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/set_memory.h> 20 #include <linux/kallsyms.h> 21 #include <linux/if_vlan.h> 22 #include <linux/vmalloc.h> 23 #include <linux/sockptr.h> 24 #include <crypto/sha1.h> 25 #include <linux/u64_stats_sync.h> 26 27 #include <net/sch_generic.h> 28 29 #include <asm/byteorder.h> 30 #include <uapi/linux/filter.h> 31 #include <uapi/linux/bpf.h> 32 33 struct sk_buff; 34 struct sock; 35 struct seccomp_data; 36 struct bpf_prog_aux; 37 struct xdp_rxq_info; 38 struct xdp_buff; 39 struct sock_reuseport; 40 struct ctl_table; 41 struct ctl_table_header; 42 43 /* ArgX, context and stack frame pointer register positions. Note, 44 * Arg1, Arg2, Arg3, etc are used as argument mappings of function 45 * calls in BPF_CALL instruction. 46 */ 47 #define BPF_REG_ARG1 BPF_REG_1 48 #define BPF_REG_ARG2 BPF_REG_2 49 #define BPF_REG_ARG3 BPF_REG_3 50 #define BPF_REG_ARG4 BPF_REG_4 51 #define BPF_REG_ARG5 BPF_REG_5 52 #define BPF_REG_CTX BPF_REG_6 53 #define BPF_REG_FP BPF_REG_10 54 55 /* Additional register mappings for converted user programs. */ 56 #define BPF_REG_A BPF_REG_0 57 #define BPF_REG_X BPF_REG_7 58 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */ 59 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */ 60 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */ 61 62 /* Kernel hidden auxiliary/helper register. */ 63 #define BPF_REG_AX MAX_BPF_REG 64 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1) 65 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG 66 67 /* unused opcode to mark special call to bpf_tail_call() helper */ 68 #define BPF_TAIL_CALL 0xf0 69 70 /* unused opcode to mark special load instruction. Same as BPF_ABS */ 71 #define BPF_PROBE_MEM 0x20 72 73 /* unused opcode to mark call to interpreter with arguments */ 74 #define BPF_CALL_ARGS 0xe0 75 76 /* unused opcode to mark speculation barrier for mitigating 77 * Speculative Store Bypass 78 */ 79 #define BPF_NOSPEC 0xc0 80 81 /* As per nm, we expose JITed images as text (code) section for 82 * kallsyms. That way, tools like perf can find it to match 83 * addresses. 84 */ 85 #define BPF_SYM_ELF_TYPE 't' 86 87 /* BPF program can access up to 512 bytes of stack space. */ 88 #define MAX_BPF_STACK 512 89 90 /* Helper macros for filter block array initializers. */ 91 92 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 93 94 #define BPF_ALU64_REG(OP, DST, SRC) \ 95 ((struct bpf_insn) { \ 96 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 97 .dst_reg = DST, \ 98 .src_reg = SRC, \ 99 .off = 0, \ 100 .imm = 0 }) 101 102 #define BPF_ALU32_REG(OP, DST, SRC) \ 103 ((struct bpf_insn) { \ 104 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 105 .dst_reg = DST, \ 106 .src_reg = SRC, \ 107 .off = 0, \ 108 .imm = 0 }) 109 110 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 111 112 #define BPF_ALU64_IMM(OP, DST, IMM) \ 113 ((struct bpf_insn) { \ 114 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 115 .dst_reg = DST, \ 116 .src_reg = 0, \ 117 .off = 0, \ 118 .imm = IMM }) 119 120 #define BPF_ALU32_IMM(OP, DST, IMM) \ 121 ((struct bpf_insn) { \ 122 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 123 .dst_reg = DST, \ 124 .src_reg = 0, \ 125 .off = 0, \ 126 .imm = IMM }) 127 128 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 129 130 #define BPF_ENDIAN(TYPE, DST, LEN) \ 131 ((struct bpf_insn) { \ 132 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 133 .dst_reg = DST, \ 134 .src_reg = 0, \ 135 .off = 0, \ 136 .imm = LEN }) 137 138 /* Short form of mov, dst_reg = src_reg */ 139 140 #define BPF_MOV64_REG(DST, SRC) \ 141 ((struct bpf_insn) { \ 142 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 143 .dst_reg = DST, \ 144 .src_reg = SRC, \ 145 .off = 0, \ 146 .imm = 0 }) 147 148 #define BPF_MOV32_REG(DST, SRC) \ 149 ((struct bpf_insn) { \ 150 .code = BPF_ALU | BPF_MOV | BPF_X, \ 151 .dst_reg = DST, \ 152 .src_reg = SRC, \ 153 .off = 0, \ 154 .imm = 0 }) 155 156 /* Short form of mov, dst_reg = imm32 */ 157 158 #define BPF_MOV64_IMM(DST, IMM) \ 159 ((struct bpf_insn) { \ 160 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 161 .dst_reg = DST, \ 162 .src_reg = 0, \ 163 .off = 0, \ 164 .imm = IMM }) 165 166 #define BPF_MOV32_IMM(DST, IMM) \ 167 ((struct bpf_insn) { \ 168 .code = BPF_ALU | BPF_MOV | BPF_K, \ 169 .dst_reg = DST, \ 170 .src_reg = 0, \ 171 .off = 0, \ 172 .imm = IMM }) 173 174 /* Special form of mov32, used for doing explicit zero extension on dst. */ 175 #define BPF_ZEXT_REG(DST) \ 176 ((struct bpf_insn) { \ 177 .code = BPF_ALU | BPF_MOV | BPF_X, \ 178 .dst_reg = DST, \ 179 .src_reg = DST, \ 180 .off = 0, \ 181 .imm = 1 }) 182 183 static inline bool insn_is_zext(const struct bpf_insn *insn) 184 { 185 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 186 } 187 188 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 189 #define BPF_LD_IMM64(DST, IMM) \ 190 BPF_LD_IMM64_RAW(DST, 0, IMM) 191 192 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 193 ((struct bpf_insn) { \ 194 .code = BPF_LD | BPF_DW | BPF_IMM, \ 195 .dst_reg = DST, \ 196 .src_reg = SRC, \ 197 .off = 0, \ 198 .imm = (__u32) (IMM) }), \ 199 ((struct bpf_insn) { \ 200 .code = 0, /* zero is reserved opcode */ \ 201 .dst_reg = 0, \ 202 .src_reg = 0, \ 203 .off = 0, \ 204 .imm = ((__u64) (IMM)) >> 32 }) 205 206 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 207 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 208 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 209 210 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 211 212 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 213 ((struct bpf_insn) { \ 214 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 215 .dst_reg = DST, \ 216 .src_reg = SRC, \ 217 .off = 0, \ 218 .imm = IMM }) 219 220 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 221 ((struct bpf_insn) { \ 222 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 223 .dst_reg = DST, \ 224 .src_reg = SRC, \ 225 .off = 0, \ 226 .imm = IMM }) 227 228 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 229 230 #define BPF_LD_ABS(SIZE, IMM) \ 231 ((struct bpf_insn) { \ 232 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 233 .dst_reg = 0, \ 234 .src_reg = 0, \ 235 .off = 0, \ 236 .imm = IMM }) 237 238 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 239 240 #define BPF_LD_IND(SIZE, SRC, IMM) \ 241 ((struct bpf_insn) { \ 242 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 243 .dst_reg = 0, \ 244 .src_reg = SRC, \ 245 .off = 0, \ 246 .imm = IMM }) 247 248 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 249 250 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 251 ((struct bpf_insn) { \ 252 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 253 .dst_reg = DST, \ 254 .src_reg = SRC, \ 255 .off = OFF, \ 256 .imm = 0 }) 257 258 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 259 260 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 261 ((struct bpf_insn) { \ 262 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 263 .dst_reg = DST, \ 264 .src_reg = SRC, \ 265 .off = OFF, \ 266 .imm = 0 }) 267 268 269 /* 270 * Atomic operations: 271 * 272 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 273 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 274 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 275 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 276 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 277 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 278 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 279 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 280 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 281 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 282 */ 283 284 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 285 ((struct bpf_insn) { \ 286 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 287 .dst_reg = DST, \ 288 .src_reg = SRC, \ 289 .off = OFF, \ 290 .imm = OP }) 291 292 /* Legacy alias */ 293 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 294 295 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 296 297 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 298 ((struct bpf_insn) { \ 299 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 300 .dst_reg = DST, \ 301 .src_reg = 0, \ 302 .off = OFF, \ 303 .imm = IMM }) 304 305 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 306 307 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 308 ((struct bpf_insn) { \ 309 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 310 .dst_reg = DST, \ 311 .src_reg = SRC, \ 312 .off = OFF, \ 313 .imm = 0 }) 314 315 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 316 317 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 318 ((struct bpf_insn) { \ 319 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 320 .dst_reg = DST, \ 321 .src_reg = 0, \ 322 .off = OFF, \ 323 .imm = IMM }) 324 325 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 326 327 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 328 ((struct bpf_insn) { \ 329 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 330 .dst_reg = DST, \ 331 .src_reg = SRC, \ 332 .off = OFF, \ 333 .imm = 0 }) 334 335 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 336 337 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 338 ((struct bpf_insn) { \ 339 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 340 .dst_reg = DST, \ 341 .src_reg = 0, \ 342 .off = OFF, \ 343 .imm = IMM }) 344 345 /* Unconditional jumps, goto pc + off16 */ 346 347 #define BPF_JMP_A(OFF) \ 348 ((struct bpf_insn) { \ 349 .code = BPF_JMP | BPF_JA, \ 350 .dst_reg = 0, \ 351 .src_reg = 0, \ 352 .off = OFF, \ 353 .imm = 0 }) 354 355 /* Relative call */ 356 357 #define BPF_CALL_REL(TGT) \ 358 ((struct bpf_insn) { \ 359 .code = BPF_JMP | BPF_CALL, \ 360 .dst_reg = 0, \ 361 .src_reg = BPF_PSEUDO_CALL, \ 362 .off = 0, \ 363 .imm = TGT }) 364 365 /* Function call */ 366 367 #define BPF_CAST_CALL(x) \ 368 ((u64 (*)(u64, u64, u64, u64, u64))(x)) 369 370 #define BPF_EMIT_CALL(FUNC) \ 371 ((struct bpf_insn) { \ 372 .code = BPF_JMP | BPF_CALL, \ 373 .dst_reg = 0, \ 374 .src_reg = 0, \ 375 .off = 0, \ 376 .imm = ((FUNC) - __bpf_call_base) }) 377 378 /* Raw code statement block */ 379 380 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 381 ((struct bpf_insn) { \ 382 .code = CODE, \ 383 .dst_reg = DST, \ 384 .src_reg = SRC, \ 385 .off = OFF, \ 386 .imm = IMM }) 387 388 /* Program exit */ 389 390 #define BPF_EXIT_INSN() \ 391 ((struct bpf_insn) { \ 392 .code = BPF_JMP | BPF_EXIT, \ 393 .dst_reg = 0, \ 394 .src_reg = 0, \ 395 .off = 0, \ 396 .imm = 0 }) 397 398 /* Speculation barrier */ 399 400 #define BPF_ST_NOSPEC() \ 401 ((struct bpf_insn) { \ 402 .code = BPF_ST | BPF_NOSPEC, \ 403 .dst_reg = 0, \ 404 .src_reg = 0, \ 405 .off = 0, \ 406 .imm = 0 }) 407 408 /* Internal classic blocks for direct assignment */ 409 410 #define __BPF_STMT(CODE, K) \ 411 ((struct sock_filter) BPF_STMT(CODE, K)) 412 413 #define __BPF_JUMP(CODE, K, JT, JF) \ 414 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 415 416 #define bytes_to_bpf_size(bytes) \ 417 ({ \ 418 int bpf_size = -EINVAL; \ 419 \ 420 if (bytes == sizeof(u8)) \ 421 bpf_size = BPF_B; \ 422 else if (bytes == sizeof(u16)) \ 423 bpf_size = BPF_H; \ 424 else if (bytes == sizeof(u32)) \ 425 bpf_size = BPF_W; \ 426 else if (bytes == sizeof(u64)) \ 427 bpf_size = BPF_DW; \ 428 \ 429 bpf_size; \ 430 }) 431 432 #define bpf_size_to_bytes(bpf_size) \ 433 ({ \ 434 int bytes = -EINVAL; \ 435 \ 436 if (bpf_size == BPF_B) \ 437 bytes = sizeof(u8); \ 438 else if (bpf_size == BPF_H) \ 439 bytes = sizeof(u16); \ 440 else if (bpf_size == BPF_W) \ 441 bytes = sizeof(u32); \ 442 else if (bpf_size == BPF_DW) \ 443 bytes = sizeof(u64); \ 444 \ 445 bytes; \ 446 }) 447 448 #define BPF_SIZEOF(type) \ 449 ({ \ 450 const int __size = bytes_to_bpf_size(sizeof(type)); \ 451 BUILD_BUG_ON(__size < 0); \ 452 __size; \ 453 }) 454 455 #define BPF_FIELD_SIZEOF(type, field) \ 456 ({ \ 457 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 458 BUILD_BUG_ON(__size < 0); \ 459 __size; \ 460 }) 461 462 #define BPF_LDST_BYTES(insn) \ 463 ({ \ 464 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 465 WARN_ON(__size < 0); \ 466 __size; \ 467 }) 468 469 #define __BPF_MAP_0(m, v, ...) v 470 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 471 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 472 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 473 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 474 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 475 476 #define __BPF_REG_0(...) __BPF_PAD(5) 477 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 478 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 479 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 480 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 481 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 482 483 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 484 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 485 486 #define __BPF_CAST(t, a) \ 487 (__force t) \ 488 (__force \ 489 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 490 (unsigned long)0, (t)0))) a 491 #define __BPF_V void 492 #define __BPF_N 493 494 #define __BPF_DECL_ARGS(t, a) t a 495 #define __BPF_DECL_REGS(t, a) u64 a 496 497 #define __BPF_PAD(n) \ 498 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 499 u64, __ur_3, u64, __ur_4, u64, __ur_5) 500 501 #define BPF_CALL_x(x, name, ...) \ 502 static __always_inline \ 503 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 504 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 505 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 506 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 507 { \ 508 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 509 } \ 510 static __always_inline \ 511 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 512 513 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) 514 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) 515 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) 516 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) 517 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) 518 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) 519 520 #define bpf_ctx_range(TYPE, MEMBER) \ 521 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 522 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 523 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 524 #if BITS_PER_LONG == 64 525 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 526 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 527 #else 528 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 529 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 530 #endif /* BITS_PER_LONG == 64 */ 531 532 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 533 ({ \ 534 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 535 *(PTR_SIZE) = (SIZE); \ 536 offsetof(TYPE, MEMBER); \ 537 }) 538 539 /* A struct sock_filter is architecture independent. */ 540 struct compat_sock_fprog { 541 u16 len; 542 compat_uptr_t filter; /* struct sock_filter * */ 543 }; 544 545 struct sock_fprog_kern { 546 u16 len; 547 struct sock_filter *filter; 548 }; 549 550 /* Some arches need doubleword alignment for their instructions and/or data */ 551 #define BPF_IMAGE_ALIGNMENT 8 552 553 struct bpf_binary_header { 554 u32 pages; 555 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 556 }; 557 558 struct bpf_prog_stats { 559 u64 cnt; 560 u64 nsecs; 561 u64 misses; 562 struct u64_stats_sync syncp; 563 } __aligned(2 * sizeof(u64)); 564 565 struct bpf_prog { 566 u16 pages; /* Number of allocated pages */ 567 u16 jited:1, /* Is our filter JIT'ed? */ 568 jit_requested:1,/* archs need to JIT the prog */ 569 gpl_compatible:1, /* Is filter GPL compatible? */ 570 cb_access:1, /* Is control block accessed? */ 571 dst_needed:1, /* Do we need dst entry? */ 572 blinded:1, /* Was blinded */ 573 is_func:1, /* program is a bpf function */ 574 kprobe_override:1, /* Do we override a kprobe? */ 575 has_callchain_buf:1, /* callchain buffer allocated? */ 576 enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */ 577 call_get_stack:1, /* Do we call bpf_get_stack() or bpf_get_stackid() */ 578 call_get_func_ip:1; /* Do we call get_func_ip() */ 579 enum bpf_prog_type type; /* Type of BPF program */ 580 enum bpf_attach_type expected_attach_type; /* For some prog types */ 581 u32 len; /* Number of filter blocks */ 582 u32 jited_len; /* Size of jited insns in bytes */ 583 u8 tag[BPF_TAG_SIZE]; 584 struct bpf_prog_stats __percpu *stats; 585 int __percpu *active; 586 unsigned int (*bpf_func)(const void *ctx, 587 const struct bpf_insn *insn); 588 struct bpf_prog_aux *aux; /* Auxiliary fields */ 589 struct sock_fprog_kern *orig_prog; /* Original BPF program */ 590 /* Instructions for interpreter */ 591 struct sock_filter insns[0]; 592 struct bpf_insn insnsi[]; 593 }; 594 595 struct sk_filter { 596 refcount_t refcnt; 597 struct rcu_head rcu; 598 struct bpf_prog *prog; 599 }; 600 601 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 602 603 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 604 const struct bpf_insn *insnsi, 605 unsigned int (*bpf_func)(const void *, 606 const struct bpf_insn *)); 607 608 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 609 const void *ctx, 610 bpf_dispatcher_fn dfunc) 611 { 612 u32 ret; 613 614 cant_migrate(); 615 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 616 struct bpf_prog_stats *stats; 617 u64 start = sched_clock(); 618 619 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 620 stats = this_cpu_ptr(prog->stats); 621 u64_stats_update_begin(&stats->syncp); 622 stats->cnt++; 623 stats->nsecs += sched_clock() - start; 624 u64_stats_update_end(&stats->syncp); 625 } else { 626 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 627 } 628 return ret; 629 } 630 631 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 632 { 633 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 634 } 635 636 /* 637 * Use in preemptible and therefore migratable context to make sure that 638 * the execution of the BPF program runs on one CPU. 639 * 640 * This uses migrate_disable/enable() explicitly to document that the 641 * invocation of a BPF program does not require reentrancy protection 642 * against a BPF program which is invoked from a preempting task. 643 * 644 * For non RT enabled kernels migrate_disable/enable() maps to 645 * preempt_disable/enable(), i.e. it disables also preemption. 646 */ 647 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 648 const void *ctx) 649 { 650 u32 ret; 651 652 migrate_disable(); 653 ret = bpf_prog_run(prog, ctx); 654 migrate_enable(); 655 return ret; 656 } 657 658 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 659 660 struct bpf_skb_data_end { 661 struct qdisc_skb_cb qdisc_cb; 662 void *data_meta; 663 void *data_end; 664 }; 665 666 struct bpf_nh_params { 667 u32 nh_family; 668 union { 669 u32 ipv4_nh; 670 struct in6_addr ipv6_nh; 671 }; 672 }; 673 674 struct bpf_redirect_info { 675 u32 flags; 676 u32 tgt_index; 677 void *tgt_value; 678 struct bpf_map *map; 679 u32 map_id; 680 enum bpf_map_type map_type; 681 u32 kern_flags; 682 struct bpf_nh_params nh; 683 }; 684 685 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 686 687 /* flags for bpf_redirect_info kern_flags */ 688 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 689 690 /* Compute the linear packet data range [data, data_end) which 691 * will be accessed by various program types (cls_bpf, act_bpf, 692 * lwt, ...). Subsystems allowing direct data access must (!) 693 * ensure that cb[] area can be written to when BPF program is 694 * invoked (otherwise cb[] save/restore is necessary). 695 */ 696 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 697 { 698 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 699 700 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 701 cb->data_meta = skb->data - skb_metadata_len(skb); 702 cb->data_end = skb->data + skb_headlen(skb); 703 } 704 705 /* Similar to bpf_compute_data_pointers(), except that save orginal 706 * data in cb->data and cb->meta_data for restore. 707 */ 708 static inline void bpf_compute_and_save_data_end( 709 struct sk_buff *skb, void **saved_data_end) 710 { 711 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 712 713 *saved_data_end = cb->data_end; 714 cb->data_end = skb->data + skb_headlen(skb); 715 } 716 717 /* Restore data saved by bpf_compute_data_pointers(). */ 718 static inline void bpf_restore_data_end( 719 struct sk_buff *skb, void *saved_data_end) 720 { 721 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 722 723 cb->data_end = saved_data_end; 724 } 725 726 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 727 { 728 /* eBPF programs may read/write skb->cb[] area to transfer meta 729 * data between tail calls. Since this also needs to work with 730 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 731 * 732 * In some socket filter cases, the cb unfortunately needs to be 733 * saved/restored so that protocol specific skb->cb[] data won't 734 * be lost. In any case, due to unpriviledged eBPF programs 735 * attached to sockets, we need to clear the bpf_skb_cb() area 736 * to not leak previous contents to user space. 737 */ 738 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 739 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 740 sizeof_field(struct qdisc_skb_cb, data)); 741 742 return qdisc_skb_cb(skb)->data; 743 } 744 745 /* Must be invoked with migration disabled */ 746 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 747 const void *ctx) 748 { 749 const struct sk_buff *skb = ctx; 750 u8 *cb_data = bpf_skb_cb(skb); 751 u8 cb_saved[BPF_SKB_CB_LEN]; 752 u32 res; 753 754 if (unlikely(prog->cb_access)) { 755 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 756 memset(cb_data, 0, sizeof(cb_saved)); 757 } 758 759 res = bpf_prog_run(prog, skb); 760 761 if (unlikely(prog->cb_access)) 762 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 763 764 return res; 765 } 766 767 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 768 struct sk_buff *skb) 769 { 770 u32 res; 771 772 migrate_disable(); 773 res = __bpf_prog_run_save_cb(prog, skb); 774 migrate_enable(); 775 return res; 776 } 777 778 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 779 struct sk_buff *skb) 780 { 781 u8 *cb_data = bpf_skb_cb(skb); 782 u32 res; 783 784 if (unlikely(prog->cb_access)) 785 memset(cb_data, 0, BPF_SKB_CB_LEN); 786 787 res = bpf_prog_run_pin_on_cpu(prog, skb); 788 return res; 789 } 790 791 DECLARE_BPF_DISPATCHER(xdp) 792 793 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 794 795 u32 xdp_master_redirect(struct xdp_buff *xdp); 796 797 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 798 struct xdp_buff *xdp) 799 { 800 /* Driver XDP hooks are invoked within a single NAPI poll cycle and thus 801 * under local_bh_disable(), which provides the needed RCU protection 802 * for accessing map entries. 803 */ 804 u32 act = __bpf_prog_run(prog, xdp, BPF_DISPATCHER_FUNC(xdp)); 805 806 if (static_branch_unlikely(&bpf_master_redirect_enabled_key)) { 807 if (act == XDP_TX && netif_is_bond_slave(xdp->rxq->dev)) 808 act = xdp_master_redirect(xdp); 809 } 810 811 return act; 812 } 813 814 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 815 816 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 817 { 818 return prog->len * sizeof(struct bpf_insn); 819 } 820 821 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 822 { 823 return round_up(bpf_prog_insn_size(prog) + 824 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 825 } 826 827 static inline unsigned int bpf_prog_size(unsigned int proglen) 828 { 829 return max(sizeof(struct bpf_prog), 830 offsetof(struct bpf_prog, insns[proglen])); 831 } 832 833 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 834 { 835 /* When classic BPF programs have been loaded and the arch 836 * does not have a classic BPF JIT (anymore), they have been 837 * converted via bpf_migrate_filter() to eBPF and thus always 838 * have an unspec program type. 839 */ 840 return prog->type == BPF_PROG_TYPE_UNSPEC; 841 } 842 843 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 844 { 845 const u32 size_machine = sizeof(unsigned long); 846 847 if (size > size_machine && size % size_machine == 0) 848 size = size_machine; 849 850 return size; 851 } 852 853 static inline bool 854 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 855 { 856 return size <= size_default && (size & (size - 1)) == 0; 857 } 858 859 static inline u8 860 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 861 { 862 u8 access_off = off & (size_default - 1); 863 864 #ifdef __LITTLE_ENDIAN 865 return access_off; 866 #else 867 return size_default - (access_off + size); 868 #endif 869 } 870 871 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 872 (size == sizeof(__u64) && \ 873 off >= offsetof(type, field) && \ 874 off + sizeof(__u64) <= offsetofend(type, field) && \ 875 off % sizeof(__u64) == 0) 876 877 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 878 879 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 880 { 881 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 882 if (!fp->jited) { 883 set_vm_flush_reset_perms(fp); 884 set_memory_ro((unsigned long)fp, fp->pages); 885 } 886 #endif 887 } 888 889 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 890 { 891 set_vm_flush_reset_perms(hdr); 892 set_memory_ro((unsigned long)hdr, hdr->pages); 893 set_memory_x((unsigned long)hdr, hdr->pages); 894 } 895 896 static inline struct bpf_binary_header * 897 bpf_jit_binary_hdr(const struct bpf_prog *fp) 898 { 899 unsigned long real_start = (unsigned long)fp->bpf_func; 900 unsigned long addr = real_start & PAGE_MASK; 901 902 return (void *)addr; 903 } 904 905 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 906 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 907 { 908 return sk_filter_trim_cap(sk, skb, 1); 909 } 910 911 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 912 void bpf_prog_free(struct bpf_prog *fp); 913 914 bool bpf_opcode_in_insntable(u8 code); 915 916 void bpf_prog_free_linfo(struct bpf_prog *prog); 917 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 918 const u32 *insn_to_jit_off); 919 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 920 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 921 922 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 923 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 924 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 925 gfp_t gfp_extra_flags); 926 void __bpf_prog_free(struct bpf_prog *fp); 927 928 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 929 { 930 __bpf_prog_free(fp); 931 } 932 933 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 934 unsigned int flen); 935 936 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 937 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 938 bpf_aux_classic_check_t trans, bool save_orig); 939 void bpf_prog_destroy(struct bpf_prog *fp); 940 941 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 942 int sk_attach_bpf(u32 ufd, struct sock *sk); 943 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 944 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 945 void sk_reuseport_prog_free(struct bpf_prog *prog); 946 int sk_detach_filter(struct sock *sk); 947 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, 948 unsigned int len); 949 950 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 951 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 952 953 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 954 #define __bpf_call_base_args \ 955 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 956 (void *)__bpf_call_base) 957 958 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 959 void bpf_jit_compile(struct bpf_prog *prog); 960 bool bpf_jit_needs_zext(void); 961 bool bpf_jit_supports_kfunc_call(void); 962 bool bpf_helper_changes_pkt_data(void *func); 963 964 static inline bool bpf_dump_raw_ok(const struct cred *cred) 965 { 966 /* Reconstruction of call-sites is dependent on kallsyms, 967 * thus make dump the same restriction. 968 */ 969 return kallsyms_show_value(cred); 970 } 971 972 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 973 const struct bpf_insn *patch, u32 len); 974 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 975 976 void bpf_clear_redirect_map(struct bpf_map *map); 977 978 static inline bool xdp_return_frame_no_direct(void) 979 { 980 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 981 982 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 983 } 984 985 static inline void xdp_set_return_frame_no_direct(void) 986 { 987 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 988 989 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 990 } 991 992 static inline void xdp_clear_return_frame_no_direct(void) 993 { 994 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 995 996 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 997 } 998 999 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 1000 unsigned int pktlen) 1001 { 1002 unsigned int len; 1003 1004 if (unlikely(!(fwd->flags & IFF_UP))) 1005 return -ENETDOWN; 1006 1007 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 1008 if (pktlen > len) 1009 return -EMSGSIZE; 1010 1011 return 0; 1012 } 1013 1014 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 1015 * same cpu context. Further for best results no more than a single map 1016 * for the do_redirect/do_flush pair should be used. This limitation is 1017 * because we only track one map and force a flush when the map changes. 1018 * This does not appear to be a real limitation for existing software. 1019 */ 1020 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 1021 struct xdp_buff *xdp, struct bpf_prog *prog); 1022 int xdp_do_redirect(struct net_device *dev, 1023 struct xdp_buff *xdp, 1024 struct bpf_prog *prog); 1025 void xdp_do_flush(void); 1026 1027 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as 1028 * it is no longer only flushing maps. Keep this define for compatibility 1029 * until all drivers are updated - do not use xdp_do_flush_map() in new code! 1030 */ 1031 #define xdp_do_flush_map xdp_do_flush 1032 1033 void bpf_warn_invalid_xdp_action(u32 act); 1034 1035 #ifdef CONFIG_INET 1036 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1037 struct bpf_prog *prog, struct sk_buff *skb, 1038 struct sock *migrating_sk, 1039 u32 hash); 1040 #else 1041 static inline struct sock * 1042 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1043 struct bpf_prog *prog, struct sk_buff *skb, 1044 struct sock *migrating_sk, 1045 u32 hash) 1046 { 1047 return NULL; 1048 } 1049 #endif 1050 1051 #ifdef CONFIG_BPF_JIT 1052 extern int bpf_jit_enable; 1053 extern int bpf_jit_harden; 1054 extern int bpf_jit_kallsyms; 1055 extern long bpf_jit_limit; 1056 1057 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1058 1059 struct bpf_binary_header * 1060 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1061 unsigned int alignment, 1062 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1063 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1064 u64 bpf_jit_alloc_exec_limit(void); 1065 void *bpf_jit_alloc_exec(unsigned long size); 1066 void bpf_jit_free_exec(void *addr); 1067 void bpf_jit_free(struct bpf_prog *fp); 1068 1069 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1070 struct bpf_jit_poke_descriptor *poke); 1071 1072 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1073 const struct bpf_insn *insn, bool extra_pass, 1074 u64 *func_addr, bool *func_addr_fixed); 1075 1076 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1077 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1078 1079 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1080 u32 pass, void *image) 1081 { 1082 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1083 proglen, pass, image, current->comm, task_pid_nr(current)); 1084 1085 if (image) 1086 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1087 16, 1, image, proglen, false); 1088 } 1089 1090 static inline bool bpf_jit_is_ebpf(void) 1091 { 1092 # ifdef CONFIG_HAVE_EBPF_JIT 1093 return true; 1094 # else 1095 return false; 1096 # endif 1097 } 1098 1099 static inline bool ebpf_jit_enabled(void) 1100 { 1101 return bpf_jit_enable && bpf_jit_is_ebpf(); 1102 } 1103 1104 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1105 { 1106 return fp->jited && bpf_jit_is_ebpf(); 1107 } 1108 1109 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1110 { 1111 /* These are the prerequisites, should someone ever have the 1112 * idea to call blinding outside of them, we make sure to 1113 * bail out. 1114 */ 1115 if (!bpf_jit_is_ebpf()) 1116 return false; 1117 if (!prog->jit_requested) 1118 return false; 1119 if (!bpf_jit_harden) 1120 return false; 1121 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) 1122 return false; 1123 1124 return true; 1125 } 1126 1127 static inline bool bpf_jit_kallsyms_enabled(void) 1128 { 1129 /* There are a couple of corner cases where kallsyms should 1130 * not be enabled f.e. on hardening. 1131 */ 1132 if (bpf_jit_harden) 1133 return false; 1134 if (!bpf_jit_kallsyms) 1135 return false; 1136 if (bpf_jit_kallsyms == 1) 1137 return true; 1138 1139 return false; 1140 } 1141 1142 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1143 unsigned long *off, char *sym); 1144 bool is_bpf_text_address(unsigned long addr); 1145 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1146 char *sym); 1147 1148 static inline const char * 1149 bpf_address_lookup(unsigned long addr, unsigned long *size, 1150 unsigned long *off, char **modname, char *sym) 1151 { 1152 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1153 1154 if (ret && modname) 1155 *modname = NULL; 1156 return ret; 1157 } 1158 1159 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1160 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1161 1162 #else /* CONFIG_BPF_JIT */ 1163 1164 static inline bool ebpf_jit_enabled(void) 1165 { 1166 return false; 1167 } 1168 1169 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1170 { 1171 return false; 1172 } 1173 1174 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1175 { 1176 return false; 1177 } 1178 1179 static inline int 1180 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1181 struct bpf_jit_poke_descriptor *poke) 1182 { 1183 return -ENOTSUPP; 1184 } 1185 1186 static inline void bpf_jit_free(struct bpf_prog *fp) 1187 { 1188 bpf_prog_unlock_free(fp); 1189 } 1190 1191 static inline bool bpf_jit_kallsyms_enabled(void) 1192 { 1193 return false; 1194 } 1195 1196 static inline const char * 1197 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1198 unsigned long *off, char *sym) 1199 { 1200 return NULL; 1201 } 1202 1203 static inline bool is_bpf_text_address(unsigned long addr) 1204 { 1205 return false; 1206 } 1207 1208 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1209 char *type, char *sym) 1210 { 1211 return -ERANGE; 1212 } 1213 1214 static inline const char * 1215 bpf_address_lookup(unsigned long addr, unsigned long *size, 1216 unsigned long *off, char **modname, char *sym) 1217 { 1218 return NULL; 1219 } 1220 1221 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1222 { 1223 } 1224 1225 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1226 { 1227 } 1228 1229 #endif /* CONFIG_BPF_JIT */ 1230 1231 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1232 1233 #define BPF_ANC BIT(15) 1234 1235 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1236 { 1237 switch (first->code) { 1238 case BPF_RET | BPF_K: 1239 case BPF_LD | BPF_W | BPF_LEN: 1240 return false; 1241 1242 case BPF_LD | BPF_W | BPF_ABS: 1243 case BPF_LD | BPF_H | BPF_ABS: 1244 case BPF_LD | BPF_B | BPF_ABS: 1245 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1246 return true; 1247 return false; 1248 1249 default: 1250 return true; 1251 } 1252 } 1253 1254 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1255 { 1256 BUG_ON(ftest->code & BPF_ANC); 1257 1258 switch (ftest->code) { 1259 case BPF_LD | BPF_W | BPF_ABS: 1260 case BPF_LD | BPF_H | BPF_ABS: 1261 case BPF_LD | BPF_B | BPF_ABS: 1262 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1263 return BPF_ANC | SKF_AD_##CODE 1264 switch (ftest->k) { 1265 BPF_ANCILLARY(PROTOCOL); 1266 BPF_ANCILLARY(PKTTYPE); 1267 BPF_ANCILLARY(IFINDEX); 1268 BPF_ANCILLARY(NLATTR); 1269 BPF_ANCILLARY(NLATTR_NEST); 1270 BPF_ANCILLARY(MARK); 1271 BPF_ANCILLARY(QUEUE); 1272 BPF_ANCILLARY(HATYPE); 1273 BPF_ANCILLARY(RXHASH); 1274 BPF_ANCILLARY(CPU); 1275 BPF_ANCILLARY(ALU_XOR_X); 1276 BPF_ANCILLARY(VLAN_TAG); 1277 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1278 BPF_ANCILLARY(PAY_OFFSET); 1279 BPF_ANCILLARY(RANDOM); 1280 BPF_ANCILLARY(VLAN_TPID); 1281 } 1282 fallthrough; 1283 default: 1284 return ftest->code; 1285 } 1286 } 1287 1288 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1289 int k, unsigned int size); 1290 1291 static inline int bpf_tell_extensions(void) 1292 { 1293 return SKF_AD_MAX; 1294 } 1295 1296 struct bpf_sock_addr_kern { 1297 struct sock *sk; 1298 struct sockaddr *uaddr; 1299 /* Temporary "register" to make indirect stores to nested structures 1300 * defined above. We need three registers to make such a store, but 1301 * only two (src and dst) are available at convert_ctx_access time 1302 */ 1303 u64 tmp_reg; 1304 void *t_ctx; /* Attach type specific context. */ 1305 }; 1306 1307 struct bpf_sock_ops_kern { 1308 struct sock *sk; 1309 union { 1310 u32 args[4]; 1311 u32 reply; 1312 u32 replylong[4]; 1313 }; 1314 struct sk_buff *syn_skb; 1315 struct sk_buff *skb; 1316 void *skb_data_end; 1317 u8 op; 1318 u8 is_fullsock; 1319 u8 remaining_opt_len; 1320 u64 temp; /* temp and everything after is not 1321 * initialized to 0 before calling 1322 * the BPF program. New fields that 1323 * should be initialized to 0 should 1324 * be inserted before temp. 1325 * temp is scratch storage used by 1326 * sock_ops_convert_ctx_access 1327 * as temporary storage of a register. 1328 */ 1329 }; 1330 1331 struct bpf_sysctl_kern { 1332 struct ctl_table_header *head; 1333 struct ctl_table *table; 1334 void *cur_val; 1335 size_t cur_len; 1336 void *new_val; 1337 size_t new_len; 1338 int new_updated; 1339 int write; 1340 loff_t *ppos; 1341 /* Temporary "register" for indirect stores to ppos. */ 1342 u64 tmp_reg; 1343 }; 1344 1345 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1346 struct bpf_sockopt_buf { 1347 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1348 }; 1349 1350 struct bpf_sockopt_kern { 1351 struct sock *sk; 1352 u8 *optval; 1353 u8 *optval_end; 1354 s32 level; 1355 s32 optname; 1356 s32 optlen; 1357 s32 retval; 1358 }; 1359 1360 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1361 1362 struct bpf_sk_lookup_kern { 1363 u16 family; 1364 u16 protocol; 1365 __be16 sport; 1366 u16 dport; 1367 struct { 1368 __be32 saddr; 1369 __be32 daddr; 1370 } v4; 1371 struct { 1372 const struct in6_addr *saddr; 1373 const struct in6_addr *daddr; 1374 } v6; 1375 struct sock *selected_sk; 1376 bool no_reuseport; 1377 }; 1378 1379 extern struct static_key_false bpf_sk_lookup_enabled; 1380 1381 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1382 * 1383 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1384 * SK_DROP. Their meaning is as follows: 1385 * 1386 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1387 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1388 * SK_DROP : terminate lookup with -ECONNREFUSED 1389 * 1390 * This macro aggregates return values and selected sockets from 1391 * multiple BPF programs according to following rules in order: 1392 * 1393 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1394 * macro result is SK_PASS and last ctx.selected_sk is used. 1395 * 2. If any program returned SK_DROP return value, 1396 * macro result is SK_DROP. 1397 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1398 * 1399 * Caller must ensure that the prog array is non-NULL, and that the 1400 * array as well as the programs it contains remain valid. 1401 */ 1402 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1403 ({ \ 1404 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1405 struct bpf_prog_array_item *_item; \ 1406 struct sock *_selected_sk = NULL; \ 1407 bool _no_reuseport = false; \ 1408 struct bpf_prog *_prog; \ 1409 bool _all_pass = true; \ 1410 u32 _ret; \ 1411 \ 1412 migrate_disable(); \ 1413 _item = &(array)->items[0]; \ 1414 while ((_prog = READ_ONCE(_item->prog))) { \ 1415 /* restore most recent selection */ \ 1416 _ctx->selected_sk = _selected_sk; \ 1417 _ctx->no_reuseport = _no_reuseport; \ 1418 \ 1419 _ret = func(_prog, _ctx); \ 1420 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1421 /* remember last non-NULL socket */ \ 1422 _selected_sk = _ctx->selected_sk; \ 1423 _no_reuseport = _ctx->no_reuseport; \ 1424 } else if (_ret == SK_DROP && _all_pass) { \ 1425 _all_pass = false; \ 1426 } \ 1427 _item++; \ 1428 } \ 1429 _ctx->selected_sk = _selected_sk; \ 1430 _ctx->no_reuseport = _no_reuseport; \ 1431 migrate_enable(); \ 1432 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1433 }) 1434 1435 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1436 const __be32 saddr, const __be16 sport, 1437 const __be32 daddr, const u16 dport, 1438 struct sock **psk) 1439 { 1440 struct bpf_prog_array *run_array; 1441 struct sock *selected_sk = NULL; 1442 bool no_reuseport = false; 1443 1444 rcu_read_lock(); 1445 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1446 if (run_array) { 1447 struct bpf_sk_lookup_kern ctx = { 1448 .family = AF_INET, 1449 .protocol = protocol, 1450 .v4.saddr = saddr, 1451 .v4.daddr = daddr, 1452 .sport = sport, 1453 .dport = dport, 1454 }; 1455 u32 act; 1456 1457 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1458 if (act == SK_PASS) { 1459 selected_sk = ctx.selected_sk; 1460 no_reuseport = ctx.no_reuseport; 1461 } else { 1462 selected_sk = ERR_PTR(-ECONNREFUSED); 1463 } 1464 } 1465 rcu_read_unlock(); 1466 *psk = selected_sk; 1467 return no_reuseport; 1468 } 1469 1470 #if IS_ENABLED(CONFIG_IPV6) 1471 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1472 const struct in6_addr *saddr, 1473 const __be16 sport, 1474 const struct in6_addr *daddr, 1475 const u16 dport, 1476 struct sock **psk) 1477 { 1478 struct bpf_prog_array *run_array; 1479 struct sock *selected_sk = NULL; 1480 bool no_reuseport = false; 1481 1482 rcu_read_lock(); 1483 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1484 if (run_array) { 1485 struct bpf_sk_lookup_kern ctx = { 1486 .family = AF_INET6, 1487 .protocol = protocol, 1488 .v6.saddr = saddr, 1489 .v6.daddr = daddr, 1490 .sport = sport, 1491 .dport = dport, 1492 }; 1493 u32 act; 1494 1495 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1496 if (act == SK_PASS) { 1497 selected_sk = ctx.selected_sk; 1498 no_reuseport = ctx.no_reuseport; 1499 } else { 1500 selected_sk = ERR_PTR(-ECONNREFUSED); 1501 } 1502 } 1503 rcu_read_unlock(); 1504 *psk = selected_sk; 1505 return no_reuseport; 1506 } 1507 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1508 1509 static __always_inline int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex, 1510 u64 flags, const u64 flag_mask, 1511 void *lookup_elem(struct bpf_map *map, u32 key)) 1512 { 1513 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1514 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1515 1516 /* Lower bits of the flags are used as return code on lookup failure */ 1517 if (unlikely(flags & ~(action_mask | flag_mask))) 1518 return XDP_ABORTED; 1519 1520 ri->tgt_value = lookup_elem(map, ifindex); 1521 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1522 /* If the lookup fails we want to clear out the state in the 1523 * redirect_info struct completely, so that if an eBPF program 1524 * performs multiple lookups, the last one always takes 1525 * precedence. 1526 */ 1527 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1528 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1529 return flags & action_mask; 1530 } 1531 1532 ri->tgt_index = ifindex; 1533 ri->map_id = map->id; 1534 ri->map_type = map->map_type; 1535 1536 if (flags & BPF_F_BROADCAST) { 1537 WRITE_ONCE(ri->map, map); 1538 ri->flags = flags; 1539 } else { 1540 WRITE_ONCE(ri->map, NULL); 1541 ri->flags = 0; 1542 } 1543 1544 return XDP_REDIRECT; 1545 } 1546 1547 #endif /* __LINUX_FILTER_H__ */ 1548