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_ALU_REG(CLASS, OP, DST, SRC)                        \
  71         ((struct bpf_insn) {                                    \
  72                 .code  = CLASS | BPF_OP(OP) | BPF_X,            \
  73                 .dst_reg = DST,                                 \
  74                 .src_reg = SRC,                                 \
  75                 .off   = 0,                                     \
  76                 .imm   = 0 })
  77
  78 #define BPF_ALU64_REG(OP, DST, SRC)                             \
  79         ((struct bpf_insn) {                                    \
  80                 .code  = BPF_ALU64 | BPF_OP(OP) | BPF_X,        \
  81                 .dst_reg = DST,                                 \
  82                 .src_reg = SRC,                                 \
  83                 .off   = 0,                                     \
  84                 .imm   = 0 })
  85
  86 #define BPF_ALU32_REG(OP, DST, SRC)                             \
  87         ((struct bpf_insn) {                                    \
  88                 .code  = BPF_ALU | BPF_OP(OP) | BPF_X,          \
  89                 .dst_reg = DST,                                 \
  90                 .src_reg = SRC,                                 \
  91                 .off   = 0,                                     \
  92                 .imm   = 0 })
  93
  94 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
  95
  96 #define BPF_ALU64_IMM(OP, DST, IMM)                             \
  97         ((struct bpf_insn) {                                    \
  98                 .code  = BPF_ALU64 | BPF_OP(OP) | BPF_K,        \
  99                 .dst_reg = DST,                                 \
 100                 .src_reg = 0,                                   \
 101                 .off   = 0,                                     \
 102                 .imm   = IMM })
 103
 104 #define BPF_ALU32_IMM(OP, DST, IMM)                             \
 105         ((struct bpf_insn) {                                    \
 106                 .code  = BPF_ALU | BPF_OP(OP) | BPF_K,          \
 107                 .dst_reg = DST,                                 \
 108                 .src_reg = 0,                                   \
 109                 .off   = 0,                                     \
 110                 .imm   = IMM })
 111
 112 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
 113
 114 #define BPF_ENDIAN(TYPE, DST, LEN)                              \
 115         ((struct bpf_insn) {                                    \
 116                 .code  = BPF_ALU | BPF_END | BPF_SRC(TYPE),     \
 117                 .dst_reg = DST,                                 \
 118                 .src_reg = 0,                                   \
 119                 .off   = 0,                                     \
 120                 .imm   = LEN })
 121
 122 /* Short form of mov, dst_reg = src_reg */
 123
 124 #define BPF_MOV_REG(CLASS, DST, SRC)                            \
 125         ((struct bpf_insn) {                                    \
 126                 .code  = CLASS | BPF_MOV | BPF_X,               \
 127                 .dst_reg = DST,                                 \
 128                 .src_reg = SRC,                                 \
 129                 .off   = 0,                                     \
 130                 .imm   = 0 })
 131
 132 #define BPF_MOV64_REG(DST, SRC)                                 \
 133         ((struct bpf_insn) {                                    \
 134                 .code  = BPF_ALU64 | BPF_MOV | BPF_X,           \
 135                 .dst_reg = DST,                                 \
 136                 .src_reg = SRC,                                 \
 137                 .off   = 0,                                     \
 138                 .imm   = 0 })
 139
 140 #define BPF_MOV32_REG(DST, SRC)                                 \
 141         ((struct bpf_insn) {                                    \
 142                 .code  = BPF_ALU | BPF_MOV | BPF_X,             \
 143                 .dst_reg = DST,                                 \
 144                 .src_reg = SRC,                                 \
 145                 .off   = 0,                                     \
 146                 .imm   = 0 })
 147
 148 /* Short form of mov, dst_reg = imm32 */
 149
 150 #define BPF_MOV64_IMM(DST, IMM)                                 \
 151         ((struct bpf_insn) {                                    \
 152                 .code  = BPF_ALU64 | BPF_MOV | BPF_K,           \
 153                 .dst_reg = DST,                                 \
 154                 .src_reg = 0,                                   \
 155                 .off   = 0,                                     \
 156                 .imm   = IMM })
 157
 158 #define BPF_MOV32_IMM(DST, IMM)                                 \
 159         ((struct bpf_insn) {                                    \
 160                 .code  = BPF_ALU | BPF_MOV | BPF_K,             \
 161                 .dst_reg = DST,                                 \
 162                 .src_reg = 0,                                   \
 163                 .off   = 0,                                     \
 164                 .imm   = IMM })
 165
 166 #define BPF_RAW_REG(insn, DST, SRC)                             \
 167         ((struct bpf_insn) {                                    \
 168                 .code  = (insn).code,                           \
 169                 .dst_reg = DST,                                 \
 170                 .src_reg = SRC,                                 \
 171                 .off   = (insn).off,                            \
 172                 .imm   = (insn).imm })
 173
 174 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
 175 #define BPF_LD_IMM64(DST, IMM)                                  \
 176         BPF_LD_IMM64_RAW(DST, 0, IMM)
 177
 178 #define BPF_LD_IMM64_RAW(DST, SRC, IMM)                         \
 179         ((struct bpf_insn) {                                    \
 180                 .code  = BPF_LD | BPF_DW | BPF_IMM,             \
 181                 .dst_reg = DST,                                 \
 182                 .src_reg = SRC,                                 \
 183                 .off   = 0,                                     \
 184                 .imm   = (__u32) (IMM) }),                      \
 185         ((struct bpf_insn) {                                    \
 186                 .code  = 0, /* zero is reserved opcode */       \
 187                 .dst_reg = 0,                                   \
 188                 .src_reg = 0,                                   \
 189                 .off   = 0,                                     \
 190                 .imm   = ((__u64) (IMM)) >> 32 })
 191
 192 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
 193 #define BPF_LD_MAP_FD(DST, MAP_FD)                              \
 194         BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
 195
 196 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
 197
 198 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM)                      \
 199         ((struct bpf_insn) {                                    \
 200                 .code  = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE),   \
 201                 .dst_reg = DST,                                 \
 202                 .src_reg = SRC,                                 \
 203                 .off   = 0,                                     \
 204                 .imm   = IMM })
 205
 206 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM)                      \
 207         ((struct bpf_insn) {                                    \
 208                 .code  = BPF_ALU | BPF_MOV | BPF_SRC(TYPE),     \
 209                 .dst_reg = DST,                                 \
 210                 .src_reg = SRC,                                 \
 211                 .off   = 0,                                     \
 212                 .imm   = IMM })
 213
 214 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
 215
 216 #define BPF_LD_ABS(SIZE, IMM)                                   \
 217         ((struct bpf_insn) {                                    \
 218                 .code  = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS,     \
 219                 .dst_reg = 0,                                   \
 220                 .src_reg = 0,                                   \
 221                 .off   = 0,                                     \
 222                 .imm   = IMM })
 223
 224 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
 225
 226 #define BPF_LD_IND(SIZE, SRC, IMM)                              \
 227         ((struct bpf_insn) {                                    \
 228                 .code  = BPF_LD | BPF_SIZE(SIZE) | BPF_IND,     \
 229                 .dst_reg = 0,                                   \
 230                 .src_reg = SRC,                                 \
 231                 .off   = 0,                                     \
 232                 .imm   = IMM })
 233
 234 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
 235
 236 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF)                        \
 237         ((struct bpf_insn) {                                    \
 238                 .code  = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM,    \
 239                 .dst_reg = DST,                                 \
 240                 .src_reg = SRC,                                 \
 241                 .off   = OFF,                                   \
 242                 .imm   = 0 })
 243
 244 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
 245
 246 #define BPF_STX_MEM(SIZE, DST, SRC, OFF)                        \
 247         ((struct bpf_insn) {                                    \
 248                 .code  = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM,    \
 249                 .dst_reg = DST,                                 \
 250                 .src_reg = SRC,                                 \
 251                 .off   = OFF,                                   \
 252                 .imm   = 0 })
 253
 254 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
 255
 256 #define BPF_STX_XADD(SIZE, DST, SRC, OFF)                       \
 257         ((struct bpf_insn) {                                    \
 258                 .code  = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD,   \
 259                 .dst_reg = DST,                                 \
 260                 .src_reg = SRC,                                 \
 261                 .off   = OFF,                                   \
 262                 .imm   = 0 })
 263
 264 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
 265
 266 #define BPF_ST_MEM(SIZE, DST, OFF, IMM)                         \
 267         ((struct bpf_insn) {                                    \
 268                 .code  = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM,     \
 269                 .dst_reg = DST,                                 \
 270                 .src_reg = 0,                                   \
 271                 .off   = OFF,                                   \
 272                 .imm   = IMM })
 273
 274 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
 275
 276 #define BPF_JMP_REG(OP, DST, SRC, OFF)                          \
 277         ((struct bpf_insn) {                                    \
 278                 .code  = BPF_JMP | BPF_OP(OP) | BPF_X,          \
 279                 .dst_reg = DST,                                 \
 280                 .src_reg = SRC,                                 \
 281                 .off   = OFF,                                   \
 282                 .imm   = 0 })
 283
 284 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
 285
 286 #define BPF_JMP_IMM(OP, DST, IMM, OFF)                          \
 287         ((struct bpf_insn) {                                    \
 288                 .code  = BPF_JMP | BPF_OP(OP) | BPF_K,          \
 289                 .dst_reg = DST,                                 \
 290                 .src_reg = 0,                                   \
 291                 .off   = OFF,                                   \
 292                 .imm   = IMM })
 293
 294 /* Unconditional jumps, goto pc + off16 */
 295
 296 #define BPF_JMP_A(OFF)                                          \
 297         ((struct bpf_insn) {                                    \
 298                 .code  = BPF_JMP | BPF_JA,                      \
 299                 .dst_reg = 0,                                   \
 300                 .src_reg = 0,                                   \
 301                 .off   = OFF,                                   \
 302                 .imm   = 0 })
 303
 304 /* Function call */
 305
 306 #define BPF_EMIT_CALL(FUNC)                                     \
 307         ((struct bpf_insn) {                                    \
 308                 .code  = BPF_JMP | BPF_CALL,                    \
 309                 .dst_reg = 0,                                   \
 310                 .src_reg = 0,                                   \
 311                 .off   = 0,                                     \
 312                 .imm   = ((FUNC) - __bpf_call_base) })
 313
 314 /* Raw code statement block */
 315
 316 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM)                  \
 317         ((struct bpf_insn) {                                    \
 318                 .code  = CODE,                                  \
 319                 .dst_reg = DST,                                 \
 320                 .src_reg = SRC,                                 \
 321                 .off   = OFF,                                   \
 322                 .imm   = IMM })
 323
 324 /* Program exit */
 325
 326 #define BPF_EXIT_INSN()                                         \
 327         ((struct bpf_insn) {                                    \
 328                 .code  = BPF_JMP | BPF_EXIT,                    \
 329                 .dst_reg = 0,                                   \
 330                 .src_reg = 0,                                   \
 331                 .off   = 0,                                     \
 332                 .imm   = 0 })
 333
 334 /* Internal classic blocks for direct assignment */
 335
 336 #define __BPF_STMT(CODE, K)                                     \
 337         ((struct sock_filter) BPF_STMT(CODE, K))
 338
 339 #define __BPF_JUMP(CODE, K, JT, JF)                             \
 340         ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
 341
 342 #define bytes_to_bpf_size(bytes)                                \
 343 ({                                                              \
 344         int bpf_size = -EINVAL;                                 \
 345                                                                 \
 346         if (bytes == sizeof(u8))                                \
 347                 bpf_size = BPF_B;                               \
 348         else if (bytes == sizeof(u16))                          \
 349                 bpf_size = BPF_H;                               \
 350         else if (bytes == sizeof(u32))                          \
 351                 bpf_size = BPF_W;                               \
 352         else if (bytes == sizeof(u64))                          \
 353                 bpf_size = BPF_DW;                              \
 354                                                                 \
 355         bpf_size;                                               \
 356 })
 357
 358 #define bpf_size_to_bytes(bpf_size)                             \
 359 ({                                                              \
 360         int bytes = -EINVAL;                                    \
 361                                                                 \
 362         if (bpf_size == BPF_B)                                  \
 363                 bytes = sizeof(u8);                             \
 364         else if (bpf_size == BPF_H)                             \
 365                 bytes = sizeof(u16);                            \
 366         else if (bpf_size == BPF_W)                             \
 367                 bytes = sizeof(u32);                            \
 368         else if (bpf_size == BPF_DW)                            \
 369                 bytes = sizeof(u64);                            \
 370                                                                 \
 371         bytes;                                                  \
 372 })
 373
 374 #define BPF_SIZEOF(type)                                        \
 375         ({                                                      \
 376                 const int __size = bytes_to_bpf_size(sizeof(type)); \
 377                 BUILD_BUG_ON(__size < 0);                       \
 378                 __size;                                         \
 379         })
 380
 381 #define BPF_FIELD_SIZEOF(type, field)                           \
 382         ({                                                      \
 383                 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
 384                 BUILD_BUG_ON(__size < 0);                       \
 385                 __size;                                         \
 386         })
 387
 388 #define BPF_LDST_BYTES(insn)                                    \
 389         ({                                                      \
 390                 const int __size = bpf_size_to_bytes(BPF_SIZE(insn->code)); \
 391                 WARN_ON(__size < 0);                            \
 392                 __size;                                         \
 393         })
 394
 395 #define __BPF_MAP_0(m, v, ...) v
 396 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
 397 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
 398 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
 399 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
 400 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
 401
 402 #define __BPF_REG_0(...) __BPF_PAD(5)
 403 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
 404 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
 405 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
 406 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
 407 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
 408
 409 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
 410 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
 411
 412 #define __BPF_CAST(t, a)                                                       \
 413         (__force t)                                                            \
 414         (__force                                                               \
 415          typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long),      \
 416                                       (unsigned long)0, (t)0))) a
 417 #define __BPF_V void
 418 #define __BPF_N
 419
 420 #define __BPF_DECL_ARGS(t, a) t   a
 421 #define __BPF_DECL_REGS(t, a) u64 a
 422
 423 #define __BPF_PAD(n)                                                           \
 424         __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2,       \
 425                   u64, __ur_3, u64, __ur_4, u64, __ur_5)
 426
 427 #define BPF_CALL_x(x, name, ...)                                               \
 428         static __always_inline                                                 \
 429         u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__));   \
 430         u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__));         \
 431         u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__))          \
 432         {                                                                      \
 433                 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
 434         }                                                                      \
 435         static __always_inline                                                 \
 436         u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
 437
 438 #define BPF_CALL_0(name, ...)   BPF_CALL_x(0, name, __VA_ARGS__)
 439 #define BPF_CALL_1(name, ...)   BPF_CALL_x(1, name, __VA_ARGS__)
 440 #define BPF_CALL_2(name, ...)   BPF_CALL_x(2, name, __VA_ARGS__)
 441 #define BPF_CALL_3(name, ...)   BPF_CALL_x(3, name, __VA_ARGS__)
 442 #define BPF_CALL_4(name, ...)   BPF_CALL_x(4, name, __VA_ARGS__)
 443 #define BPF_CALL_5(name, ...)   BPF_CALL_x(5, name, __VA_ARGS__)
 444
 445 #define bpf_ctx_range(TYPE, MEMBER)                                             \
 446         offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
 447 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2)                              \
 448         offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
 449
 450 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE)                            \
 451         ({                                                                      \
 452                 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE));             \
 453                 *(PTR_SIZE) = (SIZE);                                           \
 454                 offsetof(TYPE, MEMBER);                                         \
 455         })
 456
 457 #ifdef CONFIG_COMPAT
 458 /* A struct sock_filter is architecture independent. */
 459 struct compat_sock_fprog {
 460         u16             len;
 461         compat_uptr_t   filter; /* struct sock_filter * */
 462 };
 463 #endif
 464
 465 struct sock_fprog_kern {
 466         u16                     len;
 467         struct sock_filter      *filter;
 468 };
 469
 470 struct bpf_binary_header {
 471         unsigned int pages;
 472         u8 image[];
 473 };
 474
 475 struct bpf_prog {
 476         u16                     pages;          /* Number of allocated pages */
 477         u16                     jited:1,        /* Is our filter JIT'ed? */
 478                                 locked:1,       /* Program image locked? */
 479                                 gpl_compatible:1, /* Is filter GPL compatible? */
 480                                 cb_access:1,    /* Is control block accessed? */
 481                                 dst_needed:1;   /* Do we need dst entry? */
 482         enum bpf_prog_type      type;           /* Type of BPF program */
 483         u32                     len;            /* Number of filter blocks */
 484         u32                     jited_len;      /* Size of jited insns in bytes */
 485         u8                      tag[BPF_TAG_SIZE];
 486         struct bpf_prog_aux     *aux;           /* Auxiliary fields */
 487         struct sock_fprog_kern  *orig_prog;     /* Original BPF program */
 488         unsigned int            (*bpf_func)(const void *ctx,
 489                                             const struct bpf_insn *insn);
 490         /* Instructions for interpreter */
 491         union {
 492                 struct sock_filter      insns[0];
 493                 struct bpf_insn         insnsi[0];
 494         };
 495 };
 496
 497 struct sk_filter {
 498         refcount_t      refcnt;
 499         struct rcu_head rcu;
 500         struct bpf_prog *prog;
 501 };
 502
 503 #define BPF_PROG_RUN(filter, ctx)  (*filter->bpf_func)(ctx, filter->insnsi)
 504
 505 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
 506
 507 struct bpf_skb_data_end {
 508         struct qdisc_skb_cb qdisc_cb;
 509         void *data_end;
 510 };
 511
 512 struct xdp_buff {
 513         void *data;
 514         void *data_end;
 515         void *data_hard_start;
 516 };
 517
 518 /* compute the linear packet data range [data, data_end) which
 519  * will be accessed by cls_bpf, act_bpf and lwt programs
 520  */
 521 static inline void bpf_compute_data_end(struct sk_buff *skb)
 522 {
 523         struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
 524
 525         BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
 526         cb->data_end = skb->data + skb_headlen(skb);
 527 }
 528
 529 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
 530 {
 531         /* eBPF programs may read/write skb->cb[] area to transfer meta
 532          * data between tail calls. Since this also needs to work with
 533          * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
 534          *
 535          * In some socket filter cases, the cb unfortunately needs to be
 536          * saved/restored so that protocol specific skb->cb[] data won't
 537          * be lost. In any case, due to unpriviledged eBPF programs
 538          * attached to sockets, we need to clear the bpf_skb_cb() area
 539          * to not leak previous contents to user space.
 540          */
 541         BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
 542         BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
 543                      FIELD_SIZEOF(struct qdisc_skb_cb, data));
 544
 545         return qdisc_skb_cb(skb)->data;
 546 }
 547
 548 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
 549                                        struct sk_buff *skb)
 550 {
 551         u8 *cb_data = bpf_skb_cb(skb);
 552         u8 cb_saved[BPF_SKB_CB_LEN];
 553         u32 res;
 554
 555         if (unlikely(prog->cb_access)) {
 556                 memcpy(cb_saved, cb_data, sizeof(cb_saved));
 557                 memset(cb_data, 0, sizeof(cb_saved));
 558         }
 559
 560         res = BPF_PROG_RUN(prog, skb);
 561
 562         if (unlikely(prog->cb_access))
 563                 memcpy(cb_data, cb_saved, sizeof(cb_saved));
 564
 565         return res;
 566 }
 567
 568 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
 569                                         struct sk_buff *skb)
 570 {
 571         u8 *cb_data = bpf_skb_cb(skb);
 572
 573         if (unlikely(prog->cb_access))
 574                 memset(cb_data, 0, BPF_SKB_CB_LEN);
 575
 576         return BPF_PROG_RUN(prog, skb);
 577 }
 578
 579 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
 580                                             struct xdp_buff *xdp)
 581 {
 582         /* Caller needs to hold rcu_read_lock() (!), otherwise program
 583          * can be released while still running, or map elements could be
 584          * freed early while still having concurrent users. XDP fastpath
 585          * already takes rcu_read_lock() when fetching the program, so
 586          * it's not necessary here anymore.
 587          */
 588         return BPF_PROG_RUN(prog, xdp);
 589 }
 590
 591 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
 592 {
 593         return prog->len * sizeof(struct bpf_insn);
 594 }
 595
 596 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
 597 {
 598         return round_up(bpf_prog_insn_size(prog) +
 599                         sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
 600 }
 601
 602 static inline unsigned int bpf_prog_size(unsigned int proglen)
 603 {
 604         return max(sizeof(struct bpf_prog),
 605                    offsetof(struct bpf_prog, insns[proglen]));
 606 }
 607
 608 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
 609 {
 610         /* When classic BPF programs have been loaded and the arch
 611          * does not have a classic BPF JIT (anymore), they have been
 612          * converted via bpf_migrate_filter() to eBPF and thus always
 613          * have an unspec program type.
 614          */
 615         return prog->type == BPF_PROG_TYPE_UNSPEC;
 616 }
 617
 618 static inline bool
 619 bpf_ctx_narrow_access_ok(u32 off, u32 size, const u32 size_default)
 620 {
 621         bool off_ok;
 622 #ifdef __LITTLE_ENDIAN
 623         off_ok = (off & (size_default - 1)) == 0;
 624 #else
 625         off_ok = (off & (size_default - 1)) + size == size_default;
 626 #endif
 627         return off_ok && size <= size_default && (size & (size - 1)) == 0;
 628 }
 629
 630 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
 631
 632 #ifdef CONFIG_ARCH_HAS_SET_MEMORY
 633 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
 634 {
 635         fp->locked = 1;
 636         WARN_ON_ONCE(set_memory_ro((unsigned long)fp, fp->pages));
 637 }
 638
 639 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
 640 {
 641         if (fp->locked) {
 642                 WARN_ON_ONCE(set_memory_rw((unsigned long)fp, fp->pages));
 643                 /* In case set_memory_rw() fails, we want to be the first
 644                  * to crash here instead of some random place later on.
 645                  */
 646                 fp->locked = 0;
 647         }
 648 }
 649
 650 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
 651 {
 652         WARN_ON_ONCE(set_memory_ro((unsigned long)hdr, hdr->pages));
 653 }
 654
 655 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
 656 {
 657         WARN_ON_ONCE(set_memory_rw((unsigned long)hdr, hdr->pages));
 658 }
 659 #else
 660 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
 661 {
 662 }
 663
 664 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
 665 {
 666 }
 667
 668 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
 669 {
 670 }
 671
 672 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
 673 {
 674 }
 675 #endif /* CONFIG_ARCH_HAS_SET_MEMORY */
 676
 677 static inline struct bpf_binary_header *
 678 bpf_jit_binary_hdr(const struct bpf_prog *fp)
 679 {
 680         unsigned long real_start = (unsigned long)fp->bpf_func;
 681         unsigned long addr = real_start & PAGE_MASK;
 682
 683         return (void *)addr;
 684 }
 685
 686 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
 687 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
 688 {
 689         return sk_filter_trim_cap(sk, skb, 1);
 690 }
 691
 692 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
 693 void bpf_prog_free(struct bpf_prog *fp);
 694
 695 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
 696 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
 697                                   gfp_t gfp_extra_flags);
 698 void __bpf_prog_free(struct bpf_prog *fp);
 699
 700 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
 701 {
 702         bpf_prog_unlock_ro(fp);
 703         __bpf_prog_free(fp);
 704 }
 705
 706 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
 707                                        unsigned int flen);
 708
 709 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
 710 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
 711                               bpf_aux_classic_check_t trans, bool save_orig);
 712 void bpf_prog_destroy(struct bpf_prog *fp);
 713
 714 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
 715 int sk_attach_bpf(u32 ufd, struct sock *sk);
 716 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
 717 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
 718 int sk_detach_filter(struct sock *sk);
 719 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
 720                   unsigned int len);
 721
 722 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
 723 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
 724
 725 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
 726
 727 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
 728 void bpf_jit_compile(struct bpf_prog *prog);
 729 bool bpf_helper_changes_pkt_data(void *func);
 730
 731 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 732                                        const struct bpf_insn *patch, u32 len);
 733
 734 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
 735  * same cpu context. Further for best results no more than a single map
 736  * for the do_redirect/do_flush pair should be used. This limitation is
 737  * because we only track one map and force a flush when the map changes.
 738  * This does not appear to be a real limitation for existing software.
 739  */
 740 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
 741                             struct bpf_prog *prog);
 742 int xdp_do_redirect(struct net_device *dev,
 743                     struct xdp_buff *xdp,
 744                     struct bpf_prog *prog);
 745 void xdp_do_flush_map(void);
 746
 747 void bpf_warn_invalid_xdp_action(u32 act);
 748 void bpf_warn_invalid_xdp_redirect(u32 ifindex);
 749
 750 struct sock *do_sk_redirect_map(struct sk_buff *skb);
 751
 752 #ifdef CONFIG_BPF_JIT
 753 extern int bpf_jit_enable;
 754 extern int bpf_jit_harden;
 755 extern int bpf_jit_kallsyms;
 756 extern long bpf_jit_limit;
 757 extern long bpf_jit_limit_max;
 758
 759 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
 760
 761 struct bpf_binary_header *
 762 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
 763                      unsigned int alignment,
 764                      bpf_jit_fill_hole_t bpf_fill_ill_insns);
 765 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
 766 u64 bpf_jit_alloc_exec_limit(void);
 767 void *bpf_jit_alloc_exec(unsigned long size);
 768 void bpf_jit_free_exec(void *addr);
 769 void bpf_jit_free(struct bpf_prog *fp);
 770
 771 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
 772 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
 773
 774 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
 775                                 u32 pass, void *image)
 776 {
 777         pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
 778                proglen, pass, image, current->comm, task_pid_nr(current));
 779
 780         if (image)
 781                 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
 782                                16, 1, image, proglen, false);
 783 }
 784
 785 static inline bool bpf_jit_is_ebpf(void)
 786 {
 787 # ifdef CONFIG_HAVE_EBPF_JIT
 788         return true;
 789 # else
 790         return false;
 791 # endif
 792 }
 793
 794 static inline bool ebpf_jit_enabled(void)
 795 {
 796         return bpf_jit_enable && bpf_jit_is_ebpf();
 797 }
 798
 799 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
 800 {
 801         return fp->jited && bpf_jit_is_ebpf();
 802 }
 803
 804 static inline bool bpf_jit_blinding_enabled(void)
 805 {
 806         /* These are the prerequisites, should someone ever have the
 807          * idea to call blinding outside of them, we make sure to
 808          * bail out.
 809          */
 810         if (!bpf_jit_is_ebpf())
 811                 return false;
 812         if (!bpf_jit_enable)
 813                 return false;
 814         if (!bpf_jit_harden)
 815                 return false;
 816         if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
 817                 return false;
 818
 819         return true;
 820 }
 821
 822 static inline bool bpf_jit_kallsyms_enabled(void)
 823 {
 824         /* There are a couple of corner cases where kallsyms should
 825          * not be enabled f.e. on hardening.
 826          */
 827         if (bpf_jit_harden)
 828                 return false;
 829         if (!bpf_jit_kallsyms)
 830                 return false;
 831         if (bpf_jit_kallsyms == 1)
 832                 return true;
 833
 834         return false;
 835 }
 836
 837 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
 838                                  unsigned long *off, char *sym);
 839 bool is_bpf_text_address(unsigned long addr);
 840 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
 841                     char *sym);
 842
 843 static inline const char *
 844 bpf_address_lookup(unsigned long addr, unsigned long *size,
 845                    unsigned long *off, char **modname, char *sym)
 846 {
 847         const char *ret = __bpf_address_lookup(addr, size, off, sym);
 848
 849         if (ret && modname)
 850                 *modname = NULL;
 851         return ret;
 852 }
 853
 854 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
 855 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
 856
 857 #else /* CONFIG_BPF_JIT */
 858
 859 static inline bool ebpf_jit_enabled(void)
 860 {
 861         return false;
 862 }
 863
 864 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
 865 {
 866         return false;
 867 }
 868
 869 static inline void bpf_jit_free(struct bpf_prog *fp)
 870 {
 871         bpf_prog_unlock_free(fp);
 872 }
 873
 874 static inline bool bpf_jit_kallsyms_enabled(void)
 875 {
 876         return false;
 877 }
 878
 879 static inline const char *
 880 __bpf_address_lookup(unsigned long addr, unsigned long *size,
 881                      unsigned long *off, char *sym)
 882 {
 883         return NULL;
 884 }
 885
 886 static inline bool is_bpf_text_address(unsigned long addr)
 887 {
 888         return false;
 889 }
 890
 891 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
 892                                   char *type, char *sym)
 893 {
 894         return -ERANGE;
 895 }
 896
 897 static inline const char *
 898 bpf_address_lookup(unsigned long addr, unsigned long *size,
 899                    unsigned long *off, char **modname, char *sym)
 900 {
 901         return NULL;
 902 }
 903
 904 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
 905 {
 906 }
 907
 908 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
 909 {
 910 }
 911 #endif /* CONFIG_BPF_JIT */
 912
 913 #define BPF_ANC         BIT(15)
 914
 915 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
 916 {
 917         switch (first->code) {
 918         case BPF_RET | BPF_K:
 919         case BPF_LD | BPF_W | BPF_LEN:
 920                 return false;
 921
 922         case BPF_LD | BPF_W | BPF_ABS:
 923         case BPF_LD | BPF_H | BPF_ABS:
 924         case BPF_LD | BPF_B | BPF_ABS:
 925                 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
 926                         return true;
 927                 return false;
 928
 929         default:
 930                 return true;
 931         }
 932 }
 933
 934 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
 935 {
 936         BUG_ON(ftest->code & BPF_ANC);
 937
 938         switch (ftest->code) {
 939         case BPF_LD | BPF_W | BPF_ABS:
 940         case BPF_LD | BPF_H | BPF_ABS:
 941         case BPF_LD | BPF_B | BPF_ABS:
 942 #define BPF_ANCILLARY(CODE)     case SKF_AD_OFF + SKF_AD_##CODE:        \
 943                                 return BPF_ANC | SKF_AD_##CODE
 944                 switch (ftest->k) {
 945                 BPF_ANCILLARY(PROTOCOL);
 946                 BPF_ANCILLARY(PKTTYPE);
 947                 BPF_ANCILLARY(IFINDEX);
 948                 BPF_ANCILLARY(NLATTR);
 949                 BPF_ANCILLARY(NLATTR_NEST);
 950                 BPF_ANCILLARY(MARK);
 951                 BPF_ANCILLARY(QUEUE);
 952                 BPF_ANCILLARY(HATYPE);
 953                 BPF_ANCILLARY(RXHASH);
 954                 BPF_ANCILLARY(CPU);
 955                 BPF_ANCILLARY(ALU_XOR_X);
 956                 BPF_ANCILLARY(VLAN_TAG);
 957                 BPF_ANCILLARY(VLAN_TAG_PRESENT);
 958                 BPF_ANCILLARY(PAY_OFFSET);
 959                 BPF_ANCILLARY(RANDOM);
 960                 BPF_ANCILLARY(VLAN_TPID);
 961                 }
 962                 /* Fallthrough. */
 963         default:
 964                 return ftest->code;
 965         }
 966 }
 967
 968 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
 969                                            int k, unsigned int size);
 970
 971 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
 972                                      unsigned int size, void *buffer)
 973 {
 974         if (k >= 0)
 975                 return skb_header_pointer(skb, k, size, buffer);
 976
 977         return bpf_internal_load_pointer_neg_helper(skb, k, size);
 978 }
 979
 980 static inline int bpf_tell_extensions(void)
 981 {
 982         return SKF_AD_MAX;
 983 }
 984
 985 struct bpf_sock_ops_kern {
 986         struct  sock *sk;
 987         u32     op;
 988         union {
 989                 u32 reply;
 990                 u32 replylong[4];
 991         };
 992 };
 993
 994 #endif /* __LINUX_FILTER_H__ */