at v5.3 37 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/wait.h> 16#include <linux/u64_stats_sync.h> 17 18struct bpf_verifier_env; 19struct perf_event; 20struct bpf_prog; 21struct bpf_map; 22struct sock; 23struct seq_file; 24struct btf; 25struct btf_type; 26 27/* map is generic key/value storage optionally accesible by eBPF programs */ 28struct bpf_map_ops { 29 /* funcs callable from userspace (via syscall) */ 30 int (*map_alloc_check)(union bpf_attr *attr); 31 struct bpf_map *(*map_alloc)(union bpf_attr *attr); 32 void (*map_release)(struct bpf_map *map, struct file *map_file); 33 void (*map_free)(struct bpf_map *map); 34 int (*map_get_next_key)(struct bpf_map *map, void *key, void *next_key); 35 void (*map_release_uref)(struct bpf_map *map); 36 void *(*map_lookup_elem_sys_only)(struct bpf_map *map, void *key); 37 38 /* funcs callable from userspace and from eBPF programs */ 39 void *(*map_lookup_elem)(struct bpf_map *map, void *key); 40 int (*map_update_elem)(struct bpf_map *map, void *key, void *value, u64 flags); 41 int (*map_delete_elem)(struct bpf_map *map, void *key); 42 int (*map_push_elem)(struct bpf_map *map, void *value, u64 flags); 43 int (*map_pop_elem)(struct bpf_map *map, void *value); 44 int (*map_peek_elem)(struct bpf_map *map, void *value); 45 46 /* funcs called by prog_array and perf_event_array map */ 47 void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file, 48 int fd); 49 void (*map_fd_put_ptr)(void *ptr); 50 u32 (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf); 51 u32 (*map_fd_sys_lookup_elem)(void *ptr); 52 void (*map_seq_show_elem)(struct bpf_map *map, void *key, 53 struct seq_file *m); 54 int (*map_check_btf)(const struct bpf_map *map, 55 const struct btf *btf, 56 const struct btf_type *key_type, 57 const struct btf_type *value_type); 58 59 /* Direct value access helpers. */ 60 int (*map_direct_value_addr)(const struct bpf_map *map, 61 u64 *imm, u32 off); 62 int (*map_direct_value_meta)(const struct bpf_map *map, 63 u64 imm, u32 *off); 64}; 65 66struct bpf_map_memory { 67 u32 pages; 68 struct user_struct *user; 69}; 70 71struct bpf_map { 72 /* The first two cachelines with read-mostly members of which some 73 * are also accessed in fast-path (e.g. ops, max_entries). 74 */ 75 const struct bpf_map_ops *ops ____cacheline_aligned; 76 struct bpf_map *inner_map_meta; 77#ifdef CONFIG_SECURITY 78 void *security; 79#endif 80 enum bpf_map_type map_type; 81 u32 key_size; 82 u32 value_size; 83 u32 max_entries; 84 u32 map_flags; 85 int spin_lock_off; /* >=0 valid offset, <0 error */ 86 u32 id; 87 int numa_node; 88 u32 btf_key_type_id; 89 u32 btf_value_type_id; 90 struct btf *btf; 91 struct bpf_map_memory memory; 92 bool unpriv_array; 93 bool frozen; /* write-once */ 94 /* 48 bytes hole */ 95 96 /* The 3rd and 4th cacheline with misc members to avoid false sharing 97 * particularly with refcounting. 98 */ 99 atomic_t refcnt ____cacheline_aligned; 100 atomic_t usercnt; 101 struct work_struct work; 102 char name[BPF_OBJ_NAME_LEN]; 103}; 104 105static inline bool map_value_has_spin_lock(const struct bpf_map *map) 106{ 107 return map->spin_lock_off >= 0; 108} 109 110static inline void check_and_init_map_lock(struct bpf_map *map, void *dst) 111{ 112 if (likely(!map_value_has_spin_lock(map))) 113 return; 114 *(struct bpf_spin_lock *)(dst + map->spin_lock_off) = 115 (struct bpf_spin_lock){}; 116} 117 118/* copy everything but bpf_spin_lock */ 119static inline void copy_map_value(struct bpf_map *map, void *dst, void *src) 120{ 121 if (unlikely(map_value_has_spin_lock(map))) { 122 u32 off = map->spin_lock_off; 123 124 memcpy(dst, src, off); 125 memcpy(dst + off + sizeof(struct bpf_spin_lock), 126 src + off + sizeof(struct bpf_spin_lock), 127 map->value_size - off - sizeof(struct bpf_spin_lock)); 128 } else { 129 memcpy(dst, src, map->value_size); 130 } 131} 132void copy_map_value_locked(struct bpf_map *map, void *dst, void *src, 133 bool lock_src); 134 135struct bpf_offload_dev; 136struct bpf_offloaded_map; 137 138struct bpf_map_dev_ops { 139 int (*map_get_next_key)(struct bpf_offloaded_map *map, 140 void *key, void *next_key); 141 int (*map_lookup_elem)(struct bpf_offloaded_map *map, 142 void *key, void *value); 143 int (*map_update_elem)(struct bpf_offloaded_map *map, 144 void *key, void *value, u64 flags); 145 int (*map_delete_elem)(struct bpf_offloaded_map *map, void *key); 146}; 147 148struct bpf_offloaded_map { 149 struct bpf_map map; 150 struct net_device *netdev; 151 const struct bpf_map_dev_ops *dev_ops; 152 void *dev_priv; 153 struct list_head offloads; 154}; 155 156static inline struct bpf_offloaded_map *map_to_offmap(struct bpf_map *map) 157{ 158 return container_of(map, struct bpf_offloaded_map, map); 159} 160 161static inline bool bpf_map_offload_neutral(const struct bpf_map *map) 162{ 163 return map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; 164} 165 166static inline bool bpf_map_support_seq_show(const struct bpf_map *map) 167{ 168 return map->btf && map->ops->map_seq_show_elem; 169} 170 171int map_check_no_btf(const struct bpf_map *map, 172 const struct btf *btf, 173 const struct btf_type *key_type, 174 const struct btf_type *value_type); 175 176extern const struct bpf_map_ops bpf_map_offload_ops; 177 178/* function argument constraints */ 179enum bpf_arg_type { 180 ARG_DONTCARE = 0, /* unused argument in helper function */ 181 182 /* the following constraints used to prototype 183 * bpf_map_lookup/update/delete_elem() functions 184 */ 185 ARG_CONST_MAP_PTR, /* const argument used as pointer to bpf_map */ 186 ARG_PTR_TO_MAP_KEY, /* pointer to stack used as map key */ 187 ARG_PTR_TO_MAP_VALUE, /* pointer to stack used as map value */ 188 ARG_PTR_TO_UNINIT_MAP_VALUE, /* pointer to valid memory used to store a map value */ 189 ARG_PTR_TO_MAP_VALUE_OR_NULL, /* pointer to stack used as map value or NULL */ 190 191 /* the following constraints used to prototype bpf_memcmp() and other 192 * functions that access data on eBPF program stack 193 */ 194 ARG_PTR_TO_MEM, /* pointer to valid memory (stack, packet, map value) */ 195 ARG_PTR_TO_MEM_OR_NULL, /* pointer to valid memory or NULL */ 196 ARG_PTR_TO_UNINIT_MEM, /* pointer to memory does not need to be initialized, 197 * helper function must fill all bytes or clear 198 * them in error case. 199 */ 200 201 ARG_CONST_SIZE, /* number of bytes accessed from memory */ 202 ARG_CONST_SIZE_OR_ZERO, /* number of bytes accessed from memory or 0 */ 203 204 ARG_PTR_TO_CTX, /* pointer to context */ 205 ARG_ANYTHING, /* any (initialized) argument is ok */ 206 ARG_PTR_TO_SPIN_LOCK, /* pointer to bpf_spin_lock */ 207 ARG_PTR_TO_SOCK_COMMON, /* pointer to sock_common */ 208 ARG_PTR_TO_INT, /* pointer to int */ 209 ARG_PTR_TO_LONG, /* pointer to long */ 210 ARG_PTR_TO_SOCKET, /* pointer to bpf_sock (fullsock) */ 211}; 212 213/* type of values returned from helper functions */ 214enum bpf_return_type { 215 RET_INTEGER, /* function returns integer */ 216 RET_VOID, /* function doesn't return anything */ 217 RET_PTR_TO_MAP_VALUE, /* returns a pointer to map elem value */ 218 RET_PTR_TO_MAP_VALUE_OR_NULL, /* returns a pointer to map elem value or NULL */ 219 RET_PTR_TO_SOCKET_OR_NULL, /* returns a pointer to a socket or NULL */ 220 RET_PTR_TO_TCP_SOCK_OR_NULL, /* returns a pointer to a tcp_sock or NULL */ 221 RET_PTR_TO_SOCK_COMMON_OR_NULL, /* returns a pointer to a sock_common or NULL */ 222}; 223 224/* eBPF function prototype used by verifier to allow BPF_CALLs from eBPF programs 225 * to in-kernel helper functions and for adjusting imm32 field in BPF_CALL 226 * instructions after verifying 227 */ 228struct bpf_func_proto { 229 u64 (*func)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 230 bool gpl_only; 231 bool pkt_access; 232 enum bpf_return_type ret_type; 233 enum bpf_arg_type arg1_type; 234 enum bpf_arg_type arg2_type; 235 enum bpf_arg_type arg3_type; 236 enum bpf_arg_type arg4_type; 237 enum bpf_arg_type arg5_type; 238}; 239 240/* bpf_context is intentionally undefined structure. Pointer to bpf_context is 241 * the first argument to eBPF programs. 242 * For socket filters: 'struct bpf_context *' == 'struct sk_buff *' 243 */ 244struct bpf_context; 245 246enum bpf_access_type { 247 BPF_READ = 1, 248 BPF_WRITE = 2 249}; 250 251/* types of values stored in eBPF registers */ 252/* Pointer types represent: 253 * pointer 254 * pointer + imm 255 * pointer + (u16) var 256 * pointer + (u16) var + imm 257 * if (range > 0) then [ptr, ptr + range - off) is safe to access 258 * if (id > 0) means that some 'var' was added 259 * if (off > 0) means that 'imm' was added 260 */ 261enum bpf_reg_type { 262 NOT_INIT = 0, /* nothing was written into register */ 263 SCALAR_VALUE, /* reg doesn't contain a valid pointer */ 264 PTR_TO_CTX, /* reg points to bpf_context */ 265 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */ 266 PTR_TO_MAP_VALUE, /* reg points to map element value */ 267 PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */ 268 PTR_TO_STACK, /* reg == frame_pointer + offset */ 269 PTR_TO_PACKET_META, /* skb->data - meta_len */ 270 PTR_TO_PACKET, /* reg points to skb->data */ 271 PTR_TO_PACKET_END, /* skb->data + headlen */ 272 PTR_TO_FLOW_KEYS, /* reg points to bpf_flow_keys */ 273 PTR_TO_SOCKET, /* reg points to struct bpf_sock */ 274 PTR_TO_SOCKET_OR_NULL, /* reg points to struct bpf_sock or NULL */ 275 PTR_TO_SOCK_COMMON, /* reg points to sock_common */ 276 PTR_TO_SOCK_COMMON_OR_NULL, /* reg points to sock_common or NULL */ 277 PTR_TO_TCP_SOCK, /* reg points to struct tcp_sock */ 278 PTR_TO_TCP_SOCK_OR_NULL, /* reg points to struct tcp_sock or NULL */ 279 PTR_TO_TP_BUFFER, /* reg points to a writable raw tp's buffer */ 280 PTR_TO_XDP_SOCK, /* reg points to struct xdp_sock */ 281}; 282 283/* The information passed from prog-specific *_is_valid_access 284 * back to the verifier. 285 */ 286struct bpf_insn_access_aux { 287 enum bpf_reg_type reg_type; 288 int ctx_field_size; 289}; 290 291static inline void 292bpf_ctx_record_field_size(struct bpf_insn_access_aux *aux, u32 size) 293{ 294 aux->ctx_field_size = size; 295} 296 297struct bpf_prog_ops { 298 int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr, 299 union bpf_attr __user *uattr); 300}; 301 302struct bpf_verifier_ops { 303 /* return eBPF function prototype for verification */ 304 const struct bpf_func_proto * 305 (*get_func_proto)(enum bpf_func_id func_id, 306 const struct bpf_prog *prog); 307 308 /* return true if 'size' wide access at offset 'off' within bpf_context 309 * with 'type' (read or write) is allowed 310 */ 311 bool (*is_valid_access)(int off, int size, enum bpf_access_type type, 312 const struct bpf_prog *prog, 313 struct bpf_insn_access_aux *info); 314 int (*gen_prologue)(struct bpf_insn *insn, bool direct_write, 315 const struct bpf_prog *prog); 316 int (*gen_ld_abs)(const struct bpf_insn *orig, 317 struct bpf_insn *insn_buf); 318 u32 (*convert_ctx_access)(enum bpf_access_type type, 319 const struct bpf_insn *src, 320 struct bpf_insn *dst, 321 struct bpf_prog *prog, u32 *target_size); 322}; 323 324struct bpf_prog_offload_ops { 325 /* verifier basic callbacks */ 326 int (*insn_hook)(struct bpf_verifier_env *env, 327 int insn_idx, int prev_insn_idx); 328 int (*finalize)(struct bpf_verifier_env *env); 329 /* verifier optimization callbacks (called after .finalize) */ 330 int (*replace_insn)(struct bpf_verifier_env *env, u32 off, 331 struct bpf_insn *insn); 332 int (*remove_insns)(struct bpf_verifier_env *env, u32 off, u32 cnt); 333 /* program management callbacks */ 334 int (*prepare)(struct bpf_prog *prog); 335 int (*translate)(struct bpf_prog *prog); 336 void (*destroy)(struct bpf_prog *prog); 337}; 338 339struct bpf_prog_offload { 340 struct bpf_prog *prog; 341 struct net_device *netdev; 342 struct bpf_offload_dev *offdev; 343 void *dev_priv; 344 struct list_head offloads; 345 bool dev_state; 346 bool opt_failed; 347 void *jited_image; 348 u32 jited_len; 349}; 350 351enum bpf_cgroup_storage_type { 352 BPF_CGROUP_STORAGE_SHARED, 353 BPF_CGROUP_STORAGE_PERCPU, 354 __BPF_CGROUP_STORAGE_MAX 355}; 356 357#define MAX_BPF_CGROUP_STORAGE_TYPE __BPF_CGROUP_STORAGE_MAX 358 359struct bpf_prog_stats { 360 u64 cnt; 361 u64 nsecs; 362 struct u64_stats_sync syncp; 363}; 364 365struct bpf_prog_aux { 366 atomic_t refcnt; 367 u32 used_map_cnt; 368 u32 max_ctx_offset; 369 u32 max_pkt_offset; 370 u32 max_tp_access; 371 u32 stack_depth; 372 u32 id; 373 u32 func_cnt; /* used by non-func prog as the number of func progs */ 374 u32 func_idx; /* 0 for non-func prog, the index in func array for func prog */ 375 bool verifier_zext; /* Zero extensions has been inserted by verifier. */ 376 bool offload_requested; 377 struct bpf_prog **func; 378 void *jit_data; /* JIT specific data. arch dependent */ 379 struct latch_tree_node ksym_tnode; 380 struct list_head ksym_lnode; 381 const struct bpf_prog_ops *ops; 382 struct bpf_map **used_maps; 383 struct bpf_prog *prog; 384 struct user_struct *user; 385 u64 load_time; /* ns since boottime */ 386 struct bpf_map *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]; 387 char name[BPF_OBJ_NAME_LEN]; 388#ifdef CONFIG_SECURITY 389 void *security; 390#endif 391 struct bpf_prog_offload *offload; 392 struct btf *btf; 393 struct bpf_func_info *func_info; 394 /* bpf_line_info loaded from userspace. linfo->insn_off 395 * has the xlated insn offset. 396 * Both the main and sub prog share the same linfo. 397 * The subprog can access its first linfo by 398 * using the linfo_idx. 399 */ 400 struct bpf_line_info *linfo; 401 /* jited_linfo is the jited addr of the linfo. It has a 402 * one to one mapping to linfo: 403 * jited_linfo[i] is the jited addr for the linfo[i]->insn_off. 404 * Both the main and sub prog share the same jited_linfo. 405 * The subprog can access its first jited_linfo by 406 * using the linfo_idx. 407 */ 408 void **jited_linfo; 409 u32 func_info_cnt; 410 u32 nr_linfo; 411 /* subprog can use linfo_idx to access its first linfo and 412 * jited_linfo. 413 * main prog always has linfo_idx == 0 414 */ 415 u32 linfo_idx; 416 struct bpf_prog_stats __percpu *stats; 417 union { 418 struct work_struct work; 419 struct rcu_head rcu; 420 }; 421}; 422 423struct bpf_array { 424 struct bpf_map map; 425 u32 elem_size; 426 u32 index_mask; 427 /* 'ownership' of prog_array is claimed by the first program that 428 * is going to use this map or by the first program which FD is stored 429 * in the map to make sure that all callers and callees have the same 430 * prog_type and JITed flag 431 */ 432 enum bpf_prog_type owner_prog_type; 433 bool owner_jited; 434 union { 435 char value[0] __aligned(8); 436 void *ptrs[0] __aligned(8); 437 void __percpu *pptrs[0] __aligned(8); 438 }; 439}; 440 441#define BPF_COMPLEXITY_LIMIT_INSNS 1000000 /* yes. 1M insns */ 442#define MAX_TAIL_CALL_CNT 32 443 444#define BPF_F_ACCESS_MASK (BPF_F_RDONLY | \ 445 BPF_F_RDONLY_PROG | \ 446 BPF_F_WRONLY | \ 447 BPF_F_WRONLY_PROG) 448 449#define BPF_MAP_CAN_READ BIT(0) 450#define BPF_MAP_CAN_WRITE BIT(1) 451 452static inline u32 bpf_map_flags_to_cap(struct bpf_map *map) 453{ 454 u32 access_flags = map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG); 455 456 /* Combination of BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG is 457 * not possible. 458 */ 459 if (access_flags & BPF_F_RDONLY_PROG) 460 return BPF_MAP_CAN_READ; 461 else if (access_flags & BPF_F_WRONLY_PROG) 462 return BPF_MAP_CAN_WRITE; 463 else 464 return BPF_MAP_CAN_READ | BPF_MAP_CAN_WRITE; 465} 466 467static inline bool bpf_map_flags_access_ok(u32 access_flags) 468{ 469 return (access_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) != 470 (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG); 471} 472 473struct bpf_event_entry { 474 struct perf_event *event; 475 struct file *perf_file; 476 struct file *map_file; 477 struct rcu_head rcu; 478}; 479 480bool bpf_prog_array_compatible(struct bpf_array *array, const struct bpf_prog *fp); 481int bpf_prog_calc_tag(struct bpf_prog *fp); 482 483const struct bpf_func_proto *bpf_get_trace_printk_proto(void); 484 485typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src, 486 unsigned long off, unsigned long len); 487typedef u32 (*bpf_convert_ctx_access_t)(enum bpf_access_type type, 488 const struct bpf_insn *src, 489 struct bpf_insn *dst, 490 struct bpf_prog *prog, 491 u32 *target_size); 492 493u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 494 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy); 495 496/* an array of programs to be executed under rcu_lock. 497 * 498 * Typical usage: 499 * ret = BPF_PROG_RUN_ARRAY(&bpf_prog_array, ctx, BPF_PROG_RUN); 500 * 501 * the structure returned by bpf_prog_array_alloc() should be populated 502 * with program pointers and the last pointer must be NULL. 503 * The user has to keep refcnt on the program and make sure the program 504 * is removed from the array before bpf_prog_put(). 505 * The 'struct bpf_prog_array *' should only be replaced with xchg() 506 * since other cpus are walking the array of pointers in parallel. 507 */ 508struct bpf_prog_array_item { 509 struct bpf_prog *prog; 510 struct bpf_cgroup_storage *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]; 511}; 512 513struct bpf_prog_array { 514 struct rcu_head rcu; 515 struct bpf_prog_array_item items[0]; 516}; 517 518struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags); 519void bpf_prog_array_free(struct bpf_prog_array *progs); 520int bpf_prog_array_length(struct bpf_prog_array *progs); 521bool bpf_prog_array_is_empty(struct bpf_prog_array *array); 522int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs, 523 __u32 __user *prog_ids, u32 cnt); 524 525void bpf_prog_array_delete_safe(struct bpf_prog_array *progs, 526 struct bpf_prog *old_prog); 527int bpf_prog_array_copy_info(struct bpf_prog_array *array, 528 u32 *prog_ids, u32 request_cnt, 529 u32 *prog_cnt); 530int bpf_prog_array_copy(struct bpf_prog_array *old_array, 531 struct bpf_prog *exclude_prog, 532 struct bpf_prog *include_prog, 533 struct bpf_prog_array **new_array); 534 535#define __BPF_PROG_RUN_ARRAY(array, ctx, func, check_non_null) \ 536 ({ \ 537 struct bpf_prog_array_item *_item; \ 538 struct bpf_prog *_prog; \ 539 struct bpf_prog_array *_array; \ 540 u32 _ret = 1; \ 541 preempt_disable(); \ 542 rcu_read_lock(); \ 543 _array = rcu_dereference(array); \ 544 if (unlikely(check_non_null && !_array))\ 545 goto _out; \ 546 _item = &_array->items[0]; \ 547 while ((_prog = READ_ONCE(_item->prog))) { \ 548 bpf_cgroup_storage_set(_item->cgroup_storage); \ 549 _ret &= func(_prog, ctx); \ 550 _item++; \ 551 } \ 552_out: \ 553 rcu_read_unlock(); \ 554 preempt_enable(); \ 555 _ret; \ 556 }) 557 558/* To be used by __cgroup_bpf_run_filter_skb for EGRESS BPF progs 559 * so BPF programs can request cwr for TCP packets. 560 * 561 * Current cgroup skb programs can only return 0 or 1 (0 to drop the 562 * packet. This macro changes the behavior so the low order bit 563 * indicates whether the packet should be dropped (0) or not (1) 564 * and the next bit is a congestion notification bit. This could be 565 * used by TCP to call tcp_enter_cwr() 566 * 567 * Hence, new allowed return values of CGROUP EGRESS BPF programs are: 568 * 0: drop packet 569 * 1: keep packet 570 * 2: drop packet and cn 571 * 3: keep packet and cn 572 * 573 * This macro then converts it to one of the NET_XMIT or an error 574 * code that is then interpreted as drop packet (and no cn): 575 * 0: NET_XMIT_SUCCESS skb should be transmitted 576 * 1: NET_XMIT_DROP skb should be dropped and cn 577 * 2: NET_XMIT_CN skb should be transmitted and cn 578 * 3: -EPERM skb should be dropped 579 */ 580#define BPF_PROG_CGROUP_INET_EGRESS_RUN_ARRAY(array, ctx, func) \ 581 ({ \ 582 struct bpf_prog_array_item *_item; \ 583 struct bpf_prog *_prog; \ 584 struct bpf_prog_array *_array; \ 585 u32 ret; \ 586 u32 _ret = 1; \ 587 u32 _cn = 0; \ 588 preempt_disable(); \ 589 rcu_read_lock(); \ 590 _array = rcu_dereference(array); \ 591 _item = &_array->items[0]; \ 592 while ((_prog = READ_ONCE(_item->prog))) { \ 593 bpf_cgroup_storage_set(_item->cgroup_storage); \ 594 ret = func(_prog, ctx); \ 595 _ret &= (ret & 1); \ 596 _cn |= (ret & 2); \ 597 _item++; \ 598 } \ 599 rcu_read_unlock(); \ 600 preempt_enable(); \ 601 if (_ret) \ 602 _ret = (_cn ? NET_XMIT_CN : NET_XMIT_SUCCESS); \ 603 else \ 604 _ret = (_cn ? NET_XMIT_DROP : -EPERM); \ 605 _ret; \ 606 }) 607 608#define BPF_PROG_RUN_ARRAY(array, ctx, func) \ 609 __BPF_PROG_RUN_ARRAY(array, ctx, func, false) 610 611#define BPF_PROG_RUN_ARRAY_CHECK(array, ctx, func) \ 612 __BPF_PROG_RUN_ARRAY(array, ctx, func, true) 613 614#ifdef CONFIG_BPF_SYSCALL 615DECLARE_PER_CPU(int, bpf_prog_active); 616 617extern const struct file_operations bpf_map_fops; 618extern const struct file_operations bpf_prog_fops; 619 620#define BPF_PROG_TYPE(_id, _name) \ 621 extern const struct bpf_prog_ops _name ## _prog_ops; \ 622 extern const struct bpf_verifier_ops _name ## _verifier_ops; 623#define BPF_MAP_TYPE(_id, _ops) \ 624 extern const struct bpf_map_ops _ops; 625#include <linux/bpf_types.h> 626#undef BPF_PROG_TYPE 627#undef BPF_MAP_TYPE 628 629extern const struct bpf_prog_ops bpf_offload_prog_ops; 630extern const struct bpf_verifier_ops tc_cls_act_analyzer_ops; 631extern const struct bpf_verifier_ops xdp_analyzer_ops; 632 633struct bpf_prog *bpf_prog_get(u32 ufd); 634struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 635 bool attach_drv); 636struct bpf_prog * __must_check bpf_prog_add(struct bpf_prog *prog, int i); 637void bpf_prog_sub(struct bpf_prog *prog, int i); 638struct bpf_prog * __must_check bpf_prog_inc(struct bpf_prog *prog); 639struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog); 640void bpf_prog_put(struct bpf_prog *prog); 641int __bpf_prog_charge(struct user_struct *user, u32 pages); 642void __bpf_prog_uncharge(struct user_struct *user, u32 pages); 643 644void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock); 645void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock); 646 647struct bpf_map *bpf_map_get_with_uref(u32 ufd); 648struct bpf_map *__bpf_map_get(struct fd f); 649struct bpf_map * __must_check bpf_map_inc(struct bpf_map *map, bool uref); 650void bpf_map_put_with_uref(struct bpf_map *map); 651void bpf_map_put(struct bpf_map *map); 652int bpf_map_charge_memlock(struct bpf_map *map, u32 pages); 653void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages); 654int bpf_map_charge_init(struct bpf_map_memory *mem, size_t size); 655void bpf_map_charge_finish(struct bpf_map_memory *mem); 656void bpf_map_charge_move(struct bpf_map_memory *dst, 657 struct bpf_map_memory *src); 658void *bpf_map_area_alloc(size_t size, int numa_node); 659void bpf_map_area_free(void *base); 660void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr); 661 662extern int sysctl_unprivileged_bpf_disabled; 663 664int bpf_map_new_fd(struct bpf_map *map, int flags); 665int bpf_prog_new_fd(struct bpf_prog *prog); 666 667int bpf_obj_pin_user(u32 ufd, const char __user *pathname); 668int bpf_obj_get_user(const char __user *pathname, int flags); 669 670int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value); 671int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value); 672int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value, 673 u64 flags); 674int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value, 675 u64 flags); 676 677int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value); 678 679int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file, 680 void *key, void *value, u64 map_flags); 681int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 682int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file, 683 void *key, void *value, u64 map_flags); 684int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 685 686int bpf_get_file_flag(int flags); 687int bpf_check_uarg_tail_zero(void __user *uaddr, size_t expected_size, 688 size_t actual_size); 689 690/* memcpy that is used with 8-byte aligned pointers, power-of-8 size and 691 * forced to use 'long' read/writes to try to atomically copy long counters. 692 * Best-effort only. No barriers here, since it _will_ race with concurrent 693 * updates from BPF programs. Called from bpf syscall and mostly used with 694 * size 8 or 16 bytes, so ask compiler to inline it. 695 */ 696static inline void bpf_long_memcpy(void *dst, const void *src, u32 size) 697{ 698 const long *lsrc = src; 699 long *ldst = dst; 700 701 size /= sizeof(long); 702 while (size--) 703 *ldst++ = *lsrc++; 704} 705 706/* verify correctness of eBPF program */ 707int bpf_check(struct bpf_prog **fp, union bpf_attr *attr, 708 union bpf_attr __user *uattr); 709void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth); 710 711/* Map specifics */ 712struct xdp_buff; 713struct sk_buff; 714 715struct bpf_dtab_netdev *__dev_map_lookup_elem(struct bpf_map *map, u32 key); 716void __dev_map_insert_ctx(struct bpf_map *map, u32 index); 717void __dev_map_flush(struct bpf_map *map); 718int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, 719 struct net_device *dev_rx); 720int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, struct sk_buff *skb, 721 struct bpf_prog *xdp_prog); 722 723struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key); 724void __cpu_map_insert_ctx(struct bpf_map *map, u32 index); 725void __cpu_map_flush(struct bpf_map *map); 726int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_buff *xdp, 727 struct net_device *dev_rx); 728 729/* Return map's numa specified by userspace */ 730static inline int bpf_map_attr_numa_node(const union bpf_attr *attr) 731{ 732 return (attr->map_flags & BPF_F_NUMA_NODE) ? 733 attr->numa_node : NUMA_NO_NODE; 734} 735 736struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type); 737int array_map_alloc_check(union bpf_attr *attr); 738 739int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr, 740 union bpf_attr __user *uattr); 741int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr, 742 union bpf_attr __user *uattr); 743int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 744 const union bpf_attr *kattr, 745 union bpf_attr __user *uattr); 746#else /* !CONFIG_BPF_SYSCALL */ 747static inline struct bpf_prog *bpf_prog_get(u32 ufd) 748{ 749 return ERR_PTR(-EOPNOTSUPP); 750} 751 752static inline struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, 753 enum bpf_prog_type type, 754 bool attach_drv) 755{ 756 return ERR_PTR(-EOPNOTSUPP); 757} 758 759static inline struct bpf_prog * __must_check bpf_prog_add(struct bpf_prog *prog, 760 int i) 761{ 762 return ERR_PTR(-EOPNOTSUPP); 763} 764 765static inline void bpf_prog_sub(struct bpf_prog *prog, int i) 766{ 767} 768 769static inline void bpf_prog_put(struct bpf_prog *prog) 770{ 771} 772 773static inline struct bpf_prog * __must_check bpf_prog_inc(struct bpf_prog *prog) 774{ 775 return ERR_PTR(-EOPNOTSUPP); 776} 777 778static inline struct bpf_prog *__must_check 779bpf_prog_inc_not_zero(struct bpf_prog *prog) 780{ 781 return ERR_PTR(-EOPNOTSUPP); 782} 783 784static inline int __bpf_prog_charge(struct user_struct *user, u32 pages) 785{ 786 return 0; 787} 788 789static inline void __bpf_prog_uncharge(struct user_struct *user, u32 pages) 790{ 791} 792 793static inline int bpf_obj_get_user(const char __user *pathname, int flags) 794{ 795 return -EOPNOTSUPP; 796} 797 798static inline struct net_device *__dev_map_lookup_elem(struct bpf_map *map, 799 u32 key) 800{ 801 return NULL; 802} 803 804static inline void __dev_map_insert_ctx(struct bpf_map *map, u32 index) 805{ 806} 807 808static inline void __dev_map_flush(struct bpf_map *map) 809{ 810} 811 812struct xdp_buff; 813struct bpf_dtab_netdev; 814 815static inline 816int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, 817 struct net_device *dev_rx) 818{ 819 return 0; 820} 821 822struct sk_buff; 823 824static inline int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, 825 struct sk_buff *skb, 826 struct bpf_prog *xdp_prog) 827{ 828 return 0; 829} 830 831static inline 832struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key) 833{ 834 return NULL; 835} 836 837static inline void __cpu_map_insert_ctx(struct bpf_map *map, u32 index) 838{ 839} 840 841static inline void __cpu_map_flush(struct bpf_map *map) 842{ 843} 844 845static inline int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, 846 struct xdp_buff *xdp, 847 struct net_device *dev_rx) 848{ 849 return 0; 850} 851 852static inline struct bpf_prog *bpf_prog_get_type_path(const char *name, 853 enum bpf_prog_type type) 854{ 855 return ERR_PTR(-EOPNOTSUPP); 856} 857 858static inline int bpf_prog_test_run_xdp(struct bpf_prog *prog, 859 const union bpf_attr *kattr, 860 union bpf_attr __user *uattr) 861{ 862 return -ENOTSUPP; 863} 864 865static inline int bpf_prog_test_run_skb(struct bpf_prog *prog, 866 const union bpf_attr *kattr, 867 union bpf_attr __user *uattr) 868{ 869 return -ENOTSUPP; 870} 871 872static inline int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 873 const union bpf_attr *kattr, 874 union bpf_attr __user *uattr) 875{ 876 return -ENOTSUPP; 877} 878#endif /* CONFIG_BPF_SYSCALL */ 879 880static inline struct bpf_prog *bpf_prog_get_type(u32 ufd, 881 enum bpf_prog_type type) 882{ 883 return bpf_prog_get_type_dev(ufd, type, false); 884} 885 886bool bpf_prog_get_ok(struct bpf_prog *, enum bpf_prog_type *, bool); 887 888int bpf_prog_offload_compile(struct bpf_prog *prog); 889void bpf_prog_offload_destroy(struct bpf_prog *prog); 890int bpf_prog_offload_info_fill(struct bpf_prog_info *info, 891 struct bpf_prog *prog); 892 893int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map); 894 895int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value); 896int bpf_map_offload_update_elem(struct bpf_map *map, 897 void *key, void *value, u64 flags); 898int bpf_map_offload_delete_elem(struct bpf_map *map, void *key); 899int bpf_map_offload_get_next_key(struct bpf_map *map, 900 void *key, void *next_key); 901 902bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map); 903 904struct bpf_offload_dev * 905bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv); 906void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev); 907void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev); 908int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev, 909 struct net_device *netdev); 910void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev, 911 struct net_device *netdev); 912bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev); 913 914#if defined(CONFIG_NET) && defined(CONFIG_BPF_SYSCALL) 915int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr); 916 917static inline bool bpf_prog_is_dev_bound(const struct bpf_prog_aux *aux) 918{ 919 return aux->offload_requested; 920} 921 922static inline bool bpf_map_is_dev_bound(struct bpf_map *map) 923{ 924 return unlikely(map->ops == &bpf_map_offload_ops); 925} 926 927struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr); 928void bpf_map_offload_map_free(struct bpf_map *map); 929#else 930static inline int bpf_prog_offload_init(struct bpf_prog *prog, 931 union bpf_attr *attr) 932{ 933 return -EOPNOTSUPP; 934} 935 936static inline bool bpf_prog_is_dev_bound(struct bpf_prog_aux *aux) 937{ 938 return false; 939} 940 941static inline bool bpf_map_is_dev_bound(struct bpf_map *map) 942{ 943 return false; 944} 945 946static inline struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr) 947{ 948 return ERR_PTR(-EOPNOTSUPP); 949} 950 951static inline void bpf_map_offload_map_free(struct bpf_map *map) 952{ 953} 954#endif /* CONFIG_NET && CONFIG_BPF_SYSCALL */ 955 956#if defined(CONFIG_BPF_STREAM_PARSER) 957int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog, u32 which); 958int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog); 959#else 960static inline int sock_map_prog_update(struct bpf_map *map, 961 struct bpf_prog *prog, u32 which) 962{ 963 return -EOPNOTSUPP; 964} 965 966static inline int sock_map_get_from_fd(const union bpf_attr *attr, 967 struct bpf_prog *prog) 968{ 969 return -EINVAL; 970} 971#endif 972 973#if defined(CONFIG_XDP_SOCKETS) 974struct xdp_sock; 975struct xdp_sock *__xsk_map_lookup_elem(struct bpf_map *map, u32 key); 976int __xsk_map_redirect(struct bpf_map *map, struct xdp_buff *xdp, 977 struct xdp_sock *xs); 978void __xsk_map_flush(struct bpf_map *map); 979#else 980struct xdp_sock; 981static inline struct xdp_sock *__xsk_map_lookup_elem(struct bpf_map *map, 982 u32 key) 983{ 984 return NULL; 985} 986 987static inline int __xsk_map_redirect(struct bpf_map *map, struct xdp_buff *xdp, 988 struct xdp_sock *xs) 989{ 990 return -EOPNOTSUPP; 991} 992 993static inline void __xsk_map_flush(struct bpf_map *map) 994{ 995} 996#endif 997 998#if defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) 999void bpf_sk_reuseport_detach(struct sock *sk); 1000int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, void *key, 1001 void *value); 1002int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, void *key, 1003 void *value, u64 map_flags); 1004#else 1005static inline void bpf_sk_reuseport_detach(struct sock *sk) 1006{ 1007} 1008 1009#ifdef CONFIG_BPF_SYSCALL 1010static inline int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, 1011 void *key, void *value) 1012{ 1013 return -EOPNOTSUPP; 1014} 1015 1016static inline int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, 1017 void *key, void *value, 1018 u64 map_flags) 1019{ 1020 return -EOPNOTSUPP; 1021} 1022#endif /* CONFIG_BPF_SYSCALL */ 1023#endif /* defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) */ 1024 1025/* verifier prototypes for helper functions called from eBPF programs */ 1026extern const struct bpf_func_proto bpf_map_lookup_elem_proto; 1027extern const struct bpf_func_proto bpf_map_update_elem_proto; 1028extern const struct bpf_func_proto bpf_map_delete_elem_proto; 1029extern const struct bpf_func_proto bpf_map_push_elem_proto; 1030extern const struct bpf_func_proto bpf_map_pop_elem_proto; 1031extern const struct bpf_func_proto bpf_map_peek_elem_proto; 1032 1033extern const struct bpf_func_proto bpf_get_prandom_u32_proto; 1034extern const struct bpf_func_proto bpf_get_smp_processor_id_proto; 1035extern const struct bpf_func_proto bpf_get_numa_node_id_proto; 1036extern const struct bpf_func_proto bpf_tail_call_proto; 1037extern const struct bpf_func_proto bpf_ktime_get_ns_proto; 1038extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto; 1039extern const struct bpf_func_proto bpf_get_current_uid_gid_proto; 1040extern const struct bpf_func_proto bpf_get_current_comm_proto; 1041extern const struct bpf_func_proto bpf_get_stackid_proto; 1042extern const struct bpf_func_proto bpf_get_stack_proto; 1043extern const struct bpf_func_proto bpf_sock_map_update_proto; 1044extern const struct bpf_func_proto bpf_sock_hash_update_proto; 1045extern const struct bpf_func_proto bpf_get_current_cgroup_id_proto; 1046extern const struct bpf_func_proto bpf_msg_redirect_hash_proto; 1047extern const struct bpf_func_proto bpf_msg_redirect_map_proto; 1048extern const struct bpf_func_proto bpf_sk_redirect_hash_proto; 1049extern const struct bpf_func_proto bpf_sk_redirect_map_proto; 1050extern const struct bpf_func_proto bpf_spin_lock_proto; 1051extern const struct bpf_func_proto bpf_spin_unlock_proto; 1052extern const struct bpf_func_proto bpf_get_local_storage_proto; 1053extern const struct bpf_func_proto bpf_strtol_proto; 1054extern const struct bpf_func_proto bpf_strtoul_proto; 1055extern const struct bpf_func_proto bpf_tcp_sock_proto; 1056 1057/* Shared helpers among cBPF and eBPF. */ 1058void bpf_user_rnd_init_once(void); 1059u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 1060 1061#if defined(CONFIG_NET) 1062bool bpf_sock_common_is_valid_access(int off, int size, 1063 enum bpf_access_type type, 1064 struct bpf_insn_access_aux *info); 1065bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 1066 struct bpf_insn_access_aux *info); 1067u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 1068 const struct bpf_insn *si, 1069 struct bpf_insn *insn_buf, 1070 struct bpf_prog *prog, 1071 u32 *target_size); 1072#else 1073static inline bool bpf_sock_common_is_valid_access(int off, int size, 1074 enum bpf_access_type type, 1075 struct bpf_insn_access_aux *info) 1076{ 1077 return false; 1078} 1079static inline bool bpf_sock_is_valid_access(int off, int size, 1080 enum bpf_access_type type, 1081 struct bpf_insn_access_aux *info) 1082{ 1083 return false; 1084} 1085static inline u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 1086 const struct bpf_insn *si, 1087 struct bpf_insn *insn_buf, 1088 struct bpf_prog *prog, 1089 u32 *target_size) 1090{ 1091 return 0; 1092} 1093#endif 1094 1095#ifdef CONFIG_INET 1096bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 1097 struct bpf_insn_access_aux *info); 1098 1099u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 1100 const struct bpf_insn *si, 1101 struct bpf_insn *insn_buf, 1102 struct bpf_prog *prog, 1103 u32 *target_size); 1104 1105bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 1106 struct bpf_insn_access_aux *info); 1107 1108u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 1109 const struct bpf_insn *si, 1110 struct bpf_insn *insn_buf, 1111 struct bpf_prog *prog, 1112 u32 *target_size); 1113#else 1114static inline bool bpf_tcp_sock_is_valid_access(int off, int size, 1115 enum bpf_access_type type, 1116 struct bpf_insn_access_aux *info) 1117{ 1118 return false; 1119} 1120 1121static inline u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 1122 const struct bpf_insn *si, 1123 struct bpf_insn *insn_buf, 1124 struct bpf_prog *prog, 1125 u32 *target_size) 1126{ 1127 return 0; 1128} 1129static inline bool bpf_xdp_sock_is_valid_access(int off, int size, 1130 enum bpf_access_type type, 1131 struct bpf_insn_access_aux *info) 1132{ 1133 return false; 1134} 1135 1136static inline u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 1137 const struct bpf_insn *si, 1138 struct bpf_insn *insn_buf, 1139 struct bpf_prog *prog, 1140 u32 *target_size) 1141{ 1142 return 0; 1143} 1144#endif /* CONFIG_INET */ 1145 1146#endif /* _LINUX_BPF_H */