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