include/linux/filter.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 /*
   3  * Linux Socket Filter Data Structures
   4  */
   5 #ifndef __LINUX_FILTER_H__
   6 #define __LINUX_FILTER_H__
   7
   8 #include <stdarg.h>
   9
  10 #include <linux/atomic.h>
  11 #include <linux/refcount.h>
  12 #include <linux/compat.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/linkage.h>
  15 #include <linux/printk.h>
  16 #include <linux/workqueue.h>
  17 #include <linux/sched.h>
  18 #include <linux/capability.h>
  19 #include <linux/cryptohash.h>
  20 #include <linux/set_memory.h>
  21
  22 #include <net/sch_generic.h>
  23
  24 #include <uapi/linux/filter.h>
  25 #include <uapi/linux/bpf.h>
  26
  27 struct sk_buff;
  28 struct sock;
  29 struct seccomp_data;
  30 struct bpf_prog_aux;
  31
  32 /* ArgX, context and stack frame pointer register positions. Note,
  33  * Arg1, Arg2, Arg3, etc are used as argument mappings of function
  34  * calls in BPF_CALL instruction.
  35  */
  36 #define BPF_REG_ARG1    BPF_REG_1
  37 #define BPF_REG_ARG2    BPF_REG_2
  38 #define BPF_REG_ARG3    BPF_REG_3
  39 #define BPF_REG_ARG4    BPF_REG_4
  40 #define BPF_REG_ARG5    BPF_REG_5
  41 #define BPF_REG_CTX     BPF_REG_6
  42 #define BPF_REG_FP      BPF_REG_10
  43
  44 /* Additional register mappings for converted user programs. */
  45 #define BPF_REG_A       BPF_REG_0
  46 #define BPF_REG_X       BPF_REG_7
  47 #define BPF_REG_TMP     BPF_REG_8
  48
  49 /* Kernel hidden auxiliary/helper register. */
  50 #define BPF_REG_AX              MAX_BPF_REG
  51 #define MAX_BPF_EXT_REG         (MAX_BPF_REG + 1)
  52 #define MAX_BPF_JIT_REG         MAX_BPF_EXT_REG
  53
  54 /* unused opcode to mark special call to bpf_tail_call() helper */
  55 #define BPF_TAIL_CALL   0xf0
  56
  57 /* As per nm, we expose JITed images as text (code) section for
  58  * kallsyms. That way, tools like perf can find it to match
  59  * addresses.
  60  */
  61 #define BPF_SYM_ELF_TYPE        't'
  62
  63 /* BPF program can access up to 512 bytes of stack space. */
  64 #define MAX_BPF_STACK   512
  65
  66 /* Helper macros for filter block array initializers. */
  67
  68 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
  69
  70 #define BPF_ALU64_REG(OP, DST, SRC)                             \
  71         ((struct bpf_insn) {                                    \
  72                 .code  = BPF_ALU64 | BPF_OP(OP) | BPF_X,        \
  73                 .dst_reg = DST,                                 \
  74                 .src_reg = SRC,                                 \
  75                 .off   = 0,                                     \
  76                 .imm   = 0 })
  77
  78 #define BPF_ALU32_REG(OP, DST, SRC)                             \
  79         ((struct bpf_insn) {                                    \
  80                 .code  = BPF_ALU | BPF_OP(OP) | BPF_X,          \
  81                 .dst_reg = DST,                                 \
  82                 .src_reg = SRC,                                 \
  83                 .off   = 0,                                     \
  84                 .imm   = 0 })
  85
  86 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
  87
  88 #define BPF_ALU64_IMM(OP, DST, IMM)                             \
  89         ((struct bpf_insn) {                                    \
  90                 .code  = BPF_ALU64 | BPF_OP(OP) | BPF_K,        \
  91                 .dst_reg = DST,                                 \
  92                 .src_reg = 0,                                   \
  93                 .off   = 0,                                     \
  94                 .imm   = IMM })
  95
  96 #define BPF_ALU32_IMM(OP, DST, IMM)                             \
  97         ((struct bpf_insn) {                                    \
  98                 .code  = BPF_ALU | BPF_OP(OP) | BPF_K,          \
  99                 .dst_reg = DST,                                 \
 100                 .src_reg = 0,                                   \
 101                 .off   = 0,                                     \
 102                 .imm   = IMM })
 103
 104 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
 105
 106 #define BPF_ENDIAN(TYPE, DST, LEN)                              \
 107         ((struct bpf_insn) {                                    \
 108                 .code  = BPF_ALU | BPF_END | BPF_SRC(TYPE),     \
 109                 .dst_reg = DST,                                 \
 110                 .src_reg = 0,                                   \
 111                 .off   = 0,                                     \
 112                 .imm   = LEN })
 113
 114 /* Short form of mov, dst_reg = src_reg */
 115
 116 #define BPF_MOV64_REG(DST, SRC)                                 \
 117         ((struct bpf_insn) {                                    \
 118                 .code  = BPF_ALU64 | BPF_MOV | BPF_X,           \
 119                 .dst_reg = DST,                                 \
 120                 .src_reg = SRC,                                 \
 121                 .off   = 0,                                     \
 122                 .imm   = 0 })
 123
 124 #define BPF_MOV32_REG(DST, SRC)                                 \
 125         ((struct bpf_insn) {                                    \
 126                 .code  = BPF_ALU | BPF_MOV | BPF_X,             \
 127                 .dst_reg = DST,                                 \
 128                 .src_reg = SRC,                                 \
 129                 .off   = 0,                                     \
 130                 .imm   = 0 })
 131
 132 /* Short form of mov, dst_reg = imm32 */
 133
 134 #define BPF_MOV64_IMM(DST, IMM)                                 \
 135         ((struct bpf_insn) {                                    \
 136                 .code  = BPF_ALU64 | BPF_MOV | BPF_K,           \
 137                 .dst_reg = DST,                                 \
 138                 .src_reg = 0,                                   \
 139                 .off   = 0,                                     \
 140                 .imm   = IMM })
 141
 142 #define BPF_MOV32_IMM(DST, IMM)                                 \
 143         ((struct bpf_insn) {                                    \
 144                 .code  = BPF_ALU | BPF_MOV | BPF_K,             \
 145                 .dst_reg = DST,                                 \
 146                 .src_reg = 0,                                   \
 147                 .off   = 0,                                     \
 148                 .imm   = IMM })
 149
 150 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
 151 #define BPF_LD_IMM64(DST, IMM)                                  \
 152         BPF_LD_IMM64_RAW(DST, 0, IMM)
 153
 154 #define BPF_LD_IMM64_RAW(DST, SRC, IMM)                         \
 155         ((struct bpf_insn) {                                    \
 156                 .code  = BPF_LD | BPF_DW | BPF_IMM,             \
 157                 .dst_reg = DST,                                 \
 158                 .src_reg = SRC,                                 \
 159                 .off   = 0,                                     \
 160                 .imm   = (__u32) (IMM) }),                      \
 161         ((struct bpf_insn) {                                    \
 162                 .code  = 0, /* zero is reserved opcode */       \
 163                 .dst_reg = 0,                                   \
 164                 .src_reg = 0,                                   \
 165                 .off   = 0,                                     \
 166                 .imm   = ((__u64) (IMM)) >> 32 })
 167
 168 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
 169 #define BPF_LD_MAP_FD(DST, MAP_FD)                              \
 170         BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
 171
 172 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
 173
 174 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM)                      \
 175         ((struct bpf_insn) {                                    \
 176                 .code  = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE),   \
 177                 .dst_reg = DST,                                 \
 178                 .src_reg = SRC,                                 \
 179                 .off   = 0,                                     \
 180                 .imm   = IMM })
 181
 182 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM)                      \
 183         ((struct bpf_insn) {                                    \
 184                 .code  = BPF_ALU | BPF_MOV | BPF_SRC(TYPE),     \
 185                 .dst_reg = DST,                                 \
 186                 .src_reg = SRC,                                 \
 187                 .off   = 0,                                     \
 188                 .imm   = IMM })
 189
 190 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
 191
 192 #define BPF_LD_ABS(SIZE, IMM)                                   \
 193         ((struct bpf_insn) {                                    \
 194                 .code  = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS,     \
 195                 .dst_reg = 0,                                   \
 196                 .src_reg = 0,                                   \
 197                 .off   = 0,                                     \
 198                 .imm   = IMM })
 199
 200 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
 201
 202 #define BPF_LD_IND(SIZE, SRC, IMM)                              \
 203         ((struct bpf_insn) {                                    \
 204                 .code  = BPF_LD | BPF_SIZE(SIZE) | BPF_IND,     \
 205                 .dst_reg = 0,                                   \
 206                 .src_reg = SRC,                                 \
 207                 .off   = 0,                                     \
 208                 .imm   = IMM })
 209
 210 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
 211
 212 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF)                        \
 213         ((struct bpf_insn) {                                    \
 214                 .code  = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM,    \
 215                 .dst_reg = DST,                                 \
 216                 .src_reg = SRC,                                 \
 217                 .off   = OFF,                                   \
 218                 .imm   = 0 })
 219
 220 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
 221
 222 #define BPF_STX_MEM(SIZE, DST, SRC, OFF)                        \
 223         ((struct bpf_insn) {                                    \
 224                 .code  = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM,    \
 225                 .dst_reg = DST,                                 \
 226                 .src_reg = SRC,                                 \
 227                 .off   = OFF,                                   \
 228                 .imm   = 0 })
 229
 230 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
 231
 232 #define BPF_STX_XADD(SIZE, DST, SRC, OFF)                       \
 233         ((struct bpf_insn) {                                    \
 234                 .code  = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD,   \
 235                 .dst_reg = DST,                                 \
 236                 .src_reg = SRC,                                 \
 237                 .off   = OFF,                                   \
 238                 .imm   = 0 })
 239
 240 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
 241
 242 #define BPF_ST_MEM(SIZE, DST, OFF, IMM)                         \
 243         ((struct bpf_insn) {                                    \
 244                 .code  = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM,     \
 245                 .dst_reg = DST,                                 \
 246                 .src_reg = 0,                                   \
 247                 .off   = OFF,                                   \
 248                 .imm   = IMM })
 249
 250 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
 251
 252 #define BPF_JMP_REG(OP, DST, SRC, OFF)                          \
 253         ((struct bpf_insn) {                                    \
 254                 .code  = BPF_JMP | BPF_OP(OP) | BPF_X,          \
 255                 .dst_reg = DST,                                 \
 256                 .src_reg = SRC,                                 \
 257                 .off   = OFF,                                   \
 258                 .imm   = 0 })
 259
 260 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
 261
 262 #define BPF_JMP_IMM(OP, DST, IMM, OFF)                          \
 263         ((struct bpf_insn) {                                    \
 264                 .code  = BPF_JMP | BPF_OP(OP) | BPF_K,          \
 265                 .dst_reg = DST,                                 \
 266                 .src_reg = 0,                                   \
 267                 .off   = OFF,                                   \
 268                 .imm   = IMM })
 269
 270 /* Unconditional jumps, goto pc + off16 */
 271
 272 #define BPF_JMP_A(OFF)                                          \
 273         ((struct bpf_insn) {                                    \
 274                 .code  = BPF_JMP | BPF_JA,                      \
 275                 .dst_reg = 0,                                   \
 276                 .src_reg = 0,                                   \
 277                 .off   = OFF,                                   \
 278                 .imm   = 0 })
 279
 280 /* Function call */
 281
 282 #define BPF_EMIT_CALL(FUNC)                                     \
 283         ((struct bpf_insn) {                                    \
 284                 .code  = BPF_JMP | BPF_CALL,                    \
 285                 .dst_reg = 0,                                   \
 286                 .src_reg = 0,                                   \
 287                 .off   = 0,                                     \
 288                 .imm   = ((FUNC) - __bpf_call_base) })
 289
 290 /* Raw code statement block */
 291
 292 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM)                  \
 293         ((struct bpf_insn) {                                    \
 294                 .code  = CODE,                                  \
 295                 .dst_reg = DST,                                 \
 296                 .src_reg = SRC,                                 \
 297                 .off   = OFF,                                   \
 298                 .imm   = IMM })
 299
 300 /* Program exit */
 301
 302 #define BPF_EXIT_INSN()                                         \
 303         ((struct bpf_insn) {                                    \
 304                 .code  = BPF_JMP | BPF_EXIT,                    \
 305                 .dst_reg = 0,                                   \
 306                 .src_reg = 0,                                   \
 307                 .off   = 0,                                     \
 308                 .imm   = 0 })
 309
 310 /* Internal classic blocks for direct assignment */
 311
 312 #define __BPF_STMT(CODE, K)                                     \
 313         ((struct sock_filter) BPF_STMT(CODE, K))
 314
 315 #define __BPF_JUMP(CODE, K, JT, JF)                             \
 316         ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
 317
 318 #define bytes_to_bpf_size(bytes)                                \
 319 ({                                                              \
 320         int bpf_size = -EINVAL;                                 \
 321                                                                 \
 322         if (bytes == sizeof(u8))                                \
 323                 bpf_size = BPF_B;                               \
 324         else if (bytes == sizeof(u16))                          \
 325                 bpf_size = BPF_H;                               \
 326         else if (bytes == sizeof(u32))                          \
 327                 bpf_size = BPF_W;                               \
 328         else if (bytes == sizeof(u64))                          \
 329                 bpf_size = BPF_DW;                              \
 330                                                                 \
 331         bpf_size;                                               \
 332 })
 333
 334 #define bpf_size_to_bytes(bpf_size)                             \
 335 ({                                                              \
 336         int bytes = -EINVAL;                                    \
 337                                                                 \
 338         if (bpf_size == BPF_B)                                  \
 339                 bytes = sizeof(u8);                             \
 340         else if (bpf_size == BPF_H)                             \
 341                 bytes = sizeof(u16);                            \
 342         else if (bpf_size == BPF_W)                             \
 343                 bytes = sizeof(u32);                            \
 344         else if (bpf_size == BPF_DW)                            \
 345                 bytes = sizeof(u64);                            \
 346                                                                 \
 347         bytes;                                                  \
 348 })
 349
 350 #define BPF_SIZEOF(type)                                        \
 351         ({                                                      \
 352                 const int __size = bytes_to_bpf_size(sizeof(type)); \
 353                 BUILD_BUG_ON(__size < 0);                       \
 354                 __size;                                         \
 355         })
 356
 357 #define BPF_FIELD_SIZEOF(type, field)                           \
 358         ({                                                      \
 359                 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
 360                 BUILD_BUG_ON(__size < 0);                       \
 361                 __size;                                         \
 362         })
 363
 364 #define BPF_LDST_BYTES(insn)                                    \
 365         ({                                                      \
 366                 const int __size = bpf_size_to_bytes(BPF_SIZE(insn->code)); \
 367                 WARN_ON(__size < 0);                            \
 368                 __size;                                         \
 369         })
 370
 371 #define __BPF_MAP_0(m, v, ...) v
 372 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
 373 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
 374 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
 375 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
 376 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
 377
 378 #define __BPF_REG_0(...) __BPF_PAD(5)
 379 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
 380 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
 381 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
 382 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
 383 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
 384
 385 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
 386 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
 387
 388 #define __BPF_CAST(t, a)                                                       \
 389         (__force t)                                                            \
 390         (__force                                                               \
 391          typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long),      \
 392                                       (unsigned long)0, (t)0))) a
 393 #define __BPF_V void
 394 #define __BPF_N
 395
 396 #define __BPF_DECL_ARGS(t, a) t   a
 397 #define __BPF_DECL_REGS(t, a) u64 a
 398
 399 #define __BPF_PAD(n)                                                           \
 400         __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2,       \
 401                   u64, __ur_3, u64, __ur_4, u64, __ur_5)
 402
 403 #define BPF_CALL_x(x, name, ...)                                               \
 404         static __always_inline                                                 \
 405         u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__));   \
 406         u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__));         \
 407         u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__))          \
 408         {                                                                      \
 409                 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
 410         }                                                                      \
 411         static __always_inline                                                 \
 412         u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
 413
 414 #define BPF_CALL_0(name, ...)   BPF_CALL_x(0, name, __VA_ARGS__)
 415 #define BPF_CALL_1(name, ...)   BPF_CALL_x(1, name, __VA_ARGS__)
 416 #define BPF_CALL_2(name, ...)   BPF_CALL_x(2, name, __VA_ARGS__)
 417 #define BPF_CALL_3(name, ...)   BPF_CALL_x(3, name, __VA_ARGS__)
 418 #define BPF_CALL_4(name, ...)   BPF_CALL_x(4, name, __VA_ARGS__)
 419 #define BPF_CALL_5(name, ...)   BPF_CALL_x(5, name, __VA_ARGS__)
 420
 421 #define bpf_ctx_range(TYPE, MEMBER)                                             \
 422         offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
 423 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2)                              \
 424         offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
 425
 426 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE)                            \
 427         ({                                                                      \
 428                 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE));             \
 429                 *(PTR_SIZE) = (SIZE);                                           \
 430                 offsetof(TYPE, MEMBER);                                         \
 431         })
 432
 433 #ifdef CONFIG_COMPAT
 434 /* A struct sock_filter is architecture independent. */
 435 struct compat_sock_fprog {
 436         u16             len;
 437         compat_uptr_t   filter; /* struct sock_filter * */
 438 };
 439 #endif
 440
 441 struct sock_fprog_kern {
 442         u16                     len;
 443         struct sock_filter      *filter;
 444 };
 445
 446 struct bpf_binary_header {
 447         unsigned int pages;
 448         u8 image[];
 449 };
 450
 451 struct bpf_prog {
 452         u16                     pages;          /* Number of allocated pages */
 453         u16                     jited:1,        /* Is our filter JIT'ed? */
 454                                 locked:1,       /* Program image locked? */
 455                                 gpl_compatible:1, /* Is filter GPL compatible? */
 456                                 cb_access:1,    /* Is control block accessed? */
 457                                 dst_needed:1;   /* Do we need dst entry? */
 458         enum bpf_prog_type      type;           /* Type of BPF program */
 459         u32                     len;            /* Number of filter blocks */
 460         u32                     jited_len;      /* Size of jited insns in bytes */
 461         u8                      tag[BPF_TAG_SIZE];
 462         struct bpf_prog_aux     *aux;           /* Auxiliary fields */
 463         struct sock_fprog_kern  *orig_prog;     /* Original BPF program */
 464         unsigned int            (*bpf_func)(const void *ctx,
 465                                             const struct bpf_insn *insn);
 466         /* Instructions for interpreter */
 467         union {
 468                 struct sock_filter      insns[0];
 469                 struct bpf_insn         insnsi[0];
 470         };
 471 };
 472
 473 struct sk_filter {
 474         refcount_t      refcnt;
 475         struct rcu_head rcu;
 476         struct bpf_prog *prog;
 477 };
 478
 479 #define BPF_PROG_RUN(filter, ctx)  (*filter->bpf_func)(ctx, filter->insnsi)
 480
 481 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
 482
 483 struct bpf_skb_data_end {
 484         struct qdisc_skb_cb qdisc_cb;
 485         void *data_end;
 486 };
 487
 488 struct xdp_buff {
 489         void *data;
 490         void *data_end;
 491         void *data_hard_start;
 492 };
 493
 494 /* compute the linear packet data range [data, data_end) which
 495  * will be accessed by cls_bpf, act_bpf and lwt programs
 496  */
 497 static inline void bpf_compute_data_end(struct sk_buff *skb)
 498 {
 499         struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
 500
 501         BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
 502         cb->data_end = skb->data + skb_headlen(skb);
 503 }
 504
 505 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
 506 {
 507         /* eBPF programs may read/write skb->cb[] area to transfer meta
 508          * data between tail calls. Since this also needs to work with
 509          * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
 510          *
 511          * In some socket filter cases, the cb unfortunately needs to be
 512          * saved/restored so that protocol specific skb->cb[] data won't
 513          * be lost. In any case, due to unpriviledged eBPF programs
 514          * attached to sockets, we need to clear the bpf_skb_cb() area
 515          * to not leak previous contents to user space.
 516          */
 517         BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
 518         BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
 519                      FIELD_SIZEOF(struct qdisc_skb_cb, data));
 520
 521         return qdisc_skb_cb(skb)->data;
 522 }
 523
 524 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
 525                                        struct sk_buff *skb)
 526 {
 527         u8 *cb_data = bpf_skb_cb(skb);
 528         u8 cb_saved[BPF_SKB_CB_LEN];
 529         u32 res;
 530
 531         if (unlikely(prog->cb_access)) {
 532                 memcpy(cb_saved, cb_data, sizeof(cb_saved));
 533                 memset(cb_data, 0, sizeof(cb_saved));
 534         }
 535
 536         res = BPF_PROG_RUN(prog, skb);
 537
 538         if (unlikely(prog->cb_access))
 539                 memcpy(cb_data, cb_saved, sizeof(cb_saved));
 540
 541         return res;
 542 }
 543
 544 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
 545                                         struct sk_buff *skb)
 546 {
 547         u8 *cb_data = bpf_skb_cb(skb);
 548
 549         if (unlikely(prog->cb_access))
 550                 memset(cb_data, 0, BPF_SKB_CB_LEN);
 551
 552         return BPF_PROG_RUN(prog, skb);
 553 }
 554
 555 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
 556                                             struct xdp_buff *xdp)
 557 {
 558         /* Caller needs to hold rcu_read_lock() (!), otherwise program
 559          * can be released while still running, or map elements could be
 560          * freed early while still having concurrent users. XDP fastpath
 561          * already takes rcu_read_lock() when fetching the program, so
 562          * it's not necessary here anymore.
 563          */
 564         return BPF_PROG_RUN(prog, xdp);
 565 }
 566
 567 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
 568 {
 569         return prog->len * sizeof(struct bpf_insn);
 570 }
 571
 572 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
 573 {
 574         return round_up(bpf_prog_insn_size(prog) +
 575                         sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
 576 }
 577
 578 static inline unsigned int bpf_prog_size(unsigned int proglen)
 579 {
 580         return max(sizeof(struct bpf_prog),
 581                    offsetof(struct bpf_prog, insns[proglen]));
 582 }
 583
 584 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
 585 {
 586         /* When classic BPF programs have been loaded and the arch
 587          * does not have a classic BPF JIT (anymore), they have been
 588          * converted via bpf_migrate_filter() to eBPF and thus always
 589          * have an unspec program type.
 590          */
 591         return prog->type == BPF_PROG_TYPE_UNSPEC;
 592 }
 593
 594 static inline bool
 595 bpf_ctx_narrow_access_ok(u32 off, u32 size, const u32 size_default)
 596 {
 597         bool off_ok;
 598 #ifdef __LITTLE_ENDIAN
 599         off_ok = (off & (size_default - 1)) == 0;
 600 #else
 601         off_ok = (off & (size_default - 1)) + size == size_default;
 602 #endif
 603         return off_ok && size <= size_default && (size & (size - 1)) == 0;
 604 }
 605
 606 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
 607
 608 #ifdef CONFIG_ARCH_HAS_SET_MEMORY
 609 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
 610 {
 611         fp->locked = 1;
 612         WARN_ON_ONCE(set_memory_ro((unsigned long)fp, fp->pages));
 613 }
 614
 615 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
 616 {
 617         if (fp->locked) {
 618                 WARN_ON_ONCE(set_memory_rw((unsigned long)fp, fp->pages));
 619                 /* In case set_memory_rw() fails, we want to be the first
 620                  * to crash here instead of some random place later on.
 621                  */
 622                 fp->locked = 0;
 623         }
 624 }
 625
 626 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
 627 {
 628         WARN_ON_ONCE(set_memory_ro((unsigned long)hdr, hdr->pages));
 629 }
 630
 631 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
 632 {
 633         WARN_ON_ONCE(set_memory_rw((unsigned long)hdr, hdr->pages));
 634 }
 635 #else
 636 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
 637 {
 638 }
 639
 640 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
 641 {
 642 }
 643
 644 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
 645 {
 646 }
 647
 648 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
 649 {
 650 }
 651 #endif /* CONFIG_ARCH_HAS_SET_MEMORY */
 652
 653 static inline struct bpf_binary_header *
 654 bpf_jit_binary_hdr(const struct bpf_prog *fp)
 655 {
 656         unsigned long real_start = (unsigned long)fp->bpf_func;
 657         unsigned long addr = real_start & PAGE_MASK;
 658
 659         return (void *)addr;
 660 }
 661
 662 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
 663 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
 664 {
 665         return sk_filter_trim_cap(sk, skb, 1);
 666 }
 667
 668 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
 669 void bpf_prog_free(struct bpf_prog *fp);
 670
 671 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
 672 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
 673                                   gfp_t gfp_extra_flags);
 674 void __bpf_prog_free(struct bpf_prog *fp);
 675
 676 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
 677 {
 678         bpf_prog_unlock_ro(fp);
 679         __bpf_prog_free(fp);
 680 }
 681
 682 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
 683                                        unsigned int flen);
 684
 685 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
 686 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
 687                               bpf_aux_classic_check_t trans, bool save_orig);
 688 void bpf_prog_destroy(struct bpf_prog *fp);
 689
 690 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
 691 int sk_attach_bpf(u32 ufd, struct sock *sk);
 692 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
 693 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
 694 int sk_detach_filter(struct sock *sk);
 695 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
 696                   unsigned int len);
 697
 698 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
 699 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
 700
 701 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
 702
 703 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
 704 void bpf_jit_compile(struct bpf_prog *prog);
 705 bool bpf_helper_changes_pkt_data(void *func);
 706
 707 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 708                                        const struct bpf_insn *patch, u32 len);
 709
 710 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
 711  * same cpu context. Further for best results no more than a single map
 712  * for the do_redirect/do_flush pair should be used. This limitation is
 713  * because we only track one map and force a flush when the map changes.
 714  * This does not appear to be a real limitation for existing software.
 715  */
 716 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
 717                             struct bpf_prog *prog);
 718 int xdp_do_redirect(struct net_device *dev,
 719                     struct xdp_buff *xdp,
 720                     struct bpf_prog *prog);
 721 void xdp_do_flush_map(void);
 722
 723 void bpf_warn_invalid_xdp_action(u32 act);
 724 void bpf_warn_invalid_xdp_redirect(u32 ifindex);
 725
 726 struct sock *do_sk_redirect_map(struct sk_buff *skb);
 727
 728 #ifdef CONFIG_BPF_JIT
 729 extern int bpf_jit_enable;
 730 extern int bpf_jit_harden;
 731 extern int bpf_jit_kallsyms;
 732 extern long bpf_jit_limit;
 733 extern long bpf_jit_limit_max;
 734
 735 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
 736
 737 struct bpf_binary_header *
 738 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
 739                      unsigned int alignment,
 740                      bpf_jit_fill_hole_t bpf_fill_ill_insns);
 741 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
 742 u64 bpf_jit_alloc_exec_limit(void);
 743 void *bpf_jit_alloc_exec(unsigned long size);
 744 void bpf_jit_free_exec(void *addr);
 745 void bpf_jit_free(struct bpf_prog *fp);
 746
 747 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
 748 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
 749
 750 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
 751                                 u32 pass, void *image)
 752 {
 753         pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
 754                proglen, pass, image, current->comm, task_pid_nr(current));
 755
 756         if (image)
 757                 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
 758                                16, 1, image, proglen, false);
 759 }
 760
 761 static inline bool bpf_jit_is_ebpf(void)
 762 {
 763 # ifdef CONFIG_HAVE_EBPF_JIT
 764         return true;
 765 # else
 766         return false;
 767 # endif
 768 }
 769
 770 static inline bool ebpf_jit_enabled(void)
 771 {
 772         return bpf_jit_enable && bpf_jit_is_ebpf();
 773 }
 774
 775 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
 776 {
 777         return fp->jited && bpf_jit_is_ebpf();
 778 }
 779
 780 static inline bool bpf_jit_blinding_enabled(void)
 781 {
 782         /* These are the prerequisites, should someone ever have the
 783          * idea to call blinding outside of them, we make sure to
 784          * bail out.
 785          */
 786         if (!bpf_jit_is_ebpf())
 787                 return false;
 788         if (!bpf_jit_enable)
 789                 return false;
 790         if (!bpf_jit_harden)
 791                 return false;
 792         if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
 793                 return false;
 794
 795         return true;
 796 }
 797
 798 static inline bool bpf_jit_kallsyms_enabled(void)
 799 {
 800         /* There are a couple of corner cases where kallsyms should
 801          * not be enabled f.e. on hardening.
 802          */
 803         if (bpf_jit_harden)
 804                 return false;
 805         if (!bpf_jit_kallsyms)
 806                 return false;
 807         if (bpf_jit_kallsyms == 1)
 808                 return true;
 809
 810         return false;
 811 }
 812
 813 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
 814                                  unsigned long *off, char *sym);
 815 bool is_bpf_text_address(unsigned long addr);
 816 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
 817                     char *sym);
 818
 819 static inline const char *
 820 bpf_address_lookup(unsigned long addr, unsigned long *size,
 821                    unsigned long *off, char **modname, char *sym)
 822 {
 823         const char *ret = __bpf_address_lookup(addr, size, off, sym);
 824
 825         if (ret && modname)
 826                 *modname = NULL;
 827         return ret;
 828 }
 829
 830 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
 831 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
 832
 833 #else /* CONFIG_BPF_JIT */
 834
 835 static inline bool ebpf_jit_enabled(void)
 836 {
 837         return false;
 838 }
 839
 840 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
 841 {
 842         return false;
 843 }
 844
 845 static inline void bpf_jit_free(struct bpf_prog *fp)
 846 {
 847         bpf_prog_unlock_free(fp);
 848 }
 849
 850 static inline bool bpf_jit_kallsyms_enabled(void)
 851 {
 852         return false;
 853 }
 854
 855 static inline const char *
 856 __bpf_address_lookup(unsigned long addr, unsigned long *size,
 857                      unsigned long *off, char *sym)
 858 {
 859         return NULL;
 860 }
 861
 862 static inline bool is_bpf_text_address(unsigned long addr)
 863 {
 864         return false;
 865 }
 866
 867 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
 868                                   char *type, char *sym)
 869 {
 870         return -ERANGE;
 871 }
 872
 873 static inline const char *
 874 bpf_address_lookup(unsigned long addr, unsigned long *size,
 875                    unsigned long *off, char **modname, char *sym)
 876 {
 877         return NULL;
 878 }
 879
 880 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
 881 {
 882 }
 883
 884 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
 885 {
 886 }
 887 #endif /* CONFIG_BPF_JIT */
 888
 889 #define BPF_ANC         BIT(15)
 890
 891 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
 892 {
 893         switch (first->code) {
 894         case BPF_RET | BPF_K:
 895         case BPF_LD | BPF_W | BPF_LEN:
 896                 return false;
 897
 898         case BPF_LD | BPF_W | BPF_ABS:
 899         case BPF_LD | BPF_H | BPF_ABS:
 900         case BPF_LD | BPF_B | BPF_ABS:
 901                 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
 902                         return true;
 903                 return false;
 904
 905         default:
 906                 return true;
 907         }
 908 }
 909
 910 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
 911 {
 912         BUG_ON(ftest->code & BPF_ANC);
 913
 914         switch (ftest->code) {
 915         case BPF_LD | BPF_W | BPF_ABS:
 916         case BPF_LD | BPF_H | BPF_ABS:
 917         case BPF_LD | BPF_B | BPF_ABS:
 918 #define BPF_ANCILLARY(CODE)     case SKF_AD_OFF + SKF_AD_##CODE:        \
 919                                 return BPF_ANC | SKF_AD_##CODE
 920                 switch (ftest->k) {
 921                 BPF_ANCILLARY(PROTOCOL);
 922                 BPF_ANCILLARY(PKTTYPE);
 923                 BPF_ANCILLARY(IFINDEX);
 924                 BPF_ANCILLARY(NLATTR);
 925                 BPF_ANCILLARY(NLATTR_NEST);
 926                 BPF_ANCILLARY(MARK);
 927                 BPF_ANCILLARY(QUEUE);
 928                 BPF_ANCILLARY(HATYPE);
 929                 BPF_ANCILLARY(RXHASH);
 930                 BPF_ANCILLARY(CPU);
 931                 BPF_ANCILLARY(ALU_XOR_X);
 932                 BPF_ANCILLARY(VLAN_TAG);
 933                 BPF_ANCILLARY(VLAN_TAG_PRESENT);
 934                 BPF_ANCILLARY(PAY_OFFSET);
 935                 BPF_ANCILLARY(RANDOM);
 936                 BPF_ANCILLARY(VLAN_TPID);
 937                 }
 938                 /* Fallthrough. */
 939         default:
 940                 return ftest->code;
 941         }
 942 }
 943
 944 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
 945                                            int k, unsigned int size);
 946
 947 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
 948                                      unsigned int size, void *buffer)
 949 {
 950         if (k >= 0)
 951                 return skb_header_pointer(skb, k, size, buffer);
 952
 953         return bpf_internal_load_pointer_neg_helper(skb, k, size);
 954 }
 955
 956 static inline int bpf_tell_extensions(void)
 957 {
 958         return SKF_AD_MAX;
 959 }
 960
 961 struct bpf_sock_ops_kern {
 962         struct  sock *sk;
 963         u32     op;
 964         union {
 965                 u32 reply;
 966                 u32 replylong[4];
 967         };
 968 };
 969
 970 #endif /* __LINUX_FILTER_H__ */