include/linux/bpf.h

   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         char 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_prog_ops {
 161         int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr,
 162                         union bpf_attr __user *uattr);
 163 };
 164
 165 struct bpf_verifier_ops {
 166         /* return eBPF function prototype for verification */
 167         const struct bpf_func_proto *(*get_func_proto)(enum bpf_func_id func_id);
 168
 169         /* return true if 'size' wide access at offset 'off' within bpf_context
 170          * with 'type' (read or write) is allowed
 171          */
 172         bool (*is_valid_access)(int off, int size, enum bpf_access_type type,
 173                                 struct bpf_insn_access_aux *info);
 174         int (*gen_prologue)(struct bpf_insn *insn, bool direct_write,
 175                             const struct bpf_prog *prog);
 176         u32 (*convert_ctx_access)(enum bpf_access_type type,
 177                                   const struct bpf_insn *src,
 178                                   struct bpf_insn *dst,
 179                                   struct bpf_prog *prog, u32 *target_size);
 180 };
 181
 182 struct bpf_prog_aux {
 183         atomic_t refcnt;
 184         u32 used_map_cnt;
 185         u32 max_ctx_offset;
 186         u32 stack_depth;
 187         u32 id;
 188         struct latch_tree_node ksym_tnode;
 189         struct list_head ksym_lnode;
 190         const struct bpf_prog_ops *ops;
 191         const struct bpf_verifier_ops *vops;
 192         struct bpf_map **used_maps;
 193         struct bpf_prog *prog;
 194         struct user_struct *user;
 195         u64 load_time; /* ns since boottime */
 196         char name[BPF_OBJ_NAME_LEN];
 197         union {
 198                 struct work_struct work;
 199                 struct rcu_head rcu;
 200         };
 201 };
 202
 203 struct bpf_array {
 204         struct bpf_map map;
 205         u32 elem_size;
 206         /* 'ownership' of prog_array is claimed by the first program that
 207          * is going to use this map or by the first program which FD is stored
 208          * in the map to make sure that all callers and callees have the same
 209          * prog_type and JITed flag
 210          */
 211         enum bpf_prog_type owner_prog_type;
 212         bool owner_jited;
 213         union {
 214                 char value[0] __aligned(8);
 215                 void *ptrs[0] __aligned(8);
 216                 void __percpu *pptrs[0] __aligned(8);
 217         };
 218 };
 219
 220 #define MAX_TAIL_CALL_CNT 32
 221
 222 struct bpf_event_entry {
 223         struct perf_event *event;
 224         struct file *perf_file;
 225         struct file *map_file;
 226         struct rcu_head rcu;
 227 };
 228
 229 u64 bpf_tail_call(u64 ctx, u64 r2, u64 index, u64 r4, u64 r5);
 230 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
 231
 232 bool bpf_prog_array_compatible(struct bpf_array *array, const struct bpf_prog *fp);
 233 int bpf_prog_calc_tag(struct bpf_prog *fp);
 234
 235 const struct bpf_func_proto *bpf_get_trace_printk_proto(void);
 236
 237 typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src,
 238                                         unsigned long off, unsigned long len);
 239
 240 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 241                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy);
 242
 243 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
 244                           union bpf_attr __user *uattr);
 245 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
 246                           union bpf_attr __user *uattr);
 247
 248 /* an array of programs to be executed under rcu_lock.
 249  *
 250  * Typical usage:
 251  * ret = BPF_PROG_RUN_ARRAY(&bpf_prog_array, ctx, BPF_PROG_RUN);
 252  *
 253  * the structure returned by bpf_prog_array_alloc() should be populated
 254  * with program pointers and the last pointer must be NULL.
 255  * The user has to keep refcnt on the program and make sure the program
 256  * is removed from the array before bpf_prog_put().
 257  * The 'struct bpf_prog_array *' should only be replaced with xchg()
 258  * since other cpus are walking the array of pointers in parallel.
 259  */
 260 struct bpf_prog_array {
 261         struct rcu_head rcu;
 262         struct bpf_prog *progs[0];
 263 };
 264
 265 struct bpf_prog_array __rcu *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags);
 266 void bpf_prog_array_free(struct bpf_prog_array __rcu *progs);
 267 int bpf_prog_array_length(struct bpf_prog_array __rcu *progs);
 268 int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *progs,
 269                                 __u32 __user *prog_ids, u32 cnt);
 270
 271 #define BPF_PROG_RUN_ARRAY(array, ctx, func)            \
 272         ({                                              \
 273                 struct bpf_prog **_prog;                \
 274                 u32 _ret = 1;                           \
 275                 rcu_read_lock();                        \
 276                 _prog = rcu_dereference(array)->progs;  \
 277                 for (; *_prog; _prog++)                 \
 278                         _ret &= func(*_prog, ctx);      \
 279                 rcu_read_unlock();                      \
 280                 _ret;                                   \
 281          })
 282
 283 #ifdef CONFIG_BPF_SYSCALL
 284 DECLARE_PER_CPU(int, bpf_prog_active);
 285
 286 #define BPF_PROG_TYPE(_id, _name) \
 287         extern const struct bpf_prog_ops _name ## _prog_ops; \
 288         extern const struct bpf_verifier_ops _name ## _verifier_ops;
 289 #define BPF_MAP_TYPE(_id, _ops) \
 290         extern const struct bpf_map_ops _ops;
 291 #include <linux/bpf_types.h>
 292 #undef BPF_PROG_TYPE
 293 #undef BPF_MAP_TYPE
 294
 295 struct bpf_prog *bpf_prog_get(u32 ufd);
 296 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type);
 297 struct bpf_prog * __must_check bpf_prog_add(struct bpf_prog *prog, int i);
 298 void bpf_prog_sub(struct bpf_prog *prog, int i);
 299 struct bpf_prog * __must_check bpf_prog_inc(struct bpf_prog *prog);
 300 struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog);
 301 void bpf_prog_put(struct bpf_prog *prog);
 302 int __bpf_prog_charge(struct user_struct *user, u32 pages);
 303 void __bpf_prog_uncharge(struct user_struct *user, u32 pages);
 304
 305 struct bpf_map *bpf_map_get_with_uref(u32 ufd);
 306 struct bpf_map *__bpf_map_get(struct fd f);
 307 struct bpf_map * __must_check bpf_map_inc(struct bpf_map *map, bool uref);
 308 void bpf_map_put_with_uref(struct bpf_map *map);
 309 void bpf_map_put(struct bpf_map *map);
 310 int bpf_map_precharge_memlock(u32 pages);
 311 void *bpf_map_area_alloc(size_t size, int numa_node);
 312 void bpf_map_area_free(void *base);
 313
 314 extern int sysctl_unprivileged_bpf_disabled;
 315
 316 int bpf_map_new_fd(struct bpf_map *map);
 317 int bpf_prog_new_fd(struct bpf_prog *prog);
 318
 319 int bpf_obj_pin_user(u32 ufd, const char __user *pathname);
 320 int bpf_obj_get_user(const char __user *pathname);
 321
 322 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value);
 323 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value);
 324 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
 325                            u64 flags);
 326 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
 327                             u64 flags);
 328
 329 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value);
 330
 331 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
 332                                  void *key, void *value, u64 map_flags);
 333 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value);
 334 void bpf_fd_array_map_clear(struct bpf_map *map);
 335 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
 336                                 void *key, void *value, u64 map_flags);
 337 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value);
 338
 339 /* memcpy that is used with 8-byte aligned pointers, power-of-8 size and
 340  * forced to use 'long' read/writes to try to atomically copy long counters.
 341  * Best-effort only.  No barriers here, since it _will_ race with concurrent
 342  * updates from BPF programs. Called from bpf syscall and mostly used with
 343  * size 8 or 16 bytes, so ask compiler to inline it.
 344  */
 345 static inline void bpf_long_memcpy(void *dst, const void *src, u32 size)
 346 {
 347         const long *lsrc = src;
 348         long *ldst = dst;
 349
 350         size /= sizeof(long);
 351         while (size--)
 352                 *ldst++ = *lsrc++;
 353 }
 354
 355 /* verify correctness of eBPF program */
 356 int bpf_check(struct bpf_prog **fp, union bpf_attr *attr);
 357
 358 /* Map specifics */
 359 struct net_device  *__dev_map_lookup_elem(struct bpf_map *map, u32 key);
 360 void __dev_map_insert_ctx(struct bpf_map *map, u32 index);
 361 void __dev_map_flush(struct bpf_map *map);
 362
 363 struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key);
 364 void __cpu_map_insert_ctx(struct bpf_map *map, u32 index);
 365 void __cpu_map_flush(struct bpf_map *map);
 366 struct xdp_buff;
 367 int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_buff *xdp,
 368                     struct net_device *dev_rx);
 369
 370 /* Return map's numa specified by userspace */
 371 static inline int bpf_map_attr_numa_node(const union bpf_attr *attr)
 372 {
 373         return (attr->map_flags & BPF_F_NUMA_NODE) ?
 374                 attr->numa_node : NUMA_NO_NODE;
 375 }
 376
 377 #else /* !CONFIG_BPF_SYSCALL */
 378 static inline struct bpf_prog *bpf_prog_get(u32 ufd)
 379 {
 380         return ERR_PTR(-EOPNOTSUPP);
 381 }
 382
 383 static inline struct bpf_prog *bpf_prog_get_type(u32 ufd,
 384                                                  enum bpf_prog_type type)
 385 {
 386         return ERR_PTR(-EOPNOTSUPP);
 387 }
 388 static inline struct bpf_prog * __must_check bpf_prog_add(struct bpf_prog *prog,
 389                                                           int i)
 390 {
 391         return ERR_PTR(-EOPNOTSUPP);
 392 }
 393
 394 static inline void bpf_prog_sub(struct bpf_prog *prog, int i)
 395 {
 396 }
 397
 398 static inline void bpf_prog_put(struct bpf_prog *prog)
 399 {
 400 }
 401
 402 static inline struct bpf_prog * __must_check bpf_prog_inc(struct bpf_prog *prog)
 403 {
 404         return ERR_PTR(-EOPNOTSUPP);
 405 }
 406
 407 static inline struct bpf_prog *__must_check
 408 bpf_prog_inc_not_zero(struct bpf_prog *prog)
 409 {
 410         return ERR_PTR(-EOPNOTSUPP);
 411 }
 412
 413 static inline int __bpf_prog_charge(struct user_struct *user, u32 pages)
 414 {
 415         return 0;
 416 }
 417
 418 static inline void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
 419 {
 420 }
 421
 422 static inline int bpf_obj_get_user(const char __user *pathname)
 423 {
 424         return -EOPNOTSUPP;
 425 }
 426
 427 static inline struct net_device  *__dev_map_lookup_elem(struct bpf_map *map,
 428                                                        u32 key)
 429 {
 430         return NULL;
 431 }
 432
 433 static inline void __dev_map_insert_ctx(struct bpf_map *map, u32 index)
 434 {
 435 }
 436
 437 static inline void __dev_map_flush(struct bpf_map *map)
 438 {
 439 }
 440
 441 static inline
 442 struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key)
 443 {
 444         return NULL;
 445 }
 446
 447 static inline void __cpu_map_insert_ctx(struct bpf_map *map, u32 index)
 448 {
 449 }
 450
 451 static inline void __cpu_map_flush(struct bpf_map *map)
 452 {
 453 }
 454
 455 struct xdp_buff;
 456 static inline int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu,
 457                                   struct xdp_buff *xdp,
 458                                   struct net_device *dev_rx)
 459 {
 460         return 0;
 461 }
 462 #endif /* CONFIG_BPF_SYSCALL */
 463
 464 #if defined(CONFIG_STREAM_PARSER) && defined(CONFIG_BPF_SYSCALL)
 465 struct sock  *__sock_map_lookup_elem(struct bpf_map *map, u32 key);
 466 int sock_map_prog(struct bpf_map *map, struct bpf_prog *prog, u32 type);
 467 #else
 468 static inline struct sock  *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 469 {
 470         return NULL;
 471 }
 472
 473 static inline int sock_map_prog(struct bpf_map *map,
 474                                 struct bpf_prog *prog,
 475                                 u32 type)
 476 {
 477         return -EOPNOTSUPP;
 478 }
 479 #endif
 480
 481 /* verifier prototypes for helper functions called from eBPF programs */
 482 extern const struct bpf_func_proto bpf_map_lookup_elem_proto;
 483 extern const struct bpf_func_proto bpf_map_update_elem_proto;
 484 extern const struct bpf_func_proto bpf_map_delete_elem_proto;
 485
 486 extern const struct bpf_func_proto bpf_get_prandom_u32_proto;
 487 extern const struct bpf_func_proto bpf_get_smp_processor_id_proto;
 488 extern const struct bpf_func_proto bpf_get_numa_node_id_proto;
 489 extern const struct bpf_func_proto bpf_tail_call_proto;
 490 extern const struct bpf_func_proto bpf_ktime_get_ns_proto;
 491 extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto;
 492 extern const struct bpf_func_proto bpf_get_current_uid_gid_proto;
 493 extern const struct bpf_func_proto bpf_get_current_comm_proto;
 494 extern const struct bpf_func_proto bpf_skb_vlan_push_proto;
 495 extern const struct bpf_func_proto bpf_skb_vlan_pop_proto;
 496 extern const struct bpf_func_proto bpf_get_stackid_proto;
 497 extern const struct bpf_func_proto bpf_sock_map_update_proto;
 498
 499 /* Shared helpers among cBPF and eBPF. */
 500 void bpf_user_rnd_init_once(void);
 501 u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
 502
 503 #endif /* _LINUX_BPF_H */