include/linux/filter.h

   1 /*
   2  * Linux Socket Filter Data Structures
   3  */
   4 #ifndef __LINUX_FILTER_H__
   5 #define __LINUX_FILTER_H__
   6
   7 #include <stdarg.h>
   8
   9 #include <linux/atomic.h>
  10 #include <linux/refcount.h>
  11 #include <linux/compat.h>
  12 #include <linux/skbuff.h>
  13 #include <linux/linkage.h>
  14 #include <linux/printk.h>
  15 #include <linux/workqueue.h>
  16 #include <linux/sched.h>
  17 #include <linux/capability.h>
  18 #include <linux/cryptohash.h>
  19
  20 #include <net/sch_generic.h>
  21
  22 #ifdef CONFIG_ARCH_HAS_SET_MEMORY
  23 #include <asm/set_memory.h>
  24 #endif
  25
  26 #include <uapi/linux/filter.h>
  27 #include <uapi/linux/bpf.h>
  28
  29 struct sk_buff;
  30 struct sock;
  31 struct seccomp_data;
  32 struct bpf_prog_aux;
  33
  34 /* ArgX, context and stack frame pointer register positions. Note,
  35  * Arg1, Arg2, Arg3, etc are used as argument mappings of function
  36  * calls in BPF_CALL instruction.
  37  */
  38 #define BPF_REG_ARG1    BPF_REG_1
  39 #define BPF_REG_ARG2    BPF_REG_2
  40 #define BPF_REG_ARG3    BPF_REG_3
  41 #define BPF_REG_ARG4    BPF_REG_4
  42 #define BPF_REG_ARG5    BPF_REG_5
  43 #define BPF_REG_CTX     BPF_REG_6
  44 #define BPF_REG_FP      BPF_REG_10
  45
  46 /* Additional register mappings for converted user programs. */
  47 #define BPF_REG_A       BPF_REG_0
  48 #define BPF_REG_X       BPF_REG_7
  49 #define BPF_REG_TMP     BPF_REG_8
  50
  51 /* Kernel hidden auxiliary/helper register for hardening step.
  52  * Only used by eBPF JITs. It's nothing more than a temporary
  53  * register that JITs use internally, only that here it's part
  54  * of eBPF instructions that have been rewritten for blinding
  55  * constants. See JIT pre-step in bpf_jit_blind_constants().
  56  */
  57 #define BPF_REG_AX              MAX_BPF_REG
  58 #define MAX_BPF_JIT_REG         (MAX_BPF_REG + 1)
  59
  60 /* As per nm, we expose JITed images as text (code) section for
  61  * kallsyms. That way, tools like perf can find it to match
  62  * addresses.
  63  */
  64 #define BPF_SYM_ELF_TYPE        't'
  65
  66 /* BPF program can access up to 512 bytes of stack space. */
  67 #define MAX_BPF_STACK   512
  68
  69 #define BPF_TAG_SIZE    8
  70
  71 /* Helper macros for filter block array initializers. */
  72
  73 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
  74
  75 #define BPF_ALU64_REG(OP, DST, SRC)                             \
  76         ((struct bpf_insn) {                                    \
  77                 .code  = BPF_ALU64 | BPF_OP(OP) | BPF_X,        \
  78                 .dst_reg = DST,                                 \
  79                 .src_reg = SRC,                                 \
  80                 .off   = 0,                                     \
  81                 .imm   = 0 })
  82
  83 #define BPF_ALU32_REG(OP, DST, SRC)                             \
  84         ((struct bpf_insn) {                                    \
  85                 .code  = BPF_ALU | BPF_OP(OP) | BPF_X,          \
  86                 .dst_reg = DST,                                 \
  87                 .src_reg = SRC,                                 \
  88                 .off   = 0,                                     \
  89                 .imm   = 0 })
  90
  91 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
  92
  93 #define BPF_ALU64_IMM(OP, DST, IMM)                             \
  94         ((struct bpf_insn) {                                    \
  95                 .code  = BPF_ALU64 | BPF_OP(OP) | BPF_K,        \
  96                 .dst_reg = DST,                                 \
  97                 .src_reg = 0,                                   \
  98                 .off   = 0,                                     \
  99                 .imm   = IMM })
 100
 101 #define BPF_ALU32_IMM(OP, DST, IMM)                             \
 102         ((struct bpf_insn) {                                    \
 103                 .code  = BPF_ALU | BPF_OP(OP) | BPF_K,          \
 104                 .dst_reg = DST,                                 \
 105                 .src_reg = 0,                                   \
 106                 .off   = 0,                                     \
 107                 .imm   = IMM })
 108
 109 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
 110
 111 #define BPF_ENDIAN(TYPE, DST, LEN)                              \
 112         ((struct bpf_insn) {                                    \
 113                 .code  = BPF_ALU | BPF_END | BPF_SRC(TYPE),     \
 114                 .dst_reg = DST,                                 \
 115                 .src_reg = 0,                                   \
 116                 .off   = 0,                                     \
 117                 .imm   = LEN })
 118
 119 /* Short form of mov, dst_reg = src_reg */
 120
 121 #define BPF_MOV64_REG(DST, SRC)                                 \
 122         ((struct bpf_insn) {                                    \
 123                 .code  = BPF_ALU64 | BPF_MOV | BPF_X,           \
 124                 .dst_reg = DST,                                 \
 125                 .src_reg = SRC,                                 \
 126                 .off   = 0,                                     \
 127                 .imm   = 0 })
 128
 129 #define BPF_MOV32_REG(DST, SRC)                                 \
 130         ((struct bpf_insn) {                                    \
 131                 .code  = BPF_ALU | BPF_MOV | BPF_X,             \
 132                 .dst_reg = DST,                                 \
 133                 .src_reg = SRC,                                 \
 134                 .off   = 0,                                     \
 135                 .imm   = 0 })
 136
 137 /* Short form of mov, dst_reg = imm32 */
 138
 139 #define BPF_MOV64_IMM(DST, IMM)                                 \
 140         ((struct bpf_insn) {                                    \
 141                 .code  = BPF_ALU64 | BPF_MOV | BPF_K,           \
 142                 .dst_reg = DST,                                 \
 143                 .src_reg = 0,                                   \
 144                 .off   = 0,                                     \
 145                 .imm   = IMM })
 146
 147 #define BPF_MOV32_IMM(DST, IMM)                                 \
 148         ((struct bpf_insn) {                                    \
 149                 .code  = BPF_ALU | BPF_MOV | BPF_K,             \
 150                 .dst_reg = DST,                                 \
 151                 .src_reg = 0,                                   \
 152                 .off   = 0,                                     \
 153                 .imm   = IMM })
 154
 155 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
 156 #define BPF_LD_IMM64(DST, IMM)                                  \
 157         BPF_LD_IMM64_RAW(DST, 0, IMM)
 158
 159 #define BPF_LD_IMM64_RAW(DST, SRC, IMM)                         \
 160         ((struct bpf_insn) {                                    \
 161                 .code  = BPF_LD | BPF_DW | BPF_IMM,             \
 162                 .dst_reg = DST,                                 \
 163                 .src_reg = SRC,                                 \
 164                 .off   = 0,                                     \
 165                 .imm   = (__u32) (IMM) }),                      \
 166         ((struct bpf_insn) {                                    \
 167                 .code  = 0, /* zero is reserved opcode */       \
 168                 .dst_reg = 0,                                   \
 169                 .src_reg = 0,                                   \
 170                 .off   = 0,                                     \
 171                 .imm   = ((__u64) (IMM)) >> 32 })
 172
 173 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
 174 #define BPF_LD_MAP_FD(DST, MAP_FD)                              \
 175         BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
 176
 177 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
 178
 179 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM)                      \
 180         ((struct bpf_insn) {                                    \
 181                 .code  = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE),   \
 182                 .dst_reg = DST,                                 \
 183                 .src_reg = SRC,                                 \
 184                 .off   = 0,                                     \
 185                 .imm   = IMM })
 186
 187 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM)                      \
 188         ((struct bpf_insn) {                                    \
 189                 .code  = BPF_ALU | BPF_MOV | BPF_SRC(TYPE),     \
 190                 .dst_reg = DST,                                 \
 191                 .src_reg = SRC,                                 \
 192                 .off   = 0,                                     \
 193                 .imm   = IMM })
 194
 195 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
 196
 197 #define BPF_LD_ABS(SIZE, IMM)                                   \
 198         ((struct bpf_insn) {                                    \
 199                 .code  = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS,     \
 200                 .dst_reg = 0,                                   \
 201                 .src_reg = 0,                                   \
 202                 .off   = 0,                                     \
 203                 .imm   = IMM })
 204
 205 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
 206
 207 #define BPF_LD_IND(SIZE, SRC, IMM)                              \
 208         ((struct bpf_insn) {                                    \
 209                 .code  = BPF_LD | BPF_SIZE(SIZE) | BPF_IND,     \
 210                 .dst_reg = 0,                                   \
 211                 .src_reg = SRC,                                 \
 212                 .off   = 0,                                     \
 213                 .imm   = IMM })
 214
 215 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
 216
 217 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF)                        \
 218         ((struct bpf_insn) {                                    \
 219                 .code  = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM,    \
 220                 .dst_reg = DST,                                 \
 221                 .src_reg = SRC,                                 \
 222                 .off   = OFF,                                   \
 223                 .imm   = 0 })
 224
 225 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
 226
 227 #define BPF_STX_MEM(SIZE, DST, SRC, OFF)                        \
 228         ((struct bpf_insn) {                                    \
 229                 .code  = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM,    \
 230                 .dst_reg = DST,                                 \
 231                 .src_reg = SRC,                                 \
 232                 .off   = OFF,                                   \
 233                 .imm   = 0 })
 234
 235 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
 236
 237 #define BPF_STX_XADD(SIZE, DST, SRC, OFF)                       \
 238         ((struct bpf_insn) {                                    \
 239                 .code  = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD,   \
 240                 .dst_reg = DST,                                 \
 241                 .src_reg = SRC,                                 \
 242                 .off   = OFF,                                   \
 243                 .imm   = 0 })
 244
 245 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
 246
 247 #define BPF_ST_MEM(SIZE, DST, OFF, IMM)                         \
 248         ((struct bpf_insn) {                                    \
 249                 .code  = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM,     \
 250                 .dst_reg = DST,                                 \
 251                 .src_reg = 0,                                   \
 252                 .off   = OFF,                                   \
 253                 .imm   = IMM })
 254
 255 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
 256
 257 #define BPF_JMP_REG(OP, DST, SRC, OFF)                          \
 258         ((struct bpf_insn) {                                    \
 259                 .code  = BPF_JMP | BPF_OP(OP) | BPF_X,          \
 260                 .dst_reg = DST,                                 \
 261                 .src_reg = SRC,                                 \
 262                 .off   = OFF,                                   \
 263                 .imm   = 0 })
 264
 265 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
 266
 267 #define BPF_JMP_IMM(OP, DST, IMM, OFF)                          \
 268         ((struct bpf_insn) {                                    \
 269                 .code  = BPF_JMP | BPF_OP(OP) | BPF_K,          \
 270                 .dst_reg = DST,                                 \
 271                 .src_reg = 0,                                   \
 272                 .off   = OFF,                                   \
 273                 .imm   = IMM })
 274
 275 /* Unconditional jumps, goto pc + off16 */
 276
 277 #define BPF_JMP_A(OFF)                                          \
 278         ((struct bpf_insn) {                                    \
 279                 .code  = BPF_JMP | BPF_JA,                      \
 280                 .dst_reg = 0,                                   \
 281                 .src_reg = 0,                                   \
 282                 .off   = OFF,                                   \
 283                 .imm   = 0 })
 284
 285 /* Function call */
 286
 287 #define BPF_EMIT_CALL(FUNC)                                     \
 288         ((struct bpf_insn) {                                    \
 289                 .code  = BPF_JMP | BPF_CALL,                    \
 290                 .dst_reg = 0,                                   \
 291                 .src_reg = 0,                                   \
 292                 .off   = 0,                                     \
 293                 .imm   = ((FUNC) - __bpf_call_base) })
 294
 295 /* Raw code statement block */
 296
 297 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM)                  \
 298         ((struct bpf_insn) {                                    \
 299                 .code  = CODE,                                  \
 300                 .dst_reg = DST,                                 \
 301                 .src_reg = SRC,                                 \
 302                 .off   = OFF,                                   \
 303                 .imm   = IMM })
 304
 305 /* Program exit */
 306
 307 #define BPF_EXIT_INSN()                                         \
 308         ((struct bpf_insn) {                                    \
 309                 .code  = BPF_JMP | BPF_EXIT,                    \
 310                 .dst_reg = 0,                                   \
 311                 .src_reg = 0,                                   \
 312                 .off   = 0,                                     \
 313                 .imm   = 0 })
 314
 315 /* Internal classic blocks for direct assignment */
 316
 317 #define __BPF_STMT(CODE, K)                                     \
 318         ((struct sock_filter) BPF_STMT(CODE, K))
 319
 320 #define __BPF_JUMP(CODE, K, JT, JF)                             \
 321         ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
 322
 323 #define bytes_to_bpf_size(bytes)                                \
 324 ({                                                              \
 325         int bpf_size = -EINVAL;                                 \
 326                                                                 \
 327         if (bytes == sizeof(u8))                                \
 328                 bpf_size = BPF_B;                               \
 329         else if (bytes == sizeof(u16))                          \
 330                 bpf_size = BPF_H;                               \
 331         else if (bytes == sizeof(u32))                          \
 332                 bpf_size = BPF_W;                               \
 333         else if (bytes == sizeof(u64))                          \
 334                 bpf_size = BPF_DW;                              \
 335                                                                 \
 336         bpf_size;                                               \
 337 })
 338
 339 #define BPF_SIZEOF(type)                                        \
 340         ({                                                      \
 341                 const int __size = bytes_to_bpf_size(sizeof(type)); \
 342                 BUILD_BUG_ON(__size < 0);                       \
 343                 __size;                                         \
 344         })
 345
 346 #define BPF_FIELD_SIZEOF(type, field)                           \
 347         ({                                                      \
 348                 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
 349                 BUILD_BUG_ON(__size < 0);                       \
 350                 __size;                                         \
 351         })
 352
 353 #define __BPF_MAP_0(m, v, ...) v
 354 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
 355 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
 356 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
 357 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
 358 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
 359
 360 #define __BPF_REG_0(...) __BPF_PAD(5)
 361 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
 362 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
 363 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
 364 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
 365 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
 366
 367 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
 368 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
 369
 370 #define __BPF_CAST(t, a)                                                       \
 371         (__force t)                                                            \
 372         (__force                                                               \
 373          typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long),      \
 374                                       (unsigned long)0, (t)0))) a
 375 #define __BPF_V void
 376 #define __BPF_N
 377
 378 #define __BPF_DECL_ARGS(t, a) t   a
 379 #define __BPF_DECL_REGS(t, a) u64 a
 380
 381 #define __BPF_PAD(n)                                                           \
 382         __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2,       \
 383                   u64, __ur_3, u64, __ur_4, u64, __ur_5)
 384
 385 #define BPF_CALL_x(x, name, ...)                                               \
 386         static __always_inline                                                 \
 387         u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__));   \
 388         u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__));         \
 389         u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__))          \
 390         {                                                                      \
 391                 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
 392         }                                                                      \
 393         static __always_inline                                                 \
 394         u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
 395
 396 #define BPF_CALL_0(name, ...)   BPF_CALL_x(0, name, __VA_ARGS__)
 397 #define BPF_CALL_1(name, ...)   BPF_CALL_x(1, name, __VA_ARGS__)
 398 #define BPF_CALL_2(name, ...)   BPF_CALL_x(2, name, __VA_ARGS__)
 399 #define BPF_CALL_3(name, ...)   BPF_CALL_x(3, name, __VA_ARGS__)
 400 #define BPF_CALL_4(name, ...)   BPF_CALL_x(4, name, __VA_ARGS__)
 401 #define BPF_CALL_5(name, ...)   BPF_CALL_x(5, name, __VA_ARGS__)
 402
 403 #ifdef CONFIG_COMPAT
 404 /* A struct sock_filter is architecture independent. */
 405 struct compat_sock_fprog {
 406         u16             len;
 407         compat_uptr_t   filter; /* struct sock_filter * */
 408 };
 409 #endif
 410
 411 struct sock_fprog_kern {
 412         u16                     len;
 413         struct sock_filter      *filter;
 414 };
 415
 416 struct bpf_binary_header {
 417         unsigned int pages;
 418         u8 image[];
 419 };
 420
 421 struct bpf_prog {
 422         u16                     pages;          /* Number of allocated pages */
 423         kmemcheck_bitfield_begin(meta);
 424         u16                     jited:1,        /* Is our filter JIT'ed? */
 425                                 locked:1,       /* Program image locked? */
 426                                 gpl_compatible:1, /* Is filter GPL compatible? */
 427                                 cb_access:1,    /* Is control block accessed? */
 428                                 dst_needed:1;   /* Do we need dst entry? */
 429         kmemcheck_bitfield_end(meta);
 430         enum bpf_prog_type      type;           /* Type of BPF program */
 431         u32                     len;            /* Number of filter blocks */
 432         u8                      tag[BPF_TAG_SIZE];
 433         struct bpf_prog_aux     *aux;           /* Auxiliary fields */
 434         struct sock_fprog_kern  *orig_prog;     /* Original BPF program */
 435         unsigned int            (*bpf_func)(const void *ctx,
 436                                             const struct bpf_insn *insn);
 437         /* Instructions for interpreter */
 438         union {
 439                 struct sock_filter      insns[0];
 440                 struct bpf_insn         insnsi[0];
 441         };
 442 };
 443
 444 struct sk_filter {
 445         refcount_t      refcnt;
 446         struct rcu_head rcu;
 447         struct bpf_prog *prog;
 448 };
 449
 450 #define BPF_PROG_RUN(filter, ctx)  (*filter->bpf_func)(ctx, filter->insnsi)
 451
 452 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
 453
 454 struct bpf_skb_data_end {
 455         struct qdisc_skb_cb qdisc_cb;
 456         void *data_end;
 457 };
 458
 459 struct xdp_buff {
 460         void *data;
 461         void *data_end;
 462         void *data_hard_start;
 463 };
 464
 465 /* compute the linear packet data range [data, data_end) which
 466  * will be accessed by cls_bpf, act_bpf and lwt programs
 467  */
 468 static inline void bpf_compute_data_end(struct sk_buff *skb)
 469 {
 470         struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
 471
 472         BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
 473         cb->data_end = skb->data + skb_headlen(skb);
 474 }
 475
 476 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
 477 {
 478         /* eBPF programs may read/write skb->cb[] area to transfer meta
 479          * data between tail calls. Since this also needs to work with
 480          * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
 481          *
 482          * In some socket filter cases, the cb unfortunately needs to be
 483          * saved/restored so that protocol specific skb->cb[] data won't
 484          * be lost. In any case, due to unpriviledged eBPF programs
 485          * attached to sockets, we need to clear the bpf_skb_cb() area
 486          * to not leak previous contents to user space.
 487          */
 488         BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
 489         BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
 490                      FIELD_SIZEOF(struct qdisc_skb_cb, data));
 491
 492         return qdisc_skb_cb(skb)->data;
 493 }
 494
 495 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
 496                                        struct sk_buff *skb)
 497 {
 498         u8 *cb_data = bpf_skb_cb(skb);
 499         u8 cb_saved[BPF_SKB_CB_LEN];
 500         u32 res;
 501
 502         if (unlikely(prog->cb_access)) {
 503                 memcpy(cb_saved, cb_data, sizeof(cb_saved));
 504                 memset(cb_data, 0, sizeof(cb_saved));
 505         }
 506
 507         res = BPF_PROG_RUN(prog, skb);
 508
 509         if (unlikely(prog->cb_access))
 510                 memcpy(cb_data, cb_saved, sizeof(cb_saved));
 511
 512         return res;
 513 }
 514
 515 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
 516                                         struct sk_buff *skb)
 517 {
 518         u8 *cb_data = bpf_skb_cb(skb);
 519
 520         if (unlikely(prog->cb_access))
 521                 memset(cb_data, 0, BPF_SKB_CB_LEN);
 522
 523         return BPF_PROG_RUN(prog, skb);
 524 }
 525
 526 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
 527                                             struct xdp_buff *xdp)
 528 {
 529         /* Caller needs to hold rcu_read_lock() (!), otherwise program
 530          * can be released while still running, or map elements could be
 531          * freed early while still having concurrent users. XDP fastpath
 532          * already takes rcu_read_lock() when fetching the program, so
 533          * it's not necessary here anymore.
 534          */
 535         return BPF_PROG_RUN(prog, xdp);
 536 }
 537
 538 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
 539 {
 540         return prog->len * sizeof(struct bpf_insn);
 541 }
 542
 543 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
 544 {
 545         return round_up(bpf_prog_insn_size(prog) +
 546                         sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
 547 }
 548
 549 static inline unsigned int bpf_prog_size(unsigned int proglen)
 550 {
 551         return max(sizeof(struct bpf_prog),
 552                    offsetof(struct bpf_prog, insns[proglen]));
 553 }
 554
 555 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
 556 {
 557         /* When classic BPF programs have been loaded and the arch
 558          * does not have a classic BPF JIT (anymore), they have been
 559          * converted via bpf_migrate_filter() to eBPF and thus always
 560          * have an unspec program type.
 561          */
 562         return prog->type == BPF_PROG_TYPE_UNSPEC;
 563 }
 564
 565 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
 566
 567 #ifdef CONFIG_ARCH_HAS_SET_MEMORY
 568 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
 569 {
 570         fp->locked = 1;
 571         WARN_ON_ONCE(set_memory_ro((unsigned long)fp, fp->pages));
 572 }
 573
 574 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
 575 {
 576         if (fp->locked) {
 577                 WARN_ON_ONCE(set_memory_rw((unsigned long)fp, fp->pages));
 578                 /* In case set_memory_rw() fails, we want to be the first
 579                  * to crash here instead of some random place later on.
 580                  */
 581                 fp->locked = 0;
 582         }
 583 }
 584
 585 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
 586 {
 587         WARN_ON_ONCE(set_memory_ro((unsigned long)hdr, hdr->pages));
 588 }
 589
 590 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
 591 {
 592         WARN_ON_ONCE(set_memory_rw((unsigned long)hdr, hdr->pages));
 593 }
 594 #else
 595 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
 596 {
 597 }
 598
 599 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
 600 {
 601 }
 602
 603 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
 604 {
 605 }
 606
 607 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
 608 {
 609 }
 610 #endif /* CONFIG_ARCH_HAS_SET_MEMORY */
 611
 612 static inline struct bpf_binary_header *
 613 bpf_jit_binary_hdr(const struct bpf_prog *fp)
 614 {
 615         unsigned long real_start = (unsigned long)fp->bpf_func;
 616         unsigned long addr = real_start & PAGE_MASK;
 617
 618         return (void *)addr;
 619 }
 620
 621 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
 622 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
 623 {
 624         return sk_filter_trim_cap(sk, skb, 1);
 625 }
 626
 627 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
 628 void bpf_prog_free(struct bpf_prog *fp);
 629
 630 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
 631 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
 632                                   gfp_t gfp_extra_flags);
 633 void __bpf_prog_free(struct bpf_prog *fp);
 634
 635 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
 636 {
 637         bpf_prog_unlock_ro(fp);
 638         __bpf_prog_free(fp);
 639 }
 640
 641 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
 642                                        unsigned int flen);
 643
 644 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
 645 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
 646                               bpf_aux_classic_check_t trans, bool save_orig);
 647 void bpf_prog_destroy(struct bpf_prog *fp);
 648
 649 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
 650 int sk_attach_bpf(u32 ufd, struct sock *sk);
 651 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
 652 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
 653 int sk_detach_filter(struct sock *sk);
 654 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
 655                   unsigned int len);
 656
 657 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
 658 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
 659
 660 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
 661
 662 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
 663 void bpf_jit_compile(struct bpf_prog *prog);
 664 bool bpf_helper_changes_pkt_data(void *func);
 665
 666 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 667                                        const struct bpf_insn *patch, u32 len);
 668 void bpf_warn_invalid_xdp_action(u32 act);
 669
 670 #ifdef CONFIG_BPF_JIT
 671 extern int bpf_jit_enable;
 672 extern int bpf_jit_harden;
 673 extern int bpf_jit_kallsyms;
 674
 675 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
 676
 677 struct bpf_binary_header *
 678 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
 679                      unsigned int alignment,
 680                      bpf_jit_fill_hole_t bpf_fill_ill_insns);
 681 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
 682
 683 void bpf_jit_free(struct bpf_prog *fp);
 684
 685 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
 686 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
 687
 688 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
 689                                 u32 pass, void *image)
 690 {
 691         pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
 692                proglen, pass, image, current->comm, task_pid_nr(current));
 693
 694         if (image)
 695                 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
 696                                16, 1, image, proglen, false);
 697 }
 698
 699 static inline bool bpf_jit_is_ebpf(void)
 700 {
 701 # ifdef CONFIG_HAVE_EBPF_JIT
 702         return true;
 703 # else
 704         return false;
 705 # endif
 706 }
 707
 708 static inline bool ebpf_jit_enabled(void)
 709 {
 710         return bpf_jit_enable && bpf_jit_is_ebpf();
 711 }
 712
 713 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
 714 {
 715         return fp->jited && bpf_jit_is_ebpf();
 716 }
 717
 718 static inline bool bpf_jit_blinding_enabled(void)
 719 {
 720         /* These are the prerequisites, should someone ever have the
 721          * idea to call blinding outside of them, we make sure to
 722          * bail out.
 723          */
 724         if (!bpf_jit_is_ebpf())
 725                 return false;
 726         if (!bpf_jit_enable)
 727                 return false;
 728         if (!bpf_jit_harden)
 729                 return false;
 730         if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
 731                 return false;
 732
 733         return true;
 734 }
 735
 736 static inline bool bpf_jit_kallsyms_enabled(void)
 737 {
 738         /* There are a couple of corner cases where kallsyms should
 739          * not be enabled f.e. on hardening.
 740          */
 741         if (bpf_jit_harden)
 742                 return false;
 743         if (!bpf_jit_kallsyms)
 744                 return false;
 745         if (bpf_jit_kallsyms == 1)
 746                 return true;
 747
 748         return false;
 749 }
 750
 751 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
 752                                  unsigned long *off, char *sym);
 753 bool is_bpf_text_address(unsigned long addr);
 754 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
 755                     char *sym);
 756
 757 static inline const char *
 758 bpf_address_lookup(unsigned long addr, unsigned long *size,
 759                    unsigned long *off, char **modname, char *sym)
 760 {
 761         const char *ret = __bpf_address_lookup(addr, size, off, sym);
 762
 763         if (ret && modname)
 764                 *modname = NULL;
 765         return ret;
 766 }
 767
 768 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
 769 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
 770
 771 #else /* CONFIG_BPF_JIT */
 772
 773 static inline bool ebpf_jit_enabled(void)
 774 {
 775         return false;
 776 }
 777
 778 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
 779 {
 780         return false;
 781 }
 782
 783 static inline void bpf_jit_free(struct bpf_prog *fp)
 784 {
 785         bpf_prog_unlock_free(fp);
 786 }
 787
 788 static inline bool bpf_jit_kallsyms_enabled(void)
 789 {
 790         return false;
 791 }
 792
 793 static inline const char *
 794 __bpf_address_lookup(unsigned long addr, unsigned long *size,
 795                      unsigned long *off, char *sym)
 796 {
 797         return NULL;
 798 }
 799
 800 static inline bool is_bpf_text_address(unsigned long addr)
 801 {
 802         return false;
 803 }
 804
 805 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
 806                                   char *type, char *sym)
 807 {
 808         return -ERANGE;
 809 }
 810
 811 static inline const char *
 812 bpf_address_lookup(unsigned long addr, unsigned long *size,
 813                    unsigned long *off, char **modname, char *sym)
 814 {
 815         return NULL;
 816 }
 817
 818 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
 819 {
 820 }
 821
 822 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
 823 {
 824 }
 825 #endif /* CONFIG_BPF_JIT */
 826
 827 #define BPF_ANC         BIT(15)
 828
 829 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
 830 {
 831         switch (first->code) {
 832         case BPF_RET | BPF_K:
 833         case BPF_LD | BPF_W | BPF_LEN:
 834                 return false;
 835
 836         case BPF_LD | BPF_W | BPF_ABS:
 837         case BPF_LD | BPF_H | BPF_ABS:
 838         case BPF_LD | BPF_B | BPF_ABS:
 839                 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
 840                         return true;
 841                 return false;
 842
 843         default:
 844                 return true;
 845         }
 846 }
 847
 848 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
 849 {
 850         BUG_ON(ftest->code & BPF_ANC);
 851
 852         switch (ftest->code) {
 853         case BPF_LD | BPF_W | BPF_ABS:
 854         case BPF_LD | BPF_H | BPF_ABS:
 855         case BPF_LD | BPF_B | BPF_ABS:
 856 #define BPF_ANCILLARY(CODE)     case SKF_AD_OFF + SKF_AD_##CODE:        \
 857                                 return BPF_ANC | SKF_AD_##CODE
 858                 switch (ftest->k) {
 859                 BPF_ANCILLARY(PROTOCOL);
 860                 BPF_ANCILLARY(PKTTYPE);
 861                 BPF_ANCILLARY(IFINDEX);
 862                 BPF_ANCILLARY(NLATTR);
 863                 BPF_ANCILLARY(NLATTR_NEST);
 864                 BPF_ANCILLARY(MARK);
 865                 BPF_ANCILLARY(QUEUE);
 866                 BPF_ANCILLARY(HATYPE);
 867                 BPF_ANCILLARY(RXHASH);
 868                 BPF_ANCILLARY(CPU);
 869                 BPF_ANCILLARY(ALU_XOR_X);
 870                 BPF_ANCILLARY(VLAN_TAG);
 871                 BPF_ANCILLARY(VLAN_TAG_PRESENT);
 872                 BPF_ANCILLARY(PAY_OFFSET);
 873                 BPF_ANCILLARY(RANDOM);
 874                 BPF_ANCILLARY(VLAN_TPID);
 875                 }
 876                 /* Fallthrough. */
 877         default:
 878                 return ftest->code;
 879         }
 880 }
 881
 882 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
 883                                            int k, unsigned int size);
 884
 885 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
 886                                      unsigned int size, void *buffer)
 887 {
 888         if (k >= 0)
 889                 return skb_header_pointer(skb, k, size, buffer);
 890
 891         return bpf_internal_load_pointer_neg_helper(skb, k, size);
 892 }
 893
 894 static inline int bpf_tell_extensions(void)
 895 {
 896         return SKF_AD_MAX;
 897 }
 898
 899 #endif /* __LINUX_FILTER_H__ */