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 static void * 177 alloc_zero_tailing_info(const void *orecord, __u32 cnt, 178 __u32 actual_rec_size, __u32 expected_rec_size) 179 { 180 __u64 info_len = actual_rec_size * cnt; 181 void *info, *nrecord; 182 int i; 183 184 info = malloc(info_len); 185 if (!info) 186 return NULL; 187 188 /* zero out bytes kernel does not understand */ 189 nrecord = info; 190 for (i = 0; i < cnt; i++) { 191 memcpy(nrecord, orecord, expected_rec_size); 192 memset(nrecord + expected_rec_size, 0, 193 actual_rec_size - expected_rec_size); 194 orecord += actual_rec_size; 195 nrecord += actual_rec_size; 196 } 197 198 return info; 199 } 200 201 int bpf_load_program_xattr(const struct bpf_load_program_attr *load_attr, 202 char *log_buf, size_t log_buf_sz) 203 { 204 void *finfo = NULL, *linfo = NULL; 205 union bpf_attr attr; 206 __u32 name_len; 207 int fd; 208 209 if (!load_attr) 210 return -EINVAL; 211 212 name_len = load_attr->name ? strlen(load_attr->name) : 0; 213 214 bzero(&attr, sizeof(attr)); 215 attr.prog_type = load_attr->prog_type; 216 attr.expected_attach_type = load_attr->expected_attach_type; 217 attr.insn_cnt = (__u32)load_attr->insns_cnt; 218 attr.insns = ptr_to_u64(load_attr->insns); 219 attr.license = ptr_to_u64(load_attr->license); 220 attr.log_buf = ptr_to_u64(NULL); 221 attr.log_size = 0; 222 attr.log_level = 0; 223 attr.kern_version = load_attr->kern_version; 224 attr.prog_ifindex = load_attr->prog_ifindex; 225 attr.prog_btf_fd = load_attr->prog_btf_fd; 226 attr.func_info_rec_size = load_attr->func_info_rec_size; 227 attr.func_info_cnt = load_attr->func_info_cnt; 228 attr.func_info = ptr_to_u64(load_attr->func_info); 229 attr.line_info_rec_size = load_attr->line_info_rec_size; 230 attr.line_info_cnt = load_attr->line_info_cnt; 231 attr.line_info = ptr_to_u64(load_attr->line_info); 232 memcpy(attr.prog_name, load_attr->name, 233 min(name_len, BPF_OBJ_NAME_LEN - 1)); 234 235 fd = sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 236 if (fd >= 0) 237 return fd; 238 239 /* After bpf_prog_load, the kernel may modify certain attributes 240 * to give user space a hint how to deal with loading failure. 241 * Check to see whether we can make some changes and load again. 242 */ 243 while (errno == E2BIG && (!finfo || !linfo)) { 244 if (!finfo && attr.func_info_cnt && 245 attr.func_info_rec_size < load_attr->func_info_rec_size) { 246 /* try with corrected func info records */ 247 finfo = alloc_zero_tailing_info(load_attr->func_info, 248 load_attr->func_info_cnt, 249 load_attr->func_info_rec_size, 250 attr.func_info_rec_size); 251 if (!finfo) 252 goto done; 253 254 attr.func_info = ptr_to_u64(finfo); 255 attr.func_info_rec_size = load_attr->func_info_rec_size; 256 } else if (!linfo && attr.line_info_cnt && 257 attr.line_info_rec_size < 258 load_attr->line_info_rec_size) { 259 linfo = alloc_zero_tailing_info(load_attr->line_info, 260 load_attr->line_info_cnt, 261 load_attr->line_info_rec_size, 262 attr.line_info_rec_size); 263 if (!linfo) 264 goto done; 265 266 attr.line_info = ptr_to_u64(linfo); 267 attr.line_info_rec_size = load_attr->line_info_rec_size; 268 } else { 269 break; 270 } 271 272 fd = sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 273 274 if (fd >= 0) 275 goto done; 276 } 277 278 if (!log_buf || !log_buf_sz) 279 goto done; 280 281 /* Try again with log */ 282 attr.log_buf = ptr_to_u64(log_buf); 283 attr.log_size = log_buf_sz; 284 attr.log_level = 1; 285 log_buf[0] = 0; 286 fd = sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 287 done: 288 free(finfo); 289 free(linfo); 290 return fd; 291 } 292 293 int bpf_load_program(enum bpf_prog_type type, const struct bpf_insn *insns, 294 size_t insns_cnt, const char *license, 295 __u32 kern_version, char *log_buf, 296 size_t log_buf_sz) 297 { 298 struct bpf_load_program_attr load_attr; 299 300 memset(&load_attr, 0, sizeof(struct bpf_load_program_attr)); 301 load_attr.prog_type = type; 302 load_attr.expected_attach_type = 0; 303 load_attr.name = NULL; 304 load_attr.insns = insns; 305 load_attr.insns_cnt = insns_cnt; 306 load_attr.license = license; 307 load_attr.kern_version = kern_version; 308 309 return bpf_load_program_xattr(&load_attr, log_buf, log_buf_sz); 310 } 311 312 int bpf_verify_program(enum bpf_prog_type type, const struct bpf_insn *insns, 313 size_t insns_cnt, __u32 prog_flags, const char *license, 314 __u32 kern_version, char *log_buf, size_t log_buf_sz, 315 int log_level) 316 { 317 union bpf_attr attr; 318 319 bzero(&attr, sizeof(attr)); 320 attr.prog_type = type; 321 attr.insn_cnt = (__u32)insns_cnt; 322 attr.insns = ptr_to_u64(insns); 323 attr.license = ptr_to_u64(license); 324 attr.log_buf = ptr_to_u64(log_buf); 325 attr.log_size = log_buf_sz; 326 attr.log_level = log_level; 327 log_buf[0] = 0; 328 attr.kern_version = kern_version; 329 attr.prog_flags = prog_flags; 330 331 return sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 332 } 333 334 int bpf_map_update_elem(int fd, const void *key, const void *value, 335 __u64 flags) 336 { 337 union bpf_attr attr; 338 339 bzero(&attr, sizeof(attr)); 340 attr.map_fd = fd; 341 attr.key = ptr_to_u64(key); 342 attr.value = ptr_to_u64(value); 343 attr.flags = flags; 344 345 return sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); 346 } 347 348 int bpf_map_lookup_elem(int fd, const void *key, void *value) 349 { 350 union bpf_attr attr; 351 352 bzero(&attr, sizeof(attr)); 353 attr.map_fd = fd; 354 attr.key = ptr_to_u64(key); 355 attr.value = ptr_to_u64(value); 356 357 return sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr)); 358 } 359 360 int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value) 361 { 362 union bpf_attr attr; 363 364 bzero(&attr, sizeof(attr)); 365 attr.map_fd = fd; 366 attr.key = ptr_to_u64(key); 367 attr.value = ptr_to_u64(value); 368 369 return sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, sizeof(attr)); 370 } 371 372 int bpf_map_delete_elem(int fd, const void *key) 373 { 374 union bpf_attr attr; 375 376 bzero(&attr, sizeof(attr)); 377 attr.map_fd = fd; 378 attr.key = ptr_to_u64(key); 379 380 return sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr)); 381 } 382 383 int bpf_map_get_next_key(int fd, const void *key, void *next_key) 384 { 385 union bpf_attr attr; 386 387 bzero(&attr, sizeof(attr)); 388 attr.map_fd = fd; 389 attr.key = ptr_to_u64(key); 390 attr.next_key = ptr_to_u64(next_key); 391 392 return sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, sizeof(attr)); 393 } 394 395 int bpf_obj_pin(int fd, const char *pathname) 396 { 397 union bpf_attr attr; 398 399 bzero(&attr, sizeof(attr)); 400 attr.pathname = ptr_to_u64((void *)pathname); 401 attr.bpf_fd = fd; 402 403 return sys_bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); 404 } 405 406 int bpf_obj_get(const char *pathname) 407 { 408 union bpf_attr attr; 409 410 bzero(&attr, sizeof(attr)); 411 attr.pathname = ptr_to_u64((void *)pathname); 412 413 return sys_bpf(BPF_OBJ_GET, &attr, sizeof(attr)); 414 } 415 416 int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type, 417 unsigned int flags) 418 { 419 union bpf_attr attr; 420 421 bzero(&attr, sizeof(attr)); 422 attr.target_fd = target_fd; 423 attr.attach_bpf_fd = prog_fd; 424 attr.attach_type = type; 425 attr.attach_flags = flags; 426 427 return sys_bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); 428 } 429 430 int bpf_prog_detach(int target_fd, enum bpf_attach_type type) 431 { 432 union bpf_attr attr; 433 434 bzero(&attr, sizeof(attr)); 435 attr.target_fd = target_fd; 436 attr.attach_type = type; 437 438 return sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); 439 } 440 441 int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type) 442 { 443 union bpf_attr attr; 444 445 bzero(&attr, sizeof(attr)); 446 attr.target_fd = target_fd; 447 attr.attach_bpf_fd = prog_fd; 448 attr.attach_type = type; 449 450 return sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); 451 } 452 453 int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags, 454 __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt) 455 { 456 union bpf_attr attr; 457 int ret; 458 459 bzero(&attr, sizeof(attr)); 460 attr.query.target_fd = target_fd; 461 attr.query.attach_type = type; 462 attr.query.query_flags = query_flags; 463 attr.query.prog_cnt = *prog_cnt; 464 attr.query.prog_ids = ptr_to_u64(prog_ids); 465 466 ret = sys_bpf(BPF_PROG_QUERY, &attr, sizeof(attr)); 467 if (attach_flags) 468 *attach_flags = attr.query.attach_flags; 469 *prog_cnt = attr.query.prog_cnt; 470 return ret; 471 } 472 473 int bpf_prog_test_run(int prog_fd, int repeat, void *data, __u32 size, 474 void *data_out, __u32 *size_out, __u32 *retval, 475 __u32 *duration) 476 { 477 union bpf_attr attr; 478 int ret; 479 480 bzero(&attr, sizeof(attr)); 481 attr.test.prog_fd = prog_fd; 482 attr.test.data_in = ptr_to_u64(data); 483 attr.test.data_out = ptr_to_u64(data_out); 484 attr.test.data_size_in = size; 485 attr.test.repeat = repeat; 486 487 ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr)); 488 if (size_out) 489 *size_out = attr.test.data_size_out; 490 if (retval) 491 *retval = attr.test.retval; 492 if (duration) 493 *duration = attr.test.duration; 494 return ret; 495 } 496 497 int bpf_prog_test_run_xattr(struct bpf_prog_test_run_attr *test_attr) 498 { 499 union bpf_attr attr; 500 int ret; 501 502 if (!test_attr->data_out && test_attr->data_size_out > 0) 503 return -EINVAL; 504 505 bzero(&attr, sizeof(attr)); 506 attr.test.prog_fd = test_attr->prog_fd; 507 attr.test.data_in = ptr_to_u64(test_attr->data_in); 508 attr.test.data_out = ptr_to_u64(test_attr->data_out); 509 attr.test.data_size_in = test_attr->data_size_in; 510 attr.test.data_size_out = test_attr->data_size_out; 511 attr.test.repeat = test_attr->repeat; 512 513 ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr)); 514 test_attr->data_size_out = attr.test.data_size_out; 515 test_attr->retval = attr.test.retval; 516 test_attr->duration = attr.test.duration; 517 return ret; 518 } 519 520 int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id) 521 { 522 union bpf_attr attr; 523 int err; 524 525 bzero(&attr, sizeof(attr)); 526 attr.start_id = start_id; 527 528 err = sys_bpf(BPF_PROG_GET_NEXT_ID, &attr, sizeof(attr)); 529 if (!err) 530 *next_id = attr.next_id; 531 532 return err; 533 } 534 535 int bpf_map_get_next_id(__u32 start_id, __u32 *next_id) 536 { 537 union bpf_attr attr; 538 int err; 539 540 bzero(&attr, sizeof(attr)); 541 attr.start_id = start_id; 542 543 err = sys_bpf(BPF_MAP_GET_NEXT_ID, &attr, sizeof(attr)); 544 if (!err) 545 *next_id = attr.next_id; 546 547 return err; 548 } 549 550 int bpf_prog_get_fd_by_id(__u32 id) 551 { 552 union bpf_attr attr; 553 554 bzero(&attr, sizeof(attr)); 555 attr.prog_id = id; 556 557 return sys_bpf(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr)); 558 } 559 560 int bpf_map_get_fd_by_id(__u32 id) 561 { 562 union bpf_attr attr; 563 564 bzero(&attr, sizeof(attr)); 565 attr.map_id = id; 566 567 return sys_bpf(BPF_MAP_GET_FD_BY_ID, &attr, sizeof(attr)); 568 } 569 570 int bpf_btf_get_fd_by_id(__u32 id) 571 { 572 union bpf_attr attr; 573 574 bzero(&attr, sizeof(attr)); 575 attr.btf_id = id; 576 577 return sys_bpf(BPF_BTF_GET_FD_BY_ID, &attr, sizeof(attr)); 578 } 579 580 int bpf_obj_get_info_by_fd(int prog_fd, void *info, __u32 *info_len) 581 { 582 union bpf_attr attr; 583 int err; 584 585 bzero(&attr, sizeof(attr)); 586 attr.info.bpf_fd = prog_fd; 587 attr.info.info_len = *info_len; 588 attr.info.info = ptr_to_u64(info); 589 590 err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr)); 591 if (!err) 592 *info_len = attr.info.info_len; 593 594 return err; 595 } 596 597 int bpf_raw_tracepoint_open(const char *name, int prog_fd) 598 { 599 union bpf_attr attr; 600 601 bzero(&attr, sizeof(attr)); 602 attr.raw_tracepoint.name = ptr_to_u64(name); 603 attr.raw_tracepoint.prog_fd = prog_fd; 604 605 return sys_bpf(BPF_RAW_TRACEPOINT_OPEN, &attr, sizeof(attr)); 606 } 607 608 int bpf_load_btf(void *btf, __u32 btf_size, char *log_buf, __u32 log_buf_size, 609 bool do_log) 610 { 611 union bpf_attr attr = {}; 612 int fd; 613 614 attr.btf = ptr_to_u64(btf); 615 attr.btf_size = btf_size; 616 617 retry: 618 if (do_log && log_buf && log_buf_size) { 619 attr.btf_log_level = 1; 620 attr.btf_log_size = log_buf_size; 621 attr.btf_log_buf = ptr_to_u64(log_buf); 622 } 623 624 fd = sys_bpf(BPF_BTF_LOAD, &attr, sizeof(attr)); 625 if (fd == -1 && !do_log && log_buf && log_buf_size) { 626 do_log = true; 627 goto retry; 628 } 629 630 return fd; 631 } 632 633 int bpf_task_fd_query(int pid, int fd, __u32 flags, char *buf, __u32 *buf_len, 634 __u32 *prog_id, __u32 *fd_type, __u64 *probe_offset, 635 __u64 *probe_addr) 636 { 637 union bpf_attr attr = {}; 638 int err; 639 640 attr.task_fd_query.pid = pid; 641 attr.task_fd_query.fd = fd; 642 attr.task_fd_query.flags = flags; 643 attr.task_fd_query.buf = ptr_to_u64(buf); 644 attr.task_fd_query.buf_len = *buf_len; 645 646 err = sys_bpf(BPF_TASK_FD_QUERY, &attr, sizeof(attr)); 647 *buf_len = attr.task_fd_query.buf_len; 648 *prog_id = attr.task_fd_query.prog_id; 649 *fd_type = attr.task_fd_query.fd_type; 650 *probe_offset = attr.task_fd_query.probe_offset; 651 *probe_addr = attr.task_fd_query.probe_addr; 652 653 return err; 654 } 655