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/btf.h> 6 #include <linux/bpf-cgroup.h> 7 #include <linux/rcupdate.h> 8 #include <linux/random.h> 9 #include <linux/smp.h> 10 #include <linux/topology.h> 11 #include <linux/ktime.h> 12 #include <linux/sched.h> 13 #include <linux/uidgid.h> 14 #include <linux/filter.h> 15 #include <linux/ctype.h> 16 #include <linux/jiffies.h> 17 #include <linux/pid_namespace.h> 18 #include <linux/proc_ns.h> 19 #include <linux/security.h> 20 #include <linux/btf_ids.h> 21 22 #include "../../lib/kstrtox.h" 23 24 /* If kernel subsystem is allowing eBPF programs to call this function, 25 * inside its own verifier_ops->get_func_proto() callback it should return 26 * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments 27 * 28 * Different map implementations will rely on rcu in map methods 29 * lookup/update/delete, therefore eBPF programs must run under rcu lock 30 * if program is allowed to access maps, so check rcu_read_lock_held in 31 * all three functions. 32 */ 33 BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key) 34 { 35 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 36 return (unsigned long) map->ops->map_lookup_elem(map, key); 37 } 38 39 const struct bpf_func_proto bpf_map_lookup_elem_proto = { 40 .func = bpf_map_lookup_elem, 41 .gpl_only = false, 42 .pkt_access = true, 43 .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, 44 .arg1_type = ARG_CONST_MAP_PTR, 45 .arg2_type = ARG_PTR_TO_MAP_KEY, 46 }; 47 48 BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key, 49 void *, value, u64, flags) 50 { 51 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 52 return map->ops->map_update_elem(map, key, value, flags); 53 } 54 55 const struct bpf_func_proto bpf_map_update_elem_proto = { 56 .func = bpf_map_update_elem, 57 .gpl_only = false, 58 .pkt_access = true, 59 .ret_type = RET_INTEGER, 60 .arg1_type = ARG_CONST_MAP_PTR, 61 .arg2_type = ARG_PTR_TO_MAP_KEY, 62 .arg3_type = ARG_PTR_TO_MAP_VALUE, 63 .arg4_type = ARG_ANYTHING, 64 }; 65 66 BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key) 67 { 68 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 69 return map->ops->map_delete_elem(map, key); 70 } 71 72 const struct bpf_func_proto bpf_map_delete_elem_proto = { 73 .func = bpf_map_delete_elem, 74 .gpl_only = false, 75 .pkt_access = true, 76 .ret_type = RET_INTEGER, 77 .arg1_type = ARG_CONST_MAP_PTR, 78 .arg2_type = ARG_PTR_TO_MAP_KEY, 79 }; 80 81 BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags) 82 { 83 return map->ops->map_push_elem(map, value, flags); 84 } 85 86 const struct bpf_func_proto bpf_map_push_elem_proto = { 87 .func = bpf_map_push_elem, 88 .gpl_only = false, 89 .pkt_access = true, 90 .ret_type = RET_INTEGER, 91 .arg1_type = ARG_CONST_MAP_PTR, 92 .arg2_type = ARG_PTR_TO_MAP_VALUE, 93 .arg3_type = ARG_ANYTHING, 94 }; 95 96 BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value) 97 { 98 return map->ops->map_pop_elem(map, value); 99 } 100 101 const struct bpf_func_proto bpf_map_pop_elem_proto = { 102 .func = bpf_map_pop_elem, 103 .gpl_only = false, 104 .ret_type = RET_INTEGER, 105 .arg1_type = ARG_CONST_MAP_PTR, 106 .arg2_type = ARG_PTR_TO_UNINIT_MAP_VALUE, 107 }; 108 109 BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value) 110 { 111 return map->ops->map_peek_elem(map, value); 112 } 113 114 const struct bpf_func_proto bpf_map_peek_elem_proto = { 115 .func = bpf_map_peek_elem, 116 .gpl_only = false, 117 .ret_type = RET_INTEGER, 118 .arg1_type = ARG_CONST_MAP_PTR, 119 .arg2_type = ARG_PTR_TO_UNINIT_MAP_VALUE, 120 }; 121 122 const struct bpf_func_proto bpf_get_prandom_u32_proto = { 123 .func = bpf_user_rnd_u32, 124 .gpl_only = false, 125 .ret_type = RET_INTEGER, 126 }; 127 128 BPF_CALL_0(bpf_get_smp_processor_id) 129 { 130 return smp_processor_id(); 131 } 132 133 const struct bpf_func_proto bpf_get_smp_processor_id_proto = { 134 .func = bpf_get_smp_processor_id, 135 .gpl_only = false, 136 .ret_type = RET_INTEGER, 137 }; 138 139 BPF_CALL_0(bpf_get_numa_node_id) 140 { 141 return numa_node_id(); 142 } 143 144 const struct bpf_func_proto bpf_get_numa_node_id_proto = { 145 .func = bpf_get_numa_node_id, 146 .gpl_only = false, 147 .ret_type = RET_INTEGER, 148 }; 149 150 BPF_CALL_0(bpf_ktime_get_ns) 151 { 152 /* NMI safe access to clock monotonic */ 153 return ktime_get_mono_fast_ns(); 154 } 155 156 const struct bpf_func_proto bpf_ktime_get_ns_proto = { 157 .func = bpf_ktime_get_ns, 158 .gpl_only = false, 159 .ret_type = RET_INTEGER, 160 }; 161 162 BPF_CALL_0(bpf_ktime_get_boot_ns) 163 { 164 /* NMI safe access to clock boottime */ 165 return ktime_get_boot_fast_ns(); 166 } 167 168 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = { 169 .func = bpf_ktime_get_boot_ns, 170 .gpl_only = false, 171 .ret_type = RET_INTEGER, 172 }; 173 174 BPF_CALL_0(bpf_ktime_get_coarse_ns) 175 { 176 return ktime_get_coarse_ns(); 177 } 178 179 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = { 180 .func = bpf_ktime_get_coarse_ns, 181 .gpl_only = false, 182 .ret_type = RET_INTEGER, 183 }; 184 185 BPF_CALL_0(bpf_get_current_pid_tgid) 186 { 187 struct task_struct *task = current; 188 189 if (unlikely(!task)) 190 return -EINVAL; 191 192 return (u64) task->tgid << 32 | task->pid; 193 } 194 195 const struct bpf_func_proto bpf_get_current_pid_tgid_proto = { 196 .func = bpf_get_current_pid_tgid, 197 .gpl_only = false, 198 .ret_type = RET_INTEGER, 199 }; 200 201 BPF_CALL_0(bpf_get_current_uid_gid) 202 { 203 struct task_struct *task = current; 204 kuid_t uid; 205 kgid_t gid; 206 207 if (unlikely(!task)) 208 return -EINVAL; 209 210 current_uid_gid(&uid, &gid); 211 return (u64) from_kgid(&init_user_ns, gid) << 32 | 212 from_kuid(&init_user_ns, uid); 213 } 214 215 const struct bpf_func_proto bpf_get_current_uid_gid_proto = { 216 .func = bpf_get_current_uid_gid, 217 .gpl_only = false, 218 .ret_type = RET_INTEGER, 219 }; 220 221 BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size) 222 { 223 struct task_struct *task = current; 224 225 if (unlikely(!task)) 226 goto err_clear; 227 228 strncpy(buf, task->comm, size); 229 230 /* Verifier guarantees that size > 0. For task->comm exceeding 231 * size, guarantee that buf is %NUL-terminated. Unconditionally 232 * done here to save the size test. 233 */ 234 buf[size - 1] = 0; 235 return 0; 236 err_clear: 237 memset(buf, 0, size); 238 return -EINVAL; 239 } 240 241 const struct bpf_func_proto bpf_get_current_comm_proto = { 242 .func = bpf_get_current_comm, 243 .gpl_only = false, 244 .ret_type = RET_INTEGER, 245 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 246 .arg2_type = ARG_CONST_SIZE, 247 }; 248 249 #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK) 250 251 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock) 252 { 253 arch_spinlock_t *l = (void *)lock; 254 union { 255 __u32 val; 256 arch_spinlock_t lock; 257 } u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED }; 258 259 compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0"); 260 BUILD_BUG_ON(sizeof(*l) != sizeof(__u32)); 261 BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32)); 262 arch_spin_lock(l); 263 } 264 265 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock) 266 { 267 arch_spinlock_t *l = (void *)lock; 268 269 arch_spin_unlock(l); 270 } 271 272 #else 273 274 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock) 275 { 276 atomic_t *l = (void *)lock; 277 278 BUILD_BUG_ON(sizeof(*l) != sizeof(*lock)); 279 do { 280 atomic_cond_read_relaxed(l, !VAL); 281 } while (atomic_xchg(l, 1)); 282 } 283 284 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock) 285 { 286 atomic_t *l = (void *)lock; 287 288 atomic_set_release(l, 0); 289 } 290 291 #endif 292 293 static DEFINE_PER_CPU(unsigned long, irqsave_flags); 294 295 static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock) 296 { 297 unsigned long flags; 298 299 local_irq_save(flags); 300 __bpf_spin_lock(lock); 301 __this_cpu_write(irqsave_flags, flags); 302 } 303 304 notrace BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock) 305 { 306 __bpf_spin_lock_irqsave(lock); 307 return 0; 308 } 309 310 const struct bpf_func_proto bpf_spin_lock_proto = { 311 .func = bpf_spin_lock, 312 .gpl_only = false, 313 .ret_type = RET_VOID, 314 .arg1_type = ARG_PTR_TO_SPIN_LOCK, 315 }; 316 317 static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock) 318 { 319 unsigned long flags; 320 321 flags = __this_cpu_read(irqsave_flags); 322 __bpf_spin_unlock(lock); 323 local_irq_restore(flags); 324 } 325 326 notrace BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock) 327 { 328 __bpf_spin_unlock_irqrestore(lock); 329 return 0; 330 } 331 332 const struct bpf_func_proto bpf_spin_unlock_proto = { 333 .func = bpf_spin_unlock, 334 .gpl_only = false, 335 .ret_type = RET_VOID, 336 .arg1_type = ARG_PTR_TO_SPIN_LOCK, 337 }; 338 339 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src, 340 bool lock_src) 341 { 342 struct bpf_spin_lock *lock; 343 344 if (lock_src) 345 lock = src + map->spin_lock_off; 346 else 347 lock = dst + map->spin_lock_off; 348 preempt_disable(); 349 __bpf_spin_lock_irqsave(lock); 350 copy_map_value(map, dst, src); 351 __bpf_spin_unlock_irqrestore(lock); 352 preempt_enable(); 353 } 354 355 BPF_CALL_0(bpf_jiffies64) 356 { 357 return get_jiffies_64(); 358 } 359 360 const struct bpf_func_proto bpf_jiffies64_proto = { 361 .func = bpf_jiffies64, 362 .gpl_only = false, 363 .ret_type = RET_INTEGER, 364 }; 365 366 #ifdef CONFIG_CGROUPS 367 BPF_CALL_0(bpf_get_current_cgroup_id) 368 { 369 struct cgroup *cgrp; 370 u64 cgrp_id; 371 372 rcu_read_lock(); 373 cgrp = task_dfl_cgroup(current); 374 cgrp_id = cgroup_id(cgrp); 375 rcu_read_unlock(); 376 377 return cgrp_id; 378 } 379 380 const struct bpf_func_proto bpf_get_current_cgroup_id_proto = { 381 .func = bpf_get_current_cgroup_id, 382 .gpl_only = false, 383 .ret_type = RET_INTEGER, 384 }; 385 386 BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level) 387 { 388 struct cgroup *cgrp; 389 struct cgroup *ancestor; 390 u64 cgrp_id; 391 392 rcu_read_lock(); 393 cgrp = task_dfl_cgroup(current); 394 ancestor = cgroup_ancestor(cgrp, ancestor_level); 395 cgrp_id = ancestor ? cgroup_id(ancestor) : 0; 396 rcu_read_unlock(); 397 398 return cgrp_id; 399 } 400 401 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = { 402 .func = bpf_get_current_ancestor_cgroup_id, 403 .gpl_only = false, 404 .ret_type = RET_INTEGER, 405 .arg1_type = ARG_ANYTHING, 406 }; 407 408 #ifdef CONFIG_CGROUP_BPF 409 410 BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags) 411 { 412 /* flags argument is not used now, 413 * but provides an ability to extend the API. 414 * verifier checks that its value is correct. 415 */ 416 enum bpf_cgroup_storage_type stype = cgroup_storage_type(map); 417 struct bpf_cgroup_storage *storage; 418 struct bpf_cg_run_ctx *ctx; 419 void *ptr; 420 421 /* get current cgroup storage from BPF run context */ 422 ctx = container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx); 423 storage = ctx->prog_item->cgroup_storage[stype]; 424 425 if (stype == BPF_CGROUP_STORAGE_SHARED) 426 ptr = &READ_ONCE(storage->buf)->data[0]; 427 else 428 ptr = this_cpu_ptr(storage->percpu_buf); 429 430 return (unsigned long)ptr; 431 } 432 433 const struct bpf_func_proto bpf_get_local_storage_proto = { 434 .func = bpf_get_local_storage, 435 .gpl_only = false, 436 .ret_type = RET_PTR_TO_MAP_VALUE, 437 .arg1_type = ARG_CONST_MAP_PTR, 438 .arg2_type = ARG_ANYTHING, 439 }; 440 #endif 441 442 #define BPF_STRTOX_BASE_MASK 0x1F 443 444 static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags, 445 unsigned long long *res, bool *is_negative) 446 { 447 unsigned int base = flags & BPF_STRTOX_BASE_MASK; 448 const char *cur_buf = buf; 449 size_t cur_len = buf_len; 450 unsigned int consumed; 451 size_t val_len; 452 char str[64]; 453 454 if (!buf || !buf_len || !res || !is_negative) 455 return -EINVAL; 456 457 if (base != 0 && base != 8 && base != 10 && base != 16) 458 return -EINVAL; 459 460 if (flags & ~BPF_STRTOX_BASE_MASK) 461 return -EINVAL; 462 463 while (cur_buf < buf + buf_len && isspace(*cur_buf)) 464 ++cur_buf; 465 466 *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-'); 467 if (*is_negative) 468 ++cur_buf; 469 470 consumed = cur_buf - buf; 471 cur_len -= consumed; 472 if (!cur_len) 473 return -EINVAL; 474 475 cur_len = min(cur_len, sizeof(str) - 1); 476 memcpy(str, cur_buf, cur_len); 477 str[cur_len] = '\0'; 478 cur_buf = str; 479 480 cur_buf = _parse_integer_fixup_radix(cur_buf, &base); 481 val_len = _parse_integer(cur_buf, base, res); 482 483 if (val_len & KSTRTOX_OVERFLOW) 484 return -ERANGE; 485 486 if (val_len == 0) 487 return -EINVAL; 488 489 cur_buf += val_len; 490 consumed += cur_buf - str; 491 492 return consumed; 493 } 494 495 static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags, 496 long long *res) 497 { 498 unsigned long long _res; 499 bool is_negative; 500 int err; 501 502 err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative); 503 if (err < 0) 504 return err; 505 if (is_negative) { 506 if ((long long)-_res > 0) 507 return -ERANGE; 508 *res = -_res; 509 } else { 510 if ((long long)_res < 0) 511 return -ERANGE; 512 *res = _res; 513 } 514 return err; 515 } 516 517 BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags, 518 long *, res) 519 { 520 long long _res; 521 int err; 522 523 err = __bpf_strtoll(buf, buf_len, flags, &_res); 524 if (err < 0) 525 return err; 526 if (_res != (long)_res) 527 return -ERANGE; 528 *res = _res; 529 return err; 530 } 531 532 const struct bpf_func_proto bpf_strtol_proto = { 533 .func = bpf_strtol, 534 .gpl_only = false, 535 .ret_type = RET_INTEGER, 536 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY, 537 .arg2_type = ARG_CONST_SIZE, 538 .arg3_type = ARG_ANYTHING, 539 .arg4_type = ARG_PTR_TO_LONG, 540 }; 541 542 BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags, 543 unsigned long *, res) 544 { 545 unsigned long long _res; 546 bool is_negative; 547 int err; 548 549 err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative); 550 if (err < 0) 551 return err; 552 if (is_negative) 553 return -EINVAL; 554 if (_res != (unsigned long)_res) 555 return -ERANGE; 556 *res = _res; 557 return err; 558 } 559 560 const struct bpf_func_proto bpf_strtoul_proto = { 561 .func = bpf_strtoul, 562 .gpl_only = false, 563 .ret_type = RET_INTEGER, 564 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY, 565 .arg2_type = ARG_CONST_SIZE, 566 .arg3_type = ARG_ANYTHING, 567 .arg4_type = ARG_PTR_TO_LONG, 568 }; 569 #endif 570 571 BPF_CALL_3(bpf_strncmp, const char *, s1, u32, s1_sz, const char *, s2) 572 { 573 return strncmp(s1, s2, s1_sz); 574 } 575 576 const struct bpf_func_proto bpf_strncmp_proto = { 577 .func = bpf_strncmp, 578 .gpl_only = false, 579 .ret_type = RET_INTEGER, 580 .arg1_type = ARG_PTR_TO_MEM, 581 .arg2_type = ARG_CONST_SIZE, 582 .arg3_type = ARG_PTR_TO_CONST_STR, 583 }; 584 585 BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino, 586 struct bpf_pidns_info *, nsdata, u32, size) 587 { 588 struct task_struct *task = current; 589 struct pid_namespace *pidns; 590 int err = -EINVAL; 591 592 if (unlikely(size != sizeof(struct bpf_pidns_info))) 593 goto clear; 594 595 if (unlikely((u64)(dev_t)dev != dev)) 596 goto clear; 597 598 if (unlikely(!task)) 599 goto clear; 600 601 pidns = task_active_pid_ns(task); 602 if (unlikely(!pidns)) { 603 err = -ENOENT; 604 goto clear; 605 } 606 607 if (!ns_match(&pidns->ns, (dev_t)dev, ino)) 608 goto clear; 609 610 nsdata->pid = task_pid_nr_ns(task, pidns); 611 nsdata->tgid = task_tgid_nr_ns(task, pidns); 612 return 0; 613 clear: 614 memset((void *)nsdata, 0, (size_t) size); 615 return err; 616 } 617 618 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = { 619 .func = bpf_get_ns_current_pid_tgid, 620 .gpl_only = false, 621 .ret_type = RET_INTEGER, 622 .arg1_type = ARG_ANYTHING, 623 .arg2_type = ARG_ANYTHING, 624 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 625 .arg4_type = ARG_CONST_SIZE, 626 }; 627 628 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = { 629 .func = bpf_get_raw_cpu_id, 630 .gpl_only = false, 631 .ret_type = RET_INTEGER, 632 }; 633 634 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map, 635 u64, flags, void *, data, u64, size) 636 { 637 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) 638 return -EINVAL; 639 640 return bpf_event_output(map, flags, data, size, NULL, 0, NULL); 641 } 642 643 const struct bpf_func_proto bpf_event_output_data_proto = { 644 .func = bpf_event_output_data, 645 .gpl_only = true, 646 .ret_type = RET_INTEGER, 647 .arg1_type = ARG_PTR_TO_CTX, 648 .arg2_type = ARG_CONST_MAP_PTR, 649 .arg3_type = ARG_ANYTHING, 650 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 651 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 652 }; 653 654 BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size, 655 const void __user *, user_ptr) 656 { 657 int ret = copy_from_user(dst, user_ptr, size); 658 659 if (unlikely(ret)) { 660 memset(dst, 0, size); 661 ret = -EFAULT; 662 } 663 664 return ret; 665 } 666 667 const struct bpf_func_proto bpf_copy_from_user_proto = { 668 .func = bpf_copy_from_user, 669 .gpl_only = false, 670 .ret_type = RET_INTEGER, 671 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 672 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 673 .arg3_type = ARG_ANYTHING, 674 }; 675 676 BPF_CALL_5(bpf_copy_from_user_task, void *, dst, u32, size, 677 const void __user *, user_ptr, struct task_struct *, tsk, u64, flags) 678 { 679 int ret; 680 681 /* flags is not used yet */ 682 if (unlikely(flags)) 683 return -EINVAL; 684 685 if (unlikely(!size)) 686 return 0; 687 688 ret = access_process_vm(tsk, (unsigned long)user_ptr, dst, size, 0); 689 if (ret == size) 690 return 0; 691 692 memset(dst, 0, size); 693 /* Return -EFAULT for partial read */ 694 return ret < 0 ? ret : -EFAULT; 695 } 696 697 const struct bpf_func_proto bpf_copy_from_user_task_proto = { 698 .func = bpf_copy_from_user_task, 699 .gpl_only = true, 700 .ret_type = RET_INTEGER, 701 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 702 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 703 .arg3_type = ARG_ANYTHING, 704 .arg4_type = ARG_PTR_TO_BTF_ID, 705 .arg4_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK], 706 .arg5_type = ARG_ANYTHING 707 }; 708 709 BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu) 710 { 711 if (cpu >= nr_cpu_ids) 712 return (unsigned long)NULL; 713 714 return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu); 715 } 716 717 const struct bpf_func_proto bpf_per_cpu_ptr_proto = { 718 .func = bpf_per_cpu_ptr, 719 .gpl_only = false, 720 .ret_type = RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL | MEM_RDONLY, 721 .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID, 722 .arg2_type = ARG_ANYTHING, 723 }; 724 725 BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr) 726 { 727 return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr); 728 } 729 730 const struct bpf_func_proto bpf_this_cpu_ptr_proto = { 731 .func = bpf_this_cpu_ptr, 732 .gpl_only = false, 733 .ret_type = RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY, 734 .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID, 735 }; 736 737 static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype, 738 size_t bufsz) 739 { 740 void __user *user_ptr = (__force void __user *)unsafe_ptr; 741 742 buf[0] = 0; 743 744 switch (fmt_ptype) { 745 case 's': 746 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE 747 if ((unsigned long)unsafe_ptr < TASK_SIZE) 748 return strncpy_from_user_nofault(buf, user_ptr, bufsz); 749 fallthrough; 750 #endif 751 case 'k': 752 return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz); 753 case 'u': 754 return strncpy_from_user_nofault(buf, user_ptr, bufsz); 755 } 756 757 return -EINVAL; 758 } 759 760 /* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary 761 * arguments representation. 762 */ 763 #define MAX_BPRINTF_BUF_LEN 512 764 765 /* Support executing three nested bprintf helper calls on a given CPU */ 766 #define MAX_BPRINTF_NEST_LEVEL 3 767 struct bpf_bprintf_buffers { 768 char tmp_bufs[MAX_BPRINTF_NEST_LEVEL][MAX_BPRINTF_BUF_LEN]; 769 }; 770 static DEFINE_PER_CPU(struct bpf_bprintf_buffers, bpf_bprintf_bufs); 771 static DEFINE_PER_CPU(int, bpf_bprintf_nest_level); 772 773 static int try_get_fmt_tmp_buf(char **tmp_buf) 774 { 775 struct bpf_bprintf_buffers *bufs; 776 int nest_level; 777 778 preempt_disable(); 779 nest_level = this_cpu_inc_return(bpf_bprintf_nest_level); 780 if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) { 781 this_cpu_dec(bpf_bprintf_nest_level); 782 preempt_enable(); 783 return -EBUSY; 784 } 785 bufs = this_cpu_ptr(&bpf_bprintf_bufs); 786 *tmp_buf = bufs->tmp_bufs[nest_level - 1]; 787 788 return 0; 789 } 790 791 void bpf_bprintf_cleanup(void) 792 { 793 if (this_cpu_read(bpf_bprintf_nest_level)) { 794 this_cpu_dec(bpf_bprintf_nest_level); 795 preempt_enable(); 796 } 797 } 798 799 /* 800 * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers 801 * 802 * Returns a negative value if fmt is an invalid format string or 0 otherwise. 803 * 804 * This can be used in two ways: 805 * - Format string verification only: when bin_args is NULL 806 * - Arguments preparation: in addition to the above verification, it writes in 807 * bin_args a binary representation of arguments usable by bstr_printf where 808 * pointers from BPF have been sanitized. 809 * 810 * In argument preparation mode, if 0 is returned, safe temporary buffers are 811 * allocated and bpf_bprintf_cleanup should be called to free them after use. 812 */ 813 int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args, 814 u32 **bin_args, u32 num_args) 815 { 816 char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end; 817 size_t sizeof_cur_arg, sizeof_cur_ip; 818 int err, i, num_spec = 0; 819 u64 cur_arg; 820 char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX"; 821 822 fmt_end = strnchr(fmt, fmt_size, 0); 823 if (!fmt_end) 824 return -EINVAL; 825 fmt_size = fmt_end - fmt; 826 827 if (bin_args) { 828 if (num_args && try_get_fmt_tmp_buf(&tmp_buf)) 829 return -EBUSY; 830 831 tmp_buf_end = tmp_buf + MAX_BPRINTF_BUF_LEN; 832 *bin_args = (u32 *)tmp_buf; 833 } 834 835 for (i = 0; i < fmt_size; i++) { 836 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) { 837 err = -EINVAL; 838 goto out; 839 } 840 841 if (fmt[i] != '%') 842 continue; 843 844 if (fmt[i + 1] == '%') { 845 i++; 846 continue; 847 } 848 849 if (num_spec >= num_args) { 850 err = -EINVAL; 851 goto out; 852 } 853 854 /* The string is zero-terminated so if fmt[i] != 0, we can 855 * always access fmt[i + 1], in the worst case it will be a 0 856 */ 857 i++; 858 859 /* skip optional "[0 +-][num]" width formatting field */ 860 while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' || 861 fmt[i] == ' ') 862 i++; 863 if (fmt[i] >= '1' && fmt[i] <= '9') { 864 i++; 865 while (fmt[i] >= '0' && fmt[i] <= '9') 866 i++; 867 } 868 869 if (fmt[i] == 'p') { 870 sizeof_cur_arg = sizeof(long); 871 872 if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') && 873 fmt[i + 2] == 's') { 874 fmt_ptype = fmt[i + 1]; 875 i += 2; 876 goto fmt_str; 877 } 878 879 if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) || 880 ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' || 881 fmt[i + 1] == 'x' || fmt[i + 1] == 's' || 882 fmt[i + 1] == 'S') { 883 /* just kernel pointers */ 884 if (tmp_buf) 885 cur_arg = raw_args[num_spec]; 886 i++; 887 goto nocopy_fmt; 888 } 889 890 if (fmt[i + 1] == 'B') { 891 if (tmp_buf) { 892 err = snprintf(tmp_buf, 893 (tmp_buf_end - tmp_buf), 894 "%pB", 895 (void *)(long)raw_args[num_spec]); 896 tmp_buf += (err + 1); 897 } 898 899 i++; 900 num_spec++; 901 continue; 902 } 903 904 /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */ 905 if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') || 906 (fmt[i + 2] != '4' && fmt[i + 2] != '6')) { 907 err = -EINVAL; 908 goto out; 909 } 910 911 i += 2; 912 if (!tmp_buf) 913 goto nocopy_fmt; 914 915 sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16; 916 if (tmp_buf_end - tmp_buf < sizeof_cur_ip) { 917 err = -ENOSPC; 918 goto out; 919 } 920 921 unsafe_ptr = (char *)(long)raw_args[num_spec]; 922 err = copy_from_kernel_nofault(cur_ip, unsafe_ptr, 923 sizeof_cur_ip); 924 if (err < 0) 925 memset(cur_ip, 0, sizeof_cur_ip); 926 927 /* hack: bstr_printf expects IP addresses to be 928 * pre-formatted as strings, ironically, the easiest way 929 * to do that is to call snprintf. 930 */ 931 ip_spec[2] = fmt[i - 1]; 932 ip_spec[3] = fmt[i]; 933 err = snprintf(tmp_buf, tmp_buf_end - tmp_buf, 934 ip_spec, &cur_ip); 935 936 tmp_buf += err + 1; 937 num_spec++; 938 939 continue; 940 } else if (fmt[i] == 's') { 941 fmt_ptype = fmt[i]; 942 fmt_str: 943 if (fmt[i + 1] != 0 && 944 !isspace(fmt[i + 1]) && 945 !ispunct(fmt[i + 1])) { 946 err = -EINVAL; 947 goto out; 948 } 949 950 if (!tmp_buf) 951 goto nocopy_fmt; 952 953 if (tmp_buf_end == tmp_buf) { 954 err = -ENOSPC; 955 goto out; 956 } 957 958 unsafe_ptr = (char *)(long)raw_args[num_spec]; 959 err = bpf_trace_copy_string(tmp_buf, unsafe_ptr, 960 fmt_ptype, 961 tmp_buf_end - tmp_buf); 962 if (err < 0) { 963 tmp_buf[0] = '\0'; 964 err = 1; 965 } 966 967 tmp_buf += err; 968 num_spec++; 969 970 continue; 971 } else if (fmt[i] == 'c') { 972 if (!tmp_buf) 973 goto nocopy_fmt; 974 975 if (tmp_buf_end == tmp_buf) { 976 err = -ENOSPC; 977 goto out; 978 } 979 980 *tmp_buf = raw_args[num_spec]; 981 tmp_buf++; 982 num_spec++; 983 984 continue; 985 } 986 987 sizeof_cur_arg = sizeof(int); 988 989 if (fmt[i] == 'l') { 990 sizeof_cur_arg = sizeof(long); 991 i++; 992 } 993 if (fmt[i] == 'l') { 994 sizeof_cur_arg = sizeof(long long); 995 i++; 996 } 997 998 if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' && 999 fmt[i] != 'x' && fmt[i] != 'X') { 1000 err = -EINVAL; 1001 goto out; 1002 } 1003 1004 if (tmp_buf) 1005 cur_arg = raw_args[num_spec]; 1006 nocopy_fmt: 1007 if (tmp_buf) { 1008 tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32)); 1009 if (tmp_buf_end - tmp_buf < sizeof_cur_arg) { 1010 err = -ENOSPC; 1011 goto out; 1012 } 1013 1014 if (sizeof_cur_arg == 8) { 1015 *(u32 *)tmp_buf = *(u32 *)&cur_arg; 1016 *(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1); 1017 } else { 1018 *(u32 *)tmp_buf = (u32)(long)cur_arg; 1019 } 1020 tmp_buf += sizeof_cur_arg; 1021 } 1022 num_spec++; 1023 } 1024 1025 err = 0; 1026 out: 1027 if (err) 1028 bpf_bprintf_cleanup(); 1029 return err; 1030 } 1031 1032 BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt, 1033 const void *, data, u32, data_len) 1034 { 1035 int err, num_args; 1036 u32 *bin_args; 1037 1038 if (data_len % 8 || data_len > MAX_BPRINTF_VARARGS * 8 || 1039 (data_len && !data)) 1040 return -EINVAL; 1041 num_args = data_len / 8; 1042 1043 /* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we 1044 * can safely give an unbounded size. 1045 */ 1046 err = bpf_bprintf_prepare(fmt, UINT_MAX, data, &bin_args, num_args); 1047 if (err < 0) 1048 return err; 1049 1050 err = bstr_printf(str, str_size, fmt, bin_args); 1051 1052 bpf_bprintf_cleanup(); 1053 1054 return err + 1; 1055 } 1056 1057 const struct bpf_func_proto bpf_snprintf_proto = { 1058 .func = bpf_snprintf, 1059 .gpl_only = true, 1060 .ret_type = RET_INTEGER, 1061 .arg1_type = ARG_PTR_TO_MEM_OR_NULL, 1062 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 1063 .arg3_type = ARG_PTR_TO_CONST_STR, 1064 .arg4_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 1065 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 1066 }; 1067 1068 /* BPF map elements can contain 'struct bpf_timer'. 1069 * Such map owns all of its BPF timers. 1070 * 'struct bpf_timer' is allocated as part of map element allocation 1071 * and it's zero initialized. 1072 * That space is used to keep 'struct bpf_timer_kern'. 1073 * bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and 1074 * remembers 'struct bpf_map *' pointer it's part of. 1075 * bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn. 1076 * bpf_timer_start() arms the timer. 1077 * If user space reference to a map goes to zero at this point 1078 * ops->map_release_uref callback is responsible for cancelling the timers, 1079 * freeing their memory, and decrementing prog's refcnts. 1080 * bpf_timer_cancel() cancels the timer and decrements prog's refcnt. 1081 * Inner maps can contain bpf timers as well. ops->map_release_uref is 1082 * freeing the timers when inner map is replaced or deleted by user space. 1083 */ 1084 struct bpf_hrtimer { 1085 struct hrtimer timer; 1086 struct bpf_map *map; 1087 struct bpf_prog *prog; 1088 void __rcu *callback_fn; 1089 void *value; 1090 }; 1091 1092 /* the actual struct hidden inside uapi struct bpf_timer */ 1093 struct bpf_timer_kern { 1094 struct bpf_hrtimer *timer; 1095 /* bpf_spin_lock is used here instead of spinlock_t to make 1096 * sure that it always fits into space resereved by struct bpf_timer 1097 * regardless of LOCKDEP and spinlock debug flags. 1098 */ 1099 struct bpf_spin_lock lock; 1100 } __attribute__((aligned(8))); 1101 1102 static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running); 1103 1104 static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer) 1105 { 1106 struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer); 1107 struct bpf_map *map = t->map; 1108 void *value = t->value; 1109 bpf_callback_t callback_fn; 1110 void *key; 1111 u32 idx; 1112 1113 BTF_TYPE_EMIT(struct bpf_timer); 1114 callback_fn = rcu_dereference_check(t->callback_fn, rcu_read_lock_bh_held()); 1115 if (!callback_fn) 1116 goto out; 1117 1118 /* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and 1119 * cannot be preempted by another bpf_timer_cb() on the same cpu. 1120 * Remember the timer this callback is servicing to prevent 1121 * deadlock if callback_fn() calls bpf_timer_cancel() or 1122 * bpf_map_delete_elem() on the same timer. 1123 */ 1124 this_cpu_write(hrtimer_running, t); 1125 if (map->map_type == BPF_MAP_TYPE_ARRAY) { 1126 struct bpf_array *array = container_of(map, struct bpf_array, map); 1127 1128 /* compute the key */ 1129 idx = ((char *)value - array->value) / array->elem_size; 1130 key = &idx; 1131 } else { /* hash or lru */ 1132 key = value - round_up(map->key_size, 8); 1133 } 1134 1135 callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0); 1136 /* The verifier checked that return value is zero. */ 1137 1138 this_cpu_write(hrtimer_running, NULL); 1139 out: 1140 return HRTIMER_NORESTART; 1141 } 1142 1143 BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map, 1144 u64, flags) 1145 { 1146 clockid_t clockid = flags & (MAX_CLOCKS - 1); 1147 struct bpf_hrtimer *t; 1148 int ret = 0; 1149 1150 BUILD_BUG_ON(MAX_CLOCKS != 16); 1151 BUILD_BUG_ON(sizeof(struct bpf_timer_kern) > sizeof(struct bpf_timer)); 1152 BUILD_BUG_ON(__alignof__(struct bpf_timer_kern) != __alignof__(struct bpf_timer)); 1153 1154 if (in_nmi()) 1155 return -EOPNOTSUPP; 1156 1157 if (flags >= MAX_CLOCKS || 1158 /* similar to timerfd except _ALARM variants are not supported */ 1159 (clockid != CLOCK_MONOTONIC && 1160 clockid != CLOCK_REALTIME && 1161 clockid != CLOCK_BOOTTIME)) 1162 return -EINVAL; 1163 __bpf_spin_lock_irqsave(&timer->lock); 1164 t = timer->timer; 1165 if (t) { 1166 ret = -EBUSY; 1167 goto out; 1168 } 1169 if (!atomic64_read(&map->usercnt)) { 1170 /* maps with timers must be either held by user space 1171 * or pinned in bpffs. 1172 */ 1173 ret = -EPERM; 1174 goto out; 1175 } 1176 /* allocate hrtimer via map_kmalloc to use memcg accounting */ 1177 t = bpf_map_kmalloc_node(map, sizeof(*t), GFP_ATOMIC, map->numa_node); 1178 if (!t) { 1179 ret = -ENOMEM; 1180 goto out; 1181 } 1182 t->value = (void *)timer - map->timer_off; 1183 t->map = map; 1184 t->prog = NULL; 1185 rcu_assign_pointer(t->callback_fn, NULL); 1186 hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT); 1187 t->timer.function = bpf_timer_cb; 1188 timer->timer = t; 1189 out: 1190 __bpf_spin_unlock_irqrestore(&timer->lock); 1191 return ret; 1192 } 1193 1194 static const struct bpf_func_proto bpf_timer_init_proto = { 1195 .func = bpf_timer_init, 1196 .gpl_only = true, 1197 .ret_type = RET_INTEGER, 1198 .arg1_type = ARG_PTR_TO_TIMER, 1199 .arg2_type = ARG_CONST_MAP_PTR, 1200 .arg3_type = ARG_ANYTHING, 1201 }; 1202 1203 BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn, 1204 struct bpf_prog_aux *, aux) 1205 { 1206 struct bpf_prog *prev, *prog = aux->prog; 1207 struct bpf_hrtimer *t; 1208 int ret = 0; 1209 1210 if (in_nmi()) 1211 return -EOPNOTSUPP; 1212 __bpf_spin_lock_irqsave(&timer->lock); 1213 t = timer->timer; 1214 if (!t) { 1215 ret = -EINVAL; 1216 goto out; 1217 } 1218 if (!atomic64_read(&t->map->usercnt)) { 1219 /* maps with timers must be either held by user space 1220 * or pinned in bpffs. Otherwise timer might still be 1221 * running even when bpf prog is detached and user space 1222 * is gone, since map_release_uref won't ever be called. 1223 */ 1224 ret = -EPERM; 1225 goto out; 1226 } 1227 prev = t->prog; 1228 if (prev != prog) { 1229 /* Bump prog refcnt once. Every bpf_timer_set_callback() 1230 * can pick different callback_fn-s within the same prog. 1231 */ 1232 prog = bpf_prog_inc_not_zero(prog); 1233 if (IS_ERR(prog)) { 1234 ret = PTR_ERR(prog); 1235 goto out; 1236 } 1237 if (prev) 1238 /* Drop prev prog refcnt when swapping with new prog */ 1239 bpf_prog_put(prev); 1240 t->prog = prog; 1241 } 1242 rcu_assign_pointer(t->callback_fn, callback_fn); 1243 out: 1244 __bpf_spin_unlock_irqrestore(&timer->lock); 1245 return ret; 1246 } 1247 1248 static const struct bpf_func_proto bpf_timer_set_callback_proto = { 1249 .func = bpf_timer_set_callback, 1250 .gpl_only = true, 1251 .ret_type = RET_INTEGER, 1252 .arg1_type = ARG_PTR_TO_TIMER, 1253 .arg2_type = ARG_PTR_TO_FUNC, 1254 }; 1255 1256 BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, flags) 1257 { 1258 struct bpf_hrtimer *t; 1259 int ret = 0; 1260 1261 if (in_nmi()) 1262 return -EOPNOTSUPP; 1263 if (flags) 1264 return -EINVAL; 1265 __bpf_spin_lock_irqsave(&timer->lock); 1266 t = timer->timer; 1267 if (!t || !t->prog) { 1268 ret = -EINVAL; 1269 goto out; 1270 } 1271 hrtimer_start(&t->timer, ns_to_ktime(nsecs), HRTIMER_MODE_REL_SOFT); 1272 out: 1273 __bpf_spin_unlock_irqrestore(&timer->lock); 1274 return ret; 1275 } 1276 1277 static const struct bpf_func_proto bpf_timer_start_proto = { 1278 .func = bpf_timer_start, 1279 .gpl_only = true, 1280 .ret_type = RET_INTEGER, 1281 .arg1_type = ARG_PTR_TO_TIMER, 1282 .arg2_type = ARG_ANYTHING, 1283 .arg3_type = ARG_ANYTHING, 1284 }; 1285 1286 static void drop_prog_refcnt(struct bpf_hrtimer *t) 1287 { 1288 struct bpf_prog *prog = t->prog; 1289 1290 if (prog) { 1291 bpf_prog_put(prog); 1292 t->prog = NULL; 1293 rcu_assign_pointer(t->callback_fn, NULL); 1294 } 1295 } 1296 1297 BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer) 1298 { 1299 struct bpf_hrtimer *t; 1300 int ret = 0; 1301 1302 if (in_nmi()) 1303 return -EOPNOTSUPP; 1304 __bpf_spin_lock_irqsave(&timer->lock); 1305 t = timer->timer; 1306 if (!t) { 1307 ret = -EINVAL; 1308 goto out; 1309 } 1310 if (this_cpu_read(hrtimer_running) == t) { 1311 /* If bpf callback_fn is trying to bpf_timer_cancel() 1312 * its own timer the hrtimer_cancel() will deadlock 1313 * since it waits for callback_fn to finish 1314 */ 1315 ret = -EDEADLK; 1316 goto out; 1317 } 1318 drop_prog_refcnt(t); 1319 out: 1320 __bpf_spin_unlock_irqrestore(&timer->lock); 1321 /* Cancel the timer and wait for associated callback to finish 1322 * if it was running. 1323 */ 1324 ret = ret ?: hrtimer_cancel(&t->timer); 1325 return ret; 1326 } 1327 1328 static const struct bpf_func_proto bpf_timer_cancel_proto = { 1329 .func = bpf_timer_cancel, 1330 .gpl_only = true, 1331 .ret_type = RET_INTEGER, 1332 .arg1_type = ARG_PTR_TO_TIMER, 1333 }; 1334 1335 /* This function is called by map_delete/update_elem for individual element and 1336 * by ops->map_release_uref when the user space reference to a map reaches zero. 1337 */ 1338 void bpf_timer_cancel_and_free(void *val) 1339 { 1340 struct bpf_timer_kern *timer = val; 1341 struct bpf_hrtimer *t; 1342 1343 /* Performance optimization: read timer->timer without lock first. */ 1344 if (!READ_ONCE(timer->timer)) 1345 return; 1346 1347 __bpf_spin_lock_irqsave(&timer->lock); 1348 /* re-read it under lock */ 1349 t = timer->timer; 1350 if (!t) 1351 goto out; 1352 drop_prog_refcnt(t); 1353 /* The subsequent bpf_timer_start/cancel() helpers won't be able to use 1354 * this timer, since it won't be initialized. 1355 */ 1356 timer->timer = NULL; 1357 out: 1358 __bpf_spin_unlock_irqrestore(&timer->lock); 1359 if (!t) 1360 return; 1361 /* Cancel the timer and wait for callback to complete if it was running. 1362 * If hrtimer_cancel() can be safely called it's safe to call kfree(t) 1363 * right after for both preallocated and non-preallocated maps. 1364 * The timer->timer = NULL was already done and no code path can 1365 * see address 't' anymore. 1366 * 1367 * Check that bpf_map_delete/update_elem() wasn't called from timer 1368 * callback_fn. In such case don't call hrtimer_cancel() (since it will 1369 * deadlock) and don't call hrtimer_try_to_cancel() (since it will just 1370 * return -1). Though callback_fn is still running on this cpu it's 1371 * safe to do kfree(t) because bpf_timer_cb() read everything it needed 1372 * from 't'. The bpf subprog callback_fn won't be able to access 't', 1373 * since timer->timer = NULL was already done. The timer will be 1374 * effectively cancelled because bpf_timer_cb() will return 1375 * HRTIMER_NORESTART. 1376 */ 1377 if (this_cpu_read(hrtimer_running) != t) 1378 hrtimer_cancel(&t->timer); 1379 kfree(t); 1380 } 1381 1382 const struct bpf_func_proto bpf_get_current_task_proto __weak; 1383 const struct bpf_func_proto bpf_get_current_task_btf_proto __weak; 1384 const struct bpf_func_proto bpf_probe_read_user_proto __weak; 1385 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak; 1386 const struct bpf_func_proto bpf_probe_read_kernel_proto __weak; 1387 const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak; 1388 const struct bpf_func_proto bpf_task_pt_regs_proto __weak; 1389 1390 const struct bpf_func_proto * 1391 bpf_base_func_proto(enum bpf_func_id func_id) 1392 { 1393 switch (func_id) { 1394 case BPF_FUNC_map_lookup_elem: 1395 return &bpf_map_lookup_elem_proto; 1396 case BPF_FUNC_map_update_elem: 1397 return &bpf_map_update_elem_proto; 1398 case BPF_FUNC_map_delete_elem: 1399 return &bpf_map_delete_elem_proto; 1400 case BPF_FUNC_map_push_elem: 1401 return &bpf_map_push_elem_proto; 1402 case BPF_FUNC_map_pop_elem: 1403 return &bpf_map_pop_elem_proto; 1404 case BPF_FUNC_map_peek_elem: 1405 return &bpf_map_peek_elem_proto; 1406 case BPF_FUNC_get_prandom_u32: 1407 return &bpf_get_prandom_u32_proto; 1408 case BPF_FUNC_get_smp_processor_id: 1409 return &bpf_get_raw_smp_processor_id_proto; 1410 case BPF_FUNC_get_numa_node_id: 1411 return &bpf_get_numa_node_id_proto; 1412 case BPF_FUNC_tail_call: 1413 return &bpf_tail_call_proto; 1414 case BPF_FUNC_ktime_get_ns: 1415 return &bpf_ktime_get_ns_proto; 1416 case BPF_FUNC_ktime_get_boot_ns: 1417 return &bpf_ktime_get_boot_ns_proto; 1418 case BPF_FUNC_ringbuf_output: 1419 return &bpf_ringbuf_output_proto; 1420 case BPF_FUNC_ringbuf_reserve: 1421 return &bpf_ringbuf_reserve_proto; 1422 case BPF_FUNC_ringbuf_submit: 1423 return &bpf_ringbuf_submit_proto; 1424 case BPF_FUNC_ringbuf_discard: 1425 return &bpf_ringbuf_discard_proto; 1426 case BPF_FUNC_ringbuf_query: 1427 return &bpf_ringbuf_query_proto; 1428 case BPF_FUNC_for_each_map_elem: 1429 return &bpf_for_each_map_elem_proto; 1430 case BPF_FUNC_loop: 1431 return &bpf_loop_proto; 1432 case BPF_FUNC_strncmp: 1433 return &bpf_strncmp_proto; 1434 default: 1435 break; 1436 } 1437 1438 if (!bpf_capable()) 1439 return NULL; 1440 1441 switch (func_id) { 1442 case BPF_FUNC_spin_lock: 1443 return &bpf_spin_lock_proto; 1444 case BPF_FUNC_spin_unlock: 1445 return &bpf_spin_unlock_proto; 1446 case BPF_FUNC_jiffies64: 1447 return &bpf_jiffies64_proto; 1448 case BPF_FUNC_per_cpu_ptr: 1449 return &bpf_per_cpu_ptr_proto; 1450 case BPF_FUNC_this_cpu_ptr: 1451 return &bpf_this_cpu_ptr_proto; 1452 case BPF_FUNC_timer_init: 1453 return &bpf_timer_init_proto; 1454 case BPF_FUNC_timer_set_callback: 1455 return &bpf_timer_set_callback_proto; 1456 case BPF_FUNC_timer_start: 1457 return &bpf_timer_start_proto; 1458 case BPF_FUNC_timer_cancel: 1459 return &bpf_timer_cancel_proto; 1460 default: 1461 break; 1462 } 1463 1464 if (!perfmon_capable()) 1465 return NULL; 1466 1467 switch (func_id) { 1468 case BPF_FUNC_trace_printk: 1469 return bpf_get_trace_printk_proto(); 1470 case BPF_FUNC_get_current_task: 1471 return &bpf_get_current_task_proto; 1472 case BPF_FUNC_get_current_task_btf: 1473 return &bpf_get_current_task_btf_proto; 1474 case BPF_FUNC_probe_read_user: 1475 return &bpf_probe_read_user_proto; 1476 case BPF_FUNC_probe_read_kernel: 1477 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? 1478 NULL : &bpf_probe_read_kernel_proto; 1479 case BPF_FUNC_probe_read_user_str: 1480 return &bpf_probe_read_user_str_proto; 1481 case BPF_FUNC_probe_read_kernel_str: 1482 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? 1483 NULL : &bpf_probe_read_kernel_str_proto; 1484 case BPF_FUNC_snprintf_btf: 1485 return &bpf_snprintf_btf_proto; 1486 case BPF_FUNC_snprintf: 1487 return &bpf_snprintf_proto; 1488 case BPF_FUNC_task_pt_regs: 1489 return &bpf_task_pt_regs_proto; 1490 case BPF_FUNC_trace_vprintk: 1491 return bpf_get_trace_vprintk_proto(); 1492 default: 1493 return NULL; 1494 } 1495 } 1496