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