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 <linux/atomic.h> 9 #include <linux/bpf.h> 10 #include <linux/refcount.h> 11 #include <linux/compat.h> 12 #include <linux/skbuff.h> 13 #include <linux/linkage.h> 14 #include <linux/printk.h> 15 #include <linux/workqueue.h> 16 #include <linux/sched.h> 17 #include <linux/sched/clock.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 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 /* unused opcode to mark speculation barrier for mitigating 76 * Speculative Store Bypass 77 */ 78 #define BPF_NOSPEC 0xc0 79 80 /* As per nm, we expose JITed images as text (code) section for 81 * kallsyms. That way, tools like perf can find it to match 82 * addresses. 83 */ 84 #define BPF_SYM_ELF_TYPE 't' 85 86 /* BPF program can access up to 512 bytes of stack space. */ 87 #define MAX_BPF_STACK 512 88 89 /* Helper macros for filter block array initializers. */ 90 91 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 92 93 #define BPF_ALU64_REG(OP, DST, SRC) \ 94 ((struct bpf_insn) { \ 95 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 96 .dst_reg = DST, \ 97 .src_reg = SRC, \ 98 .off = 0, \ 99 .imm = 0 }) 100 101 #define BPF_ALU32_REG(OP, DST, SRC) \ 102 ((struct bpf_insn) { \ 103 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 104 .dst_reg = DST, \ 105 .src_reg = SRC, \ 106 .off = 0, \ 107 .imm = 0 }) 108 109 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 110 111 #define BPF_ALU64_IMM(OP, DST, IMM) \ 112 ((struct bpf_insn) { \ 113 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 114 .dst_reg = DST, \ 115 .src_reg = 0, \ 116 .off = 0, \ 117 .imm = IMM }) 118 119 #define BPF_ALU32_IMM(OP, DST, IMM) \ 120 ((struct bpf_insn) { \ 121 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 122 .dst_reg = DST, \ 123 .src_reg = 0, \ 124 .off = 0, \ 125 .imm = IMM }) 126 127 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 128 129 #define BPF_ENDIAN(TYPE, DST, LEN) \ 130 ((struct bpf_insn) { \ 131 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 132 .dst_reg = DST, \ 133 .src_reg = 0, \ 134 .off = 0, \ 135 .imm = LEN }) 136 137 /* Short form of mov, dst_reg = src_reg */ 138 139 #define BPF_MOV64_REG(DST, SRC) \ 140 ((struct bpf_insn) { \ 141 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 142 .dst_reg = DST, \ 143 .src_reg = SRC, \ 144 .off = 0, \ 145 .imm = 0 }) 146 147 #define BPF_MOV32_REG(DST, SRC) \ 148 ((struct bpf_insn) { \ 149 .code = BPF_ALU | BPF_MOV | BPF_X, \ 150 .dst_reg = DST, \ 151 .src_reg = SRC, \ 152 .off = 0, \ 153 .imm = 0 }) 154 155 /* Short form of mov, dst_reg = imm32 */ 156 157 #define BPF_MOV64_IMM(DST, IMM) \ 158 ((struct bpf_insn) { \ 159 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 160 .dst_reg = DST, \ 161 .src_reg = 0, \ 162 .off = 0, \ 163 .imm = IMM }) 164 165 #define BPF_MOV32_IMM(DST, IMM) \ 166 ((struct bpf_insn) { \ 167 .code = BPF_ALU | BPF_MOV | BPF_K, \ 168 .dst_reg = DST, \ 169 .src_reg = 0, \ 170 .off = 0, \ 171 .imm = IMM }) 172 173 /* Special form of mov32, used for doing explicit zero extension on dst. */ 174 #define BPF_ZEXT_REG(DST) \ 175 ((struct bpf_insn) { \ 176 .code = BPF_ALU | BPF_MOV | BPF_X, \ 177 .dst_reg = DST, \ 178 .src_reg = DST, \ 179 .off = 0, \ 180 .imm = 1 }) 181 182 static inline bool insn_is_zext(const struct bpf_insn *insn) 183 { 184 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 185 } 186 187 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 188 #define BPF_LD_IMM64(DST, IMM) \ 189 BPF_LD_IMM64_RAW(DST, 0, IMM) 190 191 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 192 ((struct bpf_insn) { \ 193 .code = BPF_LD | BPF_DW | BPF_IMM, \ 194 .dst_reg = DST, \ 195 .src_reg = SRC, \ 196 .off = 0, \ 197 .imm = (__u32) (IMM) }), \ 198 ((struct bpf_insn) { \ 199 .code = 0, /* zero is reserved opcode */ \ 200 .dst_reg = 0, \ 201 .src_reg = 0, \ 202 .off = 0, \ 203 .imm = ((__u64) (IMM)) >> 32 }) 204 205 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 206 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 207 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 208 209 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 210 211 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 212 ((struct bpf_insn) { \ 213 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 214 .dst_reg = DST, \ 215 .src_reg = SRC, \ 216 .off = 0, \ 217 .imm = IMM }) 218 219 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 220 ((struct bpf_insn) { \ 221 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 222 .dst_reg = DST, \ 223 .src_reg = SRC, \ 224 .off = 0, \ 225 .imm = IMM }) 226 227 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 228 229 #define BPF_LD_ABS(SIZE, IMM) \ 230 ((struct bpf_insn) { \ 231 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 232 .dst_reg = 0, \ 233 .src_reg = 0, \ 234 .off = 0, \ 235 .imm = IMM }) 236 237 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 238 239 #define BPF_LD_IND(SIZE, SRC, IMM) \ 240 ((struct bpf_insn) { \ 241 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 242 .dst_reg = 0, \ 243 .src_reg = SRC, \ 244 .off = 0, \ 245 .imm = IMM }) 246 247 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 248 249 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 250 ((struct bpf_insn) { \ 251 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 252 .dst_reg = DST, \ 253 .src_reg = SRC, \ 254 .off = OFF, \ 255 .imm = 0 }) 256 257 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 258 259 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 260 ((struct bpf_insn) { \ 261 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 262 .dst_reg = DST, \ 263 .src_reg = SRC, \ 264 .off = OFF, \ 265 .imm = 0 }) 266 267 268 /* 269 * Atomic operations: 270 * 271 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 272 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 273 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 274 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 275 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 276 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 277 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 278 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 279 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 280 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 281 */ 282 283 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 284 ((struct bpf_insn) { \ 285 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 286 .dst_reg = DST, \ 287 .src_reg = SRC, \ 288 .off = OFF, \ 289 .imm = OP }) 290 291 /* Legacy alias */ 292 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 293 294 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 295 296 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 297 ((struct bpf_insn) { \ 298 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 299 .dst_reg = DST, \ 300 .src_reg = 0, \ 301 .off = OFF, \ 302 .imm = IMM }) 303 304 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 305 306 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 307 ((struct bpf_insn) { \ 308 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 309 .dst_reg = DST, \ 310 .src_reg = SRC, \ 311 .off = OFF, \ 312 .imm = 0 }) 313 314 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 315 316 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 317 ((struct bpf_insn) { \ 318 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 319 .dst_reg = DST, \ 320 .src_reg = 0, \ 321 .off = OFF, \ 322 .imm = IMM }) 323 324 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 325 326 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 327 ((struct bpf_insn) { \ 328 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 329 .dst_reg = DST, \ 330 .src_reg = SRC, \ 331 .off = OFF, \ 332 .imm = 0 }) 333 334 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 335 336 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 337 ((struct bpf_insn) { \ 338 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 339 .dst_reg = DST, \ 340 .src_reg = 0, \ 341 .off = OFF, \ 342 .imm = IMM }) 343 344 /* Unconditional jumps, goto pc + off16 */ 345 346 #define BPF_JMP_A(OFF) \ 347 ((struct bpf_insn) { \ 348 .code = BPF_JMP | BPF_JA, \ 349 .dst_reg = 0, \ 350 .src_reg = 0, \ 351 .off = OFF, \ 352 .imm = 0 }) 353 354 /* Relative call */ 355 356 #define BPF_CALL_REL(TGT) \ 357 ((struct bpf_insn) { \ 358 .code = BPF_JMP | BPF_CALL, \ 359 .dst_reg = 0, \ 360 .src_reg = BPF_PSEUDO_CALL, \ 361 .off = 0, \ 362 .imm = TGT }) 363 364 /* Convert function address to BPF immediate */ 365 366 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base) 367 368 #define BPF_EMIT_CALL(FUNC) \ 369 ((struct bpf_insn) { \ 370 .code = BPF_JMP | BPF_CALL, \ 371 .dst_reg = 0, \ 372 .src_reg = 0, \ 373 .off = 0, \ 374 .imm = BPF_CALL_IMM(FUNC) }) 375 376 /* Raw code statement block */ 377 378 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 379 ((struct bpf_insn) { \ 380 .code = CODE, \ 381 .dst_reg = DST, \ 382 .src_reg = SRC, \ 383 .off = OFF, \ 384 .imm = IMM }) 385 386 /* Program exit */ 387 388 #define BPF_EXIT_INSN() \ 389 ((struct bpf_insn) { \ 390 .code = BPF_JMP | BPF_EXIT, \ 391 .dst_reg = 0, \ 392 .src_reg = 0, \ 393 .off = 0, \ 394 .imm = 0 }) 395 396 /* Speculation barrier */ 397 398 #define BPF_ST_NOSPEC() \ 399 ((struct bpf_insn) { \ 400 .code = BPF_ST | BPF_NOSPEC, \ 401 .dst_reg = 0, \ 402 .src_reg = 0, \ 403 .off = 0, \ 404 .imm = 0 }) 405 406 /* Internal classic blocks for direct assignment */ 407 408 #define __BPF_STMT(CODE, K) \ 409 ((struct sock_filter) BPF_STMT(CODE, K)) 410 411 #define __BPF_JUMP(CODE, K, JT, JF) \ 412 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 413 414 #define bytes_to_bpf_size(bytes) \ 415 ({ \ 416 int bpf_size = -EINVAL; \ 417 \ 418 if (bytes == sizeof(u8)) \ 419 bpf_size = BPF_B; \ 420 else if (bytes == sizeof(u16)) \ 421 bpf_size = BPF_H; \ 422 else if (bytes == sizeof(u32)) \ 423 bpf_size = BPF_W; \ 424 else if (bytes == sizeof(u64)) \ 425 bpf_size = BPF_DW; \ 426 \ 427 bpf_size; \ 428 }) 429 430 #define bpf_size_to_bytes(bpf_size) \ 431 ({ \ 432 int bytes = -EINVAL; \ 433 \ 434 if (bpf_size == BPF_B) \ 435 bytes = sizeof(u8); \ 436 else if (bpf_size == BPF_H) \ 437 bytes = sizeof(u16); \ 438 else if (bpf_size == BPF_W) \ 439 bytes = sizeof(u32); \ 440 else if (bpf_size == BPF_DW) \ 441 bytes = sizeof(u64); \ 442 \ 443 bytes; \ 444 }) 445 446 #define BPF_SIZEOF(type) \ 447 ({ \ 448 const int __size = bytes_to_bpf_size(sizeof(type)); \ 449 BUILD_BUG_ON(__size < 0); \ 450 __size; \ 451 }) 452 453 #define BPF_FIELD_SIZEOF(type, field) \ 454 ({ \ 455 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 456 BUILD_BUG_ON(__size < 0); \ 457 __size; \ 458 }) 459 460 #define BPF_LDST_BYTES(insn) \ 461 ({ \ 462 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 463 WARN_ON(__size < 0); \ 464 __size; \ 465 }) 466 467 #define __BPF_MAP_0(m, v, ...) v 468 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 469 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 470 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 471 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 472 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 473 474 #define __BPF_REG_0(...) __BPF_PAD(5) 475 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 476 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 477 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 478 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 479 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 480 481 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 482 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 483 484 #define __BPF_CAST(t, a) \ 485 (__force t) \ 486 (__force \ 487 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 488 (unsigned long)0, (t)0))) a 489 #define __BPF_V void 490 #define __BPF_N 491 492 #define __BPF_DECL_ARGS(t, a) t a 493 #define __BPF_DECL_REGS(t, a) u64 a 494 495 #define __BPF_PAD(n) \ 496 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 497 u64, __ur_3, u64, __ur_4, u64, __ur_5) 498 499 #define BPF_CALL_x(x, name, ...) \ 500 static __always_inline \ 501 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 502 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 503 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 504 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 505 { \ 506 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 507 } \ 508 static __always_inline \ 509 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 510 511 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) 512 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) 513 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) 514 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) 515 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) 516 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) 517 518 #define bpf_ctx_range(TYPE, MEMBER) \ 519 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 520 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 521 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 522 #if BITS_PER_LONG == 64 523 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 524 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 525 #else 526 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 527 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 528 #endif /* BITS_PER_LONG == 64 */ 529 530 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 531 ({ \ 532 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 533 *(PTR_SIZE) = (SIZE); \ 534 offsetof(TYPE, MEMBER); \ 535 }) 536 537 /* A struct sock_filter is architecture independent. */ 538 struct compat_sock_fprog { 539 u16 len; 540 compat_uptr_t filter; /* struct sock_filter * */ 541 }; 542 543 struct sock_fprog_kern { 544 u16 len; 545 struct sock_filter *filter; 546 }; 547 548 /* Some arches need doubleword alignment for their instructions and/or data */ 549 #define BPF_IMAGE_ALIGNMENT 8 550 551 struct bpf_binary_header { 552 u32 size; 553 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 554 }; 555 556 struct bpf_prog_stats { 557 u64_stats_t cnt; 558 u64_stats_t nsecs; 559 u64_stats_t misses; 560 struct u64_stats_sync syncp; 561 } __aligned(2 * sizeof(u64)); 562 563 struct sk_filter { 564 refcount_t refcnt; 565 struct rcu_head rcu; 566 struct bpf_prog *prog; 567 }; 568 569 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 570 571 extern struct mutex nf_conn_btf_access_lock; 572 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 573 const struct bpf_reg_state *reg, 574 int off, int size); 575 576 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 577 const struct bpf_insn *insnsi, 578 unsigned int (*bpf_func)(const void *, 579 const struct bpf_insn *)); 580 581 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 582 const void *ctx, 583 bpf_dispatcher_fn dfunc) 584 { 585 u32 ret; 586 587 cant_migrate(); 588 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 589 struct bpf_prog_stats *stats; 590 u64 start = sched_clock(); 591 unsigned long flags; 592 593 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 594 stats = this_cpu_ptr(prog->stats); 595 flags = u64_stats_update_begin_irqsave(&stats->syncp); 596 u64_stats_inc(&stats->cnt); 597 u64_stats_add(&stats->nsecs, sched_clock() - start); 598 u64_stats_update_end_irqrestore(&stats->syncp, flags); 599 } else { 600 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 601 } 602 return ret; 603 } 604 605 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 606 { 607 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 608 } 609 610 /* 611 * Use in preemptible and therefore migratable context to make sure that 612 * the execution of the BPF program runs on one CPU. 613 * 614 * This uses migrate_disable/enable() explicitly to document that the 615 * invocation of a BPF program does not require reentrancy protection 616 * against a BPF program which is invoked from a preempting task. 617 */ 618 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 619 const void *ctx) 620 { 621 u32 ret; 622 623 migrate_disable(); 624 ret = bpf_prog_run(prog, ctx); 625 migrate_enable(); 626 return ret; 627 } 628 629 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 630 631 struct bpf_skb_data_end { 632 struct qdisc_skb_cb qdisc_cb; 633 void *data_meta; 634 void *data_end; 635 }; 636 637 struct bpf_nh_params { 638 u32 nh_family; 639 union { 640 u32 ipv4_nh; 641 struct in6_addr ipv6_nh; 642 }; 643 }; 644 645 struct bpf_redirect_info { 646 u64 tgt_index; 647 void *tgt_value; 648 struct bpf_map *map; 649 u32 flags; 650 u32 kern_flags; 651 u32 map_id; 652 enum bpf_map_type map_type; 653 struct bpf_nh_params nh; 654 }; 655 656 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 657 658 /* flags for bpf_redirect_info kern_flags */ 659 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 660 661 /* Compute the linear packet data range [data, data_end) which 662 * will be accessed by various program types (cls_bpf, act_bpf, 663 * lwt, ...). Subsystems allowing direct data access must (!) 664 * ensure that cb[] area can be written to when BPF program is 665 * invoked (otherwise cb[] save/restore is necessary). 666 */ 667 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 668 { 669 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 670 671 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 672 cb->data_meta = skb->data - skb_metadata_len(skb); 673 cb->data_end = skb->data + skb_headlen(skb); 674 } 675 676 /* Similar to bpf_compute_data_pointers(), except that save orginal 677 * data in cb->data and cb->meta_data for restore. 678 */ 679 static inline void bpf_compute_and_save_data_end( 680 struct sk_buff *skb, void **saved_data_end) 681 { 682 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 683 684 *saved_data_end = cb->data_end; 685 cb->data_end = skb->data + skb_headlen(skb); 686 } 687 688 /* Restore data saved by bpf_compute_data_pointers(). */ 689 static inline void bpf_restore_data_end( 690 struct sk_buff *skb, void *saved_data_end) 691 { 692 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 693 694 cb->data_end = saved_data_end; 695 } 696 697 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 698 { 699 /* eBPF programs may read/write skb->cb[] area to transfer meta 700 * data between tail calls. Since this also needs to work with 701 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 702 * 703 * In some socket filter cases, the cb unfortunately needs to be 704 * saved/restored so that protocol specific skb->cb[] data won't 705 * be lost. In any case, due to unpriviledged eBPF programs 706 * attached to sockets, we need to clear the bpf_skb_cb() area 707 * to not leak previous contents to user space. 708 */ 709 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 710 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 711 sizeof_field(struct qdisc_skb_cb, data)); 712 713 return qdisc_skb_cb(skb)->data; 714 } 715 716 /* Must be invoked with migration disabled */ 717 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 718 const void *ctx) 719 { 720 const struct sk_buff *skb = ctx; 721 u8 *cb_data = bpf_skb_cb(skb); 722 u8 cb_saved[BPF_SKB_CB_LEN]; 723 u32 res; 724 725 if (unlikely(prog->cb_access)) { 726 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 727 memset(cb_data, 0, sizeof(cb_saved)); 728 } 729 730 res = bpf_prog_run(prog, skb); 731 732 if (unlikely(prog->cb_access)) 733 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 734 735 return res; 736 } 737 738 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 739 struct sk_buff *skb) 740 { 741 u32 res; 742 743 migrate_disable(); 744 res = __bpf_prog_run_save_cb(prog, skb); 745 migrate_enable(); 746 return res; 747 } 748 749 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 750 struct sk_buff *skb) 751 { 752 u8 *cb_data = bpf_skb_cb(skb); 753 u32 res; 754 755 if (unlikely(prog->cb_access)) 756 memset(cb_data, 0, BPF_SKB_CB_LEN); 757 758 res = bpf_prog_run_pin_on_cpu(prog, skb); 759 return res; 760 } 761 762 DECLARE_BPF_DISPATCHER(xdp) 763 764 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 765 766 u32 xdp_master_redirect(struct xdp_buff *xdp); 767 768 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 769 struct xdp_buff *xdp) 770 { 771 /* Driver XDP hooks are invoked within a single NAPI poll cycle and thus 772 * under local_bh_disable(), which provides the needed RCU protection 773 * for accessing map entries. 774 */ 775 u32 act = __bpf_prog_run(prog, xdp, BPF_DISPATCHER_FUNC(xdp)); 776 777 if (static_branch_unlikely(&bpf_master_redirect_enabled_key)) { 778 if (act == XDP_TX && netif_is_bond_slave(xdp->rxq->dev)) 779 act = xdp_master_redirect(xdp); 780 } 781 782 return act; 783 } 784 785 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 786 787 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 788 { 789 return prog->len * sizeof(struct bpf_insn); 790 } 791 792 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 793 { 794 return round_up(bpf_prog_insn_size(prog) + 795 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 796 } 797 798 static inline unsigned int bpf_prog_size(unsigned int proglen) 799 { 800 return max(sizeof(struct bpf_prog), 801 offsetof(struct bpf_prog, insns[proglen])); 802 } 803 804 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 805 { 806 /* When classic BPF programs have been loaded and the arch 807 * does not have a classic BPF JIT (anymore), they have been 808 * converted via bpf_migrate_filter() to eBPF and thus always 809 * have an unspec program type. 810 */ 811 return prog->type == BPF_PROG_TYPE_UNSPEC; 812 } 813 814 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 815 { 816 const u32 size_machine = sizeof(unsigned long); 817 818 if (size > size_machine && size % size_machine == 0) 819 size = size_machine; 820 821 return size; 822 } 823 824 static inline bool 825 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 826 { 827 return size <= size_default && (size & (size - 1)) == 0; 828 } 829 830 static inline u8 831 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 832 { 833 u8 access_off = off & (size_default - 1); 834 835 #ifdef __LITTLE_ENDIAN 836 return access_off; 837 #else 838 return size_default - (access_off + size); 839 #endif 840 } 841 842 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 843 (size == sizeof(__u64) && \ 844 off >= offsetof(type, field) && \ 845 off + sizeof(__u64) <= offsetofend(type, field) && \ 846 off % sizeof(__u64) == 0) 847 848 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 849 850 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 851 { 852 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 853 if (!fp->jited) { 854 set_vm_flush_reset_perms(fp); 855 set_memory_ro((unsigned long)fp, fp->pages); 856 } 857 #endif 858 } 859 860 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 861 { 862 set_vm_flush_reset_perms(hdr); 863 set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT); 864 } 865 866 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 867 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 868 { 869 return sk_filter_trim_cap(sk, skb, 1); 870 } 871 872 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 873 void bpf_prog_free(struct bpf_prog *fp); 874 875 bool bpf_opcode_in_insntable(u8 code); 876 877 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 878 const u32 *insn_to_jit_off); 879 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 880 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 881 882 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 883 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 884 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 885 gfp_t gfp_extra_flags); 886 void __bpf_prog_free(struct bpf_prog *fp); 887 888 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 889 { 890 __bpf_prog_free(fp); 891 } 892 893 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 894 unsigned int flen); 895 896 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 897 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 898 bpf_aux_classic_check_t trans, bool save_orig); 899 void bpf_prog_destroy(struct bpf_prog *fp); 900 901 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 902 int sk_attach_bpf(u32 ufd, struct sock *sk); 903 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 904 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 905 void sk_reuseport_prog_free(struct bpf_prog *prog); 906 int sk_detach_filter(struct sock *sk); 907 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); 908 909 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 910 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 911 912 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 913 #define __bpf_call_base_args \ 914 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 915 (void *)__bpf_call_base) 916 917 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 918 void bpf_jit_compile(struct bpf_prog *prog); 919 bool bpf_jit_needs_zext(void); 920 bool bpf_jit_supports_subprog_tailcalls(void); 921 bool bpf_jit_supports_kfunc_call(void); 922 bool bpf_jit_supports_far_kfunc_call(void); 923 bool bpf_helper_changes_pkt_data(void *func); 924 925 static inline bool bpf_dump_raw_ok(const struct cred *cred) 926 { 927 /* Reconstruction of call-sites is dependent on kallsyms, 928 * thus make dump the same restriction. 929 */ 930 return kallsyms_show_value(cred); 931 } 932 933 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 934 const struct bpf_insn *patch, u32 len); 935 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 936 937 void bpf_clear_redirect_map(struct bpf_map *map); 938 939 static inline bool xdp_return_frame_no_direct(void) 940 { 941 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 942 943 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 944 } 945 946 static inline void xdp_set_return_frame_no_direct(void) 947 { 948 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 949 950 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 951 } 952 953 static inline void xdp_clear_return_frame_no_direct(void) 954 { 955 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 956 957 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 958 } 959 960 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 961 unsigned int pktlen) 962 { 963 unsigned int len; 964 965 if (unlikely(!(fwd->flags & IFF_UP))) 966 return -ENETDOWN; 967 968 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 969 if (pktlen > len) 970 return -EMSGSIZE; 971 972 return 0; 973 } 974 975 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 976 * same cpu context. Further for best results no more than a single map 977 * for the do_redirect/do_flush pair should be used. This limitation is 978 * because we only track one map and force a flush when the map changes. 979 * This does not appear to be a real limitation for existing software. 980 */ 981 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 982 struct xdp_buff *xdp, struct bpf_prog *prog); 983 int xdp_do_redirect(struct net_device *dev, 984 struct xdp_buff *xdp, 985 struct bpf_prog *prog); 986 int xdp_do_redirect_frame(struct net_device *dev, 987 struct xdp_buff *xdp, 988 struct xdp_frame *xdpf, 989 struct bpf_prog *prog); 990 void xdp_do_flush(void); 991 992 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as 993 * it is no longer only flushing maps. Keep this define for compatibility 994 * until all drivers are updated - do not use xdp_do_flush_map() in new code! 995 */ 996 #define xdp_do_flush_map xdp_do_flush 997 998 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act); 999 1000 #ifdef CONFIG_INET 1001 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1002 struct bpf_prog *prog, struct sk_buff *skb, 1003 struct sock *migrating_sk, 1004 u32 hash); 1005 #else 1006 static inline struct sock * 1007 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1008 struct bpf_prog *prog, struct sk_buff *skb, 1009 struct sock *migrating_sk, 1010 u32 hash) 1011 { 1012 return NULL; 1013 } 1014 #endif 1015 1016 #ifdef CONFIG_BPF_JIT 1017 extern int bpf_jit_enable; 1018 extern int bpf_jit_harden; 1019 extern int bpf_jit_kallsyms; 1020 extern long bpf_jit_limit; 1021 extern long bpf_jit_limit_max; 1022 1023 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1024 1025 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); 1026 1027 struct bpf_binary_header * 1028 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1029 unsigned int alignment, 1030 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1031 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1032 u64 bpf_jit_alloc_exec_limit(void); 1033 void *bpf_jit_alloc_exec(unsigned long size); 1034 void bpf_jit_free_exec(void *addr); 1035 void bpf_jit_free(struct bpf_prog *fp); 1036 struct bpf_binary_header * 1037 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); 1038 1039 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); 1040 void bpf_prog_pack_free(struct bpf_binary_header *hdr); 1041 1042 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 1043 { 1044 return list_empty(&fp->aux->ksym.lnode) || 1045 fp->aux->ksym.lnode.prev == LIST_POISON2; 1046 } 1047 1048 struct bpf_binary_header * 1049 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, 1050 unsigned int alignment, 1051 struct bpf_binary_header **rw_hdr, 1052 u8 **rw_image, 1053 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1054 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, 1055 struct bpf_binary_header *ro_header, 1056 struct bpf_binary_header *rw_header); 1057 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, 1058 struct bpf_binary_header *rw_header); 1059 1060 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1061 struct bpf_jit_poke_descriptor *poke); 1062 1063 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1064 const struct bpf_insn *insn, bool extra_pass, 1065 u64 *func_addr, bool *func_addr_fixed); 1066 1067 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1068 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1069 1070 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1071 u32 pass, void *image) 1072 { 1073 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1074 proglen, pass, image, current->comm, task_pid_nr(current)); 1075 1076 if (image) 1077 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1078 16, 1, image, proglen, false); 1079 } 1080 1081 static inline bool bpf_jit_is_ebpf(void) 1082 { 1083 # ifdef CONFIG_HAVE_EBPF_JIT 1084 return true; 1085 # else 1086 return false; 1087 # endif 1088 } 1089 1090 static inline bool ebpf_jit_enabled(void) 1091 { 1092 return bpf_jit_enable && bpf_jit_is_ebpf(); 1093 } 1094 1095 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1096 { 1097 return fp->jited && bpf_jit_is_ebpf(); 1098 } 1099 1100 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1101 { 1102 /* These are the prerequisites, should someone ever have the 1103 * idea to call blinding outside of them, we make sure to 1104 * bail out. 1105 */ 1106 if (!bpf_jit_is_ebpf()) 1107 return false; 1108 if (!prog->jit_requested) 1109 return false; 1110 if (!bpf_jit_harden) 1111 return false; 1112 if (bpf_jit_harden == 1 && bpf_capable()) 1113 return false; 1114 1115 return true; 1116 } 1117 1118 static inline bool bpf_jit_kallsyms_enabled(void) 1119 { 1120 /* There are a couple of corner cases where kallsyms should 1121 * not be enabled f.e. on hardening. 1122 */ 1123 if (bpf_jit_harden) 1124 return false; 1125 if (!bpf_jit_kallsyms) 1126 return false; 1127 if (bpf_jit_kallsyms == 1) 1128 return true; 1129 1130 return false; 1131 } 1132 1133 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1134 unsigned long *off, char *sym); 1135 bool is_bpf_text_address(unsigned long addr); 1136 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1137 char *sym); 1138 1139 static inline const char * 1140 bpf_address_lookup(unsigned long addr, unsigned long *size, 1141 unsigned long *off, char **modname, char *sym) 1142 { 1143 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1144 1145 if (ret && modname) 1146 *modname = NULL; 1147 return ret; 1148 } 1149 1150 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1151 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1152 1153 #else /* CONFIG_BPF_JIT */ 1154 1155 static inline bool ebpf_jit_enabled(void) 1156 { 1157 return false; 1158 } 1159 1160 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1161 { 1162 return false; 1163 } 1164 1165 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1166 { 1167 return false; 1168 } 1169 1170 static inline int 1171 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1172 struct bpf_jit_poke_descriptor *poke) 1173 { 1174 return -ENOTSUPP; 1175 } 1176 1177 static inline void bpf_jit_free(struct bpf_prog *fp) 1178 { 1179 bpf_prog_unlock_free(fp); 1180 } 1181 1182 static inline bool bpf_jit_kallsyms_enabled(void) 1183 { 1184 return false; 1185 } 1186 1187 static inline const char * 1188 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1189 unsigned long *off, char *sym) 1190 { 1191 return NULL; 1192 } 1193 1194 static inline bool is_bpf_text_address(unsigned long addr) 1195 { 1196 return false; 1197 } 1198 1199 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1200 char *type, char *sym) 1201 { 1202 return -ERANGE; 1203 } 1204 1205 static inline const char * 1206 bpf_address_lookup(unsigned long addr, unsigned long *size, 1207 unsigned long *off, char **modname, char *sym) 1208 { 1209 return NULL; 1210 } 1211 1212 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1213 { 1214 } 1215 1216 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1217 { 1218 } 1219 1220 #endif /* CONFIG_BPF_JIT */ 1221 1222 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1223 1224 #define BPF_ANC BIT(15) 1225 1226 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1227 { 1228 switch (first->code) { 1229 case BPF_RET | BPF_K: 1230 case BPF_LD | BPF_W | BPF_LEN: 1231 return false; 1232 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 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1237 return true; 1238 return false; 1239 1240 default: 1241 return true; 1242 } 1243 } 1244 1245 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1246 { 1247 BUG_ON(ftest->code & BPF_ANC); 1248 1249 switch (ftest->code) { 1250 case BPF_LD | BPF_W | BPF_ABS: 1251 case BPF_LD | BPF_H | BPF_ABS: 1252 case BPF_LD | BPF_B | BPF_ABS: 1253 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1254 return BPF_ANC | SKF_AD_##CODE 1255 switch (ftest->k) { 1256 BPF_ANCILLARY(PROTOCOL); 1257 BPF_ANCILLARY(PKTTYPE); 1258 BPF_ANCILLARY(IFINDEX); 1259 BPF_ANCILLARY(NLATTR); 1260 BPF_ANCILLARY(NLATTR_NEST); 1261 BPF_ANCILLARY(MARK); 1262 BPF_ANCILLARY(QUEUE); 1263 BPF_ANCILLARY(HATYPE); 1264 BPF_ANCILLARY(RXHASH); 1265 BPF_ANCILLARY(CPU); 1266 BPF_ANCILLARY(ALU_XOR_X); 1267 BPF_ANCILLARY(VLAN_TAG); 1268 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1269 BPF_ANCILLARY(PAY_OFFSET); 1270 BPF_ANCILLARY(RANDOM); 1271 BPF_ANCILLARY(VLAN_TPID); 1272 } 1273 fallthrough; 1274 default: 1275 return ftest->code; 1276 } 1277 } 1278 1279 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1280 int k, unsigned int size); 1281 1282 static inline int bpf_tell_extensions(void) 1283 { 1284 return SKF_AD_MAX; 1285 } 1286 1287 struct bpf_sock_addr_kern { 1288 struct sock *sk; 1289 struct sockaddr *uaddr; 1290 /* Temporary "register" to make indirect stores to nested structures 1291 * defined above. We need three registers to make such a store, but 1292 * only two (src and dst) are available at convert_ctx_access time 1293 */ 1294 u64 tmp_reg; 1295 void *t_ctx; /* Attach type specific context. */ 1296 }; 1297 1298 struct bpf_sock_ops_kern { 1299 struct sock *sk; 1300 union { 1301 u32 args[4]; 1302 u32 reply; 1303 u32 replylong[4]; 1304 }; 1305 struct sk_buff *syn_skb; 1306 struct sk_buff *skb; 1307 void *skb_data_end; 1308 u8 op; 1309 u8 is_fullsock; 1310 u8 remaining_opt_len; 1311 u64 temp; /* temp and everything after is not 1312 * initialized to 0 before calling 1313 * the BPF program. New fields that 1314 * should be initialized to 0 should 1315 * be inserted before temp. 1316 * temp is scratch storage used by 1317 * sock_ops_convert_ctx_access 1318 * as temporary storage of a register. 1319 */ 1320 }; 1321 1322 struct bpf_sysctl_kern { 1323 struct ctl_table_header *head; 1324 struct ctl_table *table; 1325 void *cur_val; 1326 size_t cur_len; 1327 void *new_val; 1328 size_t new_len; 1329 int new_updated; 1330 int write; 1331 loff_t *ppos; 1332 /* Temporary "register" for indirect stores to ppos. */ 1333 u64 tmp_reg; 1334 }; 1335 1336 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1337 struct bpf_sockopt_buf { 1338 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1339 }; 1340 1341 struct bpf_sockopt_kern { 1342 struct sock *sk; 1343 u8 *optval; 1344 u8 *optval_end; 1345 s32 level; 1346 s32 optname; 1347 s32 optlen; 1348 /* for retval in struct bpf_cg_run_ctx */ 1349 struct task_struct *current_task; 1350 /* Temporary "register" for indirect stores to ppos. */ 1351 u64 tmp_reg; 1352 }; 1353 1354 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1355 1356 struct bpf_sk_lookup_kern { 1357 u16 family; 1358 u16 protocol; 1359 __be16 sport; 1360 u16 dport; 1361 struct { 1362 __be32 saddr; 1363 __be32 daddr; 1364 } v4; 1365 struct { 1366 const struct in6_addr *saddr; 1367 const struct in6_addr *daddr; 1368 } v6; 1369 struct sock *selected_sk; 1370 u32 ingress_ifindex; 1371 bool no_reuseport; 1372 }; 1373 1374 extern struct static_key_false bpf_sk_lookup_enabled; 1375 1376 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1377 * 1378 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1379 * SK_DROP. Their meaning is as follows: 1380 * 1381 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1382 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1383 * SK_DROP : terminate lookup with -ECONNREFUSED 1384 * 1385 * This macro aggregates return values and selected sockets from 1386 * multiple BPF programs according to following rules in order: 1387 * 1388 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1389 * macro result is SK_PASS and last ctx.selected_sk is used. 1390 * 2. If any program returned SK_DROP return value, 1391 * macro result is SK_DROP. 1392 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1393 * 1394 * Caller must ensure that the prog array is non-NULL, and that the 1395 * array as well as the programs it contains remain valid. 1396 */ 1397 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1398 ({ \ 1399 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1400 struct bpf_prog_array_item *_item; \ 1401 struct sock *_selected_sk = NULL; \ 1402 bool _no_reuseport = false; \ 1403 struct bpf_prog *_prog; \ 1404 bool _all_pass = true; \ 1405 u32 _ret; \ 1406 \ 1407 migrate_disable(); \ 1408 _item = &(array)->items[0]; \ 1409 while ((_prog = READ_ONCE(_item->prog))) { \ 1410 /* restore most recent selection */ \ 1411 _ctx->selected_sk = _selected_sk; \ 1412 _ctx->no_reuseport = _no_reuseport; \ 1413 \ 1414 _ret = func(_prog, _ctx); \ 1415 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1416 /* remember last non-NULL socket */ \ 1417 _selected_sk = _ctx->selected_sk; \ 1418 _no_reuseport = _ctx->no_reuseport; \ 1419 } else if (_ret == SK_DROP && _all_pass) { \ 1420 _all_pass = false; \ 1421 } \ 1422 _item++; \ 1423 } \ 1424 _ctx->selected_sk = _selected_sk; \ 1425 _ctx->no_reuseport = _no_reuseport; \ 1426 migrate_enable(); \ 1427 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1428 }) 1429 1430 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1431 const __be32 saddr, const __be16 sport, 1432 const __be32 daddr, const u16 dport, 1433 const int ifindex, struct sock **psk) 1434 { 1435 struct bpf_prog_array *run_array; 1436 struct sock *selected_sk = NULL; 1437 bool no_reuseport = false; 1438 1439 rcu_read_lock(); 1440 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1441 if (run_array) { 1442 struct bpf_sk_lookup_kern ctx = { 1443 .family = AF_INET, 1444 .protocol = protocol, 1445 .v4.saddr = saddr, 1446 .v4.daddr = daddr, 1447 .sport = sport, 1448 .dport = dport, 1449 .ingress_ifindex = ifindex, 1450 }; 1451 u32 act; 1452 1453 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1454 if (act == SK_PASS) { 1455 selected_sk = ctx.selected_sk; 1456 no_reuseport = ctx.no_reuseport; 1457 } else { 1458 selected_sk = ERR_PTR(-ECONNREFUSED); 1459 } 1460 } 1461 rcu_read_unlock(); 1462 *psk = selected_sk; 1463 return no_reuseport; 1464 } 1465 1466 #if IS_ENABLED(CONFIG_IPV6) 1467 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1468 const struct in6_addr *saddr, 1469 const __be16 sport, 1470 const struct in6_addr *daddr, 1471 const u16 dport, 1472 const int ifindex, struct sock **psk) 1473 { 1474 struct bpf_prog_array *run_array; 1475 struct sock *selected_sk = NULL; 1476 bool no_reuseport = false; 1477 1478 rcu_read_lock(); 1479 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1480 if (run_array) { 1481 struct bpf_sk_lookup_kern ctx = { 1482 .family = AF_INET6, 1483 .protocol = protocol, 1484 .v6.saddr = saddr, 1485 .v6.daddr = daddr, 1486 .sport = sport, 1487 .dport = dport, 1488 .ingress_ifindex = ifindex, 1489 }; 1490 u32 act; 1491 1492 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1493 if (act == SK_PASS) { 1494 selected_sk = ctx.selected_sk; 1495 no_reuseport = ctx.no_reuseport; 1496 } else { 1497 selected_sk = ERR_PTR(-ECONNREFUSED); 1498 } 1499 } 1500 rcu_read_unlock(); 1501 *psk = selected_sk; 1502 return no_reuseport; 1503 } 1504 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1505 1506 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, 1507 u64 flags, const u64 flag_mask, 1508 void *lookup_elem(struct bpf_map *map, u32 key)) 1509 { 1510 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1511 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1512 1513 /* Lower bits of the flags are used as return code on lookup failure */ 1514 if (unlikely(flags & ~(action_mask | flag_mask))) 1515 return XDP_ABORTED; 1516 1517 ri->tgt_value = lookup_elem(map, index); 1518 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1519 /* If the lookup fails we want to clear out the state in the 1520 * redirect_info struct completely, so that if an eBPF program 1521 * performs multiple lookups, the last one always takes 1522 * precedence. 1523 */ 1524 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1525 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1526 return flags & action_mask; 1527 } 1528 1529 ri->tgt_index = index; 1530 ri->map_id = map->id; 1531 ri->map_type = map->map_type; 1532 1533 if (flags & BPF_F_BROADCAST) { 1534 WRITE_ONCE(ri->map, map); 1535 ri->flags = flags; 1536 } else { 1537 WRITE_ONCE(ri->map, NULL); 1538 ri->flags = 0; 1539 } 1540 1541 return XDP_REDIRECT; 1542 } 1543 1544 #ifdef CONFIG_NET 1545 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); 1546 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1547 u32 len, u64 flags); 1548 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1549 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1550 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); 1551 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 1552 void *buf, unsigned long len, bool flush); 1553 #else /* CONFIG_NET */ 1554 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, 1555 void *to, u32 len) 1556 { 1557 return -EOPNOTSUPP; 1558 } 1559 1560 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, 1561 const void *from, u32 len, u64 flags) 1562 { 1563 return -EOPNOTSUPP; 1564 } 1565 1566 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, 1567 void *buf, u32 len) 1568 { 1569 return -EOPNOTSUPP; 1570 } 1571 1572 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, 1573 void *buf, u32 len) 1574 { 1575 return -EOPNOTSUPP; 1576 } 1577 1578 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 1579 { 1580 return NULL; 1581 } 1582 1583 static inline void *bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, 1584 unsigned long len, bool flush) 1585 { 1586 return NULL; 1587 } 1588 #endif /* CONFIG_NET */ 1589 1590 #endif /* __LINUX_FILTER_H__ */ 1591