1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) 2 3 /* 4 * common eBPF ELF operations. 5 * 6 * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org> 7 * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com> 8 * Copyright (C) 2015 Huawei Inc. 9 * 10 * This program is free software; you can redistribute it and/or 11 * modify it under the terms of the GNU Lesser General Public 12 * License as published by the Free Software Foundation; 13 * version 2.1 of the License (not later!) 14 * 15 * This program is distributed in the hope that it will be useful, 16 * but WITHOUT ANY WARRANTY; without even the implied warranty of 17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 18 * GNU Lesser General Public License for more details. 19 * 20 * You should have received a copy of the GNU Lesser General Public 21 * License along with this program; if not, see <http://www.gnu.org/licenses> 22 */ 23 24 #include <stdlib.h> 25 #include <memory.h> 26 #include <unistd.h> 27 #include <asm/unistd.h> 28 #include <linux/bpf.h> 29 #include "bpf.h" 30 #include "libbpf.h" 31 #include <errno.h> 32 33 /* 34 * When building perf, unistd.h is overridden. __NR_bpf is 35 * required to be defined explicitly. 36 */ 37 #ifndef __NR_bpf 38 # if defined(__i386__) 39 # define __NR_bpf 357 40 # elif defined(__x86_64__) 41 # define __NR_bpf 321 42 # elif defined(__aarch64__) 43 # define __NR_bpf 280 44 # elif defined(__sparc__) 45 # define __NR_bpf 349 46 # elif defined(__s390__) 47 # define __NR_bpf 351 48 # else 49 # error __NR_bpf not defined. libbpf does not support your arch. 50 # endif 51 #endif 52 53 #ifndef min 54 #define min(x, y) ((x) < (y) ? (x) : (y)) 55 #endif 56 57 static inline __u64 ptr_to_u64(const void *ptr) 58 { 59 return (__u64) (unsigned long) ptr; 60 } 61 62 static inline int sys_bpf(enum bpf_cmd cmd, union bpf_attr *attr, 63 unsigned int size) 64 { 65 return syscall(__NR_bpf, cmd, attr, size); 66 } 67 68 int bpf_create_map_xattr(const struct bpf_create_map_attr *create_attr) 69 { 70 __u32 name_len = create_attr->name ? strlen(create_attr->name) : 0; 71 union bpf_attr attr; 72 73 memset(&attr, '\0', sizeof(attr)); 74 75 attr.map_type = create_attr->map_type; 76 attr.key_size = create_attr->key_size; 77 attr.value_size = create_attr->value_size; 78 attr.max_entries = create_attr->max_entries; 79 attr.map_flags = create_attr->map_flags; 80 memcpy(attr.map_name, create_attr->name, 81 min(name_len, BPF_OBJ_NAME_LEN - 1)); 82 attr.numa_node = create_attr->numa_node; 83 attr.btf_fd = create_attr->btf_fd; 84 attr.btf_key_type_id = create_attr->btf_key_type_id; 85 attr.btf_value_type_id = create_attr->btf_value_type_id; 86 attr.map_ifindex = create_attr->map_ifindex; 87 attr.inner_map_fd = create_attr->inner_map_fd; 88 89 return sys_bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); 90 } 91 92 int bpf_create_map_node(enum bpf_map_type map_type, const char *name, 93 int key_size, int value_size, int max_entries, 94 __u32 map_flags, int node) 95 { 96 struct bpf_create_map_attr map_attr = {}; 97 98 map_attr.name = name; 99 map_attr.map_type = map_type; 100 map_attr.map_flags = map_flags; 101 map_attr.key_size = key_size; 102 map_attr.value_size = value_size; 103 map_attr.max_entries = max_entries; 104 if (node >= 0) { 105 map_attr.numa_node = node; 106 map_attr.map_flags |= BPF_F_NUMA_NODE; 107 } 108 109 return bpf_create_map_xattr(&map_attr); 110 } 111 112 int bpf_create_map(enum bpf_map_type map_type, int key_size, 113 int value_size, int max_entries, __u32 map_flags) 114 { 115 struct bpf_create_map_attr map_attr = {}; 116 117 map_attr.map_type = map_type; 118 map_attr.map_flags = map_flags; 119 map_attr.key_size = key_size; 120 map_attr.value_size = value_size; 121 map_attr.max_entries = max_entries; 122 123 return bpf_create_map_xattr(&map_attr); 124 } 125 126 int bpf_create_map_name(enum bpf_map_type map_type, const char *name, 127 int key_size, int value_size, int max_entries, 128 __u32 map_flags) 129 { 130 struct bpf_create_map_attr map_attr = {}; 131 132 map_attr.name = name; 133 map_attr.map_type = map_type; 134 map_attr.map_flags = map_flags; 135 map_attr.key_size = key_size; 136 map_attr.value_size = value_size; 137 map_attr.max_entries = max_entries; 138 139 return bpf_create_map_xattr(&map_attr); 140 } 141 142 int bpf_create_map_in_map_node(enum bpf_map_type map_type, const char *name, 143 int key_size, int inner_map_fd, int max_entries, 144 __u32 map_flags, int node) 145 { 146 __u32 name_len = name ? strlen(name) : 0; 147 union bpf_attr attr; 148 149 memset(&attr, '\0', sizeof(attr)); 150 151 attr.map_type = map_type; 152 attr.key_size = key_size; 153 attr.value_size = 4; 154 attr.inner_map_fd = inner_map_fd; 155 attr.max_entries = max_entries; 156 attr.map_flags = map_flags; 157 memcpy(attr.map_name, name, min(name_len, BPF_OBJ_NAME_LEN - 1)); 158 159 if (node >= 0) { 160 attr.map_flags |= BPF_F_NUMA_NODE; 161 attr.numa_node = node; 162 } 163 164 return sys_bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); 165 } 166 167 int bpf_create_map_in_map(enum bpf_map_type map_type, const char *name, 168 int key_size, int inner_map_fd, int max_entries, 169 __u32 map_flags) 170 { 171 return bpf_create_map_in_map_node(map_type, name, key_size, 172 inner_map_fd, max_entries, map_flags, 173 -1); 174 } 175 176 int bpf_load_program_xattr(const struct bpf_load_program_attr *load_attr, 177 char *log_buf, size_t log_buf_sz) 178 { 179 union bpf_attr attr; 180 void *finfo = NULL; 181 __u32 name_len; 182 int fd; 183 184 if (!load_attr) 185 return -EINVAL; 186 187 name_len = load_attr->name ? strlen(load_attr->name) : 0; 188 189 bzero(&attr, sizeof(attr)); 190 attr.prog_type = load_attr->prog_type; 191 attr.expected_attach_type = load_attr->expected_attach_type; 192 attr.insn_cnt = (__u32)load_attr->insns_cnt; 193 attr.insns = ptr_to_u64(load_attr->insns); 194 attr.license = ptr_to_u64(load_attr->license); 195 attr.log_buf = ptr_to_u64(NULL); 196 attr.log_size = 0; 197 attr.log_level = 0; 198 attr.kern_version = load_attr->kern_version; 199 attr.prog_ifindex = load_attr->prog_ifindex; 200 attr.prog_btf_fd = load_attr->prog_btf_fd; 201 attr.func_info_rec_size = load_attr->func_info_rec_size; 202 attr.func_info_cnt = load_attr->func_info_cnt; 203 attr.func_info = ptr_to_u64(load_attr->func_info); 204 memcpy(attr.prog_name, load_attr->name, 205 min(name_len, BPF_OBJ_NAME_LEN - 1)); 206 207 fd = sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 208 if (fd >= 0 || !log_buf || !log_buf_sz) 209 return fd; 210 211 /* After bpf_prog_load, the kernel may modify certain attributes 212 * to give user space a hint how to deal with loading failure. 213 * Check to see whether we can make some changes and load again. 214 */ 215 if (errno == E2BIG && attr.func_info_cnt && 216 attr.func_info_rec_size < load_attr->func_info_rec_size) { 217 __u32 actual_rec_size = load_attr->func_info_rec_size; 218 __u32 expected_rec_size = attr.func_info_rec_size; 219 __u32 finfo_cnt = load_attr->func_info_cnt; 220 __u64 finfo_len = actual_rec_size * finfo_cnt; 221 const void *orecord; 222 void *nrecord; 223 int i; 224 225 finfo = malloc(finfo_len); 226 if (!finfo) 227 /* further try with log buffer won't help */ 228 return fd; 229 230 /* zero out bytes kernel does not understand */ 231 orecord = load_attr->func_info; 232 nrecord = finfo; 233 for (i = 0; i < load_attr->func_info_cnt; i++) { 234 memcpy(nrecord, orecord, expected_rec_size); 235 memset(nrecord + expected_rec_size, 0, 236 actual_rec_size - expected_rec_size); 237 orecord += actual_rec_size; 238 nrecord += actual_rec_size; 239 } 240 241 /* try with corrected func info records */ 242 attr.func_info = ptr_to_u64(finfo); 243 attr.func_info_rec_size = load_attr->func_info_rec_size; 244 245 fd = sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 246 247 if (fd >= 0 || !log_buf || !log_buf_sz) 248 goto done; 249 } 250 251 /* Try again with log */ 252 attr.log_buf = ptr_to_u64(log_buf); 253 attr.log_size = log_buf_sz; 254 attr.log_level = 1; 255 log_buf[0] = 0; 256 fd = sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 257 done: 258 free(finfo); 259 return fd; 260 } 261 262 int bpf_load_program(enum bpf_prog_type type, const struct bpf_insn *insns, 263 size_t insns_cnt, const char *license, 264 __u32 kern_version, char *log_buf, 265 size_t log_buf_sz) 266 { 267 struct bpf_load_program_attr load_attr; 268 269 memset(&load_attr, 0, sizeof(struct bpf_load_program_attr)); 270 load_attr.prog_type = type; 271 load_attr.expected_attach_type = 0; 272 load_attr.name = NULL; 273 load_attr.insns = insns; 274 load_attr.insns_cnt = insns_cnt; 275 load_attr.license = license; 276 load_attr.kern_version = kern_version; 277 278 return bpf_load_program_xattr(&load_attr, log_buf, log_buf_sz); 279 } 280 281 int bpf_verify_program(enum bpf_prog_type type, const struct bpf_insn *insns, 282 size_t insns_cnt, int strict_alignment, 283 const char *license, __u32 kern_version, 284 char *log_buf, size_t log_buf_sz, int log_level) 285 { 286 union bpf_attr attr; 287 288 bzero(&attr, sizeof(attr)); 289 attr.prog_type = type; 290 attr.insn_cnt = (__u32)insns_cnt; 291 attr.insns = ptr_to_u64(insns); 292 attr.license = ptr_to_u64(license); 293 attr.log_buf = ptr_to_u64(log_buf); 294 attr.log_size = log_buf_sz; 295 attr.log_level = log_level; 296 log_buf[0] = 0; 297 attr.kern_version = kern_version; 298 attr.prog_flags = strict_alignment ? BPF_F_STRICT_ALIGNMENT : 0; 299 300 return sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 301 } 302 303 int bpf_map_update_elem(int fd, const void *key, const void *value, 304 __u64 flags) 305 { 306 union bpf_attr attr; 307 308 bzero(&attr, sizeof(attr)); 309 attr.map_fd = fd; 310 attr.key = ptr_to_u64(key); 311 attr.value = ptr_to_u64(value); 312 attr.flags = flags; 313 314 return sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); 315 } 316 317 int bpf_map_lookup_elem(int fd, const void *key, void *value) 318 { 319 union bpf_attr attr; 320 321 bzero(&attr, sizeof(attr)); 322 attr.map_fd = fd; 323 attr.key = ptr_to_u64(key); 324 attr.value = ptr_to_u64(value); 325 326 return sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr)); 327 } 328 329 int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value) 330 { 331 union bpf_attr attr; 332 333 bzero(&attr, sizeof(attr)); 334 attr.map_fd = fd; 335 attr.key = ptr_to_u64(key); 336 attr.value = ptr_to_u64(value); 337 338 return sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, sizeof(attr)); 339 } 340 341 int bpf_map_delete_elem(int fd, const void *key) 342 { 343 union bpf_attr attr; 344 345 bzero(&attr, sizeof(attr)); 346 attr.map_fd = fd; 347 attr.key = ptr_to_u64(key); 348 349 return sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr)); 350 } 351 352 int bpf_map_get_next_key(int fd, const void *key, void *next_key) 353 { 354 union bpf_attr attr; 355 356 bzero(&attr, sizeof(attr)); 357 attr.map_fd = fd; 358 attr.key = ptr_to_u64(key); 359 attr.next_key = ptr_to_u64(next_key); 360 361 return sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, sizeof(attr)); 362 } 363 364 int bpf_obj_pin(int fd, const char *pathname) 365 { 366 union bpf_attr attr; 367 368 bzero(&attr, sizeof(attr)); 369 attr.pathname = ptr_to_u64((void *)pathname); 370 attr.bpf_fd = fd; 371 372 return sys_bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); 373 } 374 375 int bpf_obj_get(const char *pathname) 376 { 377 union bpf_attr attr; 378 379 bzero(&attr, sizeof(attr)); 380 attr.pathname = ptr_to_u64((void *)pathname); 381 382 return sys_bpf(BPF_OBJ_GET, &attr, sizeof(attr)); 383 } 384 385 int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type, 386 unsigned int flags) 387 { 388 union bpf_attr attr; 389 390 bzero(&attr, sizeof(attr)); 391 attr.target_fd = target_fd; 392 attr.attach_bpf_fd = prog_fd; 393 attr.attach_type = type; 394 attr.attach_flags = flags; 395 396 return sys_bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); 397 } 398 399 int bpf_prog_detach(int target_fd, enum bpf_attach_type type) 400 { 401 union bpf_attr attr; 402 403 bzero(&attr, sizeof(attr)); 404 attr.target_fd = target_fd; 405 attr.attach_type = type; 406 407 return sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); 408 } 409 410 int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type) 411 { 412 union bpf_attr attr; 413 414 bzero(&attr, sizeof(attr)); 415 attr.target_fd = target_fd; 416 attr.attach_bpf_fd = prog_fd; 417 attr.attach_type = type; 418 419 return sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); 420 } 421 422 int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags, 423 __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt) 424 { 425 union bpf_attr attr; 426 int ret; 427 428 bzero(&attr, sizeof(attr)); 429 attr.query.target_fd = target_fd; 430 attr.query.attach_type = type; 431 attr.query.query_flags = query_flags; 432 attr.query.prog_cnt = *prog_cnt; 433 attr.query.prog_ids = ptr_to_u64(prog_ids); 434 435 ret = sys_bpf(BPF_PROG_QUERY, &attr, sizeof(attr)); 436 if (attach_flags) 437 *attach_flags = attr.query.attach_flags; 438 *prog_cnt = attr.query.prog_cnt; 439 return ret; 440 } 441 442 int bpf_prog_test_run(int prog_fd, int repeat, void *data, __u32 size, 443 void *data_out, __u32 *size_out, __u32 *retval, 444 __u32 *duration) 445 { 446 union bpf_attr attr; 447 int ret; 448 449 bzero(&attr, sizeof(attr)); 450 attr.test.prog_fd = prog_fd; 451 attr.test.data_in = ptr_to_u64(data); 452 attr.test.data_out = ptr_to_u64(data_out); 453 attr.test.data_size_in = size; 454 attr.test.repeat = repeat; 455 456 ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr)); 457 if (size_out) 458 *size_out = attr.test.data_size_out; 459 if (retval) 460 *retval = attr.test.retval; 461 if (duration) 462 *duration = attr.test.duration; 463 return ret; 464 } 465 466 int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id) 467 { 468 union bpf_attr attr; 469 int err; 470 471 bzero(&attr, sizeof(attr)); 472 attr.start_id = start_id; 473 474 err = sys_bpf(BPF_PROG_GET_NEXT_ID, &attr, sizeof(attr)); 475 if (!err) 476 *next_id = attr.next_id; 477 478 return err; 479 } 480 481 int bpf_map_get_next_id(__u32 start_id, __u32 *next_id) 482 { 483 union bpf_attr attr; 484 int err; 485 486 bzero(&attr, sizeof(attr)); 487 attr.start_id = start_id; 488 489 err = sys_bpf(BPF_MAP_GET_NEXT_ID, &attr, sizeof(attr)); 490 if (!err) 491 *next_id = attr.next_id; 492 493 return err; 494 } 495 496 int bpf_prog_get_fd_by_id(__u32 id) 497 { 498 union bpf_attr attr; 499 500 bzero(&attr, sizeof(attr)); 501 attr.prog_id = id; 502 503 return sys_bpf(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr)); 504 } 505 506 int bpf_map_get_fd_by_id(__u32 id) 507 { 508 union bpf_attr attr; 509 510 bzero(&attr, sizeof(attr)); 511 attr.map_id = id; 512 513 return sys_bpf(BPF_MAP_GET_FD_BY_ID, &attr, sizeof(attr)); 514 } 515 516 int bpf_btf_get_fd_by_id(__u32 id) 517 { 518 union bpf_attr attr; 519 520 bzero(&attr, sizeof(attr)); 521 attr.btf_id = id; 522 523 return sys_bpf(BPF_BTF_GET_FD_BY_ID, &attr, sizeof(attr)); 524 } 525 526 int bpf_obj_get_info_by_fd(int prog_fd, void *info, __u32 *info_len) 527 { 528 union bpf_attr attr; 529 int err; 530 531 bzero(&attr, sizeof(attr)); 532 attr.info.bpf_fd = prog_fd; 533 attr.info.info_len = *info_len; 534 attr.info.info = ptr_to_u64(info); 535 536 err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr)); 537 if (!err) 538 *info_len = attr.info.info_len; 539 540 return err; 541 } 542 543 int bpf_raw_tracepoint_open(const char *name, int prog_fd) 544 { 545 union bpf_attr attr; 546 547 bzero(&attr, sizeof(attr)); 548 attr.raw_tracepoint.name = ptr_to_u64(name); 549 attr.raw_tracepoint.prog_fd = prog_fd; 550 551 return sys_bpf(BPF_RAW_TRACEPOINT_OPEN, &attr, sizeof(attr)); 552 } 553 554 int bpf_load_btf(void *btf, __u32 btf_size, char *log_buf, __u32 log_buf_size, 555 bool do_log) 556 { 557 union bpf_attr attr = {}; 558 int fd; 559 560 attr.btf = ptr_to_u64(btf); 561 attr.btf_size = btf_size; 562 563 retry: 564 if (do_log && log_buf && log_buf_size) { 565 attr.btf_log_level = 1; 566 attr.btf_log_size = log_buf_size; 567 attr.btf_log_buf = ptr_to_u64(log_buf); 568 } 569 570 fd = sys_bpf(BPF_BTF_LOAD, &attr, sizeof(attr)); 571 if (fd == -1 && !do_log && log_buf && log_buf_size) { 572 do_log = true; 573 goto retry; 574 } 575 576 return fd; 577 } 578 579 int bpf_task_fd_query(int pid, int fd, __u32 flags, char *buf, __u32 *buf_len, 580 __u32 *prog_id, __u32 *fd_type, __u64 *probe_offset, 581 __u64 *probe_addr) 582 { 583 union bpf_attr attr = {}; 584 int err; 585 586 attr.task_fd_query.pid = pid; 587 attr.task_fd_query.fd = fd; 588 attr.task_fd_query.flags = flags; 589 attr.task_fd_query.buf = ptr_to_u64(buf); 590 attr.task_fd_query.buf_len = *buf_len; 591 592 err = sys_bpf(BPF_TASK_FD_QUERY, &attr, sizeof(attr)); 593 *buf_len = attr.task_fd_query.buf_len; 594 *prog_id = attr.task_fd_query.prog_id; 595 *fd_type = attr.task_fd_query.fd_type; 596 *probe_offset = attr.task_fd_query.probe_offset; 597 *probe_addr = attr.task_fd_query.probe_addr; 598 599 return err; 600 } 601