| 1 | // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) |
| 2 | /* Copyright (c) 2019 Netronome Systems, Inc. */ |
| 3 | |
| 4 | #include <errno.h> |
| 5 | #include <fcntl.h> |
| 6 | #include <string.h> |
| 7 | #include <stdlib.h> |
| 8 | #include <unistd.h> |
| 9 | #include <net/if.h> |
| 10 | #include <sys/utsname.h> |
| 11 | |
| 12 | #include <linux/btf.h> |
| 13 | #include <linux/filter.h> |
| 14 | #include <linux/kernel.h> |
| 15 | #include <linux/version.h> |
| 16 | |
| 17 | #include "bpf.h" |
| 18 | #include "libbpf.h" |
| 19 | #include "libbpf_internal.h" |
| 20 | |
| 21 | /* On Ubuntu LINUX_VERSION_CODE doesn't correspond to info.release, |
| 22 | * but Ubuntu provides /proc/version_signature file, as described at |
| 23 | * https://ubuntu.com/kernel, with an example contents below, which we |
| 24 | * can use to get a proper LINUX_VERSION_CODE. |
| 25 | * |
| 26 | * Ubuntu 5.4.0-12.15-generic 5.4.8 |
| 27 | * |
| 28 | * In the above, 5.4.8 is what kernel is actually expecting, while |
| 29 | * uname() call will return 5.4.0 in info.release. |
| 30 | */ |
| 31 | static __u32 get_ubuntu_kernel_version(void) |
| 32 | { |
| 33 | const char *ubuntu_kver_file = "/proc/version_signature"; |
| 34 | __u32 major, minor, patch; |
| 35 | int ret; |
| 36 | FILE *f; |
| 37 | |
| 38 | if (faccessat(AT_FDCWD, ubuntu_kver_file, R_OK, AT_EACCESS) != 0) |
| 39 | return 0; |
| 40 | |
| 41 | f = fopen(ubuntu_kver_file, "re"); |
| 42 | if (!f) |
| 43 | return 0; |
| 44 | |
| 45 | ret = fscanf(f, "%*s %*s %u.%u.%u\n", &major, &minor, &patch); |
| 46 | fclose(f); |
| 47 | if (ret != 3) |
| 48 | return 0; |
| 49 | |
| 50 | return KERNEL_VERSION(major, minor, patch); |
| 51 | } |
| 52 | |
| 53 | /* On Debian LINUX_VERSION_CODE doesn't correspond to info.release. |
| 54 | * Instead, it is provided in info.version. An example content of |
| 55 | * Debian 10 looks like the below. |
| 56 | * |
| 57 | * utsname::release 4.19.0-22-amd64 |
| 58 | * utsname::version #1 SMP Debian 4.19.260-1 (2022-09-29) |
| 59 | * |
| 60 | * In the above, 4.19.260 is what kernel is actually expecting, while |
| 61 | * uname() call will return 4.19.0 in info.release. |
| 62 | */ |
| 63 | static __u32 get_debian_kernel_version(struct utsname *info) |
| 64 | { |
| 65 | __u32 major, minor, patch; |
| 66 | char *p; |
| 67 | |
| 68 | p = strstr(info->version, "Debian "); |
| 69 | if (!p) { |
| 70 | /* This is not a Debian kernel. */ |
| 71 | return 0; |
| 72 | } |
| 73 | |
| 74 | if (sscanf(p, "Debian %u.%u.%u", &major, &minor, &patch) != 3) |
| 75 | return 0; |
| 76 | |
| 77 | return KERNEL_VERSION(major, minor, patch); |
| 78 | } |
| 79 | |
| 80 | __u32 get_kernel_version(void) |
| 81 | { |
| 82 | __u32 major, minor, patch, version; |
| 83 | struct utsname info; |
| 84 | |
| 85 | /* Check if this is an Ubuntu kernel. */ |
| 86 | version = get_ubuntu_kernel_version(); |
| 87 | if (version != 0) |
| 88 | return version; |
| 89 | |
| 90 | uname(&info); |
| 91 | |
| 92 | /* Check if this is a Debian kernel. */ |
| 93 | version = get_debian_kernel_version(&info); |
| 94 | if (version != 0) |
| 95 | return version; |
| 96 | |
| 97 | if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3) |
| 98 | return 0; |
| 99 | |
| 100 | return KERNEL_VERSION(major, minor, patch); |
| 101 | } |
| 102 | |
| 103 | static int probe_prog_load(enum bpf_prog_type prog_type, |
| 104 | const struct bpf_insn *insns, size_t insns_cnt, |
| 105 | char *log_buf, size_t log_buf_sz) |
| 106 | { |
| 107 | LIBBPF_OPTS(bpf_prog_load_opts, opts, |
| 108 | .log_buf = log_buf, |
| 109 | .log_size = log_buf_sz, |
| 110 | .log_level = log_buf ? 1 : 0, |
| 111 | ); |
| 112 | int fd, err, exp_err = 0; |
| 113 | const char *exp_msg = NULL; |
| 114 | char buf[4096]; |
| 115 | |
| 116 | switch (prog_type) { |
| 117 | case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: |
| 118 | opts.expected_attach_type = BPF_CGROUP_INET4_CONNECT; |
| 119 | break; |
| 120 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: |
| 121 | opts.expected_attach_type = BPF_CGROUP_GETSOCKOPT; |
| 122 | break; |
| 123 | case BPF_PROG_TYPE_SK_LOOKUP: |
| 124 | opts.expected_attach_type = BPF_SK_LOOKUP; |
| 125 | break; |
| 126 | case BPF_PROG_TYPE_KPROBE: |
| 127 | opts.kern_version = get_kernel_version(); |
| 128 | break; |
| 129 | case BPF_PROG_TYPE_LIRC_MODE2: |
| 130 | opts.expected_attach_type = BPF_LIRC_MODE2; |
| 131 | break; |
| 132 | case BPF_PROG_TYPE_TRACING: |
| 133 | case BPF_PROG_TYPE_LSM: |
| 134 | opts.log_buf = buf; |
| 135 | opts.log_size = sizeof(buf); |
| 136 | opts.log_level = 1; |
| 137 | if (prog_type == BPF_PROG_TYPE_TRACING) |
| 138 | opts.expected_attach_type = BPF_TRACE_FENTRY; |
| 139 | else |
| 140 | opts.expected_attach_type = BPF_MODIFY_RETURN; |
| 141 | opts.attach_btf_id = 1; |
| 142 | |
| 143 | exp_err = -EINVAL; |
| 144 | exp_msg = "attach_btf_id 1 is not a function"; |
| 145 | break; |
| 146 | case BPF_PROG_TYPE_EXT: |
| 147 | opts.log_buf = buf; |
| 148 | opts.log_size = sizeof(buf); |
| 149 | opts.log_level = 1; |
| 150 | opts.attach_btf_id = 1; |
| 151 | |
| 152 | exp_err = -EINVAL; |
| 153 | exp_msg = "Cannot replace kernel functions"; |
| 154 | break; |
| 155 | case BPF_PROG_TYPE_SYSCALL: |
| 156 | opts.prog_flags = BPF_F_SLEEPABLE; |
| 157 | break; |
| 158 | case BPF_PROG_TYPE_STRUCT_OPS: |
| 159 | exp_err = -524; /* -ENOTSUPP */ |
| 160 | break; |
| 161 | case BPF_PROG_TYPE_UNSPEC: |
| 162 | case BPF_PROG_TYPE_SOCKET_FILTER: |
| 163 | case BPF_PROG_TYPE_SCHED_CLS: |
| 164 | case BPF_PROG_TYPE_SCHED_ACT: |
| 165 | case BPF_PROG_TYPE_TRACEPOINT: |
| 166 | case BPF_PROG_TYPE_XDP: |
| 167 | case BPF_PROG_TYPE_PERF_EVENT: |
| 168 | case BPF_PROG_TYPE_CGROUP_SKB: |
| 169 | case BPF_PROG_TYPE_CGROUP_SOCK: |
| 170 | case BPF_PROG_TYPE_LWT_IN: |
| 171 | case BPF_PROG_TYPE_LWT_OUT: |
| 172 | case BPF_PROG_TYPE_LWT_XMIT: |
| 173 | case BPF_PROG_TYPE_SOCK_OPS: |
| 174 | case BPF_PROG_TYPE_SK_SKB: |
| 175 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
| 176 | case BPF_PROG_TYPE_SK_MSG: |
| 177 | case BPF_PROG_TYPE_RAW_TRACEPOINT: |
| 178 | case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE: |
| 179 | case BPF_PROG_TYPE_LWT_SEG6LOCAL: |
| 180 | case BPF_PROG_TYPE_SK_REUSEPORT: |
| 181 | case BPF_PROG_TYPE_FLOW_DISSECTOR: |
| 182 | case BPF_PROG_TYPE_CGROUP_SYSCTL: |
| 183 | break; |
| 184 | case BPF_PROG_TYPE_NETFILTER: |
| 185 | opts.expected_attach_type = BPF_NETFILTER; |
| 186 | break; |
| 187 | default: |
| 188 | return -EOPNOTSUPP; |
| 189 | } |
| 190 | |
| 191 | fd = bpf_prog_load(prog_type, NULL, "GPL", insns, insns_cnt, &opts); |
| 192 | err = -errno; |
| 193 | if (fd >= 0) |
| 194 | close(fd); |
| 195 | if (exp_err) { |
| 196 | if (fd >= 0 || err != exp_err) |
| 197 | return 0; |
| 198 | if (exp_msg && !strstr(buf, exp_msg)) |
| 199 | return 0; |
| 200 | return 1; |
| 201 | } |
| 202 | return fd >= 0 ? 1 : 0; |
| 203 | } |
| 204 | |
| 205 | int libbpf_probe_bpf_prog_type(enum bpf_prog_type prog_type, const void *opts) |
| 206 | { |
| 207 | struct bpf_insn insns[] = { |
| 208 | BPF_MOV64_IMM(BPF_REG_0, 0), |
| 209 | BPF_EXIT_INSN() |
| 210 | }; |
| 211 | const size_t insn_cnt = ARRAY_SIZE(insns); |
| 212 | int ret; |
| 213 | |
| 214 | if (opts) |
| 215 | return libbpf_err(-EINVAL); |
| 216 | |
| 217 | ret = probe_prog_load(prog_type, insns, insn_cnt, NULL, 0); |
| 218 | return libbpf_err(ret); |
| 219 | } |
| 220 | |
| 221 | int libbpf__load_raw_btf(const char *raw_types, size_t types_len, |
| 222 | const char *str_sec, size_t str_len, |
| 223 | int token_fd) |
| 224 | { |
| 225 | struct btf_header hdr = { |
| 226 | .magic = BTF_MAGIC, |
| 227 | .version = BTF_VERSION, |
| 228 | .hdr_len = sizeof(struct btf_header), |
| 229 | .type_len = types_len, |
| 230 | .str_off = types_len, |
| 231 | .str_len = str_len, |
| 232 | }; |
| 233 | LIBBPF_OPTS(bpf_btf_load_opts, opts, |
| 234 | .token_fd = token_fd, |
| 235 | .btf_flags = token_fd ? BPF_F_TOKEN_FD : 0, |
| 236 | ); |
| 237 | int btf_fd, btf_len; |
| 238 | __u8 *raw_btf; |
| 239 | |
| 240 | btf_len = hdr.hdr_len + hdr.type_len + hdr.str_len; |
| 241 | raw_btf = malloc(btf_len); |
| 242 | if (!raw_btf) |
| 243 | return -ENOMEM; |
| 244 | |
| 245 | memcpy(raw_btf, &hdr, sizeof(hdr)); |
| 246 | memcpy(raw_btf + hdr.hdr_len, raw_types, hdr.type_len); |
| 247 | memcpy(raw_btf + hdr.hdr_len + hdr.type_len, str_sec, hdr.str_len); |
| 248 | |
| 249 | btf_fd = bpf_btf_load(raw_btf, btf_len, &opts); |
| 250 | |
| 251 | free(raw_btf); |
| 252 | return btf_fd; |
| 253 | } |
| 254 | |
| 255 | static int load_local_storage_btf(void) |
| 256 | { |
| 257 | const char strs[] = "\0bpf_spin_lock\0val\0cnt\0l"; |
| 258 | /* struct bpf_spin_lock { |
| 259 | * int val; |
| 260 | * }; |
| 261 | * struct val { |
| 262 | * int cnt; |
| 263 | * struct bpf_spin_lock l; |
| 264 | * }; |
| 265 | */ |
| 266 | __u32 types[] = { |
| 267 | /* int */ |
| 268 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ |
| 269 | /* struct bpf_spin_lock */ /* [2] */ |
| 270 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 1), 4), |
| 271 | BTF_MEMBER_ENC(15, 1, 0), /* int val; */ |
| 272 | /* struct val */ /* [3] */ |
| 273 | BTF_TYPE_ENC(15, BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 2), 8), |
| 274 | BTF_MEMBER_ENC(19, 1, 0), /* int cnt; */ |
| 275 | BTF_MEMBER_ENC(23, 2, 32),/* struct bpf_spin_lock l; */ |
| 276 | }; |
| 277 | |
| 278 | return libbpf__load_raw_btf((char *)types, sizeof(types), |
| 279 | strs, sizeof(strs), 0); |
| 280 | } |
| 281 | |
| 282 | static int probe_map_create(enum bpf_map_type map_type) |
| 283 | { |
| 284 | LIBBPF_OPTS(bpf_map_create_opts, opts); |
| 285 | int key_size, value_size, max_entries; |
| 286 | __u32 btf_key_type_id = 0, btf_value_type_id = 0; |
| 287 | int fd = -1, btf_fd = -1, fd_inner = -1, exp_err = 0, err = 0; |
| 288 | |
| 289 | key_size = sizeof(__u32); |
| 290 | value_size = sizeof(__u32); |
| 291 | max_entries = 1; |
| 292 | |
| 293 | switch (map_type) { |
| 294 | case BPF_MAP_TYPE_STACK_TRACE: |
| 295 | value_size = sizeof(__u64); |
| 296 | break; |
| 297 | case BPF_MAP_TYPE_LPM_TRIE: |
| 298 | key_size = sizeof(__u64); |
| 299 | value_size = sizeof(__u64); |
| 300 | opts.map_flags = BPF_F_NO_PREALLOC; |
| 301 | break; |
| 302 | case BPF_MAP_TYPE_CGROUP_STORAGE: |
| 303 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: |
| 304 | key_size = sizeof(struct bpf_cgroup_storage_key); |
| 305 | value_size = sizeof(__u64); |
| 306 | max_entries = 0; |
| 307 | break; |
| 308 | case BPF_MAP_TYPE_QUEUE: |
| 309 | case BPF_MAP_TYPE_STACK: |
| 310 | key_size = 0; |
| 311 | break; |
| 312 | case BPF_MAP_TYPE_SK_STORAGE: |
| 313 | case BPF_MAP_TYPE_INODE_STORAGE: |
| 314 | case BPF_MAP_TYPE_TASK_STORAGE: |
| 315 | case BPF_MAP_TYPE_CGRP_STORAGE: |
| 316 | btf_key_type_id = 1; |
| 317 | btf_value_type_id = 3; |
| 318 | value_size = 8; |
| 319 | max_entries = 0; |
| 320 | opts.map_flags = BPF_F_NO_PREALLOC; |
| 321 | btf_fd = load_local_storage_btf(); |
| 322 | if (btf_fd < 0) |
| 323 | return btf_fd; |
| 324 | break; |
| 325 | case BPF_MAP_TYPE_RINGBUF: |
| 326 | case BPF_MAP_TYPE_USER_RINGBUF: |
| 327 | key_size = 0; |
| 328 | value_size = 0; |
| 329 | max_entries = sysconf(_SC_PAGE_SIZE); |
| 330 | break; |
| 331 | case BPF_MAP_TYPE_STRUCT_OPS: |
| 332 | /* we'll get -ENOTSUPP for invalid BTF type ID for struct_ops */ |
| 333 | opts.btf_vmlinux_value_type_id = 1; |
| 334 | opts.value_type_btf_obj_fd = -1; |
| 335 | exp_err = -524; /* -ENOTSUPP */ |
| 336 | break; |
| 337 | case BPF_MAP_TYPE_BLOOM_FILTER: |
| 338 | key_size = 0; |
| 339 | max_entries = 1; |
| 340 | break; |
| 341 | case BPF_MAP_TYPE_ARENA: |
| 342 | key_size = 0; |
| 343 | value_size = 0; |
| 344 | max_entries = 1; /* one page */ |
| 345 | opts.map_extra = 0; /* can mmap() at any address */ |
| 346 | opts.map_flags = BPF_F_MMAPABLE; |
| 347 | break; |
| 348 | case BPF_MAP_TYPE_HASH: |
| 349 | case BPF_MAP_TYPE_ARRAY: |
| 350 | case BPF_MAP_TYPE_PROG_ARRAY: |
| 351 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: |
| 352 | case BPF_MAP_TYPE_PERCPU_HASH: |
| 353 | case BPF_MAP_TYPE_PERCPU_ARRAY: |
| 354 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
| 355 | case BPF_MAP_TYPE_LRU_HASH: |
| 356 | case BPF_MAP_TYPE_LRU_PERCPU_HASH: |
| 357 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
| 358 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
| 359 | case BPF_MAP_TYPE_DEVMAP: |
| 360 | case BPF_MAP_TYPE_DEVMAP_HASH: |
| 361 | case BPF_MAP_TYPE_SOCKMAP: |
| 362 | case BPF_MAP_TYPE_CPUMAP: |
| 363 | case BPF_MAP_TYPE_XSKMAP: |
| 364 | case BPF_MAP_TYPE_SOCKHASH: |
| 365 | case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY: |
| 366 | break; |
| 367 | case BPF_MAP_TYPE_UNSPEC: |
| 368 | default: |
| 369 | return -EOPNOTSUPP; |
| 370 | } |
| 371 | |
| 372 | if (map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS || |
| 373 | map_type == BPF_MAP_TYPE_HASH_OF_MAPS) { |
| 374 | fd_inner = bpf_map_create(BPF_MAP_TYPE_HASH, NULL, |
| 375 | sizeof(__u32), sizeof(__u32), 1, NULL); |
| 376 | if (fd_inner < 0) |
| 377 | goto cleanup; |
| 378 | |
| 379 | opts.inner_map_fd = fd_inner; |
| 380 | } |
| 381 | |
| 382 | if (btf_fd >= 0) { |
| 383 | opts.btf_fd = btf_fd; |
| 384 | opts.btf_key_type_id = btf_key_type_id; |
| 385 | opts.btf_value_type_id = btf_value_type_id; |
| 386 | } |
| 387 | |
| 388 | fd = bpf_map_create(map_type, NULL, key_size, value_size, max_entries, &opts); |
| 389 | err = -errno; |
| 390 | |
| 391 | cleanup: |
| 392 | if (fd >= 0) |
| 393 | close(fd); |
| 394 | if (fd_inner >= 0) |
| 395 | close(fd_inner); |
| 396 | if (btf_fd >= 0) |
| 397 | close(btf_fd); |
| 398 | |
| 399 | if (exp_err) |
| 400 | return fd < 0 && err == exp_err ? 1 : 0; |
| 401 | else |
| 402 | return fd >= 0 ? 1 : 0; |
| 403 | } |
| 404 | |
| 405 | int libbpf_probe_bpf_map_type(enum bpf_map_type map_type, const void *opts) |
| 406 | { |
| 407 | int ret; |
| 408 | |
| 409 | if (opts) |
| 410 | return libbpf_err(-EINVAL); |
| 411 | |
| 412 | ret = probe_map_create(map_type); |
| 413 | return libbpf_err(ret); |
| 414 | } |
| 415 | |
| 416 | int libbpf_probe_bpf_helper(enum bpf_prog_type prog_type, enum bpf_func_id helper_id, |
| 417 | const void *opts) |
| 418 | { |
| 419 | struct bpf_insn insns[] = { |
| 420 | BPF_EMIT_CALL((__u32)helper_id), |
| 421 | BPF_EXIT_INSN(), |
| 422 | }; |
| 423 | const size_t insn_cnt = ARRAY_SIZE(insns); |
| 424 | char buf[4096]; |
| 425 | int ret; |
| 426 | |
| 427 | if (opts) |
| 428 | return libbpf_err(-EINVAL); |
| 429 | |
| 430 | /* we can't successfully load all prog types to check for BPF helper |
| 431 | * support, so bail out with -EOPNOTSUPP error |
| 432 | */ |
| 433 | switch (prog_type) { |
| 434 | case BPF_PROG_TYPE_TRACING: |
| 435 | case BPF_PROG_TYPE_EXT: |
| 436 | case BPF_PROG_TYPE_LSM: |
| 437 | case BPF_PROG_TYPE_STRUCT_OPS: |
| 438 | return -EOPNOTSUPP; |
| 439 | default: |
| 440 | break; |
| 441 | } |
| 442 | |
| 443 | buf[0] = '\0'; |
| 444 | ret = probe_prog_load(prog_type, insns, insn_cnt, buf, sizeof(buf)); |
| 445 | if (ret < 0) |
| 446 | return libbpf_err(ret); |
| 447 | |
| 448 | /* If BPF verifier doesn't recognize BPF helper ID (enum bpf_func_id) |
| 449 | * at all, it will emit something like "invalid func unknown#181". |
| 450 | * If BPF verifier recognizes BPF helper but it's not supported for |
| 451 | * given BPF program type, it will emit "unknown func bpf_sys_bpf#166" |
| 452 | * or "program of this type cannot use helper bpf_sys_bpf#166". |
| 453 | * In both cases, provided combination of BPF program type and BPF |
| 454 | * helper is not supported by the kernel. |
| 455 | * In all other cases, probe_prog_load() above will either succeed (e.g., |
| 456 | * because BPF helper happens to accept no input arguments or it |
| 457 | * accepts one input argument and initial PTR_TO_CTX is fine for |
| 458 | * that), or we'll get some more specific BPF verifier error about |
| 459 | * some unsatisfied conditions. |
| 460 | */ |
| 461 | if (ret == 0 && (strstr(buf, "invalid func ") || strstr(buf, "unknown func ") || |
| 462 | strstr(buf, "program of this type cannot use helper "))) |
| 463 | return 0; |
| 464 | return 1; /* assume supported */ |
| 465 | } |