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/refcount.h> 10 #include <linux/compat.h> 11 #include <linux/skbuff.h> 12 #include <linux/linkage.h> 13 #include <linux/printk.h> 14 #include <linux/workqueue.h> 15 #include <linux/sched.h> 16 #include <linux/capability.h> 17 #include <linux/set_memory.h> 18 #include <linux/kallsyms.h> 19 #include <linux/if_vlan.h> 20 #include <linux/vmalloc.h> 21 #include <linux/sockptr.h> 22 #include <crypto/sha1.h> 23 #include <linux/u64_stats_sync.h> 24 25 #include <net/sch_generic.h> 26 27 #include <asm/byteorder.h> 28 #include <uapi/linux/filter.h> 29 #include <uapi/linux/bpf.h> 30 31 struct sk_buff; 32 struct sock; 33 struct seccomp_data; 34 struct bpf_prog_aux; 35 struct xdp_rxq_info; 36 struct xdp_buff; 37 struct sock_reuseport; 38 struct ctl_table; 39 struct ctl_table_header; 40 41 /* ArgX, context and stack frame pointer register positions. Note, 42 * Arg1, Arg2, Arg3, etc are used as argument mappings of function 43 * calls in BPF_CALL instruction. 44 */ 45 #define BPF_REG_ARG1 BPF_REG_1 46 #define BPF_REG_ARG2 BPF_REG_2 47 #define BPF_REG_ARG3 BPF_REG_3 48 #define BPF_REG_ARG4 BPF_REG_4 49 #define BPF_REG_ARG5 BPF_REG_5 50 #define BPF_REG_CTX BPF_REG_6 51 #define BPF_REG_FP BPF_REG_10 52 53 /* Additional register mappings for converted user programs. */ 54 #define BPF_REG_A BPF_REG_0 55 #define BPF_REG_X BPF_REG_7 56 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */ 57 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */ 58 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */ 59 60 /* Kernel hidden auxiliary/helper register. */ 61 #define BPF_REG_AX MAX_BPF_REG 62 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1) 63 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG 64 65 /* unused opcode to mark special call to bpf_tail_call() helper */ 66 #define BPF_TAIL_CALL 0xf0 67 68 /* unused opcode to mark special load instruction. Same as BPF_ABS */ 69 #define BPF_PROBE_MEM 0x20 70 71 /* unused opcode to mark call to interpreter with arguments */ 72 #define BPF_CALL_ARGS 0xe0 73 74 /* unused opcode to mark speculation barrier for mitigating 75 * Speculative Store Bypass 76 */ 77 #define BPF_NOSPEC 0xc0 78 79 /* As per nm, we expose JITed images as text (code) section for 80 * kallsyms. That way, tools like perf can find it to match 81 * addresses. 82 */ 83 #define BPF_SYM_ELF_TYPE 't' 84 85 /* BPF program can access up to 512 bytes of stack space. */ 86 #define MAX_BPF_STACK 512 87 88 /* Helper macros for filter block array initializers. */ 89 90 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 91 92 #define BPF_ALU64_REG(OP, DST, SRC) \ 93 ((struct bpf_insn) { \ 94 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 95 .dst_reg = DST, \ 96 .src_reg = SRC, \ 97 .off = 0, \ 98 .imm = 0 }) 99 100 #define BPF_ALU32_REG(OP, DST, SRC) \ 101 ((struct bpf_insn) { \ 102 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 103 .dst_reg = DST, \ 104 .src_reg = SRC, \ 105 .off = 0, \ 106 .imm = 0 }) 107 108 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 109 110 #define BPF_ALU64_IMM(OP, DST, IMM) \ 111 ((struct bpf_insn) { \ 112 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 113 .dst_reg = DST, \ 114 .src_reg = 0, \ 115 .off = 0, \ 116 .imm = IMM }) 117 118 #define BPF_ALU32_IMM(OP, DST, IMM) \ 119 ((struct bpf_insn) { \ 120 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 121 .dst_reg = DST, \ 122 .src_reg = 0, \ 123 .off = 0, \ 124 .imm = IMM }) 125 126 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 127 128 #define BPF_ENDIAN(TYPE, DST, LEN) \ 129 ((struct bpf_insn) { \ 130 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 131 .dst_reg = DST, \ 132 .src_reg = 0, \ 133 .off = 0, \ 134 .imm = LEN }) 135 136 /* Short form of mov, dst_reg = src_reg */ 137 138 #define BPF_MOV64_REG(DST, SRC) \ 139 ((struct bpf_insn) { \ 140 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 141 .dst_reg = DST, \ 142 .src_reg = SRC, \ 143 .off = 0, \ 144 .imm = 0 }) 145 146 #define BPF_MOV32_REG(DST, SRC) \ 147 ((struct bpf_insn) { \ 148 .code = BPF_ALU | BPF_MOV | BPF_X, \ 149 .dst_reg = DST, \ 150 .src_reg = SRC, \ 151 .off = 0, \ 152 .imm = 0 }) 153 154 /* Short form of mov, dst_reg = imm32 */ 155 156 #define BPF_MOV64_IMM(DST, IMM) \ 157 ((struct bpf_insn) { \ 158 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 159 .dst_reg = DST, \ 160 .src_reg = 0, \ 161 .off = 0, \ 162 .imm = IMM }) 163 164 #define BPF_MOV32_IMM(DST, IMM) \ 165 ((struct bpf_insn) { \ 166 .code = BPF_ALU | BPF_MOV | BPF_K, \ 167 .dst_reg = DST, \ 168 .src_reg = 0, \ 169 .off = 0, \ 170 .imm = IMM }) 171 172 /* Special form of mov32, used for doing explicit zero extension on dst. */ 173 #define BPF_ZEXT_REG(DST) \ 174 ((struct bpf_insn) { \ 175 .code = BPF_ALU | BPF_MOV | BPF_X, \ 176 .dst_reg = DST, \ 177 .src_reg = DST, \ 178 .off = 0, \ 179 .imm = 1 }) 180 181 static inline bool insn_is_zext(const struct bpf_insn *insn) 182 { 183 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 184 } 185 186 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 187 #define BPF_LD_IMM64(DST, IMM) \ 188 BPF_LD_IMM64_RAW(DST, 0, IMM) 189 190 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 191 ((struct bpf_insn) { \ 192 .code = BPF_LD | BPF_DW | BPF_IMM, \ 193 .dst_reg = DST, \ 194 .src_reg = SRC, \ 195 .off = 0, \ 196 .imm = (__u32) (IMM) }), \ 197 ((struct bpf_insn) { \ 198 .code = 0, /* zero is reserved opcode */ \ 199 .dst_reg = 0, \ 200 .src_reg = 0, \ 201 .off = 0, \ 202 .imm = ((__u64) (IMM)) >> 32 }) 203 204 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 205 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 206 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 207 208 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 209 210 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 211 ((struct bpf_insn) { \ 212 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 213 .dst_reg = DST, \ 214 .src_reg = SRC, \ 215 .off = 0, \ 216 .imm = IMM }) 217 218 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 219 ((struct bpf_insn) { \ 220 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 221 .dst_reg = DST, \ 222 .src_reg = SRC, \ 223 .off = 0, \ 224 .imm = IMM }) 225 226 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 227 228 #define BPF_LD_ABS(SIZE, IMM) \ 229 ((struct bpf_insn) { \ 230 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 231 .dst_reg = 0, \ 232 .src_reg = 0, \ 233 .off = 0, \ 234 .imm = IMM }) 235 236 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 237 238 #define BPF_LD_IND(SIZE, SRC, IMM) \ 239 ((struct bpf_insn) { \ 240 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 241 .dst_reg = 0, \ 242 .src_reg = SRC, \ 243 .off = 0, \ 244 .imm = IMM }) 245 246 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 247 248 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 249 ((struct bpf_insn) { \ 250 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 251 .dst_reg = DST, \ 252 .src_reg = SRC, \ 253 .off = OFF, \ 254 .imm = 0 }) 255 256 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 257 258 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 259 ((struct bpf_insn) { \ 260 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 261 .dst_reg = DST, \ 262 .src_reg = SRC, \ 263 .off = OFF, \ 264 .imm = 0 }) 265 266 267 /* 268 * Atomic operations: 269 * 270 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 271 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 272 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 273 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 274 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 275 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 276 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 277 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 278 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 279 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 280 */ 281 282 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 283 ((struct bpf_insn) { \ 284 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 285 .dst_reg = DST, \ 286 .src_reg = SRC, \ 287 .off = OFF, \ 288 .imm = OP }) 289 290 /* Legacy alias */ 291 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 292 293 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 294 295 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 296 ((struct bpf_insn) { \ 297 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 298 .dst_reg = DST, \ 299 .src_reg = 0, \ 300 .off = OFF, \ 301 .imm = IMM }) 302 303 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 304 305 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 306 ((struct bpf_insn) { \ 307 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 308 .dst_reg = DST, \ 309 .src_reg = SRC, \ 310 .off = OFF, \ 311 .imm = 0 }) 312 313 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 314 315 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 316 ((struct bpf_insn) { \ 317 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 318 .dst_reg = DST, \ 319 .src_reg = 0, \ 320 .off = OFF, \ 321 .imm = IMM }) 322 323 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 324 325 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 326 ((struct bpf_insn) { \ 327 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 328 .dst_reg = DST, \ 329 .src_reg = SRC, \ 330 .off = OFF, \ 331 .imm = 0 }) 332 333 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 334 335 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 336 ((struct bpf_insn) { \ 337 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 338 .dst_reg = DST, \ 339 .src_reg = 0, \ 340 .off = OFF, \ 341 .imm = IMM }) 342 343 /* Unconditional jumps, goto pc + off16 */ 344 345 #define BPF_JMP_A(OFF) \ 346 ((struct bpf_insn) { \ 347 .code = BPF_JMP | BPF_JA, \ 348 .dst_reg = 0, \ 349 .src_reg = 0, \ 350 .off = OFF, \ 351 .imm = 0 }) 352 353 /* Relative call */ 354 355 #define BPF_CALL_REL(TGT) \ 356 ((struct bpf_insn) { \ 357 .code = BPF_JMP | BPF_CALL, \ 358 .dst_reg = 0, \ 359 .src_reg = BPF_PSEUDO_CALL, \ 360 .off = 0, \ 361 .imm = TGT }) 362 363 /* Convert function address to BPF immediate */ 364 365 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base) 366 367 #define BPF_EMIT_CALL(FUNC) \ 368 ((struct bpf_insn) { \ 369 .code = BPF_JMP | BPF_CALL, \ 370 .dst_reg = 0, \ 371 .src_reg = 0, \ 372 .off = 0, \ 373 .imm = BPF_CALL_IMM(FUNC) }) 374 375 /* Raw code statement block */ 376 377 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 378 ((struct bpf_insn) { \ 379 .code = CODE, \ 380 .dst_reg = DST, \ 381 .src_reg = SRC, \ 382 .off = OFF, \ 383 .imm = IMM }) 384 385 /* Program exit */ 386 387 #define BPF_EXIT_INSN() \ 388 ((struct bpf_insn) { \ 389 .code = BPF_JMP | BPF_EXIT, \ 390 .dst_reg = 0, \ 391 .src_reg = 0, \ 392 .off = 0, \ 393 .imm = 0 }) 394 395 /* Speculation barrier */ 396 397 #define BPF_ST_NOSPEC() \ 398 ((struct bpf_insn) { \ 399 .code = BPF_ST | BPF_NOSPEC, \ 400 .dst_reg = 0, \ 401 .src_reg = 0, \ 402 .off = 0, \ 403 .imm = 0 }) 404 405 /* Internal classic blocks for direct assignment */ 406 407 #define __BPF_STMT(CODE, K) \ 408 ((struct sock_filter) BPF_STMT(CODE, K)) 409 410 #define __BPF_JUMP(CODE, K, JT, JF) \ 411 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 412 413 #define bytes_to_bpf_size(bytes) \ 414 ({ \ 415 int bpf_size = -EINVAL; \ 416 \ 417 if (bytes == sizeof(u8)) \ 418 bpf_size = BPF_B; \ 419 else if (bytes == sizeof(u16)) \ 420 bpf_size = BPF_H; \ 421 else if (bytes == sizeof(u32)) \ 422 bpf_size = BPF_W; \ 423 else if (bytes == sizeof(u64)) \ 424 bpf_size = BPF_DW; \ 425 \ 426 bpf_size; \ 427 }) 428 429 #define bpf_size_to_bytes(bpf_size) \ 430 ({ \ 431 int bytes = -EINVAL; \ 432 \ 433 if (bpf_size == BPF_B) \ 434 bytes = sizeof(u8); \ 435 else if (bpf_size == BPF_H) \ 436 bytes = sizeof(u16); \ 437 else if (bpf_size == BPF_W) \ 438 bytes = sizeof(u32); \ 439 else if (bpf_size == BPF_DW) \ 440 bytes = sizeof(u64); \ 441 \ 442 bytes; \ 443 }) 444 445 #define BPF_SIZEOF(type) \ 446 ({ \ 447 const int __size = bytes_to_bpf_size(sizeof(type)); \ 448 BUILD_BUG_ON(__size < 0); \ 449 __size; \ 450 }) 451 452 #define BPF_FIELD_SIZEOF(type, field) \ 453 ({ \ 454 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 455 BUILD_BUG_ON(__size < 0); \ 456 __size; \ 457 }) 458 459 #define BPF_LDST_BYTES(insn) \ 460 ({ \ 461 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 462 WARN_ON(__size < 0); \ 463 __size; \ 464 }) 465 466 #define __BPF_MAP_0(m, v, ...) v 467 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 468 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 469 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 470 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 471 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 472 473 #define __BPF_REG_0(...) __BPF_PAD(5) 474 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 475 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 476 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 477 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 478 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 479 480 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 481 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 482 483 #define __BPF_CAST(t, a) \ 484 (__force t) \ 485 (__force \ 486 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 487 (unsigned long)0, (t)0))) a 488 #define __BPF_V void 489 #define __BPF_N 490 491 #define __BPF_DECL_ARGS(t, a) t a 492 #define __BPF_DECL_REGS(t, a) u64 a 493 494 #define __BPF_PAD(n) \ 495 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 496 u64, __ur_3, u64, __ur_4, u64, __ur_5) 497 498 #define BPF_CALL_x(x, name, ...) \ 499 static __always_inline \ 500 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 501 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 502 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 503 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 504 { \ 505 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 506 } \ 507 static __always_inline \ 508 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 509 510 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) 511 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) 512 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) 513 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) 514 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) 515 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) 516 517 #define bpf_ctx_range(TYPE, MEMBER) \ 518 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 519 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 520 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 521 #if BITS_PER_LONG == 64 522 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 523 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 524 #else 525 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 526 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 527 #endif /* BITS_PER_LONG == 64 */ 528 529 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 530 ({ \ 531 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 532 *(PTR_SIZE) = (SIZE); \ 533 offsetof(TYPE, MEMBER); \ 534 }) 535 536 /* A struct sock_filter is architecture independent. */ 537 struct compat_sock_fprog { 538 u16 len; 539 compat_uptr_t filter; /* struct sock_filter * */ 540 }; 541 542 struct sock_fprog_kern { 543 u16 len; 544 struct sock_filter *filter; 545 }; 546 547 /* Some arches need doubleword alignment for their instructions and/or data */ 548 #define BPF_IMAGE_ALIGNMENT 8 549 550 struct bpf_binary_header { 551 u32 pages; 552 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 553 }; 554 555 struct bpf_prog_stats { 556 u64 cnt; 557 u64 nsecs; 558 u64 misses; 559 struct u64_stats_sync syncp; 560 } __aligned(2 * sizeof(u64)); 561 562 struct bpf_prog { 563 u16 pages; /* Number of allocated pages */ 564 u16 jited:1, /* Is our filter JIT'ed? */ 565 jit_requested:1,/* archs need to JIT the prog */ 566 gpl_compatible:1, /* Is filter GPL compatible? */ 567 cb_access:1, /* Is control block accessed? */ 568 dst_needed:1, /* Do we need dst entry? */ 569 blinded:1, /* Was blinded */ 570 is_func:1, /* program is a bpf function */ 571 kprobe_override:1, /* Do we override a kprobe? */ 572 has_callchain_buf:1, /* callchain buffer allocated? */ 573 enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */ 574 call_get_stack:1, /* Do we call bpf_get_stack() or bpf_get_stackid() */ 575 call_get_func_ip:1; /* Do we call get_func_ip() */ 576 enum bpf_prog_type type; /* Type of BPF program */ 577 enum bpf_attach_type expected_attach_type; /* For some prog types */ 578 u32 len; /* Number of filter blocks */ 579 u32 jited_len; /* Size of jited insns in bytes */ 580 u8 tag[BPF_TAG_SIZE]; 581 struct bpf_prog_stats __percpu *stats; 582 int __percpu *active; 583 unsigned int (*bpf_func)(const void *ctx, 584 const struct bpf_insn *insn); 585 struct bpf_prog_aux *aux; /* Auxiliary fields */ 586 struct sock_fprog_kern *orig_prog; /* Original BPF program */ 587 /* Instructions for interpreter */ 588 struct sock_filter insns[0]; 589 struct bpf_insn insnsi[]; 590 }; 591 592 struct sk_filter { 593 refcount_t refcnt; 594 struct rcu_head rcu; 595 struct bpf_prog *prog; 596 }; 597 598 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 599 600 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 601 const struct bpf_insn *insnsi, 602 unsigned int (*bpf_func)(const void *, 603 const struct bpf_insn *)); 604 605 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 606 const void *ctx, 607 bpf_dispatcher_fn dfunc) 608 { 609 u32 ret; 610 611 cant_migrate(); 612 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 613 struct bpf_prog_stats *stats; 614 u64 start = sched_clock(); 615 616 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 617 stats = this_cpu_ptr(prog->stats); 618 u64_stats_update_begin(&stats->syncp); 619 stats->cnt++; 620 stats->nsecs += sched_clock() - start; 621 u64_stats_update_end(&stats->syncp); 622 } else { 623 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 624 } 625 return ret; 626 } 627 628 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 629 { 630 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 631 } 632 633 /* 634 * Use in preemptible and therefore migratable context to make sure that 635 * the execution of the BPF program runs on one CPU. 636 * 637 * This uses migrate_disable/enable() explicitly to document that the 638 * invocation of a BPF program does not require reentrancy protection 639 * against a BPF program which is invoked from a preempting task. 640 * 641 * For non RT enabled kernels migrate_disable/enable() maps to 642 * preempt_disable/enable(), i.e. it disables also preemption. 643 */ 644 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 645 const void *ctx) 646 { 647 u32 ret; 648 649 migrate_disable(); 650 ret = bpf_prog_run(prog, ctx); 651 migrate_enable(); 652 return ret; 653 } 654 655 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 656 657 struct bpf_skb_data_end { 658 struct qdisc_skb_cb qdisc_cb; 659 void *data_meta; 660 void *data_end; 661 }; 662 663 struct bpf_nh_params { 664 u32 nh_family; 665 union { 666 u32 ipv4_nh; 667 struct in6_addr ipv6_nh; 668 }; 669 }; 670 671 struct bpf_redirect_info { 672 u32 flags; 673 u32 tgt_index; 674 void *tgt_value; 675 struct bpf_map *map; 676 u32 map_id; 677 enum bpf_map_type map_type; 678 u32 kern_flags; 679 struct bpf_nh_params nh; 680 }; 681 682 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 683 684 /* flags for bpf_redirect_info kern_flags */ 685 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 686 687 /* Compute the linear packet data range [data, data_end) which 688 * will be accessed by various program types (cls_bpf, act_bpf, 689 * lwt, ...). Subsystems allowing direct data access must (!) 690 * ensure that cb[] area can be written to when BPF program is 691 * invoked (otherwise cb[] save/restore is necessary). 692 */ 693 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 694 { 695 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 696 697 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 698 cb->data_meta = skb->data - skb_metadata_len(skb); 699 cb->data_end = skb->data + skb_headlen(skb); 700 } 701 702 /* Similar to bpf_compute_data_pointers(), except that save orginal 703 * data in cb->data and cb->meta_data for restore. 704 */ 705 static inline void bpf_compute_and_save_data_end( 706 struct sk_buff *skb, void **saved_data_end) 707 { 708 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 709 710 *saved_data_end = cb->data_end; 711 cb->data_end = skb->data + skb_headlen(skb); 712 } 713 714 /* Restore data saved by bpf_compute_data_pointers(). */ 715 static inline void bpf_restore_data_end( 716 struct sk_buff *skb, void *saved_data_end) 717 { 718 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 719 720 cb->data_end = saved_data_end; 721 } 722 723 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 724 { 725 /* eBPF programs may read/write skb->cb[] area to transfer meta 726 * data between tail calls. Since this also needs to work with 727 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 728 * 729 * In some socket filter cases, the cb unfortunately needs to be 730 * saved/restored so that protocol specific skb->cb[] data won't 731 * be lost. In any case, due to unpriviledged eBPF programs 732 * attached to sockets, we need to clear the bpf_skb_cb() area 733 * to not leak previous contents to user space. 734 */ 735 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 736 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 737 sizeof_field(struct qdisc_skb_cb, data)); 738 739 return qdisc_skb_cb(skb)->data; 740 } 741 742 /* Must be invoked with migration disabled */ 743 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 744 const void *ctx) 745 { 746 const struct sk_buff *skb = ctx; 747 u8 *cb_data = bpf_skb_cb(skb); 748 u8 cb_saved[BPF_SKB_CB_LEN]; 749 u32 res; 750 751 if (unlikely(prog->cb_access)) { 752 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 753 memset(cb_data, 0, sizeof(cb_saved)); 754 } 755 756 res = bpf_prog_run(prog, skb); 757 758 if (unlikely(prog->cb_access)) 759 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 760 761 return res; 762 } 763 764 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 765 struct sk_buff *skb) 766 { 767 u32 res; 768 769 migrate_disable(); 770 res = __bpf_prog_run_save_cb(prog, skb); 771 migrate_enable(); 772 return res; 773 } 774 775 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 776 struct sk_buff *skb) 777 { 778 u8 *cb_data = bpf_skb_cb(skb); 779 u32 res; 780 781 if (unlikely(prog->cb_access)) 782 memset(cb_data, 0, BPF_SKB_CB_LEN); 783 784 res = bpf_prog_run_pin_on_cpu(prog, skb); 785 return res; 786 } 787 788 DECLARE_BPF_DISPATCHER(xdp) 789 790 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 791 792 u32 xdp_master_redirect(struct xdp_buff *xdp); 793 794 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 795 struct xdp_buff *xdp) 796 { 797 /* Driver XDP hooks are invoked within a single NAPI poll cycle and thus 798 * under local_bh_disable(), which provides the needed RCU protection 799 * for accessing map entries. 800 */ 801 u32 act = __bpf_prog_run(prog, xdp, BPF_DISPATCHER_FUNC(xdp)); 802 803 if (static_branch_unlikely(&bpf_master_redirect_enabled_key)) { 804 if (act == XDP_TX && netif_is_bond_slave(xdp->rxq->dev)) 805 act = xdp_master_redirect(xdp); 806 } 807 808 return act; 809 } 810 811 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 812 813 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 814 { 815 return prog->len * sizeof(struct bpf_insn); 816 } 817 818 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 819 { 820 return round_up(bpf_prog_insn_size(prog) + 821 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 822 } 823 824 static inline unsigned int bpf_prog_size(unsigned int proglen) 825 { 826 return max(sizeof(struct bpf_prog), 827 offsetof(struct bpf_prog, insns[proglen])); 828 } 829 830 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 831 { 832 /* When classic BPF programs have been loaded and the arch 833 * does not have a classic BPF JIT (anymore), they have been 834 * converted via bpf_migrate_filter() to eBPF and thus always 835 * have an unspec program type. 836 */ 837 return prog->type == BPF_PROG_TYPE_UNSPEC; 838 } 839 840 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 841 { 842 const u32 size_machine = sizeof(unsigned long); 843 844 if (size > size_machine && size % size_machine == 0) 845 size = size_machine; 846 847 return size; 848 } 849 850 static inline bool 851 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 852 { 853 return size <= size_default && (size & (size - 1)) == 0; 854 } 855 856 static inline u8 857 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 858 { 859 u8 access_off = off & (size_default - 1); 860 861 #ifdef __LITTLE_ENDIAN 862 return access_off; 863 #else 864 return size_default - (access_off + size); 865 #endif 866 } 867 868 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 869 (size == sizeof(__u64) && \ 870 off >= offsetof(type, field) && \ 871 off + sizeof(__u64) <= offsetofend(type, field) && \ 872 off % sizeof(__u64) == 0) 873 874 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 875 876 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 877 { 878 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 879 if (!fp->jited) { 880 set_vm_flush_reset_perms(fp); 881 set_memory_ro((unsigned long)fp, fp->pages); 882 } 883 #endif 884 } 885 886 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 887 { 888 set_vm_flush_reset_perms(hdr); 889 set_memory_ro((unsigned long)hdr, hdr->pages); 890 set_memory_x((unsigned long)hdr, hdr->pages); 891 } 892 893 static inline struct bpf_binary_header * 894 bpf_jit_binary_hdr(const struct bpf_prog *fp) 895 { 896 unsigned long real_start = (unsigned long)fp->bpf_func; 897 unsigned long addr = real_start & PAGE_MASK; 898 899 return (void *)addr; 900 } 901 902 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 903 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 904 { 905 return sk_filter_trim_cap(sk, skb, 1); 906 } 907 908 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 909 void bpf_prog_free(struct bpf_prog *fp); 910 911 bool bpf_opcode_in_insntable(u8 code); 912 913 void bpf_prog_free_linfo(struct bpf_prog *prog); 914 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 915 const u32 *insn_to_jit_off); 916 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 917 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 918 919 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 920 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 921 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 922 gfp_t gfp_extra_flags); 923 void __bpf_prog_free(struct bpf_prog *fp); 924 925 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 926 { 927 __bpf_prog_free(fp); 928 } 929 930 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 931 unsigned int flen); 932 933 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 934 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 935 bpf_aux_classic_check_t trans, bool save_orig); 936 void bpf_prog_destroy(struct bpf_prog *fp); 937 938 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 939 int sk_attach_bpf(u32 ufd, struct sock *sk); 940 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 941 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 942 void sk_reuseport_prog_free(struct bpf_prog *prog); 943 int sk_detach_filter(struct sock *sk); 944 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, 945 unsigned int len); 946 947 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 948 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 949 950 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 951 #define __bpf_call_base_args \ 952 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 953 (void *)__bpf_call_base) 954 955 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 956 void bpf_jit_compile(struct bpf_prog *prog); 957 bool bpf_jit_needs_zext(void); 958 bool bpf_jit_supports_kfunc_call(void); 959 bool bpf_helper_changes_pkt_data(void *func); 960 961 static inline bool bpf_dump_raw_ok(const struct cred *cred) 962 { 963 /* Reconstruction of call-sites is dependent on kallsyms, 964 * thus make dump the same restriction. 965 */ 966 return kallsyms_show_value(cred); 967 } 968 969 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 970 const struct bpf_insn *patch, u32 len); 971 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 972 973 void bpf_clear_redirect_map(struct bpf_map *map); 974 975 static inline bool xdp_return_frame_no_direct(void) 976 { 977 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 978 979 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 980 } 981 982 static inline void xdp_set_return_frame_no_direct(void) 983 { 984 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 985 986 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 987 } 988 989 static inline void xdp_clear_return_frame_no_direct(void) 990 { 991 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 992 993 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 994 } 995 996 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 997 unsigned int pktlen) 998 { 999 unsigned int len; 1000 1001 if (unlikely(!(fwd->flags & IFF_UP))) 1002 return -ENETDOWN; 1003 1004 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 1005 if (pktlen > len) 1006 return -EMSGSIZE; 1007 1008 return 0; 1009 } 1010 1011 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 1012 * same cpu context. Further for best results no more than a single map 1013 * for the do_redirect/do_flush pair should be used. This limitation is 1014 * because we only track one map and force a flush when the map changes. 1015 * This does not appear to be a real limitation for existing software. 1016 */ 1017 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 1018 struct xdp_buff *xdp, struct bpf_prog *prog); 1019 int xdp_do_redirect(struct net_device *dev, 1020 struct xdp_buff *xdp, 1021 struct bpf_prog *prog); 1022 void xdp_do_flush(void); 1023 1024 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as 1025 * it is no longer only flushing maps. Keep this define for compatibility 1026 * until all drivers are updated - do not use xdp_do_flush_map() in new code! 1027 */ 1028 #define xdp_do_flush_map xdp_do_flush 1029 1030 void bpf_warn_invalid_xdp_action(u32 act); 1031 1032 #ifdef CONFIG_INET 1033 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1034 struct bpf_prog *prog, struct sk_buff *skb, 1035 struct sock *migrating_sk, 1036 u32 hash); 1037 #else 1038 static inline struct sock * 1039 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1040 struct bpf_prog *prog, struct sk_buff *skb, 1041 struct sock *migrating_sk, 1042 u32 hash) 1043 { 1044 return NULL; 1045 } 1046 #endif 1047 1048 #ifdef CONFIG_BPF_JIT 1049 extern int bpf_jit_enable; 1050 extern int bpf_jit_harden; 1051 extern int bpf_jit_kallsyms; 1052 extern long bpf_jit_limit; 1053 1054 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1055 1056 struct bpf_binary_header * 1057 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1058 unsigned int alignment, 1059 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1060 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1061 u64 bpf_jit_alloc_exec_limit(void); 1062 void *bpf_jit_alloc_exec(unsigned long size); 1063 void bpf_jit_free_exec(void *addr); 1064 void bpf_jit_free(struct bpf_prog *fp); 1065 1066 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1067 struct bpf_jit_poke_descriptor *poke); 1068 1069 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1070 const struct bpf_insn *insn, bool extra_pass, 1071 u64 *func_addr, bool *func_addr_fixed); 1072 1073 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1074 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1075 1076 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1077 u32 pass, void *image) 1078 { 1079 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1080 proglen, pass, image, current->comm, task_pid_nr(current)); 1081 1082 if (image) 1083 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1084 16, 1, image, proglen, false); 1085 } 1086 1087 static inline bool bpf_jit_is_ebpf(void) 1088 { 1089 # ifdef CONFIG_HAVE_EBPF_JIT 1090 return true; 1091 # else 1092 return false; 1093 # endif 1094 } 1095 1096 static inline bool ebpf_jit_enabled(void) 1097 { 1098 return bpf_jit_enable && bpf_jit_is_ebpf(); 1099 } 1100 1101 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1102 { 1103 return fp->jited && bpf_jit_is_ebpf(); 1104 } 1105 1106 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1107 { 1108 /* These are the prerequisites, should someone ever have the 1109 * idea to call blinding outside of them, we make sure to 1110 * bail out. 1111 */ 1112 if (!bpf_jit_is_ebpf()) 1113 return false; 1114 if (!prog->jit_requested) 1115 return false; 1116 if (!bpf_jit_harden) 1117 return false; 1118 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) 1119 return false; 1120 1121 return true; 1122 } 1123 1124 static inline bool bpf_jit_kallsyms_enabled(void) 1125 { 1126 /* There are a couple of corner cases where kallsyms should 1127 * not be enabled f.e. on hardening. 1128 */ 1129 if (bpf_jit_harden) 1130 return false; 1131 if (!bpf_jit_kallsyms) 1132 return false; 1133 if (bpf_jit_kallsyms == 1) 1134 return true; 1135 1136 return false; 1137 } 1138 1139 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1140 unsigned long *off, char *sym); 1141 bool is_bpf_text_address(unsigned long addr); 1142 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1143 char *sym); 1144 1145 static inline const char * 1146 bpf_address_lookup(unsigned long addr, unsigned long *size, 1147 unsigned long *off, char **modname, char *sym) 1148 { 1149 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1150 1151 if (ret && modname) 1152 *modname = NULL; 1153 return ret; 1154 } 1155 1156 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1157 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1158 1159 #else /* CONFIG_BPF_JIT */ 1160 1161 static inline bool ebpf_jit_enabled(void) 1162 { 1163 return false; 1164 } 1165 1166 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1167 { 1168 return false; 1169 } 1170 1171 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1172 { 1173 return false; 1174 } 1175 1176 static inline int 1177 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1178 struct bpf_jit_poke_descriptor *poke) 1179 { 1180 return -ENOTSUPP; 1181 } 1182 1183 static inline void bpf_jit_free(struct bpf_prog *fp) 1184 { 1185 bpf_prog_unlock_free(fp); 1186 } 1187 1188 static inline bool bpf_jit_kallsyms_enabled(void) 1189 { 1190 return false; 1191 } 1192 1193 static inline const char * 1194 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1195 unsigned long *off, char *sym) 1196 { 1197 return NULL; 1198 } 1199 1200 static inline bool is_bpf_text_address(unsigned long addr) 1201 { 1202 return false; 1203 } 1204 1205 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1206 char *type, char *sym) 1207 { 1208 return -ERANGE; 1209 } 1210 1211 static inline const char * 1212 bpf_address_lookup(unsigned long addr, unsigned long *size, 1213 unsigned long *off, char **modname, char *sym) 1214 { 1215 return NULL; 1216 } 1217 1218 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1219 { 1220 } 1221 1222 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1223 { 1224 } 1225 1226 #endif /* CONFIG_BPF_JIT */ 1227 1228 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1229 1230 #define BPF_ANC BIT(15) 1231 1232 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1233 { 1234 switch (first->code) { 1235 case BPF_RET | BPF_K: 1236 case BPF_LD | BPF_W | BPF_LEN: 1237 return false; 1238 1239 case BPF_LD | BPF_W | BPF_ABS: 1240 case BPF_LD | BPF_H | BPF_ABS: 1241 case BPF_LD | BPF_B | BPF_ABS: 1242 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1243 return true; 1244 return false; 1245 1246 default: 1247 return true; 1248 } 1249 } 1250 1251 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1252 { 1253 BUG_ON(ftest->code & BPF_ANC); 1254 1255 switch (ftest->code) { 1256 case BPF_LD | BPF_W | BPF_ABS: 1257 case BPF_LD | BPF_H | BPF_ABS: 1258 case BPF_LD | BPF_B | BPF_ABS: 1259 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1260 return BPF_ANC | SKF_AD_##CODE 1261 switch (ftest->k) { 1262 BPF_ANCILLARY(PROTOCOL); 1263 BPF_ANCILLARY(PKTTYPE); 1264 BPF_ANCILLARY(IFINDEX); 1265 BPF_ANCILLARY(NLATTR); 1266 BPF_ANCILLARY(NLATTR_NEST); 1267 BPF_ANCILLARY(MARK); 1268 BPF_ANCILLARY(QUEUE); 1269 BPF_ANCILLARY(HATYPE); 1270 BPF_ANCILLARY(RXHASH); 1271 BPF_ANCILLARY(CPU); 1272 BPF_ANCILLARY(ALU_XOR_X); 1273 BPF_ANCILLARY(VLAN_TAG); 1274 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1275 BPF_ANCILLARY(PAY_OFFSET); 1276 BPF_ANCILLARY(RANDOM); 1277 BPF_ANCILLARY(VLAN_TPID); 1278 } 1279 fallthrough; 1280 default: 1281 return ftest->code; 1282 } 1283 } 1284 1285 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1286 int k, unsigned int size); 1287 1288 static inline int bpf_tell_extensions(void) 1289 { 1290 return SKF_AD_MAX; 1291 } 1292 1293 struct bpf_sock_addr_kern { 1294 struct sock *sk; 1295 struct sockaddr *uaddr; 1296 /* Temporary "register" to make indirect stores to nested structures 1297 * defined above. We need three registers to make such a store, but 1298 * only two (src and dst) are available at convert_ctx_access time 1299 */ 1300 u64 tmp_reg; 1301 void *t_ctx; /* Attach type specific context. */ 1302 }; 1303 1304 struct bpf_sock_ops_kern { 1305 struct sock *sk; 1306 union { 1307 u32 args[4]; 1308 u32 reply; 1309 u32 replylong[4]; 1310 }; 1311 struct sk_buff *syn_skb; 1312 struct sk_buff *skb; 1313 void *skb_data_end; 1314 u8 op; 1315 u8 is_fullsock; 1316 u8 remaining_opt_len; 1317 u64 temp; /* temp and everything after is not 1318 * initialized to 0 before calling 1319 * the BPF program. New fields that 1320 * should be initialized to 0 should 1321 * be inserted before temp. 1322 * temp is scratch storage used by 1323 * sock_ops_convert_ctx_access 1324 * as temporary storage of a register. 1325 */ 1326 }; 1327 1328 struct bpf_sysctl_kern { 1329 struct ctl_table_header *head; 1330 struct ctl_table *table; 1331 void *cur_val; 1332 size_t cur_len; 1333 void *new_val; 1334 size_t new_len; 1335 int new_updated; 1336 int write; 1337 loff_t *ppos; 1338 /* Temporary "register" for indirect stores to ppos. */ 1339 u64 tmp_reg; 1340 }; 1341 1342 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1343 struct bpf_sockopt_buf { 1344 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1345 }; 1346 1347 struct bpf_sockopt_kern { 1348 struct sock *sk; 1349 u8 *optval; 1350 u8 *optval_end; 1351 s32 level; 1352 s32 optname; 1353 s32 optlen; 1354 s32 retval; 1355 }; 1356 1357 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1358 1359 struct bpf_sk_lookup_kern { 1360 u16 family; 1361 u16 protocol; 1362 __be16 sport; 1363 u16 dport; 1364 struct { 1365 __be32 saddr; 1366 __be32 daddr; 1367 } v4; 1368 struct { 1369 const struct in6_addr *saddr; 1370 const struct in6_addr *daddr; 1371 } v6; 1372 struct sock *selected_sk; 1373 bool no_reuseport; 1374 }; 1375 1376 extern struct static_key_false bpf_sk_lookup_enabled; 1377 1378 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1379 * 1380 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1381 * SK_DROP. Their meaning is as follows: 1382 * 1383 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1384 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1385 * SK_DROP : terminate lookup with -ECONNREFUSED 1386 * 1387 * This macro aggregates return values and selected sockets from 1388 * multiple BPF programs according to following rules in order: 1389 * 1390 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1391 * macro result is SK_PASS and last ctx.selected_sk is used. 1392 * 2. If any program returned SK_DROP return value, 1393 * macro result is SK_DROP. 1394 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1395 * 1396 * Caller must ensure that the prog array is non-NULL, and that the 1397 * array as well as the programs it contains remain valid. 1398 */ 1399 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1400 ({ \ 1401 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1402 struct bpf_prog_array_item *_item; \ 1403 struct sock *_selected_sk = NULL; \ 1404 bool _no_reuseport = false; \ 1405 struct bpf_prog *_prog; \ 1406 bool _all_pass = true; \ 1407 u32 _ret; \ 1408 \ 1409 migrate_disable(); \ 1410 _item = &(array)->items[0]; \ 1411 while ((_prog = READ_ONCE(_item->prog))) { \ 1412 /* restore most recent selection */ \ 1413 _ctx->selected_sk = _selected_sk; \ 1414 _ctx->no_reuseport = _no_reuseport; \ 1415 \ 1416 _ret = func(_prog, _ctx); \ 1417 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1418 /* remember last non-NULL socket */ \ 1419 _selected_sk = _ctx->selected_sk; \ 1420 _no_reuseport = _ctx->no_reuseport; \ 1421 } else if (_ret == SK_DROP && _all_pass) { \ 1422 _all_pass = false; \ 1423 } \ 1424 _item++; \ 1425 } \ 1426 _ctx->selected_sk = _selected_sk; \ 1427 _ctx->no_reuseport = _no_reuseport; \ 1428 migrate_enable(); \ 1429 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1430 }) 1431 1432 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1433 const __be32 saddr, const __be16 sport, 1434 const __be32 daddr, const u16 dport, 1435 struct sock **psk) 1436 { 1437 struct bpf_prog_array *run_array; 1438 struct sock *selected_sk = NULL; 1439 bool no_reuseport = false; 1440 1441 rcu_read_lock(); 1442 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1443 if (run_array) { 1444 struct bpf_sk_lookup_kern ctx = { 1445 .family = AF_INET, 1446 .protocol = protocol, 1447 .v4.saddr = saddr, 1448 .v4.daddr = daddr, 1449 .sport = sport, 1450 .dport = dport, 1451 }; 1452 u32 act; 1453 1454 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1455 if (act == SK_PASS) { 1456 selected_sk = ctx.selected_sk; 1457 no_reuseport = ctx.no_reuseport; 1458 } else { 1459 selected_sk = ERR_PTR(-ECONNREFUSED); 1460 } 1461 } 1462 rcu_read_unlock(); 1463 *psk = selected_sk; 1464 return no_reuseport; 1465 } 1466 1467 #if IS_ENABLED(CONFIG_IPV6) 1468 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1469 const struct in6_addr *saddr, 1470 const __be16 sport, 1471 const struct in6_addr *daddr, 1472 const u16 dport, 1473 struct sock **psk) 1474 { 1475 struct bpf_prog_array *run_array; 1476 struct sock *selected_sk = NULL; 1477 bool no_reuseport = false; 1478 1479 rcu_read_lock(); 1480 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1481 if (run_array) { 1482 struct bpf_sk_lookup_kern ctx = { 1483 .family = AF_INET6, 1484 .protocol = protocol, 1485 .v6.saddr = saddr, 1486 .v6.daddr = daddr, 1487 .sport = sport, 1488 .dport = dport, 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 int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex, 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, ifindex); 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 = ifindex; 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 #endif /* __LINUX_FILTER_H__ */ 1545