPEDEF; | |
8621 | break; | |
8622 | case BPF_LSM_MAC: | |
8623 | *prefix = BTF_LSM_PREFIX; | |
8624 | *kind = BTF_KIND_FUNC; | |
8625 | break; | |
8626 | case BPF_TRACE_ITER: | |
8627 | *prefix = BTF_ITER_PREFIX; | |
8628 | *kind = BTF_KIND_FUNC; | |
8629 | break; | |
8630 | default: | |
8631 | *prefix = ""; | |
8632 | *kind = BTF_KIND_FUNC; | |
8633 | } | |
8634 | } | |
8635 | ||
a6ed02ca KS |
8636 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
8637 | const char *name, __u32 kind) | |
8638 | { | |
8639 | char btf_type_name[BTF_MAX_NAME_SIZE]; | |
8640 | int ret; | |
8641 | ||
8642 | ret = snprintf(btf_type_name, sizeof(btf_type_name), | |
8643 | "%s%s", prefix, name); | |
8644 | /* snprintf returns the number of characters written excluding the | |
c139e40a | 8645 | * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it |
a6ed02ca KS |
8646 | * indicates truncation. |
8647 | */ | |
8648 | if (ret < 0 || ret >= sizeof(btf_type_name)) | |
8649 | return -ENAMETOOLONG; | |
8650 | return btf__find_by_name_kind(btf, btf_type_name, kind); | |
8651 | } | |
8652 | ||
91abb4a6 AN |
8653 | static inline int find_attach_btf_id(struct btf *btf, const char *name, |
8654 | enum bpf_attach_type attach_type) | |
a6ed02ca | 8655 | { |
67234743 AS |
8656 | const char *prefix; |
8657 | int kind; | |
a6ed02ca | 8658 | |
67234743 AS |
8659 | btf_get_kernel_prefix_kind(attach_type, &prefix, &kind); |
8660 | return find_btf_by_prefix_kind(btf, prefix, name, kind); | |
a6ed02ca KS |
8661 | } |
8662 | ||
b8c54ea4 AS |
8663 | int libbpf_find_vmlinux_btf_id(const char *name, |
8664 | enum bpf_attach_type attach_type) | |
12a8654b | 8665 | { |
a6ed02ca | 8666 | struct btf *btf; |
3521ffa2 | 8667 | int err; |
12a8654b | 8668 | |
a710eed3 | 8669 | btf = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
8670 | err = libbpf_get_error(btf); |
8671 | if (err) { | |
12a8654b | 8672 | pr_warn("vmlinux BTF is not found\n"); |
e9fc3ce9 | 8673 | return libbpf_err(err); |
12a8654b AS |
8674 | } |
8675 | ||
91abb4a6 AN |
8676 | err = find_attach_btf_id(btf, name, attach_type); |
8677 | if (err <= 0) | |
8678 | pr_warn("%s is not found in vmlinux BTF\n", name); | |
8679 | ||
3521ffa2 | 8680 | btf__free(btf); |
e9fc3ce9 | 8681 | return libbpf_err(err); |
b8c54ea4 AS |
8682 | } |
8683 | ||
e7bf94db AS |
8684 | static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd) |
8685 | { | |
ebc7b50a DM |
8686 | struct bpf_prog_info info = {}; |
8687 | __u32 info_len = sizeof(info); | |
6cc93e2f | 8688 | struct btf *btf; |
6d2d73cd | 8689 | int err; |
e7bf94db | 8690 | |
ebc7b50a | 8691 | err = bpf_obj_get_info_by_fd(attach_prog_fd, &info, &info_len); |
e9fc3ce9 | 8692 | if (err) { |
ebc7b50a DM |
8693 | pr_warn("failed bpf_obj_get_info_by_fd for FD %d: %d\n", |
8694 | attach_prog_fd, err); | |
e9fc3ce9 | 8695 | return err; |
e7bf94db | 8696 | } |
6d2d73cd QM |
8697 | |
8698 | err = -EINVAL; | |
ebc7b50a | 8699 | if (!info.btf_id) { |
e7bf94db AS |
8700 | pr_warn("The target program doesn't have BTF\n"); |
8701 | goto out; | |
8702 | } | |
ebc7b50a DM |
8703 | btf = btf__load_from_kernel_by_id(info.btf_id); |
8704 | err = libbpf_get_error(btf); | |
8705 | if (err) { | |
8706 | pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err); | |
e7bf94db AS |
8707 | goto out; |
8708 | } | |
8709 | err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC); | |
8710 | btf__free(btf); | |
8711 | if (err <= 0) { | |
8712 | pr_warn("%s is not found in prog's BTF\n", name); | |
8713 | goto out; | |
8714 | } | |
8715 | out: | |
e7bf94db AS |
8716 | return err; |
8717 | } | |
8718 | ||
91abb4a6 AN |
8719 | static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name, |
8720 | enum bpf_attach_type attach_type, | |
8721 | int *btf_obj_fd, int *btf_type_id) | |
8722 | { | |
8723 | int ret, i; | |
8724 | ||
8725 | ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type); | |
8726 | if (ret > 0) { | |
8727 | *btf_obj_fd = 0; /* vmlinux BTF */ | |
8728 | *btf_type_id = ret; | |
8729 | return 0; | |
8730 | } | |
8731 | if (ret != -ENOENT) | |
8732 | return ret; | |
8733 | ||
8734 | ret = load_module_btfs(obj); | |
8735 | if (ret) | |
8736 | return ret; | |
8737 | ||
8738 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
8739 | const struct module_btf *mod = &obj->btf_modules[i]; | |
8740 | ||
8741 | ret = find_attach_btf_id(mod->btf, attach_name, attach_type); | |
8742 | if (ret > 0) { | |
8743 | *btf_obj_fd = mod->fd; | |
8744 | *btf_type_id = ret; | |
8745 | return 0; | |
8746 | } | |
8747 | if (ret == -ENOENT) | |
8748 | continue; | |
8749 | ||
8750 | return ret; | |
8751 | } | |
8752 | ||
8753 | return -ESRCH; | |
8754 | } | |
8755 | ||
15ea31fa AN |
8756 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
8757 | int *btf_obj_fd, int *btf_type_id) | |
b8c54ea4 | 8758 | { |
a6ed02ca KS |
8759 | enum bpf_attach_type attach_type = prog->expected_attach_type; |
8760 | __u32 attach_prog_fd = prog->attach_prog_fd; | |
b6291a6f | 8761 | int err = 0; |
b8c54ea4 | 8762 | |
91abb4a6 AN |
8763 | /* BPF program's BTF ID */ |
8764 | if (attach_prog_fd) { | |
8765 | err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd); | |
8766 | if (err < 0) { | |
8767 | pr_warn("failed to find BPF program (FD %d) BTF ID for '%s': %d\n", | |
8768 | attach_prog_fd, attach_name, err); | |
8769 | return err; | |
8770 | } | |
8771 | *btf_obj_fd = 0; | |
8772 | *btf_type_id = err; | |
8773 | return 0; | |
8774 | } | |
8775 | ||
8776 | /* kernel/module BTF ID */ | |
67234743 AS |
8777 | if (prog->obj->gen_loader) { |
8778 | bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type); | |
8779 | *btf_obj_fd = 0; | |
8780 | *btf_type_id = 1; | |
8781 | } else { | |
8782 | err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id); | |
8783 | } | |
91abb4a6 AN |
8784 | if (err) { |
8785 | pr_warn("failed to find kernel BTF type ID of '%s': %d\n", attach_name, err); | |
b8c54ea4 | 8786 | return err; |
12a8654b | 8787 | } |
91abb4a6 | 8788 | return 0; |
12a8654b AS |
8789 | } |
8790 | ||
956b620f AI |
8791 | int libbpf_attach_type_by_name(const char *name, |
8792 | enum bpf_attach_type *attach_type) | |
8793 | { | |
c76e4c22 | 8794 | char *type_names; |
b6291a6f | 8795 | const struct bpf_sec_def *sec_def; |
956b620f AI |
8796 | |
8797 | if (!name) | |
e9fc3ce9 | 8798 | return libbpf_err(-EINVAL); |
956b620f | 8799 | |
b6291a6f AN |
8800 | sec_def = find_sec_def(name); |
8801 | if (!sec_def) { | |
8802 | pr_debug("failed to guess attach type based on ELF section name '%s'\n", name); | |
8803 | type_names = libbpf_get_type_names(true); | |
8804 | if (type_names != NULL) { | |
8805 | pr_debug("attachable section(type) names are:%s\n", type_names); | |
8806 | free(type_names); | |
8807 | } | |
8808 | ||
8809 | return libbpf_err(-EINVAL); | |
c76e4c22 TS |
8810 | } |
8811 | ||
15ea31fa AN |
8812 | if (sec_def->preload_fn != libbpf_preload_prog) |
8813 | return libbpf_err(-EINVAL); | |
8814 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
b6291a6f AN |
8815 | return libbpf_err(-EINVAL); |
8816 | ||
8817 | *attach_type = sec_def->expected_attach_type; | |
8818 | return 0; | |
956b620f AI |
8819 | } |
8820 | ||
a324aae3 | 8821 | int bpf_map__fd(const struct bpf_map *map) |
9d759a9b | 8822 | { |
e9fc3ce9 | 8823 | return map ? map->fd : libbpf_err(-EINVAL); |
9d759a9b WN |
8824 | } |
8825 | ||
a324aae3 | 8826 | const struct bpf_map_def *bpf_map__def(const struct bpf_map *map) |
9d759a9b | 8827 | { |
e9fc3ce9 | 8828 | return map ? &map->def : libbpf_err_ptr(-EINVAL); |
9d759a9b WN |
8829 | } |
8830 | ||
aed65917 AN |
8831 | static bool map_uses_real_name(const struct bpf_map *map) |
8832 | { | |
8833 | /* Since libbpf started to support custom .data.* and .rodata.* maps, | |
8834 | * their user-visible name differs from kernel-visible name. Users see | |
8835 | * such map's corresponding ELF section name as a map name. | |
8836 | * This check distinguishes .data/.rodata from .data.* and .rodata.* | |
8837 | * maps to know which name has to be returned to the user. | |
8838 | */ | |
8839 | if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0) | |
8840 | return true; | |
8841 | if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0) | |
8842 | return true; | |
8843 | return false; | |
8844 | } | |
8845 | ||
a324aae3 | 8846 | const char *bpf_map__name(const struct bpf_map *map) |
561bbcca | 8847 | { |
aed65917 AN |
8848 | if (!map) |
8849 | return NULL; | |
8850 | ||
8851 | if (map_uses_real_name(map)) | |
8852 | return map->real_name; | |
8853 | ||
8854 | return map->name; | |
561bbcca WN |
8855 | } |
8856 | ||
1bdb6c9a AN |
8857 | enum bpf_map_type bpf_map__type(const struct bpf_map *map) |
8858 | { | |
8859 | return map->def.type; | |
8860 | } | |
8861 | ||
8862 | int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type) | |
8863 | { | |
8864 | if (map->fd >= 0) | |
e9fc3ce9 | 8865 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
8866 | map->def.type = type; |
8867 | return 0; | |
8868 | } | |
8869 | ||
8870 | __u32 bpf_map__map_flags(const struct bpf_map *map) | |
8871 | { | |
8872 | return map->def.map_flags; | |
8873 | } | |
8874 | ||
8875 | int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags) | |
8876 | { | |
8877 | if (map->fd >= 0) | |
e9fc3ce9 | 8878 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
8879 | map->def.map_flags = flags; |
8880 | return 0; | |
8881 | } | |
8882 | ||
8883 | __u32 bpf_map__numa_node(const struct bpf_map *map) | |
8884 | { | |
8885 | return map->numa_node; | |
8886 | } | |
8887 | ||
8888 | int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node) | |
8889 | { | |
8890 | if (map->fd >= 0) | |
e9fc3ce9 | 8891 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
8892 | map->numa_node = numa_node; |
8893 | return 0; | |
8894 | } | |
8895 | ||
8896 | __u32 bpf_map__key_size(const struct bpf_map *map) | |
8897 | { | |
8898 | return map->def.key_size; | |
8899 | } | |
8900 | ||
8901 | int bpf_map__set_key_size(struct bpf_map *map, __u32 size) | |
8902 | { | |
8903 | if (map->fd >= 0) | |
e9fc3ce9 | 8904 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
8905 | map->def.key_size = size; |
8906 | return 0; | |
8907 | } | |
8908 | ||
8909 | __u32 bpf_map__value_size(const struct bpf_map *map) | |
8910 | { | |
8911 | return map->def.value_size; | |
8912 | } | |
8913 | ||
8914 | int bpf_map__set_value_size(struct bpf_map *map, __u32 size) | |
8915 | { | |
8916 | if (map->fd >= 0) | |
e9fc3ce9 | 8917 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
8918 | map->def.value_size = size; |
8919 | return 0; | |
8920 | } | |
8921 | ||
5b891af7 | 8922 | __u32 bpf_map__btf_key_type_id(const struct bpf_map *map) |
8a138aed | 8923 | { |
61746dbe | 8924 | return map ? map->btf_key_type_id : 0; |
8a138aed MKL |
8925 | } |
8926 | ||
5b891af7 | 8927 | __u32 bpf_map__btf_value_type_id(const struct bpf_map *map) |
8a138aed | 8928 | { |
61746dbe | 8929 | return map ? map->btf_value_type_id : 0; |
8a138aed MKL |
8930 | } |
8931 | ||
edb13ed4 ACM |
8932 | int bpf_map__set_priv(struct bpf_map *map, void *priv, |
8933 | bpf_map_clear_priv_t clear_priv) | |
9d759a9b WN |
8934 | { |
8935 | if (!map) | |
e9fc3ce9 | 8936 | return libbpf_err(-EINVAL); |
9d759a9b WN |
8937 | |
8938 | if (map->priv) { | |
8939 | if (map->clear_priv) | |
8940 | map->clear_priv(map, map->priv); | |
8941 | } | |
8942 | ||
8943 | map->priv = priv; | |
8944 | map->clear_priv = clear_priv; | |
8945 | return 0; | |
8946 | } | |
8947 | ||
a324aae3 | 8948 | void *bpf_map__priv(const struct bpf_map *map) |
9d759a9b | 8949 | { |
e9fc3ce9 | 8950 | return map ? map->priv : libbpf_err_ptr(-EINVAL); |
9d759a9b WN |
8951 | } |
8952 | ||
e2842be5 THJ |
8953 | int bpf_map__set_initial_value(struct bpf_map *map, |
8954 | const void *data, size_t size) | |
8955 | { | |
8956 | if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG || | |
8957 | size != map->def.value_size || map->fd >= 0) | |
e9fc3ce9 | 8958 | return libbpf_err(-EINVAL); |
e2842be5 THJ |
8959 | |
8960 | memcpy(map->mmaped, data, size); | |
8961 | return 0; | |
8962 | } | |
8963 | ||
7723256b AS |
8964 | const void *bpf_map__initial_value(struct bpf_map *map, size_t *psize) |
8965 | { | |
8966 | if (!map->mmaped) | |
8967 | return NULL; | |
8968 | *psize = map->def.value_size; | |
8969 | return map->mmaped; | |
8970 | } | |
8971 | ||
a324aae3 | 8972 | bool bpf_map__is_offload_neutral(const struct bpf_map *map) |
f83fb22c JK |
8973 | { |
8974 | return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; | |
8975 | } | |
8976 | ||
a324aae3 | 8977 | bool bpf_map__is_internal(const struct bpf_map *map) |
d859900c DB |
8978 | { |
8979 | return map->libbpf_type != LIBBPF_MAP_UNSPEC; | |
8980 | } | |
8981 | ||
1bdb6c9a AN |
8982 | __u32 bpf_map__ifindex(const struct bpf_map *map) |
8983 | { | |
8984 | return map->map_ifindex; | |
8985 | } | |
8986 | ||
8987 | int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex) | |
9aba3613 | 8988 | { |
1bdb6c9a | 8989 | if (map->fd >= 0) |
e9fc3ce9 | 8990 | return libbpf_err(-EBUSY); |
9aba3613 | 8991 | map->map_ifindex = ifindex; |
1bdb6c9a | 8992 | return 0; |
9aba3613 JK |
8993 | } |
8994 | ||
addb9fc9 NS |
8995 | int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd) |
8996 | { | |
8997 | if (!bpf_map_type__is_map_in_map(map->def.type)) { | |
be18010e | 8998 | pr_warn("error: unsupported map type\n"); |
e9fc3ce9 | 8999 | return libbpf_err(-EINVAL); |
addb9fc9 NS |
9000 | } |
9001 | if (map->inner_map_fd != -1) { | |
be18010e | 9002 | pr_warn("error: inner_map_fd already specified\n"); |
e9fc3ce9 | 9003 | return libbpf_err(-EINVAL); |
addb9fc9 | 9004 | } |
b3278099 | 9005 | zfree(&map->inner_map); |
addb9fc9 NS |
9006 | map->inner_map_fd = fd; |
9007 | return 0; | |
9008 | } | |
9009 | ||
0c19a9fb | 9010 | static struct bpf_map * |
a324aae3 | 9011 | __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i) |
9d759a9b | 9012 | { |
0c19a9fb | 9013 | ssize_t idx; |
9d759a9b WN |
9014 | struct bpf_map *s, *e; |
9015 | ||
9016 | if (!obj || !obj->maps) | |
e9fc3ce9 | 9017 | return errno = EINVAL, NULL; |
9d759a9b WN |
9018 | |
9019 | s = obj->maps; | |
9020 | e = obj->maps + obj->nr_maps; | |
9021 | ||
0c19a9fb | 9022 | if ((m < s) || (m >= e)) { |
be18010e KW |
9023 | pr_warn("error in %s: map handler doesn't belong to object\n", |
9024 | __func__); | |
e9fc3ce9 | 9025 | return errno = EINVAL, NULL; |
9d759a9b WN |
9026 | } |
9027 | ||
0c19a9fb SF |
9028 | idx = (m - obj->maps) + i; |
9029 | if (idx >= obj->nr_maps || idx < 0) | |
9d759a9b WN |
9030 | return NULL; |
9031 | return &obj->maps[idx]; | |
9032 | } | |
561bbcca | 9033 | |
0c19a9fb | 9034 | struct bpf_map * |
a324aae3 | 9035 | bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj) |
2088a3a7 HC |
9036 | { |
9037 | return bpf_object__next_map(obj, prev); | |
9038 | } | |
9039 | ||
9040 | struct bpf_map * | |
9041 | bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev) | |
0c19a9fb SF |
9042 | { |
9043 | if (prev == NULL) | |
9044 | return obj->maps; | |
9045 | ||
9046 | return __bpf_map__iter(prev, obj, 1); | |
9047 | } | |
9048 | ||
9049 | struct bpf_map * | |
a324aae3 | 9050 | bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj) |
2088a3a7 HC |
9051 | { |
9052 | return bpf_object__prev_map(obj, next); | |
9053 | } | |
9054 | ||
9055 | struct bpf_map * | |
9056 | bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next) | |
0c19a9fb SF |
9057 | { |
9058 | if (next == NULL) { | |
9059 | if (!obj->nr_maps) | |
9060 | return NULL; | |
9061 | return obj->maps + obj->nr_maps - 1; | |
9062 | } | |
9063 | ||
9064 | return __bpf_map__iter(next, obj, -1); | |
9065 | } | |
9066 | ||
561bbcca | 9067 | struct bpf_map * |
a324aae3 | 9068 | bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name) |
561bbcca WN |
9069 | { |
9070 | struct bpf_map *pos; | |
9071 | ||
f74a53d9 | 9072 | bpf_object__for_each_map(pos, obj) { |
26071635 AN |
9073 | /* if it's a special internal map name (which always starts |
9074 | * with dot) then check if that special name matches the | |
9075 | * real map name (ELF section name) | |
9076 | */ | |
9077 | if (name[0] == '.') { | |
9078 | if (pos->real_name && strcmp(pos->real_name, name) == 0) | |
9079 | return pos; | |
9080 | continue; | |
9081 | } | |
9082 | /* otherwise map name has to be an exact match */ | |
aed65917 AN |
9083 | if (map_uses_real_name(pos)) { |
9084 | if (strcmp(pos->real_name, name) == 0) | |
9085 | return pos; | |
9086 | continue; | |
9087 | } | |
9088 | if (strcmp(pos->name, name) == 0) | |
561bbcca WN |
9089 | return pos; |
9090 | } | |
e9fc3ce9 | 9091 | return errno = ENOENT, NULL; |
561bbcca | 9092 | } |
5a6acad1 | 9093 | |
f3cea32d | 9094 | int |
a324aae3 | 9095 | bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name) |
f3cea32d MF |
9096 | { |
9097 | return bpf_map__fd(bpf_object__find_map_by_name(obj, name)); | |
9098 | } | |
9099 | ||
5a6acad1 WN |
9100 | struct bpf_map * |
9101 | bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset) | |
9102 | { | |
e9fc3ce9 | 9103 | return libbpf_err_ptr(-ENOTSUP); |
5a6acad1 | 9104 | } |
e28ff1a8 JS |
9105 | |
9106 | long libbpf_get_error(const void *ptr) | |
9107 | { | |
e9fc3ce9 AN |
9108 | if (!IS_ERR_OR_NULL(ptr)) |
9109 | return 0; | |
9110 | ||
9111 | if (IS_ERR(ptr)) | |
9112 | errno = -PTR_ERR(ptr); | |
9113 | ||
9114 | /* If ptr == NULL, then errno should be already set by the failing | |
9115 | * API, because libbpf never returns NULL on success and it now always | |
9116 | * sets errno on error. So no extra errno handling for ptr == NULL | |
9117 | * case. | |
9118 | */ | |
9119 | return -errno; | |
e28ff1a8 | 9120 | } |
6f6d33f3 JF |
9121 | |
9122 | int bpf_prog_load(const char *file, enum bpf_prog_type type, | |
9123 | struct bpf_object **pobj, int *prog_fd) | |
d7be143b AI |
9124 | { |
9125 | struct bpf_prog_load_attr attr; | |
9126 | ||
9127 | memset(&attr, 0, sizeof(struct bpf_prog_load_attr)); | |
9128 | attr.file = file; | |
9129 | attr.prog_type = type; | |
9130 | attr.expected_attach_type = 0; | |
9131 | ||
9132 | return bpf_prog_load_xattr(&attr, pobj, prog_fd); | |
9133 | } | |
9134 | ||
9135 | int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr, | |
9136 | struct bpf_object **pobj, int *prog_fd) | |
6f6d33f3 | 9137 | { |
33bae185 | 9138 | struct bpf_object_open_attr open_attr = {}; |
48cca7e4 | 9139 | struct bpf_program *prog, *first_prog = NULL; |
6f6d33f3 | 9140 | struct bpf_object *obj; |
f0307a7e | 9141 | struct bpf_map *map; |
6f6d33f3 JF |
9142 | int err; |
9143 | ||
d7be143b | 9144 | if (!attr) |
e9fc3ce9 | 9145 | return libbpf_err(-EINVAL); |
17387dd5 | 9146 | if (!attr->file) |
e9fc3ce9 | 9147 | return libbpf_err(-EINVAL); |
d7be143b | 9148 | |
33bae185 LY |
9149 | open_attr.file = attr->file; |
9150 | open_attr.prog_type = attr->prog_type; | |
9151 | ||
07f2d4ea | 9152 | obj = bpf_object__open_xattr(&open_attr); |
e9fc3ce9 AN |
9153 | err = libbpf_get_error(obj); |
9154 | if (err) | |
9155 | return libbpf_err(-ENOENT); | |
6f6d33f3 | 9156 | |
48cca7e4 | 9157 | bpf_object__for_each_program(prog, obj) { |
dd4436bb | 9158 | enum bpf_attach_type attach_type = attr->expected_attach_type; |
48cca7e4 | 9159 | /* |
dd4436bb AN |
9160 | * to preserve backwards compatibility, bpf_prog_load treats |
9161 | * attr->prog_type, if specified, as an override to whatever | |
9162 | * bpf_object__open guessed | |
48cca7e4 | 9163 | */ |
dd4436bb AN |
9164 | if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) { |
9165 | bpf_program__set_type(prog, attr->prog_type); | |
9166 | bpf_program__set_expected_attach_type(prog, | |
9167 | attach_type); | |
9168 | } | |
9169 | if (bpf_program__get_type(prog) == BPF_PROG_TYPE_UNSPEC) { | |
9170 | /* | |
9171 | * we haven't guessed from section name and user | |
9172 | * didn't provide a fallback type, too bad... | |
9173 | */ | |
9174 | bpf_object__close(obj); | |
e9fc3ce9 | 9175 | return libbpf_err(-EINVAL); |
583c9009 | 9176 | } |
48cca7e4 | 9177 | |
dd4436bb | 9178 | prog->prog_ifindex = attr->ifindex; |
da11b417 | 9179 | prog->log_level = attr->log_level; |
2b288740 | 9180 | prog->prog_flags |= attr->prog_flags; |
69495d2a | 9181 | if (!first_prog) |
48cca7e4 AS |
9182 | first_prog = prog; |
9183 | } | |
9184 | ||
f74a53d9 | 9185 | bpf_object__for_each_map(map, obj) { |
f83fb22c JK |
9186 | if (!bpf_map__is_offload_neutral(map)) |
9187 | map->map_ifindex = attr->ifindex; | |
f0307a7e DB |
9188 | } |
9189 | ||
48cca7e4 | 9190 | if (!first_prog) { |
be18010e | 9191 | pr_warn("object file doesn't contain bpf program\n"); |
48cca7e4 | 9192 | bpf_object__close(obj); |
e9fc3ce9 | 9193 | return libbpf_err(-ENOENT); |
583c9009 RG |
9194 | } |
9195 | ||
6f6d33f3 JF |
9196 | err = bpf_object__load(obj); |
9197 | if (err) { | |
9198 | bpf_object__close(obj); | |
e9fc3ce9 | 9199 | return libbpf_err(err); |
6f6d33f3 JF |
9200 | } |
9201 | ||
9202 | *pobj = obj; | |
48cca7e4 | 9203 | *prog_fd = bpf_program__fd(first_prog); |
6f6d33f3 JF |
9204 | return 0; |
9205 | } | |
d0cabbb0 | 9206 | |
1c2e9efc | 9207 | struct bpf_link { |
d6958706 | 9208 | int (*detach)(struct bpf_link *link); |
d88b71d4 | 9209 | void (*dealloc)(struct bpf_link *link); |
c016b68e AN |
9210 | char *pin_path; /* NULL, if not pinned */ |
9211 | int fd; /* hook FD, -1 if not applicable */ | |
d6958706 | 9212 | bool disconnected; |
1c2e9efc AN |
9213 | }; |
9214 | ||
cc4f864b AN |
9215 | /* Replace link's underlying BPF program with the new one */ |
9216 | int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog) | |
9217 | { | |
e9fc3ce9 | 9218 | int ret; |
c139e40a | 9219 | |
e9fc3ce9 AN |
9220 | ret = bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL); |
9221 | return libbpf_err_errno(ret); | |
cc4f864b AN |
9222 | } |
9223 | ||
d6958706 AN |
9224 | /* Release "ownership" of underlying BPF resource (typically, BPF program |
9225 | * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected | |
9226 | * link, when destructed through bpf_link__destroy() call won't attempt to | |
9227 | * detach/unregisted that BPF resource. This is useful in situations where, | |
9228 | * say, attached BPF program has to outlive userspace program that attached it | |
9229 | * in the system. Depending on type of BPF program, though, there might be | |
9230 | * additional steps (like pinning BPF program in BPF FS) necessary to ensure | |
9231 | * exit of userspace program doesn't trigger automatic detachment and clean up | |
9232 | * inside the kernel. | |
9233 | */ | |
9234 | void bpf_link__disconnect(struct bpf_link *link) | |
9235 | { | |
9236 | link->disconnected = true; | |
9237 | } | |
9238 | ||
1c2e9efc AN |
9239 | int bpf_link__destroy(struct bpf_link *link) |
9240 | { | |
d6958706 | 9241 | int err = 0; |
1c2e9efc | 9242 | |
50450fc7 | 9243 | if (IS_ERR_OR_NULL(link)) |
1c2e9efc AN |
9244 | return 0; |
9245 | ||
d6958706 AN |
9246 | if (!link->disconnected && link->detach) |
9247 | err = link->detach(link); | |
c016b68e AN |
9248 | if (link->pin_path) |
9249 | free(link->pin_path); | |
d88b71d4 AN |
9250 | if (link->dealloc) |
9251 | link->dealloc(link); | |
9252 | else | |
9253 | free(link); | |
1c2e9efc | 9254 | |
e9fc3ce9 | 9255 | return libbpf_err(err); |
1c2e9efc AN |
9256 | } |
9257 | ||
c016b68e AN |
9258 | int bpf_link__fd(const struct bpf_link *link) |
9259 | { | |
9260 | return link->fd; | |
9261 | } | |
9262 | ||
9263 | const char *bpf_link__pin_path(const struct bpf_link *link) | |
9264 | { | |
9265 | return link->pin_path; | |
9266 | } | |
9267 | ||
9268 | static int bpf_link__detach_fd(struct bpf_link *link) | |
9269 | { | |
e9fc3ce9 | 9270 | return libbpf_err_errno(close(link->fd)); |
c016b68e AN |
9271 | } |
9272 | ||
9273 | struct bpf_link *bpf_link__open(const char *path) | |
9274 | { | |
9275 | struct bpf_link *link; | |
9276 | int fd; | |
9277 | ||
9278 | fd = bpf_obj_get(path); | |
9279 | if (fd < 0) { | |
9280 | fd = -errno; | |
9281 | pr_warn("failed to open link at %s: %d\n", path, fd); | |
e9fc3ce9 | 9282 | return libbpf_err_ptr(fd); |
c016b68e AN |
9283 | } |
9284 | ||
9285 | link = calloc(1, sizeof(*link)); | |
9286 | if (!link) { | |
9287 | close(fd); | |
e9fc3ce9 | 9288 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
9289 | } |
9290 | link->detach = &bpf_link__detach_fd; | |
9291 | link->fd = fd; | |
9292 | ||
9293 | link->pin_path = strdup(path); | |
9294 | if (!link->pin_path) { | |
9295 | bpf_link__destroy(link); | |
e9fc3ce9 | 9296 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
9297 | } |
9298 | ||
9299 | return link; | |
9300 | } | |
9301 | ||
2e49527e AN |
9302 | int bpf_link__detach(struct bpf_link *link) |
9303 | { | |
9304 | return bpf_link_detach(link->fd) ? -errno : 0; | |
9305 | } | |
9306 | ||
c016b68e AN |
9307 | int bpf_link__pin(struct bpf_link *link, const char *path) |
9308 | { | |
9309 | int err; | |
9310 | ||
9311 | if (link->pin_path) | |
e9fc3ce9 | 9312 | return libbpf_err(-EBUSY); |
c016b68e AN |
9313 | err = make_parent_dir(path); |
9314 | if (err) | |
e9fc3ce9 | 9315 | return libbpf_err(err); |
c016b68e AN |
9316 | err = check_path(path); |
9317 | if (err) | |
e9fc3ce9 | 9318 | return libbpf_err(err); |
c016b68e AN |
9319 | |
9320 | link->pin_path = strdup(path); | |
9321 | if (!link->pin_path) | |
e9fc3ce9 | 9322 | return libbpf_err(-ENOMEM); |
c016b68e AN |
9323 | |
9324 | if (bpf_obj_pin(link->fd, link->pin_path)) { | |
9325 | err = -errno; | |
9326 | zfree(&link->pin_path); | |
e9fc3ce9 | 9327 | return libbpf_err(err); |
c016b68e AN |
9328 | } |
9329 | ||
9330 | pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path); | |
9331 | return 0; | |
9332 | } | |
9333 | ||
9334 | int bpf_link__unpin(struct bpf_link *link) | |
9335 | { | |
9336 | int err; | |
9337 | ||
9338 | if (!link->pin_path) | |
e9fc3ce9 | 9339 | return libbpf_err(-EINVAL); |
c016b68e AN |
9340 | |
9341 | err = unlink(link->pin_path); | |
9342 | if (err != 0) | |
af0efa05 | 9343 | return -errno; |
c016b68e AN |
9344 | |
9345 | pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path); | |
9346 | zfree(&link->pin_path); | |
9347 | return 0; | |
9348 | } | |
63f2f5ee | 9349 | |
668ace0e AN |
9350 | struct bpf_link_perf { |
9351 | struct bpf_link link; | |
9352 | int perf_event_fd; | |
ca304b40 RDT |
9353 | /* legacy kprobe support: keep track of probe identifier and type */ |
9354 | char *legacy_probe_name; | |
46ed5fc3 | 9355 | bool legacy_is_kprobe; |
ca304b40 | 9356 | bool legacy_is_retprobe; |
668ace0e AN |
9357 | }; |
9358 | ||
46ed5fc3 | 9359 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe); |
cc10623c | 9360 | static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe); |
46ed5fc3 | 9361 | |
668ace0e | 9362 | static int bpf_link_perf_detach(struct bpf_link *link) |
63f2f5ee | 9363 | { |
668ace0e AN |
9364 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); |
9365 | int err = 0; | |
63f2f5ee | 9366 | |
668ace0e | 9367 | if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0) |
63f2f5ee AN |
9368 | err = -errno; |
9369 | ||
668ace0e AN |
9370 | if (perf_link->perf_event_fd != link->fd) |
9371 | close(perf_link->perf_event_fd); | |
c016b68e | 9372 | close(link->fd); |
668ace0e | 9373 | |
cc10623c | 9374 | /* legacy uprobe/kprobe needs to be removed after perf event fd closure */ |
46ed5fc3 AN |
9375 | if (perf_link->legacy_probe_name) { |
9376 | if (perf_link->legacy_is_kprobe) { | |
9377 | err = remove_kprobe_event_legacy(perf_link->legacy_probe_name, | |
9378 | perf_link->legacy_is_retprobe); | |
cc10623c AN |
9379 | } else { |
9380 | err = remove_uprobe_event_legacy(perf_link->legacy_probe_name, | |
9381 | perf_link->legacy_is_retprobe); | |
46ed5fc3 AN |
9382 | } |
9383 | } | |
ca304b40 RDT |
9384 | |
9385 | return err; | |
63f2f5ee AN |
9386 | } |
9387 | ||
668ace0e AN |
9388 | static void bpf_link_perf_dealloc(struct bpf_link *link) |
9389 | { | |
9390 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
9391 | ||
ca304b40 | 9392 | free(perf_link->legacy_probe_name); |
668ace0e AN |
9393 | free(perf_link); |
9394 | } | |
9395 | ||
942025c9 | 9396 | struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd, |
47faff37 | 9397 | const struct bpf_perf_event_opts *opts) |
63f2f5ee AN |
9398 | { |
9399 | char errmsg[STRERR_BUFSIZE]; | |
668ace0e AN |
9400 | struct bpf_link_perf *link; |
9401 | int prog_fd, link_fd = -1, err; | |
63f2f5ee | 9402 | |
47faff37 AN |
9403 | if (!OPTS_VALID(opts, bpf_perf_event_opts)) |
9404 | return libbpf_err_ptr(-EINVAL); | |
9405 | ||
63f2f5ee | 9406 | if (pfd < 0) { |
52109584 AN |
9407 | pr_warn("prog '%s': invalid perf event FD %d\n", |
9408 | prog->name, pfd); | |
e9fc3ce9 | 9409 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
9410 | } |
9411 | prog_fd = bpf_program__fd(prog); | |
9412 | if (prog_fd < 0) { | |
52109584 AN |
9413 | pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n", |
9414 | prog->name); | |
e9fc3ce9 | 9415 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
9416 | } |
9417 | ||
d6958706 | 9418 | link = calloc(1, sizeof(*link)); |
63f2f5ee | 9419 | if (!link) |
e9fc3ce9 | 9420 | return libbpf_err_ptr(-ENOMEM); |
668ace0e AN |
9421 | link->link.detach = &bpf_link_perf_detach; |
9422 | link->link.dealloc = &bpf_link_perf_dealloc; | |
9423 | link->perf_event_fd = pfd; | |
63f2f5ee | 9424 | |
668ace0e | 9425 | if (kernel_supports(prog->obj, FEAT_PERF_LINK)) { |
47faff37 AN |
9426 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts, |
9427 | .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0)); | |
9428 | ||
9429 | link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts); | |
668ace0e AN |
9430 | if (link_fd < 0) { |
9431 | err = -errno; | |
9432 | pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n", | |
9433 | prog->name, pfd, | |
9434 | err, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
9435 | goto err_out; | |
9436 | } | |
9437 | link->link.fd = link_fd; | |
9438 | } else { | |
47faff37 AN |
9439 | if (OPTS_GET(opts, bpf_cookie, 0)) { |
9440 | pr_warn("prog '%s': user context value is not supported\n", prog->name); | |
9441 | err = -EOPNOTSUPP; | |
9442 | goto err_out; | |
9443 | } | |
9444 | ||
668ace0e AN |
9445 | if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) { |
9446 | err = -errno; | |
9447 | pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n", | |
9448 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
9449 | if (err == -EPROTO) | |
9450 | pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n", | |
9451 | prog->name, pfd); | |
9452 | goto err_out; | |
9453 | } | |
9454 | link->link.fd = pfd; | |
63f2f5ee AN |
9455 | } |
9456 | if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
9457 | err = -errno; | |
668ace0e | 9458 | pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n", |
52109584 | 9459 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
668ace0e | 9460 | goto err_out; |
63f2f5ee | 9461 | } |
668ace0e AN |
9462 | |
9463 | return &link->link; | |
9464 | err_out: | |
9465 | if (link_fd >= 0) | |
9466 | close(link_fd); | |
9467 | free(link); | |
9468 | return libbpf_err_ptr(err); | |
63f2f5ee AN |
9469 | } |
9470 | ||
942025c9 | 9471 | struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd) |
47faff37 AN |
9472 | { |
9473 | return bpf_program__attach_perf_event_opts(prog, pfd, NULL); | |
9474 | } | |
9475 | ||
b2650027 AN |
9476 | /* |
9477 | * this function is expected to parse integer in the range of [0, 2^31-1] from | |
9478 | * given file using scanf format string fmt. If actual parsed value is | |
9479 | * negative, the result might be indistinguishable from error | |
9480 | */ | |
9481 | static int parse_uint_from_file(const char *file, const char *fmt) | |
9482 | { | |
9483 | char buf[STRERR_BUFSIZE]; | |
9484 | int err, ret; | |
9485 | FILE *f; | |
9486 | ||
9487 | f = fopen(file, "r"); | |
9488 | if (!f) { | |
9489 | err = -errno; | |
9490 | pr_debug("failed to open '%s': %s\n", file, | |
9491 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
9492 | return err; | |
9493 | } | |
9494 | err = fscanf(f, fmt, &ret); | |
9495 | if (err != 1) { | |
9496 | err = err == EOF ? -EIO : -errno; | |
9497 | pr_debug("failed to parse '%s': %s\n", file, | |
9498 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
9499 | fclose(f); | |
9500 | return err; | |
9501 | } | |
9502 | fclose(f); | |
9503 | return ret; | |
9504 | } | |
9505 | ||
9506 | static int determine_kprobe_perf_type(void) | |
9507 | { | |
9508 | const char *file = "/sys/bus/event_source/devices/kprobe/type"; | |
9509 | ||
9510 | return parse_uint_from_file(file, "%d\n"); | |
9511 | } | |
9512 | ||
9513 | static int determine_uprobe_perf_type(void) | |
9514 | { | |
9515 | const char *file = "/sys/bus/event_source/devices/uprobe/type"; | |
9516 | ||
9517 | return parse_uint_from_file(file, "%d\n"); | |
9518 | } | |
9519 | ||
9520 | static int determine_kprobe_retprobe_bit(void) | |
9521 | { | |
9522 | const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe"; | |
9523 | ||
9524 | return parse_uint_from_file(file, "config:%d\n"); | |
9525 | } | |
9526 | ||
9527 | static int determine_uprobe_retprobe_bit(void) | |
9528 | { | |
9529 | const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe"; | |
9530 | ||
9531 | return parse_uint_from_file(file, "config:%d\n"); | |
9532 | } | |
9533 | ||
5e3b8356 AN |
9534 | #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32 |
9535 | #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32 | |
9536 | ||
b2650027 | 9537 | static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name, |
5e3b8356 | 9538 | uint64_t offset, int pid, size_t ref_ctr_off) |
b2650027 AN |
9539 | { |
9540 | struct perf_event_attr attr = {}; | |
9541 | char errmsg[STRERR_BUFSIZE]; | |
9542 | int type, pfd, err; | |
9543 | ||
5e3b8356 AN |
9544 | if (ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS)) |
9545 | return -EINVAL; | |
9546 | ||
b2650027 AN |
9547 | type = uprobe ? determine_uprobe_perf_type() |
9548 | : determine_kprobe_perf_type(); | |
9549 | if (type < 0) { | |
be18010e KW |
9550 | pr_warn("failed to determine %s perf type: %s\n", |
9551 | uprobe ? "uprobe" : "kprobe", | |
9552 | libbpf_strerror_r(type, errmsg, sizeof(errmsg))); | |
b2650027 AN |
9553 | return type; |
9554 | } | |
9555 | if (retprobe) { | |
9556 | int bit = uprobe ? determine_uprobe_retprobe_bit() | |
9557 | : determine_kprobe_retprobe_bit(); | |
9558 | ||
9559 | if (bit < 0) { | |
be18010e KW |
9560 | pr_warn("failed to determine %s retprobe bit: %s\n", |
9561 | uprobe ? "uprobe" : "kprobe", | |
9562 | libbpf_strerror_r(bit, errmsg, sizeof(errmsg))); | |
b2650027 AN |
9563 | return bit; |
9564 | } | |
9565 | attr.config |= 1 << bit; | |
9566 | } | |
9567 | attr.size = sizeof(attr); | |
9568 | attr.type = type; | |
5e3b8356 | 9569 | attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT; |
36db2a94 AN |
9570 | attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */ |
9571 | attr.config2 = offset; /* kprobe_addr or probe_offset */ | |
b2650027 AN |
9572 | |
9573 | /* pid filter is meaningful only for uprobes */ | |
9574 | pfd = syscall(__NR_perf_event_open, &attr, | |
9575 | pid < 0 ? -1 : pid /* pid */, | |
9576 | pid == -1 ? 0 : -1 /* cpu */, | |
9577 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
9578 | if (pfd < 0) { | |
9579 | err = -errno; | |
be18010e KW |
9580 | pr_warn("%s perf_event_open() failed: %s\n", |
9581 | uprobe ? "uprobe" : "kprobe", | |
9582 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
b2650027 AN |
9583 | return err; |
9584 | } | |
9585 | return pfd; | |
9586 | } | |
9587 | ||
46ed5fc3 AN |
9588 | static int append_to_file(const char *file, const char *fmt, ...) |
9589 | { | |
9590 | int fd, n, err = 0; | |
9591 | va_list ap; | |
9592 | ||
9593 | fd = open(file, O_WRONLY | O_APPEND, 0); | |
9594 | if (fd < 0) | |
9595 | return -errno; | |
9596 | ||
9597 | va_start(ap, fmt); | |
9598 | n = vdprintf(fd, fmt, ap); | |
9599 | va_end(ap); | |
9600 | ||
9601 | if (n < 0) | |
9602 | err = -errno; | |
9603 | ||
9604 | close(fd); | |
9605 | return err; | |
9606 | } | |
9607 | ||
9608 | static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz, | |
9609 | const char *kfunc_name, size_t offset) | |
9610 | { | |
9611 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), kfunc_name, offset); | |
9612 | } | |
9613 | ||
9614 | static int add_kprobe_event_legacy(const char *probe_name, bool retprobe, | |
9615 | const char *kfunc_name, size_t offset) | |
9616 | { | |
9617 | const char *file = "/sys/kernel/debug/tracing/kprobe_events"; | |
9618 | ||
9619 | return append_to_file(file, "%c:%s/%s %s+0x%zx", | |
9620 | retprobe ? 'r' : 'p', | |
9621 | retprobe ? "kretprobes" : "kprobes", | |
9622 | probe_name, kfunc_name, offset); | |
9623 | } | |
9624 | ||
9625 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe) | |
9626 | { | |
9627 | const char *file = "/sys/kernel/debug/tracing/kprobe_events"; | |
9628 | ||
9629 | return append_to_file(file, "-:%s/%s", retprobe ? "kretprobes" : "kprobes", probe_name); | |
9630 | } | |
9631 | ||
9632 | static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
9633 | { | |
9634 | char file[256]; | |
9635 | ||
9636 | snprintf(file, sizeof(file), | |
9637 | "/sys/kernel/debug/tracing/events/%s/%s/id", | |
9638 | retprobe ? "kretprobes" : "kprobes", probe_name); | |
9639 | ||
9640 | return parse_uint_from_file(file, "%d\n"); | |
9641 | } | |
9642 | ||
9643 | static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe, | |
9644 | const char *kfunc_name, size_t offset, int pid) | |
ca304b40 RDT |
9645 | { |
9646 | struct perf_event_attr attr = {}; | |
9647 | char errmsg[STRERR_BUFSIZE]; | |
9648 | int type, pfd, err; | |
9649 | ||
46ed5fc3 | 9650 | err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset); |
ca304b40 | 9651 | if (err < 0) { |
46ed5fc3 AN |
9652 | pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n", |
9653 | kfunc_name, offset, | |
ca304b40 RDT |
9654 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
9655 | return err; | |
9656 | } | |
46ed5fc3 | 9657 | type = determine_kprobe_perf_type_legacy(probe_name, retprobe); |
ca304b40 | 9658 | if (type < 0) { |
46ed5fc3 AN |
9659 | pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n", |
9660 | kfunc_name, offset, | |
ca304b40 RDT |
9661 | libbpf_strerror_r(type, errmsg, sizeof(errmsg))); |
9662 | return type; | |
9663 | } | |
9664 | attr.size = sizeof(attr); | |
9665 | attr.config = type; | |
9666 | attr.type = PERF_TYPE_TRACEPOINT; | |
9667 | ||
9668 | pfd = syscall(__NR_perf_event_open, &attr, | |
9669 | pid < 0 ? -1 : pid, /* pid */ | |
9670 | pid == -1 ? 0 : -1, /* cpu */ | |
9671 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
9672 | if (pfd < 0) { | |
9673 | err = -errno; | |
9674 | pr_warn("legacy kprobe perf_event_open() failed: %s\n", | |
9675 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
9676 | return err; | |
9677 | } | |
9678 | return pfd; | |
9679 | } | |
9680 | ||
da97553e | 9681 | struct bpf_link * |
942025c9 | 9682 | bpf_program__attach_kprobe_opts(const struct bpf_program *prog, |
ac0ed488 | 9683 | const char *func_name, |
47faff37 | 9684 | const struct bpf_kprobe_opts *opts) |
b2650027 | 9685 | { |
47faff37 | 9686 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
b2650027 | 9687 | char errmsg[STRERR_BUFSIZE]; |
ca304b40 | 9688 | char *legacy_probe = NULL; |
b2650027 | 9689 | struct bpf_link *link; |
46ed5fc3 | 9690 | size_t offset; |
ca304b40 | 9691 | bool retprobe, legacy; |
b2650027 AN |
9692 | int pfd, err; |
9693 | ||
da97553e JO |
9694 | if (!OPTS_VALID(opts, bpf_kprobe_opts)) |
9695 | return libbpf_err_ptr(-EINVAL); | |
9696 | ||
9697 | retprobe = OPTS_GET(opts, retprobe, false); | |
9698 | offset = OPTS_GET(opts, offset, 0); | |
47faff37 | 9699 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
da97553e | 9700 | |
ca304b40 RDT |
9701 | legacy = determine_kprobe_perf_type() < 0; |
9702 | if (!legacy) { | |
9703 | pfd = perf_event_open_probe(false /* uprobe */, retprobe, | |
9704 | func_name, offset, | |
9705 | -1 /* pid */, 0 /* ref_ctr_off */); | |
9706 | } else { | |
46ed5fc3 AN |
9707 | char probe_name[256]; |
9708 | ||
9709 | gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
9710 | func_name, offset); | |
9711 | ||
ca304b40 RDT |
9712 | legacy_probe = strdup(func_name); |
9713 | if (!legacy_probe) | |
9714 | return libbpf_err_ptr(-ENOMEM); | |
9715 | ||
46ed5fc3 | 9716 | pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name, |
ca304b40 RDT |
9717 | offset, -1 /* pid */); |
9718 | } | |
b2650027 | 9719 | if (pfd < 0) { |
46ed5fc3 AN |
9720 | err = -errno; |
9721 | pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n", | |
9722 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
9723 | func_name, offset, | |
303a2572 AN |
9724 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
9725 | goto err_out; | |
b2650027 | 9726 | } |
47faff37 | 9727 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
9728 | err = libbpf_get_error(link); |
9729 | if (err) { | |
b2650027 | 9730 | close(pfd); |
46ed5fc3 AN |
9731 | pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n", |
9732 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
9733 | func_name, offset, | |
be18010e | 9734 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
303a2572 | 9735 | goto err_out; |
b2650027 | 9736 | } |
ca304b40 RDT |
9737 | if (legacy) { |
9738 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
9739 | ||
9740 | perf_link->legacy_probe_name = legacy_probe; | |
46ed5fc3 | 9741 | perf_link->legacy_is_kprobe = true; |
ca304b40 RDT |
9742 | perf_link->legacy_is_retprobe = retprobe; |
9743 | } | |
9744 | ||
b2650027 | 9745 | return link; |
303a2572 AN |
9746 | err_out: |
9747 | free(legacy_probe); | |
9748 | return libbpf_err_ptr(err); | |
b2650027 AN |
9749 | } |
9750 | ||
942025c9 | 9751 | struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog, |
ac0ed488 JO |
9752 | bool retprobe, |
9753 | const char *func_name) | |
9754 | { | |
da97553e | 9755 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts, |
ac0ed488 | 9756 | .retprobe = retprobe, |
da97553e | 9757 | ); |
ac0ed488 JO |
9758 | |
9759 | return bpf_program__attach_kprobe_opts(prog, func_name, &opts); | |
9760 | } | |
9761 | ||
12d9466d | 9762 | static struct bpf_link *attach_kprobe(const struct bpf_program *prog, long cookie) |
d7a18ea7 | 9763 | { |
da97553e | 9764 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts); |
a2488b5f AM |
9765 | unsigned long offset = 0; |
9766 | struct bpf_link *link; | |
d7a18ea7 | 9767 | const char *func_name; |
a2488b5f AM |
9768 | char *func; |
9769 | int n, err; | |
d7a18ea7 | 9770 | |
13d35a0c AN |
9771 | opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/"); |
9772 | if (opts.retprobe) | |
9773 | func_name = prog->sec_name + sizeof("kretprobe/") - 1; | |
9774 | else | |
9775 | func_name = prog->sec_name + sizeof("kprobe/") - 1; | |
d7a18ea7 | 9776 | |
e3f9bc35 | 9777 | n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset); |
a2488b5f AM |
9778 | if (n < 1) { |
9779 | err = -EINVAL; | |
9780 | pr_warn("kprobe name is invalid: %s\n", func_name); | |
9781 | return libbpf_err_ptr(err); | |
9782 | } | |
9783 | if (opts.retprobe && offset != 0) { | |
1f71a468 | 9784 | free(func); |
a2488b5f AM |
9785 | err = -EINVAL; |
9786 | pr_warn("kretprobes do not support offset specification\n"); | |
9787 | return libbpf_err_ptr(err); | |
9788 | } | |
d7a18ea7 | 9789 | |
a2488b5f AM |
9790 | opts.offset = offset; |
9791 | link = bpf_program__attach_kprobe_opts(prog, func, &opts); | |
9792 | free(func); | |
9793 | return link; | |
d7a18ea7 AN |
9794 | } |
9795 | ||
cc10623c AN |
9796 | static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz, |
9797 | const char *binary_path, uint64_t offset) | |
9798 | { | |
9799 | int i; | |
9800 | ||
9801 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset); | |
9802 | ||
9803 | /* sanitize binary_path in the probe name */ | |
9804 | for (i = 0; buf[i]; i++) { | |
9805 | if (!isalnum(buf[i])) | |
9806 | buf[i] = '_'; | |
9807 | } | |
9808 | } | |
9809 | ||
9810 | static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe, | |
9811 | const char *binary_path, size_t offset) | |
9812 | { | |
9813 | const char *file = "/sys/kernel/debug/tracing/uprobe_events"; | |
9814 | ||
9815 | return append_to_file(file, "%c:%s/%s %s:0x%zx", | |
9816 | retprobe ? 'r' : 'p', | |
9817 | retprobe ? "uretprobes" : "uprobes", | |
9818 | probe_name, binary_path, offset); | |
9819 | } | |
9820 | ||
9821 | static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe) | |
9822 | { | |
9823 | const char *file = "/sys/kernel/debug/tracing/uprobe_events"; | |
9824 | ||
9825 | return append_to_file(file, "-:%s/%s", retprobe ? "uretprobes" : "uprobes", probe_name); | |
9826 | } | |
9827 | ||
9828 | static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
9829 | { | |
9830 | char file[512]; | |
9831 | ||
9832 | snprintf(file, sizeof(file), | |
9833 | "/sys/kernel/debug/tracing/events/%s/%s/id", | |
9834 | retprobe ? "uretprobes" : "uprobes", probe_name); | |
9835 | ||
9836 | return parse_uint_from_file(file, "%d\n"); | |
9837 | } | |
9838 | ||
9839 | static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe, | |
9840 | const char *binary_path, size_t offset, int pid) | |
9841 | { | |
9842 | struct perf_event_attr attr; | |
9843 | int type, pfd, err; | |
9844 | ||
9845 | err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset); | |
9846 | if (err < 0) { | |
9847 | pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n", | |
9848 | binary_path, (size_t)offset, err); | |
9849 | return err; | |
9850 | } | |
9851 | type = determine_uprobe_perf_type_legacy(probe_name, retprobe); | |
9852 | if (type < 0) { | |
9853 | pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n", | |
9854 | binary_path, offset, err); | |
9855 | return type; | |
9856 | } | |
9857 | ||
9858 | memset(&attr, 0, sizeof(attr)); | |
9859 | attr.size = sizeof(attr); | |
9860 | attr.config = type; | |
9861 | attr.type = PERF_TYPE_TRACEPOINT; | |
9862 | ||
9863 | pfd = syscall(__NR_perf_event_open, &attr, | |
9864 | pid < 0 ? -1 : pid, /* pid */ | |
9865 | pid == -1 ? 0 : -1, /* cpu */ | |
9866 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
9867 | if (pfd < 0) { | |
9868 | err = -errno; | |
9869 | pr_warn("legacy uprobe perf_event_open() failed: %d\n", err); | |
9870 | return err; | |
9871 | } | |
9872 | return pfd; | |
9873 | } | |
9874 | ||
47faff37 | 9875 | LIBBPF_API struct bpf_link * |
942025c9 | 9876 | bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid, |
47faff37 AN |
9877 | const char *binary_path, size_t func_offset, |
9878 | const struct bpf_uprobe_opts *opts) | |
b2650027 | 9879 | { |
47faff37 | 9880 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
cc10623c | 9881 | char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL; |
b2650027 | 9882 | struct bpf_link *link; |
5e3b8356 | 9883 | size_t ref_ctr_off; |
b2650027 | 9884 | int pfd, err; |
cc10623c | 9885 | bool retprobe, legacy; |
47faff37 AN |
9886 | |
9887 | if (!OPTS_VALID(opts, bpf_uprobe_opts)) | |
9888 | return libbpf_err_ptr(-EINVAL); | |
9889 | ||
9890 | retprobe = OPTS_GET(opts, retprobe, false); | |
5e3b8356 | 9891 | ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0); |
47faff37 | 9892 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
b2650027 | 9893 | |
cc10623c AN |
9894 | legacy = determine_uprobe_perf_type() < 0; |
9895 | if (!legacy) { | |
9896 | pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path, | |
9897 | func_offset, pid, ref_ctr_off); | |
9898 | } else { | |
9899 | char probe_name[512]; | |
9900 | ||
9901 | if (ref_ctr_off) | |
9902 | return libbpf_err_ptr(-EINVAL); | |
9903 | ||
9904 | gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
9905 | binary_path, func_offset); | |
9906 | ||
9907 | legacy_probe = strdup(probe_name); | |
9908 | if (!legacy_probe) | |
9909 | return libbpf_err_ptr(-ENOMEM); | |
9910 | ||
9911 | pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe, | |
9912 | binary_path, func_offset, pid); | |
9913 | } | |
b2650027 | 9914 | if (pfd < 0) { |
cc10623c | 9915 | err = -errno; |
52109584 AN |
9916 | pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n", |
9917 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e | 9918 | binary_path, func_offset, |
cc10623c AN |
9919 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
9920 | goto err_out; | |
b2650027 | 9921 | } |
cc10623c | 9922 | |
47faff37 | 9923 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
9924 | err = libbpf_get_error(link); |
9925 | if (err) { | |
b2650027 | 9926 | close(pfd); |
52109584 AN |
9927 | pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n", |
9928 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e KW |
9929 | binary_path, func_offset, |
9930 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
cc10623c AN |
9931 | goto err_out; |
9932 | } | |
9933 | if (legacy) { | |
9934 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
9935 | ||
9936 | perf_link->legacy_probe_name = legacy_probe; | |
9937 | perf_link->legacy_is_kprobe = false; | |
9938 | perf_link->legacy_is_retprobe = retprobe; | |
b2650027 AN |
9939 | } |
9940 | return link; | |
cc10623c AN |
9941 | err_out: |
9942 | free(legacy_probe); | |
9943 | return libbpf_err_ptr(err); | |
9944 | ||
b2650027 AN |
9945 | } |
9946 | ||
942025c9 | 9947 | struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog, |
47faff37 AN |
9948 | bool retprobe, pid_t pid, |
9949 | const char *binary_path, | |
9950 | size_t func_offset) | |
9951 | { | |
9952 | DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe); | |
9953 | ||
9954 | return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts); | |
9955 | } | |
9956 | ||
f6de59c1 AN |
9957 | static int determine_tracepoint_id(const char *tp_category, |
9958 | const char *tp_name) | |
9959 | { | |
9960 | char file[PATH_MAX]; | |
9961 | int ret; | |
9962 | ||
9963 | ret = snprintf(file, sizeof(file), | |
9964 | "/sys/kernel/debug/tracing/events/%s/%s/id", | |
9965 | tp_category, tp_name); | |
9966 | if (ret < 0) | |
9967 | return -errno; | |
9968 | if (ret >= sizeof(file)) { | |
9969 | pr_debug("tracepoint %s/%s path is too long\n", | |
9970 | tp_category, tp_name); | |
9971 | return -E2BIG; | |
9972 | } | |
9973 | return parse_uint_from_file(file, "%d\n"); | |
9974 | } | |
9975 | ||
9976 | static int perf_event_open_tracepoint(const char *tp_category, | |
9977 | const char *tp_name) | |
9978 | { | |
9979 | struct perf_event_attr attr = {}; | |
9980 | char errmsg[STRERR_BUFSIZE]; | |
9981 | int tp_id, pfd, err; | |
9982 | ||
9983 | tp_id = determine_tracepoint_id(tp_category, tp_name); | |
9984 | if (tp_id < 0) { | |
be18010e KW |
9985 | pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n", |
9986 | tp_category, tp_name, | |
9987 | libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
9988 | return tp_id; |
9989 | } | |
9990 | ||
9991 | attr.type = PERF_TYPE_TRACEPOINT; | |
9992 | attr.size = sizeof(attr); | |
9993 | attr.config = tp_id; | |
9994 | ||
9995 | pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */, | |
9996 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
9997 | if (pfd < 0) { | |
9998 | err = -errno; | |
be18010e KW |
9999 | pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n", |
10000 | tp_category, tp_name, | |
10001 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
10002 | return err; |
10003 | } | |
10004 | return pfd; | |
10005 | } | |
10006 | ||
942025c9 | 10007 | struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog, |
47faff37 AN |
10008 | const char *tp_category, |
10009 | const char *tp_name, | |
10010 | const struct bpf_tracepoint_opts *opts) | |
f6de59c1 | 10011 | { |
47faff37 | 10012 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f6de59c1 AN |
10013 | char errmsg[STRERR_BUFSIZE]; |
10014 | struct bpf_link *link; | |
10015 | int pfd, err; | |
10016 | ||
47faff37 AN |
10017 | if (!OPTS_VALID(opts, bpf_tracepoint_opts)) |
10018 | return libbpf_err_ptr(-EINVAL); | |
10019 | ||
10020 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); | |
10021 | ||
f6de59c1 AN |
10022 | pfd = perf_event_open_tracepoint(tp_category, tp_name); |
10023 | if (pfd < 0) { | |
52109584 AN |
10024 | pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n", |
10025 | prog->name, tp_category, tp_name, | |
be18010e | 10026 | libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 10027 | return libbpf_err_ptr(pfd); |
f6de59c1 | 10028 | } |
47faff37 | 10029 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
10030 | err = libbpf_get_error(link); |
10031 | if (err) { | |
f6de59c1 | 10032 | close(pfd); |
52109584 AN |
10033 | pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n", |
10034 | prog->name, tp_category, tp_name, | |
be18010e | 10035 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 10036 | return libbpf_err_ptr(err); |
f6de59c1 AN |
10037 | } |
10038 | return link; | |
10039 | } | |
10040 | ||
942025c9 | 10041 | struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog, |
47faff37 AN |
10042 | const char *tp_category, |
10043 | const char *tp_name) | |
10044 | { | |
10045 | return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL); | |
10046 | } | |
10047 | ||
12d9466d | 10048 | static struct bpf_link *attach_tp(const struct bpf_program *prog, long cookie) |
d7a18ea7 AN |
10049 | { |
10050 | char *sec_name, *tp_cat, *tp_name; | |
10051 | struct bpf_link *link; | |
10052 | ||
52109584 | 10053 | sec_name = strdup(prog->sec_name); |
d7a18ea7 | 10054 | if (!sec_name) |
e9fc3ce9 | 10055 | return libbpf_err_ptr(-ENOMEM); |
d7a18ea7 | 10056 | |
13d35a0c AN |
10057 | /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */ |
10058 | if (str_has_pfx(prog->sec_name, "tp/")) | |
10059 | tp_cat = sec_name + sizeof("tp/") - 1; | |
10060 | else | |
10061 | tp_cat = sec_name + sizeof("tracepoint/") - 1; | |
d7a18ea7 AN |
10062 | tp_name = strchr(tp_cat, '/'); |
10063 | if (!tp_name) { | |
e9fc3ce9 AN |
10064 | free(sec_name); |
10065 | return libbpf_err_ptr(-EINVAL); | |
d7a18ea7 AN |
10066 | } |
10067 | *tp_name = '\0'; | |
10068 | tp_name++; | |
10069 | ||
10070 | link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name); | |
d7a18ea7 AN |
10071 | free(sec_name); |
10072 | return link; | |
10073 | } | |
10074 | ||
942025c9 | 10075 | struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog, |
84bf5e1f AN |
10076 | const char *tp_name) |
10077 | { | |
10078 | char errmsg[STRERR_BUFSIZE]; | |
c016b68e | 10079 | struct bpf_link *link; |
84bf5e1f AN |
10080 | int prog_fd, pfd; |
10081 | ||
10082 | prog_fd = bpf_program__fd(prog); | |
10083 | if (prog_fd < 0) { | |
52109584 | 10084 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 10085 | return libbpf_err_ptr(-EINVAL); |
84bf5e1f AN |
10086 | } |
10087 | ||
d6958706 | 10088 | link = calloc(1, sizeof(*link)); |
84bf5e1f | 10089 | if (!link) |
e9fc3ce9 | 10090 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 10091 | link->detach = &bpf_link__detach_fd; |
84bf5e1f AN |
10092 | |
10093 | pfd = bpf_raw_tracepoint_open(tp_name, prog_fd); | |
10094 | if (pfd < 0) { | |
10095 | pfd = -errno; | |
10096 | free(link); | |
52109584 AN |
10097 | pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n", |
10098 | prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 10099 | return libbpf_err_ptr(pfd); |
84bf5e1f AN |
10100 | } |
10101 | link->fd = pfd; | |
c016b68e | 10102 | return link; |
84bf5e1f AN |
10103 | } |
10104 | ||
12d9466d | 10105 | static struct bpf_link *attach_raw_tp(const struct bpf_program *prog, long cookie) |
d7a18ea7 | 10106 | { |
ccaf12d6 HT |
10107 | static const char *const prefixes[] = { |
10108 | "raw_tp/", | |
10109 | "raw_tracepoint/", | |
10110 | "raw_tp.w/", | |
10111 | "raw_tracepoint.w/", | |
10112 | }; | |
10113 | size_t i; | |
10114 | const char *tp_name = NULL; | |
13d35a0c | 10115 | |
ccaf12d6 HT |
10116 | for (i = 0; i < ARRAY_SIZE(prefixes); i++) { |
10117 | if (str_has_pfx(prog->sec_name, prefixes[i])) { | |
10118 | tp_name = prog->sec_name + strlen(prefixes[i]); | |
10119 | break; | |
10120 | } | |
10121 | } | |
10122 | if (!tp_name) { | |
10123 | pr_warn("prog '%s': invalid section name '%s'\n", | |
10124 | prog->name, prog->sec_name); | |
10125 | return libbpf_err_ptr(-EINVAL); | |
10126 | } | |
d7a18ea7 AN |
10127 | |
10128 | return bpf_program__attach_raw_tracepoint(prog, tp_name); | |
10129 | } | |
10130 | ||
1e092a03 | 10131 | /* Common logic for all BPF program types that attach to a btf_id */ |
942025c9 | 10132 | static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog) |
b8c54ea4 AS |
10133 | { |
10134 | char errmsg[STRERR_BUFSIZE]; | |
c016b68e | 10135 | struct bpf_link *link; |
b8c54ea4 AS |
10136 | int prog_fd, pfd; |
10137 | ||
10138 | prog_fd = bpf_program__fd(prog); | |
10139 | if (prog_fd < 0) { | |
52109584 | 10140 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 10141 | return libbpf_err_ptr(-EINVAL); |
b8c54ea4 AS |
10142 | } |
10143 | ||
d6958706 | 10144 | link = calloc(1, sizeof(*link)); |
b8c54ea4 | 10145 | if (!link) |
e9fc3ce9 | 10146 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 10147 | link->detach = &bpf_link__detach_fd; |
b8c54ea4 AS |
10148 | |
10149 | pfd = bpf_raw_tracepoint_open(NULL, prog_fd); | |
10150 | if (pfd < 0) { | |
10151 | pfd = -errno; | |
10152 | free(link); | |
52109584 AN |
10153 | pr_warn("prog '%s': failed to attach: %s\n", |
10154 | prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 10155 | return libbpf_err_ptr(pfd); |
b8c54ea4 AS |
10156 | } |
10157 | link->fd = pfd; | |
10158 | return (struct bpf_link *)link; | |
10159 | } | |
10160 | ||
942025c9 | 10161 | struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog) |
1e092a03 KS |
10162 | { |
10163 | return bpf_program__attach_btf_id(prog); | |
10164 | } | |
10165 | ||
942025c9 | 10166 | struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog) |
1e092a03 KS |
10167 | { |
10168 | return bpf_program__attach_btf_id(prog); | |
10169 | } | |
10170 | ||
12d9466d | 10171 | static struct bpf_link *attach_trace(const struct bpf_program *prog, long cookie) |
d7a18ea7 AN |
10172 | { |
10173 | return bpf_program__attach_trace(prog); | |
10174 | } | |
10175 | ||
12d9466d | 10176 | static struct bpf_link *attach_lsm(const struct bpf_program *prog, long cookie) |
1e092a03 KS |
10177 | { |
10178 | return bpf_program__attach_lsm(prog); | |
10179 | } | |
10180 | ||
d60d81ac | 10181 | static struct bpf_link * |
942025c9 | 10182 | bpf_program__attach_fd(const struct bpf_program *prog, int target_fd, int btf_id, |
d60d81ac | 10183 | const char *target_name) |
cc4f864b | 10184 | { |
a5359091 THJ |
10185 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts, |
10186 | .target_btf_id = btf_id); | |
cc4f864b AN |
10187 | enum bpf_attach_type attach_type; |
10188 | char errmsg[STRERR_BUFSIZE]; | |
10189 | struct bpf_link *link; | |
10190 | int prog_fd, link_fd; | |
10191 | ||
10192 | prog_fd = bpf_program__fd(prog); | |
10193 | if (prog_fd < 0) { | |
52109584 | 10194 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 10195 | return libbpf_err_ptr(-EINVAL); |
cc4f864b AN |
10196 | } |
10197 | ||
10198 | link = calloc(1, sizeof(*link)); | |
10199 | if (!link) | |
e9fc3ce9 | 10200 | return libbpf_err_ptr(-ENOMEM); |
cc4f864b AN |
10201 | link->detach = &bpf_link__detach_fd; |
10202 | ||
10203 | attach_type = bpf_program__get_expected_attach_type(prog); | |
a5359091 | 10204 | link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts); |
cc4f864b AN |
10205 | if (link_fd < 0) { |
10206 | link_fd = -errno; | |
10207 | free(link); | |
52109584 AN |
10208 | pr_warn("prog '%s': failed to attach to %s: %s\n", |
10209 | prog->name, target_name, | |
cc4f864b | 10210 | libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 10211 | return libbpf_err_ptr(link_fd); |
cc4f864b AN |
10212 | } |
10213 | link->fd = link_fd; | |
10214 | return link; | |
10215 | } | |
10216 | ||
d60d81ac | 10217 | struct bpf_link * |
942025c9 | 10218 | bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd) |
d60d81ac | 10219 | { |
a5359091 | 10220 | return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup"); |
d60d81ac JS |
10221 | } |
10222 | ||
10223 | struct bpf_link * | |
942025c9 | 10224 | bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd) |
d60d81ac | 10225 | { |
a5359091 | 10226 | return bpf_program__attach_fd(prog, netns_fd, 0, "netns"); |
d60d81ac JS |
10227 | } |
10228 | ||
942025c9 | 10229 | struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex) |
dc8698ca AN |
10230 | { |
10231 | /* target_fd/target_ifindex use the same field in LINK_CREATE */ | |
a5359091 THJ |
10232 | return bpf_program__attach_fd(prog, ifindex, 0, "xdp"); |
10233 | } | |
10234 | ||
942025c9 | 10235 | struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog, |
a5359091 THJ |
10236 | int target_fd, |
10237 | const char *attach_func_name) | |
10238 | { | |
10239 | int btf_id; | |
10240 | ||
10241 | if (!!target_fd != !!attach_func_name) { | |
10242 | pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n", | |
10243 | prog->name); | |
e9fc3ce9 | 10244 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
10245 | } |
10246 | ||
10247 | if (prog->type != BPF_PROG_TYPE_EXT) { | |
10248 | pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace", | |
10249 | prog->name); | |
e9fc3ce9 | 10250 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
10251 | } |
10252 | ||
10253 | if (target_fd) { | |
10254 | btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd); | |
10255 | if (btf_id < 0) | |
e9fc3ce9 | 10256 | return libbpf_err_ptr(btf_id); |
a5359091 THJ |
10257 | |
10258 | return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace"); | |
10259 | } else { | |
10260 | /* no target, so use raw_tracepoint_open for compatibility | |
10261 | * with old kernels | |
10262 | */ | |
10263 | return bpf_program__attach_trace(prog); | |
10264 | } | |
dc8698ca AN |
10265 | } |
10266 | ||
c09add2f | 10267 | struct bpf_link * |
942025c9 | 10268 | bpf_program__attach_iter(const struct bpf_program *prog, |
c09add2f YS |
10269 | const struct bpf_iter_attach_opts *opts) |
10270 | { | |
cd31039a | 10271 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); |
c09add2f YS |
10272 | char errmsg[STRERR_BUFSIZE]; |
10273 | struct bpf_link *link; | |
10274 | int prog_fd, link_fd; | |
cd31039a | 10275 | __u32 target_fd = 0; |
c09add2f YS |
10276 | |
10277 | if (!OPTS_VALID(opts, bpf_iter_attach_opts)) | |
e9fc3ce9 | 10278 | return libbpf_err_ptr(-EINVAL); |
c09add2f | 10279 | |
74fc097d YS |
10280 | link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0); |
10281 | link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0); | |
cd31039a | 10282 | |
c09add2f YS |
10283 | prog_fd = bpf_program__fd(prog); |
10284 | if (prog_fd < 0) { | |
52109584 | 10285 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 10286 | return libbpf_err_ptr(-EINVAL); |
c09add2f YS |
10287 | } |
10288 | ||
10289 | link = calloc(1, sizeof(*link)); | |
10290 | if (!link) | |
e9fc3ce9 | 10291 | return libbpf_err_ptr(-ENOMEM); |
c09add2f YS |
10292 | link->detach = &bpf_link__detach_fd; |
10293 | ||
cd31039a YS |
10294 | link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER, |
10295 | &link_create_opts); | |
c09add2f YS |
10296 | if (link_fd < 0) { |
10297 | link_fd = -errno; | |
10298 | free(link); | |
52109584 AN |
10299 | pr_warn("prog '%s': failed to attach to iterator: %s\n", |
10300 | prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 10301 | return libbpf_err_ptr(link_fd); |
c09add2f YS |
10302 | } |
10303 | link->fd = link_fd; | |
10304 | return link; | |
10305 | } | |
10306 | ||
12d9466d | 10307 | static struct bpf_link *attach_iter(const struct bpf_program *prog, long cookie) |
e9fc3ce9 AN |
10308 | { |
10309 | return bpf_program__attach_iter(prog, NULL); | |
10310 | } | |
10311 | ||
942025c9 | 10312 | struct bpf_link *bpf_program__attach(const struct bpf_program *prog) |
d7a18ea7 | 10313 | { |
5532dfd4 | 10314 | if (!prog->sec_def || !prog->sec_def->attach_fn) |
e9fc3ce9 | 10315 | return libbpf_err_ptr(-ESRCH); |
d7a18ea7 | 10316 | |
12d9466d | 10317 | return prog->sec_def->attach_fn(prog, prog->sec_def->cookie); |
d7a18ea7 AN |
10318 | } |
10319 | ||
590a0088 MKL |
10320 | static int bpf_link__detach_struct_ops(struct bpf_link *link) |
10321 | { | |
590a0088 MKL |
10322 | __u32 zero = 0; |
10323 | ||
c016b68e | 10324 | if (bpf_map_delete_elem(link->fd, &zero)) |
590a0088 MKL |
10325 | return -errno; |
10326 | ||
10327 | return 0; | |
10328 | } | |
10329 | ||
942025c9 | 10330 | struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map) |
590a0088 MKL |
10331 | { |
10332 | struct bpf_struct_ops *st_ops; | |
c016b68e | 10333 | struct bpf_link *link; |
590a0088 MKL |
10334 | __u32 i, zero = 0; |
10335 | int err; | |
10336 | ||
10337 | if (!bpf_map__is_struct_ops(map) || map->fd == -1) | |
e9fc3ce9 | 10338 | return libbpf_err_ptr(-EINVAL); |
590a0088 MKL |
10339 | |
10340 | link = calloc(1, sizeof(*link)); | |
10341 | if (!link) | |
e9fc3ce9 | 10342 | return libbpf_err_ptr(-EINVAL); |
590a0088 MKL |
10343 | |
10344 | st_ops = map->st_ops; | |
10345 | for (i = 0; i < btf_vlen(st_ops->type); i++) { | |
10346 | struct bpf_program *prog = st_ops->progs[i]; | |
10347 | void *kern_data; | |
10348 | int prog_fd; | |
10349 | ||
10350 | if (!prog) | |
10351 | continue; | |
10352 | ||
10353 | prog_fd = bpf_program__fd(prog); | |
10354 | kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i]; | |
10355 | *(unsigned long *)kern_data = prog_fd; | |
10356 | } | |
10357 | ||
10358 | err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0); | |
10359 | if (err) { | |
10360 | err = -errno; | |
10361 | free(link); | |
e9fc3ce9 | 10362 | return libbpf_err_ptr(err); |
590a0088 MKL |
10363 | } |
10364 | ||
c016b68e | 10365 | link->detach = bpf_link__detach_struct_ops; |
590a0088 MKL |
10366 | link->fd = map->fd; |
10367 | ||
c016b68e | 10368 | return link; |
590a0088 MKL |
10369 | } |
10370 | ||
d0cabbb0 | 10371 | enum bpf_perf_event_ret |
3dca2115 DB |
10372 | bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size, |
10373 | void **copy_mem, size_t *copy_size, | |
10374 | bpf_perf_event_print_t fn, void *private_data) | |
d0cabbb0 | 10375 | { |
3dca2115 | 10376 | struct perf_event_mmap_page *header = mmap_mem; |
a64af0ef | 10377 | __u64 data_head = ring_buffer_read_head(header); |
d0cabbb0 | 10378 | __u64 data_tail = header->data_tail; |
3dca2115 DB |
10379 | void *base = ((__u8 *)header) + page_size; |
10380 | int ret = LIBBPF_PERF_EVENT_CONT; | |
10381 | struct perf_event_header *ehdr; | |
10382 | size_t ehdr_size; | |
10383 | ||
10384 | while (data_head != data_tail) { | |
10385 | ehdr = base + (data_tail & (mmap_size - 1)); | |
10386 | ehdr_size = ehdr->size; | |
10387 | ||
10388 | if (((void *)ehdr) + ehdr_size > base + mmap_size) { | |
10389 | void *copy_start = ehdr; | |
10390 | size_t len_first = base + mmap_size - copy_start; | |
10391 | size_t len_secnd = ehdr_size - len_first; | |
10392 | ||
10393 | if (*copy_size < ehdr_size) { | |
10394 | free(*copy_mem); | |
10395 | *copy_mem = malloc(ehdr_size); | |
10396 | if (!*copy_mem) { | |
10397 | *copy_size = 0; | |
d0cabbb0 JK |
10398 | ret = LIBBPF_PERF_EVENT_ERROR; |
10399 | break; | |
10400 | } | |
3dca2115 | 10401 | *copy_size = ehdr_size; |
d0cabbb0 JK |
10402 | } |
10403 | ||
3dca2115 DB |
10404 | memcpy(*copy_mem, copy_start, len_first); |
10405 | memcpy(*copy_mem + len_first, base, len_secnd); | |
10406 | ehdr = *copy_mem; | |
d0cabbb0 JK |
10407 | } |
10408 | ||
3dca2115 DB |
10409 | ret = fn(ehdr, private_data); |
10410 | data_tail += ehdr_size; | |
d0cabbb0 JK |
10411 | if (ret != LIBBPF_PERF_EVENT_CONT) |
10412 | break; | |
d0cabbb0 JK |
10413 | } |
10414 | ||
a64af0ef | 10415 | ring_buffer_write_tail(header, data_tail); |
e9fc3ce9 | 10416 | return libbpf_err(ret); |
d0cabbb0 | 10417 | } |
34be1646 | 10418 | |
fb84b822 AN |
10419 | struct perf_buffer; |
10420 | ||
10421 | struct perf_buffer_params { | |
10422 | struct perf_event_attr *attr; | |
10423 | /* if event_cb is specified, it takes precendence */ | |
10424 | perf_buffer_event_fn event_cb; | |
10425 | /* sample_cb and lost_cb are higher-level common-case callbacks */ | |
10426 | perf_buffer_sample_fn sample_cb; | |
10427 | perf_buffer_lost_fn lost_cb; | |
10428 | void *ctx; | |
10429 | int cpu_cnt; | |
10430 | int *cpus; | |
10431 | int *map_keys; | |
10432 | }; | |
10433 | ||
10434 | struct perf_cpu_buf { | |
10435 | struct perf_buffer *pb; | |
10436 | void *base; /* mmap()'ed memory */ | |
10437 | void *buf; /* for reconstructing segmented data */ | |
10438 | size_t buf_size; | |
10439 | int fd; | |
10440 | int cpu; | |
10441 | int map_key; | |
10442 | }; | |
10443 | ||
10444 | struct perf_buffer { | |
10445 | perf_buffer_event_fn event_cb; | |
10446 | perf_buffer_sample_fn sample_cb; | |
10447 | perf_buffer_lost_fn lost_cb; | |
10448 | void *ctx; /* passed into callbacks */ | |
10449 | ||
10450 | size_t page_size; | |
10451 | size_t mmap_size; | |
10452 | struct perf_cpu_buf **cpu_bufs; | |
10453 | struct epoll_event *events; | |
783b8f01 | 10454 | int cpu_cnt; /* number of allocated CPU buffers */ |
fb84b822 AN |
10455 | int epoll_fd; /* perf event FD */ |
10456 | int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */ | |
10457 | }; | |
10458 | ||
10459 | static void perf_buffer__free_cpu_buf(struct perf_buffer *pb, | |
10460 | struct perf_cpu_buf *cpu_buf) | |
10461 | { | |
10462 | if (!cpu_buf) | |
10463 | return; | |
10464 | if (cpu_buf->base && | |
10465 | munmap(cpu_buf->base, pb->mmap_size + pb->page_size)) | |
be18010e | 10466 | pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu); |
fb84b822 AN |
10467 | if (cpu_buf->fd >= 0) { |
10468 | ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0); | |
10469 | close(cpu_buf->fd); | |
10470 | } | |
10471 | free(cpu_buf->buf); | |
10472 | free(cpu_buf); | |
10473 | } | |
10474 | ||
10475 | void perf_buffer__free(struct perf_buffer *pb) | |
10476 | { | |
10477 | int i; | |
10478 | ||
50450fc7 | 10479 | if (IS_ERR_OR_NULL(pb)) |
fb84b822 AN |
10480 | return; |
10481 | if (pb->cpu_bufs) { | |
601b05ca | 10482 | for (i = 0; i < pb->cpu_cnt; i++) { |
fb84b822 AN |
10483 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; |
10484 | ||
601b05ca EC |
10485 | if (!cpu_buf) |
10486 | continue; | |
10487 | ||
fb84b822 AN |
10488 | bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key); |
10489 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
10490 | } | |
10491 | free(pb->cpu_bufs); | |
10492 | } | |
10493 | if (pb->epoll_fd >= 0) | |
10494 | close(pb->epoll_fd); | |
10495 | free(pb->events); | |
10496 | free(pb); | |
10497 | } | |
10498 | ||
10499 | static struct perf_cpu_buf * | |
10500 | perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr, | |
10501 | int cpu, int map_key) | |
10502 | { | |
10503 | struct perf_cpu_buf *cpu_buf; | |
10504 | char msg[STRERR_BUFSIZE]; | |
10505 | int err; | |
10506 | ||
10507 | cpu_buf = calloc(1, sizeof(*cpu_buf)); | |
10508 | if (!cpu_buf) | |
10509 | return ERR_PTR(-ENOMEM); | |
10510 | ||
10511 | cpu_buf->pb = pb; | |
10512 | cpu_buf->cpu = cpu; | |
10513 | cpu_buf->map_key = map_key; | |
10514 | ||
10515 | cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu, | |
10516 | -1, PERF_FLAG_FD_CLOEXEC); | |
10517 | if (cpu_buf->fd < 0) { | |
10518 | err = -errno; | |
be18010e KW |
10519 | pr_warn("failed to open perf buffer event on cpu #%d: %s\n", |
10520 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10521 | goto error; |
10522 | } | |
10523 | ||
10524 | cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size, | |
10525 | PROT_READ | PROT_WRITE, MAP_SHARED, | |
10526 | cpu_buf->fd, 0); | |
10527 | if (cpu_buf->base == MAP_FAILED) { | |
10528 | cpu_buf->base = NULL; | |
10529 | err = -errno; | |
be18010e KW |
10530 | pr_warn("failed to mmap perf buffer on cpu #%d: %s\n", |
10531 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10532 | goto error; |
10533 | } | |
10534 | ||
10535 | if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
10536 | err = -errno; | |
be18010e KW |
10537 | pr_warn("failed to enable perf buffer event on cpu #%d: %s\n", |
10538 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10539 | goto error; |
10540 | } | |
10541 | ||
10542 | return cpu_buf; | |
10543 | ||
10544 | error: | |
10545 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
10546 | return (struct perf_cpu_buf *)ERR_PTR(err); | |
10547 | } | |
10548 | ||
10549 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
10550 | struct perf_buffer_params *p); | |
10551 | ||
10552 | struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt, | |
10553 | const struct perf_buffer_opts *opts) | |
10554 | { | |
10555 | struct perf_buffer_params p = {}; | |
4be6e05c ACM |
10556 | struct perf_event_attr attr = { 0, }; |
10557 | ||
65bb2e0f | 10558 | attr.config = PERF_COUNT_SW_BPF_OUTPUT; |
4be6e05c ACM |
10559 | attr.type = PERF_TYPE_SOFTWARE; |
10560 | attr.sample_type = PERF_SAMPLE_RAW; | |
10561 | attr.sample_period = 1; | |
10562 | attr.wakeup_events = 1; | |
fb84b822 AN |
10563 | |
10564 | p.attr = &attr; | |
10565 | p.sample_cb = opts ? opts->sample_cb : NULL; | |
10566 | p.lost_cb = opts ? opts->lost_cb : NULL; | |
10567 | p.ctx = opts ? opts->ctx : NULL; | |
10568 | ||
e9fc3ce9 | 10569 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
10570 | } |
10571 | ||
10572 | struct perf_buffer * | |
10573 | perf_buffer__new_raw(int map_fd, size_t page_cnt, | |
10574 | const struct perf_buffer_raw_opts *opts) | |
10575 | { | |
10576 | struct perf_buffer_params p = {}; | |
10577 | ||
10578 | p.attr = opts->attr; | |
10579 | p.event_cb = opts->event_cb; | |
10580 | p.ctx = opts->ctx; | |
10581 | p.cpu_cnt = opts->cpu_cnt; | |
10582 | p.cpus = opts->cpus; | |
10583 | p.map_keys = opts->map_keys; | |
10584 | ||
e9fc3ce9 | 10585 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
10586 | } |
10587 | ||
10588 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
10589 | struct perf_buffer_params *p) | |
10590 | { | |
783b8f01 | 10591 | const char *online_cpus_file = "/sys/devices/system/cpu/online"; |
0e289487 | 10592 | struct bpf_map_info map; |
fb84b822 AN |
10593 | char msg[STRERR_BUFSIZE]; |
10594 | struct perf_buffer *pb; | |
783b8f01 | 10595 | bool *online = NULL; |
fb84b822 | 10596 | __u32 map_info_len; |
783b8f01 | 10597 | int err, i, j, n; |
fb84b822 AN |
10598 | |
10599 | if (page_cnt & (page_cnt - 1)) { | |
be18010e KW |
10600 | pr_warn("page count should be power of two, but is %zu\n", |
10601 | page_cnt); | |
fb84b822 AN |
10602 | return ERR_PTR(-EINVAL); |
10603 | } | |
10604 | ||
0e289487 AN |
10605 | /* best-effort sanity checks */ |
10606 | memset(&map, 0, sizeof(map)); | |
fb84b822 AN |
10607 | map_info_len = sizeof(map); |
10608 | err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len); | |
10609 | if (err) { | |
10610 | err = -errno; | |
0e289487 AN |
10611 | /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return |
10612 | * -EBADFD, -EFAULT, or -E2BIG on real error | |
10613 | */ | |
10614 | if (err != -EINVAL) { | |
10615 | pr_warn("failed to get map info for map FD %d: %s\n", | |
10616 | map_fd, libbpf_strerror_r(err, msg, sizeof(msg))); | |
10617 | return ERR_PTR(err); | |
10618 | } | |
10619 | pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n", | |
10620 | map_fd); | |
10621 | } else { | |
10622 | if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) { | |
10623 | pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n", | |
10624 | map.name); | |
10625 | return ERR_PTR(-EINVAL); | |
10626 | } | |
fb84b822 AN |
10627 | } |
10628 | ||
10629 | pb = calloc(1, sizeof(*pb)); | |
10630 | if (!pb) | |
10631 | return ERR_PTR(-ENOMEM); | |
10632 | ||
10633 | pb->event_cb = p->event_cb; | |
10634 | pb->sample_cb = p->sample_cb; | |
10635 | pb->lost_cb = p->lost_cb; | |
10636 | pb->ctx = p->ctx; | |
10637 | ||
10638 | pb->page_size = getpagesize(); | |
10639 | pb->mmap_size = pb->page_size * page_cnt; | |
10640 | pb->map_fd = map_fd; | |
10641 | ||
10642 | pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
10643 | if (pb->epoll_fd < 0) { | |
10644 | err = -errno; | |
be18010e KW |
10645 | pr_warn("failed to create epoll instance: %s\n", |
10646 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10647 | goto error; |
10648 | } | |
10649 | ||
10650 | if (p->cpu_cnt > 0) { | |
10651 | pb->cpu_cnt = p->cpu_cnt; | |
10652 | } else { | |
10653 | pb->cpu_cnt = libbpf_num_possible_cpus(); | |
10654 | if (pb->cpu_cnt < 0) { | |
10655 | err = pb->cpu_cnt; | |
10656 | goto error; | |
10657 | } | |
0e289487 | 10658 | if (map.max_entries && map.max_entries < pb->cpu_cnt) |
fb84b822 AN |
10659 | pb->cpu_cnt = map.max_entries; |
10660 | } | |
10661 | ||
10662 | pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events)); | |
10663 | if (!pb->events) { | |
10664 | err = -ENOMEM; | |
be18010e | 10665 | pr_warn("failed to allocate events: out of memory\n"); |
fb84b822 AN |
10666 | goto error; |
10667 | } | |
10668 | pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs)); | |
10669 | if (!pb->cpu_bufs) { | |
10670 | err = -ENOMEM; | |
be18010e | 10671 | pr_warn("failed to allocate buffers: out of memory\n"); |
fb84b822 AN |
10672 | goto error; |
10673 | } | |
10674 | ||
783b8f01 AN |
10675 | err = parse_cpu_mask_file(online_cpus_file, &online, &n); |
10676 | if (err) { | |
10677 | pr_warn("failed to get online CPU mask: %d\n", err); | |
10678 | goto error; | |
10679 | } | |
10680 | ||
10681 | for (i = 0, j = 0; i < pb->cpu_cnt; i++) { | |
fb84b822 AN |
10682 | struct perf_cpu_buf *cpu_buf; |
10683 | int cpu, map_key; | |
10684 | ||
10685 | cpu = p->cpu_cnt > 0 ? p->cpus[i] : i; | |
10686 | map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i; | |
10687 | ||
783b8f01 AN |
10688 | /* in case user didn't explicitly requested particular CPUs to |
10689 | * be attached to, skip offline/not present CPUs | |
10690 | */ | |
10691 | if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu])) | |
10692 | continue; | |
10693 | ||
fb84b822 AN |
10694 | cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key); |
10695 | if (IS_ERR(cpu_buf)) { | |
10696 | err = PTR_ERR(cpu_buf); | |
10697 | goto error; | |
10698 | } | |
10699 | ||
783b8f01 | 10700 | pb->cpu_bufs[j] = cpu_buf; |
fb84b822 AN |
10701 | |
10702 | err = bpf_map_update_elem(pb->map_fd, &map_key, | |
10703 | &cpu_buf->fd, 0); | |
10704 | if (err) { | |
10705 | err = -errno; | |
be18010e KW |
10706 | pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n", |
10707 | cpu, map_key, cpu_buf->fd, | |
10708 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10709 | goto error; |
10710 | } | |
10711 | ||
783b8f01 AN |
10712 | pb->events[j].events = EPOLLIN; |
10713 | pb->events[j].data.ptr = cpu_buf; | |
fb84b822 | 10714 | if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd, |
783b8f01 | 10715 | &pb->events[j]) < 0) { |
fb84b822 | 10716 | err = -errno; |
be18010e KW |
10717 | pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n", |
10718 | cpu, cpu_buf->fd, | |
10719 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
10720 | goto error; |
10721 | } | |
783b8f01 | 10722 | j++; |
fb84b822 | 10723 | } |
783b8f01 AN |
10724 | pb->cpu_cnt = j; |
10725 | free(online); | |
fb84b822 AN |
10726 | |
10727 | return pb; | |
10728 | ||
10729 | error: | |
783b8f01 | 10730 | free(online); |
fb84b822 AN |
10731 | if (pb) |
10732 | perf_buffer__free(pb); | |
10733 | return ERR_PTR(err); | |
10734 | } | |
10735 | ||
10736 | struct perf_sample_raw { | |
10737 | struct perf_event_header header; | |
10738 | uint32_t size; | |
385bbf7b | 10739 | char data[]; |
fb84b822 AN |
10740 | }; |
10741 | ||
10742 | struct perf_sample_lost { | |
10743 | struct perf_event_header header; | |
10744 | uint64_t id; | |
10745 | uint64_t lost; | |
10746 | uint64_t sample_id; | |
10747 | }; | |
10748 | ||
10749 | static enum bpf_perf_event_ret | |
10750 | perf_buffer__process_record(struct perf_event_header *e, void *ctx) | |
10751 | { | |
10752 | struct perf_cpu_buf *cpu_buf = ctx; | |
10753 | struct perf_buffer *pb = cpu_buf->pb; | |
10754 | void *data = e; | |
10755 | ||
10756 | /* user wants full control over parsing perf event */ | |
10757 | if (pb->event_cb) | |
10758 | return pb->event_cb(pb->ctx, cpu_buf->cpu, e); | |
10759 | ||
10760 | switch (e->type) { | |
10761 | case PERF_RECORD_SAMPLE: { | |
10762 | struct perf_sample_raw *s = data; | |
10763 | ||
10764 | if (pb->sample_cb) | |
10765 | pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size); | |
10766 | break; | |
10767 | } | |
10768 | case PERF_RECORD_LOST: { | |
10769 | struct perf_sample_lost *s = data; | |
10770 | ||
10771 | if (pb->lost_cb) | |
10772 | pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost); | |
10773 | break; | |
10774 | } | |
10775 | default: | |
be18010e | 10776 | pr_warn("unknown perf sample type %d\n", e->type); |
fb84b822 AN |
10777 | return LIBBPF_PERF_EVENT_ERROR; |
10778 | } | |
10779 | return LIBBPF_PERF_EVENT_CONT; | |
10780 | } | |
10781 | ||
10782 | static int perf_buffer__process_records(struct perf_buffer *pb, | |
10783 | struct perf_cpu_buf *cpu_buf) | |
10784 | { | |
10785 | enum bpf_perf_event_ret ret; | |
10786 | ||
10787 | ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size, | |
10788 | pb->page_size, &cpu_buf->buf, | |
10789 | &cpu_buf->buf_size, | |
10790 | perf_buffer__process_record, cpu_buf); | |
10791 | if (ret != LIBBPF_PERF_EVENT_CONT) | |
10792 | return ret; | |
10793 | return 0; | |
10794 | } | |
10795 | ||
dca5612f AN |
10796 | int perf_buffer__epoll_fd(const struct perf_buffer *pb) |
10797 | { | |
10798 | return pb->epoll_fd; | |
10799 | } | |
10800 | ||
fb84b822 AN |
10801 | int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms) |
10802 | { | |
10803 | int i, cnt, err; | |
10804 | ||
10805 | cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms); | |
e9fc3ce9 | 10806 | if (cnt < 0) |
af0efa05 | 10807 | return -errno; |
e9fc3ce9 | 10808 | |
fb84b822 AN |
10809 | for (i = 0; i < cnt; i++) { |
10810 | struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr; | |
10811 | ||
10812 | err = perf_buffer__process_records(pb, cpu_buf); | |
10813 | if (err) { | |
be18010e | 10814 | pr_warn("error while processing records: %d\n", err); |
e9fc3ce9 | 10815 | return libbpf_err(err); |
fb84b822 AN |
10816 | } |
10817 | } | |
e9fc3ce9 | 10818 | return cnt; |
fb84b822 AN |
10819 | } |
10820 | ||
dca5612f AN |
10821 | /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer |
10822 | * manager. | |
10823 | */ | |
10824 | size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb) | |
10825 | { | |
10826 | return pb->cpu_cnt; | |
10827 | } | |
10828 | ||
10829 | /* | |
10830 | * Return perf_event FD of a ring buffer in *buf_idx* slot of | |
10831 | * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using | |
10832 | * select()/poll()/epoll() Linux syscalls. | |
10833 | */ | |
10834 | int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx) | |
10835 | { | |
10836 | struct perf_cpu_buf *cpu_buf; | |
10837 | ||
10838 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 10839 | return libbpf_err(-EINVAL); |
dca5612f AN |
10840 | |
10841 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
10842 | if (!cpu_buf) | |
e9fc3ce9 | 10843 | return libbpf_err(-ENOENT); |
dca5612f AN |
10844 | |
10845 | return cpu_buf->fd; | |
10846 | } | |
10847 | ||
10848 | /* | |
10849 | * Consume data from perf ring buffer corresponding to slot *buf_idx* in | |
10850 | * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to | |
10851 | * consume, do nothing and return success. | |
10852 | * Returns: | |
10853 | * - 0 on success; | |
10854 | * - <0 on failure. | |
10855 | */ | |
10856 | int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx) | |
10857 | { | |
10858 | struct perf_cpu_buf *cpu_buf; | |
10859 | ||
10860 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 10861 | return libbpf_err(-EINVAL); |
dca5612f AN |
10862 | |
10863 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
10864 | if (!cpu_buf) | |
e9fc3ce9 | 10865 | return libbpf_err(-ENOENT); |
dca5612f AN |
10866 | |
10867 | return perf_buffer__process_records(pb, cpu_buf); | |
10868 | } | |
10869 | ||
272d51af EC |
10870 | int perf_buffer__consume(struct perf_buffer *pb) |
10871 | { | |
10872 | int i, err; | |
10873 | ||
10874 | for (i = 0; i < pb->cpu_cnt; i++) { | |
10875 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; | |
10876 | ||
10877 | if (!cpu_buf) | |
10878 | continue; | |
10879 | ||
10880 | err = perf_buffer__process_records(pb, cpu_buf); | |
10881 | if (err) { | |
dca5612f | 10882 | pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err); |
e9fc3ce9 | 10883 | return libbpf_err(err); |
272d51af EC |
10884 | } |
10885 | } | |
10886 | return 0; | |
10887 | } | |
10888 | ||
34be1646 SL |
10889 | struct bpf_prog_info_array_desc { |
10890 | int array_offset; /* e.g. offset of jited_prog_insns */ | |
10891 | int count_offset; /* e.g. offset of jited_prog_len */ | |
10892 | int size_offset; /* > 0: offset of rec size, | |
10893 | * < 0: fix size of -size_offset | |
10894 | */ | |
10895 | }; | |
10896 | ||
10897 | static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = { | |
10898 | [BPF_PROG_INFO_JITED_INSNS] = { | |
10899 | offsetof(struct bpf_prog_info, jited_prog_insns), | |
10900 | offsetof(struct bpf_prog_info, jited_prog_len), | |
10901 | -1, | |
10902 | }, | |
10903 | [BPF_PROG_INFO_XLATED_INSNS] = { | |
10904 | offsetof(struct bpf_prog_info, xlated_prog_insns), | |
10905 | offsetof(struct bpf_prog_info, xlated_prog_len), | |
10906 | -1, | |
10907 | }, | |
10908 | [BPF_PROG_INFO_MAP_IDS] = { | |
10909 | offsetof(struct bpf_prog_info, map_ids), | |
10910 | offsetof(struct bpf_prog_info, nr_map_ids), | |
10911 | -(int)sizeof(__u32), | |
10912 | }, | |
10913 | [BPF_PROG_INFO_JITED_KSYMS] = { | |
10914 | offsetof(struct bpf_prog_info, jited_ksyms), | |
10915 | offsetof(struct bpf_prog_info, nr_jited_ksyms), | |
10916 | -(int)sizeof(__u64), | |
10917 | }, | |
10918 | [BPF_PROG_INFO_JITED_FUNC_LENS] = { | |
10919 | offsetof(struct bpf_prog_info, jited_func_lens), | |
10920 | offsetof(struct bpf_prog_info, nr_jited_func_lens), | |
10921 | -(int)sizeof(__u32), | |
10922 | }, | |
10923 | [BPF_PROG_INFO_FUNC_INFO] = { | |
10924 | offsetof(struct bpf_prog_info, func_info), | |
10925 | offsetof(struct bpf_prog_info, nr_func_info), | |
10926 | offsetof(struct bpf_prog_info, func_info_rec_size), | |
10927 | }, | |
10928 | [BPF_PROG_INFO_LINE_INFO] = { | |
10929 | offsetof(struct bpf_prog_info, line_info), | |
10930 | offsetof(struct bpf_prog_info, nr_line_info), | |
10931 | offsetof(struct bpf_prog_info, line_info_rec_size), | |
10932 | }, | |
10933 | [BPF_PROG_INFO_JITED_LINE_INFO] = { | |
10934 | offsetof(struct bpf_prog_info, jited_line_info), | |
10935 | offsetof(struct bpf_prog_info, nr_jited_line_info), | |
10936 | offsetof(struct bpf_prog_info, jited_line_info_rec_size), | |
10937 | }, | |
10938 | [BPF_PROG_INFO_PROG_TAGS] = { | |
10939 | offsetof(struct bpf_prog_info, prog_tags), | |
10940 | offsetof(struct bpf_prog_info, nr_prog_tags), | |
10941 | -(int)sizeof(__u8) * BPF_TAG_SIZE, | |
10942 | }, | |
10943 | ||
10944 | }; | |
10945 | ||
8983b731 AN |
10946 | static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, |
10947 | int offset) | |
34be1646 SL |
10948 | { |
10949 | __u32 *array = (__u32 *)info; | |
10950 | ||
10951 | if (offset >= 0) | |
10952 | return array[offset / sizeof(__u32)]; | |
10953 | return -(int)offset; | |
10954 | } | |
10955 | ||
8983b731 AN |
10956 | static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, |
10957 | int offset) | |
34be1646 SL |
10958 | { |
10959 | __u64 *array = (__u64 *)info; | |
10960 | ||
10961 | if (offset >= 0) | |
10962 | return array[offset / sizeof(__u64)]; | |
10963 | return -(int)offset; | |
10964 | } | |
10965 | ||
10966 | static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset, | |
10967 | __u32 val) | |
10968 | { | |
10969 | __u32 *array = (__u32 *)info; | |
10970 | ||
10971 | if (offset >= 0) | |
10972 | array[offset / sizeof(__u32)] = val; | |
10973 | } | |
10974 | ||
10975 | static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset, | |
10976 | __u64 val) | |
10977 | { | |
10978 | __u64 *array = (__u64 *)info; | |
10979 | ||
10980 | if (offset >= 0) | |
10981 | array[offset / sizeof(__u64)] = val; | |
10982 | } | |
10983 | ||
10984 | struct bpf_prog_info_linear * | |
10985 | bpf_program__get_prog_info_linear(int fd, __u64 arrays) | |
10986 | { | |
10987 | struct bpf_prog_info_linear *info_linear; | |
10988 | struct bpf_prog_info info = {}; | |
10989 | __u32 info_len = sizeof(info); | |
10990 | __u32 data_len = 0; | |
10991 | int i, err; | |
10992 | void *ptr; | |
10993 | ||
10994 | if (arrays >> BPF_PROG_INFO_LAST_ARRAY) | |
e9fc3ce9 | 10995 | return libbpf_err_ptr(-EINVAL); |
34be1646 SL |
10996 | |
10997 | /* step 1: get array dimensions */ | |
10998 | err = bpf_obj_get_info_by_fd(fd, &info, &info_len); | |
10999 | if (err) { | |
11000 | pr_debug("can't get prog info: %s", strerror(errno)); | |
e9fc3ce9 | 11001 | return libbpf_err_ptr(-EFAULT); |
34be1646 SL |
11002 | } |
11003 | ||
11004 | /* step 2: calculate total size of all arrays */ | |
11005 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
11006 | bool include_array = (arrays & (1UL << i)) > 0; | |
11007 | struct bpf_prog_info_array_desc *desc; | |
11008 | __u32 count, size; | |
11009 | ||
11010 | desc = bpf_prog_info_array_desc + i; | |
11011 | ||
11012 | /* kernel is too old to support this field */ | |
11013 | if (info_len < desc->array_offset + sizeof(__u32) || | |
11014 | info_len < desc->count_offset + sizeof(__u32) || | |
11015 | (desc->size_offset > 0 && info_len < desc->size_offset)) | |
11016 | include_array = false; | |
11017 | ||
11018 | if (!include_array) { | |
11019 | arrays &= ~(1UL << i); /* clear the bit */ | |
11020 | continue; | |
11021 | } | |
11022 | ||
11023 | count = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
11024 | size = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
11025 | ||
11026 | data_len += count * size; | |
11027 | } | |
11028 | ||
11029 | /* step 3: allocate continuous memory */ | |
11030 | data_len = roundup(data_len, sizeof(__u64)); | |
11031 | info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len); | |
11032 | if (!info_linear) | |
e9fc3ce9 | 11033 | return libbpf_err_ptr(-ENOMEM); |
34be1646 SL |
11034 | |
11035 | /* step 4: fill data to info_linear->info */ | |
11036 | info_linear->arrays = arrays; | |
11037 | memset(&info_linear->info, 0, sizeof(info)); | |
11038 | ptr = info_linear->data; | |
11039 | ||
11040 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
11041 | struct bpf_prog_info_array_desc *desc; | |
11042 | __u32 count, size; | |
11043 | ||
11044 | if ((arrays & (1UL << i)) == 0) | |
11045 | continue; | |
11046 | ||
11047 | desc = bpf_prog_info_array_desc + i; | |
11048 | count = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
11049 | size = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
11050 | bpf_prog_info_set_offset_u32(&info_linear->info, | |
11051 | desc->count_offset, count); | |
11052 | bpf_prog_info_set_offset_u32(&info_linear->info, | |
11053 | desc->size_offset, size); | |
11054 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
11055 | desc->array_offset, | |
11056 | ptr_to_u64(ptr)); | |
11057 | ptr += count * size; | |
11058 | } | |
11059 | ||
11060 | /* step 5: call syscall again to get required arrays */ | |
11061 | err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len); | |
11062 | if (err) { | |
11063 | pr_debug("can't get prog info: %s", strerror(errno)); | |
11064 | free(info_linear); | |
e9fc3ce9 | 11065 | return libbpf_err_ptr(-EFAULT); |
34be1646 SL |
11066 | } |
11067 | ||
11068 | /* step 6: verify the data */ | |
11069 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
11070 | struct bpf_prog_info_array_desc *desc; | |
11071 | __u32 v1, v2; | |
11072 | ||
11073 | if ((arrays & (1UL << i)) == 0) | |
11074 | continue; | |
11075 | ||
11076 | desc = bpf_prog_info_array_desc + i; | |
11077 | v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset); | |
11078 | v2 = bpf_prog_info_read_offset_u32(&info_linear->info, | |
11079 | desc->count_offset); | |
11080 | if (v1 != v2) | |
be18010e | 11081 | pr_warn("%s: mismatch in element count\n", __func__); |
34be1646 SL |
11082 | |
11083 | v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset); | |
11084 | v2 = bpf_prog_info_read_offset_u32(&info_linear->info, | |
11085 | desc->size_offset); | |
11086 | if (v1 != v2) | |
be18010e | 11087 | pr_warn("%s: mismatch in rec size\n", __func__); |
34be1646 SL |
11088 | } |
11089 | ||
11090 | /* step 7: update info_len and data_len */ | |
11091 | info_linear->info_len = sizeof(struct bpf_prog_info); | |
11092 | info_linear->data_len = data_len; | |
11093 | ||
11094 | return info_linear; | |
11095 | } | |
11096 | ||
11097 | void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear) | |
11098 | { | |
11099 | int i; | |
11100 | ||
11101 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
11102 | struct bpf_prog_info_array_desc *desc; | |
11103 | __u64 addr, offs; | |
11104 | ||
11105 | if ((info_linear->arrays & (1UL << i)) == 0) | |
11106 | continue; | |
11107 | ||
11108 | desc = bpf_prog_info_array_desc + i; | |
11109 | addr = bpf_prog_info_read_offset_u64(&info_linear->info, | |
11110 | desc->array_offset); | |
11111 | offs = addr - ptr_to_u64(info_linear->data); | |
11112 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
11113 | desc->array_offset, offs); | |
11114 | } | |
11115 | } | |
11116 | ||
11117 | void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear) | |
11118 | { | |
11119 | int i; | |
11120 | ||
11121 | for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) { | |
11122 | struct bpf_prog_info_array_desc *desc; | |
11123 | __u64 addr, offs; | |
11124 | ||
11125 | if ((info_linear->arrays & (1UL << i)) == 0) | |
11126 | continue; | |
11127 | ||
11128 | desc = bpf_prog_info_array_desc + i; | |
11129 | offs = bpf_prog_info_read_offset_u64(&info_linear->info, | |
11130 | desc->array_offset); | |
11131 | addr = offs + ptr_to_u64(info_linear->data); | |
11132 | bpf_prog_info_set_offset_u64(&info_linear->info, | |
11133 | desc->array_offset, addr); | |
11134 | } | |
11135 | } | |
6446b315 | 11136 | |
ff26ce5c EC |
11137 | int bpf_program__set_attach_target(struct bpf_program *prog, |
11138 | int attach_prog_fd, | |
11139 | const char *attach_func_name) | |
11140 | { | |
fe62de31 | 11141 | int btf_obj_fd = 0, btf_id = 0, err; |
ff26ce5c | 11142 | |
2d5ec1c6 | 11143 | if (!prog || attach_prog_fd < 0) |
e9fc3ce9 | 11144 | return libbpf_err(-EINVAL); |
ff26ce5c | 11145 | |
fe62de31 | 11146 | if (prog->obj->loaded) |
e9fc3ce9 | 11147 | return libbpf_err(-EINVAL); |
fe62de31 | 11148 | |
2d5ec1c6 AN |
11149 | if (attach_prog_fd && !attach_func_name) { |
11150 | /* remember attach_prog_fd and let bpf_program__load() find | |
11151 | * BTF ID during the program load | |
11152 | */ | |
11153 | prog->attach_prog_fd = attach_prog_fd; | |
11154 | return 0; | |
11155 | } | |
11156 | ||
fe62de31 | 11157 | if (attach_prog_fd) { |
ff26ce5c EC |
11158 | btf_id = libbpf_find_prog_btf_id(attach_func_name, |
11159 | attach_prog_fd); | |
fe62de31 | 11160 | if (btf_id < 0) |
e9fc3ce9 | 11161 | return libbpf_err(btf_id); |
fe62de31 | 11162 | } else { |
2d5ec1c6 AN |
11163 | if (!attach_func_name) |
11164 | return libbpf_err(-EINVAL); | |
11165 | ||
fe62de31 AN |
11166 | /* load btf_vmlinux, if not yet */ |
11167 | err = bpf_object__load_vmlinux_btf(prog->obj, true); | |
11168 | if (err) | |
e9fc3ce9 | 11169 | return libbpf_err(err); |
fe62de31 AN |
11170 | err = find_kernel_btf_id(prog->obj, attach_func_name, |
11171 | prog->expected_attach_type, | |
11172 | &btf_obj_fd, &btf_id); | |
11173 | if (err) | |
e9fc3ce9 | 11174 | return libbpf_err(err); |
fe62de31 | 11175 | } |
ff26ce5c EC |
11176 | |
11177 | prog->attach_btf_id = btf_id; | |
fe62de31 | 11178 | prog->attach_btf_obj_fd = btf_obj_fd; |
ff26ce5c EC |
11179 | prog->attach_prog_fd = attach_prog_fd; |
11180 | return 0; | |
11181 | } | |
11182 | ||
6803ee25 | 11183 | int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz) |
6446b315 | 11184 | { |
6803ee25 AN |
11185 | int err = 0, n, len, start, end = -1; |
11186 | bool *tmp; | |
6446b315 | 11187 | |
6803ee25 AN |
11188 | *mask = NULL; |
11189 | *mask_sz = 0; | |
11190 | ||
11191 | /* Each sub string separated by ',' has format \d+-\d+ or \d+ */ | |
11192 | while (*s) { | |
11193 | if (*s == ',' || *s == '\n') { | |
11194 | s++; | |
11195 | continue; | |
11196 | } | |
11197 | n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len); | |
11198 | if (n <= 0 || n > 2) { | |
11199 | pr_warn("Failed to get CPU range %s: %d\n", s, n); | |
11200 | err = -EINVAL; | |
11201 | goto cleanup; | |
11202 | } else if (n == 1) { | |
11203 | end = start; | |
11204 | } | |
11205 | if (start < 0 || start > end) { | |
11206 | pr_warn("Invalid CPU range [%d,%d] in %s\n", | |
11207 | start, end, s); | |
11208 | err = -EINVAL; | |
11209 | goto cleanup; | |
11210 | } | |
11211 | tmp = realloc(*mask, end + 1); | |
11212 | if (!tmp) { | |
11213 | err = -ENOMEM; | |
11214 | goto cleanup; | |
11215 | } | |
11216 | *mask = tmp; | |
11217 | memset(tmp + *mask_sz, 0, start - *mask_sz); | |
11218 | memset(tmp + start, 1, end - start + 1); | |
11219 | *mask_sz = end + 1; | |
11220 | s += len; | |
11221 | } | |
11222 | if (!*mask_sz) { | |
11223 | pr_warn("Empty CPU range\n"); | |
11224 | return -EINVAL; | |
11225 | } | |
11226 | return 0; | |
11227 | cleanup: | |
11228 | free(*mask); | |
11229 | *mask = NULL; | |
11230 | return err; | |
11231 | } | |
11232 | ||
11233 | int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz) | |
11234 | { | |
11235 | int fd, err = 0, len; | |
11236 | char buf[128]; | |
6446b315 HL |
11237 | |
11238 | fd = open(fcpu, O_RDONLY); | |
11239 | if (fd < 0) { | |
6803ee25 AN |
11240 | err = -errno; |
11241 | pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err); | |
11242 | return err; | |
6446b315 HL |
11243 | } |
11244 | len = read(fd, buf, sizeof(buf)); | |
11245 | close(fd); | |
11246 | if (len <= 0) { | |
6803ee25 AN |
11247 | err = len ? -errno : -EINVAL; |
11248 | pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err); | |
11249 | return err; | |
6446b315 | 11250 | } |
6803ee25 AN |
11251 | if (len >= sizeof(buf)) { |
11252 | pr_warn("CPU mask is too big in file %s\n", fcpu); | |
11253 | return -E2BIG; | |
6446b315 HL |
11254 | } |
11255 | buf[len] = '\0'; | |
11256 | ||
6803ee25 AN |
11257 | return parse_cpu_mask_str(buf, mask, mask_sz); |
11258 | } | |
11259 | ||
11260 | int libbpf_num_possible_cpus(void) | |
11261 | { | |
11262 | static const char *fcpu = "/sys/devices/system/cpu/possible"; | |
11263 | static int cpus; | |
11264 | int err, n, i, tmp_cpus; | |
11265 | bool *mask; | |
11266 | ||
11267 | tmp_cpus = READ_ONCE(cpus); | |
11268 | if (tmp_cpus > 0) | |
11269 | return tmp_cpus; | |
11270 | ||
11271 | err = parse_cpu_mask_file(fcpu, &mask, &n); | |
11272 | if (err) | |
e9fc3ce9 | 11273 | return libbpf_err(err); |
6803ee25 AN |
11274 | |
11275 | tmp_cpus = 0; | |
11276 | for (i = 0; i < n; i++) { | |
11277 | if (mask[i]) | |
11278 | tmp_cpus++; | |
6446b315 | 11279 | } |
6803ee25 | 11280 | free(mask); |
56fbc241 TC |
11281 | |
11282 | WRITE_ONCE(cpus, tmp_cpus); | |
11283 | return tmp_cpus; | |
6446b315 | 11284 | } |
d66562fb AN |
11285 | |
11286 | int bpf_object__open_skeleton(struct bpf_object_skeleton *s, | |
11287 | const struct bpf_object_open_opts *opts) | |
11288 | { | |
11289 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts, | |
11290 | .object_name = s->name, | |
11291 | ); | |
11292 | struct bpf_object *obj; | |
e9fc3ce9 | 11293 | int i, err; |
d66562fb AN |
11294 | |
11295 | /* Attempt to preserve opts->object_name, unless overriden by user | |
11296 | * explicitly. Overwriting object name for skeletons is discouraged, | |
11297 | * as it breaks global data maps, because they contain object name | |
11298 | * prefix as their own map name prefix. When skeleton is generated, | |
11299 | * bpftool is making an assumption that this name will stay the same. | |
11300 | */ | |
11301 | if (opts) { | |
11302 | memcpy(&skel_opts, opts, sizeof(*opts)); | |
11303 | if (!opts->object_name) | |
11304 | skel_opts.object_name = s->name; | |
11305 | } | |
11306 | ||
11307 | obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts); | |
e9fc3ce9 AN |
11308 | err = libbpf_get_error(obj); |
11309 | if (err) { | |
11310 | pr_warn("failed to initialize skeleton BPF object '%s': %d\n", | |
11311 | s->name, err); | |
11312 | return libbpf_err(err); | |
d66562fb AN |
11313 | } |
11314 | ||
11315 | *s->obj = obj; | |
11316 | ||
11317 | for (i = 0; i < s->map_cnt; i++) { | |
11318 | struct bpf_map **map = s->maps[i].map; | |
11319 | const char *name = s->maps[i].name; | |
11320 | void **mmaped = s->maps[i].mmaped; | |
11321 | ||
11322 | *map = bpf_object__find_map_by_name(obj, name); | |
11323 | if (!*map) { | |
11324 | pr_warn("failed to find skeleton map '%s'\n", name); | |
e9fc3ce9 | 11325 | return libbpf_err(-ESRCH); |
d66562fb AN |
11326 | } |
11327 | ||
2ad97d47 | 11328 | /* externs shouldn't be pre-setup from user code */ |
81bfdd08 | 11329 | if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG) |
d66562fb AN |
11330 | *mmaped = (*map)->mmaped; |
11331 | } | |
11332 | ||
11333 | for (i = 0; i < s->prog_cnt; i++) { | |
11334 | struct bpf_program **prog = s->progs[i].prog; | |
11335 | const char *name = s->progs[i].name; | |
11336 | ||
11337 | *prog = bpf_object__find_program_by_name(obj, name); | |
11338 | if (!*prog) { | |
11339 | pr_warn("failed to find skeleton program '%s'\n", name); | |
e9fc3ce9 | 11340 | return libbpf_err(-ESRCH); |
d66562fb AN |
11341 | } |
11342 | } | |
11343 | ||
11344 | return 0; | |
11345 | } | |
11346 | ||
11347 | int bpf_object__load_skeleton(struct bpf_object_skeleton *s) | |
11348 | { | |
11349 | int i, err; | |
11350 | ||
11351 | err = bpf_object__load(*s->obj); | |
11352 | if (err) { | |
11353 | pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err); | |
e9fc3ce9 | 11354 | return libbpf_err(err); |
d66562fb AN |
11355 | } |
11356 | ||
11357 | for (i = 0; i < s->map_cnt; i++) { | |
11358 | struct bpf_map *map = *s->maps[i].map; | |
11359 | size_t mmap_sz = bpf_map_mmap_sz(map); | |
11360 | int prot, map_fd = bpf_map__fd(map); | |
11361 | void **mmaped = s->maps[i].mmaped; | |
d66562fb AN |
11362 | |
11363 | if (!mmaped) | |
11364 | continue; | |
11365 | ||
11366 | if (!(map->def.map_flags & BPF_F_MMAPABLE)) { | |
11367 | *mmaped = NULL; | |
11368 | continue; | |
11369 | } | |
11370 | ||
11371 | if (map->def.map_flags & BPF_F_RDONLY_PROG) | |
11372 | prot = PROT_READ; | |
11373 | else | |
11374 | prot = PROT_READ | PROT_WRITE; | |
11375 | ||
11376 | /* Remap anonymous mmap()-ed "map initialization image" as | |
11377 | * a BPF map-backed mmap()-ed memory, but preserving the same | |
11378 | * memory address. This will cause kernel to change process' | |
11379 | * page table to point to a different piece of kernel memory, | |
11380 | * but from userspace point of view memory address (and its | |
11381 | * contents, being identical at this point) will stay the | |
11382 | * same. This mapping will be released by bpf_object__close() | |
11383 | * as per normal clean up procedure, so we don't need to worry | |
11384 | * about it from skeleton's clean up perspective. | |
11385 | */ | |
2ad97d47 AN |
11386 | *mmaped = mmap(map->mmaped, mmap_sz, prot, |
11387 | MAP_SHARED | MAP_FIXED, map_fd, 0); | |
11388 | if (*mmaped == MAP_FAILED) { | |
d66562fb AN |
11389 | err = -errno; |
11390 | *mmaped = NULL; | |
11391 | pr_warn("failed to re-mmap() map '%s': %d\n", | |
11392 | bpf_map__name(map), err); | |
e9fc3ce9 | 11393 | return libbpf_err(err); |
d66562fb AN |
11394 | } |
11395 | } | |
11396 | ||
11397 | return 0; | |
11398 | } | |
11399 | ||
11400 | int bpf_object__attach_skeleton(struct bpf_object_skeleton *s) | |
11401 | { | |
e9fc3ce9 | 11402 | int i, err; |
d66562fb AN |
11403 | |
11404 | for (i = 0; i < s->prog_cnt; i++) { | |
11405 | struct bpf_program *prog = *s->progs[i].prog; | |
11406 | struct bpf_link **link = s->progs[i].link; | |
d66562fb | 11407 | |
d9297581 AN |
11408 | if (!prog->load) |
11409 | continue; | |
11410 | ||
5532dfd4 AN |
11411 | /* auto-attaching not supported for this program */ |
11412 | if (!prog->sec_def || !prog->sec_def->attach_fn) | |
d66562fb AN |
11413 | continue; |
11414 | ||
942025c9 | 11415 | *link = bpf_program__attach(prog); |
e9fc3ce9 AN |
11416 | err = libbpf_get_error(*link); |
11417 | if (err) { | |
11418 | pr_warn("failed to auto-attach program '%s': %d\n", | |
11419 | bpf_program__name(prog), err); | |
11420 | return libbpf_err(err); | |
d66562fb AN |
11421 | } |
11422 | } | |
11423 | ||
11424 | return 0; | |
11425 | } | |
11426 | ||
11427 | void bpf_object__detach_skeleton(struct bpf_object_skeleton *s) | |
11428 | { | |
11429 | int i; | |
11430 | ||
11431 | for (i = 0; i < s->prog_cnt; i++) { | |
11432 | struct bpf_link **link = s->progs[i].link; | |
11433 | ||
50450fc7 | 11434 | bpf_link__destroy(*link); |
d66562fb AN |
11435 | *link = NULL; |
11436 | } | |
11437 | } | |
11438 | ||
11439 | void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s) | |
11440 | { | |
11441 | if (s->progs) | |
11442 | bpf_object__detach_skeleton(s); | |
11443 | if (s->obj) | |
11444 | bpf_object__close(*s->obj); | |
11445 | free(s->maps); | |
11446 | free(s->progs); | |
11447 | free(s); | |
11448 | } |