1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com 3 */ 4 #ifndef _LINUX_BPF_H 5 #define _LINUX_BPF_H 1 6 7 #include <uapi/linux/bpf.h> 8 #include <uapi/linux/filter.h> 9 10 #include <linux/workqueue.h> 11 #include <linux/file.h> 12 #include <linux/percpu.h> 13 #include <linux/err.h> 14 #include <linux/rbtree_latch.h> 15 #include <linux/numa.h> 16 #include <linux/mm_types.h> 17 #include <linux/wait.h> 18 #include <linux/refcount.h> 19 #include <linux/mutex.h> 20 #include <linux/module.h> 21 #include <linux/kallsyms.h> 22 #include <linux/capability.h> 23 #include <linux/sched/mm.h> 24 #include <linux/slab.h> 25 #include <linux/percpu-refcount.h> 26 #include <linux/stddef.h> 27 #include <linux/bpfptr.h> 28 #include <linux/btf.h> 29 #include <linux/rcupdate_trace.h> 30 #include <linux/static_call.h> 31 #include <linux/memcontrol.h> 32 33 struct bpf_verifier_env; 34 struct bpf_verifier_log; 35 struct perf_event; 36 struct bpf_prog; 37 struct bpf_prog_aux; 38 struct bpf_map; 39 struct sock; 40 struct seq_file; 41 struct btf; 42 struct btf_type; 43 struct exception_table_entry; 44 struct seq_operations; 45 struct bpf_iter_aux_info; 46 struct bpf_local_storage; 47 struct bpf_local_storage_map; 48 struct kobject; 49 struct mem_cgroup; 50 struct module; 51 struct bpf_func_state; 52 struct ftrace_ops; 53 struct cgroup; 54 struct bpf_token; 55 struct user_namespace; 56 struct super_block; 57 struct inode; 58 59 extern struct idr btf_idr; 60 extern spinlock_t btf_idr_lock; 61 extern struct kobject *btf_kobj; 62 extern struct bpf_mem_alloc bpf_global_ma, bpf_global_percpu_ma; 63 extern bool bpf_global_ma_set; 64 65 typedef u64 (*bpf_callback_t)(u64, u64, u64, u64, u64); 66 typedef int (*bpf_iter_init_seq_priv_t)(void *private_data, 67 struct bpf_iter_aux_info *aux); 68 typedef void (*bpf_iter_fini_seq_priv_t)(void *private_data); 69 typedef unsigned int (*bpf_func_t)(const void *, 70 const struct bpf_insn *); 71 struct bpf_iter_seq_info { 72 const struct seq_operations *seq_ops; 73 bpf_iter_init_seq_priv_t init_seq_private; 74 bpf_iter_fini_seq_priv_t fini_seq_private; 75 u32 seq_priv_size; 76 }; 77 78 /* map is generic key/value storage optionally accessible by eBPF programs */ 79 struct bpf_map_ops { 80 /* funcs callable from userspace (via syscall) */ 81 int (*map_alloc_check)(union bpf_attr *attr); 82 struct bpf_map *(*map_alloc)(union bpf_attr *attr); 83 void (*map_release)(struct bpf_map *map, struct file *map_file); 84 void (*map_free)(struct bpf_map *map); 85 int (*map_get_next_key)(struct bpf_map *map, void *key, void *next_key); 86 void (*map_release_uref)(struct bpf_map *map); 87 void *(*map_lookup_elem_sys_only)(struct bpf_map *map, void *key); 88 int (*map_lookup_batch)(struct bpf_map *map, const union bpf_attr *attr, 89 union bpf_attr __user *uattr); 90 int (*map_lookup_and_delete_elem)(struct bpf_map *map, void *key, 91 void *value, u64 flags); 92 int (*map_lookup_and_delete_batch)(struct bpf_map *map, 93 const union bpf_attr *attr, 94 union bpf_attr __user *uattr); 95 int (*map_update_batch)(struct bpf_map *map, struct file *map_file, 96 const union bpf_attr *attr, 97 union bpf_attr __user *uattr); 98 int (*map_delete_batch)(struct bpf_map *map, const union bpf_attr *attr, 99 union bpf_attr __user *uattr); 100 101 /* funcs callable from userspace and from eBPF programs */ 102 void *(*map_lookup_elem)(struct bpf_map *map, void *key); 103 long (*map_update_elem)(struct bpf_map *map, void *key, void *value, u64 flags); 104 long (*map_delete_elem)(struct bpf_map *map, void *key); 105 long (*map_push_elem)(struct bpf_map *map, void *value, u64 flags); 106 long (*map_pop_elem)(struct bpf_map *map, void *value); 107 long (*map_peek_elem)(struct bpf_map *map, void *value); 108 void *(*map_lookup_percpu_elem)(struct bpf_map *map, void *key, u32 cpu); 109 110 /* funcs called by prog_array and perf_event_array map */ 111 void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file, 112 int fd); 113 /* If need_defer is true, the implementation should guarantee that 114 * the to-be-put element is still alive before the bpf program, which 115 * may manipulate it, exists. 116 */ 117 void (*map_fd_put_ptr)(struct bpf_map *map, void *ptr, bool need_defer); 118 int (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf); 119 u32 (*map_fd_sys_lookup_elem)(void *ptr); 120 void (*map_seq_show_elem)(struct bpf_map *map, void *key, 121 struct seq_file *m); 122 int (*map_check_btf)(const struct bpf_map *map, 123 const struct btf *btf, 124 const struct btf_type *key_type, 125 const struct btf_type *value_type); 126 127 /* Prog poke tracking helpers. */ 128 int (*map_poke_track)(struct bpf_map *map, struct bpf_prog_aux *aux); 129 void (*map_poke_untrack)(struct bpf_map *map, struct bpf_prog_aux *aux); 130 void (*map_poke_run)(struct bpf_map *map, u32 key, struct bpf_prog *old, 131 struct bpf_prog *new); 132 133 /* Direct value access helpers. */ 134 int (*map_direct_value_addr)(const struct bpf_map *map, 135 u64 *imm, u32 off); 136 int (*map_direct_value_meta)(const struct bpf_map *map, 137 u64 imm, u32 *off); 138 int (*map_mmap)(struct bpf_map *map, struct vm_area_struct *vma); 139 __poll_t (*map_poll)(struct bpf_map *map, struct file *filp, 140 struct poll_table_struct *pts); 141 142 /* Functions called by bpf_local_storage maps */ 143 int (*map_local_storage_charge)(struct bpf_local_storage_map *smap, 144 void *owner, u32 size); 145 void (*map_local_storage_uncharge)(struct bpf_local_storage_map *smap, 146 void *owner, u32 size); 147 struct bpf_local_storage __rcu ** (*map_owner_storage_ptr)(void *owner); 148 149 /* Misc helpers.*/ 150 long (*map_redirect)(struct bpf_map *map, u64 key, u64 flags); 151 152 /* map_meta_equal must be implemented for maps that can be 153 * used as an inner map. It is a runtime check to ensure 154 * an inner map can be inserted to an outer map. 155 * 156 * Some properties of the inner map has been used during the 157 * verification time. When inserting an inner map at the runtime, 158 * map_meta_equal has to ensure the inserting map has the same 159 * properties that the verifier has used earlier. 160 */ 161 bool (*map_meta_equal)(const struct bpf_map *meta0, 162 const struct bpf_map *meta1); 163 164 165 int (*map_set_for_each_callback_args)(struct bpf_verifier_env *env, 166 struct bpf_func_state *caller, 167 struct bpf_func_state *callee); 168 long (*map_for_each_callback)(struct bpf_map *map, 169 bpf_callback_t callback_fn, 170 void *callback_ctx, u64 flags); 171 172 u64 (*map_mem_usage)(const struct bpf_map *map); 173 174 /* BTF id of struct allocated by map_alloc */ 175 int *map_btf_id; 176 177 /* bpf_iter info used to open a seq_file */ 178 const struct bpf_iter_seq_info *iter_seq_info; 179 }; 180 181 enum { 182 /* Support at most 10 fields in a BTF type */ 183 BTF_FIELDS_MAX = 10, 184 }; 185 186 enum btf_field_type { 187 BPF_SPIN_LOCK = (1 << 0), 188 BPF_TIMER = (1 << 1), 189 BPF_KPTR_UNREF = (1 << 2), 190 BPF_KPTR_REF = (1 << 3), 191 BPF_KPTR_PERCPU = (1 << 4), 192 BPF_KPTR = BPF_KPTR_UNREF | BPF_KPTR_REF | BPF_KPTR_PERCPU, 193 BPF_LIST_HEAD = (1 << 5), 194 BPF_LIST_NODE = (1 << 6), 195 BPF_RB_ROOT = (1 << 7), 196 BPF_RB_NODE = (1 << 8), 197 BPF_GRAPH_NODE = BPF_RB_NODE | BPF_LIST_NODE, 198 BPF_GRAPH_ROOT = BPF_RB_ROOT | BPF_LIST_HEAD, 199 BPF_REFCOUNT = (1 << 9), 200 }; 201 202 typedef void (*btf_dtor_kfunc_t)(void *); 203 204 struct btf_field_kptr { 205 struct btf *btf; 206 struct module *module; 207 /* dtor used if btf_is_kernel(btf), otherwise the type is 208 * program-allocated, dtor is NULL, and __bpf_obj_drop_impl is used 209 */ 210 btf_dtor_kfunc_t dtor; 211 u32 btf_id; 212 }; 213 214 struct btf_field_graph_root { 215 struct btf *btf; 216 u32 value_btf_id; 217 u32 node_offset; 218 struct btf_record *value_rec; 219 }; 220 221 struct btf_field { 222 u32 offset; 223 u32 size; 224 enum btf_field_type type; 225 union { 226 struct btf_field_kptr kptr; 227 struct btf_field_graph_root graph_root; 228 }; 229 }; 230 231 struct btf_record { 232 u32 cnt; 233 u32 field_mask; 234 int spin_lock_off; 235 int timer_off; 236 int refcount_off; 237 struct btf_field fields[]; 238 }; 239 240 /* Non-opaque version of bpf_rb_node in uapi/linux/bpf.h */ 241 struct bpf_rb_node_kern { 242 struct rb_node rb_node; 243 void *owner; 244 } __attribute__((aligned(8))); 245 246 /* Non-opaque version of bpf_list_node in uapi/linux/bpf.h */ 247 struct bpf_list_node_kern { 248 struct list_head list_head; 249 void *owner; 250 } __attribute__((aligned(8))); 251 252 struct bpf_map { 253 /* The first two cachelines with read-mostly members of which some 254 * are also accessed in fast-path (e.g. ops, max_entries). 255 */ 256 const struct bpf_map_ops *ops ____cacheline_aligned; 257 struct bpf_map *inner_map_meta; 258 #ifdef CONFIG_SECURITY 259 void *security; 260 #endif 261 enum bpf_map_type map_type; 262 u32 key_size; 263 u32 value_size; 264 u32 max_entries; 265 u64 map_extra; /* any per-map-type extra fields */ 266 u32 map_flags; 267 u32 id; 268 struct btf_record *record; 269 int numa_node; 270 u32 btf_key_type_id; 271 u32 btf_value_type_id; 272 u32 btf_vmlinux_value_type_id; 273 struct btf *btf; 274 #ifdef CONFIG_MEMCG_KMEM 275 struct obj_cgroup *objcg; 276 #endif 277 char name[BPF_OBJ_NAME_LEN]; 278 /* The 3rd and 4th cacheline with misc members to avoid false sharing 279 * particularly with refcounting. 280 */ 281 atomic64_t refcnt ____cacheline_aligned; 282 atomic64_t usercnt; 283 /* rcu is used before freeing and work is only used during freeing */ 284 union { 285 struct work_struct work; 286 struct rcu_head rcu; 287 }; 288 struct mutex freeze_mutex; 289 atomic64_t writecnt; 290 /* 'Ownership' of program-containing map is claimed by the first program 291 * that is going to use this map or by the first program which FD is 292 * stored in the map to make sure that all callers and callees have the 293 * same prog type, JITed flag and xdp_has_frags flag. 294 */ 295 struct { 296 spinlock_t lock; 297 enum bpf_prog_type type; 298 bool jited; 299 bool xdp_has_frags; 300 } owner; 301 bool bypass_spec_v1; 302 bool frozen; /* write-once; write-protected by freeze_mutex */ 303 bool free_after_mult_rcu_gp; 304 bool free_after_rcu_gp; 305 atomic64_t sleepable_refcnt; 306 s64 __percpu *elem_count; 307 }; 308 309 static inline const char *btf_field_type_name(enum btf_field_type type) 310 { 311 switch (type) { 312 case BPF_SPIN_LOCK: 313 return "bpf_spin_lock"; 314 case BPF_TIMER: 315 return "bpf_timer"; 316 case BPF_KPTR_UNREF: 317 case BPF_KPTR_REF: 318 return "kptr"; 319 case BPF_KPTR_PERCPU: 320 return "percpu_kptr"; 321 case BPF_LIST_HEAD: 322 return "bpf_list_head"; 323 case BPF_LIST_NODE: 324 return "bpf_list_node"; 325 case BPF_RB_ROOT: 326 return "bpf_rb_root"; 327 case BPF_RB_NODE: 328 return "bpf_rb_node"; 329 case BPF_REFCOUNT: 330 return "bpf_refcount"; 331 default: 332 WARN_ON_ONCE(1); 333 return "unknown"; 334 } 335 } 336 337 static inline u32 btf_field_type_size(enum btf_field_type type) 338 { 339 switch (type) { 340 case BPF_SPIN_LOCK: 341 return sizeof(struct bpf_spin_lock); 342 case BPF_TIMER: 343 return sizeof(struct bpf_timer); 344 case BPF_KPTR_UNREF: 345 case BPF_KPTR_REF: 346 case BPF_KPTR_PERCPU: 347 return sizeof(u64); 348 case BPF_LIST_HEAD: 349 return sizeof(struct bpf_list_head); 350 case BPF_LIST_NODE: 351 return sizeof(struct bpf_list_node); 352 case BPF_RB_ROOT: 353 return sizeof(struct bpf_rb_root); 354 case BPF_RB_NODE: 355 return sizeof(struct bpf_rb_node); 356 case BPF_REFCOUNT: 357 return sizeof(struct bpf_refcount); 358 default: 359 WARN_ON_ONCE(1); 360 return 0; 361 } 362 } 363 364 static inline u32 btf_field_type_align(enum btf_field_type type) 365 { 366 switch (type) { 367 case BPF_SPIN_LOCK: 368 return __alignof__(struct bpf_spin_lock); 369 case BPF_TIMER: 370 return __alignof__(struct bpf_timer); 371 case BPF_KPTR_UNREF: 372 case BPF_KPTR_REF: 373 case BPF_KPTR_PERCPU: 374 return __alignof__(u64); 375 case BPF_LIST_HEAD: 376 return __alignof__(struct bpf_list_head); 377 case BPF_LIST_NODE: 378 return __alignof__(struct bpf_list_node); 379 case BPF_RB_ROOT: 380 return __alignof__(struct bpf_rb_root); 381 case BPF_RB_NODE: 382 return __alignof__(struct bpf_rb_node); 383 case BPF_REFCOUNT: 384 return __alignof__(struct bpf_refcount); 385 default: 386 WARN_ON_ONCE(1); 387 return 0; 388 } 389 } 390 391 static inline void bpf_obj_init_field(const struct btf_field *field, void *addr) 392 { 393 memset(addr, 0, field->size); 394 395 switch (field->type) { 396 case BPF_REFCOUNT: 397 refcount_set((refcount_t *)addr, 1); 398 break; 399 case BPF_RB_NODE: 400 RB_CLEAR_NODE((struct rb_node *)addr); 401 break; 402 case BPF_LIST_HEAD: 403 case BPF_LIST_NODE: 404 INIT_LIST_HEAD((struct list_head *)addr); 405 break; 406 case BPF_RB_ROOT: 407 /* RB_ROOT_CACHED 0-inits, no need to do anything after memset */ 408 case BPF_SPIN_LOCK: 409 case BPF_TIMER: 410 case BPF_KPTR_UNREF: 411 case BPF_KPTR_REF: 412 case BPF_KPTR_PERCPU: 413 break; 414 default: 415 WARN_ON_ONCE(1); 416 return; 417 } 418 } 419 420 static inline bool btf_record_has_field(const struct btf_record *rec, enum btf_field_type type) 421 { 422 if (IS_ERR_OR_NULL(rec)) 423 return false; 424 return rec->field_mask & type; 425 } 426 427 static inline void bpf_obj_init(const struct btf_record *rec, void *obj) 428 { 429 int i; 430 431 if (IS_ERR_OR_NULL(rec)) 432 return; 433 for (i = 0; i < rec->cnt; i++) 434 bpf_obj_init_field(&rec->fields[i], obj + rec->fields[i].offset); 435 } 436 437 /* 'dst' must be a temporary buffer and should not point to memory that is being 438 * used in parallel by a bpf program or bpf syscall, otherwise the access from 439 * the bpf program or bpf syscall may be corrupted by the reinitialization, 440 * leading to weird problems. Even 'dst' is newly-allocated from bpf memory 441 * allocator, it is still possible for 'dst' to be used in parallel by a bpf 442 * program or bpf syscall. 443 */ 444 static inline void check_and_init_map_value(struct bpf_map *map, void *dst) 445 { 446 bpf_obj_init(map->record, dst); 447 } 448 449 /* memcpy that is used with 8-byte aligned pointers, power-of-8 size and 450 * forced to use 'long' read/writes to try to atomically copy long counters. 451 * Best-effort only. No barriers here, since it _will_ race with concurrent 452 * updates from BPF programs. Called from bpf syscall and mostly used with 453 * size 8 or 16 bytes, so ask compiler to inline it. 454 */ 455 static inline void bpf_long_memcpy(void *dst, const void *src, u32 size) 456 { 457 const long *lsrc = src; 458 long *ldst = dst; 459 460 size /= sizeof(long); 461 while (size--) 462 data_race(*ldst++ = *lsrc++); 463 } 464 465 /* copy everything but bpf_spin_lock, bpf_timer, and kptrs. There could be one of each. */ 466 static inline void bpf_obj_memcpy(struct btf_record *rec, 467 void *dst, void *src, u32 size, 468 bool long_memcpy) 469 { 470 u32 curr_off = 0; 471 int i; 472 473 if (IS_ERR_OR_NULL(rec)) { 474 if (long_memcpy) 475 bpf_long_memcpy(dst, src, round_up(size, 8)); 476 else 477 memcpy(dst, src, size); 478 return; 479 } 480 481 for (i = 0; i < rec->cnt; i++) { 482 u32 next_off = rec->fields[i].offset; 483 u32 sz = next_off - curr_off; 484 485 memcpy(dst + curr_off, src + curr_off, sz); 486 curr_off += rec->fields[i].size + sz; 487 } 488 memcpy(dst + curr_off, src + curr_off, size - curr_off); 489 } 490 491 static inline void copy_map_value(struct bpf_map *map, void *dst, void *src) 492 { 493 bpf_obj_memcpy(map->record, dst, src, map->value_size, false); 494 } 495 496 static inline void copy_map_value_long(struct bpf_map *map, void *dst, void *src) 497 { 498 bpf_obj_memcpy(map->record, dst, src, map->value_size, true); 499 } 500 501 static inline void bpf_obj_memzero(struct btf_record *rec, void *dst, u32 size) 502 { 503 u32 curr_off = 0; 504 int i; 505 506 if (IS_ERR_OR_NULL(rec)) { 507 memset(dst, 0, size); 508 return; 509 } 510 511 for (i = 0; i < rec->cnt; i++) { 512 u32 next_off = rec->fields[i].offset; 513 u32 sz = next_off - curr_off; 514 515 memset(dst + curr_off, 0, sz); 516 curr_off += rec->fields[i].size + sz; 517 } 518 memset(dst + curr_off, 0, size - curr_off); 519 } 520 521 static inline void zero_map_value(struct bpf_map *map, void *dst) 522 { 523 bpf_obj_memzero(map->record, dst, map->value_size); 524 } 525 526 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src, 527 bool lock_src); 528 void bpf_timer_cancel_and_free(void *timer); 529 void bpf_list_head_free(const struct btf_field *field, void *list_head, 530 struct bpf_spin_lock *spin_lock); 531 void bpf_rb_root_free(const struct btf_field *field, void *rb_root, 532 struct bpf_spin_lock *spin_lock); 533 534 535 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size); 536 537 struct bpf_offload_dev; 538 struct bpf_offloaded_map; 539 540 struct bpf_map_dev_ops { 541 int (*map_get_next_key)(struct bpf_offloaded_map *map, 542 void *key, void *next_key); 543 int (*map_lookup_elem)(struct bpf_offloaded_map *map, 544 void *key, void *value); 545 int (*map_update_elem)(struct bpf_offloaded_map *map, 546 void *key, void *value, u64 flags); 547 int (*map_delete_elem)(struct bpf_offloaded_map *map, void *key); 548 }; 549 550 struct bpf_offloaded_map { 551 struct bpf_map map; 552 struct net_device *netdev; 553 const struct bpf_map_dev_ops *dev_ops; 554 void *dev_priv; 555 struct list_head offloads; 556 }; 557 558 static inline struct bpf_offloaded_map *map_to_offmap(struct bpf_map *map) 559 { 560 return container_of(map, struct bpf_offloaded_map, map); 561 } 562 563 static inline bool bpf_map_offload_neutral(const struct bpf_map *map) 564 { 565 return map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; 566 } 567 568 static inline bool bpf_map_support_seq_show(const struct bpf_map *map) 569 { 570 return (map->btf_value_type_id || map->btf_vmlinux_value_type_id) && 571 map->ops->map_seq_show_elem; 572 } 573 574 int map_check_no_btf(const struct bpf_map *map, 575 const struct btf *btf, 576 const struct btf_type *key_type, 577 const struct btf_type *value_type); 578 579 bool bpf_map_meta_equal(const struct bpf_map *meta0, 580 const struct bpf_map *meta1); 581 582 extern const struct bpf_map_ops bpf_map_offload_ops; 583 584 /* bpf_type_flag contains a set of flags that are applicable to the values of 585 * arg_type, ret_type and reg_type. For example, a pointer value may be null, 586 * or a memory is read-only. We classify types into two categories: base types 587 * and extended types. Extended types are base types combined with a type flag. 588 * 589 * Currently there are no more than 32 base types in arg_type, ret_type and 590 * reg_types. 591 */ 592 #define BPF_BASE_TYPE_BITS 8 593 594 enum bpf_type_flag { 595 /* PTR may be NULL. */ 596 PTR_MAYBE_NULL = BIT(0 + BPF_BASE_TYPE_BITS), 597 598 /* MEM is read-only. When applied on bpf_arg, it indicates the arg is 599 * compatible with both mutable and immutable memory. 600 */ 601 MEM_RDONLY = BIT(1 + BPF_BASE_TYPE_BITS), 602 603 /* MEM points to BPF ring buffer reservation. */ 604 MEM_RINGBUF = BIT(2 + BPF_BASE_TYPE_BITS), 605 606 /* MEM is in user address space. */ 607 MEM_USER = BIT(3 + BPF_BASE_TYPE_BITS), 608 609 /* MEM is a percpu memory. MEM_PERCPU tags PTR_TO_BTF_ID. When tagged 610 * with MEM_PERCPU, PTR_TO_BTF_ID _cannot_ be directly accessed. In 611 * order to drop this tag, it must be passed into bpf_per_cpu_ptr() 612 * or bpf_this_cpu_ptr(), which will return the pointer corresponding 613 * to the specified cpu. 614 */ 615 MEM_PERCPU = BIT(4 + BPF_BASE_TYPE_BITS), 616 617 /* Indicates that the argument will be released. */ 618 OBJ_RELEASE = BIT(5 + BPF_BASE_TYPE_BITS), 619 620 /* PTR is not trusted. This is only used with PTR_TO_BTF_ID, to mark 621 * unreferenced and referenced kptr loaded from map value using a load 622 * instruction, so that they can only be dereferenced but not escape the 623 * BPF program into the kernel (i.e. cannot be passed as arguments to 624 * kfunc or bpf helpers). 625 */ 626 PTR_UNTRUSTED = BIT(6 + BPF_BASE_TYPE_BITS), 627 628 MEM_UNINIT = BIT(7 + BPF_BASE_TYPE_BITS), 629 630 /* DYNPTR points to memory local to the bpf program. */ 631 DYNPTR_TYPE_LOCAL = BIT(8 + BPF_BASE_TYPE_BITS), 632 633 /* DYNPTR points to a kernel-produced ringbuf record. */ 634 DYNPTR_TYPE_RINGBUF = BIT(9 + BPF_BASE_TYPE_BITS), 635 636 /* Size is known at compile time. */ 637 MEM_FIXED_SIZE = BIT(10 + BPF_BASE_TYPE_BITS), 638 639 /* MEM is of an allocated object of type in program BTF. This is used to 640 * tag PTR_TO_BTF_ID allocated using bpf_obj_new. 641 */ 642 MEM_ALLOC = BIT(11 + BPF_BASE_TYPE_BITS), 643 644 /* PTR was passed from the kernel in a trusted context, and may be 645 * passed to KF_TRUSTED_ARGS kfuncs or BPF helper functions. 646 * Confusingly, this is _not_ the opposite of PTR_UNTRUSTED above. 647 * PTR_UNTRUSTED refers to a kptr that was read directly from a map 648 * without invoking bpf_kptr_xchg(). What we really need to know is 649 * whether a pointer is safe to pass to a kfunc or BPF helper function. 650 * While PTR_UNTRUSTED pointers are unsafe to pass to kfuncs and BPF 651 * helpers, they do not cover all possible instances of unsafe 652 * pointers. For example, a pointer that was obtained from walking a 653 * struct will _not_ get the PTR_UNTRUSTED type modifier, despite the 654 * fact that it may be NULL, invalid, etc. This is due to backwards 655 * compatibility requirements, as this was the behavior that was first 656 * introduced when kptrs were added. The behavior is now considered 657 * deprecated, and PTR_UNTRUSTED will eventually be removed. 658 * 659 * PTR_TRUSTED, on the other hand, is a pointer that the kernel 660 * guarantees to be valid and safe to pass to kfuncs and BPF helpers. 661 * For example, pointers passed to tracepoint arguments are considered 662 * PTR_TRUSTED, as are pointers that are passed to struct_ops 663 * callbacks. As alluded to above, pointers that are obtained from 664 * walking PTR_TRUSTED pointers are _not_ trusted. For example, if a 665 * struct task_struct *task is PTR_TRUSTED, then accessing 666 * task->last_wakee will lose the PTR_TRUSTED modifier when it's stored 667 * in a BPF register. Similarly, pointers passed to certain programs 668 * types such as kretprobes are not guaranteed to be valid, as they may 669 * for example contain an object that was recently freed. 670 */ 671 PTR_TRUSTED = BIT(12 + BPF_BASE_TYPE_BITS), 672 673 /* MEM is tagged with rcu and memory access needs rcu_read_lock protection. */ 674 MEM_RCU = BIT(13 + BPF_BASE_TYPE_BITS), 675 676 /* Used to tag PTR_TO_BTF_ID | MEM_ALLOC references which are non-owning. 677 * Currently only valid for linked-list and rbtree nodes. If the nodes 678 * have a bpf_refcount_field, they must be tagged MEM_RCU as well. 679 */ 680 NON_OWN_REF = BIT(14 + BPF_BASE_TYPE_BITS), 681 682 /* DYNPTR points to sk_buff */ 683 DYNPTR_TYPE_SKB = BIT(15 + BPF_BASE_TYPE_BITS), 684 685 /* DYNPTR points to xdp_buff */ 686 DYNPTR_TYPE_XDP = BIT(16 + BPF_BASE_TYPE_BITS), 687 688 __BPF_TYPE_FLAG_MAX, 689 __BPF_TYPE_LAST_FLAG = __BPF_TYPE_FLAG_MAX - 1, 690 }; 691 692 #define DYNPTR_TYPE_FLAG_MASK (DYNPTR_TYPE_LOCAL | DYNPTR_TYPE_RINGBUF | DYNPTR_TYPE_SKB \ 693 | DYNPTR_TYPE_XDP) 694 695 /* Max number of base types. */ 696 #define BPF_BASE_TYPE_LIMIT (1UL << BPF_BASE_TYPE_BITS) 697 698 /* Max number of all types. */ 699 #define BPF_TYPE_LIMIT (__BPF_TYPE_LAST_FLAG | (__BPF_TYPE_LAST_FLAG - 1)) 700 701 /* function argument constraints */ 702 enum bpf_arg_type { 703 ARG_DONTCARE = 0, /* unused argument in helper function */ 704 705 /* the following constraints used to prototype 706 * bpf_map_lookup/update/delete_elem() functions 707 */ 708 ARG_CONST_MAP_PTR, /* const argument used as pointer to bpf_map */ 709 ARG_PTR_TO_MAP_KEY, /* pointer to stack used as map key */ 710 ARG_PTR_TO_MAP_VALUE, /* pointer to stack used as map value */ 711 712 /* Used to prototype bpf_memcmp() and other functions that access data 713 * on eBPF program stack 714 */ 715 ARG_PTR_TO_MEM, /* pointer to valid memory (stack, packet, map value) */ 716 717 ARG_CONST_SIZE, /* number of bytes accessed from memory */ 718 ARG_CONST_SIZE_OR_ZERO, /* number of bytes accessed from memory or 0 */ 719 720 ARG_PTR_TO_CTX, /* pointer to context */ 721 ARG_ANYTHING, /* any (initialized) argument is ok */ 722 ARG_PTR_TO_SPIN_LOCK, /* pointer to bpf_spin_lock */ 723 ARG_PTR_TO_SOCK_COMMON, /* pointer to sock_common */ 724 ARG_PTR_TO_INT, /* pointer to int */ 725 ARG_PTR_TO_LONG, /* pointer to long */ 726 ARG_PTR_TO_SOCKET, /* pointer to bpf_sock (fullsock) */ 727 ARG_PTR_TO_BTF_ID, /* pointer to in-kernel struct */ 728 ARG_PTR_TO_RINGBUF_MEM, /* pointer to dynamically reserved ringbuf memory */ 729 ARG_CONST_ALLOC_SIZE_OR_ZERO, /* number of allocated bytes requested */ 730 ARG_PTR_TO_BTF_ID_SOCK_COMMON, /* pointer to in-kernel sock_common or bpf-mirrored bpf_sock */ 731 ARG_PTR_TO_PERCPU_BTF_ID, /* pointer to in-kernel percpu type */ 732 ARG_PTR_TO_FUNC, /* pointer to a bpf program function */ 733 ARG_PTR_TO_STACK, /* pointer to stack */ 734 ARG_PTR_TO_CONST_STR, /* pointer to a null terminated read-only string */ 735 ARG_PTR_TO_TIMER, /* pointer to bpf_timer */ 736 ARG_PTR_TO_KPTR, /* pointer to referenced kptr */ 737 ARG_PTR_TO_DYNPTR, /* pointer to bpf_dynptr. See bpf_type_flag for dynptr type */ 738 __BPF_ARG_TYPE_MAX, 739 740 /* Extended arg_types. */ 741 ARG_PTR_TO_MAP_VALUE_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_MAP_VALUE, 742 ARG_PTR_TO_MEM_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_MEM, 743 ARG_PTR_TO_CTX_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_CTX, 744 ARG_PTR_TO_SOCKET_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_SOCKET, 745 ARG_PTR_TO_STACK_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_STACK, 746 ARG_PTR_TO_BTF_ID_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_BTF_ID, 747 /* pointer to memory does not need to be initialized, helper function must fill 748 * all bytes or clear them in error case. 749 */ 750 ARG_PTR_TO_UNINIT_MEM = MEM_UNINIT | ARG_PTR_TO_MEM, 751 /* Pointer to valid memory of size known at compile time. */ 752 ARG_PTR_TO_FIXED_SIZE_MEM = MEM_FIXED_SIZE | ARG_PTR_TO_MEM, 753 754 /* This must be the last entry. Its purpose is to ensure the enum is 755 * wide enough to hold the higher bits reserved for bpf_type_flag. 756 */ 757 __BPF_ARG_TYPE_LIMIT = BPF_TYPE_LIMIT, 758 }; 759 static_assert(__BPF_ARG_TYPE_MAX <= BPF_BASE_TYPE_LIMIT); 760 761 /* type of values returned from helper functions */ 762 enum bpf_return_type { 763 RET_INTEGER, /* function returns integer */ 764 RET_VOID, /* function doesn't return anything */ 765 RET_PTR_TO_MAP_VALUE, /* returns a pointer to map elem value */ 766 RET_PTR_TO_SOCKET, /* returns a pointer to a socket */ 767 RET_PTR_TO_TCP_SOCK, /* returns a pointer to a tcp_sock */ 768 RET_PTR_TO_SOCK_COMMON, /* returns a pointer to a sock_common */ 769 RET_PTR_TO_MEM, /* returns a pointer to memory */ 770 RET_PTR_TO_MEM_OR_BTF_ID, /* returns a pointer to a valid memory or a btf_id */ 771 RET_PTR_TO_BTF_ID, /* returns a pointer to a btf_id */ 772 __BPF_RET_TYPE_MAX, 773 774 /* Extended ret_types. */ 775 RET_PTR_TO_MAP_VALUE_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_MAP_VALUE, 776 RET_PTR_TO_SOCKET_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_SOCKET, 777 RET_PTR_TO_TCP_SOCK_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_TCP_SOCK, 778 RET_PTR_TO_SOCK_COMMON_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_SOCK_COMMON, 779 RET_PTR_TO_RINGBUF_MEM_OR_NULL = PTR_MAYBE_NULL | MEM_RINGBUF | RET_PTR_TO_MEM, 780 RET_PTR_TO_DYNPTR_MEM_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_MEM, 781 RET_PTR_TO_BTF_ID_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_BTF_ID, 782 RET_PTR_TO_BTF_ID_TRUSTED = PTR_TRUSTED | RET_PTR_TO_BTF_ID, 783 784 /* This must be the last entry. Its purpose is to ensure the enum is 785 * wide enough to hold the higher bits reserved for bpf_type_flag. 786 */ 787 __BPF_RET_TYPE_LIMIT = BPF_TYPE_LIMIT, 788 }; 789 static_assert(__BPF_RET_TYPE_MAX <= BPF_BASE_TYPE_LIMIT); 790 791 /* eBPF function prototype used by verifier to allow BPF_CALLs from eBPF programs 792 * to in-kernel helper functions and for adjusting imm32 field in BPF_CALL 793 * instructions after verifying 794 */ 795 struct bpf_func_proto { 796 u64 (*func)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 797 bool gpl_only; 798 bool pkt_access; 799 bool might_sleep; 800 enum bpf_return_type ret_type; 801 union { 802 struct { 803 enum bpf_arg_type arg1_type; 804 enum bpf_arg_type arg2_type; 805 enum bpf_arg_type arg3_type; 806 enum bpf_arg_type arg4_type; 807 enum bpf_arg_type arg5_type; 808 }; 809 enum bpf_arg_type arg_type[5]; 810 }; 811 union { 812 struct { 813 u32 *arg1_btf_id; 814 u32 *arg2_btf_id; 815 u32 *arg3_btf_id; 816 u32 *arg4_btf_id; 817 u32 *arg5_btf_id; 818 }; 819 u32 *arg_btf_id[5]; 820 struct { 821 size_t arg1_size; 822 size_t arg2_size; 823 size_t arg3_size; 824 size_t arg4_size; 825 size_t arg5_size; 826 }; 827 size_t arg_size[5]; 828 }; 829 int *ret_btf_id; /* return value btf_id */ 830 bool (*allowed)(const struct bpf_prog *prog); 831 }; 832 833 /* bpf_context is intentionally undefined structure. Pointer to bpf_context is 834 * the first argument to eBPF programs. 835 * For socket filters: 'struct bpf_context *' == 'struct sk_buff *' 836 */ 837 struct bpf_context; 838 839 enum bpf_access_type { 840 BPF_READ = 1, 841 BPF_WRITE = 2 842 }; 843 844 /* types of values stored in eBPF registers */ 845 /* Pointer types represent: 846 * pointer 847 * pointer + imm 848 * pointer + (u16) var 849 * pointer + (u16) var + imm 850 * if (range > 0) then [ptr, ptr + range - off) is safe to access 851 * if (id > 0) means that some 'var' was added 852 * if (off > 0) means that 'imm' was added 853 */ 854 enum bpf_reg_type { 855 NOT_INIT = 0, /* nothing was written into register */ 856 SCALAR_VALUE, /* reg doesn't contain a valid pointer */ 857 PTR_TO_CTX, /* reg points to bpf_context */ 858 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */ 859 PTR_TO_MAP_VALUE, /* reg points to map element value */ 860 PTR_TO_MAP_KEY, /* reg points to a map element key */ 861 PTR_TO_STACK, /* reg == frame_pointer + offset */ 862 PTR_TO_PACKET_META, /* skb->data - meta_len */ 863 PTR_TO_PACKET, /* reg points to skb->data */ 864 PTR_TO_PACKET_END, /* skb->data + headlen */ 865 PTR_TO_FLOW_KEYS, /* reg points to bpf_flow_keys */ 866 PTR_TO_SOCKET, /* reg points to struct bpf_sock */ 867 PTR_TO_SOCK_COMMON, /* reg points to sock_common */ 868 PTR_TO_TCP_SOCK, /* reg points to struct tcp_sock */ 869 PTR_TO_TP_BUFFER, /* reg points to a writable raw tp's buffer */ 870 PTR_TO_XDP_SOCK, /* reg points to struct xdp_sock */ 871 /* PTR_TO_BTF_ID points to a kernel struct that does not need 872 * to be null checked by the BPF program. This does not imply the 873 * pointer is _not_ null and in practice this can easily be a null 874 * pointer when reading pointer chains. The assumption is program 875 * context will handle null pointer dereference typically via fault 876 * handling. The verifier must keep this in mind and can make no 877 * assumptions about null or non-null when doing branch analysis. 878 * Further, when passed into helpers the helpers can not, without 879 * additional context, assume the value is non-null. 880 */ 881 PTR_TO_BTF_ID, 882 /* PTR_TO_BTF_ID_OR_NULL points to a kernel struct that has not 883 * been checked for null. Used primarily to inform the verifier 884 * an explicit null check is required for this struct. 885 */ 886 PTR_TO_MEM, /* reg points to valid memory region */ 887 PTR_TO_BUF, /* reg points to a read/write buffer */ 888 PTR_TO_FUNC, /* reg points to a bpf program function */ 889 CONST_PTR_TO_DYNPTR, /* reg points to a const struct bpf_dynptr */ 890 __BPF_REG_TYPE_MAX, 891 892 /* Extended reg_types. */ 893 PTR_TO_MAP_VALUE_OR_NULL = PTR_MAYBE_NULL | PTR_TO_MAP_VALUE, 894 PTR_TO_SOCKET_OR_NULL = PTR_MAYBE_NULL | PTR_TO_SOCKET, 895 PTR_TO_SOCK_COMMON_OR_NULL = PTR_MAYBE_NULL | PTR_TO_SOCK_COMMON, 896 PTR_TO_TCP_SOCK_OR_NULL = PTR_MAYBE_NULL | PTR_TO_TCP_SOCK, 897 PTR_TO_BTF_ID_OR_NULL = PTR_MAYBE_NULL | PTR_TO_BTF_ID, 898 899 /* This must be the last entry. Its purpose is to ensure the enum is 900 * wide enough to hold the higher bits reserved for bpf_type_flag. 901 */ 902 __BPF_REG_TYPE_LIMIT = BPF_TYPE_LIMIT, 903 }; 904 static_assert(__BPF_REG_TYPE_MAX <= BPF_BASE_TYPE_LIMIT); 905 906 /* The information passed from prog-specific *_is_valid_access 907 * back to the verifier. 908 */ 909 struct bpf_insn_access_aux { 910 enum bpf_reg_type reg_type; 911 union { 912 int ctx_field_size; 913 struct { 914 struct btf *btf; 915 u32 btf_id; 916 }; 917 }; 918 struct bpf_verifier_log *log; /* for verbose logs */ 919 }; 920 921 static inline void 922 bpf_ctx_record_field_size(struct bpf_insn_access_aux *aux, u32 size) 923 { 924 aux->ctx_field_size = size; 925 } 926 927 static bool bpf_is_ldimm64(const struct bpf_insn *insn) 928 { 929 return insn->code == (BPF_LD | BPF_IMM | BPF_DW); 930 } 931 932 static inline bool bpf_pseudo_func(const struct bpf_insn *insn) 933 { 934 return bpf_is_ldimm64(insn) && insn->src_reg == BPF_PSEUDO_FUNC; 935 } 936 937 struct bpf_prog_ops { 938 int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr, 939 union bpf_attr __user *uattr); 940 }; 941 942 struct bpf_reg_state; 943 struct bpf_verifier_ops { 944 /* return eBPF function prototype for verification */ 945 const struct bpf_func_proto * 946 (*get_func_proto)(enum bpf_func_id func_id, 947 const struct bpf_prog *prog); 948 949 /* return true if 'size' wide access at offset 'off' within bpf_context 950 * with 'type' (read or write) is allowed 951 */ 952 bool (*is_valid_access)(int off, int size, enum bpf_access_type type, 953 const struct bpf_prog *prog, 954 struct bpf_insn_access_aux *info); 955 int (*gen_prologue)(struct bpf_insn *insn, bool direct_write, 956 const struct bpf_prog *prog); 957 int (*gen_ld_abs)(const struct bpf_insn *orig, 958 struct bpf_insn *insn_buf); 959 u32 (*convert_ctx_access)(enum bpf_access_type type, 960 const struct bpf_insn *src, 961 struct bpf_insn *dst, 962 struct bpf_prog *prog, u32 *target_size); 963 int (*btf_struct_access)(struct bpf_verifier_log *log, 964 const struct bpf_reg_state *reg, 965 int off, int size); 966 }; 967 968 struct bpf_prog_offload_ops { 969 /* verifier basic callbacks */ 970 int (*insn_hook)(struct bpf_verifier_env *env, 971 int insn_idx, int prev_insn_idx); 972 int (*finalize)(struct bpf_verifier_env *env); 973 /* verifier optimization callbacks (called after .finalize) */ 974 int (*replace_insn)(struct bpf_verifier_env *env, u32 off, 975 struct bpf_insn *insn); 976 int (*remove_insns)(struct bpf_verifier_env *env, u32 off, u32 cnt); 977 /* program management callbacks */ 978 int (*prepare)(struct bpf_prog *prog); 979 int (*translate)(struct bpf_prog *prog); 980 void (*destroy)(struct bpf_prog *prog); 981 }; 982 983 struct bpf_prog_offload { 984 struct bpf_prog *prog; 985 struct net_device *netdev; 986 struct bpf_offload_dev *offdev; 987 void *dev_priv; 988 struct list_head offloads; 989 bool dev_state; 990 bool opt_failed; 991 void *jited_image; 992 u32 jited_len; 993 }; 994 995 enum bpf_cgroup_storage_type { 996 BPF_CGROUP_STORAGE_SHARED, 997 BPF_CGROUP_STORAGE_PERCPU, 998 __BPF_CGROUP_STORAGE_MAX 999 }; 1000 1001 #define MAX_BPF_CGROUP_STORAGE_TYPE __BPF_CGROUP_STORAGE_MAX 1002 1003 /* The longest tracepoint has 12 args. 1004 * See include/trace/bpf_probe.h 1005 */ 1006 #define MAX_BPF_FUNC_ARGS 12 1007 1008 /* The maximum number of arguments passed through registers 1009 * a single function may have. 1010 */ 1011 #define MAX_BPF_FUNC_REG_ARGS 5 1012 1013 /* The argument is a structure. */ 1014 #define BTF_FMODEL_STRUCT_ARG BIT(0) 1015 1016 /* The argument is signed. */ 1017 #define BTF_FMODEL_SIGNED_ARG BIT(1) 1018 1019 struct btf_func_model { 1020 u8 ret_size; 1021 u8 ret_flags; 1022 u8 nr_args; 1023 u8 arg_size[MAX_BPF_FUNC_ARGS]; 1024 u8 arg_flags[MAX_BPF_FUNC_ARGS]; 1025 }; 1026 1027 /* Restore arguments before returning from trampoline to let original function 1028 * continue executing. This flag is used for fentry progs when there are no 1029 * fexit progs. 1030 */ 1031 #define BPF_TRAMP_F_RESTORE_REGS BIT(0) 1032 /* Call original function after fentry progs, but before fexit progs. 1033 * Makes sense for fentry/fexit, normal calls and indirect calls. 1034 */ 1035 #define BPF_TRAMP_F_CALL_ORIG BIT(1) 1036 /* Skip current frame and return to parent. Makes sense for fentry/fexit 1037 * programs only. Should not be used with normal calls and indirect calls. 1038 */ 1039 #define BPF_TRAMP_F_SKIP_FRAME BIT(2) 1040 /* Store IP address of the caller on the trampoline stack, 1041 * so it's available for trampoline's programs. 1042 */ 1043 #define BPF_TRAMP_F_IP_ARG BIT(3) 1044 /* Return the return value of fentry prog. Only used by bpf_struct_ops. */ 1045 #define BPF_TRAMP_F_RET_FENTRY_RET BIT(4) 1046 1047 /* Get original function from stack instead of from provided direct address. 1048 * Makes sense for trampolines with fexit or fmod_ret programs. 1049 */ 1050 #define BPF_TRAMP_F_ORIG_STACK BIT(5) 1051 1052 /* This trampoline is on a function with another ftrace_ops with IPMODIFY, 1053 * e.g., a live patch. This flag is set and cleared by ftrace call backs, 1054 */ 1055 #define BPF_TRAMP_F_SHARE_IPMODIFY BIT(6) 1056 1057 /* Indicate that current trampoline is in a tail call context. Then, it has to 1058 * cache and restore tail_call_cnt to avoid infinite tail call loop. 1059 */ 1060 #define BPF_TRAMP_F_TAIL_CALL_CTX BIT(7) 1061 1062 /* Each call __bpf_prog_enter + call bpf_func + call __bpf_prog_exit is ~50 1063 * bytes on x86. 1064 */ 1065 enum { 1066 #if defined(__s390x__) 1067 BPF_MAX_TRAMP_LINKS = 27, 1068 #else 1069 BPF_MAX_TRAMP_LINKS = 38, 1070 #endif 1071 }; 1072 1073 struct bpf_tramp_links { 1074 struct bpf_tramp_link *links[BPF_MAX_TRAMP_LINKS]; 1075 int nr_links; 1076 }; 1077 1078 struct bpf_tramp_run_ctx; 1079 1080 /* Different use cases for BPF trampoline: 1081 * 1. replace nop at the function entry (kprobe equivalent) 1082 * flags = BPF_TRAMP_F_RESTORE_REGS 1083 * fentry = a set of programs to run before returning from trampoline 1084 * 1085 * 2. replace nop at the function entry (kprobe + kretprobe equivalent) 1086 * flags = BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_SKIP_FRAME 1087 * orig_call = fentry_ip + MCOUNT_INSN_SIZE 1088 * fentry = a set of program to run before calling original function 1089 * fexit = a set of program to run after original function 1090 * 1091 * 3. replace direct call instruction anywhere in the function body 1092 * or assign a function pointer for indirect call (like tcp_congestion_ops->cong_avoid) 1093 * With flags = 0 1094 * fentry = a set of programs to run before returning from trampoline 1095 * With flags = BPF_TRAMP_F_CALL_ORIG 1096 * orig_call = original callback addr or direct function addr 1097 * fentry = a set of program to run before calling original function 1098 * fexit = a set of program to run after original function 1099 */ 1100 struct bpf_tramp_image; 1101 int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *image_end, 1102 const struct btf_func_model *m, u32 flags, 1103 struct bpf_tramp_links *tlinks, 1104 void *func_addr); 1105 void *arch_alloc_bpf_trampoline(unsigned int size); 1106 void arch_free_bpf_trampoline(void *image, unsigned int size); 1107 void arch_protect_bpf_trampoline(void *image, unsigned int size); 1108 void arch_unprotect_bpf_trampoline(void *image, unsigned int size); 1109 int arch_bpf_trampoline_size(const struct btf_func_model *m, u32 flags, 1110 struct bpf_tramp_links *tlinks, void *func_addr); 1111 1112 u64 notrace __bpf_prog_enter_sleepable_recur(struct bpf_prog *prog, 1113 struct bpf_tramp_run_ctx *run_ctx); 1114 void notrace __bpf_prog_exit_sleepable_recur(struct bpf_prog *prog, u64 start, 1115 struct bpf_tramp_run_ctx *run_ctx); 1116 void notrace __bpf_tramp_enter(struct bpf_tramp_image *tr); 1117 void notrace __bpf_tramp_exit(struct bpf_tramp_image *tr); 1118 typedef u64 (*bpf_trampoline_enter_t)(struct bpf_prog *prog, 1119 struct bpf_tramp_run_ctx *run_ctx); 1120 typedef void (*bpf_trampoline_exit_t)(struct bpf_prog *prog, u64 start, 1121 struct bpf_tramp_run_ctx *run_ctx); 1122 bpf_trampoline_enter_t bpf_trampoline_enter(const struct bpf_prog *prog); 1123 bpf_trampoline_exit_t bpf_trampoline_exit(const struct bpf_prog *prog); 1124 1125 struct bpf_ksym { 1126 unsigned long start; 1127 unsigned long end; 1128 char name[KSYM_NAME_LEN]; 1129 struct list_head lnode; 1130 struct latch_tree_node tnode; 1131 bool prog; 1132 }; 1133 1134 enum bpf_tramp_prog_type { 1135 BPF_TRAMP_FENTRY, 1136 BPF_TRAMP_FEXIT, 1137 BPF_TRAMP_MODIFY_RETURN, 1138 BPF_TRAMP_MAX, 1139 BPF_TRAMP_REPLACE, /* more than MAX */ 1140 }; 1141 1142 struct bpf_tramp_image { 1143 void *image; 1144 int size; 1145 struct bpf_ksym ksym; 1146 struct percpu_ref pcref; 1147 void *ip_after_call; 1148 void *ip_epilogue; 1149 union { 1150 struct rcu_head rcu; 1151 struct work_struct work; 1152 }; 1153 }; 1154 1155 struct bpf_trampoline { 1156 /* hlist for trampoline_table */ 1157 struct hlist_node hlist; 1158 struct ftrace_ops *fops; 1159 /* serializes access to fields of this trampoline */ 1160 struct mutex mutex; 1161 refcount_t refcnt; 1162 u32 flags; 1163 u64 key; 1164 struct { 1165 struct btf_func_model model; 1166 void *addr; 1167 bool ftrace_managed; 1168 } func; 1169 /* if !NULL this is BPF_PROG_TYPE_EXT program that extends another BPF 1170 * program by replacing one of its functions. func.addr is the address 1171 * of the function it replaced. 1172 */ 1173 struct bpf_prog *extension_prog; 1174 /* list of BPF programs using this trampoline */ 1175 struct hlist_head progs_hlist[BPF_TRAMP_MAX]; 1176 /* Number of attached programs. A counter per kind. */ 1177 int progs_cnt[BPF_TRAMP_MAX]; 1178 /* Executable image of trampoline */ 1179 struct bpf_tramp_image *cur_image; 1180 struct module *mod; 1181 }; 1182 1183 struct bpf_attach_target_info { 1184 struct btf_func_model fmodel; 1185 long tgt_addr; 1186 struct module *tgt_mod; 1187 const char *tgt_name; 1188 const struct btf_type *tgt_type; 1189 }; 1190 1191 #define BPF_DISPATCHER_MAX 48 /* Fits in 2048B */ 1192 1193 struct bpf_dispatcher_prog { 1194 struct bpf_prog *prog; 1195 refcount_t users; 1196 }; 1197 1198 struct bpf_dispatcher { 1199 /* dispatcher mutex */ 1200 struct mutex mutex; 1201 void *func; 1202 struct bpf_dispatcher_prog progs[BPF_DISPATCHER_MAX]; 1203 int num_progs; 1204 void *image; 1205 void *rw_image; 1206 u32 image_off; 1207 struct bpf_ksym ksym; 1208 #ifdef CONFIG_HAVE_STATIC_CALL 1209 struct static_call_key *sc_key; 1210 void *sc_tramp; 1211 #endif 1212 }; 1213 1214 static __always_inline __nocfi unsigned int bpf_dispatcher_nop_func( 1215 const void *ctx, 1216 const struct bpf_insn *insnsi, 1217 bpf_func_t bpf_func) 1218 { 1219 return bpf_func(ctx, insnsi); 1220 } 1221 1222 /* the implementation of the opaque uapi struct bpf_dynptr */ 1223 struct bpf_dynptr_kern { 1224 void *data; 1225 /* Size represents the number of usable bytes of dynptr data. 1226 * If for example the offset is at 4 for a local dynptr whose data is 1227 * of type u64, the number of usable bytes is 4. 1228 * 1229 * The upper 8 bits are reserved. It is as follows: 1230 * Bits 0 - 23 = size 1231 * Bits 24 - 30 = dynptr type 1232 * Bit 31 = whether dynptr is read-only 1233 */ 1234 u32 size; 1235 u32 offset; 1236 } __aligned(8); 1237 1238 enum bpf_dynptr_type { 1239 BPF_DYNPTR_TYPE_INVALID, 1240 /* Points to memory that is local to the bpf program */ 1241 BPF_DYNPTR_TYPE_LOCAL, 1242 /* Underlying data is a ringbuf record */ 1243 BPF_DYNPTR_TYPE_RINGBUF, 1244 /* Underlying data is a sk_buff */ 1245 BPF_DYNPTR_TYPE_SKB, 1246 /* Underlying data is a xdp_buff */ 1247 BPF_DYNPTR_TYPE_XDP, 1248 }; 1249 1250 int bpf_dynptr_check_size(u32 size); 1251 u32 __bpf_dynptr_size(const struct bpf_dynptr_kern *ptr); 1252 const void *__bpf_dynptr_data(const struct bpf_dynptr_kern *ptr, u32 len); 1253 void *__bpf_dynptr_data_rw(const struct bpf_dynptr_kern *ptr, u32 len); 1254 1255 #ifdef CONFIG_BPF_JIT 1256 int bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_trampoline *tr); 1257 int bpf_trampoline_unlink_prog(struct bpf_tramp_link *link, struct bpf_trampoline *tr); 1258 struct bpf_trampoline *bpf_trampoline_get(u64 key, 1259 struct bpf_attach_target_info *tgt_info); 1260 void bpf_trampoline_put(struct bpf_trampoline *tr); 1261 int arch_prepare_bpf_dispatcher(void *image, void *buf, s64 *funcs, int num_funcs); 1262 1263 /* 1264 * When the architecture supports STATIC_CALL replace the bpf_dispatcher_fn 1265 * indirection with a direct call to the bpf program. If the architecture does 1266 * not have STATIC_CALL, avoid a double-indirection. 1267 */ 1268 #ifdef CONFIG_HAVE_STATIC_CALL 1269 1270 #define __BPF_DISPATCHER_SC_INIT(_name) \ 1271 .sc_key = &STATIC_CALL_KEY(_name), \ 1272 .sc_tramp = STATIC_CALL_TRAMP_ADDR(_name), 1273 1274 #define __BPF_DISPATCHER_SC(name) \ 1275 DEFINE_STATIC_CALL(bpf_dispatcher_##name##_call, bpf_dispatcher_nop_func) 1276 1277 #define __BPF_DISPATCHER_CALL(name) \ 1278 static_call(bpf_dispatcher_##name##_call)(ctx, insnsi, bpf_func) 1279 1280 #define __BPF_DISPATCHER_UPDATE(_d, _new) \ 1281 __static_call_update((_d)->sc_key, (_d)->sc_tramp, (_new)) 1282 1283 #else 1284 #define __BPF_DISPATCHER_SC_INIT(name) 1285 #define __BPF_DISPATCHER_SC(name) 1286 #define __BPF_DISPATCHER_CALL(name) bpf_func(ctx, insnsi) 1287 #define __BPF_DISPATCHER_UPDATE(_d, _new) 1288 #endif 1289 1290 #define BPF_DISPATCHER_INIT(_name) { \ 1291 .mutex = __MUTEX_INITIALIZER(_name.mutex), \ 1292 .func = &_name##_func, \ 1293 .progs = {}, \ 1294 .num_progs = 0, \ 1295 .image = NULL, \ 1296 .image_off = 0, \ 1297 .ksym = { \ 1298 .name = #_name, \ 1299 .lnode = LIST_HEAD_INIT(_name.ksym.lnode), \ 1300 }, \ 1301 __BPF_DISPATCHER_SC_INIT(_name##_call) \ 1302 } 1303 1304 #define DEFINE_BPF_DISPATCHER(name) \ 1305 __BPF_DISPATCHER_SC(name); \ 1306 noinline __nocfi unsigned int bpf_dispatcher_##name##_func( \ 1307 const void *ctx, \ 1308 const struct bpf_insn *insnsi, \ 1309 bpf_func_t bpf_func) \ 1310 { \ 1311 return __BPF_DISPATCHER_CALL(name); \ 1312 } \ 1313 EXPORT_SYMBOL(bpf_dispatcher_##name##_func); \ 1314 struct bpf_dispatcher bpf_dispatcher_##name = \ 1315 BPF_DISPATCHER_INIT(bpf_dispatcher_##name); 1316 1317 #define DECLARE_BPF_DISPATCHER(name) \ 1318 unsigned int bpf_dispatcher_##name##_func( \ 1319 const void *ctx, \ 1320 const struct bpf_insn *insnsi, \ 1321 bpf_func_t bpf_func); \ 1322 extern struct bpf_dispatcher bpf_dispatcher_##name; 1323 1324 #define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_##name##_func 1325 #define BPF_DISPATCHER_PTR(name) (&bpf_dispatcher_##name) 1326 void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, struct bpf_prog *from, 1327 struct bpf_prog *to); 1328 /* Called only from JIT-enabled code, so there's no need for stubs. */ 1329 void bpf_image_ksym_add(void *data, unsigned int size, struct bpf_ksym *ksym); 1330 void bpf_image_ksym_del(struct bpf_ksym *ksym); 1331 void bpf_ksym_add(struct bpf_ksym *ksym); 1332 void bpf_ksym_del(struct bpf_ksym *ksym); 1333 int bpf_jit_charge_modmem(u32 size); 1334 void bpf_jit_uncharge_modmem(u32 size); 1335 bool bpf_prog_has_trampoline(const struct bpf_prog *prog); 1336 #else 1337 static inline int bpf_trampoline_link_prog(struct bpf_tramp_link *link, 1338 struct bpf_trampoline *tr) 1339 { 1340 return -ENOTSUPP; 1341 } 1342 static inline int bpf_trampoline_unlink_prog(struct bpf_tramp_link *link, 1343 struct bpf_trampoline *tr) 1344 { 1345 return -ENOTSUPP; 1346 } 1347 static inline struct bpf_trampoline *bpf_trampoline_get(u64 key, 1348 struct bpf_attach_target_info *tgt_info) 1349 { 1350 return NULL; 1351 } 1352 static inline void bpf_trampoline_put(struct bpf_trampoline *tr) {} 1353 #define DEFINE_BPF_DISPATCHER(name) 1354 #define DECLARE_BPF_DISPATCHER(name) 1355 #define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_nop_func 1356 #define BPF_DISPATCHER_PTR(name) NULL 1357 static inline void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, 1358 struct bpf_prog *from, 1359 struct bpf_prog *to) {} 1360 static inline bool is_bpf_image_address(unsigned long address) 1361 { 1362 return false; 1363 } 1364 static inline bool bpf_prog_has_trampoline(const struct bpf_prog *prog) 1365 { 1366 return false; 1367 } 1368 #endif 1369 1370 struct bpf_func_info_aux { 1371 u16 linkage; 1372 bool unreliable; 1373 bool called : 1; 1374 bool verified : 1; 1375 }; 1376 1377 enum bpf_jit_poke_reason { 1378 BPF_POKE_REASON_TAIL_CALL, 1379 }; 1380 1381 /* Descriptor of pokes pointing /into/ the JITed image. */ 1382 struct bpf_jit_poke_descriptor { 1383 void *tailcall_target; 1384 void *tailcall_bypass; 1385 void *bypass_addr; 1386 void *aux; 1387 union { 1388 struct { 1389 struct bpf_map *map; 1390 u32 key; 1391 } tail_call; 1392 }; 1393 bool tailcall_target_stable; 1394 u8 adj_off; 1395 u16 reason; 1396 u32 insn_idx; 1397 }; 1398 1399 /* reg_type info for ctx arguments */ 1400 struct bpf_ctx_arg_aux { 1401 u32 offset; 1402 enum bpf_reg_type reg_type; 1403 u32 btf_id; 1404 }; 1405 1406 struct btf_mod_pair { 1407 struct btf *btf; 1408 struct module *module; 1409 }; 1410 1411 struct bpf_kfunc_desc_tab; 1412 1413 struct bpf_prog_aux { 1414 atomic64_t refcnt; 1415 u32 used_map_cnt; 1416 u32 used_btf_cnt; 1417 u32 max_ctx_offset; 1418 u32 max_pkt_offset; 1419 u32 max_tp_access; 1420 u32 stack_depth; 1421 u32 id; 1422 u32 func_cnt; /* used by non-func prog as the number of func progs */ 1423 u32 real_func_cnt; /* includes hidden progs, only used for JIT and freeing progs */ 1424 u32 func_idx; /* 0 for non-func prog, the index in func array for func prog */ 1425 u32 attach_btf_id; /* in-kernel BTF type id to attach to */ 1426 u32 ctx_arg_info_size; 1427 u32 max_rdonly_access; 1428 u32 max_rdwr_access; 1429 struct btf *attach_btf; 1430 const struct bpf_ctx_arg_aux *ctx_arg_info; 1431 struct mutex dst_mutex; /* protects dst_* pointers below, *after* prog becomes visible */ 1432 struct bpf_prog *dst_prog; 1433 struct bpf_trampoline *dst_trampoline; 1434 enum bpf_prog_type saved_dst_prog_type; 1435 enum bpf_attach_type saved_dst_attach_type; 1436 bool verifier_zext; /* Zero extensions has been inserted by verifier. */ 1437 bool dev_bound; /* Program is bound to the netdev. */ 1438 bool offload_requested; /* Program is bound and offloaded to the netdev. */ 1439 bool attach_btf_trace; /* true if attaching to BTF-enabled raw tp */ 1440 bool func_proto_unreliable; 1441 bool sleepable; 1442 bool tail_call_reachable; 1443 bool xdp_has_frags; 1444 bool exception_cb; 1445 bool exception_boundary; 1446 /* BTF_KIND_FUNC_PROTO for valid attach_btf_id */ 1447 const struct btf_type *attach_func_proto; 1448 /* function name for valid attach_btf_id */ 1449 const char *attach_func_name; 1450 struct bpf_prog **func; 1451 void *jit_data; /* JIT specific data. arch dependent */ 1452 struct bpf_jit_poke_descriptor *poke_tab; 1453 struct bpf_kfunc_desc_tab *kfunc_tab; 1454 struct bpf_kfunc_btf_tab *kfunc_btf_tab; 1455 u32 size_poke_tab; 1456 struct bpf_ksym ksym; 1457 const struct bpf_prog_ops *ops; 1458 struct bpf_map **used_maps; 1459 struct mutex used_maps_mutex; /* mutex for used_maps and used_map_cnt */ 1460 struct btf_mod_pair *used_btfs; 1461 struct bpf_prog *prog; 1462 struct user_struct *user; 1463 u64 load_time; /* ns since boottime */ 1464 u32 verified_insns; 1465 int cgroup_atype; /* enum cgroup_bpf_attach_type */ 1466 struct bpf_map *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]; 1467 char name[BPF_OBJ_NAME_LEN]; 1468 unsigned int (*bpf_exception_cb)(u64 cookie, u64 sp, u64 bp); 1469 #ifdef CONFIG_SECURITY 1470 void *security; 1471 #endif 1472 struct bpf_token *token; 1473 struct bpf_prog_offload *offload; 1474 struct btf *btf; 1475 struct bpf_func_info *func_info; 1476 struct bpf_func_info_aux *func_info_aux; 1477 /* bpf_line_info loaded from userspace. linfo->insn_off 1478 * has the xlated insn offset. 1479 * Both the main and sub prog share the same linfo. 1480 * The subprog can access its first linfo by 1481 * using the linfo_idx. 1482 */ 1483 struct bpf_line_info *linfo; 1484 /* jited_linfo is the jited addr of the linfo. It has a 1485 * one to one mapping to linfo: 1486 * jited_linfo[i] is the jited addr for the linfo[i]->insn_off. 1487 * Both the main and sub prog share the same jited_linfo. 1488 * The subprog can access its first jited_linfo by 1489 * using the linfo_idx. 1490 */ 1491 void **jited_linfo; 1492 u32 func_info_cnt; 1493 u32 nr_linfo; 1494 /* subprog can use linfo_idx to access its first linfo and 1495 * jited_linfo. 1496 * main prog always has linfo_idx == 0 1497 */ 1498 u32 linfo_idx; 1499 struct module *mod; 1500 u32 num_exentries; 1501 struct exception_table_entry *extable; 1502 union { 1503 struct work_struct work; 1504 struct rcu_head rcu; 1505 }; 1506 }; 1507 1508 struct bpf_prog { 1509 u16 pages; /* Number of allocated pages */ 1510 u16 jited:1, /* Is our filter JIT'ed? */ 1511 jit_requested:1,/* archs need to JIT the prog */ 1512 gpl_compatible:1, /* Is filter GPL compatible? */ 1513 cb_access:1, /* Is control block accessed? */ 1514 dst_needed:1, /* Do we need dst entry? */ 1515 blinding_requested:1, /* needs constant blinding */ 1516 blinded:1, /* Was blinded */ 1517 is_func:1, /* program is a bpf function */ 1518 kprobe_override:1, /* Do we override a kprobe? */ 1519 has_callchain_buf:1, /* callchain buffer allocated? */ 1520 enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */ 1521 call_get_stack:1, /* Do we call bpf_get_stack() or bpf_get_stackid() */ 1522 call_get_func_ip:1, /* Do we call get_func_ip() */ 1523 tstamp_type_access:1; /* Accessed __sk_buff->tstamp_type */ 1524 enum bpf_prog_type type; /* Type of BPF program */ 1525 enum bpf_attach_type expected_attach_type; /* For some prog types */ 1526 u32 len; /* Number of filter blocks */ 1527 u32 jited_len; /* Size of jited insns in bytes */ 1528 u8 tag[BPF_TAG_SIZE]; 1529 struct bpf_prog_stats __percpu *stats; 1530 int __percpu *active; 1531 unsigned int (*bpf_func)(const void *ctx, 1532 const struct bpf_insn *insn); 1533 struct bpf_prog_aux *aux; /* Auxiliary fields */ 1534 struct sock_fprog_kern *orig_prog; /* Original BPF program */ 1535 /* Instructions for interpreter */ 1536 union { 1537 DECLARE_FLEX_ARRAY(struct sock_filter, insns); 1538 DECLARE_FLEX_ARRAY(struct bpf_insn, insnsi); 1539 }; 1540 }; 1541 1542 struct bpf_array_aux { 1543 /* Programs with direct jumps into programs part of this array. */ 1544 struct list_head poke_progs; 1545 struct bpf_map *map; 1546 struct mutex poke_mutex; 1547 struct work_struct work; 1548 }; 1549 1550 struct bpf_link { 1551 atomic64_t refcnt; 1552 u32 id; 1553 enum bpf_link_type type; 1554 const struct bpf_link_ops *ops; 1555 struct bpf_prog *prog; 1556 struct work_struct work; 1557 }; 1558 1559 struct bpf_link_ops { 1560 void (*release)(struct bpf_link *link); 1561 void (*dealloc)(struct bpf_link *link); 1562 int (*detach)(struct bpf_link *link); 1563 int (*update_prog)(struct bpf_link *link, struct bpf_prog *new_prog, 1564 struct bpf_prog *old_prog); 1565 void (*show_fdinfo)(const struct bpf_link *link, struct seq_file *seq); 1566 int (*fill_link_info)(const struct bpf_link *link, 1567 struct bpf_link_info *info); 1568 int (*update_map)(struct bpf_link *link, struct bpf_map *new_map, 1569 struct bpf_map *old_map); 1570 }; 1571 1572 struct bpf_tramp_link { 1573 struct bpf_link link; 1574 struct hlist_node tramp_hlist; 1575 u64 cookie; 1576 }; 1577 1578 struct bpf_shim_tramp_link { 1579 struct bpf_tramp_link link; 1580 struct bpf_trampoline *trampoline; 1581 }; 1582 1583 struct bpf_tracing_link { 1584 struct bpf_tramp_link link; 1585 enum bpf_attach_type attach_type; 1586 struct bpf_trampoline *trampoline; 1587 struct bpf_prog *tgt_prog; 1588 }; 1589 1590 struct bpf_link_primer { 1591 struct bpf_link *link; 1592 struct file *file; 1593 int fd; 1594 u32 id; 1595 }; 1596 1597 struct bpf_mount_opts { 1598 umode_t mode; 1599 1600 /* BPF token-related delegation options */ 1601 u64 delegate_cmds; 1602 u64 delegate_maps; 1603 u64 delegate_progs; 1604 u64 delegate_attachs; 1605 }; 1606 1607 struct bpf_token { 1608 struct work_struct work; 1609 atomic64_t refcnt; 1610 struct user_namespace *userns; 1611 u64 allowed_cmds; 1612 u64 allowed_maps; 1613 u64 allowed_progs; 1614 u64 allowed_attachs; 1615 #ifdef CONFIG_SECURITY 1616 void *security; 1617 #endif 1618 }; 1619 1620 struct bpf_struct_ops_value; 1621 struct btf_member; 1622 1623 #define BPF_STRUCT_OPS_MAX_NR_MEMBERS 64 1624 /** 1625 * struct bpf_struct_ops - A structure of callbacks allowing a subsystem to 1626 * define a BPF_MAP_TYPE_STRUCT_OPS map type composed 1627 * of BPF_PROG_TYPE_STRUCT_OPS progs. 1628 * @verifier_ops: A structure of callbacks that are invoked by the verifier 1629 * when determining whether the struct_ops progs in the 1630 * struct_ops map are valid. 1631 * @init: A callback that is invoked a single time, and before any other 1632 * callback, to initialize the structure. A nonzero return value means 1633 * the subsystem could not be initialized. 1634 * @check_member: When defined, a callback invoked by the verifier to allow 1635 * the subsystem to determine if an entry in the struct_ops map 1636 * is valid. A nonzero return value means that the map is 1637 * invalid and should be rejected by the verifier. 1638 * @init_member: A callback that is invoked for each member of the struct_ops 1639 * map to allow the subsystem to initialize the member. A nonzero 1640 * value means the member could not be initialized. This callback 1641 * is exclusive with the @type, @type_id, @value_type, and 1642 * @value_id fields. 1643 * @reg: A callback that is invoked when the struct_ops map has been 1644 * initialized and is being attached to. Zero means the struct_ops map 1645 * has been successfully registered and is live. A nonzero return value 1646 * means the struct_ops map could not be registered. 1647 * @unreg: A callback that is invoked when the struct_ops map should be 1648 * unregistered. 1649 * @update: A callback that is invoked when the live struct_ops map is being 1650 * updated to contain new values. This callback is only invoked when 1651 * the struct_ops map is loaded with BPF_F_LINK. If not defined, the 1652 * it is assumed that the struct_ops map cannot be updated. 1653 * @validate: A callback that is invoked after all of the members have been 1654 * initialized. This callback should perform static checks on the 1655 * map, meaning that it should either fail or succeed 1656 * deterministically. A struct_ops map that has been validated may 1657 * not necessarily succeed in being registered if the call to @reg 1658 * fails. For example, a valid struct_ops map may be loaded, but 1659 * then fail to be registered due to there being another active 1660 * struct_ops map on the system in the subsystem already. For this 1661 * reason, if this callback is not defined, the check is skipped as 1662 * the struct_ops map will have final verification performed in 1663 * @reg. 1664 * @type: BTF type. 1665 * @value_type: Value type. 1666 * @name: The name of the struct bpf_struct_ops object. 1667 * @func_models: Func models 1668 * @type_id: BTF type id. 1669 * @value_id: BTF value id. 1670 */ 1671 struct bpf_struct_ops { 1672 const struct bpf_verifier_ops *verifier_ops; 1673 int (*init)(struct btf *btf); 1674 int (*check_member)(const struct btf_type *t, 1675 const struct btf_member *member, 1676 const struct bpf_prog *prog); 1677 int (*init_member)(const struct btf_type *t, 1678 const struct btf_member *member, 1679 void *kdata, const void *udata); 1680 int (*reg)(void *kdata); 1681 void (*unreg)(void *kdata); 1682 int (*update)(void *kdata, void *old_kdata); 1683 int (*validate)(void *kdata); 1684 const struct btf_type *type; 1685 const struct btf_type *value_type; 1686 const char *name; 1687 struct btf_func_model func_models[BPF_STRUCT_OPS_MAX_NR_MEMBERS]; 1688 u32 type_id; 1689 u32 value_id; 1690 }; 1691 1692 #if defined(CONFIG_BPF_JIT) && defined(CONFIG_BPF_SYSCALL) 1693 #define BPF_MODULE_OWNER ((void *)((0xeB9FUL << 2) + POISON_POINTER_DELTA)) 1694 const struct bpf_struct_ops *bpf_struct_ops_find(u32 type_id); 1695 void bpf_struct_ops_init(struct btf *btf, struct bpf_verifier_log *log); 1696 bool bpf_struct_ops_get(const void *kdata); 1697 void bpf_struct_ops_put(const void *kdata); 1698 int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, void *key, 1699 void *value); 1700 int bpf_struct_ops_prepare_trampoline(struct bpf_tramp_links *tlinks, 1701 struct bpf_tramp_link *link, 1702 const struct btf_func_model *model, 1703 void *image, void *image_end); 1704 static inline bool bpf_try_module_get(const void *data, struct module *owner) 1705 { 1706 if (owner == BPF_MODULE_OWNER) 1707 return bpf_struct_ops_get(data); 1708 else 1709 return try_module_get(owner); 1710 } 1711 static inline void bpf_module_put(const void *data, struct module *owner) 1712 { 1713 if (owner == BPF_MODULE_OWNER) 1714 bpf_struct_ops_put(data); 1715 else 1716 module_put(owner); 1717 } 1718 int bpf_struct_ops_link_create(union bpf_attr *attr); 1719 1720 #ifdef CONFIG_NET 1721 /* Define it here to avoid the use of forward declaration */ 1722 struct bpf_dummy_ops_state { 1723 int val; 1724 }; 1725 1726 struct bpf_dummy_ops { 1727 int (*test_1)(struct bpf_dummy_ops_state *cb); 1728 int (*test_2)(struct bpf_dummy_ops_state *cb, int a1, unsigned short a2, 1729 char a3, unsigned long a4); 1730 int (*test_sleepable)(struct bpf_dummy_ops_state *cb); 1731 }; 1732 1733 int bpf_struct_ops_test_run(struct bpf_prog *prog, const union bpf_attr *kattr, 1734 union bpf_attr __user *uattr); 1735 #endif 1736 #else 1737 static inline const struct bpf_struct_ops *bpf_struct_ops_find(u32 type_id) 1738 { 1739 return NULL; 1740 } 1741 static inline void bpf_struct_ops_init(struct btf *btf, 1742 struct bpf_verifier_log *log) 1743 { 1744 } 1745 static inline bool bpf_try_module_get(const void *data, struct module *owner) 1746 { 1747 return try_module_get(owner); 1748 } 1749 static inline void bpf_module_put(const void *data, struct module *owner) 1750 { 1751 module_put(owner); 1752 } 1753 static inline int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, 1754 void *key, 1755 void *value) 1756 { 1757 return -EINVAL; 1758 } 1759 static inline int bpf_struct_ops_link_create(union bpf_attr *attr) 1760 { 1761 return -EOPNOTSUPP; 1762 } 1763 1764 #endif 1765 1766 #if defined(CONFIG_CGROUP_BPF) && defined(CONFIG_BPF_LSM) 1767 int bpf_trampoline_link_cgroup_shim(struct bpf_prog *prog, 1768 int cgroup_atype); 1769 void bpf_trampoline_unlink_cgroup_shim(struct bpf_prog *prog); 1770 #else 1771 static inline int bpf_trampoline_link_cgroup_shim(struct bpf_prog *prog, 1772 int cgroup_atype) 1773 { 1774 return -EOPNOTSUPP; 1775 } 1776 static inline void bpf_trampoline_unlink_cgroup_shim(struct bpf_prog *prog) 1777 { 1778 } 1779 #endif 1780 1781 struct bpf_array { 1782 struct bpf_map map; 1783 u32 elem_size; 1784 u32 index_mask; 1785 struct bpf_array_aux *aux; 1786 union { 1787 DECLARE_FLEX_ARRAY(char, value) __aligned(8); 1788 DECLARE_FLEX_ARRAY(void *, ptrs) __aligned(8); 1789 DECLARE_FLEX_ARRAY(void __percpu *, pptrs) __aligned(8); 1790 }; 1791 }; 1792 1793 #define BPF_COMPLEXITY_LIMIT_INSNS 1000000 /* yes. 1M insns */ 1794 #define MAX_TAIL_CALL_CNT 33 1795 1796 /* Maximum number of loops for bpf_loop and bpf_iter_num. 1797 * It's enum to expose it (and thus make it discoverable) through BTF. 1798 */ 1799 enum { 1800 BPF_MAX_LOOPS = 8 * 1024 * 1024, 1801 }; 1802 1803 #define BPF_F_ACCESS_MASK (BPF_F_RDONLY | \ 1804 BPF_F_RDONLY_PROG | \ 1805 BPF_F_WRONLY | \ 1806 BPF_F_WRONLY_PROG) 1807 1808 #define BPF_MAP_CAN_READ BIT(0) 1809 #define BPF_MAP_CAN_WRITE BIT(1) 1810 1811 /* Maximum number of user-producer ring buffer samples that can be drained in 1812 * a call to bpf_user_ringbuf_drain(). 1813 */ 1814 #define BPF_MAX_USER_RINGBUF_SAMPLES (128 * 1024) 1815 1816 static inline u32 bpf_map_flags_to_cap(struct bpf_map *map) 1817 { 1818 u32 access_flags = map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG); 1819 1820 /* Combination of BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG is 1821 * not possible. 1822 */ 1823 if (access_flags & BPF_F_RDONLY_PROG) 1824 return BPF_MAP_CAN_READ; 1825 else if (access_flags & BPF_F_WRONLY_PROG) 1826 return BPF_MAP_CAN_WRITE; 1827 else 1828 return BPF_MAP_CAN_READ | BPF_MAP_CAN_WRITE; 1829 } 1830 1831 static inline bool bpf_map_flags_access_ok(u32 access_flags) 1832 { 1833 return (access_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) != 1834 (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG); 1835 } 1836 1837 struct bpf_event_entry { 1838 struct perf_event *event; 1839 struct file *perf_file; 1840 struct file *map_file; 1841 struct rcu_head rcu; 1842 }; 1843 1844 static inline bool map_type_contains_progs(struct bpf_map *map) 1845 { 1846 return map->map_type == BPF_MAP_TYPE_PROG_ARRAY || 1847 map->map_type == BPF_MAP_TYPE_DEVMAP || 1848 map->map_type == BPF_MAP_TYPE_CPUMAP; 1849 } 1850 1851 bool bpf_prog_map_compatible(struct bpf_map *map, const struct bpf_prog *fp); 1852 int bpf_prog_calc_tag(struct bpf_prog *fp); 1853 1854 const struct bpf_func_proto *bpf_get_trace_printk_proto(void); 1855 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void); 1856 1857 typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src, 1858 unsigned long off, unsigned long len); 1859 typedef u32 (*bpf_convert_ctx_access_t)(enum bpf_access_type type, 1860 const struct bpf_insn *src, 1861 struct bpf_insn *dst, 1862 struct bpf_prog *prog, 1863 u32 *target_size); 1864 1865 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 1866 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy); 1867 1868 /* an array of programs to be executed under rcu_lock. 1869 * 1870 * Typical usage: 1871 * ret = bpf_prog_run_array(rcu_dereference(&bpf_prog_array), ctx, bpf_prog_run); 1872 * 1873 * the structure returned by bpf_prog_array_alloc() should be populated 1874 * with program pointers and the last pointer must be NULL. 1875 * The user has to keep refcnt on the program and make sure the program 1876 * is removed from the array before bpf_prog_put(). 1877 * The 'struct bpf_prog_array *' should only be replaced with xchg() 1878 * since other cpus are walking the array of pointers in parallel. 1879 */ 1880 struct bpf_prog_array_item { 1881 struct bpf_prog *prog; 1882 union { 1883 struct bpf_cgroup_storage *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]; 1884 u64 bpf_cookie; 1885 }; 1886 }; 1887 1888 struct bpf_prog_array { 1889 struct rcu_head rcu; 1890 struct bpf_prog_array_item items[]; 1891 }; 1892 1893 struct bpf_empty_prog_array { 1894 struct bpf_prog_array hdr; 1895 struct bpf_prog *null_prog; 1896 }; 1897 1898 /* to avoid allocating empty bpf_prog_array for cgroups that 1899 * don't have bpf program attached use one global 'bpf_empty_prog_array' 1900 * It will not be modified the caller of bpf_prog_array_alloc() 1901 * (since caller requested prog_cnt == 0) 1902 * that pointer should be 'freed' by bpf_prog_array_free() 1903 */ 1904 extern struct bpf_empty_prog_array bpf_empty_prog_array; 1905 1906 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags); 1907 void bpf_prog_array_free(struct bpf_prog_array *progs); 1908 /* Use when traversal over the bpf_prog_array uses tasks_trace rcu */ 1909 void bpf_prog_array_free_sleepable(struct bpf_prog_array *progs); 1910 int bpf_prog_array_length(struct bpf_prog_array *progs); 1911 bool bpf_prog_array_is_empty(struct bpf_prog_array *array); 1912 int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs, 1913 __u32 __user *prog_ids, u32 cnt); 1914 1915 void bpf_prog_array_delete_safe(struct bpf_prog_array *progs, 1916 struct bpf_prog *old_prog); 1917 int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index); 1918 int bpf_prog_array_update_at(struct bpf_prog_array *array, int index, 1919 struct bpf_prog *prog); 1920 int bpf_prog_array_copy_info(struct bpf_prog_array *array, 1921 u32 *prog_ids, u32 request_cnt, 1922 u32 *prog_cnt); 1923 int bpf_prog_array_copy(struct bpf_prog_array *old_array, 1924 struct bpf_prog *exclude_prog, 1925 struct bpf_prog *include_prog, 1926 u64 bpf_cookie, 1927 struct bpf_prog_array **new_array); 1928 1929 struct bpf_run_ctx {}; 1930 1931 struct bpf_cg_run_ctx { 1932 struct bpf_run_ctx run_ctx; 1933 const struct bpf_prog_array_item *prog_item; 1934 int retval; 1935 }; 1936 1937 struct bpf_trace_run_ctx { 1938 struct bpf_run_ctx run_ctx; 1939 u64 bpf_cookie; 1940 bool is_uprobe; 1941 }; 1942 1943 struct bpf_tramp_run_ctx { 1944 struct bpf_run_ctx run_ctx; 1945 u64 bpf_cookie; 1946 struct bpf_run_ctx *saved_run_ctx; 1947 }; 1948 1949 static inline struct bpf_run_ctx *bpf_set_run_ctx(struct bpf_run_ctx *new_ctx) 1950 { 1951 struct bpf_run_ctx *old_ctx = NULL; 1952 1953 #ifdef CONFIG_BPF_SYSCALL 1954 old_ctx = current->bpf_ctx; 1955 current->bpf_ctx = new_ctx; 1956 #endif 1957 return old_ctx; 1958 } 1959 1960 static inline void bpf_reset_run_ctx(struct bpf_run_ctx *old_ctx) 1961 { 1962 #ifdef CONFIG_BPF_SYSCALL 1963 current->bpf_ctx = old_ctx; 1964 #endif 1965 } 1966 1967 /* BPF program asks to bypass CAP_NET_BIND_SERVICE in bind. */ 1968 #define BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE (1 << 0) 1969 /* BPF program asks to set CN on the packet. */ 1970 #define BPF_RET_SET_CN (1 << 0) 1971 1972 typedef u32 (*bpf_prog_run_fn)(const struct bpf_prog *prog, const void *ctx); 1973 1974 static __always_inline u32 1975 bpf_prog_run_array(const struct bpf_prog_array *array, 1976 const void *ctx, bpf_prog_run_fn run_prog) 1977 { 1978 const struct bpf_prog_array_item *item; 1979 const struct bpf_prog *prog; 1980 struct bpf_run_ctx *old_run_ctx; 1981 struct bpf_trace_run_ctx run_ctx; 1982 u32 ret = 1; 1983 1984 RCU_LOCKDEP_WARN(!rcu_read_lock_held(), "no rcu lock held"); 1985 1986 if (unlikely(!array)) 1987 return ret; 1988 1989 run_ctx.is_uprobe = false; 1990 1991 migrate_disable(); 1992 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); 1993 item = &array->items[0]; 1994 while ((prog = READ_ONCE(item->prog))) { 1995 run_ctx.bpf_cookie = item->bpf_cookie; 1996 ret &= run_prog(prog, ctx); 1997 item++; 1998 } 1999 bpf_reset_run_ctx(old_run_ctx); 2000 migrate_enable(); 2001 return ret; 2002 } 2003 2004 /* Notes on RCU design for bpf_prog_arrays containing sleepable programs: 2005 * 2006 * We use the tasks_trace rcu flavor read section to protect the bpf_prog_array 2007 * overall. As a result, we must use the bpf_prog_array_free_sleepable 2008 * in order to use the tasks_trace rcu grace period. 2009 * 2010 * When a non-sleepable program is inside the array, we take the rcu read 2011 * section and disable preemption for that program alone, so it can access 2012 * rcu-protected dynamically sized maps. 2013 */ 2014 static __always_inline u32 2015 bpf_prog_run_array_uprobe(const struct bpf_prog_array __rcu *array_rcu, 2016 const void *ctx, bpf_prog_run_fn run_prog) 2017 { 2018 const struct bpf_prog_array_item *item; 2019 const struct bpf_prog *prog; 2020 const struct bpf_prog_array *array; 2021 struct bpf_run_ctx *old_run_ctx; 2022 struct bpf_trace_run_ctx run_ctx; 2023 u32 ret = 1; 2024 2025 might_fault(); 2026 2027 rcu_read_lock_trace(); 2028 migrate_disable(); 2029 2030 run_ctx.is_uprobe = true; 2031 2032 array = rcu_dereference_check(array_rcu, rcu_read_lock_trace_held()); 2033 if (unlikely(!array)) 2034 goto out; 2035 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); 2036 item = &array->items[0]; 2037 while ((prog = READ_ONCE(item->prog))) { 2038 if (!prog->aux->sleepable) 2039 rcu_read_lock(); 2040 2041 run_ctx.bpf_cookie = item->bpf_cookie; 2042 ret &= run_prog(prog, ctx); 2043 item++; 2044 2045 if (!prog->aux->sleepable) 2046 rcu_read_unlock(); 2047 } 2048 bpf_reset_run_ctx(old_run_ctx); 2049 out: 2050 migrate_enable(); 2051 rcu_read_unlock_trace(); 2052 return ret; 2053 } 2054 2055 #ifdef CONFIG_BPF_SYSCALL 2056 DECLARE_PER_CPU(int, bpf_prog_active); 2057 extern struct mutex bpf_stats_enabled_mutex; 2058 2059 /* 2060 * Block execution of BPF programs attached to instrumentation (perf, 2061 * kprobes, tracepoints) to prevent deadlocks on map operations as any of 2062 * these events can happen inside a region which holds a map bucket lock 2063 * and can deadlock on it. 2064 */ 2065 static inline void bpf_disable_instrumentation(void) 2066 { 2067 migrate_disable(); 2068 this_cpu_inc(bpf_prog_active); 2069 } 2070 2071 static inline void bpf_enable_instrumentation(void) 2072 { 2073 this_cpu_dec(bpf_prog_active); 2074 migrate_enable(); 2075 } 2076 2077 extern const struct super_operations bpf_super_ops; 2078 extern const struct file_operations bpf_map_fops; 2079 extern const struct file_operations bpf_prog_fops; 2080 extern const struct file_operations bpf_iter_fops; 2081 2082 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ 2083 extern const struct bpf_prog_ops _name ## _prog_ops; \ 2084 extern const struct bpf_verifier_ops _name ## _verifier_ops; 2085 #define BPF_MAP_TYPE(_id, _ops) \ 2086 extern const struct bpf_map_ops _ops; 2087 #define BPF_LINK_TYPE(_id, _name) 2088 #include <linux/bpf_types.h> 2089 #undef BPF_PROG_TYPE 2090 #undef BPF_MAP_TYPE 2091 #undef BPF_LINK_TYPE 2092 2093 extern const struct bpf_prog_ops bpf_offload_prog_ops; 2094 extern const struct bpf_verifier_ops tc_cls_act_analyzer_ops; 2095 extern const struct bpf_verifier_ops xdp_analyzer_ops; 2096 2097 struct bpf_prog *bpf_prog_get(u32 ufd); 2098 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 2099 bool attach_drv); 2100 void bpf_prog_add(struct bpf_prog *prog, int i); 2101 void bpf_prog_sub(struct bpf_prog *prog, int i); 2102 void bpf_prog_inc(struct bpf_prog *prog); 2103 struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog); 2104 void bpf_prog_put(struct bpf_prog *prog); 2105 2106 void bpf_prog_free_id(struct bpf_prog *prog); 2107 void bpf_map_free_id(struct bpf_map *map); 2108 2109 struct btf_field *btf_record_find(const struct btf_record *rec, 2110 u32 offset, u32 field_mask); 2111 void btf_record_free(struct btf_record *rec); 2112 void bpf_map_free_record(struct bpf_map *map); 2113 struct btf_record *btf_record_dup(const struct btf_record *rec); 2114 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b); 2115 void bpf_obj_free_timer(const struct btf_record *rec, void *obj); 2116 void bpf_obj_free_fields(const struct btf_record *rec, void *obj); 2117 void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu); 2118 2119 struct bpf_map *bpf_map_get(u32 ufd); 2120 struct bpf_map *bpf_map_get_with_uref(u32 ufd); 2121 struct bpf_map *__bpf_map_get(struct fd f); 2122 void bpf_map_inc(struct bpf_map *map); 2123 void bpf_map_inc_with_uref(struct bpf_map *map); 2124 struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref); 2125 struct bpf_map * __must_check bpf_map_inc_not_zero(struct bpf_map *map); 2126 void bpf_map_put_with_uref(struct bpf_map *map); 2127 void bpf_map_put(struct bpf_map *map); 2128 void *bpf_map_area_alloc(u64 size, int numa_node); 2129 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node); 2130 void bpf_map_area_free(void *base); 2131 bool bpf_map_write_active(const struct bpf_map *map); 2132 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr); 2133 int generic_map_lookup_batch(struct bpf_map *map, 2134 const union bpf_attr *attr, 2135 union bpf_attr __user *uattr); 2136 int generic_map_update_batch(struct bpf_map *map, struct file *map_file, 2137 const union bpf_attr *attr, 2138 union bpf_attr __user *uattr); 2139 int generic_map_delete_batch(struct bpf_map *map, 2140 const union bpf_attr *attr, 2141 union bpf_attr __user *uattr); 2142 struct bpf_map *bpf_map_get_curr_or_next(u32 *id); 2143 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id); 2144 2145 #ifdef CONFIG_MEMCG_KMEM 2146 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags, 2147 int node); 2148 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags); 2149 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size, 2150 gfp_t flags); 2151 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size, 2152 size_t align, gfp_t flags); 2153 #else 2154 static inline void * 2155 bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags, 2156 int node) 2157 { 2158 return kmalloc_node(size, flags, node); 2159 } 2160 2161 static inline void * 2162 bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags) 2163 { 2164 return kzalloc(size, flags); 2165 } 2166 2167 static inline void * 2168 bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size, gfp_t flags) 2169 { 2170 return kvcalloc(n, size, flags); 2171 } 2172 2173 static inline void __percpu * 2174 bpf_map_alloc_percpu(const struct bpf_map *map, size_t size, size_t align, 2175 gfp_t flags) 2176 { 2177 return __alloc_percpu_gfp(size, align, flags); 2178 } 2179 #endif 2180 2181 static inline int 2182 bpf_map_init_elem_count(struct bpf_map *map) 2183 { 2184 size_t size = sizeof(*map->elem_count), align = size; 2185 gfp_t flags = GFP_USER | __GFP_NOWARN; 2186 2187 map->elem_count = bpf_map_alloc_percpu(map, size, align, flags); 2188 if (!map->elem_count) 2189 return -ENOMEM; 2190 2191 return 0; 2192 } 2193 2194 static inline void 2195 bpf_map_free_elem_count(struct bpf_map *map) 2196 { 2197 free_percpu(map->elem_count); 2198 } 2199 2200 static inline void bpf_map_inc_elem_count(struct bpf_map *map) 2201 { 2202 this_cpu_inc(*map->elem_count); 2203 } 2204 2205 static inline void bpf_map_dec_elem_count(struct bpf_map *map) 2206 { 2207 this_cpu_dec(*map->elem_count); 2208 } 2209 2210 extern int sysctl_unprivileged_bpf_disabled; 2211 2212 bool bpf_token_capable(const struct bpf_token *token, int cap); 2213 2214 static inline bool bpf_allow_ptr_leaks(const struct bpf_token *token) 2215 { 2216 return bpf_token_capable(token, CAP_PERFMON); 2217 } 2218 2219 static inline bool bpf_allow_uninit_stack(const struct bpf_token *token) 2220 { 2221 return bpf_token_capable(token, CAP_PERFMON); 2222 } 2223 2224 static inline bool bpf_bypass_spec_v1(const struct bpf_token *token) 2225 { 2226 return cpu_mitigations_off() || bpf_token_capable(token, CAP_PERFMON); 2227 } 2228 2229 static inline bool bpf_bypass_spec_v4(const struct bpf_token *token) 2230 { 2231 return cpu_mitigations_off() || bpf_token_capable(token, CAP_PERFMON); 2232 } 2233 2234 int bpf_map_new_fd(struct bpf_map *map, int flags); 2235 int bpf_prog_new_fd(struct bpf_prog *prog); 2236 2237 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type, 2238 const struct bpf_link_ops *ops, struct bpf_prog *prog); 2239 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer); 2240 int bpf_link_settle(struct bpf_link_primer *primer); 2241 void bpf_link_cleanup(struct bpf_link_primer *primer); 2242 void bpf_link_inc(struct bpf_link *link); 2243 void bpf_link_put(struct bpf_link *link); 2244 int bpf_link_new_fd(struct bpf_link *link); 2245 struct bpf_link *bpf_link_get_from_fd(u32 ufd); 2246 struct bpf_link *bpf_link_get_curr_or_next(u32 *id); 2247 2248 void bpf_token_inc(struct bpf_token *token); 2249 void bpf_token_put(struct bpf_token *token); 2250 int bpf_token_create(union bpf_attr *attr); 2251 struct bpf_token *bpf_token_get_from_fd(u32 ufd); 2252 2253 bool bpf_token_allow_cmd(const struct bpf_token *token, enum bpf_cmd cmd); 2254 bool bpf_token_allow_map_type(const struct bpf_token *token, enum bpf_map_type type); 2255 bool bpf_token_allow_prog_type(const struct bpf_token *token, 2256 enum bpf_prog_type prog_type, 2257 enum bpf_attach_type attach_type); 2258 2259 int bpf_obj_pin_user(u32 ufd, int path_fd, const char __user *pathname); 2260 int bpf_obj_get_user(int path_fd, const char __user *pathname, int flags); 2261 struct inode *bpf_get_inode(struct super_block *sb, const struct inode *dir, 2262 umode_t mode); 2263 2264 #define BPF_ITER_FUNC_PREFIX "bpf_iter_" 2265 #define DEFINE_BPF_ITER_FUNC(target, args...) \ 2266 extern int bpf_iter_ ## target(args); \ 2267 int __init bpf_iter_ ## target(args) { return 0; } 2268 2269 /* 2270 * The task type of iterators. 2271 * 2272 * For BPF task iterators, they can be parameterized with various 2273 * parameters to visit only some of tasks. 2274 * 2275 * BPF_TASK_ITER_ALL (default) 2276 * Iterate over resources of every task. 2277 * 2278 * BPF_TASK_ITER_TID 2279 * Iterate over resources of a task/tid. 2280 * 2281 * BPF_TASK_ITER_TGID 2282 * Iterate over resources of every task of a process / task group. 2283 */ 2284 enum bpf_iter_task_type { 2285 BPF_TASK_ITER_ALL = 0, 2286 BPF_TASK_ITER_TID, 2287 BPF_TASK_ITER_TGID, 2288 }; 2289 2290 struct bpf_iter_aux_info { 2291 /* for map_elem iter */ 2292 struct bpf_map *map; 2293 2294 /* for cgroup iter */ 2295 struct { 2296 struct cgroup *start; /* starting cgroup */ 2297 enum bpf_cgroup_iter_order order; 2298 } cgroup; 2299 struct { 2300 enum bpf_iter_task_type type; 2301 u32 pid; 2302 } task; 2303 }; 2304 2305 typedef int (*bpf_iter_attach_target_t)(struct bpf_prog *prog, 2306 union bpf_iter_link_info *linfo, 2307 struct bpf_iter_aux_info *aux); 2308 typedef void (*bpf_iter_detach_target_t)(struct bpf_iter_aux_info *aux); 2309 typedef void (*bpf_iter_show_fdinfo_t) (const struct bpf_iter_aux_info *aux, 2310 struct seq_file *seq); 2311 typedef int (*bpf_iter_fill_link_info_t)(const struct bpf_iter_aux_info *aux, 2312 struct bpf_link_info *info); 2313 typedef const struct bpf_func_proto * 2314 (*bpf_iter_get_func_proto_t)(enum bpf_func_id func_id, 2315 const struct bpf_prog *prog); 2316 2317 enum bpf_iter_feature { 2318 BPF_ITER_RESCHED = BIT(0), 2319 }; 2320 2321 #define BPF_ITER_CTX_ARG_MAX 2 2322 struct bpf_iter_reg { 2323 const char *target; 2324 bpf_iter_attach_target_t attach_target; 2325 bpf_iter_detach_target_t detach_target; 2326 bpf_iter_show_fdinfo_t show_fdinfo; 2327 bpf_iter_fill_link_info_t fill_link_info; 2328 bpf_iter_get_func_proto_t get_func_proto; 2329 u32 ctx_arg_info_size; 2330 u32 feature; 2331 struct bpf_ctx_arg_aux ctx_arg_info[BPF_ITER_CTX_ARG_MAX]; 2332 const struct bpf_iter_seq_info *seq_info; 2333 }; 2334 2335 struct bpf_iter_meta { 2336 __bpf_md_ptr(struct seq_file *, seq); 2337 u64 session_id; 2338 u64 seq_num; 2339 }; 2340 2341 struct bpf_iter__bpf_map_elem { 2342 __bpf_md_ptr(struct bpf_iter_meta *, meta); 2343 __bpf_md_ptr(struct bpf_map *, map); 2344 __bpf_md_ptr(void *, key); 2345 __bpf_md_ptr(void *, value); 2346 }; 2347 2348 int bpf_iter_reg_target(const struct bpf_iter_reg *reg_info); 2349 void bpf_iter_unreg_target(const struct bpf_iter_reg *reg_info); 2350 bool bpf_iter_prog_supported(struct bpf_prog *prog); 2351 const struct bpf_func_proto * 2352 bpf_iter_get_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog); 2353 int bpf_iter_link_attach(const union bpf_attr *attr, bpfptr_t uattr, struct bpf_prog *prog); 2354 int bpf_iter_new_fd(struct bpf_link *link); 2355 bool bpf_link_is_iter(struct bpf_link *link); 2356 struct bpf_prog *bpf_iter_get_info(struct bpf_iter_meta *meta, bool in_stop); 2357 int bpf_iter_run_prog(struct bpf_prog *prog, void *ctx); 2358 void bpf_iter_map_show_fdinfo(const struct bpf_iter_aux_info *aux, 2359 struct seq_file *seq); 2360 int bpf_iter_map_fill_link_info(const struct bpf_iter_aux_info *aux, 2361 struct bpf_link_info *info); 2362 2363 int map_set_for_each_callback_args(struct bpf_verifier_env *env, 2364 struct bpf_func_state *caller, 2365 struct bpf_func_state *callee); 2366 2367 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value); 2368 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value); 2369 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value, 2370 u64 flags); 2371 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value, 2372 u64 flags); 2373 2374 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value); 2375 2376 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file, 2377 void *key, void *value, u64 map_flags); 2378 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 2379 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file, 2380 void *key, void *value, u64 map_flags); 2381 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 2382 2383 int bpf_get_file_flag(int flags); 2384 int bpf_check_uarg_tail_zero(bpfptr_t uaddr, size_t expected_size, 2385 size_t actual_size); 2386 2387 /* verify correctness of eBPF program */ 2388 int bpf_check(struct bpf_prog **fp, union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size); 2389 2390 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 2391 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth); 2392 #endif 2393 2394 struct btf *bpf_get_btf_vmlinux(void); 2395 2396 /* Map specifics */ 2397 struct xdp_frame; 2398 struct sk_buff; 2399 struct bpf_dtab_netdev; 2400 struct bpf_cpu_map_entry; 2401 2402 void __dev_flush(void); 2403 int dev_xdp_enqueue(struct net_device *dev, struct xdp_frame *xdpf, 2404 struct net_device *dev_rx); 2405 int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_frame *xdpf, 2406 struct net_device *dev_rx); 2407 int dev_map_enqueue_multi(struct xdp_frame *xdpf, struct net_device *dev_rx, 2408 struct bpf_map *map, bool exclude_ingress); 2409 int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, struct sk_buff *skb, 2410 struct bpf_prog *xdp_prog); 2411 int dev_map_redirect_multi(struct net_device *dev, struct sk_buff *skb, 2412 struct bpf_prog *xdp_prog, struct bpf_map *map, 2413 bool exclude_ingress); 2414 2415 void __cpu_map_flush(void); 2416 int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_frame *xdpf, 2417 struct net_device *dev_rx); 2418 int cpu_map_generic_redirect(struct bpf_cpu_map_entry *rcpu, 2419 struct sk_buff *skb); 2420 2421 /* Return map's numa specified by userspace */ 2422 static inline int bpf_map_attr_numa_node(const union bpf_attr *attr) 2423 { 2424 return (attr->map_flags & BPF_F_NUMA_NODE) ? 2425 attr->numa_node : NUMA_NO_NODE; 2426 } 2427 2428 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type); 2429 int array_map_alloc_check(union bpf_attr *attr); 2430 2431 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr, 2432 union bpf_attr __user *uattr); 2433 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr, 2434 union bpf_attr __user *uattr); 2435 int bpf_prog_test_run_tracing(struct bpf_prog *prog, 2436 const union bpf_attr *kattr, 2437 union bpf_attr __user *uattr); 2438 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 2439 const union bpf_attr *kattr, 2440 union bpf_attr __user *uattr); 2441 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog, 2442 const union bpf_attr *kattr, 2443 union bpf_attr __user *uattr); 2444 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, 2445 const union bpf_attr *kattr, 2446 union bpf_attr __user *uattr); 2447 int bpf_prog_test_run_nf(struct bpf_prog *prog, 2448 const union bpf_attr *kattr, 2449 union bpf_attr __user *uattr); 2450 bool btf_ctx_access(int off, int size, enum bpf_access_type type, 2451 const struct bpf_prog *prog, 2452 struct bpf_insn_access_aux *info); 2453 2454 static inline bool bpf_tracing_ctx_access(int off, int size, 2455 enum bpf_access_type type) 2456 { 2457 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS) 2458 return false; 2459 if (type != BPF_READ) 2460 return false; 2461 if (off % size != 0) 2462 return false; 2463 return true; 2464 } 2465 2466 static inline bool bpf_tracing_btf_ctx_access(int off, int size, 2467 enum bpf_access_type type, 2468 const struct bpf_prog *prog, 2469 struct bpf_insn_access_aux *info) 2470 { 2471 if (!bpf_tracing_ctx_access(off, size, type)) 2472 return false; 2473 return btf_ctx_access(off, size, type, prog, info); 2474 } 2475 2476 int btf_struct_access(struct bpf_verifier_log *log, 2477 const struct bpf_reg_state *reg, 2478 int off, int size, enum bpf_access_type atype, 2479 u32 *next_btf_id, enum bpf_type_flag *flag, const char **field_name); 2480 bool btf_struct_ids_match(struct bpf_verifier_log *log, 2481 const struct btf *btf, u32 id, int off, 2482 const struct btf *need_btf, u32 need_type_id, 2483 bool strict); 2484 2485 int btf_distill_func_proto(struct bpf_verifier_log *log, 2486 struct btf *btf, 2487 const struct btf_type *func_proto, 2488 const char *func_name, 2489 struct btf_func_model *m); 2490 2491 struct bpf_reg_state; 2492 int btf_check_subprog_arg_match(struct bpf_verifier_env *env, int subprog, 2493 struct bpf_reg_state *regs); 2494 int btf_check_subprog_call(struct bpf_verifier_env *env, int subprog, 2495 struct bpf_reg_state *regs); 2496 int btf_prepare_func_args(struct bpf_verifier_env *env, int subprog, 2497 struct bpf_reg_state *reg, bool is_ex_cb); 2498 int btf_check_type_match(struct bpf_verifier_log *log, const struct bpf_prog *prog, 2499 struct btf *btf, const struct btf_type *t); 2500 const char *btf_find_decl_tag_value(const struct btf *btf, const struct btf_type *pt, 2501 int comp_idx, const char *tag_key); 2502 2503 struct bpf_prog *bpf_prog_by_id(u32 id); 2504 struct bpf_link *bpf_link_by_id(u32 id); 2505 2506 const struct bpf_func_proto *bpf_base_func_proto(enum bpf_func_id func_id, 2507 const struct bpf_prog *prog); 2508 void bpf_task_storage_free(struct task_struct *task); 2509 void bpf_cgrp_storage_free(struct cgroup *cgroup); 2510 bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog); 2511 const struct btf_func_model * 2512 bpf_jit_find_kfunc_model(const struct bpf_prog *prog, 2513 const struct bpf_insn *insn); 2514 int bpf_get_kfunc_addr(const struct bpf_prog *prog, u32 func_id, 2515 u16 btf_fd_idx, u8 **func_addr); 2516 2517 struct bpf_core_ctx { 2518 struct bpf_verifier_log *log; 2519 const struct btf *btf; 2520 }; 2521 2522 bool btf_nested_type_is_trusted(struct bpf_verifier_log *log, 2523 const struct bpf_reg_state *reg, 2524 const char *field_name, u32 btf_id, const char *suffix); 2525 2526 bool btf_type_ids_nocast_alias(struct bpf_verifier_log *log, 2527 const struct btf *reg_btf, u32 reg_id, 2528 const struct btf *arg_btf, u32 arg_id); 2529 2530 int bpf_core_apply(struct bpf_core_ctx *ctx, const struct bpf_core_relo *relo, 2531 int relo_idx, void *insn); 2532 2533 static inline bool unprivileged_ebpf_enabled(void) 2534 { 2535 return !sysctl_unprivileged_bpf_disabled; 2536 } 2537 2538 /* Not all bpf prog type has the bpf_ctx. 2539 * For the bpf prog type that has initialized the bpf_ctx, 2540 * this function can be used to decide if a kernel function 2541 * is called by a bpf program. 2542 */ 2543 static inline bool has_current_bpf_ctx(void) 2544 { 2545 return !!current->bpf_ctx; 2546 } 2547 2548 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog); 2549 2550 void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data, 2551 enum bpf_dynptr_type type, u32 offset, u32 size); 2552 void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr); 2553 void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr); 2554 2555 bool dev_check_flush(void); 2556 bool cpu_map_check_flush(void); 2557 #else /* !CONFIG_BPF_SYSCALL */ 2558 static inline struct bpf_prog *bpf_prog_get(u32 ufd) 2559 { 2560 return ERR_PTR(-EOPNOTSUPP); 2561 } 2562 2563 static inline struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, 2564 enum bpf_prog_type type, 2565 bool attach_drv) 2566 { 2567 return ERR_PTR(-EOPNOTSUPP); 2568 } 2569 2570 static inline void bpf_prog_add(struct bpf_prog *prog, int i) 2571 { 2572 } 2573 2574 static inline void bpf_prog_sub(struct bpf_prog *prog, int i) 2575 { 2576 } 2577 2578 static inline void bpf_prog_put(struct bpf_prog *prog) 2579 { 2580 } 2581 2582 static inline void bpf_prog_inc(struct bpf_prog *prog) 2583 { 2584 } 2585 2586 static inline struct bpf_prog *__must_check 2587 bpf_prog_inc_not_zero(struct bpf_prog *prog) 2588 { 2589 return ERR_PTR(-EOPNOTSUPP); 2590 } 2591 2592 static inline void bpf_link_init(struct bpf_link *link, enum bpf_link_type type, 2593 const struct bpf_link_ops *ops, 2594 struct bpf_prog *prog) 2595 { 2596 } 2597 2598 static inline int bpf_link_prime(struct bpf_link *link, 2599 struct bpf_link_primer *primer) 2600 { 2601 return -EOPNOTSUPP; 2602 } 2603 2604 static inline int bpf_link_settle(struct bpf_link_primer *primer) 2605 { 2606 return -EOPNOTSUPP; 2607 } 2608 2609 static inline void bpf_link_cleanup(struct bpf_link_primer *primer) 2610 { 2611 } 2612 2613 static inline void bpf_link_inc(struct bpf_link *link) 2614 { 2615 } 2616 2617 static inline void bpf_link_put(struct bpf_link *link) 2618 { 2619 } 2620 2621 static inline int bpf_obj_get_user(const char __user *pathname, int flags) 2622 { 2623 return -EOPNOTSUPP; 2624 } 2625 2626 static inline bool bpf_token_capable(const struct bpf_token *token, int cap) 2627 { 2628 return capable(cap) || (cap != CAP_SYS_ADMIN && capable(CAP_SYS_ADMIN)); 2629 } 2630 2631 static inline void bpf_token_inc(struct bpf_token *token) 2632 { 2633 } 2634 2635 static inline void bpf_token_put(struct bpf_token *token) 2636 { 2637 } 2638 2639 static inline struct bpf_token *bpf_token_get_from_fd(u32 ufd) 2640 { 2641 return ERR_PTR(-EOPNOTSUPP); 2642 } 2643 2644 static inline void __dev_flush(void) 2645 { 2646 } 2647 2648 struct xdp_frame; 2649 struct bpf_dtab_netdev; 2650 struct bpf_cpu_map_entry; 2651 2652 static inline 2653 int dev_xdp_enqueue(struct net_device *dev, struct xdp_frame *xdpf, 2654 struct net_device *dev_rx) 2655 { 2656 return 0; 2657 } 2658 2659 static inline 2660 int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_frame *xdpf, 2661 struct net_device *dev_rx) 2662 { 2663 return 0; 2664 } 2665 2666 static inline 2667 int dev_map_enqueue_multi(struct xdp_frame *xdpf, struct net_device *dev_rx, 2668 struct bpf_map *map, bool exclude_ingress) 2669 { 2670 return 0; 2671 } 2672 2673 struct sk_buff; 2674 2675 static inline int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, 2676 struct sk_buff *skb, 2677 struct bpf_prog *xdp_prog) 2678 { 2679 return 0; 2680 } 2681 2682 static inline 2683 int dev_map_redirect_multi(struct net_device *dev, struct sk_buff *skb, 2684 struct bpf_prog *xdp_prog, struct bpf_map *map, 2685 bool exclude_ingress) 2686 { 2687 return 0; 2688 } 2689 2690 static inline void __cpu_map_flush(void) 2691 { 2692 } 2693 2694 static inline int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, 2695 struct xdp_frame *xdpf, 2696 struct net_device *dev_rx) 2697 { 2698 return 0; 2699 } 2700 2701 static inline int cpu_map_generic_redirect(struct bpf_cpu_map_entry *rcpu, 2702 struct sk_buff *skb) 2703 { 2704 return -EOPNOTSUPP; 2705 } 2706 2707 static inline struct bpf_prog *bpf_prog_get_type_path(const char *name, 2708 enum bpf_prog_type type) 2709 { 2710 return ERR_PTR(-EOPNOTSUPP); 2711 } 2712 2713 static inline int bpf_prog_test_run_xdp(struct bpf_prog *prog, 2714 const union bpf_attr *kattr, 2715 union bpf_attr __user *uattr) 2716 { 2717 return -ENOTSUPP; 2718 } 2719 2720 static inline int bpf_prog_test_run_skb(struct bpf_prog *prog, 2721 const union bpf_attr *kattr, 2722 union bpf_attr __user *uattr) 2723 { 2724 return -ENOTSUPP; 2725 } 2726 2727 static inline int bpf_prog_test_run_tracing(struct bpf_prog *prog, 2728 const union bpf_attr *kattr, 2729 union bpf_attr __user *uattr) 2730 { 2731 return -ENOTSUPP; 2732 } 2733 2734 static inline int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 2735 const union bpf_attr *kattr, 2736 union bpf_attr __user *uattr) 2737 { 2738 return -ENOTSUPP; 2739 } 2740 2741 static inline int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, 2742 const union bpf_attr *kattr, 2743 union bpf_attr __user *uattr) 2744 { 2745 return -ENOTSUPP; 2746 } 2747 2748 static inline void bpf_map_put(struct bpf_map *map) 2749 { 2750 } 2751 2752 static inline struct bpf_prog *bpf_prog_by_id(u32 id) 2753 { 2754 return ERR_PTR(-ENOTSUPP); 2755 } 2756 2757 static inline int btf_struct_access(struct bpf_verifier_log *log, 2758 const struct bpf_reg_state *reg, 2759 int off, int size, enum bpf_access_type atype, 2760 u32 *next_btf_id, enum bpf_type_flag *flag, 2761 const char **field_name) 2762 { 2763 return -EACCES; 2764 } 2765 2766 static inline const struct bpf_func_proto * 2767 bpf_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 2768 { 2769 return NULL; 2770 } 2771 2772 static inline void bpf_task_storage_free(struct task_struct *task) 2773 { 2774 } 2775 2776 static inline bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog) 2777 { 2778 return false; 2779 } 2780 2781 static inline const struct btf_func_model * 2782 bpf_jit_find_kfunc_model(const struct bpf_prog *prog, 2783 const struct bpf_insn *insn) 2784 { 2785 return NULL; 2786 } 2787 2788 static inline int 2789 bpf_get_kfunc_addr(const struct bpf_prog *prog, u32 func_id, 2790 u16 btf_fd_idx, u8 **func_addr) 2791 { 2792 return -ENOTSUPP; 2793 } 2794 2795 static inline bool unprivileged_ebpf_enabled(void) 2796 { 2797 return false; 2798 } 2799 2800 static inline bool has_current_bpf_ctx(void) 2801 { 2802 return false; 2803 } 2804 2805 static inline void bpf_prog_inc_misses_counter(struct bpf_prog *prog) 2806 { 2807 } 2808 2809 static inline void bpf_cgrp_storage_free(struct cgroup *cgroup) 2810 { 2811 } 2812 2813 static inline void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data, 2814 enum bpf_dynptr_type type, u32 offset, u32 size) 2815 { 2816 } 2817 2818 static inline void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr) 2819 { 2820 } 2821 2822 static inline void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr) 2823 { 2824 } 2825 #endif /* CONFIG_BPF_SYSCALL */ 2826 2827 static __always_inline int 2828 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr) 2829 { 2830 int ret = -EFAULT; 2831 2832 if (IS_ENABLED(CONFIG_BPF_EVENTS)) 2833 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size); 2834 if (unlikely(ret < 0)) 2835 memset(dst, 0, size); 2836 return ret; 2837 } 2838 2839 void __bpf_free_used_btfs(struct bpf_prog_aux *aux, 2840 struct btf_mod_pair *used_btfs, u32 len); 2841 2842 static inline struct bpf_prog *bpf_prog_get_type(u32 ufd, 2843 enum bpf_prog_type type) 2844 { 2845 return bpf_prog_get_type_dev(ufd, type, false); 2846 } 2847 2848 void __bpf_free_used_maps(struct bpf_prog_aux *aux, 2849 struct bpf_map **used_maps, u32 len); 2850 2851 bool bpf_prog_get_ok(struct bpf_prog *, enum bpf_prog_type *, bool); 2852 2853 int bpf_prog_offload_compile(struct bpf_prog *prog); 2854 void bpf_prog_dev_bound_destroy(struct bpf_prog *prog); 2855 int bpf_prog_offload_info_fill(struct bpf_prog_info *info, 2856 struct bpf_prog *prog); 2857 2858 int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map); 2859 2860 int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value); 2861 int bpf_map_offload_update_elem(struct bpf_map *map, 2862 void *key, void *value, u64 flags); 2863 int bpf_map_offload_delete_elem(struct bpf_map *map, void *key); 2864 int bpf_map_offload_get_next_key(struct bpf_map *map, 2865 void *key, void *next_key); 2866 2867 bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map); 2868 2869 struct bpf_offload_dev * 2870 bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv); 2871 void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev); 2872 void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev); 2873 int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev, 2874 struct net_device *netdev); 2875 void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev, 2876 struct net_device *netdev); 2877 bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev); 2878 2879 void unpriv_ebpf_notify(int new_state); 2880 2881 #if defined(CONFIG_NET) && defined(CONFIG_BPF_SYSCALL) 2882 int bpf_dev_bound_kfunc_check(struct bpf_verifier_log *log, 2883 struct bpf_prog_aux *prog_aux); 2884 void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog, u32 func_id); 2885 int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr); 2886 int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog, struct bpf_prog *old_prog); 2887 void bpf_dev_bound_netdev_unregister(struct net_device *dev); 2888 2889 static inline bool bpf_prog_is_dev_bound(const struct bpf_prog_aux *aux) 2890 { 2891 return aux->dev_bound; 2892 } 2893 2894 static inline bool bpf_prog_is_offloaded(const struct bpf_prog_aux *aux) 2895 { 2896 return aux->offload_requested; 2897 } 2898 2899 bool bpf_prog_dev_bound_match(const struct bpf_prog *lhs, const struct bpf_prog *rhs); 2900 2901 static inline bool bpf_map_is_offloaded(struct bpf_map *map) 2902 { 2903 return unlikely(map->ops == &bpf_map_offload_ops); 2904 } 2905 2906 struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr); 2907 void bpf_map_offload_map_free(struct bpf_map *map); 2908 u64 bpf_map_offload_map_mem_usage(const struct bpf_map *map); 2909 int bpf_prog_test_run_syscall(struct bpf_prog *prog, 2910 const union bpf_attr *kattr, 2911 union bpf_attr __user *uattr); 2912 2913 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog); 2914 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype); 2915 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value, u64 flags); 2916 int sock_map_bpf_prog_query(const union bpf_attr *attr, 2917 union bpf_attr __user *uattr); 2918 2919 void sock_map_unhash(struct sock *sk); 2920 void sock_map_destroy(struct sock *sk); 2921 void sock_map_close(struct sock *sk, long timeout); 2922 #else 2923 static inline int bpf_dev_bound_kfunc_check(struct bpf_verifier_log *log, 2924 struct bpf_prog_aux *prog_aux) 2925 { 2926 return -EOPNOTSUPP; 2927 } 2928 2929 static inline void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog, 2930 u32 func_id) 2931 { 2932 return NULL; 2933 } 2934 2935 static inline int bpf_prog_dev_bound_init(struct bpf_prog *prog, 2936 union bpf_attr *attr) 2937 { 2938 return -EOPNOTSUPP; 2939 } 2940 2941 static inline int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog, 2942 struct bpf_prog *old_prog) 2943 { 2944 return -EOPNOTSUPP; 2945 } 2946 2947 static inline void bpf_dev_bound_netdev_unregister(struct net_device *dev) 2948 { 2949 } 2950 2951 static inline bool bpf_prog_is_dev_bound(const struct bpf_prog_aux *aux) 2952 { 2953 return false; 2954 } 2955 2956 static inline bool bpf_prog_is_offloaded(struct bpf_prog_aux *aux) 2957 { 2958 return false; 2959 } 2960 2961 static inline bool bpf_prog_dev_bound_match(const struct bpf_prog *lhs, const struct bpf_prog *rhs) 2962 { 2963 return false; 2964 } 2965 2966 static inline bool bpf_map_is_offloaded(struct bpf_map *map) 2967 { 2968 return false; 2969 } 2970 2971 static inline struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr) 2972 { 2973 return ERR_PTR(-EOPNOTSUPP); 2974 } 2975 2976 static inline void bpf_map_offload_map_free(struct bpf_map *map) 2977 { 2978 } 2979 2980 static inline u64 bpf_map_offload_map_mem_usage(const struct bpf_map *map) 2981 { 2982 return 0; 2983 } 2984 2985 static inline int bpf_prog_test_run_syscall(struct bpf_prog *prog, 2986 const union bpf_attr *kattr, 2987 union bpf_attr __user *uattr) 2988 { 2989 return -ENOTSUPP; 2990 } 2991 2992 #ifdef CONFIG_BPF_SYSCALL 2993 static inline int sock_map_get_from_fd(const union bpf_attr *attr, 2994 struct bpf_prog *prog) 2995 { 2996 return -EINVAL; 2997 } 2998 2999 static inline int sock_map_prog_detach(const union bpf_attr *attr, 3000 enum bpf_prog_type ptype) 3001 { 3002 return -EOPNOTSUPP; 3003 } 3004 3005 static inline int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value, 3006 u64 flags) 3007 { 3008 return -EOPNOTSUPP; 3009 } 3010 3011 static inline int sock_map_bpf_prog_query(const union bpf_attr *attr, 3012 union bpf_attr __user *uattr) 3013 { 3014 return -EINVAL; 3015 } 3016 #endif /* CONFIG_BPF_SYSCALL */ 3017 #endif /* CONFIG_NET && CONFIG_BPF_SYSCALL */ 3018 3019 static __always_inline void 3020 bpf_prog_inc_misses_counters(const struct bpf_prog_array *array) 3021 { 3022 const struct bpf_prog_array_item *item; 3023 struct bpf_prog *prog; 3024 3025 if (unlikely(!array)) 3026 return; 3027 3028 item = &array->items[0]; 3029 while ((prog = READ_ONCE(item->prog))) { 3030 bpf_prog_inc_misses_counter(prog); 3031 item++; 3032 } 3033 } 3034 3035 #if defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) 3036 void bpf_sk_reuseport_detach(struct sock *sk); 3037 int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, void *key, 3038 void *value); 3039 int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, void *key, 3040 void *value, u64 map_flags); 3041 #else 3042 static inline void bpf_sk_reuseport_detach(struct sock *sk) 3043 { 3044 } 3045 3046 #ifdef CONFIG_BPF_SYSCALL 3047 static inline int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, 3048 void *key, void *value) 3049 { 3050 return -EOPNOTSUPP; 3051 } 3052 3053 static inline int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, 3054 void *key, void *value, 3055 u64 map_flags) 3056 { 3057 return -EOPNOTSUPP; 3058 } 3059 #endif /* CONFIG_BPF_SYSCALL */ 3060 #endif /* defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) */ 3061 3062 /* verifier prototypes for helper functions called from eBPF programs */ 3063 extern const struct bpf_func_proto bpf_map_lookup_elem_proto; 3064 extern const struct bpf_func_proto bpf_map_update_elem_proto; 3065 extern const struct bpf_func_proto bpf_map_delete_elem_proto; 3066 extern const struct bpf_func_proto bpf_map_push_elem_proto; 3067 extern const struct bpf_func_proto bpf_map_pop_elem_proto; 3068 extern const struct bpf_func_proto bpf_map_peek_elem_proto; 3069 extern const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto; 3070 3071 extern const struct bpf_func_proto bpf_get_prandom_u32_proto; 3072 extern const struct bpf_func_proto bpf_get_smp_processor_id_proto; 3073 extern const struct bpf_func_proto bpf_get_numa_node_id_proto; 3074 extern const struct bpf_func_proto bpf_tail_call_proto; 3075 extern const struct bpf_func_proto bpf_ktime_get_ns_proto; 3076 extern const struct bpf_func_proto bpf_ktime_get_boot_ns_proto; 3077 extern const struct bpf_func_proto bpf_ktime_get_tai_ns_proto; 3078 extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto; 3079 extern const struct bpf_func_proto bpf_get_current_uid_gid_proto; 3080 extern const struct bpf_func_proto bpf_get_current_comm_proto; 3081 extern const struct bpf_func_proto bpf_get_stackid_proto; 3082 extern const struct bpf_func_proto bpf_get_stack_proto; 3083 extern const struct bpf_func_proto bpf_get_task_stack_proto; 3084 extern const struct bpf_func_proto bpf_get_stackid_proto_pe; 3085 extern const struct bpf_func_proto bpf_get_stack_proto_pe; 3086 extern const struct bpf_func_proto bpf_sock_map_update_proto; 3087 extern const struct bpf_func_proto bpf_sock_hash_update_proto; 3088 extern const struct bpf_func_proto bpf_get_current_cgroup_id_proto; 3089 extern const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto; 3090 extern const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto; 3091 extern const struct bpf_func_proto bpf_msg_redirect_hash_proto; 3092 extern const struct bpf_func_proto bpf_msg_redirect_map_proto; 3093 extern const struct bpf_func_proto bpf_sk_redirect_hash_proto; 3094 extern const struct bpf_func_proto bpf_sk_redirect_map_proto; 3095 extern const struct bpf_func_proto bpf_spin_lock_proto; 3096 extern const struct bpf_func_proto bpf_spin_unlock_proto; 3097 extern const struct bpf_func_proto bpf_get_local_storage_proto; 3098 extern const struct bpf_func_proto bpf_strtol_proto; 3099 extern const struct bpf_func_proto bpf_strtoul_proto; 3100 extern const struct bpf_func_proto bpf_tcp_sock_proto; 3101 extern const struct bpf_func_proto bpf_jiffies64_proto; 3102 extern const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto; 3103 extern const struct bpf_func_proto bpf_event_output_data_proto; 3104 extern const struct bpf_func_proto bpf_ringbuf_output_proto; 3105 extern const struct bpf_func_proto bpf_ringbuf_reserve_proto; 3106 extern const struct bpf_func_proto bpf_ringbuf_submit_proto; 3107 extern const struct bpf_func_proto bpf_ringbuf_discard_proto; 3108 extern const struct bpf_func_proto bpf_ringbuf_query_proto; 3109 extern const struct bpf_func_proto bpf_ringbuf_reserve_dynptr_proto; 3110 extern const struct bpf_func_proto bpf_ringbuf_submit_dynptr_proto; 3111 extern const struct bpf_func_proto bpf_ringbuf_discard_dynptr_proto; 3112 extern const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto; 3113 extern const struct bpf_func_proto bpf_skc_to_tcp_sock_proto; 3114 extern const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto; 3115 extern const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto; 3116 extern const struct bpf_func_proto bpf_skc_to_udp6_sock_proto; 3117 extern const struct bpf_func_proto bpf_skc_to_unix_sock_proto; 3118 extern const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto; 3119 extern const struct bpf_func_proto bpf_copy_from_user_proto; 3120 extern const struct bpf_func_proto bpf_snprintf_btf_proto; 3121 extern const struct bpf_func_proto bpf_snprintf_proto; 3122 extern const struct bpf_func_proto bpf_per_cpu_ptr_proto; 3123 extern const struct bpf_func_proto bpf_this_cpu_ptr_proto; 3124 extern const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto; 3125 extern const struct bpf_func_proto bpf_sock_from_file_proto; 3126 extern const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto; 3127 extern const struct bpf_func_proto bpf_task_storage_get_recur_proto; 3128 extern const struct bpf_func_proto bpf_task_storage_get_proto; 3129 extern const struct bpf_func_proto bpf_task_storage_delete_recur_proto; 3130 extern const struct bpf_func_proto bpf_task_storage_delete_proto; 3131 extern const struct bpf_func_proto bpf_for_each_map_elem_proto; 3132 extern const struct bpf_func_proto bpf_btf_find_by_name_kind_proto; 3133 extern const struct bpf_func_proto bpf_sk_setsockopt_proto; 3134 extern const struct bpf_func_proto bpf_sk_getsockopt_proto; 3135 extern const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto; 3136 extern const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto; 3137 extern const struct bpf_func_proto bpf_find_vma_proto; 3138 extern const struct bpf_func_proto bpf_loop_proto; 3139 extern const struct bpf_func_proto bpf_copy_from_user_task_proto; 3140 extern const struct bpf_func_proto bpf_set_retval_proto; 3141 extern const struct bpf_func_proto bpf_get_retval_proto; 3142 extern const struct bpf_func_proto bpf_user_ringbuf_drain_proto; 3143 extern const struct bpf_func_proto bpf_cgrp_storage_get_proto; 3144 extern const struct bpf_func_proto bpf_cgrp_storage_delete_proto; 3145 3146 const struct bpf_func_proto *tracing_prog_func_proto( 3147 enum bpf_func_id func_id, const struct bpf_prog *prog); 3148 3149 /* Shared helpers among cBPF and eBPF. */ 3150 void bpf_user_rnd_init_once(void); 3151 u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 3152 u64 bpf_get_raw_cpu_id(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 3153 3154 #if defined(CONFIG_NET) 3155 bool bpf_sock_common_is_valid_access(int off, int size, 3156 enum bpf_access_type type, 3157 struct bpf_insn_access_aux *info); 3158 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 3159 struct bpf_insn_access_aux *info); 3160 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 3161 const struct bpf_insn *si, 3162 struct bpf_insn *insn_buf, 3163 struct bpf_prog *prog, 3164 u32 *target_size); 3165 int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags, 3166 struct bpf_dynptr_kern *ptr); 3167 #else 3168 static inline bool bpf_sock_common_is_valid_access(int off, int size, 3169 enum bpf_access_type type, 3170 struct bpf_insn_access_aux *info) 3171 { 3172 return false; 3173 } 3174 static inline bool bpf_sock_is_valid_access(int off, int size, 3175 enum bpf_access_type type, 3176 struct bpf_insn_access_aux *info) 3177 { 3178 return false; 3179 } 3180 static inline u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 3181 const struct bpf_insn *si, 3182 struct bpf_insn *insn_buf, 3183 struct bpf_prog *prog, 3184 u32 *target_size) 3185 { 3186 return 0; 3187 } 3188 static inline int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags, 3189 struct bpf_dynptr_kern *ptr) 3190 { 3191 return -EOPNOTSUPP; 3192 } 3193 #endif 3194 3195 #ifdef CONFIG_INET 3196 struct sk_reuseport_kern { 3197 struct sk_buff *skb; 3198 struct sock *sk; 3199 struct sock *selected_sk; 3200 struct sock *migrating_sk; 3201 void *data_end; 3202 u32 hash; 3203 u32 reuseport_id; 3204 bool bind_inany; 3205 }; 3206 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 3207 struct bpf_insn_access_aux *info); 3208 3209 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 3210 const struct bpf_insn *si, 3211 struct bpf_insn *insn_buf, 3212 struct bpf_prog *prog, 3213 u32 *target_size); 3214 3215 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 3216 struct bpf_insn_access_aux *info); 3217 3218 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 3219 const struct bpf_insn *si, 3220 struct bpf_insn *insn_buf, 3221 struct bpf_prog *prog, 3222 u32 *target_size); 3223 #else 3224 static inline bool bpf_tcp_sock_is_valid_access(int off, int size, 3225 enum bpf_access_type type, 3226 struct bpf_insn_access_aux *info) 3227 { 3228 return false; 3229 } 3230 3231 static inline u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 3232 const struct bpf_insn *si, 3233 struct bpf_insn *insn_buf, 3234 struct bpf_prog *prog, 3235 u32 *target_size) 3236 { 3237 return 0; 3238 } 3239 static inline bool bpf_xdp_sock_is_valid_access(int off, int size, 3240 enum bpf_access_type type, 3241 struct bpf_insn_access_aux *info) 3242 { 3243 return false; 3244 } 3245 3246 static inline u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 3247 const struct bpf_insn *si, 3248 struct bpf_insn *insn_buf, 3249 struct bpf_prog *prog, 3250 u32 *target_size) 3251 { 3252 return 0; 3253 } 3254 #endif /* CONFIG_INET */ 3255 3256 enum bpf_text_poke_type { 3257 BPF_MOD_CALL, 3258 BPF_MOD_JUMP, 3259 }; 3260 3261 int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, 3262 void *addr1, void *addr2); 3263 3264 void *bpf_arch_text_copy(void *dst, void *src, size_t len); 3265 int bpf_arch_text_invalidate(void *dst, size_t len); 3266 3267 struct btf_id_set; 3268 bool btf_id_set_contains(const struct btf_id_set *set, u32 id); 3269 3270 #define MAX_BPRINTF_VARARGS 12 3271 #define MAX_BPRINTF_BUF 1024 3272 3273 struct bpf_bprintf_data { 3274 u32 *bin_args; 3275 char *buf; 3276 bool get_bin_args; 3277 bool get_buf; 3278 }; 3279 3280 int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args, 3281 u32 num_args, struct bpf_bprintf_data *data); 3282 void bpf_bprintf_cleanup(struct bpf_bprintf_data *data); 3283 3284 #ifdef CONFIG_BPF_LSM 3285 void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype); 3286 void bpf_cgroup_atype_put(int cgroup_atype); 3287 #else 3288 static inline void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype) {} 3289 static inline void bpf_cgroup_atype_put(int cgroup_atype) {} 3290 #endif /* CONFIG_BPF_LSM */ 3291 3292 struct key; 3293 3294 #ifdef CONFIG_KEYS 3295 struct bpf_key { 3296 struct key *key; 3297 bool has_ref; 3298 }; 3299 #endif /* CONFIG_KEYS */ 3300 3301 static inline bool type_is_alloc(u32 type) 3302 { 3303 return type & MEM_ALLOC; 3304 } 3305 3306 static inline gfp_t bpf_memcg_flags(gfp_t flags) 3307 { 3308 if (memcg_bpf_enabled()) 3309 return flags | __GFP_ACCOUNT; 3310 return flags; 3311 } 3312 3313 static inline bool bpf_is_subprog(const struct bpf_prog *prog) 3314 { 3315 return prog->aux->func_idx != 0; 3316 } 3317 3318 #endif /* _LINUX_BPF_H */ 3319