Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux
at v4.20-rc4 552 lines 14 kB view raw
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 57static inline __u64 ptr_to_u64(const void *ptr) 58{ 59 return (__u64) (unsigned long) ptr; 60} 61 62static 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 68int 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 92int 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 112int 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 126int 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 142int 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 167int 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 176int 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 __u32 name_len; 181 int fd; 182 183 if (!load_attr) 184 return -EINVAL; 185 186 name_len = load_attr->name ? strlen(load_attr->name) : 0; 187 188 bzero(&attr, sizeof(attr)); 189 attr.prog_type = load_attr->prog_type; 190 attr.expected_attach_type = load_attr->expected_attach_type; 191 attr.insn_cnt = (__u32)load_attr->insns_cnt; 192 attr.insns = ptr_to_u64(load_attr->insns); 193 attr.license = ptr_to_u64(load_attr->license); 194 attr.log_buf = ptr_to_u64(NULL); 195 attr.log_size = 0; 196 attr.log_level = 0; 197 attr.kern_version = load_attr->kern_version; 198 attr.prog_ifindex = load_attr->prog_ifindex; 199 memcpy(attr.prog_name, load_attr->name, 200 min(name_len, BPF_OBJ_NAME_LEN - 1)); 201 202 fd = sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 203 if (fd >= 0 || !log_buf || !log_buf_sz) 204 return fd; 205 206 /* Try again with log */ 207 attr.log_buf = ptr_to_u64(log_buf); 208 attr.log_size = log_buf_sz; 209 attr.log_level = 1; 210 log_buf[0] = 0; 211 return sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 212} 213 214int bpf_load_program(enum bpf_prog_type type, const struct bpf_insn *insns, 215 size_t insns_cnt, const char *license, 216 __u32 kern_version, char *log_buf, 217 size_t log_buf_sz) 218{ 219 struct bpf_load_program_attr load_attr; 220 221 memset(&load_attr, 0, sizeof(struct bpf_load_program_attr)); 222 load_attr.prog_type = type; 223 load_attr.expected_attach_type = 0; 224 load_attr.name = NULL; 225 load_attr.insns = insns; 226 load_attr.insns_cnt = insns_cnt; 227 load_attr.license = license; 228 load_attr.kern_version = kern_version; 229 230 return bpf_load_program_xattr(&load_attr, log_buf, log_buf_sz); 231} 232 233int bpf_verify_program(enum bpf_prog_type type, const struct bpf_insn *insns, 234 size_t insns_cnt, int strict_alignment, 235 const char *license, __u32 kern_version, 236 char *log_buf, size_t log_buf_sz, int log_level) 237{ 238 union bpf_attr attr; 239 240 bzero(&attr, sizeof(attr)); 241 attr.prog_type = type; 242 attr.insn_cnt = (__u32)insns_cnt; 243 attr.insns = ptr_to_u64(insns); 244 attr.license = ptr_to_u64(license); 245 attr.log_buf = ptr_to_u64(log_buf); 246 attr.log_size = log_buf_sz; 247 attr.log_level = log_level; 248 log_buf[0] = 0; 249 attr.kern_version = kern_version; 250 attr.prog_flags = strict_alignment ? BPF_F_STRICT_ALIGNMENT : 0; 251 252 return sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 253} 254 255int bpf_map_update_elem(int fd, const void *key, const void *value, 256 __u64 flags) 257{ 258 union bpf_attr attr; 259 260 bzero(&attr, sizeof(attr)); 261 attr.map_fd = fd; 262 attr.key = ptr_to_u64(key); 263 attr.value = ptr_to_u64(value); 264 attr.flags = flags; 265 266 return sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); 267} 268 269int bpf_map_lookup_elem(int fd, const void *key, void *value) 270{ 271 union bpf_attr attr; 272 273 bzero(&attr, sizeof(attr)); 274 attr.map_fd = fd; 275 attr.key = ptr_to_u64(key); 276 attr.value = ptr_to_u64(value); 277 278 return sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr)); 279} 280 281int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value) 282{ 283 union bpf_attr attr; 284 285 bzero(&attr, sizeof(attr)); 286 attr.map_fd = fd; 287 attr.key = ptr_to_u64(key); 288 attr.value = ptr_to_u64(value); 289 290 return sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, sizeof(attr)); 291} 292 293int bpf_map_delete_elem(int fd, const void *key) 294{ 295 union bpf_attr attr; 296 297 bzero(&attr, sizeof(attr)); 298 attr.map_fd = fd; 299 attr.key = ptr_to_u64(key); 300 301 return sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr)); 302} 303 304int bpf_map_get_next_key(int fd, const void *key, void *next_key) 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.next_key = ptr_to_u64(next_key); 312 313 return sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, sizeof(attr)); 314} 315 316int bpf_obj_pin(int fd, const char *pathname) 317{ 318 union bpf_attr attr; 319 320 bzero(&attr, sizeof(attr)); 321 attr.pathname = ptr_to_u64((void *)pathname); 322 attr.bpf_fd = fd; 323 324 return sys_bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); 325} 326 327int bpf_obj_get(const char *pathname) 328{ 329 union bpf_attr attr; 330 331 bzero(&attr, sizeof(attr)); 332 attr.pathname = ptr_to_u64((void *)pathname); 333 334 return sys_bpf(BPF_OBJ_GET, &attr, sizeof(attr)); 335} 336 337int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type, 338 unsigned int flags) 339{ 340 union bpf_attr attr; 341 342 bzero(&attr, sizeof(attr)); 343 attr.target_fd = target_fd; 344 attr.attach_bpf_fd = prog_fd; 345 attr.attach_type = type; 346 attr.attach_flags = flags; 347 348 return sys_bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); 349} 350 351int bpf_prog_detach(int target_fd, enum bpf_attach_type type) 352{ 353 union bpf_attr attr; 354 355 bzero(&attr, sizeof(attr)); 356 attr.target_fd = target_fd; 357 attr.attach_type = type; 358 359 return sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); 360} 361 362int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type) 363{ 364 union bpf_attr attr; 365 366 bzero(&attr, sizeof(attr)); 367 attr.target_fd = target_fd; 368 attr.attach_bpf_fd = prog_fd; 369 attr.attach_type = type; 370 371 return sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); 372} 373 374int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags, 375 __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt) 376{ 377 union bpf_attr attr; 378 int ret; 379 380 bzero(&attr, sizeof(attr)); 381 attr.query.target_fd = target_fd; 382 attr.query.attach_type = type; 383 attr.query.query_flags = query_flags; 384 attr.query.prog_cnt = *prog_cnt; 385 attr.query.prog_ids = ptr_to_u64(prog_ids); 386 387 ret = sys_bpf(BPF_PROG_QUERY, &attr, sizeof(attr)); 388 if (attach_flags) 389 *attach_flags = attr.query.attach_flags; 390 *prog_cnt = attr.query.prog_cnt; 391 return ret; 392} 393 394int bpf_prog_test_run(int prog_fd, int repeat, void *data, __u32 size, 395 void *data_out, __u32 *size_out, __u32 *retval, 396 __u32 *duration) 397{ 398 union bpf_attr attr; 399 int ret; 400 401 bzero(&attr, sizeof(attr)); 402 attr.test.prog_fd = prog_fd; 403 attr.test.data_in = ptr_to_u64(data); 404 attr.test.data_out = ptr_to_u64(data_out); 405 attr.test.data_size_in = size; 406 attr.test.repeat = repeat; 407 408 ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr)); 409 if (size_out) 410 *size_out = attr.test.data_size_out; 411 if (retval) 412 *retval = attr.test.retval; 413 if (duration) 414 *duration = attr.test.duration; 415 return ret; 416} 417 418int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id) 419{ 420 union bpf_attr attr; 421 int err; 422 423 bzero(&attr, sizeof(attr)); 424 attr.start_id = start_id; 425 426 err = sys_bpf(BPF_PROG_GET_NEXT_ID, &attr, sizeof(attr)); 427 if (!err) 428 *next_id = attr.next_id; 429 430 return err; 431} 432 433int bpf_map_get_next_id(__u32 start_id, __u32 *next_id) 434{ 435 union bpf_attr attr; 436 int err; 437 438 bzero(&attr, sizeof(attr)); 439 attr.start_id = start_id; 440 441 err = sys_bpf(BPF_MAP_GET_NEXT_ID, &attr, sizeof(attr)); 442 if (!err) 443 *next_id = attr.next_id; 444 445 return err; 446} 447 448int bpf_prog_get_fd_by_id(__u32 id) 449{ 450 union bpf_attr attr; 451 452 bzero(&attr, sizeof(attr)); 453 attr.prog_id = id; 454 455 return sys_bpf(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr)); 456} 457 458int bpf_map_get_fd_by_id(__u32 id) 459{ 460 union bpf_attr attr; 461 462 bzero(&attr, sizeof(attr)); 463 attr.map_id = id; 464 465 return sys_bpf(BPF_MAP_GET_FD_BY_ID, &attr, sizeof(attr)); 466} 467 468int bpf_btf_get_fd_by_id(__u32 id) 469{ 470 union bpf_attr attr; 471 472 bzero(&attr, sizeof(attr)); 473 attr.btf_id = id; 474 475 return sys_bpf(BPF_BTF_GET_FD_BY_ID, &attr, sizeof(attr)); 476} 477 478int bpf_obj_get_info_by_fd(int prog_fd, void *info, __u32 *info_len) 479{ 480 union bpf_attr attr; 481 int err; 482 483 bzero(&attr, sizeof(attr)); 484 attr.info.bpf_fd = prog_fd; 485 attr.info.info_len = *info_len; 486 attr.info.info = ptr_to_u64(info); 487 488 err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr)); 489 if (!err) 490 *info_len = attr.info.info_len; 491 492 return err; 493} 494 495int bpf_raw_tracepoint_open(const char *name, int prog_fd) 496{ 497 union bpf_attr attr; 498 499 bzero(&attr, sizeof(attr)); 500 attr.raw_tracepoint.name = ptr_to_u64(name); 501 attr.raw_tracepoint.prog_fd = prog_fd; 502 503 return sys_bpf(BPF_RAW_TRACEPOINT_OPEN, &attr, sizeof(attr)); 504} 505 506int bpf_load_btf(void *btf, __u32 btf_size, char *log_buf, __u32 log_buf_size, 507 bool do_log) 508{ 509 union bpf_attr attr = {}; 510 int fd; 511 512 attr.btf = ptr_to_u64(btf); 513 attr.btf_size = btf_size; 514 515retry: 516 if (do_log && log_buf && log_buf_size) { 517 attr.btf_log_level = 1; 518 attr.btf_log_size = log_buf_size; 519 attr.btf_log_buf = ptr_to_u64(log_buf); 520 } 521 522 fd = sys_bpf(BPF_BTF_LOAD, &attr, sizeof(attr)); 523 if (fd == -1 && !do_log && log_buf && log_buf_size) { 524 do_log = true; 525 goto retry; 526 } 527 528 return fd; 529} 530 531int bpf_task_fd_query(int pid, int fd, __u32 flags, char *buf, __u32 *buf_len, 532 __u32 *prog_id, __u32 *fd_type, __u64 *probe_offset, 533 __u64 *probe_addr) 534{ 535 union bpf_attr attr = {}; 536 int err; 537 538 attr.task_fd_query.pid = pid; 539 attr.task_fd_query.fd = fd; 540 attr.task_fd_query.flags = flags; 541 attr.task_fd_query.buf = ptr_to_u64(buf); 542 attr.task_fd_query.buf_len = *buf_len; 543 544 err = sys_bpf(BPF_TASK_FD_QUERY, &attr, sizeof(attr)); 545 *buf_len = attr.task_fd_query.buf_len; 546 *prog_id = attr.task_fd_query.prog_id; 547 *fd_type = attr.task_fd_query.fd_type; 548 *probe_offset = attr.task_fd_query.probe_offset; 549 *probe_addr = attr.task_fd_query.probe_addr; 550 551 return err; 552}