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 for hardening step.
  50  * Only used by eBPF JITs. It's nothing more than a temporary
  51  * register that JITs use internally, only that here it's part
  52  * of eBPF instructions that have been rewritten for blinding
  53  * constants. See JIT pre-step in bpf_jit_blind_constants().
  54  */
  55 #define BPF_REG_AX              MAX_BPF_REG
  56 #define MAX_BPF_JIT_REG         (MAX_BPF_REG + 1)
  57
  58 /* unused opcode to mark special call to bpf_tail_call() helper */
  59 #define BPF_TAIL_CALL   0xf0
  60
  61 /* As per nm, we expose JITed images as text (code) section for
  62  * kallsyms. That way, tools like perf can find it to match
  63  * addresses.
  64  */
  65 #define BPF_SYM_ELF_TYPE        't'
  66
  67 /* BPF program can access up to 512 bytes of stack space. */
  68 #define MAX_BPF_STACK   512
  69
  70 /* Helper macros for filter block array initializers. */
  71
  72 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
  73
  74 #define BPF_ALU64_REG(OP, DST, SRC)                             \
  75         ((struct bpf_insn) {                                    \
  76                 .code  = BPF_ALU64 | BPF_OP(OP) | BPF_X,        \
  77                 .dst_reg = DST,                                 \
  78                 .src_reg = SRC,                                 \
  79                 .off   = 0,                                     \
  80                 .imm   = 0 })
  81
  82 #define BPF_ALU32_REG(OP, DST, SRC)                             \
  83         ((struct bpf_insn) {                                    \
  84                 .code  = BPF_ALU | BPF_OP(OP) | BPF_X,          \
  85                 .dst_reg = DST,                                 \
  86                 .src_reg = SRC,                                 \
  87                 .off   = 0,                                     \
  88                 .imm   = 0 })
  89
  90 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
  91
  92 #define BPF_ALU64_IMM(OP, DST, IMM)                             \
  93         ((struct bpf_insn) {                                    \
  94                 .code  = BPF_ALU64 | BPF_OP(OP) | BPF_K,        \
  95                 .dst_reg = DST,                                 \
  96                 .src_reg = 0,                                   \
  97                 .off   = 0,                                     \
  98                 .imm   = IMM })
  99
 100 #define BPF_ALU32_IMM(OP, DST, IMM)                             \
 101         ((struct bpf_insn) {                                    \
 102                 .code  = BPF_ALU | BPF_OP(OP) | BPF_K,          \
 103                 .dst_reg = DST,                                 \
 104                 .src_reg = 0,                                   \
 105                 .off   = 0,                                     \
 106                 .imm   = IMM })
 107
 108 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
 109
 110 #define BPF_ENDIAN(TYPE, DST, LEN)                              \
 111         ((struct bpf_insn) {                                    \
 112                 .code  = BPF_ALU | BPF_END | BPF_SRC(TYPE),     \
 113                 .dst_reg = DST,                                 \
 114                 .src_reg = 0,                                   \
 115                 .off   = 0,                                     \
 116                 .imm   = LEN })
 117
 118 /* Short form of mov, dst_reg = src_reg */
 119
 120 #define BPF_MOV64_REG(DST, SRC)                                 \
 121         ((struct bpf_insn) {                                    \
 122                 .code  = BPF_ALU64 | BPF_MOV | BPF_X,           \
 123                 .dst_reg = DST,                                 \
 124                 .src_reg = SRC,                                 \
 125                 .off   = 0,                                     \
 126                 .imm   = 0 })
 127
 128 #define BPF_MOV32_REG(DST, SRC)                                 \
 129         ((struct bpf_insn) {                                    \
 130                 .code  = BPF_ALU | BPF_MOV | BPF_X,             \
 131                 .dst_reg = DST,                                 \
 132                 .src_reg = SRC,                                 \
 133                 .off   = 0,                                     \
 134                 .imm   = 0 })
 135
 136 /* Short form of mov, dst_reg = imm32 */
 137
 138 #define BPF_MOV64_IMM(DST, IMM)                                 \
 139         ((struct bpf_insn) {                                    \
 140                 .code  = BPF_ALU64 | BPF_MOV | BPF_K,           \
 141                 .dst_reg = DST,                                 \
 142                 .src_reg = 0,                                   \
 143                 .off   = 0,                                     \
 144                 .imm   = IMM })
 145
 146 #define BPF_MOV32_IMM(DST, IMM)                                 \
 147         ((struct bpf_insn) {                                    \
 148                 .code  = BPF_ALU | BPF_MOV | BPF_K,             \
 149                 .dst_reg = DST,                                 \
 150                 .src_reg = 0,                                   \
 151                 .off   = 0,                                     \
 152                 .imm   = IMM })
 153
 154 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
 155 #define BPF_LD_IMM64(DST, IMM)                                  \
 156         BPF_LD_IMM64_RAW(DST, 0, IMM)
 157
 158 #define BPF_LD_IMM64_RAW(DST, SRC, IMM)                         \
 159         ((struct bpf_insn) {                                    \
 160                 .code  = BPF_LD | BPF_DW | BPF_IMM,             \
 161                 .dst_reg = DST,                                 \
 162                 .src_reg = SRC,                                 \
 163                 .off   = 0,                                     \
 164                 .imm   = (__u32) (IMM) }),                      \
 165         ((struct bpf_insn) {                                    \
 166                 .code  = 0, /* zero is reserved opcode */       \
 167                 .dst_reg = 0,                                   \
 168                 .src_reg = 0,                                   \
 169                 .off   = 0,                                     \
 170                 .imm   = ((__u64) (IMM)) >> 32 })
 171
 172 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
 173 #define BPF_LD_MAP_FD(DST, MAP_FD)                              \
 174         BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
 175
 176 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
 177
 178 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM)                      \
 179         ((struct bpf_insn) {                                    \
 180                 .code  = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE),   \
 181                 .dst_reg = DST,                                 \
 182                 .src_reg = SRC,                                 \
 183                 .off   = 0,                                     \
 184                 .imm   = IMM })
 185
 186 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM)                      \
 187         ((struct bpf_insn) {                                    \
 188                 .code  = BPF_ALU | BPF_MOV | BPF_SRC(TYPE),     \
 189                 .dst_reg = DST,                                 \
 190                 .src_reg = SRC,                                 \
 191                 .off   = 0,                                     \
 192                 .imm   = IMM })
 193
 194 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
 195
 196 #define BPF_LD_ABS(SIZE, IMM)                                   \
 197         ((struct bpf_insn) {                                    \
 198                 .code  = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS,     \
 199                 .dst_reg = 0,                                   \
 200                 .src_reg = 0,                                   \
 201                 .off   = 0,                                     \
 202                 .imm   = IMM })
 203
 204 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
 205
 206 #define BPF_LD_IND(SIZE, SRC, IMM)                              \
 207         ((struct bpf_insn) {                                    \
 208                 .code  = BPF_LD | BPF_SIZE(SIZE) | BPF_IND,     \
 209                 .dst_reg = 0,                                   \
 210                 .src_reg = SRC,                                 \
 211                 .off   = 0,                                     \
 212                 .imm   = IMM })
 213
 214 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
 215
 216 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF)                        \
 217         ((struct bpf_insn) {                                    \
 218                 .code  = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM,    \
 219                 .dst_reg = DST,                                 \
 220                 .src_reg = SRC,                                 \
 221                 .off   = OFF,                                   \
 222                 .imm   = 0 })
 223
 224 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
 225
 226 #define BPF_STX_MEM(SIZE, DST, SRC, OFF)                        \
 227         ((struct bpf_insn) {                                    \
 228                 .code  = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM,    \
 229                 .dst_reg = DST,                                 \
 230                 .src_reg = SRC,                                 \
 231                 .off   = OFF,                                   \
 232                 .imm   = 0 })
 233
 234 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
 235
 236 #define BPF_STX_XADD(SIZE, DST, SRC, OFF)                       \
 237         ((struct bpf_insn) {                                    \
 238                 .code  = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD,   \
 239                 .dst_reg = DST,                                 \
 240                 .src_reg = SRC,                                 \
 241                 .off   = OFF,                                   \
 242                 .imm   = 0 })
 243
 244 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
 245
 246 #define BPF_ST_MEM(SIZE, DST, OFF, IMM)                         \
 247         ((struct bpf_insn) {                                    \
 248                 .code  = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM,     \
 249                 .dst_reg = DST,                                 \
 250                 .src_reg = 0,                                   \
 251                 .off   = OFF,                                   \
 252                 .imm   = IMM })
 253
 254 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
 255
 256 #define BPF_JMP_REG(OP, DST, SRC, OFF)                          \
 257         ((struct bpf_insn) {                                    \
 258                 .code  = BPF_JMP | BPF_OP(OP) | BPF_X,          \
 259                 .dst_reg = DST,                                 \
 260                 .src_reg = SRC,                                 \
 261                 .off   = OFF,                                   \
 262                 .imm   = 0 })
 263
 264 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
 265
 266 #define BPF_JMP_IMM(OP, DST, IMM, OFF)                          \
 267         ((struct bpf_insn) {                                    \
 268                 .code  = BPF_JMP | BPF_OP(OP) | BPF_K,          \
 269                 .dst_reg = DST,                                 \
 270                 .src_reg = 0,                                   \
 271                 .off   = OFF,                                   \
 272                 .imm   = IMM })
 273
 274 /* Unconditional jumps, goto pc + off16 */
 275
 276 #define BPF_JMP_A(OFF)                                          \
 277         ((struct bpf_insn) {                                    \
 278                 .code  = BPF_JMP | BPF_JA,                      \
 279                 .dst_reg = 0,                                   \
 280                 .src_reg = 0,                                   \
 281                 .off   = OFF,                                   \
 282                 .imm   = 0 })
 283
 284 /* Function call */
 285
 286 #define BPF_EMIT_CALL(FUNC)                                     \
 287         ((struct bpf_insn) {                                    \
 288                 .code  = BPF_JMP | BPF_CALL,                    \
 289                 .dst_reg = 0,                                   \
 290                 .src_reg = 0,                                   \
 291                 .off   = 0,                                     \
 292                 .imm   = ((FUNC) - __bpf_call_base) })
 293
 294 /* Raw code statement block */
 295
 296 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM)                  \
 297         ((struct bpf_insn) {                                    \
 298                 .code  = CODE,                                  \
 299                 .dst_reg = DST,                                 \
 300                 .src_reg = SRC,                                 \
 301                 .off   = OFF,                                   \
 302                 .imm   = IMM })
 303
 304 /* Program exit */
 305
 306 #define BPF_EXIT_INSN()                                         \
 307         ((struct bpf_insn) {                                    \
 308                 .code  = BPF_JMP | BPF_EXIT,                    \
 309                 .dst_reg = 0,                                   \
 310                 .src_reg = 0,                                   \
 311                 .off   = 0,                                     \
 312                 .imm   = 0 })
 313
 314 /* Internal classic blocks for direct assignment */
 315
 316 #define __BPF_STMT(CODE, K)                                     \
 317         ((struct sock_filter) BPF_STMT(CODE, K))
 318
 319 #define __BPF_JUMP(CODE, K, JT, JF)                             \
 320         ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
 321
 322 #define bytes_to_bpf_size(bytes)                                \
 323 ({                                                              \
 324         int bpf_size = -EINVAL;                                 \
 325                                                                 \
 326         if (bytes == sizeof(u8))                                \
 327                 bpf_size = BPF_B;                               \
 328         else if (bytes == sizeof(u16))                          \
 329                 bpf_size = BPF_H;                               \
 330         else if (bytes == sizeof(u32))                          \
 331                 bpf_size = BPF_W;                               \
 332         else if (bytes == sizeof(u64))                          \
 333                 bpf_size = BPF_DW;                              \
 334                                                                 \
 335         bpf_size;                                               \
 336 })
 337
 338 #define bpf_size_to_bytes(bpf_size)                             \
 339 ({                                                              \
 340         int bytes = -EINVAL;                                    \
 341                                                                 \
 342         if (bpf_size == BPF_B)                                  \
 343                 bytes = sizeof(u8);                             \
 344         else if (bpf_size == BPF_H)                             \
 345                 bytes = sizeof(u16);                            \
 346         else if (bpf_size == BPF_W)                             \
 347                 bytes = sizeof(u32);                            \
 348         else if (bpf_size == BPF_DW)                            \
 349                 bytes = sizeof(u64);                            \
 350                                                                 \
 351         bytes;                                                  \
 352 })
 353
 354 #define BPF_SIZEOF(type)                                        \
 355         ({                                                      \
 356                 const int __size = bytes_to_bpf_size(sizeof(type)); \
 357                 BUILD_BUG_ON(__size < 0);                       \
 358                 __size;                                         \
 359         })
 360
 361 #define BPF_FIELD_SIZEOF(type, field)                           \
 362         ({                                                      \
 363                 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
 364                 BUILD_BUG_ON(__size < 0);                       \
 365                 __size;                                         \
 366         })
 367
 368 #define BPF_LDST_BYTES(insn)                                    \
 369         ({                                                      \
 370                 const int __size = bpf_size_to_bytes(BPF_SIZE(insn->code)); \
 371                 WARN_ON(__size < 0);                            \
 372                 __size;                                         \
 373         })
 374
 375 #define __BPF_MAP_0(m, v, ...) v
 376 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
 377 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
 378 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
 379 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
 380 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
 381
 382 #define __BPF_REG_0(...) __BPF_PAD(5)
 383 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
 384 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
 385 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
 386 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
 387 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
 388
 389 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
 390 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
 391
 392 #define __BPF_CAST(t, a)                                                       \
 393         (__force t)                                                            \
 394         (__force                                                               \
 395          typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long),      \
 396                                       (unsigned long)0, (t)0))) a
 397 #define __BPF_V void
 398 #define __BPF_N
 399
 400 #define __BPF_DECL_ARGS(t, a) t   a
 401 #define __BPF_DECL_REGS(t, a) u64 a
 402
 403 #define __BPF_PAD(n)                                                           \
 404         __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2,       \
 405                   u64, __ur_3, u64, __ur_4, u64, __ur_5)
 406
 407 #define BPF_CALL_x(x, name, ...)                                               \
 408         static __always_inline                                                 \
 409         u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__));   \
 410         u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__));         \
 411         u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__))          \
 412         {                                                                      \
 413                 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
 414         }                                                                      \
 415         static __always_inline                                                 \
 416         u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
 417
 418 #define BPF_CALL_0(name, ...)   BPF_CALL_x(0, name, __VA_ARGS__)
 419 #define BPF_CALL_1(name, ...)   BPF_CALL_x(1, name, __VA_ARGS__)
 420 #define BPF_CALL_2(name, ...)   BPF_CALL_x(2, name, __VA_ARGS__)
 421 #define BPF_CALL_3(name, ...)   BPF_CALL_x(3, name, __VA_ARGS__)
 422 #define BPF_CALL_4(name, ...)   BPF_CALL_x(4, name, __VA_ARGS__)
 423 #define BPF_CALL_5(name, ...)   BPF_CALL_x(5, name, __VA_ARGS__)
 424
 425 #define bpf_ctx_range(TYPE, MEMBER)                                             \
 426         offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
 427 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2)                              \
 428         offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
 429
 430 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE)                            \
 431         ({                                                                      \
 432                 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE));             \
 433                 *(PTR_SIZE) = (SIZE);                                           \
 434                 offsetof(TYPE, MEMBER);                                         \
 435         })
 436
 437 #ifdef CONFIG_COMPAT
 438 /* A struct sock_filter is architecture independent. */
 439 struct compat_sock_fprog {
 440         u16             len;
 441         compat_uptr_t   filter; /* struct sock_filter * */
 442 };
 443 #endif
 444
 445 struct sock_fprog_kern {
 446         u16                     len;
 447         struct sock_filter      *filter;
 448 };
 449
 450 struct bpf_binary_header {
 451         unsigned int pages;
 452         u8 image[];
 453 };
 454
 455 struct bpf_prog {
 456         u16                     pages;          /* Number of allocated pages */
 457         kmemcheck_bitfield_begin(meta);
 458         u16                     jited:1,        /* Is our filter JIT'ed? */
 459                                 locked:1,       /* Program image locked? */
 460                                 gpl_compatible:1, /* Is filter GPL compatible? */
 461                                 cb_access:1,    /* Is control block accessed? */
 462                                 dst_needed:1;   /* Do we need dst entry? */
 463         kmemcheck_bitfield_end(meta);
 464         enum bpf_prog_type      type;           /* Type of BPF program */
 465         u32                     len;            /* Number of filter blocks */
 466         u32                     jited_len;      /* Size of jited insns in bytes */
 467         u8                      tag[BPF_TAG_SIZE];
 468         struct bpf_prog_aux     *aux;           /* Auxiliary fields */
 469         struct sock_fprog_kern  *orig_prog;     /* Original BPF program */
 470         unsigned int            (*bpf_func)(const void *ctx,
 471                                             const struct bpf_insn *insn);
 472         /* Instructions for interpreter */
 473         union {
 474                 struct sock_filter      insns[0];
 475                 struct bpf_insn         insnsi[0];
 476         };
 477 };
 478
 479 struct sk_filter {
 480         refcount_t      refcnt;
 481         struct rcu_head rcu;
 482         struct bpf_prog *prog;
 483 };
 484
 485 #define BPF_PROG_RUN(filter, ctx)  (*filter->bpf_func)(ctx, filter->insnsi)
 486
 487 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
 488
 489 struct bpf_skb_data_end {
 490         struct qdisc_skb_cb qdisc_cb;
 491         void *data_end;
 492 };
 493
 494 struct xdp_buff {
 495         void *data;
 496         void *data_end;
 497         void *data_hard_start;
 498 };
 499
 500 /* compute the linear packet data range [data, data_end) which
 501  * will be accessed by cls_bpf, act_bpf and lwt programs
 502  */
 503 static inline void bpf_compute_data_end(struct sk_buff *skb)
 504 {
 505         struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
 506
 507         BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
 508         cb->data_end = skb->data + skb_headlen(skb);
 509 }
 510
 511 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
 512 {
 513         /* eBPF programs may read/write skb->cb[] area to transfer meta
 514          * data between tail calls. Since this also needs to work with
 515          * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
 516          *
 517          * In some socket filter cases, the cb unfortunately needs to be
 518          * saved/restored so that protocol specific skb->cb[] data won't
 519          * be lost. In any case, due to unpriviledged eBPF programs
 520          * attached to sockets, we need to clear the bpf_skb_cb() area
 521          * to not leak previous contents to user space.
 522          */
 523         BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
 524         BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
 525                      FIELD_SIZEOF(struct qdisc_skb_cb, data));
 526
 527         return qdisc_skb_cb(skb)->data;
 528 }
 529
 530 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
 531                                        struct sk_buff *skb)
 532 {
 533         u8 *cb_data = bpf_skb_cb(skb);
 534         u8 cb_saved[BPF_SKB_CB_LEN];
 535         u32 res;
 536
 537         if (unlikely(prog->cb_access)) {
 538                 memcpy(cb_saved, cb_data, sizeof(cb_saved));
 539                 memset(cb_data, 0, sizeof(cb_saved));
 540         }
 541
 542         res = BPF_PROG_RUN(prog, skb);
 543
 544         if (unlikely(prog->cb_access))
 545                 memcpy(cb_data, cb_saved, sizeof(cb_saved));
 546
 547         return res;
 548 }
 549
 550 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
 551                                         struct sk_buff *skb)
 552 {
 553         u8 *cb_data = bpf_skb_cb(skb);
 554
 555         if (unlikely(prog->cb_access))
 556                 memset(cb_data, 0, BPF_SKB_CB_LEN);
 557
 558         return BPF_PROG_RUN(prog, skb);
 559 }
 560
 561 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
 562                                             struct xdp_buff *xdp)
 563 {
 564         /* Caller needs to hold rcu_read_lock() (!), otherwise program
 565          * can be released while still running, or map elements could be
 566          * freed early while still having concurrent users. XDP fastpath
 567          * already takes rcu_read_lock() when fetching the program, so
 568          * it's not necessary here anymore.
 569          */
 570         return BPF_PROG_RUN(prog, xdp);
 571 }
 572
 573 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
 574 {
 575         return prog->len * sizeof(struct bpf_insn);
 576 }
 577
 578 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
 579 {
 580         return round_up(bpf_prog_insn_size(prog) +
 581                         sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
 582 }
 583
 584 static inline unsigned int bpf_prog_size(unsigned int proglen)
 585 {
 586         return max(sizeof(struct bpf_prog),
 587                    offsetof(struct bpf_prog, insns[proglen]));
 588 }
 589
 590 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
 591 {
 592         /* When classic BPF programs have been loaded and the arch
 593          * does not have a classic BPF JIT (anymore), they have been
 594          * converted via bpf_migrate_filter() to eBPF and thus always
 595          * have an unspec program type.
 596          */
 597         return prog->type == BPF_PROG_TYPE_UNSPEC;
 598 }
 599
 600 static inline bool
 601 bpf_ctx_narrow_access_ok(u32 off, u32 size, const u32 size_default)
 602 {
 603         bool off_ok;
 604 #ifdef __LITTLE_ENDIAN
 605         off_ok = (off & (size_default - 1)) == 0;
 606 #else
 607         off_ok = (off & (size_default - 1)) + size == size_default;
 608 #endif
 609         return off_ok && size <= size_default && (size & (size - 1)) == 0;
 610 }
 611
 612 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
 613
 614 #ifdef CONFIG_ARCH_HAS_SET_MEMORY
 615 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
 616 {
 617         fp->locked = 1;
 618         WARN_ON_ONCE(set_memory_ro((unsigned long)fp, fp->pages));
 619 }
 620
 621 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
 622 {
 623         if (fp->locked) {
 624                 WARN_ON_ONCE(set_memory_rw((unsigned long)fp, fp->pages));
 625                 /* In case set_memory_rw() fails, we want to be the first
 626                  * to crash here instead of some random place later on.
 627                  */
 628                 fp->locked = 0;
 629         }
 630 }
 631
 632 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
 633 {
 634         WARN_ON_ONCE(set_memory_ro((unsigned long)hdr, hdr->pages));
 635 }
 636
 637 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
 638 {
 639         WARN_ON_ONCE(set_memory_rw((unsigned long)hdr, hdr->pages));
 640 }
 641 #else
 642 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
 643 {
 644 }
 645
 646 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
 647 {
 648 }
 649
 650 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
 651 {
 652 }
 653
 654 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
 655 {
 656 }
 657 #endif /* CONFIG_ARCH_HAS_SET_MEMORY */
 658
 659 static inline struct bpf_binary_header *
 660 bpf_jit_binary_hdr(const struct bpf_prog *fp)
 661 {
 662         unsigned long real_start = (unsigned long)fp->bpf_func;
 663         unsigned long addr = real_start & PAGE_MASK;
 664
 665         return (void *)addr;
 666 }
 667
 668 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
 669 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
 670 {
 671         return sk_filter_trim_cap(sk, skb, 1);
 672 }
 673
 674 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
 675 void bpf_prog_free(struct bpf_prog *fp);
 676
 677 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
 678 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
 679                                   gfp_t gfp_extra_flags);
 680 void __bpf_prog_free(struct bpf_prog *fp);
 681
 682 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
 683 {
 684         bpf_prog_unlock_ro(fp);
 685         __bpf_prog_free(fp);
 686 }
 687
 688 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
 689                                        unsigned int flen);
 690
 691 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
 692 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
 693                               bpf_aux_classic_check_t trans, bool save_orig);
 694 void bpf_prog_destroy(struct bpf_prog *fp);
 695
 696 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
 697 int sk_attach_bpf(u32 ufd, struct sock *sk);
 698 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
 699 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
 700 int sk_detach_filter(struct sock *sk);
 701 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
 702                   unsigned int len);
 703
 704 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
 705 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
 706
 707 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
 708
 709 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
 710 void bpf_jit_compile(struct bpf_prog *prog);
 711 bool bpf_helper_changes_pkt_data(void *func);
 712
 713 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 714                                        const struct bpf_insn *patch, u32 len);
 715
 716 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
 717  * same cpu context. Further for best results no more than a single map
 718  * for the do_redirect/do_flush pair should be used. This limitation is
 719  * because we only track one map and force a flush when the map changes.
 720  * This does not appear to be a real limitation for existing software.
 721  */
 722 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
 723                             struct bpf_prog *prog);
 724 int xdp_do_redirect(struct net_device *dev,
 725                     struct xdp_buff *xdp,
 726                     struct bpf_prog *prog);
 727 void xdp_do_flush_map(void);
 728
 729 void bpf_warn_invalid_xdp_action(u32 act);
 730 void bpf_warn_invalid_xdp_redirect(u32 ifindex);
 731
 732 struct sock *do_sk_redirect_map(struct sk_buff *skb);
 733
 734 #ifdef CONFIG_BPF_JIT
 735 extern int bpf_jit_enable;
 736 extern int bpf_jit_harden;
 737 extern int bpf_jit_kallsyms;
 738
 739 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
 740
 741 struct bpf_binary_header *
 742 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
 743                      unsigned int alignment,
 744                      bpf_jit_fill_hole_t bpf_fill_ill_insns);
 745 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
 746
 747 void bpf_jit_free(struct bpf_prog *fp);
 748
 749 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
 750 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
 751
 752 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
 753                                 u32 pass, void *image)
 754 {
 755         pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
 756                proglen, pass, image, current->comm, task_pid_nr(current));
 757
 758         if (image)
 759                 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
 760                                16, 1, image, proglen, false);
 761 }
 762
 763 static inline bool bpf_jit_is_ebpf(void)
 764 {
 765 # ifdef CONFIG_HAVE_EBPF_JIT
 766         return true;
 767 # else
 768         return false;
 769 # endif
 770 }
 771
 772 static inline bool ebpf_jit_enabled(void)
 773 {
 774         return bpf_jit_enable && bpf_jit_is_ebpf();
 775 }
 776
 777 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
 778 {
 779         return fp->jited && bpf_jit_is_ebpf();
 780 }
 781
 782 static inline bool bpf_jit_blinding_enabled(void)
 783 {
 784         /* These are the prerequisites, should someone ever have the
 785          * idea to call blinding outside of them, we make sure to
 786          * bail out.
 787          */
 788         if (!bpf_jit_is_ebpf())
 789                 return false;
 790         if (!bpf_jit_enable)
 791                 return false;
 792         if (!bpf_jit_harden)
 793                 return false;
 794         if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
 795                 return false;
 796
 797         return true;
 798 }
 799
 800 static inline bool bpf_jit_kallsyms_enabled(void)
 801 {
 802         /* There are a couple of corner cases where kallsyms should
 803          * not be enabled f.e. on hardening.
 804          */
 805         if (bpf_jit_harden)
 806                 return false;
 807         if (!bpf_jit_kallsyms)
 808                 return false;
 809         if (bpf_jit_kallsyms == 1)
 810                 return true;
 811
 812         return false;
 813 }
 814
 815 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
 816                                  unsigned long *off, char *sym);
 817 bool is_bpf_text_address(unsigned long addr);
 818 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
 819                     char *sym);
 820
 821 static inline const char *
 822 bpf_address_lookup(unsigned long addr, unsigned long *size,
 823                    unsigned long *off, char **modname, char *sym)
 824 {
 825         const char *ret = __bpf_address_lookup(addr, size, off, sym);
 826
 827         if (ret && modname)
 828                 *modname = NULL;
 829         return ret;
 830 }
 831
 832 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
 833 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
 834
 835 #else /* CONFIG_BPF_JIT */
 836
 837 static inline bool ebpf_jit_enabled(void)
 838 {
 839         return false;
 840 }
 841
 842 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
 843 {
 844         return false;
 845 }
 846
 847 static inline void bpf_jit_free(struct bpf_prog *fp)
 848 {
 849         bpf_prog_unlock_free(fp);
 850 }
 851
 852 static inline bool bpf_jit_kallsyms_enabled(void)
 853 {
 854         return false;
 855 }
 856
 857 static inline const char *
 858 __bpf_address_lookup(unsigned long addr, unsigned long *size,
 859                      unsigned long *off, char *sym)
 860 {
 861         return NULL;
 862 }
 863
 864 static inline bool is_bpf_text_address(unsigned long addr)
 865 {
 866         return false;
 867 }
 868
 869 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
 870                                   char *type, char *sym)
 871 {
 872         return -ERANGE;
 873 }
 874
 875 static inline const char *
 876 bpf_address_lookup(unsigned long addr, unsigned long *size,
 877                    unsigned long *off, char **modname, char *sym)
 878 {
 879         return NULL;
 880 }
 881
 882 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
 883 {
 884 }
 885
 886 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
 887 {
 888 }
 889 #endif /* CONFIG_BPF_JIT */
 890
 891 #define BPF_ANC         BIT(15)
 892
 893 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
 894 {
 895         switch (first->code) {
 896         case BPF_RET | BPF_K:
 897         case BPF_LD | BPF_W | BPF_LEN:
 898                 return false;
 899
 900         case BPF_LD | BPF_W | BPF_ABS:
 901         case BPF_LD | BPF_H | BPF_ABS:
 902         case BPF_LD | BPF_B | BPF_ABS:
 903                 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
 904                         return true;
 905                 return false;
 906
 907         default:
 908                 return true;
 909         }
 910 }
 911
 912 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
 913 {
 914         BUG_ON(ftest->code & BPF_ANC);
 915
 916         switch (ftest->code) {
 917         case BPF_LD | BPF_W | BPF_ABS:
 918         case BPF_LD | BPF_H | BPF_ABS:
 919         case BPF_LD | BPF_B | BPF_ABS:
 920 #define BPF_ANCILLARY(CODE)     case SKF_AD_OFF + SKF_AD_##CODE:        \
 921                                 return BPF_ANC | SKF_AD_##CODE
 922                 switch (ftest->k) {
 923                 BPF_ANCILLARY(PROTOCOL);
 924                 BPF_ANCILLARY(PKTTYPE);
 925                 BPF_ANCILLARY(IFINDEX);
 926                 BPF_ANCILLARY(NLATTR);
 927                 BPF_ANCILLARY(NLATTR_NEST);
 928                 BPF_ANCILLARY(MARK);
 929                 BPF_ANCILLARY(QUEUE);
 930                 BPF_ANCILLARY(HATYPE);
 931                 BPF_ANCILLARY(RXHASH);
 932                 BPF_ANCILLARY(CPU);
 933                 BPF_ANCILLARY(ALU_XOR_X);
 934                 BPF_ANCILLARY(VLAN_TAG);
 935                 BPF_ANCILLARY(VLAN_TAG_PRESENT);
 936                 BPF_ANCILLARY(PAY_OFFSET);
 937                 BPF_ANCILLARY(RANDOM);
 938                 BPF_ANCILLARY(VLAN_TPID);
 939                 }
 940                 /* Fallthrough. */
 941         default:
 942                 return ftest->code;
 943         }
 944 }
 945
 946 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
 947                                            int k, unsigned int size);
 948
 949 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
 950                                      unsigned int size, void *buffer)
 951 {
 952         if (k >= 0)
 953                 return skb_header_pointer(skb, k, size, buffer);
 954
 955         return bpf_internal_load_pointer_neg_helper(skb, k, size);
 956 }
 957
 958 static inline int bpf_tell_extensions(void)
 959 {
 960         return SKF_AD_MAX;
 961 }
 962
 963 struct bpf_sock_ops_kern {
 964         struct  sock *sk;
 965         u32     op;
 966         union {
 967                 u32 reply;
 968                 u32 replylong[4];
 969         };
 970 };
 971
 972 #endif /* __LINUX_FILTER_H__ */