1 // SPDX-License-Identifier: LGPL-2.1 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 "nlattr.h" 32 #include <linux/rtnetlink.h> 33 #include <linux/if_link.h> 34 #include <sys/socket.h> 35 #include <errno.h> 36 37 #ifndef SOL_NETLINK 38 #define SOL_NETLINK 270 39 #endif 40 41 /* 42 * When building perf, unistd.h is overridden. __NR_bpf is 43 * required to be defined explicitly. 44 */ 45 #ifndef __NR_bpf 46 # if defined(__i386__) 47 # define __NR_bpf 357 48 # elif defined(__x86_64__) 49 # define __NR_bpf 321 50 # elif defined(__aarch64__) 51 # define __NR_bpf 280 52 # elif defined(__sparc__) 53 # define __NR_bpf 349 54 # elif defined(__s390__) 55 # define __NR_bpf 351 56 # else 57 # error __NR_bpf not defined. libbpf does not support your arch. 58 # endif 59 #endif 60 61 #ifndef min 62 #define min(x, y) ((x) < (y) ? (x) : (y)) 63 #endif 64 65 static inline __u64 ptr_to_u64(const void *ptr) 66 { 67 return (__u64) (unsigned long) ptr; 68 } 69 70 static inline int sys_bpf(enum bpf_cmd cmd, union bpf_attr *attr, 71 unsigned int size) 72 { 73 return syscall(__NR_bpf, cmd, attr, size); 74 } 75 76 int bpf_create_map_node(enum bpf_map_type map_type, const char *name, 77 int key_size, int value_size, int max_entries, 78 __u32 map_flags, int node) 79 { 80 __u32 name_len = name ? strlen(name) : 0; 81 union bpf_attr attr; 82 83 memset(&attr, '\0', sizeof(attr)); 84 85 attr.map_type = map_type; 86 attr.key_size = key_size; 87 attr.value_size = value_size; 88 attr.max_entries = max_entries; 89 attr.map_flags = map_flags; 90 memcpy(attr.map_name, name, min(name_len, BPF_OBJ_NAME_LEN - 1)); 91 92 if (node >= 0) { 93 attr.map_flags |= BPF_F_NUMA_NODE; 94 attr.numa_node = node; 95 } 96 97 return sys_bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); 98 } 99 100 int bpf_create_map(enum bpf_map_type map_type, int key_size, 101 int value_size, int max_entries, __u32 map_flags) 102 { 103 return bpf_create_map_node(map_type, NULL, key_size, value_size, 104 max_entries, map_flags, -1); 105 } 106 107 int bpf_create_map_name(enum bpf_map_type map_type, const char *name, 108 int key_size, int value_size, int max_entries, 109 __u32 map_flags) 110 { 111 return bpf_create_map_node(map_type, name, key_size, value_size, 112 max_entries, map_flags, -1); 113 } 114 115 int bpf_create_map_in_map_node(enum bpf_map_type map_type, const char *name, 116 int key_size, int inner_map_fd, int max_entries, 117 __u32 map_flags, int node) 118 { 119 __u32 name_len = name ? strlen(name) : 0; 120 union bpf_attr attr; 121 122 memset(&attr, '\0', sizeof(attr)); 123 124 attr.map_type = map_type; 125 attr.key_size = key_size; 126 attr.value_size = 4; 127 attr.inner_map_fd = inner_map_fd; 128 attr.max_entries = max_entries; 129 attr.map_flags = map_flags; 130 memcpy(attr.map_name, name, min(name_len, BPF_OBJ_NAME_LEN - 1)); 131 132 if (node >= 0) { 133 attr.map_flags |= BPF_F_NUMA_NODE; 134 attr.numa_node = node; 135 } 136 137 return sys_bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); 138 } 139 140 int bpf_create_map_in_map(enum bpf_map_type map_type, const char *name, 141 int key_size, int inner_map_fd, int max_entries, 142 __u32 map_flags) 143 { 144 return bpf_create_map_in_map_node(map_type, name, key_size, 145 inner_map_fd, max_entries, map_flags, 146 -1); 147 } 148 149 int bpf_load_program_xattr(const struct bpf_load_program_attr *load_attr, 150 char *log_buf, size_t log_buf_sz) 151 { 152 union bpf_attr attr; 153 __u32 name_len; 154 int fd; 155 156 if (!load_attr) 157 return -EINVAL; 158 159 name_len = load_attr->name ? strlen(load_attr->name) : 0; 160 161 bzero(&attr, sizeof(attr)); 162 attr.prog_type = load_attr->prog_type; 163 attr.expected_attach_type = load_attr->expected_attach_type; 164 attr.insn_cnt = (__u32)load_attr->insns_cnt; 165 attr.insns = ptr_to_u64(load_attr->insns); 166 attr.license = ptr_to_u64(load_attr->license); 167 attr.log_buf = ptr_to_u64(NULL); 168 attr.log_size = 0; 169 attr.log_level = 0; 170 attr.kern_version = load_attr->kern_version; 171 memcpy(attr.prog_name, load_attr->name, 172 min(name_len, BPF_OBJ_NAME_LEN - 1)); 173 174 fd = sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 175 if (fd >= 0 || !log_buf || !log_buf_sz) 176 return fd; 177 178 /* Try again with log */ 179 attr.log_buf = ptr_to_u64(log_buf); 180 attr.log_size = log_buf_sz; 181 attr.log_level = 1; 182 log_buf[0] = 0; 183 return sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 184 } 185 186 int bpf_load_program(enum bpf_prog_type type, const struct bpf_insn *insns, 187 size_t insns_cnt, const char *license, 188 __u32 kern_version, char *log_buf, 189 size_t log_buf_sz) 190 { 191 struct bpf_load_program_attr load_attr; 192 193 memset(&load_attr, 0, sizeof(struct bpf_load_program_attr)); 194 load_attr.prog_type = type; 195 load_attr.expected_attach_type = 0; 196 load_attr.name = NULL; 197 load_attr.insns = insns; 198 load_attr.insns_cnt = insns_cnt; 199 load_attr.license = license; 200 load_attr.kern_version = kern_version; 201 202 return bpf_load_program_xattr(&load_attr, log_buf, log_buf_sz); 203 } 204 205 int bpf_verify_program(enum bpf_prog_type type, const struct bpf_insn *insns, 206 size_t insns_cnt, int strict_alignment, 207 const char *license, __u32 kern_version, 208 char *log_buf, size_t log_buf_sz, int log_level) 209 { 210 union bpf_attr attr; 211 212 bzero(&attr, sizeof(attr)); 213 attr.prog_type = type; 214 attr.insn_cnt = (__u32)insns_cnt; 215 attr.insns = ptr_to_u64(insns); 216 attr.license = ptr_to_u64(license); 217 attr.log_buf = ptr_to_u64(log_buf); 218 attr.log_size = log_buf_sz; 219 attr.log_level = log_level; 220 log_buf[0] = 0; 221 attr.kern_version = kern_version; 222 attr.prog_flags = strict_alignment ? BPF_F_STRICT_ALIGNMENT : 0; 223 224 return sys_bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); 225 } 226 227 int bpf_map_update_elem(int fd, const void *key, const void *value, 228 __u64 flags) 229 { 230 union bpf_attr attr; 231 232 bzero(&attr, sizeof(attr)); 233 attr.map_fd = fd; 234 attr.key = ptr_to_u64(key); 235 attr.value = ptr_to_u64(value); 236 attr.flags = flags; 237 238 return sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); 239 } 240 241 int bpf_map_lookup_elem(int fd, const void *key, void *value) 242 { 243 union bpf_attr attr; 244 245 bzero(&attr, sizeof(attr)); 246 attr.map_fd = fd; 247 attr.key = ptr_to_u64(key); 248 attr.value = ptr_to_u64(value); 249 250 return sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr)); 251 } 252 253 int bpf_map_delete_elem(int fd, const void *key) 254 { 255 union bpf_attr attr; 256 257 bzero(&attr, sizeof(attr)); 258 attr.map_fd = fd; 259 attr.key = ptr_to_u64(key); 260 261 return sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr)); 262 } 263 264 int bpf_map_get_next_key(int fd, const void *key, void *next_key) 265 { 266 union bpf_attr attr; 267 268 bzero(&attr, sizeof(attr)); 269 attr.map_fd = fd; 270 attr.key = ptr_to_u64(key); 271 attr.next_key = ptr_to_u64(next_key); 272 273 return sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, sizeof(attr)); 274 } 275 276 int bpf_obj_pin(int fd, const char *pathname) 277 { 278 union bpf_attr attr; 279 280 bzero(&attr, sizeof(attr)); 281 attr.pathname = ptr_to_u64((void *)pathname); 282 attr.bpf_fd = fd; 283 284 return sys_bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); 285 } 286 287 int bpf_obj_get(const char *pathname) 288 { 289 union bpf_attr attr; 290 291 bzero(&attr, sizeof(attr)); 292 attr.pathname = ptr_to_u64((void *)pathname); 293 294 return sys_bpf(BPF_OBJ_GET, &attr, sizeof(attr)); 295 } 296 297 int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type, 298 unsigned int flags) 299 { 300 union bpf_attr attr; 301 302 bzero(&attr, sizeof(attr)); 303 attr.target_fd = target_fd; 304 attr.attach_bpf_fd = prog_fd; 305 attr.attach_type = type; 306 attr.attach_flags = flags; 307 308 return sys_bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); 309 } 310 311 int bpf_prog_detach(int target_fd, enum bpf_attach_type type) 312 { 313 union bpf_attr attr; 314 315 bzero(&attr, sizeof(attr)); 316 attr.target_fd = target_fd; 317 attr.attach_type = type; 318 319 return sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); 320 } 321 322 int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type) 323 { 324 union bpf_attr attr; 325 326 bzero(&attr, sizeof(attr)); 327 attr.target_fd = target_fd; 328 attr.attach_bpf_fd = prog_fd; 329 attr.attach_type = type; 330 331 return sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); 332 } 333 334 int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags, 335 __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt) 336 { 337 union bpf_attr attr; 338 int ret; 339 340 bzero(&attr, sizeof(attr)); 341 attr.query.target_fd = target_fd; 342 attr.query.attach_type = type; 343 attr.query.query_flags = query_flags; 344 attr.query.prog_cnt = *prog_cnt; 345 attr.query.prog_ids = ptr_to_u64(prog_ids); 346 347 ret = sys_bpf(BPF_PROG_QUERY, &attr, sizeof(attr)); 348 if (attach_flags) 349 *attach_flags = attr.query.attach_flags; 350 *prog_cnt = attr.query.prog_cnt; 351 return ret; 352 } 353 354 int bpf_prog_test_run(int prog_fd, int repeat, void *data, __u32 size, 355 void *data_out, __u32 *size_out, __u32 *retval, 356 __u32 *duration) 357 { 358 union bpf_attr attr; 359 int ret; 360 361 bzero(&attr, sizeof(attr)); 362 attr.test.prog_fd = prog_fd; 363 attr.test.data_in = ptr_to_u64(data); 364 attr.test.data_out = ptr_to_u64(data_out); 365 attr.test.data_size_in = size; 366 attr.test.repeat = repeat; 367 368 ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr)); 369 if (size_out) 370 *size_out = attr.test.data_size_out; 371 if (retval) 372 *retval = attr.test.retval; 373 if (duration) 374 *duration = attr.test.duration; 375 return ret; 376 } 377 378 int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id) 379 { 380 union bpf_attr attr; 381 int err; 382 383 bzero(&attr, sizeof(attr)); 384 attr.start_id = start_id; 385 386 err = sys_bpf(BPF_PROG_GET_NEXT_ID, &attr, sizeof(attr)); 387 if (!err) 388 *next_id = attr.next_id; 389 390 return err; 391 } 392 393 int bpf_map_get_next_id(__u32 start_id, __u32 *next_id) 394 { 395 union bpf_attr attr; 396 int err; 397 398 bzero(&attr, sizeof(attr)); 399 attr.start_id = start_id; 400 401 err = sys_bpf(BPF_MAP_GET_NEXT_ID, &attr, sizeof(attr)); 402 if (!err) 403 *next_id = attr.next_id; 404 405 return err; 406 } 407 408 int bpf_prog_get_fd_by_id(__u32 id) 409 { 410 union bpf_attr attr; 411 412 bzero(&attr, sizeof(attr)); 413 attr.prog_id = id; 414 415 return sys_bpf(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr)); 416 } 417 418 int bpf_map_get_fd_by_id(__u32 id) 419 { 420 union bpf_attr attr; 421 422 bzero(&attr, sizeof(attr)); 423 attr.map_id = id; 424 425 return sys_bpf(BPF_MAP_GET_FD_BY_ID, &attr, sizeof(attr)); 426 } 427 428 int bpf_obj_get_info_by_fd(int prog_fd, void *info, __u32 *info_len) 429 { 430 union bpf_attr attr; 431 int err; 432 433 bzero(&attr, sizeof(attr)); 434 attr.info.bpf_fd = prog_fd; 435 attr.info.info_len = *info_len; 436 attr.info.info = ptr_to_u64(info); 437 438 err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr)); 439 if (!err) 440 *info_len = attr.info.info_len; 441 442 return err; 443 } 444 445 int bpf_raw_tracepoint_open(const char *name, int prog_fd) 446 { 447 union bpf_attr attr; 448 449 bzero(&attr, sizeof(attr)); 450 attr.raw_tracepoint.name = ptr_to_u64(name); 451 attr.raw_tracepoint.prog_fd = prog_fd; 452 453 return sys_bpf(BPF_RAW_TRACEPOINT_OPEN, &attr, sizeof(attr)); 454 } 455 456 int bpf_set_link_xdp_fd(int ifindex, int fd, __u32 flags) 457 { 458 struct sockaddr_nl sa; 459 int sock, seq = 0, len, ret = -1; 460 char buf[4096]; 461 struct nlattr *nla, *nla_xdp; 462 struct { 463 struct nlmsghdr nh; 464 struct ifinfomsg ifinfo; 465 char attrbuf[64]; 466 } req; 467 struct nlmsghdr *nh; 468 struct nlmsgerr *err; 469 socklen_t addrlen; 470 int one = 1; 471 472 memset(&sa, 0, sizeof(sa)); 473 sa.nl_family = AF_NETLINK; 474 475 sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE); 476 if (sock < 0) { 477 return -errno; 478 } 479 480 if (setsockopt(sock, SOL_NETLINK, NETLINK_EXT_ACK, 481 &one, sizeof(one)) < 0) { 482 fprintf(stderr, "Netlink error reporting not supported\n"); 483 } 484 485 if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) { 486 ret = -errno; 487 goto cleanup; 488 } 489 490 addrlen = sizeof(sa); 491 if (getsockname(sock, (struct sockaddr *)&sa, &addrlen) < 0) { 492 ret = -errno; 493 goto cleanup; 494 } 495 496 if (addrlen != sizeof(sa)) { 497 ret = -LIBBPF_ERRNO__INTERNAL; 498 goto cleanup; 499 } 500 501 memset(&req, 0, sizeof(req)); 502 req.nh.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg)); 503 req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; 504 req.nh.nlmsg_type = RTM_SETLINK; 505 req.nh.nlmsg_pid = 0; 506 req.nh.nlmsg_seq = ++seq; 507 req.ifinfo.ifi_family = AF_UNSPEC; 508 req.ifinfo.ifi_index = ifindex; 509 510 /* started nested attribute for XDP */ 511 nla = (struct nlattr *)(((char *)&req) 512 + NLMSG_ALIGN(req.nh.nlmsg_len)); 513 nla->nla_type = NLA_F_NESTED | IFLA_XDP; 514 nla->nla_len = NLA_HDRLEN; 515 516 /* add XDP fd */ 517 nla_xdp = (struct nlattr *)((char *)nla + nla->nla_len); 518 nla_xdp->nla_type = IFLA_XDP_FD; 519 nla_xdp->nla_len = NLA_HDRLEN + sizeof(int); 520 memcpy((char *)nla_xdp + NLA_HDRLEN, &fd, sizeof(fd)); 521 nla->nla_len += nla_xdp->nla_len; 522 523 /* if user passed in any flags, add those too */ 524 if (flags) { 525 nla_xdp = (struct nlattr *)((char *)nla + nla->nla_len); 526 nla_xdp->nla_type = IFLA_XDP_FLAGS; 527 nla_xdp->nla_len = NLA_HDRLEN + sizeof(flags); 528 memcpy((char *)nla_xdp + NLA_HDRLEN, &flags, sizeof(flags)); 529 nla->nla_len += nla_xdp->nla_len; 530 } 531 532 req.nh.nlmsg_len += NLA_ALIGN(nla->nla_len); 533 534 if (send(sock, &req, req.nh.nlmsg_len, 0) < 0) { 535 ret = -errno; 536 goto cleanup; 537 } 538 539 len = recv(sock, buf, sizeof(buf), 0); 540 if (len < 0) { 541 ret = -errno; 542 goto cleanup; 543 } 544 545 for (nh = (struct nlmsghdr *)buf; NLMSG_OK(nh, len); 546 nh = NLMSG_NEXT(nh, len)) { 547 if (nh->nlmsg_pid != sa.nl_pid) { 548 ret = -LIBBPF_ERRNO__WRNGPID; 549 goto cleanup; 550 } 551 if (nh->nlmsg_seq != seq) { 552 ret = -LIBBPF_ERRNO__INVSEQ; 553 goto cleanup; 554 } 555 switch (nh->nlmsg_type) { 556 case NLMSG_ERROR: 557 err = (struct nlmsgerr *)NLMSG_DATA(nh); 558 if (!err->error) 559 continue; 560 ret = err->error; 561 nla_dump_errormsg(nh); 562 goto cleanup; 563 case NLMSG_DONE: 564 break; 565 default: 566 break; 567 } 568 } 569 570 ret = 0; 571 572 cleanup: 573 close(sock); 574 return ret; 575 } 576