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         void (*map_release_uref)(struct bpf_map *map);
  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         /* 1st cacheline with read-mostly members of which some
  47          * are also accessed in fast-path (e.g. ops, max_entries).
  48          */
  49         const struct bpf_map_ops *ops ____cacheline_aligned;
  50         struct bpf_map *inner_map_meta;
  51 #ifdef CONFIG_SECURITY
  52         void *security;
  53 #endif
  54         enum bpf_map_type map_type;
  55         u32 key_size;
  56         u32 value_size;
  57         u32 max_entries;
  58         u32 map_flags;
  59         u32 pages;
  60         u32 id;
  61         int numa_node;
  62         bool unpriv_array;
  63         /* 7 bytes hole */
  64
  65         /* 2nd cacheline with misc members to avoid false sharing
  66          * particularly with refcounting.
  67          */
  68         struct user_struct *user ____cacheline_aligned;
  69         atomic_t refcnt;
  70         atomic_t usercnt;
  71         struct work_struct work;
  72 };
  73
  74 /* function argument constraints */
  75 enum bpf_arg_type {
  76         ARG_DONTCARE = 0,       /* unused argument in helper function */
  77
  78         /* the following constraints used to prototype
  79          * bpf_map_lookup/update/delete_elem() functions
  80          */
  81         ARG_CONST_MAP_PTR,      /* const argument used as pointer to bpf_map */
  82         ARG_PTR_TO_MAP_KEY,     /* pointer to stack used as map key */
  83         ARG_PTR_TO_MAP_VALUE,   /* pointer to stack used as map value */
  84
  85         /* the following constraints used to prototype bpf_memcmp() and other
  86          * functions that access data on eBPF program stack
  87          */
  88         ARG_PTR_TO_MEM,         /* pointer to valid memory (stack, packet, map value) */
  89         ARG_PTR_TO_UNINIT_MEM,  /* pointer to memory does not need to be initialized,
  90                                  * helper function must fill all bytes or clear
  91                                  * them in error case.
  92                                  */
  93
  94         ARG_CONST_SIZE,         /* number of bytes accessed from memory */
  95         ARG_CONST_SIZE_OR_ZERO, /* number of bytes accessed from memory or 0 */
  96
  97         ARG_PTR_TO_CTX,         /* pointer to context */
  98         ARG_ANYTHING,           /* any (initialized) argument is ok */
  99 };
 100
 101 /* type of values returned from helper functions */
 102 enum bpf_return_type {
 103         RET_INTEGER,                    /* function returns integer */
 104         RET_VOID,                       /* function doesn't return anything */
 105         RET_PTR_TO_MAP_VALUE_OR_NULL,   /* returns a pointer to map elem value or NULL */
 106 };
 107
 108 /* eBPF function prototype used by verifier to allow BPF_CALLs from eBPF programs
 109  * to in-kernel helper functions and for adjusting imm32 field in BPF_CALL
 110  * instructions after verifying
 111  */
 112 struct bpf_func_proto {
 113         u64 (*func)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
 114         bool gpl_only;
 115         bool pkt_access;
 116         enum bpf_return_type ret_type;
 117         enum bpf_arg_type arg1_type;
 118         enum bpf_arg_type arg2_type;
 119         enum bpf_arg_type arg3_type;
 120         enum bpf_arg_type arg4_type;
 121         enum bpf_arg_type arg5_type;
 122 };
 123
 124 /* bpf_context is intentionally undefined structure. Pointer to bpf_context is
 125  * the first argument to eBPF programs.
 126  * For socket filters: 'struct bpf_context *' == 'struct sk_buff *'
 127  */
 128 struct bpf_context;
 129
 130 enum bpf_access_type {
 131         BPF_READ = 1,
 132         BPF_WRITE = 2
 133 };
 134
 135 /* types of values stored in eBPF registers */
 136 /* Pointer types represent:
 137  * pointer
 138  * pointer + imm
 139  * pointer + (u16) var
 140  * pointer + (u16) var + imm
 141  * if (range > 0) then [ptr, ptr + range - off) is safe to access
 142  * if (id > 0) means that some 'var' was added
 143  * if (off > 0) means that 'imm' was added
 144  */
 145 enum bpf_reg_type {
 146         NOT_INIT = 0,            /* nothing was written into register */
 147         SCALAR_VALUE,            /* reg doesn't contain a valid pointer */
 148         PTR_TO_CTX,              /* reg points to bpf_context */
 149         CONST_PTR_TO_MAP,        /* reg points to struct bpf_map */
 150         PTR_TO_MAP_VALUE,        /* reg points to map element value */
 151         PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */
 152         PTR_TO_STACK,            /* reg == frame_pointer + offset */
 153         PTR_TO_PACKET,           /* reg points to skb->data */
 154         PTR_TO_PACKET_END,       /* skb->data + headlen */
 155 };
 156
 157 /* The information passed from prog-specific *_is_valid_access
 158  * back to the verifier.
 159  */
 160 struct bpf_insn_access_aux {
 161         enum bpf_reg_type reg_type;
 162         int ctx_field_size;
 163 };
 164
 165 static inline void
 166 bpf_ctx_record_field_size(struct bpf_insn_access_aux *aux, u32 size)
 167 {
 168         aux->ctx_field_size = size;
 169 }
 170
 171 struct bpf_verifier_ops {
 172         /* return eBPF function prototype for verification */
 173         const struct bpf_func_proto *(*get_func_proto)(enum bpf_func_id func_id);
 174
 175         /* return true if 'size' wide access at offset 'off' within bpf_context
 176          * with 'type' (read or write) is allowed
 177          */
 178         bool (*is_valid_access)(int off, int size, enum bpf_access_type type,
 179                                 struct bpf_insn_access_aux *info);
 180         int (*gen_prologue)(struct bpf_insn *insn, bool direct_write,
 181                             const struct bpf_prog *prog);
 182         u32 (*convert_ctx_access)(enum bpf_access_type type,
 183                                   const struct bpf_insn *src,
 184                                   struct bpf_insn *dst,
 185                                   struct bpf_prog *prog, u32 *target_size);
 186         int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr,
 187                         union bpf_attr __user *uattr);
 188 };
 189
 190 struct bpf_prog_aux {
 191         atomic_t refcnt;
 192         u32 used_map_cnt;
 193         u32 max_ctx_offset;
 194         u32 stack_depth;
 195         u32 id;
 196         struct latch_tree_node ksym_tnode;
 197         struct list_head ksym_lnode;
 198         const struct bpf_verifier_ops *ops;
 199         struct bpf_map **used_maps;
 200         struct bpf_prog *prog;
 201         struct user_struct *user;
 202         union {
 203                 struct work_struct work;
 204                 struct rcu_head rcu;
 205         };
 206 };
 207
 208 struct bpf_array {
 209         struct bpf_map map;
 210         u32 elem_size;
 211         u32 index_mask;
 212         /* 'ownership' of prog_array is claimed by the first program that
 213          * is going to use this map or by the first program which FD is stored
 214          * in the map to make sure that all callers and callees have the same
 215          * prog_type and JITed flag
 216          */
 217         enum bpf_prog_type owner_prog_type;
 218         bool owner_jited;
 219         union {
 220                 char value[0] __aligned(8);
 221                 void *ptrs[0] __aligned(8);
 222                 void __percpu *pptrs[0] __aligned(8);
 223         };
 224 };
 225
 226 #define MAX_TAIL_CALL_CNT 32
 227
 228 struct bpf_event_entry {
 229         struct perf_event *event;
 230         struct file *perf_file;
 231         struct file *map_file;
 232         struct rcu_head rcu;
 233 };
 234
 235 u64 bpf_tail_call(u64 ctx, u64 r2, u64 index, u64 r4, u64 r5);
 236 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
 237
 238 bool bpf_prog_array_compatible(struct bpf_array *array, const struct bpf_prog *fp);
 239 int bpf_prog_calc_tag(struct bpf_prog *fp);
 240
 241 const struct bpf_func_proto *bpf_get_trace_printk_proto(void);
 242
 243 typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src,
 244                                         unsigned long off, unsigned long len);
 245
 246 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 247                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy);
 248
 249 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
 250                           union bpf_attr __user *uattr);
 251 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
 252                           union bpf_attr __user *uattr);
 253
 254 #ifdef CONFIG_BPF_SYSCALL
 255 DECLARE_PER_CPU(int, bpf_prog_active);
 256
 257 #define BPF_PROG_TYPE(_id, _ops) \
 258         extern const struct bpf_verifier_ops _ops;
 259 #define BPF_MAP_TYPE(_id, _ops) \
 260         extern const struct bpf_map_ops _ops;
 261 #include <linux/bpf_types.h>
 262 #undef BPF_PROG_TYPE
 263 #undef BPF_MAP_TYPE
 264
 265 struct bpf_prog *bpf_prog_get(u32 ufd);
 266 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type);
 267 struct bpf_prog * __must_check bpf_prog_add(struct bpf_prog *prog, int i);
 268 void bpf_prog_sub(struct bpf_prog *prog, int i);
 269 struct bpf_prog * __must_check bpf_prog_inc(struct bpf_prog *prog);
 270 struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog);
 271 void bpf_prog_put(struct bpf_prog *prog);
 272 int __bpf_prog_charge(struct user_struct *user, u32 pages);
 273 void __bpf_prog_uncharge(struct user_struct *user, u32 pages);
 274
 275 struct bpf_map *bpf_map_get_with_uref(u32 ufd);
 276 struct bpf_map *__bpf_map_get(struct fd f);
 277 struct bpf_map * __must_check bpf_map_inc(struct bpf_map *map, bool uref);
 278 void bpf_map_put_with_uref(struct bpf_map *map);
 279 void bpf_map_put(struct bpf_map *map);
 280 int bpf_map_precharge_memlock(u32 pages);
 281 void *bpf_map_area_alloc(size_t size, int numa_node);
 282 void bpf_map_area_free(void *base);
 283
 284 extern int sysctl_unprivileged_bpf_disabled;
 285
 286 int bpf_map_new_fd(struct bpf_map *map);
 287 int bpf_prog_new_fd(struct bpf_prog *prog);
 288
 289 int bpf_obj_pin_user(u32 ufd, const char __user *pathname);
 290 int bpf_obj_get_user(const char __user *pathname);
 291
 292 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value);
 293 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value);
 294 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
 295                            u64 flags);
 296 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
 297                             u64 flags);
 298
 299 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value);
 300
 301 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
 302                                  void *key, void *value, u64 map_flags);
 303 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value);
 304 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
 305                                 void *key, void *value, u64 map_flags);
 306 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value);
 307
 308 /* memcpy that is used with 8-byte aligned pointers, power-of-8 size and
 309  * forced to use 'long' read/writes to try to atomically copy long counters.
 310  * Best-effort only.  No barriers here, since it _will_ race with concurrent
 311  * updates from BPF programs. Called from bpf syscall and mostly used with
 312  * size 8 or 16 bytes, so ask compiler to inline it.
 313  */
 314 static inline void bpf_long_memcpy(void *dst, const void *src, u32 size)
 315 {
 316         const long *lsrc = src;
 317         long *ldst = dst;
 318
 319         size /= sizeof(long);
 320         while (size--)
 321                 *ldst++ = *lsrc++;
 322 }
 323
 324 /* verify correctness of eBPF program */
 325 int bpf_check(struct bpf_prog **fp, union bpf_attr *attr);
 326
 327 /* Map specifics */
 328 struct net_device  *__dev_map_lookup_elem(struct bpf_map *map, u32 key);
 329 void __dev_map_insert_ctx(struct bpf_map *map, u32 index);
 330 void __dev_map_flush(struct bpf_map *map);
 331
 332 /* Return map's numa specified by userspace */
 333 static inline int bpf_map_attr_numa_node(const union bpf_attr *attr)
 334 {
 335         return (attr->map_flags & BPF_F_NUMA_NODE) ?
 336                 attr->numa_node : NUMA_NO_NODE;
 337 }
 338
 339 #else
 340 static inline struct bpf_prog *bpf_prog_get(u32 ufd)
 341 {
 342         return ERR_PTR(-EOPNOTSUPP);
 343 }
 344
 345 static inline struct bpf_prog *bpf_prog_get_type(u32 ufd,
 346                                                  enum bpf_prog_type type)
 347 {
 348         return ERR_PTR(-EOPNOTSUPP);
 349 }
 350 static inline struct bpf_prog * __must_check bpf_prog_add(struct bpf_prog *prog,
 351                                                           int i)
 352 {
 353         return ERR_PTR(-EOPNOTSUPP);
 354 }
 355
 356 static inline void bpf_prog_sub(struct bpf_prog *prog, int i)
 357 {
 358 }
 359
 360 static inline void bpf_prog_put(struct bpf_prog *prog)
 361 {
 362 }
 363
 364 static inline struct bpf_prog * __must_check bpf_prog_inc(struct bpf_prog *prog)
 365 {
 366         return ERR_PTR(-EOPNOTSUPP);
 367 }
 368
 369 static inline struct bpf_prog *__must_check
 370 bpf_prog_inc_not_zero(struct bpf_prog *prog)
 371 {
 372         return ERR_PTR(-EOPNOTSUPP);
 373 }
 374
 375 static inline int __bpf_prog_charge(struct user_struct *user, u32 pages)
 376 {
 377         return 0;
 378 }
 379
 380 static inline void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
 381 {
 382 }
 383
 384 static inline int bpf_obj_get_user(const char __user *pathname)
 385 {
 386         return -EOPNOTSUPP;
 387 }
 388
 389 static inline struct net_device  *__dev_map_lookup_elem(struct bpf_map *map,
 390                                                        u32 key)
 391 {
 392         return NULL;
 393 }
 394
 395 static inline void __dev_map_insert_ctx(struct bpf_map *map, u32 index)
 396 {
 397 }
 398
 399 static inline void __dev_map_flush(struct bpf_map *map)
 400 {
 401 }
 402 #endif /* CONFIG_BPF_SYSCALL */
 403
 404 #if defined(CONFIG_STREAM_PARSER) && defined(CONFIG_BPF_SYSCALL)
 405 struct sock  *__sock_map_lookup_elem(struct bpf_map *map, u32 key);
 406 int sock_map_prog(struct bpf_map *map, struct bpf_prog *prog, u32 type);
 407 #else
 408 static inline struct sock  *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 409 {
 410         return NULL;
 411 }
 412
 413 static inline int sock_map_prog(struct bpf_map *map,
 414                                 struct bpf_prog *prog,
 415                                 u32 type)
 416 {
 417         return -EOPNOTSUPP;
 418 }
 419 #endif
 420
 421 /* verifier prototypes for helper functions called from eBPF programs */
 422 extern const struct bpf_func_proto bpf_map_lookup_elem_proto;
 423 extern const struct bpf_func_proto bpf_map_update_elem_proto;
 424 extern const struct bpf_func_proto bpf_map_delete_elem_proto;
 425
 426 extern const struct bpf_func_proto bpf_get_prandom_u32_proto;
 427 extern const struct bpf_func_proto bpf_get_smp_processor_id_proto;
 428 extern const struct bpf_func_proto bpf_get_numa_node_id_proto;
 429 extern const struct bpf_func_proto bpf_tail_call_proto;
 430 extern const struct bpf_func_proto bpf_ktime_get_ns_proto;
 431 extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto;
 432 extern const struct bpf_func_proto bpf_get_current_uid_gid_proto;
 433 extern const struct bpf_func_proto bpf_get_current_comm_proto;
 434 extern const struct bpf_func_proto bpf_skb_vlan_push_proto;
 435 extern const struct bpf_func_proto bpf_skb_vlan_pop_proto;
 436 extern const struct bpf_func_proto bpf_get_stackid_proto;
 437 extern const struct bpf_func_proto bpf_sock_map_update_proto;
 438
 439 /* Shared helpers among cBPF and eBPF. */
 440 void bpf_user_rnd_init_once(void);
 441 u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
 442
 443 #endif /* _LINUX_BPF_H */