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 19 struct perf_event; 20 struct bpf_prog; 21 struct bpf_map; 22 23 /* map is generic key/value storage optionally accesible by eBPF programs */ 24 struct bpf_map_ops { 25 /* funcs callable from userspace (via syscall) */ 26 struct bpf_map *(*map_alloc)(union bpf_attr *attr); 27 void (*map_release)(struct bpf_map *map, struct file *map_file); 28 void (*map_free)(struct bpf_map *map); 29 int (*map_get_next_key)(struct bpf_map *map, void *key, void *next_key); 30 31 /* funcs callable from userspace and from eBPF programs */ 32 void *(*map_lookup_elem)(struct bpf_map *map, void *key); 33 int (*map_update_elem)(struct bpf_map *map, void *key, void *value, u64 flags); 34 int (*map_delete_elem)(struct bpf_map *map, void *key); 35 36 /* funcs called by prog_array and perf_event_array map */ 37 void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file, 38 int fd); 39 void (*map_fd_put_ptr)(void *ptr); 40 u32 (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf); 41 u32 (*map_fd_sys_lookup_elem)(void *ptr); 42 }; 43 44 struct bpf_map { 45 atomic_t refcnt; 46 enum bpf_map_type map_type; 47 u32 key_size; 48 u32 value_size; 49 u32 max_entries; 50 u32 map_flags; 51 u32 pages; 52 u32 id; 53 int numa_node; 54 struct user_struct *user; 55 const struct bpf_map_ops *ops; 56 struct work_struct work; 57 atomic_t usercnt; 58 struct bpf_map *inner_map_meta; 59 u8 name[BPF_OBJ_NAME_LEN]; 60 }; 61 62 /* function argument constraints */ 63 enum bpf_arg_type { 64 ARG_DONTCARE = 0, /* unused argument in helper function */ 65 66 /* the following constraints used to prototype 67 * bpf_map_lookup/update/delete_elem() functions 68 */ 69 ARG_CONST_MAP_PTR, /* const argument used as pointer to bpf_map */ 70 ARG_PTR_TO_MAP_KEY, /* pointer to stack used as map key */ 71 ARG_PTR_TO_MAP_VALUE, /* pointer to stack used as map value */ 72 73 /* the following constraints used to prototype bpf_memcmp() and other 74 * functions that access data on eBPF program stack 75 */ 76 ARG_PTR_TO_MEM, /* pointer to valid memory (stack, packet, map value) */ 77 ARG_PTR_TO_UNINIT_MEM, /* pointer to memory does not need to be initialized, 78 * helper function must fill all bytes or clear 79 * them in error case. 80 */ 81 82 ARG_CONST_SIZE, /* number of bytes accessed from memory */ 83 ARG_CONST_SIZE_OR_ZERO, /* number of bytes accessed from memory or 0 */ 84 85 ARG_PTR_TO_CTX, /* pointer to context */ 86 ARG_ANYTHING, /* any (initialized) argument is ok */ 87 }; 88 89 /* type of values returned from helper functions */ 90 enum bpf_return_type { 91 RET_INTEGER, /* function returns integer */ 92 RET_VOID, /* function doesn't return anything */ 93 RET_PTR_TO_MAP_VALUE_OR_NULL, /* returns a pointer to map elem value or NULL */ 94 }; 95 96 /* eBPF function prototype used by verifier to allow BPF_CALLs from eBPF programs 97 * to in-kernel helper functions and for adjusting imm32 field in BPF_CALL 98 * instructions after verifying 99 */ 100 struct bpf_func_proto { 101 u64 (*func)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 102 bool gpl_only; 103 bool pkt_access; 104 enum bpf_return_type ret_type; 105 enum bpf_arg_type arg1_type; 106 enum bpf_arg_type arg2_type; 107 enum bpf_arg_type arg3_type; 108 enum bpf_arg_type arg4_type; 109 enum bpf_arg_type arg5_type; 110 }; 111 112 /* bpf_context is intentionally undefined structure. Pointer to bpf_context is 113 * the first argument to eBPF programs. 114 * For socket filters: 'struct bpf_context *' == 'struct sk_buff *' 115 */ 116 struct bpf_context; 117 118 enum bpf_access_type { 119 BPF_READ = 1, 120 BPF_WRITE = 2 121 }; 122 123 /* types of values stored in eBPF registers */ 124 /* Pointer types represent: 125 * pointer 126 * pointer + imm 127 * pointer + (u16) var 128 * pointer + (u16) var + imm 129 * if (range > 0) then [ptr, ptr + range - off) is safe to access 130 * if (id > 0) means that some 'var' was added 131 * if (off > 0) means that 'imm' was added 132 */ 133 enum bpf_reg_type { 134 NOT_INIT = 0, /* nothing was written into register */ 135 SCALAR_VALUE, /* reg doesn't contain a valid pointer */ 136 PTR_TO_CTX, /* reg points to bpf_context */ 137 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */ 138 PTR_TO_MAP_VALUE, /* reg points to map element value */ 139 PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */ 140 PTR_TO_STACK, /* reg == frame_pointer + offset */ 141 PTR_TO_PACKET_META, /* skb->data - meta_len */ 142 PTR_TO_PACKET, /* reg points to skb->data */ 143 PTR_TO_PACKET_END, /* skb->data + headlen */ 144 }; 145 146 /* The information passed from prog-specific *_is_valid_access 147 * back to the verifier. 148 */ 149 struct bpf_insn_access_aux { 150 enum bpf_reg_type reg_type; 151 int ctx_field_size; 152 }; 153 154 static inline void 155 bpf_ctx_record_field_size(struct bpf_insn_access_aux *aux, u32 size) 156 { 157 aux->ctx_field_size = size; 158 } 159 160 struct bpf_verifier_ops { 161 /* return eBPF function prototype for verification */ 162 const struct bpf_func_proto *(*get_func_proto)(enum bpf_func_id func_id); 163 164 /* return true if 'size' wide access at offset 'off' within bpf_context 165 * with 'type' (read or write) is allowed 166 */ 167 bool (*is_valid_access)(int off, int size, enum bpf_access_type type, 168 struct bpf_insn_access_aux *info); 169 int (*gen_prologue)(struct bpf_insn *insn, bool direct_write, 170 const struct bpf_prog *prog); 171 u32 (*convert_ctx_access)(enum bpf_access_type type, 172 const struct bpf_insn *src, 173 struct bpf_insn *dst, 174 struct bpf_prog *prog, u32 *target_size); 175 int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr, 176 union bpf_attr __user *uattr); 177 }; 178 179 struct bpf_prog_aux { 180 atomic_t refcnt; 181 u32 used_map_cnt; 182 u32 max_ctx_offset; 183 u32 stack_depth; 184 u32 id; 185 struct latch_tree_node ksym_tnode; 186 struct list_head ksym_lnode; 187 const struct bpf_verifier_ops *ops; 188 struct bpf_map **used_maps; 189 struct bpf_prog *prog; 190 struct user_struct *user; 191 u64 load_time; /* ns since boottime */ 192 u8 name[BPF_OBJ_NAME_LEN]; 193 union { 194 struct work_struct work; 195 struct rcu_head rcu; 196 }; 197 }; 198 199 struct bpf_array { 200 struct bpf_map map; 201 u32 elem_size; 202 /* 'ownership' of prog_array is claimed by the first program that 203 * is going to use this map or by the first program which FD is stored 204 * in the map to make sure that all callers and callees have the same 205 * prog_type and JITed flag 206 */ 207 enum bpf_prog_type owner_prog_type; 208 bool owner_jited; 209 union { 210 char value[0] __aligned(8); 211 void *ptrs[0] __aligned(8); 212 void __percpu *pptrs[0] __aligned(8); 213 }; 214 }; 215 216 #define MAX_TAIL_CALL_CNT 32 217 218 struct bpf_event_entry { 219 struct perf_event *event; 220 struct file *perf_file; 221 struct file *map_file; 222 struct rcu_head rcu; 223 }; 224 225 u64 bpf_tail_call(u64 ctx, u64 r2, u64 index, u64 r4, u64 r5); 226 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 227 228 bool bpf_prog_array_compatible(struct bpf_array *array, const struct bpf_prog *fp); 229 int bpf_prog_calc_tag(struct bpf_prog *fp); 230 231 const struct bpf_func_proto *bpf_get_trace_printk_proto(void); 232 233 typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src, 234 unsigned long off, unsigned long len); 235 236 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 237 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy); 238 239 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr, 240 union bpf_attr __user *uattr); 241 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr, 242 union bpf_attr __user *uattr); 243 244 /* an array of programs to be executed under rcu_lock. 245 * 246 * Typical usage: 247 * ret = BPF_PROG_RUN_ARRAY(&bpf_prog_array, ctx, BPF_PROG_RUN); 248 * 249 * the structure returned by bpf_prog_array_alloc() should be populated 250 * with program pointers and the last pointer must be NULL. 251 * The user has to keep refcnt on the program and make sure the program 252 * is removed from the array before bpf_prog_put(). 253 * The 'struct bpf_prog_array *' should only be replaced with xchg() 254 * since other cpus are walking the array of pointers in parallel. 255 */ 256 struct bpf_prog_array { 257 struct rcu_head rcu; 258 struct bpf_prog *progs[0]; 259 }; 260 261 struct bpf_prog_array __rcu *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags); 262 void bpf_prog_array_free(struct bpf_prog_array __rcu *progs); 263 int bpf_prog_array_length(struct bpf_prog_array __rcu *progs); 264 int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *progs, 265 __u32 __user *prog_ids, u32 cnt); 266 267 #define BPF_PROG_RUN_ARRAY(array, ctx, func) \ 268 ({ \ 269 struct bpf_prog **_prog; \ 270 u32 _ret = 1; \ 271 rcu_read_lock(); \ 272 _prog = rcu_dereference(array)->progs; \ 273 for (; *_prog; _prog++) \ 274 _ret &= func(*_prog, ctx); \ 275 rcu_read_unlock(); \ 276 _ret; \ 277 }) 278 279 #ifdef CONFIG_BPF_SYSCALL 280 DECLARE_PER_CPU(int, bpf_prog_active); 281 282 #define BPF_PROG_TYPE(_id, _ops) \ 283 extern const struct bpf_verifier_ops _ops; 284 #define BPF_MAP_TYPE(_id, _ops) \ 285 extern const struct bpf_map_ops _ops; 286 #include <linux/bpf_types.h> 287 #undef BPF_PROG_TYPE 288 #undef BPF_MAP_TYPE 289 290 struct bpf_prog *bpf_prog_get(u32 ufd); 291 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type); 292 struct bpf_prog * __must_check bpf_prog_add(struct bpf_prog *prog, int i); 293 void bpf_prog_sub(struct bpf_prog *prog, int i); 294 struct bpf_prog * __must_check bpf_prog_inc(struct bpf_prog *prog); 295 struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog); 296 void bpf_prog_put(struct bpf_prog *prog); 297 int __bpf_prog_charge(struct user_struct *user, u32 pages); 298 void __bpf_prog_uncharge(struct user_struct *user, u32 pages); 299 300 struct bpf_map *bpf_map_get_with_uref(u32 ufd); 301 struct bpf_map *__bpf_map_get(struct fd f); 302 struct bpf_map * __must_check bpf_map_inc(struct bpf_map *map, bool uref); 303 void bpf_map_put_with_uref(struct bpf_map *map); 304 void bpf_map_put(struct bpf_map *map); 305 int bpf_map_precharge_memlock(u32 pages); 306 void *bpf_map_area_alloc(size_t size, int numa_node); 307 void bpf_map_area_free(void *base); 308 309 extern int sysctl_unprivileged_bpf_disabled; 310 311 int bpf_map_new_fd(struct bpf_map *map); 312 int bpf_prog_new_fd(struct bpf_prog *prog); 313 314 int bpf_obj_pin_user(u32 ufd, const char __user *pathname); 315 int bpf_obj_get_user(const char __user *pathname); 316 317 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value); 318 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value); 319 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value, 320 u64 flags); 321 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value, 322 u64 flags); 323 324 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value); 325 326 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file, 327 void *key, void *value, u64 map_flags); 328 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 329 void bpf_fd_array_map_clear(struct bpf_map *map); 330 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file, 331 void *key, void *value, u64 map_flags); 332 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 333 334 /* memcpy that is used with 8-byte aligned pointers, power-of-8 size and 335 * forced to use 'long' read/writes to try to atomically copy long counters. 336 * Best-effort only. No barriers here, since it _will_ race with concurrent 337 * updates from BPF programs. Called from bpf syscall and mostly used with 338 * size 8 or 16 bytes, so ask compiler to inline it. 339 */ 340 static inline void bpf_long_memcpy(void *dst, const void *src, u32 size) 341 { 342 const long *lsrc = src; 343 long *ldst = dst; 344 345 size /= sizeof(long); 346 while (size--) 347 *ldst++ = *lsrc++; 348 } 349 350 /* verify correctness of eBPF program */ 351 int bpf_check(struct bpf_prog **fp, union bpf_attr *attr); 352 353 /* Map specifics */ 354 struct net_device *__dev_map_lookup_elem(struct bpf_map *map, u32 key); 355 void __dev_map_insert_ctx(struct bpf_map *map, u32 index); 356 void __dev_map_flush(struct bpf_map *map); 357 358 /* Return map's numa specified by userspace */ 359 static inline int bpf_map_attr_numa_node(const union bpf_attr *attr) 360 { 361 return (attr->map_flags & BPF_F_NUMA_NODE) ? 362 attr->numa_node : NUMA_NO_NODE; 363 } 364 365 #else 366 static inline struct bpf_prog *bpf_prog_get(u32 ufd) 367 { 368 return ERR_PTR(-EOPNOTSUPP); 369 } 370 371 static inline struct bpf_prog *bpf_prog_get_type(u32 ufd, 372 enum bpf_prog_type type) 373 { 374 return ERR_PTR(-EOPNOTSUPP); 375 } 376 static inline struct bpf_prog * __must_check bpf_prog_add(struct bpf_prog *prog, 377 int i) 378 { 379 return ERR_PTR(-EOPNOTSUPP); 380 } 381 382 static inline void bpf_prog_sub(struct bpf_prog *prog, int i) 383 { 384 } 385 386 static inline void bpf_prog_put(struct bpf_prog *prog) 387 { 388 } 389 390 static inline struct bpf_prog * __must_check bpf_prog_inc(struct bpf_prog *prog) 391 { 392 return ERR_PTR(-EOPNOTSUPP); 393 } 394 395 static inline struct bpf_prog *__must_check 396 bpf_prog_inc_not_zero(struct bpf_prog *prog) 397 { 398 return ERR_PTR(-EOPNOTSUPP); 399 } 400 401 static inline int __bpf_prog_charge(struct user_struct *user, u32 pages) 402 { 403 return 0; 404 } 405 406 static inline void __bpf_prog_uncharge(struct user_struct *user, u32 pages) 407 { 408 } 409 410 static inline struct net_device *__dev_map_lookup_elem(struct bpf_map *map, 411 u32 key) 412 { 413 return NULL; 414 } 415 416 static inline void __dev_map_insert_ctx(struct bpf_map *map, u32 index) 417 { 418 } 419 420 static inline void __dev_map_flush(struct bpf_map *map) 421 { 422 } 423 #endif /* CONFIG_BPF_SYSCALL */ 424 425 #if defined(CONFIG_STREAM_PARSER) && defined(CONFIG_BPF_SYSCALL) 426 struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key); 427 int sock_map_prog(struct bpf_map *map, struct bpf_prog *prog, u32 type); 428 #else 429 static inline struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key) 430 { 431 return NULL; 432 } 433 434 static inline int sock_map_prog(struct bpf_map *map, 435 struct bpf_prog *prog, 436 u32 type) 437 { 438 return -EOPNOTSUPP; 439 } 440 #endif 441 442 /* verifier prototypes for helper functions called from eBPF programs */ 443 extern const struct bpf_func_proto bpf_map_lookup_elem_proto; 444 extern const struct bpf_func_proto bpf_map_update_elem_proto; 445 extern const struct bpf_func_proto bpf_map_delete_elem_proto; 446 447 extern const struct bpf_func_proto bpf_get_prandom_u32_proto; 448 extern const struct bpf_func_proto bpf_get_smp_processor_id_proto; 449 extern const struct bpf_func_proto bpf_get_numa_node_id_proto; 450 extern const struct bpf_func_proto bpf_tail_call_proto; 451 extern const struct bpf_func_proto bpf_ktime_get_ns_proto; 452 extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto; 453 extern const struct bpf_func_proto bpf_get_current_uid_gid_proto; 454 extern const struct bpf_func_proto bpf_get_current_comm_proto; 455 extern const struct bpf_func_proto bpf_skb_vlan_push_proto; 456 extern const struct bpf_func_proto bpf_skb_vlan_pop_proto; 457 extern const struct bpf_func_proto bpf_get_stackid_proto; 458 extern const struct bpf_func_proto bpf_sock_map_update_proto; 459 460 /* Shared helpers among cBPF and eBPF. */ 461 void bpf_user_rnd_init_once(void); 462 u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 463 464 #endif /* _LINUX_BPF_H */ 465