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