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 enum bpf_prog_type type; /* Type of BPF program */ 579 enum bpf_attach_type expected_attach_type; /* For some prog types */ 580 u32 len; /* Number of filter blocks */ 581 u32 jited_len; /* Size of jited insns in bytes */ 582 u8 tag[BPF_TAG_SIZE]; 583 struct bpf_prog_stats __percpu *stats; 584 int __percpu *active; 585 unsigned int (*bpf_func)(const void *ctx, 586 const struct bpf_insn *insn); 587 struct bpf_prog_aux *aux; /* Auxiliary fields */ 588 struct sock_fprog_kern *orig_prog; /* Original BPF program */ 589 /* Instructions for interpreter */ 590 struct sock_filter insns[0]; 591 struct bpf_insn insnsi[]; 592 }; 593 594 struct sk_filter { 595 refcount_t refcnt; 596 struct rcu_head rcu; 597 struct bpf_prog *prog; 598 }; 599 600 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 601 602 #define __BPF_PROG_RUN(prog, ctx, dfunc) ({ \ 603 u32 __ret; \ 604 cant_migrate(); \ 605 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \ 606 struct bpf_prog_stats *__stats; \ 607 u64 __start = sched_clock(); \ 608 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \ 609 __stats = this_cpu_ptr(prog->stats); \ 610 u64_stats_update_begin(&__stats->syncp); \ 611 __stats->cnt++; \ 612 __stats->nsecs += sched_clock() - __start; \ 613 u64_stats_update_end(&__stats->syncp); \ 614 } else { \ 615 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \ 616 } \ 617 __ret; }) 618 619 #define BPF_PROG_RUN(prog, ctx) \ 620 __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func) 621 622 /* 623 * Use in preemptible and therefore migratable context to make sure that 624 * the execution of the BPF program runs on one CPU. 625 * 626 * This uses migrate_disable/enable() explicitly to document that the 627 * invocation of a BPF program does not require reentrancy protection 628 * against a BPF program which is invoked from a preempting task. 629 * 630 * For non RT enabled kernels migrate_disable/enable() maps to 631 * preempt_disable/enable(), i.e. it disables also preemption. 632 */ 633 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 634 const void *ctx) 635 { 636 u32 ret; 637 638 migrate_disable(); 639 ret = __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func); 640 migrate_enable(); 641 return ret; 642 } 643 644 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 645 646 struct bpf_skb_data_end { 647 struct qdisc_skb_cb qdisc_cb; 648 void *data_meta; 649 void *data_end; 650 }; 651 652 struct bpf_nh_params { 653 u32 nh_family; 654 union { 655 u32 ipv4_nh; 656 struct in6_addr ipv6_nh; 657 }; 658 }; 659 660 struct bpf_redirect_info { 661 u32 flags; 662 u32 tgt_index; 663 void *tgt_value; 664 struct bpf_map *map; 665 u32 map_id; 666 enum bpf_map_type map_type; 667 u32 kern_flags; 668 struct bpf_nh_params nh; 669 }; 670 671 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 672 673 /* flags for bpf_redirect_info kern_flags */ 674 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 675 676 /* Compute the linear packet data range [data, data_end) which 677 * will be accessed by various program types (cls_bpf, act_bpf, 678 * lwt, ...). Subsystems allowing direct data access must (!) 679 * ensure that cb[] area can be written to when BPF program is 680 * invoked (otherwise cb[] save/restore is necessary). 681 */ 682 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 683 { 684 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 685 686 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 687 cb->data_meta = skb->data - skb_metadata_len(skb); 688 cb->data_end = skb->data + skb_headlen(skb); 689 } 690 691 /* Similar to bpf_compute_data_pointers(), except that save orginal 692 * data in cb->data and cb->meta_data for restore. 693 */ 694 static inline void bpf_compute_and_save_data_end( 695 struct sk_buff *skb, void **saved_data_end) 696 { 697 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 698 699 *saved_data_end = cb->data_end; 700 cb->data_end = skb->data + skb_headlen(skb); 701 } 702 703 /* Restore data saved by bpf_compute_data_pointers(). */ 704 static inline void bpf_restore_data_end( 705 struct sk_buff *skb, void *saved_data_end) 706 { 707 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 708 709 cb->data_end = saved_data_end; 710 } 711 712 static inline u8 *bpf_skb_cb(struct sk_buff *skb) 713 { 714 /* eBPF programs may read/write skb->cb[] area to transfer meta 715 * data between tail calls. Since this also needs to work with 716 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 717 * 718 * In some socket filter cases, the cb unfortunately needs to be 719 * saved/restored so that protocol specific skb->cb[] data won't 720 * be lost. In any case, due to unpriviledged eBPF programs 721 * attached to sockets, we need to clear the bpf_skb_cb() area 722 * to not leak previous contents to user space. 723 */ 724 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 725 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 726 sizeof_field(struct qdisc_skb_cb, data)); 727 728 return qdisc_skb_cb(skb)->data; 729 } 730 731 /* Must be invoked with migration disabled */ 732 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 733 struct sk_buff *skb) 734 { 735 u8 *cb_data = bpf_skb_cb(skb); 736 u8 cb_saved[BPF_SKB_CB_LEN]; 737 u32 res; 738 739 if (unlikely(prog->cb_access)) { 740 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 741 memset(cb_data, 0, sizeof(cb_saved)); 742 } 743 744 res = BPF_PROG_RUN(prog, skb); 745 746 if (unlikely(prog->cb_access)) 747 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 748 749 return res; 750 } 751 752 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 753 struct sk_buff *skb) 754 { 755 u32 res; 756 757 migrate_disable(); 758 res = __bpf_prog_run_save_cb(prog, skb); 759 migrate_enable(); 760 return res; 761 } 762 763 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 764 struct sk_buff *skb) 765 { 766 u8 *cb_data = bpf_skb_cb(skb); 767 u32 res; 768 769 if (unlikely(prog->cb_access)) 770 memset(cb_data, 0, BPF_SKB_CB_LEN); 771 772 res = bpf_prog_run_pin_on_cpu(prog, skb); 773 return res; 774 } 775 776 DECLARE_BPF_DISPATCHER(xdp) 777 778 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 779 struct xdp_buff *xdp) 780 { 781 /* Driver XDP hooks are invoked within a single NAPI poll cycle and thus 782 * under local_bh_disable(), which provides the needed RCU protection 783 * for accessing map entries. 784 */ 785 return __BPF_PROG_RUN(prog, xdp, BPF_DISPATCHER_FUNC(xdp)); 786 } 787 788 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 789 790 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 791 { 792 return prog->len * sizeof(struct bpf_insn); 793 } 794 795 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 796 { 797 return round_up(bpf_prog_insn_size(prog) + 798 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 799 } 800 801 static inline unsigned int bpf_prog_size(unsigned int proglen) 802 { 803 return max(sizeof(struct bpf_prog), 804 offsetof(struct bpf_prog, insns[proglen])); 805 } 806 807 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 808 { 809 /* When classic BPF programs have been loaded and the arch 810 * does not have a classic BPF JIT (anymore), they have been 811 * converted via bpf_migrate_filter() to eBPF and thus always 812 * have an unspec program type. 813 */ 814 return prog->type == BPF_PROG_TYPE_UNSPEC; 815 } 816 817 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 818 { 819 const u32 size_machine = sizeof(unsigned long); 820 821 if (size > size_machine && size % size_machine == 0) 822 size = size_machine; 823 824 return size; 825 } 826 827 static inline bool 828 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 829 { 830 return size <= size_default && (size & (size - 1)) == 0; 831 } 832 833 static inline u8 834 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 835 { 836 u8 access_off = off & (size_default - 1); 837 838 #ifdef __LITTLE_ENDIAN 839 return access_off; 840 #else 841 return size_default - (access_off + size); 842 #endif 843 } 844 845 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 846 (size == sizeof(__u64) && \ 847 off >= offsetof(type, field) && \ 848 off + sizeof(__u64) <= offsetofend(type, field) && \ 849 off % sizeof(__u64) == 0) 850 851 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 852 853 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 854 { 855 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 856 if (!fp->jited) { 857 set_vm_flush_reset_perms(fp); 858 set_memory_ro((unsigned long)fp, fp->pages); 859 } 860 #endif 861 } 862 863 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 864 { 865 set_vm_flush_reset_perms(hdr); 866 set_memory_ro((unsigned long)hdr, hdr->pages); 867 set_memory_x((unsigned long)hdr, hdr->pages); 868 } 869 870 static inline struct bpf_binary_header * 871 bpf_jit_binary_hdr(const struct bpf_prog *fp) 872 { 873 unsigned long real_start = (unsigned long)fp->bpf_func; 874 unsigned long addr = real_start & PAGE_MASK; 875 876 return (void *)addr; 877 } 878 879 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 880 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 881 { 882 return sk_filter_trim_cap(sk, skb, 1); 883 } 884 885 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 886 void bpf_prog_free(struct bpf_prog *fp); 887 888 bool bpf_opcode_in_insntable(u8 code); 889 890 void bpf_prog_free_linfo(struct bpf_prog *prog); 891 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 892 const u32 *insn_to_jit_off); 893 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 894 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 895 896 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 897 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 898 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 899 gfp_t gfp_extra_flags); 900 void __bpf_prog_free(struct bpf_prog *fp); 901 902 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 903 { 904 __bpf_prog_free(fp); 905 } 906 907 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 908 unsigned int flen); 909 910 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 911 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 912 bpf_aux_classic_check_t trans, bool save_orig); 913 void bpf_prog_destroy(struct bpf_prog *fp); 914 915 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 916 int sk_attach_bpf(u32 ufd, struct sock *sk); 917 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 918 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 919 void sk_reuseport_prog_free(struct bpf_prog *prog); 920 int sk_detach_filter(struct sock *sk); 921 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, 922 unsigned int len); 923 924 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 925 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 926 927 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 928 #define __bpf_call_base_args \ 929 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 930 (void *)__bpf_call_base) 931 932 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 933 void bpf_jit_compile(struct bpf_prog *prog); 934 bool bpf_jit_needs_zext(void); 935 bool bpf_jit_supports_kfunc_call(void); 936 bool bpf_helper_changes_pkt_data(void *func); 937 938 static inline bool bpf_dump_raw_ok(const struct cred *cred) 939 { 940 /* Reconstruction of call-sites is dependent on kallsyms, 941 * thus make dump the same restriction. 942 */ 943 return kallsyms_show_value(cred); 944 } 945 946 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 947 const struct bpf_insn *patch, u32 len); 948 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 949 950 void bpf_clear_redirect_map(struct bpf_map *map); 951 952 static inline bool xdp_return_frame_no_direct(void) 953 { 954 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 955 956 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 957 } 958 959 static inline void xdp_set_return_frame_no_direct(void) 960 { 961 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 962 963 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 964 } 965 966 static inline void xdp_clear_return_frame_no_direct(void) 967 { 968 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 969 970 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 971 } 972 973 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 974 unsigned int pktlen) 975 { 976 unsigned int len; 977 978 if (unlikely(!(fwd->flags & IFF_UP))) 979 return -ENETDOWN; 980 981 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 982 if (pktlen > len) 983 return -EMSGSIZE; 984 985 return 0; 986 } 987 988 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 989 * same cpu context. Further for best results no more than a single map 990 * for the do_redirect/do_flush pair should be used. This limitation is 991 * because we only track one map and force a flush when the map changes. 992 * This does not appear to be a real limitation for existing software. 993 */ 994 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 995 struct xdp_buff *xdp, struct bpf_prog *prog); 996 int xdp_do_redirect(struct net_device *dev, 997 struct xdp_buff *xdp, 998 struct bpf_prog *prog); 999 void xdp_do_flush(void); 1000 1001 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as 1002 * it is no longer only flushing maps. Keep this define for compatibility 1003 * until all drivers are updated - do not use xdp_do_flush_map() in new code! 1004 */ 1005 #define xdp_do_flush_map xdp_do_flush 1006 1007 void bpf_warn_invalid_xdp_action(u32 act); 1008 1009 #ifdef CONFIG_INET 1010 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1011 struct bpf_prog *prog, struct sk_buff *skb, 1012 struct sock *migrating_sk, 1013 u32 hash); 1014 #else 1015 static inline struct sock * 1016 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1017 struct bpf_prog *prog, struct sk_buff *skb, 1018 struct sock *migrating_sk, 1019 u32 hash) 1020 { 1021 return NULL; 1022 } 1023 #endif 1024 1025 #ifdef CONFIG_BPF_JIT 1026 extern int bpf_jit_enable; 1027 extern int bpf_jit_harden; 1028 extern int bpf_jit_kallsyms; 1029 extern long bpf_jit_limit; 1030 1031 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1032 1033 struct bpf_binary_header * 1034 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1035 unsigned int alignment, 1036 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1037 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1038 u64 bpf_jit_alloc_exec_limit(void); 1039 void *bpf_jit_alloc_exec(unsigned long size); 1040 void bpf_jit_free_exec(void *addr); 1041 void bpf_jit_free(struct bpf_prog *fp); 1042 1043 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1044 struct bpf_jit_poke_descriptor *poke); 1045 1046 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1047 const struct bpf_insn *insn, bool extra_pass, 1048 u64 *func_addr, bool *func_addr_fixed); 1049 1050 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1051 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1052 1053 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1054 u32 pass, void *image) 1055 { 1056 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1057 proglen, pass, image, current->comm, task_pid_nr(current)); 1058 1059 if (image) 1060 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1061 16, 1, image, proglen, false); 1062 } 1063 1064 static inline bool bpf_jit_is_ebpf(void) 1065 { 1066 # ifdef CONFIG_HAVE_EBPF_JIT 1067 return true; 1068 # else 1069 return false; 1070 # endif 1071 } 1072 1073 static inline bool ebpf_jit_enabled(void) 1074 { 1075 return bpf_jit_enable && bpf_jit_is_ebpf(); 1076 } 1077 1078 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1079 { 1080 return fp->jited && bpf_jit_is_ebpf(); 1081 } 1082 1083 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1084 { 1085 /* These are the prerequisites, should someone ever have the 1086 * idea to call blinding outside of them, we make sure to 1087 * bail out. 1088 */ 1089 if (!bpf_jit_is_ebpf()) 1090 return false; 1091 if (!prog->jit_requested) 1092 return false; 1093 if (!bpf_jit_harden) 1094 return false; 1095 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) 1096 return false; 1097 1098 return true; 1099 } 1100 1101 static inline bool bpf_jit_kallsyms_enabled(void) 1102 { 1103 /* There are a couple of corner cases where kallsyms should 1104 * not be enabled f.e. on hardening. 1105 */ 1106 if (bpf_jit_harden) 1107 return false; 1108 if (!bpf_jit_kallsyms) 1109 return false; 1110 if (bpf_jit_kallsyms == 1) 1111 return true; 1112 1113 return false; 1114 } 1115 1116 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1117 unsigned long *off, char *sym); 1118 bool is_bpf_text_address(unsigned long addr); 1119 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1120 char *sym); 1121 1122 static inline const char * 1123 bpf_address_lookup(unsigned long addr, unsigned long *size, 1124 unsigned long *off, char **modname, char *sym) 1125 { 1126 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1127 1128 if (ret && modname) 1129 *modname = NULL; 1130 return ret; 1131 } 1132 1133 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1134 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1135 1136 #else /* CONFIG_BPF_JIT */ 1137 1138 static inline bool ebpf_jit_enabled(void) 1139 { 1140 return false; 1141 } 1142 1143 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1144 { 1145 return false; 1146 } 1147 1148 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1149 { 1150 return false; 1151 } 1152 1153 static inline int 1154 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1155 struct bpf_jit_poke_descriptor *poke) 1156 { 1157 return -ENOTSUPP; 1158 } 1159 1160 static inline void bpf_jit_free(struct bpf_prog *fp) 1161 { 1162 bpf_prog_unlock_free(fp); 1163 } 1164 1165 static inline bool bpf_jit_kallsyms_enabled(void) 1166 { 1167 return false; 1168 } 1169 1170 static inline const char * 1171 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1172 unsigned long *off, char *sym) 1173 { 1174 return NULL; 1175 } 1176 1177 static inline bool is_bpf_text_address(unsigned long addr) 1178 { 1179 return false; 1180 } 1181 1182 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1183 char *type, char *sym) 1184 { 1185 return -ERANGE; 1186 } 1187 1188 static inline const char * 1189 bpf_address_lookup(unsigned long addr, unsigned long *size, 1190 unsigned long *off, char **modname, char *sym) 1191 { 1192 return NULL; 1193 } 1194 1195 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1196 { 1197 } 1198 1199 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1200 { 1201 } 1202 1203 #endif /* CONFIG_BPF_JIT */ 1204 1205 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1206 1207 #define BPF_ANC BIT(15) 1208 1209 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1210 { 1211 switch (first->code) { 1212 case BPF_RET | BPF_K: 1213 case BPF_LD | BPF_W | BPF_LEN: 1214 return false; 1215 1216 case BPF_LD | BPF_W | BPF_ABS: 1217 case BPF_LD | BPF_H | BPF_ABS: 1218 case BPF_LD | BPF_B | BPF_ABS: 1219 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1220 return true; 1221 return false; 1222 1223 default: 1224 return true; 1225 } 1226 } 1227 1228 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1229 { 1230 BUG_ON(ftest->code & BPF_ANC); 1231 1232 switch (ftest->code) { 1233 case BPF_LD | BPF_W | BPF_ABS: 1234 case BPF_LD | BPF_H | BPF_ABS: 1235 case BPF_LD | BPF_B | BPF_ABS: 1236 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1237 return BPF_ANC | SKF_AD_##CODE 1238 switch (ftest->k) { 1239 BPF_ANCILLARY(PROTOCOL); 1240 BPF_ANCILLARY(PKTTYPE); 1241 BPF_ANCILLARY(IFINDEX); 1242 BPF_ANCILLARY(NLATTR); 1243 BPF_ANCILLARY(NLATTR_NEST); 1244 BPF_ANCILLARY(MARK); 1245 BPF_ANCILLARY(QUEUE); 1246 BPF_ANCILLARY(HATYPE); 1247 BPF_ANCILLARY(RXHASH); 1248 BPF_ANCILLARY(CPU); 1249 BPF_ANCILLARY(ALU_XOR_X); 1250 BPF_ANCILLARY(VLAN_TAG); 1251 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1252 BPF_ANCILLARY(PAY_OFFSET); 1253 BPF_ANCILLARY(RANDOM); 1254 BPF_ANCILLARY(VLAN_TPID); 1255 } 1256 fallthrough; 1257 default: 1258 return ftest->code; 1259 } 1260 } 1261 1262 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1263 int k, unsigned int size); 1264 1265 static inline int bpf_tell_extensions(void) 1266 { 1267 return SKF_AD_MAX; 1268 } 1269 1270 struct bpf_sock_addr_kern { 1271 struct sock *sk; 1272 struct sockaddr *uaddr; 1273 /* Temporary "register" to make indirect stores to nested structures 1274 * defined above. We need three registers to make such a store, but 1275 * only two (src and dst) are available at convert_ctx_access time 1276 */ 1277 u64 tmp_reg; 1278 void *t_ctx; /* Attach type specific context. */ 1279 }; 1280 1281 struct bpf_sock_ops_kern { 1282 struct sock *sk; 1283 union { 1284 u32 args[4]; 1285 u32 reply; 1286 u32 replylong[4]; 1287 }; 1288 struct sk_buff *syn_skb; 1289 struct sk_buff *skb; 1290 void *skb_data_end; 1291 u8 op; 1292 u8 is_fullsock; 1293 u8 remaining_opt_len; 1294 u64 temp; /* temp and everything after is not 1295 * initialized to 0 before calling 1296 * the BPF program. New fields that 1297 * should be initialized to 0 should 1298 * be inserted before temp. 1299 * temp is scratch storage used by 1300 * sock_ops_convert_ctx_access 1301 * as temporary storage of a register. 1302 */ 1303 }; 1304 1305 struct bpf_sysctl_kern { 1306 struct ctl_table_header *head; 1307 struct ctl_table *table; 1308 void *cur_val; 1309 size_t cur_len; 1310 void *new_val; 1311 size_t new_len; 1312 int new_updated; 1313 int write; 1314 loff_t *ppos; 1315 /* Temporary "register" for indirect stores to ppos. */ 1316 u64 tmp_reg; 1317 }; 1318 1319 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1320 struct bpf_sockopt_buf { 1321 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1322 }; 1323 1324 struct bpf_sockopt_kern { 1325 struct sock *sk; 1326 u8 *optval; 1327 u8 *optval_end; 1328 s32 level; 1329 s32 optname; 1330 s32 optlen; 1331 s32 retval; 1332 }; 1333 1334 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1335 1336 struct bpf_sk_lookup_kern { 1337 u16 family; 1338 u16 protocol; 1339 __be16 sport; 1340 u16 dport; 1341 struct { 1342 __be32 saddr; 1343 __be32 daddr; 1344 } v4; 1345 struct { 1346 const struct in6_addr *saddr; 1347 const struct in6_addr *daddr; 1348 } v6; 1349 struct sock *selected_sk; 1350 bool no_reuseport; 1351 }; 1352 1353 extern struct static_key_false bpf_sk_lookup_enabled; 1354 1355 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1356 * 1357 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1358 * SK_DROP. Their meaning is as follows: 1359 * 1360 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1361 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1362 * SK_DROP : terminate lookup with -ECONNREFUSED 1363 * 1364 * This macro aggregates return values and selected sockets from 1365 * multiple BPF programs according to following rules in order: 1366 * 1367 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1368 * macro result is SK_PASS and last ctx.selected_sk is used. 1369 * 2. If any program returned SK_DROP return value, 1370 * macro result is SK_DROP. 1371 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1372 * 1373 * Caller must ensure that the prog array is non-NULL, and that the 1374 * array as well as the programs it contains remain valid. 1375 */ 1376 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1377 ({ \ 1378 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1379 struct bpf_prog_array_item *_item; \ 1380 struct sock *_selected_sk = NULL; \ 1381 bool _no_reuseport = false; \ 1382 struct bpf_prog *_prog; \ 1383 bool _all_pass = true; \ 1384 u32 _ret; \ 1385 \ 1386 migrate_disable(); \ 1387 _item = &(array)->items[0]; \ 1388 while ((_prog = READ_ONCE(_item->prog))) { \ 1389 /* restore most recent selection */ \ 1390 _ctx->selected_sk = _selected_sk; \ 1391 _ctx->no_reuseport = _no_reuseport; \ 1392 \ 1393 _ret = func(_prog, _ctx); \ 1394 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1395 /* remember last non-NULL socket */ \ 1396 _selected_sk = _ctx->selected_sk; \ 1397 _no_reuseport = _ctx->no_reuseport; \ 1398 } else if (_ret == SK_DROP && _all_pass) { \ 1399 _all_pass = false; \ 1400 } \ 1401 _item++; \ 1402 } \ 1403 _ctx->selected_sk = _selected_sk; \ 1404 _ctx->no_reuseport = _no_reuseport; \ 1405 migrate_enable(); \ 1406 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1407 }) 1408 1409 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1410 const __be32 saddr, const __be16 sport, 1411 const __be32 daddr, const u16 dport, 1412 struct sock **psk) 1413 { 1414 struct bpf_prog_array *run_array; 1415 struct sock *selected_sk = NULL; 1416 bool no_reuseport = false; 1417 1418 rcu_read_lock(); 1419 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1420 if (run_array) { 1421 struct bpf_sk_lookup_kern ctx = { 1422 .family = AF_INET, 1423 .protocol = protocol, 1424 .v4.saddr = saddr, 1425 .v4.daddr = daddr, 1426 .sport = sport, 1427 .dport = dport, 1428 }; 1429 u32 act; 1430 1431 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN); 1432 if (act == SK_PASS) { 1433 selected_sk = ctx.selected_sk; 1434 no_reuseport = ctx.no_reuseport; 1435 } else { 1436 selected_sk = ERR_PTR(-ECONNREFUSED); 1437 } 1438 } 1439 rcu_read_unlock(); 1440 *psk = selected_sk; 1441 return no_reuseport; 1442 } 1443 1444 #if IS_ENABLED(CONFIG_IPV6) 1445 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1446 const struct in6_addr *saddr, 1447 const __be16 sport, 1448 const struct in6_addr *daddr, 1449 const u16 dport, 1450 struct sock **psk) 1451 { 1452 struct bpf_prog_array *run_array; 1453 struct sock *selected_sk = NULL; 1454 bool no_reuseport = false; 1455 1456 rcu_read_lock(); 1457 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1458 if (run_array) { 1459 struct bpf_sk_lookup_kern ctx = { 1460 .family = AF_INET6, 1461 .protocol = protocol, 1462 .v6.saddr = saddr, 1463 .v6.daddr = daddr, 1464 .sport = sport, 1465 .dport = dport, 1466 }; 1467 u32 act; 1468 1469 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN); 1470 if (act == SK_PASS) { 1471 selected_sk = ctx.selected_sk; 1472 no_reuseport = ctx.no_reuseport; 1473 } else { 1474 selected_sk = ERR_PTR(-ECONNREFUSED); 1475 } 1476 } 1477 rcu_read_unlock(); 1478 *psk = selected_sk; 1479 return no_reuseport; 1480 } 1481 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1482 1483 static __always_inline int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex, 1484 u64 flags, const u64 flag_mask, 1485 void *lookup_elem(struct bpf_map *map, u32 key)) 1486 { 1487 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1488 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1489 1490 /* Lower bits of the flags are used as return code on lookup failure */ 1491 if (unlikely(flags & ~(action_mask | flag_mask))) 1492 return XDP_ABORTED; 1493 1494 ri->tgt_value = lookup_elem(map, ifindex); 1495 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1496 /* If the lookup fails we want to clear out the state in the 1497 * redirect_info struct completely, so that if an eBPF program 1498 * performs multiple lookups, the last one always takes 1499 * precedence. 1500 */ 1501 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1502 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1503 return flags & action_mask; 1504 } 1505 1506 ri->tgt_index = ifindex; 1507 ri->map_id = map->id; 1508 ri->map_type = map->map_type; 1509 1510 if (flags & BPF_F_BROADCAST) { 1511 WRITE_ONCE(ri->map, map); 1512 ri->flags = flags; 1513 } else { 1514 WRITE_ONCE(ri->map, NULL); 1515 ri->flags = 0; 1516 } 1517 1518 return XDP_REDIRECT; 1519 } 1520 1521 #endif /* __LINUX_FILTER_H__ */ 1522