b.liu | e958203 | 2025-04-17 19:18:16 +0800 | [diff] [blame^] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
| 3 | */ |
| 4 | #include <linux/bpf.h> |
| 5 | #include <linux/rcupdate.h> |
| 6 | #include <linux/random.h> |
| 7 | #include <linux/smp.h> |
| 8 | #include <linux/topology.h> |
| 9 | #include <linux/ktime.h> |
| 10 | #include <linux/sched.h> |
| 11 | #include <linux/uidgid.h> |
| 12 | #include <linux/filter.h> |
| 13 | #include <linux/ctype.h> |
| 14 | |
| 15 | #include "../../lib/kstrtox.h" |
| 16 | |
| 17 | /* If kernel subsystem is allowing eBPF programs to call this function, |
| 18 | * inside its own verifier_ops->get_func_proto() callback it should return |
| 19 | * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments |
| 20 | * |
| 21 | * Different map implementations will rely on rcu in map methods |
| 22 | * lookup/update/delete, therefore eBPF programs must run under rcu lock |
| 23 | * if program is allowed to access maps, so check rcu_read_lock_held in |
| 24 | * all three functions. |
| 25 | */ |
| 26 | BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key) |
| 27 | { |
| 28 | WARN_ON_ONCE(!rcu_read_lock_held()); |
| 29 | return (unsigned long) map->ops->map_lookup_elem(map, key); |
| 30 | } |
| 31 | |
| 32 | const struct bpf_func_proto bpf_map_lookup_elem_proto = { |
| 33 | .func = bpf_map_lookup_elem, |
| 34 | .gpl_only = false, |
| 35 | .pkt_access = true, |
| 36 | .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, |
| 37 | .arg1_type = ARG_CONST_MAP_PTR, |
| 38 | .arg2_type = ARG_PTR_TO_MAP_KEY, |
| 39 | }; |
| 40 | |
| 41 | BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key, |
| 42 | void *, value, u64, flags) |
| 43 | { |
| 44 | WARN_ON_ONCE(!rcu_read_lock_held()); |
| 45 | return map->ops->map_update_elem(map, key, value, flags); |
| 46 | } |
| 47 | |
| 48 | const struct bpf_func_proto bpf_map_update_elem_proto = { |
| 49 | .func = bpf_map_update_elem, |
| 50 | .gpl_only = false, |
| 51 | .pkt_access = true, |
| 52 | .ret_type = RET_INTEGER, |
| 53 | .arg1_type = ARG_CONST_MAP_PTR, |
| 54 | .arg2_type = ARG_PTR_TO_MAP_KEY, |
| 55 | .arg3_type = ARG_PTR_TO_MAP_VALUE, |
| 56 | .arg4_type = ARG_ANYTHING, |
| 57 | }; |
| 58 | |
| 59 | BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key) |
| 60 | { |
| 61 | WARN_ON_ONCE(!rcu_read_lock_held()); |
| 62 | return map->ops->map_delete_elem(map, key); |
| 63 | } |
| 64 | |
| 65 | const struct bpf_func_proto bpf_map_delete_elem_proto = { |
| 66 | .func = bpf_map_delete_elem, |
| 67 | .gpl_only = false, |
| 68 | .pkt_access = true, |
| 69 | .ret_type = RET_INTEGER, |
| 70 | .arg1_type = ARG_CONST_MAP_PTR, |
| 71 | .arg2_type = ARG_PTR_TO_MAP_KEY, |
| 72 | }; |
| 73 | |
| 74 | BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags) |
| 75 | { |
| 76 | return map->ops->map_push_elem(map, value, flags); |
| 77 | } |
| 78 | |
| 79 | const struct bpf_func_proto bpf_map_push_elem_proto = { |
| 80 | .func = bpf_map_push_elem, |
| 81 | .gpl_only = false, |
| 82 | .pkt_access = true, |
| 83 | .ret_type = RET_INTEGER, |
| 84 | .arg1_type = ARG_CONST_MAP_PTR, |
| 85 | .arg2_type = ARG_PTR_TO_MAP_VALUE, |
| 86 | .arg3_type = ARG_ANYTHING, |
| 87 | }; |
| 88 | |
| 89 | BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value) |
| 90 | { |
| 91 | return map->ops->map_pop_elem(map, value); |
| 92 | } |
| 93 | |
| 94 | const struct bpf_func_proto bpf_map_pop_elem_proto = { |
| 95 | .func = bpf_map_pop_elem, |
| 96 | .gpl_only = false, |
| 97 | .ret_type = RET_INTEGER, |
| 98 | .arg1_type = ARG_CONST_MAP_PTR, |
| 99 | .arg2_type = ARG_PTR_TO_UNINIT_MAP_VALUE, |
| 100 | }; |
| 101 | |
| 102 | BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value) |
| 103 | { |
| 104 | return map->ops->map_peek_elem(map, value); |
| 105 | } |
| 106 | |
| 107 | const struct bpf_func_proto bpf_map_peek_elem_proto = { |
| 108 | .func = bpf_map_peek_elem, |
| 109 | .gpl_only = false, |
| 110 | .ret_type = RET_INTEGER, |
| 111 | .arg1_type = ARG_CONST_MAP_PTR, |
| 112 | .arg2_type = ARG_PTR_TO_UNINIT_MAP_VALUE, |
| 113 | }; |
| 114 | |
| 115 | const struct bpf_func_proto bpf_get_prandom_u32_proto = { |
| 116 | .func = bpf_user_rnd_u32, |
| 117 | .gpl_only = false, |
| 118 | .ret_type = RET_INTEGER, |
| 119 | }; |
| 120 | |
| 121 | BPF_CALL_0(bpf_get_smp_processor_id) |
| 122 | { |
| 123 | return smp_processor_id(); |
| 124 | } |
| 125 | |
| 126 | const struct bpf_func_proto bpf_get_smp_processor_id_proto = { |
| 127 | .func = bpf_get_smp_processor_id, |
| 128 | .gpl_only = false, |
| 129 | .ret_type = RET_INTEGER, |
| 130 | }; |
| 131 | |
| 132 | BPF_CALL_0(bpf_get_numa_node_id) |
| 133 | { |
| 134 | return numa_node_id(); |
| 135 | } |
| 136 | |
| 137 | const struct bpf_func_proto bpf_get_numa_node_id_proto = { |
| 138 | .func = bpf_get_numa_node_id, |
| 139 | .gpl_only = false, |
| 140 | .ret_type = RET_INTEGER, |
| 141 | }; |
| 142 | |
| 143 | BPF_CALL_0(bpf_ktime_get_ns) |
| 144 | { |
| 145 | /* NMI safe access to clock monotonic */ |
| 146 | return ktime_get_mono_fast_ns(); |
| 147 | } |
| 148 | |
| 149 | const struct bpf_func_proto bpf_ktime_get_ns_proto = { |
| 150 | .func = bpf_ktime_get_ns, |
| 151 | .gpl_only = false, |
| 152 | .ret_type = RET_INTEGER, |
| 153 | }; |
| 154 | |
| 155 | BPF_CALL_0(bpf_ktime_get_boot_ns) |
| 156 | { |
| 157 | /* NMI safe access to clock boottime */ |
| 158 | return ktime_get_boot_fast_ns(); |
| 159 | } |
| 160 | |
| 161 | const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = { |
| 162 | .func = bpf_ktime_get_boot_ns, |
| 163 | .gpl_only = false, |
| 164 | .ret_type = RET_INTEGER, |
| 165 | }; |
| 166 | |
| 167 | BPF_CALL_0(bpf_get_current_pid_tgid) |
| 168 | { |
| 169 | struct task_struct *task = current; |
| 170 | |
| 171 | if (unlikely(!task)) |
| 172 | return -EINVAL; |
| 173 | |
| 174 | return (u64) task->tgid << 32 | task->pid; |
| 175 | } |
| 176 | |
| 177 | const struct bpf_func_proto bpf_get_current_pid_tgid_proto = { |
| 178 | .func = bpf_get_current_pid_tgid, |
| 179 | .gpl_only = false, |
| 180 | .ret_type = RET_INTEGER, |
| 181 | }; |
| 182 | |
| 183 | BPF_CALL_0(bpf_get_current_uid_gid) |
| 184 | { |
| 185 | struct task_struct *task = current; |
| 186 | kuid_t uid; |
| 187 | kgid_t gid; |
| 188 | |
| 189 | if (unlikely(!task)) |
| 190 | return -EINVAL; |
| 191 | |
| 192 | current_uid_gid(&uid, &gid); |
| 193 | return (u64) from_kgid(&init_user_ns, gid) << 32 | |
| 194 | from_kuid(&init_user_ns, uid); |
| 195 | } |
| 196 | |
| 197 | const struct bpf_func_proto bpf_get_current_uid_gid_proto = { |
| 198 | .func = bpf_get_current_uid_gid, |
| 199 | .gpl_only = false, |
| 200 | .ret_type = RET_INTEGER, |
| 201 | }; |
| 202 | |
| 203 | BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size) |
| 204 | { |
| 205 | struct task_struct *task = current; |
| 206 | |
| 207 | if (unlikely(!task)) |
| 208 | goto err_clear; |
| 209 | |
| 210 | strncpy(buf, task->comm, size); |
| 211 | |
| 212 | /* Verifier guarantees that size > 0. For task->comm exceeding |
| 213 | * size, guarantee that buf is %NUL-terminated. Unconditionally |
| 214 | * done here to save the size test. |
| 215 | */ |
| 216 | buf[size - 1] = 0; |
| 217 | return 0; |
| 218 | err_clear: |
| 219 | memset(buf, 0, size); |
| 220 | return -EINVAL; |
| 221 | } |
| 222 | |
| 223 | const struct bpf_func_proto bpf_get_current_comm_proto = { |
| 224 | .func = bpf_get_current_comm, |
| 225 | .gpl_only = false, |
| 226 | .ret_type = RET_INTEGER, |
| 227 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, |
| 228 | .arg2_type = ARG_CONST_SIZE, |
| 229 | }; |
| 230 | |
| 231 | #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK) |
| 232 | |
| 233 | static inline void __bpf_spin_lock(struct bpf_spin_lock *lock) |
| 234 | { |
| 235 | arch_spinlock_t *l = (void *)lock; |
| 236 | union { |
| 237 | __u32 val; |
| 238 | arch_spinlock_t lock; |
| 239 | } u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED }; |
| 240 | |
| 241 | compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0"); |
| 242 | BUILD_BUG_ON(sizeof(*l) != sizeof(__u32)); |
| 243 | BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32)); |
| 244 | arch_spin_lock(l); |
| 245 | } |
| 246 | |
| 247 | static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock) |
| 248 | { |
| 249 | arch_spinlock_t *l = (void *)lock; |
| 250 | |
| 251 | arch_spin_unlock(l); |
| 252 | } |
| 253 | |
| 254 | #else |
| 255 | |
| 256 | static inline void __bpf_spin_lock(struct bpf_spin_lock *lock) |
| 257 | { |
| 258 | atomic_t *l = (void *)lock; |
| 259 | |
| 260 | BUILD_BUG_ON(sizeof(*l) != sizeof(*lock)); |
| 261 | do { |
| 262 | atomic_cond_read_relaxed(l, !VAL); |
| 263 | } while (atomic_xchg(l, 1)); |
| 264 | } |
| 265 | |
| 266 | static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock) |
| 267 | { |
| 268 | atomic_t *l = (void *)lock; |
| 269 | |
| 270 | atomic_set_release(l, 0); |
| 271 | } |
| 272 | |
| 273 | #endif |
| 274 | |
| 275 | static DEFINE_PER_CPU(unsigned long, irqsave_flags); |
| 276 | |
| 277 | static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock) |
| 278 | { |
| 279 | unsigned long flags; |
| 280 | |
| 281 | local_irq_save(flags); |
| 282 | __bpf_spin_lock(lock); |
| 283 | __this_cpu_write(irqsave_flags, flags); |
| 284 | } |
| 285 | |
| 286 | NOTRACE_BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock) |
| 287 | { |
| 288 | __bpf_spin_lock_irqsave(lock); |
| 289 | return 0; |
| 290 | } |
| 291 | |
| 292 | const struct bpf_func_proto bpf_spin_lock_proto = { |
| 293 | .func = bpf_spin_lock, |
| 294 | .gpl_only = false, |
| 295 | .ret_type = RET_VOID, |
| 296 | .arg1_type = ARG_PTR_TO_SPIN_LOCK, |
| 297 | }; |
| 298 | |
| 299 | static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock) |
| 300 | { |
| 301 | unsigned long flags; |
| 302 | |
| 303 | flags = __this_cpu_read(irqsave_flags); |
| 304 | __bpf_spin_unlock(lock); |
| 305 | local_irq_restore(flags); |
| 306 | } |
| 307 | |
| 308 | NOTRACE_BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock) |
| 309 | { |
| 310 | __bpf_spin_unlock_irqrestore(lock); |
| 311 | return 0; |
| 312 | } |
| 313 | |
| 314 | const struct bpf_func_proto bpf_spin_unlock_proto = { |
| 315 | .func = bpf_spin_unlock, |
| 316 | .gpl_only = false, |
| 317 | .ret_type = RET_VOID, |
| 318 | .arg1_type = ARG_PTR_TO_SPIN_LOCK, |
| 319 | }; |
| 320 | |
| 321 | void copy_map_value_locked(struct bpf_map *map, void *dst, void *src, |
| 322 | bool lock_src) |
| 323 | { |
| 324 | struct bpf_spin_lock *lock; |
| 325 | |
| 326 | if (lock_src) |
| 327 | lock = src + map->spin_lock_off; |
| 328 | else |
| 329 | lock = dst + map->spin_lock_off; |
| 330 | preempt_disable(); |
| 331 | __bpf_spin_lock_irqsave(lock); |
| 332 | copy_map_value(map, dst, src); |
| 333 | __bpf_spin_unlock_irqrestore(lock); |
| 334 | preempt_enable(); |
| 335 | } |
| 336 | |
| 337 | #ifdef CONFIG_CGROUPS |
| 338 | BPF_CALL_0(bpf_get_current_cgroup_id) |
| 339 | { |
| 340 | struct cgroup *cgrp = task_dfl_cgroup(current); |
| 341 | |
| 342 | return cgrp->kn->id.id; |
| 343 | } |
| 344 | |
| 345 | const struct bpf_func_proto bpf_get_current_cgroup_id_proto = { |
| 346 | .func = bpf_get_current_cgroup_id, |
| 347 | .gpl_only = false, |
| 348 | .ret_type = RET_INTEGER, |
| 349 | }; |
| 350 | |
| 351 | #ifdef CONFIG_CGROUP_BPF |
| 352 | DECLARE_PER_CPU(struct bpf_cgroup_storage*, |
| 353 | bpf_cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]); |
| 354 | |
| 355 | BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags) |
| 356 | { |
| 357 | /* flags argument is not used now, |
| 358 | * but provides an ability to extend the API. |
| 359 | * verifier checks that its value is correct. |
| 360 | */ |
| 361 | enum bpf_cgroup_storage_type stype = cgroup_storage_type(map); |
| 362 | struct bpf_cgroup_storage *storage; |
| 363 | void *ptr; |
| 364 | |
| 365 | storage = this_cpu_read(bpf_cgroup_storage[stype]); |
| 366 | |
| 367 | if (stype == BPF_CGROUP_STORAGE_SHARED) |
| 368 | ptr = &READ_ONCE(storage->buf)->data[0]; |
| 369 | else |
| 370 | ptr = this_cpu_ptr(storage->percpu_buf); |
| 371 | |
| 372 | return (unsigned long)ptr; |
| 373 | } |
| 374 | |
| 375 | const struct bpf_func_proto bpf_get_local_storage_proto = { |
| 376 | .func = bpf_get_local_storage, |
| 377 | .gpl_only = false, |
| 378 | .ret_type = RET_PTR_TO_MAP_VALUE, |
| 379 | .arg1_type = ARG_CONST_MAP_PTR, |
| 380 | .arg2_type = ARG_ANYTHING, |
| 381 | }; |
| 382 | #endif |
| 383 | |
| 384 | #define BPF_STRTOX_BASE_MASK 0x1F |
| 385 | |
| 386 | static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags, |
| 387 | unsigned long long *res, bool *is_negative) |
| 388 | { |
| 389 | unsigned int base = flags & BPF_STRTOX_BASE_MASK; |
| 390 | const char *cur_buf = buf; |
| 391 | size_t cur_len = buf_len; |
| 392 | unsigned int consumed; |
| 393 | size_t val_len; |
| 394 | char str[64]; |
| 395 | |
| 396 | if (!buf || !buf_len || !res || !is_negative) |
| 397 | return -EINVAL; |
| 398 | |
| 399 | if (base != 0 && base != 8 && base != 10 && base != 16) |
| 400 | return -EINVAL; |
| 401 | |
| 402 | if (flags & ~BPF_STRTOX_BASE_MASK) |
| 403 | return -EINVAL; |
| 404 | |
| 405 | while (cur_buf < buf + buf_len && isspace(*cur_buf)) |
| 406 | ++cur_buf; |
| 407 | |
| 408 | *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-'); |
| 409 | if (*is_negative) |
| 410 | ++cur_buf; |
| 411 | |
| 412 | consumed = cur_buf - buf; |
| 413 | cur_len -= consumed; |
| 414 | if (!cur_len) |
| 415 | return -EINVAL; |
| 416 | |
| 417 | cur_len = min(cur_len, sizeof(str) - 1); |
| 418 | memcpy(str, cur_buf, cur_len); |
| 419 | str[cur_len] = '\0'; |
| 420 | cur_buf = str; |
| 421 | |
| 422 | cur_buf = _parse_integer_fixup_radix(cur_buf, &base); |
| 423 | val_len = _parse_integer(cur_buf, base, res); |
| 424 | |
| 425 | if (val_len & KSTRTOX_OVERFLOW) |
| 426 | return -ERANGE; |
| 427 | |
| 428 | if (val_len == 0) |
| 429 | return -EINVAL; |
| 430 | |
| 431 | cur_buf += val_len; |
| 432 | consumed += cur_buf - str; |
| 433 | |
| 434 | return consumed; |
| 435 | } |
| 436 | |
| 437 | static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags, |
| 438 | long long *res) |
| 439 | { |
| 440 | unsigned long long _res; |
| 441 | bool is_negative; |
| 442 | int err; |
| 443 | |
| 444 | err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative); |
| 445 | if (err < 0) |
| 446 | return err; |
| 447 | if (is_negative) { |
| 448 | if ((long long)-_res > 0) |
| 449 | return -ERANGE; |
| 450 | *res = -_res; |
| 451 | } else { |
| 452 | if ((long long)_res < 0) |
| 453 | return -ERANGE; |
| 454 | *res = _res; |
| 455 | } |
| 456 | return err; |
| 457 | } |
| 458 | |
| 459 | BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags, |
| 460 | s64 *, res) |
| 461 | { |
| 462 | long long _res; |
| 463 | int err; |
| 464 | |
| 465 | err = __bpf_strtoll(buf, buf_len, flags, &_res); |
| 466 | if (err < 0) |
| 467 | return err; |
| 468 | if (_res != (long)_res) |
| 469 | return -ERANGE; |
| 470 | *res = _res; |
| 471 | return err; |
| 472 | } |
| 473 | |
| 474 | const struct bpf_func_proto bpf_strtol_proto = { |
| 475 | .func = bpf_strtol, |
| 476 | .gpl_only = false, |
| 477 | .ret_type = RET_INTEGER, |
| 478 | .arg1_type = ARG_PTR_TO_MEM, |
| 479 | .arg2_type = ARG_CONST_SIZE, |
| 480 | .arg3_type = ARG_ANYTHING, |
| 481 | .arg4_type = ARG_PTR_TO_LONG, |
| 482 | }; |
| 483 | |
| 484 | BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags, |
| 485 | u64 *, res) |
| 486 | { |
| 487 | unsigned long long _res; |
| 488 | bool is_negative; |
| 489 | int err; |
| 490 | |
| 491 | err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative); |
| 492 | if (err < 0) |
| 493 | return err; |
| 494 | if (is_negative) |
| 495 | return -EINVAL; |
| 496 | if (_res != (unsigned long)_res) |
| 497 | return -ERANGE; |
| 498 | *res = _res; |
| 499 | return err; |
| 500 | } |
| 501 | |
| 502 | const struct bpf_func_proto bpf_strtoul_proto = { |
| 503 | .func = bpf_strtoul, |
| 504 | .gpl_only = false, |
| 505 | .ret_type = RET_INTEGER, |
| 506 | .arg1_type = ARG_PTR_TO_MEM, |
| 507 | .arg2_type = ARG_CONST_SIZE, |
| 508 | .arg3_type = ARG_ANYTHING, |
| 509 | .arg4_type = ARG_PTR_TO_LONG, |
| 510 | }; |
| 511 | #endif |