1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com 2 * 3 * This program is free software; you can redistribute it and/or 4 * modify it under the terms of version 2 of the GNU General Public 5 * License as published by the Free Software Foundation. 6 */ 7 #ifndef _LINUX_BPF_H 8 #define _LINUX_BPF_H 1 9 10 #include <uapi/linux/bpf.h> 11 12 #include <linux/workqueue.h> 13 #include <linux/file.h> 14 #include <linux/percpu.h> 15 #include <linux/err.h> 16 #include <linux/rbtree_latch.h> 17 #include <linux/numa.h> 18 #include <linux/wait.h> 19 20 struct perf_event; 21 struct bpf_prog; 22 struct bpf_map; 23 24 /* map is generic key/value storage optionally accesible by eBPF programs */ 25 struct bpf_map_ops { 26 /* funcs callable from userspace (via syscall) */ 27 struct bpf_map *(*map_alloc)(union bpf_attr *attr); 28 void (*map_release)(struct bpf_map *map, struct file *map_file); 29 void (*map_free)(struct bpf_map *map); 30 int (*map_get_next_key)(struct bpf_map *map, void *key, void *next_key); 31 32 /* funcs callable from userspace and from eBPF programs */ 33 void *(*map_lookup_elem)(struct bpf_map *map, void *key); 34 int (*map_update_elem)(struct bpf_map *map, void *key, void *value, u64 flags); 35 int (*map_delete_elem)(struct bpf_map *map, void *key); 36 37 /* funcs called by prog_array and perf_event_array map */ 38 void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file, 39 int fd); 40 void (*map_fd_put_ptr)(void *ptr); 41 u32 (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf); 42 u32 (*map_fd_sys_lookup_elem)(void *ptr); 43 }; 44 45 struct bpf_map { 46 atomic_t refcnt; 47 enum bpf_map_type map_type; 48 u32 key_size; 49 u32 value_size; 50 u32 max_entries; 51 u32 map_flags; 52 u32 pages; 53 u32 id; 54 int numa_node; 55 struct user_struct *user; 56 const struct bpf_map_ops *ops; 57 struct work_struct work; 58 atomic_t usercnt; 59 struct bpf_map *inner_map_meta; 60 char name[BPF_OBJ_NAME_LEN]; 61 #ifdef CONFIG_SECURITY 62 void *security; 63 #endif 64 }; 65 66 /* function argument constraints */ 67 enum bpf_arg_type { 68 ARG_DONTCARE = 0, /* unused argument in helper function */ 69 70 /* the following constraints used to prototype 71 * bpf_map_lookup/update/delete_elem() functions 72 */ 73 ARG_CONST_MAP_PTR, /* const argument used as pointer to bpf_map */ 74 ARG_PTR_TO_MAP_KEY, /* pointer to stack used as map key */ 75 ARG_PTR_TO_MAP_VALUE, /* pointer to stack used as map value */ 76 77 /* the following constraints used to prototype bpf_memcmp() and other 78 * functions that access data on eBPF program stack 79 */ 80 ARG_PTR_TO_MEM, /* pointer to valid memory (stack, packet, map value) */ 81 ARG_PTR_TO_MEM_OR_NULL, /* pointer to valid memory or NULL */ 82 ARG_PTR_TO_UNINIT_MEM, /* pointer to memory does not need to be initialized, 83 * helper function must fill all bytes or clear 84 * them in error case. 85 */ 86 87 ARG_CONST_SIZE, /* number of bytes accessed from memory */ 88 ARG_CONST_SIZE_OR_ZERO, /* number of bytes accessed from memory or 0 */ 89 90 ARG_PTR_TO_CTX, /* pointer to context */ 91 ARG_ANYTHING, /* any (initialized) argument is ok */ 92 }; 93 94 /* type of values returned from helper functions */ 95 enum bpf_return_type { 96 RET_INTEGER, /* function returns integer */ 97 RET_VOID, /* function doesn't return anything */ 98 RET_PTR_TO_MAP_VALUE_OR_NULL, /* returns a pointer to map elem value or NULL */ 99 }; 100 101 /* eBPF function prototype used by verifier to allow BPF_CALLs from eBPF programs 102 * to in-kernel helper functions and for adjusting imm32 field in BPF_CALL 103 * instructions after verifying 104 */ 105 struct bpf_func_proto { 106 u64 (*func)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 107 bool gpl_only; 108 bool pkt_access; 109 enum bpf_return_type ret_type; 110 enum bpf_arg_type arg1_type; 111 enum bpf_arg_type arg2_type; 112 enum bpf_arg_type arg3_type; 113 enum bpf_arg_type arg4_type; 114 enum bpf_arg_type arg5_type; 115 }; 116 117 /* bpf_context is intentionally undefined structure. Pointer to bpf_context is 118 * the first argument to eBPF programs. 119 * For socket filters: 'struct bpf_context *' == 'struct sk_buff *' 120 */ 121 struct bpf_context; 122 123 enum bpf_access_type { 124 BPF_READ = 1, 125 BPF_WRITE = 2 126 }; 127 128 /* types of values stored in eBPF registers */ 129 /* Pointer types represent: 130 * pointer 131 * pointer + imm 132 * pointer + (u16) var 133 * pointer + (u16) var + imm 134 * if (range > 0) then [ptr, ptr + range - off) is safe to access 135 * if (id > 0) means that some 'var' was added 136 * if (off > 0) means that 'imm' was added 137 */ 138 enum bpf_reg_type { 139 NOT_INIT = 0, /* nothing was written into register */ 140 SCALAR_VALUE, /* reg doesn't contain a valid pointer */ 141 PTR_TO_CTX, /* reg points to bpf_context */ 142 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */ 143 PTR_TO_MAP_VALUE, /* reg points to map element value */ 144 PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */ 145 PTR_TO_STACK, /* reg == frame_pointer + offset */ 146 PTR_TO_PACKET_META, /* skb->data - meta_len */ 147 PTR_TO_PACKET, /* reg points to skb->data */ 148 PTR_TO_PACKET_END, /* skb->data + headlen */ 149 }; 150 151 /* The information passed from prog-specific *_is_valid_access 152 * back to the verifier. 153 */ 154 struct bpf_insn_access_aux { 155 enum bpf_reg_type reg_type; 156 int ctx_field_size; 157 }; 158 159 static inline void 160 bpf_ctx_record_field_size(struct bpf_insn_access_aux *aux, u32 size) 161 { 162 aux->ctx_field_size = size; 163 } 164 165 struct bpf_prog_ops { 166 int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr, 167 union bpf_attr __user *uattr); 168 }; 169 170 struct bpf_verifier_ops { 171 /* return eBPF function prototype for verification */ 172 const struct bpf_func_proto *(*get_func_proto)(enum bpf_func_id func_id); 173 174 /* return true if 'size' wide access at offset 'off' within bpf_context 175 * with 'type' (read or write) is allowed 176 */ 177 bool (*is_valid_access)(int off, int size, enum bpf_access_type type, 178 struct bpf_insn_access_aux *info); 179 int (*gen_prologue)(struct bpf_insn *insn, bool direct_write, 180 const struct bpf_prog *prog); 181 u32 (*convert_ctx_access)(enum bpf_access_type type, 182 const struct bpf_insn *src, 183 struct bpf_insn *dst, 184 struct bpf_prog *prog, u32 *target_size); 185 }; 186 187 struct bpf_dev_offload { 188 struct bpf_prog *prog; 189 struct net_device *netdev; 190 void *dev_priv; 191 struct list_head offloads; 192 bool dev_state; 193 bool verifier_running; 194 wait_queue_head_t verifier_done; 195 }; 196 197 struct bpf_prog_aux { 198 atomic_t refcnt; 199 u32 used_map_cnt; 200 u32 max_ctx_offset; 201 u32 stack_depth; 202 u32 id; 203 struct latch_tree_node ksym_tnode; 204 struct list_head ksym_lnode; 205 const struct bpf_prog_ops *ops; 206 struct bpf_map **used_maps; 207 struct bpf_prog *prog; 208 struct user_struct *user; 209 u64 load_time; /* ns since boottime */ 210 char name[BPF_OBJ_NAME_LEN]; 211 #ifdef CONFIG_SECURITY 212 void *security; 213 #endif 214 struct bpf_dev_offload *offload; 215 union { 216 struct work_struct work; 217 struct rcu_head rcu; 218 }; 219 }; 220 221 struct bpf_array { 222 struct bpf_map map; 223 u32 elem_size; 224 /* 'ownership' of prog_array is claimed by the first program that 225 * is going to use this map or by the first program which FD is stored 226 * in the map to make sure that all callers and callees have the same 227 * prog_type and JITed flag 228 */ 229 enum bpf_prog_type owner_prog_type; 230 bool owner_jited; 231 union { 232 char value[0] __aligned(8); 233 void *ptrs[0] __aligned(8); 234 void __percpu *pptrs[0] __aligned(8); 235 }; 236 }; 237 238 #define MAX_TAIL_CALL_CNT 32 239 240 struct bpf_event_entry { 241 struct perf_event *event; 242 struct file *perf_file; 243 struct file *map_file; 244 struct rcu_head rcu; 245 }; 246 247 bool bpf_prog_array_compatible(struct bpf_array *array, const struct bpf_prog *fp); 248 int bpf_prog_calc_tag(struct bpf_prog *fp); 249 250 const struct bpf_func_proto *bpf_get_trace_printk_proto(void); 251 252 typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src, 253 unsigned long off, unsigned long len); 254 255 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 256 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy); 257 258 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr, 259 union bpf_attr __user *uattr); 260 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr, 261 union bpf_attr __user *uattr); 262 263 /* an array of programs to be executed under rcu_lock. 264 * 265 * Typical usage: 266 * ret = BPF_PROG_RUN_ARRAY(&bpf_prog_array, ctx, BPF_PROG_RUN); 267 * 268 * the structure returned by bpf_prog_array_alloc() should be populated 269 * with program pointers and the last pointer must be NULL. 270 * The user has to keep refcnt on the program and make sure the program 271 * is removed from the array before bpf_prog_put(). 272 * The 'struct bpf_prog_array *' should only be replaced with xchg() 273 * since other cpus are walking the array of pointers in parallel. 274 */ 275 struct bpf_prog_array { 276 struct rcu_head rcu; 277 struct bpf_prog *progs[0]; 278 }; 279 280 struct bpf_prog_array __rcu *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags); 281 void bpf_prog_array_free(struct bpf_prog_array __rcu *progs); 282 int bpf_prog_array_length(struct bpf_prog_array __rcu *progs); 283 int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *progs, 284 __u32 __user *prog_ids, u32 cnt); 285 286 void bpf_prog_array_delete_safe(struct bpf_prog_array __rcu *progs, 287 struct bpf_prog *old_prog); 288 int bpf_prog_array_copy(struct bpf_prog_array __rcu *old_array, 289 struct bpf_prog *exclude_prog, 290 struct bpf_prog *include_prog, 291 struct bpf_prog_array **new_array); 292 293 #define __BPF_PROG_RUN_ARRAY(array, ctx, func, check_non_null) \ 294 ({ \ 295 struct bpf_prog **_prog, *__prog; \ 296 struct bpf_prog_array *_array; \ 297 u32 _ret = 1; \ 298 rcu_read_lock(); \ 299 _array = rcu_dereference(array); \ 300 if (unlikely(check_non_null && !_array))\ 301 goto _out; \ 302 _prog = _array->progs; \ 303 while ((__prog = READ_ONCE(*_prog))) { \ 304 _ret &= func(__prog, ctx); \ 305 _prog++; \ 306 } \ 307 _out: \ 308 rcu_read_unlock(); \ 309 _ret; \ 310 }) 311 312 #define BPF_PROG_RUN_ARRAY(array, ctx, func) \ 313 __BPF_PROG_RUN_ARRAY(array, ctx, func, false) 314 315 #define BPF_PROG_RUN_ARRAY_CHECK(array, ctx, func) \ 316 __BPF_PROG_RUN_ARRAY(array, ctx, func, true) 317 318 #ifdef CONFIG_BPF_SYSCALL 319 DECLARE_PER_CPU(int, bpf_prog_active); 320 321 extern const struct file_operations bpf_map_fops; 322 extern const struct file_operations bpf_prog_fops; 323 324 #define BPF_PROG_TYPE(_id, _name) \ 325 extern const struct bpf_prog_ops _name ## _prog_ops; \ 326 extern const struct bpf_verifier_ops _name ## _verifier_ops; 327 #define BPF_MAP_TYPE(_id, _ops) \ 328 extern const struct bpf_map_ops _ops; 329 #include <linux/bpf_types.h> 330 #undef BPF_PROG_TYPE 331 #undef BPF_MAP_TYPE 332 333 extern const struct bpf_prog_ops bpf_offload_prog_ops; 334 extern const struct bpf_verifier_ops tc_cls_act_analyzer_ops; 335 extern const struct bpf_verifier_ops xdp_analyzer_ops; 336 337 struct bpf_prog *bpf_prog_get(u32 ufd); 338 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 339 bool attach_drv); 340 struct bpf_prog * __must_check bpf_prog_add(struct bpf_prog *prog, int i); 341 void bpf_prog_sub(struct bpf_prog *prog, int i); 342 struct bpf_prog * __must_check bpf_prog_inc(struct bpf_prog *prog); 343 struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog); 344 void bpf_prog_put(struct bpf_prog *prog); 345 int __bpf_prog_charge(struct user_struct *user, u32 pages); 346 void __bpf_prog_uncharge(struct user_struct *user, u32 pages); 347 348 struct bpf_map *bpf_map_get_with_uref(u32 ufd); 349 struct bpf_map *__bpf_map_get(struct fd f); 350 struct bpf_map * __must_check bpf_map_inc(struct bpf_map *map, bool uref); 351 void bpf_map_put_with_uref(struct bpf_map *map); 352 void bpf_map_put(struct bpf_map *map); 353 int bpf_map_precharge_memlock(u32 pages); 354 void *bpf_map_area_alloc(size_t size, int numa_node); 355 void bpf_map_area_free(void *base); 356 357 extern int sysctl_unprivileged_bpf_disabled; 358 359 int bpf_map_new_fd(struct bpf_map *map, int flags); 360 int bpf_prog_new_fd(struct bpf_prog *prog); 361 362 int bpf_obj_pin_user(u32 ufd, const char __user *pathname); 363 int bpf_obj_get_user(const char __user *pathname, int flags); 364 365 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value); 366 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value); 367 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value, 368 u64 flags); 369 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value, 370 u64 flags); 371 372 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value); 373 374 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file, 375 void *key, void *value, u64 map_flags); 376 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 377 void bpf_fd_array_map_clear(struct bpf_map *map); 378 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file, 379 void *key, void *value, u64 map_flags); 380 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 381 382 int bpf_get_file_flag(int flags); 383 384 /* memcpy that is used with 8-byte aligned pointers, power-of-8 size and 385 * forced to use 'long' read/writes to try to atomically copy long counters. 386 * Best-effort only. No barriers here, since it _will_ race with concurrent 387 * updates from BPF programs. Called from bpf syscall and mostly used with 388 * size 8 or 16 bytes, so ask compiler to inline it. 389 */ 390 static inline void bpf_long_memcpy(void *dst, const void *src, u32 size) 391 { 392 const long *lsrc = src; 393 long *ldst = dst; 394 395 size /= sizeof(long); 396 while (size--) 397 *ldst++ = *lsrc++; 398 } 399 400 /* verify correctness of eBPF program */ 401 int bpf_check(struct bpf_prog **fp, union bpf_attr *attr); 402 403 /* Map specifics */ 404 struct net_device *__dev_map_lookup_elem(struct bpf_map *map, u32 key); 405 void __dev_map_insert_ctx(struct bpf_map *map, u32 index); 406 void __dev_map_flush(struct bpf_map *map); 407 408 struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key); 409 void __cpu_map_insert_ctx(struct bpf_map *map, u32 index); 410 void __cpu_map_flush(struct bpf_map *map); 411 struct xdp_buff; 412 int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_buff *xdp, 413 struct net_device *dev_rx); 414 415 /* Return map's numa specified by userspace */ 416 static inline int bpf_map_attr_numa_node(const union bpf_attr *attr) 417 { 418 return (attr->map_flags & BPF_F_NUMA_NODE) ? 419 attr->numa_node : NUMA_NO_NODE; 420 } 421 422 #else /* !CONFIG_BPF_SYSCALL */ 423 static inline struct bpf_prog *bpf_prog_get(u32 ufd) 424 { 425 return ERR_PTR(-EOPNOTSUPP); 426 } 427 428 static inline struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, 429 enum bpf_prog_type type, 430 bool attach_drv) 431 { 432 return ERR_PTR(-EOPNOTSUPP); 433 } 434 435 static inline struct bpf_prog * __must_check bpf_prog_add(struct bpf_prog *prog, 436 int i) 437 { 438 return ERR_PTR(-EOPNOTSUPP); 439 } 440 441 static inline void bpf_prog_sub(struct bpf_prog *prog, int i) 442 { 443 } 444 445 static inline void bpf_prog_put(struct bpf_prog *prog) 446 { 447 } 448 449 static inline struct bpf_prog * __must_check bpf_prog_inc(struct bpf_prog *prog) 450 { 451 return ERR_PTR(-EOPNOTSUPP); 452 } 453 454 static inline struct bpf_prog *__must_check 455 bpf_prog_inc_not_zero(struct bpf_prog *prog) 456 { 457 return ERR_PTR(-EOPNOTSUPP); 458 } 459 460 static inline int __bpf_prog_charge(struct user_struct *user, u32 pages) 461 { 462 return 0; 463 } 464 465 static inline void __bpf_prog_uncharge(struct user_struct *user, u32 pages) 466 { 467 } 468 469 static inline int bpf_obj_get_user(const char __user *pathname, int flags) 470 { 471 return -EOPNOTSUPP; 472 } 473 474 static inline struct net_device *__dev_map_lookup_elem(struct bpf_map *map, 475 u32 key) 476 { 477 return NULL; 478 } 479 480 static inline void __dev_map_insert_ctx(struct bpf_map *map, u32 index) 481 { 482 } 483 484 static inline void __dev_map_flush(struct bpf_map *map) 485 { 486 } 487 488 static inline 489 struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key) 490 { 491 return NULL; 492 } 493 494 static inline void __cpu_map_insert_ctx(struct bpf_map *map, u32 index) 495 { 496 } 497 498 static inline void __cpu_map_flush(struct bpf_map *map) 499 { 500 } 501 502 struct xdp_buff; 503 static inline int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, 504 struct xdp_buff *xdp, 505 struct net_device *dev_rx) 506 { 507 return 0; 508 } 509 #endif /* CONFIG_BPF_SYSCALL */ 510 511 static inline struct bpf_prog *bpf_prog_get_type(u32 ufd, 512 enum bpf_prog_type type) 513 { 514 return bpf_prog_get_type_dev(ufd, type, false); 515 } 516 517 int bpf_prog_offload_compile(struct bpf_prog *prog); 518 void bpf_prog_offload_destroy(struct bpf_prog *prog); 519 520 #if defined(CONFIG_NET) && defined(CONFIG_BPF_SYSCALL) 521 int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr); 522 523 static inline bool bpf_prog_is_dev_bound(struct bpf_prog_aux *aux) 524 { 525 return aux->offload; 526 } 527 #else 528 static inline int bpf_prog_offload_init(struct bpf_prog *prog, 529 union bpf_attr *attr) 530 { 531 return -EOPNOTSUPP; 532 } 533 534 static inline bool bpf_prog_is_dev_bound(struct bpf_prog_aux *aux) 535 { 536 return false; 537 } 538 #endif /* CONFIG_NET && CONFIG_BPF_SYSCALL */ 539 540 #if defined(CONFIG_STREAM_PARSER) && defined(CONFIG_BPF_SYSCALL) 541 struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key); 542 int sock_map_prog(struct bpf_map *map, struct bpf_prog *prog, u32 type); 543 #else 544 static inline struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key) 545 { 546 return NULL; 547 } 548 549 static inline int sock_map_prog(struct bpf_map *map, 550 struct bpf_prog *prog, 551 u32 type) 552 { 553 return -EOPNOTSUPP; 554 } 555 #endif 556 557 /* verifier prototypes for helper functions called from eBPF programs */ 558 extern const struct bpf_func_proto bpf_map_lookup_elem_proto; 559 extern const struct bpf_func_proto bpf_map_update_elem_proto; 560 extern const struct bpf_func_proto bpf_map_delete_elem_proto; 561 562 extern const struct bpf_func_proto bpf_get_prandom_u32_proto; 563 extern const struct bpf_func_proto bpf_get_smp_processor_id_proto; 564 extern const struct bpf_func_proto bpf_get_numa_node_id_proto; 565 extern const struct bpf_func_proto bpf_tail_call_proto; 566 extern const struct bpf_func_proto bpf_ktime_get_ns_proto; 567 extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto; 568 extern const struct bpf_func_proto bpf_get_current_uid_gid_proto; 569 extern const struct bpf_func_proto bpf_get_current_comm_proto; 570 extern const struct bpf_func_proto bpf_skb_vlan_push_proto; 571 extern const struct bpf_func_proto bpf_skb_vlan_pop_proto; 572 extern const struct bpf_func_proto bpf_get_stackid_proto; 573 extern const struct bpf_func_proto bpf_sock_map_update_proto; 574 575 /* Shared helpers among cBPF and eBPF. */ 576 void bpf_user_rnd_init_once(void); 577 u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 578 579 #endif /* _LINUX_BPF_H */ 580