at v5.16-rc2 73 kB view raw
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 9#include <linux/workqueue.h> 10#include <linux/file.h> 11#include <linux/percpu.h> 12#include <linux/err.h> 13#include <linux/rbtree_latch.h> 14#include <linux/numa.h> 15#include <linux/mm_types.h> 16#include <linux/wait.h> 17#include <linux/refcount.h> 18#include <linux/mutex.h> 19#include <linux/module.h> 20#include <linux/kallsyms.h> 21#include <linux/capability.h> 22#include <linux/sched/mm.h> 23#include <linux/slab.h> 24#include <linux/percpu-refcount.h> 25#include <linux/bpfptr.h> 26 27struct bpf_verifier_env; 28struct bpf_verifier_log; 29struct perf_event; 30struct bpf_prog; 31struct bpf_prog_aux; 32struct bpf_map; 33struct sock; 34struct seq_file; 35struct btf; 36struct btf_type; 37struct exception_table_entry; 38struct seq_operations; 39struct bpf_iter_aux_info; 40struct bpf_local_storage; 41struct bpf_local_storage_map; 42struct kobject; 43struct mem_cgroup; 44struct module; 45struct bpf_func_state; 46 47extern struct idr btf_idr; 48extern spinlock_t btf_idr_lock; 49extern struct kobject *btf_kobj; 50 51typedef u64 (*bpf_callback_t)(u64, u64, u64, u64, u64); 52typedef int (*bpf_iter_init_seq_priv_t)(void *private_data, 53 struct bpf_iter_aux_info *aux); 54typedef void (*bpf_iter_fini_seq_priv_t)(void *private_data); 55struct bpf_iter_seq_info { 56 const struct seq_operations *seq_ops; 57 bpf_iter_init_seq_priv_t init_seq_private; 58 bpf_iter_fini_seq_priv_t fini_seq_private; 59 u32 seq_priv_size; 60}; 61 62/* map is generic key/value storage optionally accessible by eBPF programs */ 63struct bpf_map_ops { 64 /* funcs callable from userspace (via syscall) */ 65 int (*map_alloc_check)(union bpf_attr *attr); 66 struct bpf_map *(*map_alloc)(union bpf_attr *attr); 67 void (*map_release)(struct bpf_map *map, struct file *map_file); 68 void (*map_free)(struct bpf_map *map); 69 int (*map_get_next_key)(struct bpf_map *map, void *key, void *next_key); 70 void (*map_release_uref)(struct bpf_map *map); 71 void *(*map_lookup_elem_sys_only)(struct bpf_map *map, void *key); 72 int (*map_lookup_batch)(struct bpf_map *map, const union bpf_attr *attr, 73 union bpf_attr __user *uattr); 74 int (*map_lookup_and_delete_elem)(struct bpf_map *map, void *key, 75 void *value, u64 flags); 76 int (*map_lookup_and_delete_batch)(struct bpf_map *map, 77 const union bpf_attr *attr, 78 union bpf_attr __user *uattr); 79 int (*map_update_batch)(struct bpf_map *map, const union bpf_attr *attr, 80 union bpf_attr __user *uattr); 81 int (*map_delete_batch)(struct bpf_map *map, const union bpf_attr *attr, 82 union bpf_attr __user *uattr); 83 84 /* funcs callable from userspace and from eBPF programs */ 85 void *(*map_lookup_elem)(struct bpf_map *map, void *key); 86 int (*map_update_elem)(struct bpf_map *map, void *key, void *value, u64 flags); 87 int (*map_delete_elem)(struct bpf_map *map, void *key); 88 int (*map_push_elem)(struct bpf_map *map, void *value, u64 flags); 89 int (*map_pop_elem)(struct bpf_map *map, void *value); 90 int (*map_peek_elem)(struct bpf_map *map, void *value); 91 92 /* funcs called by prog_array and perf_event_array map */ 93 void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file, 94 int fd); 95 void (*map_fd_put_ptr)(void *ptr); 96 int (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf); 97 u32 (*map_fd_sys_lookup_elem)(void *ptr); 98 void (*map_seq_show_elem)(struct bpf_map *map, void *key, 99 struct seq_file *m); 100 int (*map_check_btf)(const struct bpf_map *map, 101 const struct btf *btf, 102 const struct btf_type *key_type, 103 const struct btf_type *value_type); 104 105 /* Prog poke tracking helpers. */ 106 int (*map_poke_track)(struct bpf_map *map, struct bpf_prog_aux *aux); 107 void (*map_poke_untrack)(struct bpf_map *map, struct bpf_prog_aux *aux); 108 void (*map_poke_run)(struct bpf_map *map, u32 key, struct bpf_prog *old, 109 struct bpf_prog *new); 110 111 /* Direct value access helpers. */ 112 int (*map_direct_value_addr)(const struct bpf_map *map, 113 u64 *imm, u32 off); 114 int (*map_direct_value_meta)(const struct bpf_map *map, 115 u64 imm, u32 *off); 116 int (*map_mmap)(struct bpf_map *map, struct vm_area_struct *vma); 117 __poll_t (*map_poll)(struct bpf_map *map, struct file *filp, 118 struct poll_table_struct *pts); 119 120 /* Functions called by bpf_local_storage maps */ 121 int (*map_local_storage_charge)(struct bpf_local_storage_map *smap, 122 void *owner, u32 size); 123 void (*map_local_storage_uncharge)(struct bpf_local_storage_map *smap, 124 void *owner, u32 size); 125 struct bpf_local_storage __rcu ** (*map_owner_storage_ptr)(void *owner); 126 127 /* Misc helpers.*/ 128 int (*map_redirect)(struct bpf_map *map, u32 ifindex, u64 flags); 129 130 /* map_meta_equal must be implemented for maps that can be 131 * used as an inner map. It is a runtime check to ensure 132 * an inner map can be inserted to an outer map. 133 * 134 * Some properties of the inner map has been used during the 135 * verification time. When inserting an inner map at the runtime, 136 * map_meta_equal has to ensure the inserting map has the same 137 * properties that the verifier has used earlier. 138 */ 139 bool (*map_meta_equal)(const struct bpf_map *meta0, 140 const struct bpf_map *meta1); 141 142 143 int (*map_set_for_each_callback_args)(struct bpf_verifier_env *env, 144 struct bpf_func_state *caller, 145 struct bpf_func_state *callee); 146 int (*map_for_each_callback)(struct bpf_map *map, 147 bpf_callback_t callback_fn, 148 void *callback_ctx, u64 flags); 149 150 /* BTF name and id of struct allocated by map_alloc */ 151 const char * const map_btf_name; 152 int *map_btf_id; 153 154 /* bpf_iter info used to open a seq_file */ 155 const struct bpf_iter_seq_info *iter_seq_info; 156}; 157 158struct bpf_map { 159 /* The first two cachelines with read-mostly members of which some 160 * are also accessed in fast-path (e.g. ops, max_entries). 161 */ 162 const struct bpf_map_ops *ops ____cacheline_aligned; 163 struct bpf_map *inner_map_meta; 164#ifdef CONFIG_SECURITY 165 void *security; 166#endif 167 enum bpf_map_type map_type; 168 u32 key_size; 169 u32 value_size; 170 u32 max_entries; 171 u64 map_extra; /* any per-map-type extra fields */ 172 u32 map_flags; 173 int spin_lock_off; /* >=0 valid offset, <0 error */ 174 int timer_off; /* >=0 valid offset, <0 error */ 175 u32 id; 176 int numa_node; 177 u32 btf_key_type_id; 178 u32 btf_value_type_id; 179 u32 btf_vmlinux_value_type_id; 180 struct btf *btf; 181#ifdef CONFIG_MEMCG_KMEM 182 struct mem_cgroup *memcg; 183#endif 184 char name[BPF_OBJ_NAME_LEN]; 185 bool bypass_spec_v1; 186 bool frozen; /* write-once; write-protected by freeze_mutex */ 187 /* 14 bytes hole */ 188 189 /* The 3rd and 4th cacheline with misc members to avoid false sharing 190 * particularly with refcounting. 191 */ 192 atomic64_t refcnt ____cacheline_aligned; 193 atomic64_t usercnt; 194 struct work_struct work; 195 struct mutex freeze_mutex; 196 atomic64_t writecnt; 197}; 198 199static inline bool map_value_has_spin_lock(const struct bpf_map *map) 200{ 201 return map->spin_lock_off >= 0; 202} 203 204static inline bool map_value_has_timer(const struct bpf_map *map) 205{ 206 return map->timer_off >= 0; 207} 208 209static inline void check_and_init_map_value(struct bpf_map *map, void *dst) 210{ 211 if (unlikely(map_value_has_spin_lock(map))) 212 *(struct bpf_spin_lock *)(dst + map->spin_lock_off) = 213 (struct bpf_spin_lock){}; 214 if (unlikely(map_value_has_timer(map))) 215 *(struct bpf_timer *)(dst + map->timer_off) = 216 (struct bpf_timer){}; 217} 218 219/* copy everything but bpf_spin_lock and bpf_timer. There could be one of each. */ 220static inline void copy_map_value(struct bpf_map *map, void *dst, void *src) 221{ 222 u32 s_off = 0, s_sz = 0, t_off = 0, t_sz = 0; 223 224 if (unlikely(map_value_has_spin_lock(map))) { 225 s_off = map->spin_lock_off; 226 s_sz = sizeof(struct bpf_spin_lock); 227 } else if (unlikely(map_value_has_timer(map))) { 228 t_off = map->timer_off; 229 t_sz = sizeof(struct bpf_timer); 230 } 231 232 if (unlikely(s_sz || t_sz)) { 233 if (s_off < t_off || !s_sz) { 234 swap(s_off, t_off); 235 swap(s_sz, t_sz); 236 } 237 memcpy(dst, src, t_off); 238 memcpy(dst + t_off + t_sz, 239 src + t_off + t_sz, 240 s_off - t_off - t_sz); 241 memcpy(dst + s_off + s_sz, 242 src + s_off + s_sz, 243 map->value_size - s_off - s_sz); 244 } else { 245 memcpy(dst, src, map->value_size); 246 } 247} 248void copy_map_value_locked(struct bpf_map *map, void *dst, void *src, 249 bool lock_src); 250void bpf_timer_cancel_and_free(void *timer); 251int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size); 252 253struct bpf_offload_dev; 254struct bpf_offloaded_map; 255 256struct bpf_map_dev_ops { 257 int (*map_get_next_key)(struct bpf_offloaded_map *map, 258 void *key, void *next_key); 259 int (*map_lookup_elem)(struct bpf_offloaded_map *map, 260 void *key, void *value); 261 int (*map_update_elem)(struct bpf_offloaded_map *map, 262 void *key, void *value, u64 flags); 263 int (*map_delete_elem)(struct bpf_offloaded_map *map, void *key); 264}; 265 266struct bpf_offloaded_map { 267 struct bpf_map map; 268 struct net_device *netdev; 269 const struct bpf_map_dev_ops *dev_ops; 270 void *dev_priv; 271 struct list_head offloads; 272}; 273 274static inline struct bpf_offloaded_map *map_to_offmap(struct bpf_map *map) 275{ 276 return container_of(map, struct bpf_offloaded_map, map); 277} 278 279static inline bool bpf_map_offload_neutral(const struct bpf_map *map) 280{ 281 return map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; 282} 283 284static inline bool bpf_map_support_seq_show(const struct bpf_map *map) 285{ 286 return (map->btf_value_type_id || map->btf_vmlinux_value_type_id) && 287 map->ops->map_seq_show_elem; 288} 289 290int map_check_no_btf(const struct bpf_map *map, 291 const struct btf *btf, 292 const struct btf_type *key_type, 293 const struct btf_type *value_type); 294 295bool bpf_map_meta_equal(const struct bpf_map *meta0, 296 const struct bpf_map *meta1); 297 298extern const struct bpf_map_ops bpf_map_offload_ops; 299 300/* function argument constraints */ 301enum bpf_arg_type { 302 ARG_DONTCARE = 0, /* unused argument in helper function */ 303 304 /* the following constraints used to prototype 305 * bpf_map_lookup/update/delete_elem() functions 306 */ 307 ARG_CONST_MAP_PTR, /* const argument used as pointer to bpf_map */ 308 ARG_PTR_TO_MAP_KEY, /* pointer to stack used as map key */ 309 ARG_PTR_TO_MAP_VALUE, /* pointer to stack used as map value */ 310 ARG_PTR_TO_UNINIT_MAP_VALUE, /* pointer to valid memory used to store a map value */ 311 ARG_PTR_TO_MAP_VALUE_OR_NULL, /* pointer to stack used as map value or NULL */ 312 313 /* the following constraints used to prototype bpf_memcmp() and other 314 * functions that access data on eBPF program stack 315 */ 316 ARG_PTR_TO_MEM, /* pointer to valid memory (stack, packet, map value) */ 317 ARG_PTR_TO_MEM_OR_NULL, /* pointer to valid memory or NULL */ 318 ARG_PTR_TO_UNINIT_MEM, /* pointer to memory does not need to be initialized, 319 * helper function must fill all bytes or clear 320 * them in error case. 321 */ 322 323 ARG_CONST_SIZE, /* number of bytes accessed from memory */ 324 ARG_CONST_SIZE_OR_ZERO, /* number of bytes accessed from memory or 0 */ 325 326 ARG_PTR_TO_CTX, /* pointer to context */ 327 ARG_PTR_TO_CTX_OR_NULL, /* pointer to context or NULL */ 328 ARG_ANYTHING, /* any (initialized) argument is ok */ 329 ARG_PTR_TO_SPIN_LOCK, /* pointer to bpf_spin_lock */ 330 ARG_PTR_TO_SOCK_COMMON, /* pointer to sock_common */ 331 ARG_PTR_TO_INT, /* pointer to int */ 332 ARG_PTR_TO_LONG, /* pointer to long */ 333 ARG_PTR_TO_SOCKET, /* pointer to bpf_sock (fullsock) */ 334 ARG_PTR_TO_SOCKET_OR_NULL, /* pointer to bpf_sock (fullsock) or NULL */ 335 ARG_PTR_TO_BTF_ID, /* pointer to in-kernel struct */ 336 ARG_PTR_TO_ALLOC_MEM, /* pointer to dynamically allocated memory */ 337 ARG_PTR_TO_ALLOC_MEM_OR_NULL, /* pointer to dynamically allocated memory or NULL */ 338 ARG_CONST_ALLOC_SIZE_OR_ZERO, /* number of allocated bytes requested */ 339 ARG_PTR_TO_BTF_ID_SOCK_COMMON, /* pointer to in-kernel sock_common or bpf-mirrored bpf_sock */ 340 ARG_PTR_TO_PERCPU_BTF_ID, /* pointer to in-kernel percpu type */ 341 ARG_PTR_TO_FUNC, /* pointer to a bpf program function */ 342 ARG_PTR_TO_STACK_OR_NULL, /* pointer to stack or NULL */ 343 ARG_PTR_TO_CONST_STR, /* pointer to a null terminated read-only string */ 344 ARG_PTR_TO_TIMER, /* pointer to bpf_timer */ 345 __BPF_ARG_TYPE_MAX, 346}; 347 348/* type of values returned from helper functions */ 349enum bpf_return_type { 350 RET_INTEGER, /* function returns integer */ 351 RET_VOID, /* function doesn't return anything */ 352 RET_PTR_TO_MAP_VALUE, /* returns a pointer to map elem value */ 353 RET_PTR_TO_MAP_VALUE_OR_NULL, /* returns a pointer to map elem value or NULL */ 354 RET_PTR_TO_SOCKET_OR_NULL, /* returns a pointer to a socket or NULL */ 355 RET_PTR_TO_TCP_SOCK_OR_NULL, /* returns a pointer to a tcp_sock or NULL */ 356 RET_PTR_TO_SOCK_COMMON_OR_NULL, /* returns a pointer to a sock_common or NULL */ 357 RET_PTR_TO_ALLOC_MEM_OR_NULL, /* returns a pointer to dynamically allocated memory or NULL */ 358 RET_PTR_TO_BTF_ID_OR_NULL, /* returns a pointer to a btf_id or NULL */ 359 RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL, /* returns a pointer to a valid memory or a btf_id or NULL */ 360 RET_PTR_TO_MEM_OR_BTF_ID, /* returns a pointer to a valid memory or a btf_id */ 361 RET_PTR_TO_BTF_ID, /* returns a pointer to a btf_id */ 362}; 363 364/* eBPF function prototype used by verifier to allow BPF_CALLs from eBPF programs 365 * to in-kernel helper functions and for adjusting imm32 field in BPF_CALL 366 * instructions after verifying 367 */ 368struct bpf_func_proto { 369 u64 (*func)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 370 bool gpl_only; 371 bool pkt_access; 372 enum bpf_return_type ret_type; 373 union { 374 struct { 375 enum bpf_arg_type arg1_type; 376 enum bpf_arg_type arg2_type; 377 enum bpf_arg_type arg3_type; 378 enum bpf_arg_type arg4_type; 379 enum bpf_arg_type arg5_type; 380 }; 381 enum bpf_arg_type arg_type[5]; 382 }; 383 union { 384 struct { 385 u32 *arg1_btf_id; 386 u32 *arg2_btf_id; 387 u32 *arg3_btf_id; 388 u32 *arg4_btf_id; 389 u32 *arg5_btf_id; 390 }; 391 u32 *arg_btf_id[5]; 392 }; 393 int *ret_btf_id; /* return value btf_id */ 394 bool (*allowed)(const struct bpf_prog *prog); 395}; 396 397/* bpf_context is intentionally undefined structure. Pointer to bpf_context is 398 * the first argument to eBPF programs. 399 * For socket filters: 'struct bpf_context *' == 'struct sk_buff *' 400 */ 401struct bpf_context; 402 403enum bpf_access_type { 404 BPF_READ = 1, 405 BPF_WRITE = 2 406}; 407 408/* types of values stored in eBPF registers */ 409/* Pointer types represent: 410 * pointer 411 * pointer + imm 412 * pointer + (u16) var 413 * pointer + (u16) var + imm 414 * if (range > 0) then [ptr, ptr + range - off) is safe to access 415 * if (id > 0) means that some 'var' was added 416 * if (off > 0) means that 'imm' was added 417 */ 418enum bpf_reg_type { 419 NOT_INIT = 0, /* nothing was written into register */ 420 SCALAR_VALUE, /* reg doesn't contain a valid pointer */ 421 PTR_TO_CTX, /* reg points to bpf_context */ 422 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */ 423 PTR_TO_MAP_VALUE, /* reg points to map element value */ 424 PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */ 425 PTR_TO_STACK, /* reg == frame_pointer + offset */ 426 PTR_TO_PACKET_META, /* skb->data - meta_len */ 427 PTR_TO_PACKET, /* reg points to skb->data */ 428 PTR_TO_PACKET_END, /* skb->data + headlen */ 429 PTR_TO_FLOW_KEYS, /* reg points to bpf_flow_keys */ 430 PTR_TO_SOCKET, /* reg points to struct bpf_sock */ 431 PTR_TO_SOCKET_OR_NULL, /* reg points to struct bpf_sock or NULL */ 432 PTR_TO_SOCK_COMMON, /* reg points to sock_common */ 433 PTR_TO_SOCK_COMMON_OR_NULL, /* reg points to sock_common or NULL */ 434 PTR_TO_TCP_SOCK, /* reg points to struct tcp_sock */ 435 PTR_TO_TCP_SOCK_OR_NULL, /* reg points to struct tcp_sock or NULL */ 436 PTR_TO_TP_BUFFER, /* reg points to a writable raw tp's buffer */ 437 PTR_TO_XDP_SOCK, /* reg points to struct xdp_sock */ 438 /* PTR_TO_BTF_ID points to a kernel struct that does not need 439 * to be null checked by the BPF program. This does not imply the 440 * pointer is _not_ null and in practice this can easily be a null 441 * pointer when reading pointer chains. The assumption is program 442 * context will handle null pointer dereference typically via fault 443 * handling. The verifier must keep this in mind and can make no 444 * assumptions about null or non-null when doing branch analysis. 445 * Further, when passed into helpers the helpers can not, without 446 * additional context, assume the value is non-null. 447 */ 448 PTR_TO_BTF_ID, 449 /* PTR_TO_BTF_ID_OR_NULL points to a kernel struct that has not 450 * been checked for null. Used primarily to inform the verifier 451 * an explicit null check is required for this struct. 452 */ 453 PTR_TO_BTF_ID_OR_NULL, 454 PTR_TO_MEM, /* reg points to valid memory region */ 455 PTR_TO_MEM_OR_NULL, /* reg points to valid memory region or NULL */ 456 PTR_TO_RDONLY_BUF, /* reg points to a readonly buffer */ 457 PTR_TO_RDONLY_BUF_OR_NULL, /* reg points to a readonly buffer or NULL */ 458 PTR_TO_RDWR_BUF, /* reg points to a read/write buffer */ 459 PTR_TO_RDWR_BUF_OR_NULL, /* reg points to a read/write buffer or NULL */ 460 PTR_TO_PERCPU_BTF_ID, /* reg points to a percpu kernel variable */ 461 PTR_TO_FUNC, /* reg points to a bpf program function */ 462 PTR_TO_MAP_KEY, /* reg points to a map element key */ 463 __BPF_REG_TYPE_MAX, 464}; 465 466/* The information passed from prog-specific *_is_valid_access 467 * back to the verifier. 468 */ 469struct bpf_insn_access_aux { 470 enum bpf_reg_type reg_type; 471 union { 472 int ctx_field_size; 473 struct { 474 struct btf *btf; 475 u32 btf_id; 476 }; 477 }; 478 struct bpf_verifier_log *log; /* for verbose logs */ 479}; 480 481static inline void 482bpf_ctx_record_field_size(struct bpf_insn_access_aux *aux, u32 size) 483{ 484 aux->ctx_field_size = size; 485} 486 487static inline bool bpf_pseudo_func(const struct bpf_insn *insn) 488{ 489 return insn->code == (BPF_LD | BPF_IMM | BPF_DW) && 490 insn->src_reg == BPF_PSEUDO_FUNC; 491} 492 493struct bpf_prog_ops { 494 int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr, 495 union bpf_attr __user *uattr); 496}; 497 498struct bpf_verifier_ops { 499 /* return eBPF function prototype for verification */ 500 const struct bpf_func_proto * 501 (*get_func_proto)(enum bpf_func_id func_id, 502 const struct bpf_prog *prog); 503 504 /* return true if 'size' wide access at offset 'off' within bpf_context 505 * with 'type' (read or write) is allowed 506 */ 507 bool (*is_valid_access)(int off, int size, enum bpf_access_type type, 508 const struct bpf_prog *prog, 509 struct bpf_insn_access_aux *info); 510 int (*gen_prologue)(struct bpf_insn *insn, bool direct_write, 511 const struct bpf_prog *prog); 512 int (*gen_ld_abs)(const struct bpf_insn *orig, 513 struct bpf_insn *insn_buf); 514 u32 (*convert_ctx_access)(enum bpf_access_type type, 515 const struct bpf_insn *src, 516 struct bpf_insn *dst, 517 struct bpf_prog *prog, u32 *target_size); 518 int (*btf_struct_access)(struct bpf_verifier_log *log, 519 const struct btf *btf, 520 const struct btf_type *t, int off, int size, 521 enum bpf_access_type atype, 522 u32 *next_btf_id); 523 bool (*check_kfunc_call)(u32 kfunc_btf_id, struct module *owner); 524}; 525 526struct bpf_prog_offload_ops { 527 /* verifier basic callbacks */ 528 int (*insn_hook)(struct bpf_verifier_env *env, 529 int insn_idx, int prev_insn_idx); 530 int (*finalize)(struct bpf_verifier_env *env); 531 /* verifier optimization callbacks (called after .finalize) */ 532 int (*replace_insn)(struct bpf_verifier_env *env, u32 off, 533 struct bpf_insn *insn); 534 int (*remove_insns)(struct bpf_verifier_env *env, u32 off, u32 cnt); 535 /* program management callbacks */ 536 int (*prepare)(struct bpf_prog *prog); 537 int (*translate)(struct bpf_prog *prog); 538 void (*destroy)(struct bpf_prog *prog); 539}; 540 541struct bpf_prog_offload { 542 struct bpf_prog *prog; 543 struct net_device *netdev; 544 struct bpf_offload_dev *offdev; 545 void *dev_priv; 546 struct list_head offloads; 547 bool dev_state; 548 bool opt_failed; 549 void *jited_image; 550 u32 jited_len; 551}; 552 553enum bpf_cgroup_storage_type { 554 BPF_CGROUP_STORAGE_SHARED, 555 BPF_CGROUP_STORAGE_PERCPU, 556 __BPF_CGROUP_STORAGE_MAX 557}; 558 559#define MAX_BPF_CGROUP_STORAGE_TYPE __BPF_CGROUP_STORAGE_MAX 560 561/* The longest tracepoint has 12 args. 562 * See include/trace/bpf_probe.h 563 */ 564#define MAX_BPF_FUNC_ARGS 12 565 566/* The maximum number of arguments passed through registers 567 * a single function may have. 568 */ 569#define MAX_BPF_FUNC_REG_ARGS 5 570 571struct btf_func_model { 572 u8 ret_size; 573 u8 nr_args; 574 u8 arg_size[MAX_BPF_FUNC_ARGS]; 575}; 576 577/* Restore arguments before returning from trampoline to let original function 578 * continue executing. This flag is used for fentry progs when there are no 579 * fexit progs. 580 */ 581#define BPF_TRAMP_F_RESTORE_REGS BIT(0) 582/* Call original function after fentry progs, but before fexit progs. 583 * Makes sense for fentry/fexit, normal calls and indirect calls. 584 */ 585#define BPF_TRAMP_F_CALL_ORIG BIT(1) 586/* Skip current frame and return to parent. Makes sense for fentry/fexit 587 * programs only. Should not be used with normal calls and indirect calls. 588 */ 589#define BPF_TRAMP_F_SKIP_FRAME BIT(2) 590/* Store IP address of the caller on the trampoline stack, 591 * so it's available for trampoline's programs. 592 */ 593#define BPF_TRAMP_F_IP_ARG BIT(3) 594/* Return the return value of fentry prog. Only used by bpf_struct_ops. */ 595#define BPF_TRAMP_F_RET_FENTRY_RET BIT(4) 596 597/* Each call __bpf_prog_enter + call bpf_func + call __bpf_prog_exit is ~50 598 * bytes on x86. Pick a number to fit into BPF_IMAGE_SIZE / 2 599 */ 600#define BPF_MAX_TRAMP_PROGS 38 601 602struct bpf_tramp_progs { 603 struct bpf_prog *progs[BPF_MAX_TRAMP_PROGS]; 604 int nr_progs; 605}; 606 607/* Different use cases for BPF trampoline: 608 * 1. replace nop at the function entry (kprobe equivalent) 609 * flags = BPF_TRAMP_F_RESTORE_REGS 610 * fentry = a set of programs to run before returning from trampoline 611 * 612 * 2. replace nop at the function entry (kprobe + kretprobe equivalent) 613 * flags = BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_SKIP_FRAME 614 * orig_call = fentry_ip + MCOUNT_INSN_SIZE 615 * fentry = a set of program to run before calling original function 616 * fexit = a set of program to run after original function 617 * 618 * 3. replace direct call instruction anywhere in the function body 619 * or assign a function pointer for indirect call (like tcp_congestion_ops->cong_avoid) 620 * With flags = 0 621 * fentry = a set of programs to run before returning from trampoline 622 * With flags = BPF_TRAMP_F_CALL_ORIG 623 * orig_call = original callback addr or direct function addr 624 * fentry = a set of program to run before calling original function 625 * fexit = a set of program to run after original function 626 */ 627struct bpf_tramp_image; 628int arch_prepare_bpf_trampoline(struct bpf_tramp_image *tr, void *image, void *image_end, 629 const struct btf_func_model *m, u32 flags, 630 struct bpf_tramp_progs *tprogs, 631 void *orig_call); 632/* these two functions are called from generated trampoline */ 633u64 notrace __bpf_prog_enter(struct bpf_prog *prog); 634void notrace __bpf_prog_exit(struct bpf_prog *prog, u64 start); 635u64 notrace __bpf_prog_enter_sleepable(struct bpf_prog *prog); 636void notrace __bpf_prog_exit_sleepable(struct bpf_prog *prog, u64 start); 637void notrace __bpf_tramp_enter(struct bpf_tramp_image *tr); 638void notrace __bpf_tramp_exit(struct bpf_tramp_image *tr); 639 640struct bpf_ksym { 641 unsigned long start; 642 unsigned long end; 643 char name[KSYM_NAME_LEN]; 644 struct list_head lnode; 645 struct latch_tree_node tnode; 646 bool prog; 647}; 648 649enum bpf_tramp_prog_type { 650 BPF_TRAMP_FENTRY, 651 BPF_TRAMP_FEXIT, 652 BPF_TRAMP_MODIFY_RETURN, 653 BPF_TRAMP_MAX, 654 BPF_TRAMP_REPLACE, /* more than MAX */ 655}; 656 657struct bpf_tramp_image { 658 void *image; 659 struct bpf_ksym ksym; 660 struct percpu_ref pcref; 661 void *ip_after_call; 662 void *ip_epilogue; 663 union { 664 struct rcu_head rcu; 665 struct work_struct work; 666 }; 667}; 668 669struct bpf_trampoline { 670 /* hlist for trampoline_table */ 671 struct hlist_node hlist; 672 /* serializes access to fields of this trampoline */ 673 struct mutex mutex; 674 refcount_t refcnt; 675 u64 key; 676 struct { 677 struct btf_func_model model; 678 void *addr; 679 bool ftrace_managed; 680 } func; 681 /* if !NULL this is BPF_PROG_TYPE_EXT program that extends another BPF 682 * program by replacing one of its functions. func.addr is the address 683 * of the function it replaced. 684 */ 685 struct bpf_prog *extension_prog; 686 /* list of BPF programs using this trampoline */ 687 struct hlist_head progs_hlist[BPF_TRAMP_MAX]; 688 /* Number of attached programs. A counter per kind. */ 689 int progs_cnt[BPF_TRAMP_MAX]; 690 /* Executable image of trampoline */ 691 struct bpf_tramp_image *cur_image; 692 u64 selector; 693 struct module *mod; 694}; 695 696struct bpf_attach_target_info { 697 struct btf_func_model fmodel; 698 long tgt_addr; 699 const char *tgt_name; 700 const struct btf_type *tgt_type; 701}; 702 703#define BPF_DISPATCHER_MAX 48 /* Fits in 2048B */ 704 705struct bpf_dispatcher_prog { 706 struct bpf_prog *prog; 707 refcount_t users; 708}; 709 710struct bpf_dispatcher { 711 /* dispatcher mutex */ 712 struct mutex mutex; 713 void *func; 714 struct bpf_dispatcher_prog progs[BPF_DISPATCHER_MAX]; 715 int num_progs; 716 void *image; 717 u32 image_off; 718 struct bpf_ksym ksym; 719}; 720 721static __always_inline __nocfi unsigned int bpf_dispatcher_nop_func( 722 const void *ctx, 723 const struct bpf_insn *insnsi, 724 unsigned int (*bpf_func)(const void *, 725 const struct bpf_insn *)) 726{ 727 return bpf_func(ctx, insnsi); 728} 729#ifdef CONFIG_BPF_JIT 730int bpf_trampoline_link_prog(struct bpf_prog *prog, struct bpf_trampoline *tr); 731int bpf_trampoline_unlink_prog(struct bpf_prog *prog, struct bpf_trampoline *tr); 732struct bpf_trampoline *bpf_trampoline_get(u64 key, 733 struct bpf_attach_target_info *tgt_info); 734void bpf_trampoline_put(struct bpf_trampoline *tr); 735#define BPF_DISPATCHER_INIT(_name) { \ 736 .mutex = __MUTEX_INITIALIZER(_name.mutex), \ 737 .func = &_name##_func, \ 738 .progs = {}, \ 739 .num_progs = 0, \ 740 .image = NULL, \ 741 .image_off = 0, \ 742 .ksym = { \ 743 .name = #_name, \ 744 .lnode = LIST_HEAD_INIT(_name.ksym.lnode), \ 745 }, \ 746} 747 748#define DEFINE_BPF_DISPATCHER(name) \ 749 noinline __nocfi unsigned int bpf_dispatcher_##name##_func( \ 750 const void *ctx, \ 751 const struct bpf_insn *insnsi, \ 752 unsigned int (*bpf_func)(const void *, \ 753 const struct bpf_insn *)) \ 754 { \ 755 return bpf_func(ctx, insnsi); \ 756 } \ 757 EXPORT_SYMBOL(bpf_dispatcher_##name##_func); \ 758 struct bpf_dispatcher bpf_dispatcher_##name = \ 759 BPF_DISPATCHER_INIT(bpf_dispatcher_##name); 760#define DECLARE_BPF_DISPATCHER(name) \ 761 unsigned int bpf_dispatcher_##name##_func( \ 762 const void *ctx, \ 763 const struct bpf_insn *insnsi, \ 764 unsigned int (*bpf_func)(const void *, \ 765 const struct bpf_insn *)); \ 766 extern struct bpf_dispatcher bpf_dispatcher_##name; 767#define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_##name##_func 768#define BPF_DISPATCHER_PTR(name) (&bpf_dispatcher_##name) 769void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, struct bpf_prog *from, 770 struct bpf_prog *to); 771/* Called only from JIT-enabled code, so there's no need for stubs. */ 772void *bpf_jit_alloc_exec_page(void); 773void bpf_image_ksym_add(void *data, struct bpf_ksym *ksym); 774void bpf_image_ksym_del(struct bpf_ksym *ksym); 775void bpf_ksym_add(struct bpf_ksym *ksym); 776void bpf_ksym_del(struct bpf_ksym *ksym); 777int bpf_jit_charge_modmem(u32 pages); 778void bpf_jit_uncharge_modmem(u32 pages); 779#else 780static inline int bpf_trampoline_link_prog(struct bpf_prog *prog, 781 struct bpf_trampoline *tr) 782{ 783 return -ENOTSUPP; 784} 785static inline int bpf_trampoline_unlink_prog(struct bpf_prog *prog, 786 struct bpf_trampoline *tr) 787{ 788 return -ENOTSUPP; 789} 790static inline struct bpf_trampoline *bpf_trampoline_get(u64 key, 791 struct bpf_attach_target_info *tgt_info) 792{ 793 return ERR_PTR(-EOPNOTSUPP); 794} 795static inline void bpf_trampoline_put(struct bpf_trampoline *tr) {} 796#define DEFINE_BPF_DISPATCHER(name) 797#define DECLARE_BPF_DISPATCHER(name) 798#define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_nop_func 799#define BPF_DISPATCHER_PTR(name) NULL 800static inline void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, 801 struct bpf_prog *from, 802 struct bpf_prog *to) {} 803static inline bool is_bpf_image_address(unsigned long address) 804{ 805 return false; 806} 807#endif 808 809struct bpf_func_info_aux { 810 u16 linkage; 811 bool unreliable; 812}; 813 814enum bpf_jit_poke_reason { 815 BPF_POKE_REASON_TAIL_CALL, 816}; 817 818/* Descriptor of pokes pointing /into/ the JITed image. */ 819struct bpf_jit_poke_descriptor { 820 void *tailcall_target; 821 void *tailcall_bypass; 822 void *bypass_addr; 823 void *aux; 824 union { 825 struct { 826 struct bpf_map *map; 827 u32 key; 828 } tail_call; 829 }; 830 bool tailcall_target_stable; 831 u8 adj_off; 832 u16 reason; 833 u32 insn_idx; 834}; 835 836/* reg_type info for ctx arguments */ 837struct bpf_ctx_arg_aux { 838 u32 offset; 839 enum bpf_reg_type reg_type; 840 u32 btf_id; 841}; 842 843struct btf_mod_pair { 844 struct btf *btf; 845 struct module *module; 846}; 847 848struct bpf_kfunc_desc_tab; 849 850struct bpf_prog_aux { 851 atomic64_t refcnt; 852 u32 used_map_cnt; 853 u32 used_btf_cnt; 854 u32 max_ctx_offset; 855 u32 max_pkt_offset; 856 u32 max_tp_access; 857 u32 stack_depth; 858 u32 id; 859 u32 func_cnt; /* used by non-func prog as the number of func progs */ 860 u32 func_idx; /* 0 for non-func prog, the index in func array for func prog */ 861 u32 attach_btf_id; /* in-kernel BTF type id to attach to */ 862 u32 ctx_arg_info_size; 863 u32 max_rdonly_access; 864 u32 max_rdwr_access; 865 struct btf *attach_btf; 866 const struct bpf_ctx_arg_aux *ctx_arg_info; 867 struct mutex dst_mutex; /* protects dst_* pointers below, *after* prog becomes visible */ 868 struct bpf_prog *dst_prog; 869 struct bpf_trampoline *dst_trampoline; 870 enum bpf_prog_type saved_dst_prog_type; 871 enum bpf_attach_type saved_dst_attach_type; 872 bool verifier_zext; /* Zero extensions has been inserted by verifier. */ 873 bool offload_requested; 874 bool attach_btf_trace; /* true if attaching to BTF-enabled raw tp */ 875 bool func_proto_unreliable; 876 bool sleepable; 877 bool tail_call_reachable; 878 struct hlist_node tramp_hlist; 879 /* BTF_KIND_FUNC_PROTO for valid attach_btf_id */ 880 const struct btf_type *attach_func_proto; 881 /* function name for valid attach_btf_id */ 882 const char *attach_func_name; 883 struct bpf_prog **func; 884 void *jit_data; /* JIT specific data. arch dependent */ 885 struct bpf_jit_poke_descriptor *poke_tab; 886 struct bpf_kfunc_desc_tab *kfunc_tab; 887 struct bpf_kfunc_btf_tab *kfunc_btf_tab; 888 u32 size_poke_tab; 889 struct bpf_ksym ksym; 890 const struct bpf_prog_ops *ops; 891 struct bpf_map **used_maps; 892 struct mutex used_maps_mutex; /* mutex for used_maps and used_map_cnt */ 893 struct btf_mod_pair *used_btfs; 894 struct bpf_prog *prog; 895 struct user_struct *user; 896 u64 load_time; /* ns since boottime */ 897 u32 verified_insns; 898 struct bpf_map *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]; 899 char name[BPF_OBJ_NAME_LEN]; 900#ifdef CONFIG_SECURITY 901 void *security; 902#endif 903 struct bpf_prog_offload *offload; 904 struct btf *btf; 905 struct bpf_func_info *func_info; 906 struct bpf_func_info_aux *func_info_aux; 907 /* bpf_line_info loaded from userspace. linfo->insn_off 908 * has the xlated insn offset. 909 * Both the main and sub prog share the same linfo. 910 * The subprog can access its first linfo by 911 * using the linfo_idx. 912 */ 913 struct bpf_line_info *linfo; 914 /* jited_linfo is the jited addr of the linfo. It has a 915 * one to one mapping to linfo: 916 * jited_linfo[i] is the jited addr for the linfo[i]->insn_off. 917 * Both the main and sub prog share the same jited_linfo. 918 * The subprog can access its first jited_linfo by 919 * using the linfo_idx. 920 */ 921 void **jited_linfo; 922 u32 func_info_cnt; 923 u32 nr_linfo; 924 /* subprog can use linfo_idx to access its first linfo and 925 * jited_linfo. 926 * main prog always has linfo_idx == 0 927 */ 928 u32 linfo_idx; 929 u32 num_exentries; 930 struct exception_table_entry *extable; 931 union { 932 struct work_struct work; 933 struct rcu_head rcu; 934 }; 935}; 936 937struct bpf_array_aux { 938 /* 'Ownership' of prog array is claimed by the first program that 939 * is going to use this map or by the first program which FD is 940 * stored in the map to make sure that all callers and callees have 941 * the same prog type and JITed flag. 942 */ 943 struct { 944 spinlock_t lock; 945 enum bpf_prog_type type; 946 bool jited; 947 } owner; 948 /* Programs with direct jumps into programs part of this array. */ 949 struct list_head poke_progs; 950 struct bpf_map *map; 951 struct mutex poke_mutex; 952 struct work_struct work; 953}; 954 955struct bpf_link { 956 atomic64_t refcnt; 957 u32 id; 958 enum bpf_link_type type; 959 const struct bpf_link_ops *ops; 960 struct bpf_prog *prog; 961 struct work_struct work; 962}; 963 964struct bpf_link_ops { 965 void (*release)(struct bpf_link *link); 966 void (*dealloc)(struct bpf_link *link); 967 int (*detach)(struct bpf_link *link); 968 int (*update_prog)(struct bpf_link *link, struct bpf_prog *new_prog, 969 struct bpf_prog *old_prog); 970 void (*show_fdinfo)(const struct bpf_link *link, struct seq_file *seq); 971 int (*fill_link_info)(const struct bpf_link *link, 972 struct bpf_link_info *info); 973}; 974 975struct bpf_link_primer { 976 struct bpf_link *link; 977 struct file *file; 978 int fd; 979 u32 id; 980}; 981 982struct bpf_struct_ops_value; 983struct btf_member; 984 985#define BPF_STRUCT_OPS_MAX_NR_MEMBERS 64 986struct bpf_struct_ops { 987 const struct bpf_verifier_ops *verifier_ops; 988 int (*init)(struct btf *btf); 989 int (*check_member)(const struct btf_type *t, 990 const struct btf_member *member); 991 int (*init_member)(const struct btf_type *t, 992 const struct btf_member *member, 993 void *kdata, const void *udata); 994 int (*reg)(void *kdata); 995 void (*unreg)(void *kdata); 996 const struct btf_type *type; 997 const struct btf_type *value_type; 998 const char *name; 999 struct btf_func_model func_models[BPF_STRUCT_OPS_MAX_NR_MEMBERS]; 1000 u32 type_id; 1001 u32 value_id; 1002}; 1003 1004#if defined(CONFIG_BPF_JIT) && defined(CONFIG_BPF_SYSCALL) 1005#define BPF_MODULE_OWNER ((void *)((0xeB9FUL << 2) + POISON_POINTER_DELTA)) 1006const struct bpf_struct_ops *bpf_struct_ops_find(u32 type_id); 1007void bpf_struct_ops_init(struct btf *btf, struct bpf_verifier_log *log); 1008bool bpf_struct_ops_get(const void *kdata); 1009void bpf_struct_ops_put(const void *kdata); 1010int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, void *key, 1011 void *value); 1012int bpf_struct_ops_prepare_trampoline(struct bpf_tramp_progs *tprogs, 1013 struct bpf_prog *prog, 1014 const struct btf_func_model *model, 1015 void *image, void *image_end); 1016static inline bool bpf_try_module_get(const void *data, struct module *owner) 1017{ 1018 if (owner == BPF_MODULE_OWNER) 1019 return bpf_struct_ops_get(data); 1020 else 1021 return try_module_get(owner); 1022} 1023static inline void bpf_module_put(const void *data, struct module *owner) 1024{ 1025 if (owner == BPF_MODULE_OWNER) 1026 bpf_struct_ops_put(data); 1027 else 1028 module_put(owner); 1029} 1030 1031#ifdef CONFIG_NET 1032/* Define it here to avoid the use of forward declaration */ 1033struct bpf_dummy_ops_state { 1034 int val; 1035}; 1036 1037struct bpf_dummy_ops { 1038 int (*test_1)(struct bpf_dummy_ops_state *cb); 1039 int (*test_2)(struct bpf_dummy_ops_state *cb, int a1, unsigned short a2, 1040 char a3, unsigned long a4); 1041}; 1042 1043int bpf_struct_ops_test_run(struct bpf_prog *prog, const union bpf_attr *kattr, 1044 union bpf_attr __user *uattr); 1045#endif 1046#else 1047static inline const struct bpf_struct_ops *bpf_struct_ops_find(u32 type_id) 1048{ 1049 return NULL; 1050} 1051static inline void bpf_struct_ops_init(struct btf *btf, 1052 struct bpf_verifier_log *log) 1053{ 1054} 1055static inline bool bpf_try_module_get(const void *data, struct module *owner) 1056{ 1057 return try_module_get(owner); 1058} 1059static inline void bpf_module_put(const void *data, struct module *owner) 1060{ 1061 module_put(owner); 1062} 1063static inline int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, 1064 void *key, 1065 void *value) 1066{ 1067 return -EINVAL; 1068} 1069#endif 1070 1071struct bpf_array { 1072 struct bpf_map map; 1073 u32 elem_size; 1074 u32 index_mask; 1075 struct bpf_array_aux *aux; 1076 union { 1077 char value[0] __aligned(8); 1078 void *ptrs[0] __aligned(8); 1079 void __percpu *pptrs[0] __aligned(8); 1080 }; 1081}; 1082 1083#define BPF_COMPLEXITY_LIMIT_INSNS 1000000 /* yes. 1M insns */ 1084#define MAX_TAIL_CALL_CNT 32 1085 1086#define BPF_F_ACCESS_MASK (BPF_F_RDONLY | \ 1087 BPF_F_RDONLY_PROG | \ 1088 BPF_F_WRONLY | \ 1089 BPF_F_WRONLY_PROG) 1090 1091#define BPF_MAP_CAN_READ BIT(0) 1092#define BPF_MAP_CAN_WRITE BIT(1) 1093 1094static inline u32 bpf_map_flags_to_cap(struct bpf_map *map) 1095{ 1096 u32 access_flags = map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG); 1097 1098 /* Combination of BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG is 1099 * not possible. 1100 */ 1101 if (access_flags & BPF_F_RDONLY_PROG) 1102 return BPF_MAP_CAN_READ; 1103 else if (access_flags & BPF_F_WRONLY_PROG) 1104 return BPF_MAP_CAN_WRITE; 1105 else 1106 return BPF_MAP_CAN_READ | BPF_MAP_CAN_WRITE; 1107} 1108 1109static inline bool bpf_map_flags_access_ok(u32 access_flags) 1110{ 1111 return (access_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) != 1112 (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG); 1113} 1114 1115struct bpf_event_entry { 1116 struct perf_event *event; 1117 struct file *perf_file; 1118 struct file *map_file; 1119 struct rcu_head rcu; 1120}; 1121 1122bool bpf_prog_array_compatible(struct bpf_array *array, const struct bpf_prog *fp); 1123int bpf_prog_calc_tag(struct bpf_prog *fp); 1124 1125const struct bpf_func_proto *bpf_get_trace_printk_proto(void); 1126const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void); 1127 1128typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src, 1129 unsigned long off, unsigned long len); 1130typedef u32 (*bpf_convert_ctx_access_t)(enum bpf_access_type type, 1131 const struct bpf_insn *src, 1132 struct bpf_insn *dst, 1133 struct bpf_prog *prog, 1134 u32 *target_size); 1135 1136u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 1137 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy); 1138 1139/* an array of programs to be executed under rcu_lock. 1140 * 1141 * Typical usage: 1142 * ret = BPF_PROG_RUN_ARRAY(&bpf_prog_array, ctx, bpf_prog_run); 1143 * 1144 * the structure returned by bpf_prog_array_alloc() should be populated 1145 * with program pointers and the last pointer must be NULL. 1146 * The user has to keep refcnt on the program and make sure the program 1147 * is removed from the array before bpf_prog_put(). 1148 * The 'struct bpf_prog_array *' should only be replaced with xchg() 1149 * since other cpus are walking the array of pointers in parallel. 1150 */ 1151struct bpf_prog_array_item { 1152 struct bpf_prog *prog; 1153 union { 1154 struct bpf_cgroup_storage *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]; 1155 u64 bpf_cookie; 1156 }; 1157}; 1158 1159struct bpf_prog_array { 1160 struct rcu_head rcu; 1161 struct bpf_prog_array_item items[]; 1162}; 1163 1164struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags); 1165void bpf_prog_array_free(struct bpf_prog_array *progs); 1166int bpf_prog_array_length(struct bpf_prog_array *progs); 1167bool bpf_prog_array_is_empty(struct bpf_prog_array *array); 1168int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs, 1169 __u32 __user *prog_ids, u32 cnt); 1170 1171void bpf_prog_array_delete_safe(struct bpf_prog_array *progs, 1172 struct bpf_prog *old_prog); 1173int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index); 1174int bpf_prog_array_update_at(struct bpf_prog_array *array, int index, 1175 struct bpf_prog *prog); 1176int bpf_prog_array_copy_info(struct bpf_prog_array *array, 1177 u32 *prog_ids, u32 request_cnt, 1178 u32 *prog_cnt); 1179int bpf_prog_array_copy(struct bpf_prog_array *old_array, 1180 struct bpf_prog *exclude_prog, 1181 struct bpf_prog *include_prog, 1182 u64 bpf_cookie, 1183 struct bpf_prog_array **new_array); 1184 1185struct bpf_run_ctx {}; 1186 1187struct bpf_cg_run_ctx { 1188 struct bpf_run_ctx run_ctx; 1189 const struct bpf_prog_array_item *prog_item; 1190}; 1191 1192struct bpf_trace_run_ctx { 1193 struct bpf_run_ctx run_ctx; 1194 u64 bpf_cookie; 1195}; 1196 1197static inline struct bpf_run_ctx *bpf_set_run_ctx(struct bpf_run_ctx *new_ctx) 1198{ 1199 struct bpf_run_ctx *old_ctx = NULL; 1200 1201#ifdef CONFIG_BPF_SYSCALL 1202 old_ctx = current->bpf_ctx; 1203 current->bpf_ctx = new_ctx; 1204#endif 1205 return old_ctx; 1206} 1207 1208static inline void bpf_reset_run_ctx(struct bpf_run_ctx *old_ctx) 1209{ 1210#ifdef CONFIG_BPF_SYSCALL 1211 current->bpf_ctx = old_ctx; 1212#endif 1213} 1214 1215/* BPF program asks to bypass CAP_NET_BIND_SERVICE in bind. */ 1216#define BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE (1 << 0) 1217/* BPF program asks to set CN on the packet. */ 1218#define BPF_RET_SET_CN (1 << 0) 1219 1220typedef u32 (*bpf_prog_run_fn)(const struct bpf_prog *prog, const void *ctx); 1221 1222static __always_inline u32 1223BPF_PROG_RUN_ARRAY_CG_FLAGS(const struct bpf_prog_array __rcu *array_rcu, 1224 const void *ctx, bpf_prog_run_fn run_prog, 1225 u32 *ret_flags) 1226{ 1227 const struct bpf_prog_array_item *item; 1228 const struct bpf_prog *prog; 1229 const struct bpf_prog_array *array; 1230 struct bpf_run_ctx *old_run_ctx; 1231 struct bpf_cg_run_ctx run_ctx; 1232 u32 ret = 1; 1233 u32 func_ret; 1234 1235 migrate_disable(); 1236 rcu_read_lock(); 1237 array = rcu_dereference(array_rcu); 1238 item = &array->items[0]; 1239 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); 1240 while ((prog = READ_ONCE(item->prog))) { 1241 run_ctx.prog_item = item; 1242 func_ret = run_prog(prog, ctx); 1243 ret &= (func_ret & 1); 1244 *(ret_flags) |= (func_ret >> 1); 1245 item++; 1246 } 1247 bpf_reset_run_ctx(old_run_ctx); 1248 rcu_read_unlock(); 1249 migrate_enable(); 1250 return ret; 1251} 1252 1253static __always_inline u32 1254BPF_PROG_RUN_ARRAY_CG(const struct bpf_prog_array __rcu *array_rcu, 1255 const void *ctx, bpf_prog_run_fn run_prog) 1256{ 1257 const struct bpf_prog_array_item *item; 1258 const struct bpf_prog *prog; 1259 const struct bpf_prog_array *array; 1260 struct bpf_run_ctx *old_run_ctx; 1261 struct bpf_cg_run_ctx run_ctx; 1262 u32 ret = 1; 1263 1264 migrate_disable(); 1265 rcu_read_lock(); 1266 array = rcu_dereference(array_rcu); 1267 item = &array->items[0]; 1268 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); 1269 while ((prog = READ_ONCE(item->prog))) { 1270 run_ctx.prog_item = item; 1271 ret &= run_prog(prog, ctx); 1272 item++; 1273 } 1274 bpf_reset_run_ctx(old_run_ctx); 1275 rcu_read_unlock(); 1276 migrate_enable(); 1277 return ret; 1278} 1279 1280static __always_inline u32 1281BPF_PROG_RUN_ARRAY(const struct bpf_prog_array __rcu *array_rcu, 1282 const void *ctx, bpf_prog_run_fn run_prog) 1283{ 1284 const struct bpf_prog_array_item *item; 1285 const struct bpf_prog *prog; 1286 const struct bpf_prog_array *array; 1287 struct bpf_run_ctx *old_run_ctx; 1288 struct bpf_trace_run_ctx run_ctx; 1289 u32 ret = 1; 1290 1291 migrate_disable(); 1292 rcu_read_lock(); 1293 array = rcu_dereference(array_rcu); 1294 if (unlikely(!array)) 1295 goto out; 1296 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); 1297 item = &array->items[0]; 1298 while ((prog = READ_ONCE(item->prog))) { 1299 run_ctx.bpf_cookie = item->bpf_cookie; 1300 ret &= run_prog(prog, ctx); 1301 item++; 1302 } 1303 bpf_reset_run_ctx(old_run_ctx); 1304out: 1305 rcu_read_unlock(); 1306 migrate_enable(); 1307 return ret; 1308} 1309 1310/* To be used by __cgroup_bpf_run_filter_skb for EGRESS BPF progs 1311 * so BPF programs can request cwr for TCP packets. 1312 * 1313 * Current cgroup skb programs can only return 0 or 1 (0 to drop the 1314 * packet. This macro changes the behavior so the low order bit 1315 * indicates whether the packet should be dropped (0) or not (1) 1316 * and the next bit is a congestion notification bit. This could be 1317 * used by TCP to call tcp_enter_cwr() 1318 * 1319 * Hence, new allowed return values of CGROUP EGRESS BPF programs are: 1320 * 0: drop packet 1321 * 1: keep packet 1322 * 2: drop packet and cn 1323 * 3: keep packet and cn 1324 * 1325 * This macro then converts it to one of the NET_XMIT or an error 1326 * code that is then interpreted as drop packet (and no cn): 1327 * 0: NET_XMIT_SUCCESS skb should be transmitted 1328 * 1: NET_XMIT_DROP skb should be dropped and cn 1329 * 2: NET_XMIT_CN skb should be transmitted and cn 1330 * 3: -EPERM skb should be dropped 1331 */ 1332#define BPF_PROG_CGROUP_INET_EGRESS_RUN_ARRAY(array, ctx, func) \ 1333 ({ \ 1334 u32 _flags = 0; \ 1335 bool _cn; \ 1336 u32 _ret; \ 1337 _ret = BPF_PROG_RUN_ARRAY_CG_FLAGS(array, ctx, func, &_flags); \ 1338 _cn = _flags & BPF_RET_SET_CN; \ 1339 if (_ret) \ 1340 _ret = (_cn ? NET_XMIT_CN : NET_XMIT_SUCCESS); \ 1341 else \ 1342 _ret = (_cn ? NET_XMIT_DROP : -EPERM); \ 1343 _ret; \ 1344 }) 1345 1346#ifdef CONFIG_BPF_SYSCALL 1347DECLARE_PER_CPU(int, bpf_prog_active); 1348extern struct mutex bpf_stats_enabled_mutex; 1349 1350/* 1351 * Block execution of BPF programs attached to instrumentation (perf, 1352 * kprobes, tracepoints) to prevent deadlocks on map operations as any of 1353 * these events can happen inside a region which holds a map bucket lock 1354 * and can deadlock on it. 1355 * 1356 * Use the preemption safe inc/dec variants on RT because migrate disable 1357 * is preemptible on RT and preemption in the middle of the RMW operation 1358 * might lead to inconsistent state. Use the raw variants for non RT 1359 * kernels as migrate_disable() maps to preempt_disable() so the slightly 1360 * more expensive save operation can be avoided. 1361 */ 1362static inline void bpf_disable_instrumentation(void) 1363{ 1364 migrate_disable(); 1365 if (IS_ENABLED(CONFIG_PREEMPT_RT)) 1366 this_cpu_inc(bpf_prog_active); 1367 else 1368 __this_cpu_inc(bpf_prog_active); 1369} 1370 1371static inline void bpf_enable_instrumentation(void) 1372{ 1373 if (IS_ENABLED(CONFIG_PREEMPT_RT)) 1374 this_cpu_dec(bpf_prog_active); 1375 else 1376 __this_cpu_dec(bpf_prog_active); 1377 migrate_enable(); 1378} 1379 1380extern const struct file_operations bpf_map_fops; 1381extern const struct file_operations bpf_prog_fops; 1382extern const struct file_operations bpf_iter_fops; 1383 1384#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ 1385 extern const struct bpf_prog_ops _name ## _prog_ops; \ 1386 extern const struct bpf_verifier_ops _name ## _verifier_ops; 1387#define BPF_MAP_TYPE(_id, _ops) \ 1388 extern const struct bpf_map_ops _ops; 1389#define BPF_LINK_TYPE(_id, _name) 1390#include <linux/bpf_types.h> 1391#undef BPF_PROG_TYPE 1392#undef BPF_MAP_TYPE 1393#undef BPF_LINK_TYPE 1394 1395extern const struct bpf_prog_ops bpf_offload_prog_ops; 1396extern const struct bpf_verifier_ops tc_cls_act_analyzer_ops; 1397extern const struct bpf_verifier_ops xdp_analyzer_ops; 1398 1399struct bpf_prog *bpf_prog_get(u32 ufd); 1400struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 1401 bool attach_drv); 1402void bpf_prog_add(struct bpf_prog *prog, int i); 1403void bpf_prog_sub(struct bpf_prog *prog, int i); 1404void bpf_prog_inc(struct bpf_prog *prog); 1405struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog); 1406void bpf_prog_put(struct bpf_prog *prog); 1407 1408void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock); 1409void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock); 1410 1411struct bpf_map *bpf_map_get(u32 ufd); 1412struct bpf_map *bpf_map_get_with_uref(u32 ufd); 1413struct bpf_map *__bpf_map_get(struct fd f); 1414void bpf_map_inc(struct bpf_map *map); 1415void bpf_map_inc_with_uref(struct bpf_map *map); 1416struct bpf_map * __must_check bpf_map_inc_not_zero(struct bpf_map *map); 1417void bpf_map_put_with_uref(struct bpf_map *map); 1418void bpf_map_put(struct bpf_map *map); 1419void *bpf_map_area_alloc(u64 size, int numa_node); 1420void *bpf_map_area_mmapable_alloc(u64 size, int numa_node); 1421void bpf_map_area_free(void *base); 1422bool bpf_map_write_active(const struct bpf_map *map); 1423void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr); 1424int generic_map_lookup_batch(struct bpf_map *map, 1425 const union bpf_attr *attr, 1426 union bpf_attr __user *uattr); 1427int generic_map_update_batch(struct bpf_map *map, 1428 const union bpf_attr *attr, 1429 union bpf_attr __user *uattr); 1430int generic_map_delete_batch(struct bpf_map *map, 1431 const union bpf_attr *attr, 1432 union bpf_attr __user *uattr); 1433struct bpf_map *bpf_map_get_curr_or_next(u32 *id); 1434struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id); 1435 1436#ifdef CONFIG_MEMCG_KMEM 1437void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags, 1438 int node); 1439void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags); 1440void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size, 1441 size_t align, gfp_t flags); 1442#else 1443static inline void * 1444bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags, 1445 int node) 1446{ 1447 return kmalloc_node(size, flags, node); 1448} 1449 1450static inline void * 1451bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags) 1452{ 1453 return kzalloc(size, flags); 1454} 1455 1456static inline void __percpu * 1457bpf_map_alloc_percpu(const struct bpf_map *map, size_t size, size_t align, 1458 gfp_t flags) 1459{ 1460 return __alloc_percpu_gfp(size, align, flags); 1461} 1462#endif 1463 1464extern int sysctl_unprivileged_bpf_disabled; 1465 1466static inline bool bpf_allow_ptr_leaks(void) 1467{ 1468 return perfmon_capable(); 1469} 1470 1471static inline bool bpf_allow_uninit_stack(void) 1472{ 1473 return perfmon_capable(); 1474} 1475 1476static inline bool bpf_allow_ptr_to_map_access(void) 1477{ 1478 return perfmon_capable(); 1479} 1480 1481static inline bool bpf_bypass_spec_v1(void) 1482{ 1483 return perfmon_capable(); 1484} 1485 1486static inline bool bpf_bypass_spec_v4(void) 1487{ 1488 return perfmon_capable(); 1489} 1490 1491int bpf_map_new_fd(struct bpf_map *map, int flags); 1492int bpf_prog_new_fd(struct bpf_prog *prog); 1493 1494void bpf_link_init(struct bpf_link *link, enum bpf_link_type type, 1495 const struct bpf_link_ops *ops, struct bpf_prog *prog); 1496int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer); 1497int bpf_link_settle(struct bpf_link_primer *primer); 1498void bpf_link_cleanup(struct bpf_link_primer *primer); 1499void bpf_link_inc(struct bpf_link *link); 1500void bpf_link_put(struct bpf_link *link); 1501int bpf_link_new_fd(struct bpf_link *link); 1502struct file *bpf_link_new_file(struct bpf_link *link, int *reserved_fd); 1503struct bpf_link *bpf_link_get_from_fd(u32 ufd); 1504 1505int bpf_obj_pin_user(u32 ufd, const char __user *pathname); 1506int bpf_obj_get_user(const char __user *pathname, int flags); 1507 1508#define BPF_ITER_FUNC_PREFIX "bpf_iter_" 1509#define DEFINE_BPF_ITER_FUNC(target, args...) \ 1510 extern int bpf_iter_ ## target(args); \ 1511 int __init bpf_iter_ ## target(args) { return 0; } 1512 1513struct bpf_iter_aux_info { 1514 struct bpf_map *map; 1515}; 1516 1517typedef int (*bpf_iter_attach_target_t)(struct bpf_prog *prog, 1518 union bpf_iter_link_info *linfo, 1519 struct bpf_iter_aux_info *aux); 1520typedef void (*bpf_iter_detach_target_t)(struct bpf_iter_aux_info *aux); 1521typedef void (*bpf_iter_show_fdinfo_t) (const struct bpf_iter_aux_info *aux, 1522 struct seq_file *seq); 1523typedef int (*bpf_iter_fill_link_info_t)(const struct bpf_iter_aux_info *aux, 1524 struct bpf_link_info *info); 1525typedef const struct bpf_func_proto * 1526(*bpf_iter_get_func_proto_t)(enum bpf_func_id func_id, 1527 const struct bpf_prog *prog); 1528 1529enum bpf_iter_feature { 1530 BPF_ITER_RESCHED = BIT(0), 1531}; 1532 1533#define BPF_ITER_CTX_ARG_MAX 2 1534struct bpf_iter_reg { 1535 const char *target; 1536 bpf_iter_attach_target_t attach_target; 1537 bpf_iter_detach_target_t detach_target; 1538 bpf_iter_show_fdinfo_t show_fdinfo; 1539 bpf_iter_fill_link_info_t fill_link_info; 1540 bpf_iter_get_func_proto_t get_func_proto; 1541 u32 ctx_arg_info_size; 1542 u32 feature; 1543 struct bpf_ctx_arg_aux ctx_arg_info[BPF_ITER_CTX_ARG_MAX]; 1544 const struct bpf_iter_seq_info *seq_info; 1545}; 1546 1547struct bpf_iter_meta { 1548 __bpf_md_ptr(struct seq_file *, seq); 1549 u64 session_id; 1550 u64 seq_num; 1551}; 1552 1553struct bpf_iter__bpf_map_elem { 1554 __bpf_md_ptr(struct bpf_iter_meta *, meta); 1555 __bpf_md_ptr(struct bpf_map *, map); 1556 __bpf_md_ptr(void *, key); 1557 __bpf_md_ptr(void *, value); 1558}; 1559 1560int bpf_iter_reg_target(const struct bpf_iter_reg *reg_info); 1561void bpf_iter_unreg_target(const struct bpf_iter_reg *reg_info); 1562bool bpf_iter_prog_supported(struct bpf_prog *prog); 1563const struct bpf_func_proto * 1564bpf_iter_get_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog); 1565int bpf_iter_link_attach(const union bpf_attr *attr, bpfptr_t uattr, struct bpf_prog *prog); 1566int bpf_iter_new_fd(struct bpf_link *link); 1567bool bpf_link_is_iter(struct bpf_link *link); 1568struct bpf_prog *bpf_iter_get_info(struct bpf_iter_meta *meta, bool in_stop); 1569int bpf_iter_run_prog(struct bpf_prog *prog, void *ctx); 1570void bpf_iter_map_show_fdinfo(const struct bpf_iter_aux_info *aux, 1571 struct seq_file *seq); 1572int bpf_iter_map_fill_link_info(const struct bpf_iter_aux_info *aux, 1573 struct bpf_link_info *info); 1574 1575int map_set_for_each_callback_args(struct bpf_verifier_env *env, 1576 struct bpf_func_state *caller, 1577 struct bpf_func_state *callee); 1578 1579int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value); 1580int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value); 1581int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value, 1582 u64 flags); 1583int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value, 1584 u64 flags); 1585 1586int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value); 1587 1588int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file, 1589 void *key, void *value, u64 map_flags); 1590int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 1591int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file, 1592 void *key, void *value, u64 map_flags); 1593int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 1594 1595int bpf_get_file_flag(int flags); 1596int bpf_check_uarg_tail_zero(bpfptr_t uaddr, size_t expected_size, 1597 size_t actual_size); 1598 1599/* memcpy that is used with 8-byte aligned pointers, power-of-8 size and 1600 * forced to use 'long' read/writes to try to atomically copy long counters. 1601 * Best-effort only. No barriers here, since it _will_ race with concurrent 1602 * updates from BPF programs. Called from bpf syscall and mostly used with 1603 * size 8 or 16 bytes, so ask compiler to inline it. 1604 */ 1605static inline void bpf_long_memcpy(void *dst, const void *src, u32 size) 1606{ 1607 const long *lsrc = src; 1608 long *ldst = dst; 1609 1610 size /= sizeof(long); 1611 while (size--) 1612 *ldst++ = *lsrc++; 1613} 1614 1615/* verify correctness of eBPF program */ 1616int bpf_check(struct bpf_prog **fp, union bpf_attr *attr, bpfptr_t uattr); 1617 1618#ifndef CONFIG_BPF_JIT_ALWAYS_ON 1619void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth); 1620#endif 1621 1622struct btf *bpf_get_btf_vmlinux(void); 1623 1624/* Map specifics */ 1625struct xdp_buff; 1626struct sk_buff; 1627struct bpf_dtab_netdev; 1628struct bpf_cpu_map_entry; 1629 1630void __dev_flush(void); 1631int dev_xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp, 1632 struct net_device *dev_rx); 1633int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, 1634 struct net_device *dev_rx); 1635int dev_map_enqueue_multi(struct xdp_buff *xdp, struct net_device *dev_rx, 1636 struct bpf_map *map, bool exclude_ingress); 1637int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, struct sk_buff *skb, 1638 struct bpf_prog *xdp_prog); 1639int dev_map_redirect_multi(struct net_device *dev, struct sk_buff *skb, 1640 struct bpf_prog *xdp_prog, struct bpf_map *map, 1641 bool exclude_ingress); 1642 1643void __cpu_map_flush(void); 1644int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_buff *xdp, 1645 struct net_device *dev_rx); 1646int cpu_map_generic_redirect(struct bpf_cpu_map_entry *rcpu, 1647 struct sk_buff *skb); 1648 1649/* Return map's numa specified by userspace */ 1650static inline int bpf_map_attr_numa_node(const union bpf_attr *attr) 1651{ 1652 return (attr->map_flags & BPF_F_NUMA_NODE) ? 1653 attr->numa_node : NUMA_NO_NODE; 1654} 1655 1656struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type); 1657int array_map_alloc_check(union bpf_attr *attr); 1658 1659int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr, 1660 union bpf_attr __user *uattr); 1661int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr, 1662 union bpf_attr __user *uattr); 1663int bpf_prog_test_run_tracing(struct bpf_prog *prog, 1664 const union bpf_attr *kattr, 1665 union bpf_attr __user *uattr); 1666int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 1667 const union bpf_attr *kattr, 1668 union bpf_attr __user *uattr); 1669int bpf_prog_test_run_raw_tp(struct bpf_prog *prog, 1670 const union bpf_attr *kattr, 1671 union bpf_attr __user *uattr); 1672int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, 1673 const union bpf_attr *kattr, 1674 union bpf_attr __user *uattr); 1675bool bpf_prog_test_check_kfunc_call(u32 kfunc_id, struct module *owner); 1676bool btf_ctx_access(int off, int size, enum bpf_access_type type, 1677 const struct bpf_prog *prog, 1678 struct bpf_insn_access_aux *info); 1679 1680static inline bool bpf_tracing_ctx_access(int off, int size, 1681 enum bpf_access_type type) 1682{ 1683 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS) 1684 return false; 1685 if (type != BPF_READ) 1686 return false; 1687 if (off % size != 0) 1688 return false; 1689 return true; 1690} 1691 1692static inline bool bpf_tracing_btf_ctx_access(int off, int size, 1693 enum bpf_access_type type, 1694 const struct bpf_prog *prog, 1695 struct bpf_insn_access_aux *info) 1696{ 1697 if (!bpf_tracing_ctx_access(off, size, type)) 1698 return false; 1699 return btf_ctx_access(off, size, type, prog, info); 1700} 1701 1702int btf_struct_access(struct bpf_verifier_log *log, const struct btf *btf, 1703 const struct btf_type *t, int off, int size, 1704 enum bpf_access_type atype, 1705 u32 *next_btf_id); 1706bool btf_struct_ids_match(struct bpf_verifier_log *log, 1707 const struct btf *btf, u32 id, int off, 1708 const struct btf *need_btf, u32 need_type_id); 1709 1710int btf_distill_func_proto(struct bpf_verifier_log *log, 1711 struct btf *btf, 1712 const struct btf_type *func_proto, 1713 const char *func_name, 1714 struct btf_func_model *m); 1715 1716struct bpf_reg_state; 1717int btf_check_subprog_arg_match(struct bpf_verifier_env *env, int subprog, 1718 struct bpf_reg_state *regs); 1719int btf_check_kfunc_arg_match(struct bpf_verifier_env *env, 1720 const struct btf *btf, u32 func_id, 1721 struct bpf_reg_state *regs); 1722int btf_prepare_func_args(struct bpf_verifier_env *env, int subprog, 1723 struct bpf_reg_state *reg); 1724int btf_check_type_match(struct bpf_verifier_log *log, const struct bpf_prog *prog, 1725 struct btf *btf, const struct btf_type *t); 1726 1727struct bpf_prog *bpf_prog_by_id(u32 id); 1728struct bpf_link *bpf_link_by_id(u32 id); 1729 1730const struct bpf_func_proto *bpf_base_func_proto(enum bpf_func_id func_id); 1731void bpf_task_storage_free(struct task_struct *task); 1732bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog); 1733const struct btf_func_model * 1734bpf_jit_find_kfunc_model(const struct bpf_prog *prog, 1735 const struct bpf_insn *insn); 1736#else /* !CONFIG_BPF_SYSCALL */ 1737static inline struct bpf_prog *bpf_prog_get(u32 ufd) 1738{ 1739 return ERR_PTR(-EOPNOTSUPP); 1740} 1741 1742static inline struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, 1743 enum bpf_prog_type type, 1744 bool attach_drv) 1745{ 1746 return ERR_PTR(-EOPNOTSUPP); 1747} 1748 1749static inline void bpf_prog_add(struct bpf_prog *prog, int i) 1750{ 1751} 1752 1753static inline void bpf_prog_sub(struct bpf_prog *prog, int i) 1754{ 1755} 1756 1757static inline void bpf_prog_put(struct bpf_prog *prog) 1758{ 1759} 1760 1761static inline void bpf_prog_inc(struct bpf_prog *prog) 1762{ 1763} 1764 1765static inline struct bpf_prog *__must_check 1766bpf_prog_inc_not_zero(struct bpf_prog *prog) 1767{ 1768 return ERR_PTR(-EOPNOTSUPP); 1769} 1770 1771static inline void bpf_link_init(struct bpf_link *link, enum bpf_link_type type, 1772 const struct bpf_link_ops *ops, 1773 struct bpf_prog *prog) 1774{ 1775} 1776 1777static inline int bpf_link_prime(struct bpf_link *link, 1778 struct bpf_link_primer *primer) 1779{ 1780 return -EOPNOTSUPP; 1781} 1782 1783static inline int bpf_link_settle(struct bpf_link_primer *primer) 1784{ 1785 return -EOPNOTSUPP; 1786} 1787 1788static inline void bpf_link_cleanup(struct bpf_link_primer *primer) 1789{ 1790} 1791 1792static inline void bpf_link_inc(struct bpf_link *link) 1793{ 1794} 1795 1796static inline void bpf_link_put(struct bpf_link *link) 1797{ 1798} 1799 1800static inline int bpf_obj_get_user(const char __user *pathname, int flags) 1801{ 1802 return -EOPNOTSUPP; 1803} 1804 1805static inline bool dev_map_can_have_prog(struct bpf_map *map) 1806{ 1807 return false; 1808} 1809 1810static inline void __dev_flush(void) 1811{ 1812} 1813 1814struct xdp_buff; 1815struct bpf_dtab_netdev; 1816struct bpf_cpu_map_entry; 1817 1818static inline 1819int dev_xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp, 1820 struct net_device *dev_rx) 1821{ 1822 return 0; 1823} 1824 1825static inline 1826int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, 1827 struct net_device *dev_rx) 1828{ 1829 return 0; 1830} 1831 1832static inline 1833int dev_map_enqueue_multi(struct xdp_buff *xdp, struct net_device *dev_rx, 1834 struct bpf_map *map, bool exclude_ingress) 1835{ 1836 return 0; 1837} 1838 1839struct sk_buff; 1840 1841static inline int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, 1842 struct sk_buff *skb, 1843 struct bpf_prog *xdp_prog) 1844{ 1845 return 0; 1846} 1847 1848static inline 1849int dev_map_redirect_multi(struct net_device *dev, struct sk_buff *skb, 1850 struct bpf_prog *xdp_prog, struct bpf_map *map, 1851 bool exclude_ingress) 1852{ 1853 return 0; 1854} 1855 1856static inline void __cpu_map_flush(void) 1857{ 1858} 1859 1860static inline int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, 1861 struct xdp_buff *xdp, 1862 struct net_device *dev_rx) 1863{ 1864 return 0; 1865} 1866 1867static inline int cpu_map_generic_redirect(struct bpf_cpu_map_entry *rcpu, 1868 struct sk_buff *skb) 1869{ 1870 return -EOPNOTSUPP; 1871} 1872 1873static inline bool cpu_map_prog_allowed(struct bpf_map *map) 1874{ 1875 return false; 1876} 1877 1878static inline struct bpf_prog *bpf_prog_get_type_path(const char *name, 1879 enum bpf_prog_type type) 1880{ 1881 return ERR_PTR(-EOPNOTSUPP); 1882} 1883 1884static inline int bpf_prog_test_run_xdp(struct bpf_prog *prog, 1885 const union bpf_attr *kattr, 1886 union bpf_attr __user *uattr) 1887{ 1888 return -ENOTSUPP; 1889} 1890 1891static inline int bpf_prog_test_run_skb(struct bpf_prog *prog, 1892 const union bpf_attr *kattr, 1893 union bpf_attr __user *uattr) 1894{ 1895 return -ENOTSUPP; 1896} 1897 1898static inline int bpf_prog_test_run_tracing(struct bpf_prog *prog, 1899 const union bpf_attr *kattr, 1900 union bpf_attr __user *uattr) 1901{ 1902 return -ENOTSUPP; 1903} 1904 1905static inline int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 1906 const union bpf_attr *kattr, 1907 union bpf_attr __user *uattr) 1908{ 1909 return -ENOTSUPP; 1910} 1911 1912static inline int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, 1913 const union bpf_attr *kattr, 1914 union bpf_attr __user *uattr) 1915{ 1916 return -ENOTSUPP; 1917} 1918 1919static inline bool bpf_prog_test_check_kfunc_call(u32 kfunc_id, 1920 struct module *owner) 1921{ 1922 return false; 1923} 1924 1925static inline void bpf_map_put(struct bpf_map *map) 1926{ 1927} 1928 1929static inline struct bpf_prog *bpf_prog_by_id(u32 id) 1930{ 1931 return ERR_PTR(-ENOTSUPP); 1932} 1933 1934static inline const struct bpf_func_proto * 1935bpf_base_func_proto(enum bpf_func_id func_id) 1936{ 1937 return NULL; 1938} 1939 1940static inline void bpf_task_storage_free(struct task_struct *task) 1941{ 1942} 1943 1944static inline bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog) 1945{ 1946 return false; 1947} 1948 1949static inline const struct btf_func_model * 1950bpf_jit_find_kfunc_model(const struct bpf_prog *prog, 1951 const struct bpf_insn *insn) 1952{ 1953 return NULL; 1954} 1955#endif /* CONFIG_BPF_SYSCALL */ 1956 1957void __bpf_free_used_btfs(struct bpf_prog_aux *aux, 1958 struct btf_mod_pair *used_btfs, u32 len); 1959 1960static inline struct bpf_prog *bpf_prog_get_type(u32 ufd, 1961 enum bpf_prog_type type) 1962{ 1963 return bpf_prog_get_type_dev(ufd, type, false); 1964} 1965 1966void __bpf_free_used_maps(struct bpf_prog_aux *aux, 1967 struct bpf_map **used_maps, u32 len); 1968 1969bool bpf_prog_get_ok(struct bpf_prog *, enum bpf_prog_type *, bool); 1970 1971int bpf_prog_offload_compile(struct bpf_prog *prog); 1972void bpf_prog_offload_destroy(struct bpf_prog *prog); 1973int bpf_prog_offload_info_fill(struct bpf_prog_info *info, 1974 struct bpf_prog *prog); 1975 1976int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map); 1977 1978int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value); 1979int bpf_map_offload_update_elem(struct bpf_map *map, 1980 void *key, void *value, u64 flags); 1981int bpf_map_offload_delete_elem(struct bpf_map *map, void *key); 1982int bpf_map_offload_get_next_key(struct bpf_map *map, 1983 void *key, void *next_key); 1984 1985bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map); 1986 1987struct bpf_offload_dev * 1988bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv); 1989void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev); 1990void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev); 1991int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev, 1992 struct net_device *netdev); 1993void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev, 1994 struct net_device *netdev); 1995bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev); 1996 1997#if defined(CONFIG_NET) && defined(CONFIG_BPF_SYSCALL) 1998int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr); 1999 2000static inline bool bpf_prog_is_dev_bound(const struct bpf_prog_aux *aux) 2001{ 2002 return aux->offload_requested; 2003} 2004 2005static inline bool bpf_map_is_dev_bound(struct bpf_map *map) 2006{ 2007 return unlikely(map->ops == &bpf_map_offload_ops); 2008} 2009 2010struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr); 2011void bpf_map_offload_map_free(struct bpf_map *map); 2012int bpf_prog_test_run_syscall(struct bpf_prog *prog, 2013 const union bpf_attr *kattr, 2014 union bpf_attr __user *uattr); 2015 2016int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog); 2017int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype); 2018int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value, u64 flags); 2019void sock_map_unhash(struct sock *sk); 2020void sock_map_close(struct sock *sk, long timeout); 2021#else 2022static inline int bpf_prog_offload_init(struct bpf_prog *prog, 2023 union bpf_attr *attr) 2024{ 2025 return -EOPNOTSUPP; 2026} 2027 2028static inline bool bpf_prog_is_dev_bound(struct bpf_prog_aux *aux) 2029{ 2030 return false; 2031} 2032 2033static inline bool bpf_map_is_dev_bound(struct bpf_map *map) 2034{ 2035 return false; 2036} 2037 2038static inline struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr) 2039{ 2040 return ERR_PTR(-EOPNOTSUPP); 2041} 2042 2043static inline void bpf_map_offload_map_free(struct bpf_map *map) 2044{ 2045} 2046 2047static inline int bpf_prog_test_run_syscall(struct bpf_prog *prog, 2048 const union bpf_attr *kattr, 2049 union bpf_attr __user *uattr) 2050{ 2051 return -ENOTSUPP; 2052} 2053 2054#ifdef CONFIG_BPF_SYSCALL 2055static inline int sock_map_get_from_fd(const union bpf_attr *attr, 2056 struct bpf_prog *prog) 2057{ 2058 return -EINVAL; 2059} 2060 2061static inline int sock_map_prog_detach(const union bpf_attr *attr, 2062 enum bpf_prog_type ptype) 2063{ 2064 return -EOPNOTSUPP; 2065} 2066 2067static inline int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value, 2068 u64 flags) 2069{ 2070 return -EOPNOTSUPP; 2071} 2072#endif /* CONFIG_BPF_SYSCALL */ 2073#endif /* CONFIG_NET && CONFIG_BPF_SYSCALL */ 2074 2075#if defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) 2076void bpf_sk_reuseport_detach(struct sock *sk); 2077int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, void *key, 2078 void *value); 2079int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, void *key, 2080 void *value, u64 map_flags); 2081#else 2082static inline void bpf_sk_reuseport_detach(struct sock *sk) 2083{ 2084} 2085 2086#ifdef CONFIG_BPF_SYSCALL 2087static inline int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, 2088 void *key, void *value) 2089{ 2090 return -EOPNOTSUPP; 2091} 2092 2093static inline int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, 2094 void *key, void *value, 2095 u64 map_flags) 2096{ 2097 return -EOPNOTSUPP; 2098} 2099#endif /* CONFIG_BPF_SYSCALL */ 2100#endif /* defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) */ 2101 2102/* verifier prototypes for helper functions called from eBPF programs */ 2103extern const struct bpf_func_proto bpf_map_lookup_elem_proto; 2104extern const struct bpf_func_proto bpf_map_update_elem_proto; 2105extern const struct bpf_func_proto bpf_map_delete_elem_proto; 2106extern const struct bpf_func_proto bpf_map_push_elem_proto; 2107extern const struct bpf_func_proto bpf_map_pop_elem_proto; 2108extern const struct bpf_func_proto bpf_map_peek_elem_proto; 2109 2110extern const struct bpf_func_proto bpf_get_prandom_u32_proto; 2111extern const struct bpf_func_proto bpf_get_smp_processor_id_proto; 2112extern const struct bpf_func_proto bpf_get_numa_node_id_proto; 2113extern const struct bpf_func_proto bpf_tail_call_proto; 2114extern const struct bpf_func_proto bpf_ktime_get_ns_proto; 2115extern const struct bpf_func_proto bpf_ktime_get_boot_ns_proto; 2116extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto; 2117extern const struct bpf_func_proto bpf_get_current_uid_gid_proto; 2118extern const struct bpf_func_proto bpf_get_current_comm_proto; 2119extern const struct bpf_func_proto bpf_get_stackid_proto; 2120extern const struct bpf_func_proto bpf_get_stack_proto; 2121extern const struct bpf_func_proto bpf_get_task_stack_proto; 2122extern const struct bpf_func_proto bpf_get_stackid_proto_pe; 2123extern const struct bpf_func_proto bpf_get_stack_proto_pe; 2124extern const struct bpf_func_proto bpf_sock_map_update_proto; 2125extern const struct bpf_func_proto bpf_sock_hash_update_proto; 2126extern const struct bpf_func_proto bpf_get_current_cgroup_id_proto; 2127extern const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto; 2128extern const struct bpf_func_proto bpf_msg_redirect_hash_proto; 2129extern const struct bpf_func_proto bpf_msg_redirect_map_proto; 2130extern const struct bpf_func_proto bpf_sk_redirect_hash_proto; 2131extern const struct bpf_func_proto bpf_sk_redirect_map_proto; 2132extern const struct bpf_func_proto bpf_spin_lock_proto; 2133extern const struct bpf_func_proto bpf_spin_unlock_proto; 2134extern const struct bpf_func_proto bpf_get_local_storage_proto; 2135extern const struct bpf_func_proto bpf_strtol_proto; 2136extern const struct bpf_func_proto bpf_strtoul_proto; 2137extern const struct bpf_func_proto bpf_tcp_sock_proto; 2138extern const struct bpf_func_proto bpf_jiffies64_proto; 2139extern const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto; 2140extern const struct bpf_func_proto bpf_event_output_data_proto; 2141extern const struct bpf_func_proto bpf_ringbuf_output_proto; 2142extern const struct bpf_func_proto bpf_ringbuf_reserve_proto; 2143extern const struct bpf_func_proto bpf_ringbuf_submit_proto; 2144extern const struct bpf_func_proto bpf_ringbuf_discard_proto; 2145extern const struct bpf_func_proto bpf_ringbuf_query_proto; 2146extern const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto; 2147extern const struct bpf_func_proto bpf_skc_to_tcp_sock_proto; 2148extern const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto; 2149extern const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto; 2150extern const struct bpf_func_proto bpf_skc_to_udp6_sock_proto; 2151extern const struct bpf_func_proto bpf_skc_to_unix_sock_proto; 2152extern const struct bpf_func_proto bpf_copy_from_user_proto; 2153extern const struct bpf_func_proto bpf_snprintf_btf_proto; 2154extern const struct bpf_func_proto bpf_snprintf_proto; 2155extern const struct bpf_func_proto bpf_per_cpu_ptr_proto; 2156extern const struct bpf_func_proto bpf_this_cpu_ptr_proto; 2157extern const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto; 2158extern const struct bpf_func_proto bpf_sock_from_file_proto; 2159extern const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto; 2160extern const struct bpf_func_proto bpf_task_storage_get_proto; 2161extern const struct bpf_func_proto bpf_task_storage_delete_proto; 2162extern const struct bpf_func_proto bpf_for_each_map_elem_proto; 2163extern const struct bpf_func_proto bpf_btf_find_by_name_kind_proto; 2164extern const struct bpf_func_proto bpf_sk_setsockopt_proto; 2165extern const struct bpf_func_proto bpf_sk_getsockopt_proto; 2166extern const struct bpf_func_proto bpf_kallsyms_lookup_name_proto; 2167 2168const struct bpf_func_proto *tracing_prog_func_proto( 2169 enum bpf_func_id func_id, const struct bpf_prog *prog); 2170 2171/* Shared helpers among cBPF and eBPF. */ 2172void bpf_user_rnd_init_once(void); 2173u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 2174u64 bpf_get_raw_cpu_id(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 2175 2176#if defined(CONFIG_NET) 2177bool bpf_sock_common_is_valid_access(int off, int size, 2178 enum bpf_access_type type, 2179 struct bpf_insn_access_aux *info); 2180bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 2181 struct bpf_insn_access_aux *info); 2182u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 2183 const struct bpf_insn *si, 2184 struct bpf_insn *insn_buf, 2185 struct bpf_prog *prog, 2186 u32 *target_size); 2187#else 2188static inline bool bpf_sock_common_is_valid_access(int off, int size, 2189 enum bpf_access_type type, 2190 struct bpf_insn_access_aux *info) 2191{ 2192 return false; 2193} 2194static inline bool bpf_sock_is_valid_access(int off, int size, 2195 enum bpf_access_type type, 2196 struct bpf_insn_access_aux *info) 2197{ 2198 return false; 2199} 2200static inline u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 2201 const struct bpf_insn *si, 2202 struct bpf_insn *insn_buf, 2203 struct bpf_prog *prog, 2204 u32 *target_size) 2205{ 2206 return 0; 2207} 2208#endif 2209 2210#ifdef CONFIG_INET 2211struct sk_reuseport_kern { 2212 struct sk_buff *skb; 2213 struct sock *sk; 2214 struct sock *selected_sk; 2215 struct sock *migrating_sk; 2216 void *data_end; 2217 u32 hash; 2218 u32 reuseport_id; 2219 bool bind_inany; 2220}; 2221bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 2222 struct bpf_insn_access_aux *info); 2223 2224u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 2225 const struct bpf_insn *si, 2226 struct bpf_insn *insn_buf, 2227 struct bpf_prog *prog, 2228 u32 *target_size); 2229 2230bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 2231 struct bpf_insn_access_aux *info); 2232 2233u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 2234 const struct bpf_insn *si, 2235 struct bpf_insn *insn_buf, 2236 struct bpf_prog *prog, 2237 u32 *target_size); 2238#else 2239static inline bool bpf_tcp_sock_is_valid_access(int off, int size, 2240 enum bpf_access_type type, 2241 struct bpf_insn_access_aux *info) 2242{ 2243 return false; 2244} 2245 2246static inline u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 2247 const struct bpf_insn *si, 2248 struct bpf_insn *insn_buf, 2249 struct bpf_prog *prog, 2250 u32 *target_size) 2251{ 2252 return 0; 2253} 2254static inline bool bpf_xdp_sock_is_valid_access(int off, int size, 2255 enum bpf_access_type type, 2256 struct bpf_insn_access_aux *info) 2257{ 2258 return false; 2259} 2260 2261static inline u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 2262 const struct bpf_insn *si, 2263 struct bpf_insn *insn_buf, 2264 struct bpf_prog *prog, 2265 u32 *target_size) 2266{ 2267 return 0; 2268} 2269#endif /* CONFIG_INET */ 2270 2271enum bpf_text_poke_type { 2272 BPF_MOD_CALL, 2273 BPF_MOD_JUMP, 2274}; 2275 2276int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, 2277 void *addr1, void *addr2); 2278 2279struct btf_id_set; 2280bool btf_id_set_contains(const struct btf_id_set *set, u32 id); 2281 2282#define MAX_BPRINTF_VARARGS 12 2283 2284int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args, 2285 u32 **bin_buf, u32 num_args); 2286void bpf_bprintf_cleanup(void); 2287 2288#endif /* _LINUX_BPF_H */