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 /* Function call */ 364 365 #define BPF_CAST_CALL(x) \ 366 ((u64 (*)(u64, u64, u64, u64, u64))(x)) 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 = ((FUNC) - __bpf_call_base) }) 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 pages; 553 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 554 }; 555 556 struct bpf_prog_stats { 557 u64 cnt; 558 u64 nsecs; 559 u64 misses; 560 struct u64_stats_sync syncp; 561 } __aligned(2 * sizeof(u64)); 562 563 struct bpf_prog { 564 u16 pages; /* Number of allocated pages */ 565 u16 jited:1, /* Is our filter JIT'ed? */ 566 jit_requested:1,/* archs need to JIT the prog */ 567 gpl_compatible:1, /* Is filter GPL compatible? */ 568 cb_access:1, /* Is control block accessed? */ 569 dst_needed:1, /* Do we need dst entry? */ 570 blinded:1, /* Was blinded */ 571 is_func:1, /* program is a bpf function */ 572 kprobe_override:1, /* Do we override a kprobe? */ 573 has_callchain_buf:1, /* callchain buffer allocated? */ 574 enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */ 575 call_get_stack:1; /* Do we call bpf_get_stack() or bpf_get_stackid() */ 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 #define __BPF_PROG_RUN(prog, ctx, dfunc) ({ \ 601 u32 __ret; \ 602 cant_migrate(); \ 603 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \ 604 struct bpf_prog_stats *__stats; \ 605 u64 __start = sched_clock(); \ 606 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \ 607 __stats = this_cpu_ptr(prog->stats); \ 608 u64_stats_update_begin(&__stats->syncp); \ 609 __stats->cnt++; \ 610 __stats->nsecs += sched_clock() - __start; \ 611 u64_stats_update_end(&__stats->syncp); \ 612 } else { \ 613 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \ 614 } \ 615 __ret; }) 616 617 #define BPF_PROG_RUN(prog, ctx) \ 618 __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func) 619 620 /* 621 * Use in preemptible and therefore migratable context to make sure that 622 * the execution of the BPF program runs on one CPU. 623 * 624 * This uses migrate_disable/enable() explicitly to document that the 625 * invocation of a BPF program does not require reentrancy protection 626 * against a BPF program which is invoked from a preempting task. 627 * 628 * For non RT enabled kernels migrate_disable/enable() maps to 629 * preempt_disable/enable(), i.e. it disables also preemption. 630 */ 631 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 632 const void *ctx) 633 { 634 u32 ret; 635 636 migrate_disable(); 637 ret = __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func); 638 migrate_enable(); 639 return ret; 640 } 641 642 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 643 644 struct bpf_skb_data_end { 645 struct qdisc_skb_cb qdisc_cb; 646 void *data_meta; 647 void *data_end; 648 }; 649 650 struct bpf_nh_params { 651 u32 nh_family; 652 union { 653 u32 ipv4_nh; 654 struct in6_addr ipv6_nh; 655 }; 656 }; 657 658 struct bpf_redirect_info { 659 u32 flags; 660 u32 tgt_index; 661 void *tgt_value; 662 struct bpf_map *map; 663 u32 map_id; 664 enum bpf_map_type map_type; 665 u32 kern_flags; 666 struct bpf_nh_params nh; 667 }; 668 669 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 670 671 /* flags for bpf_redirect_info kern_flags */ 672 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 673 674 /* Compute the linear packet data range [data, data_end) which 675 * will be accessed by various program types (cls_bpf, act_bpf, 676 * lwt, ...). Subsystems allowing direct data access must (!) 677 * ensure that cb[] area can be written to when BPF program is 678 * invoked (otherwise cb[] save/restore is necessary). 679 */ 680 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 681 { 682 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 683 684 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 685 cb->data_meta = skb->data - skb_metadata_len(skb); 686 cb->data_end = skb->data + skb_headlen(skb); 687 } 688 689 /* Similar to bpf_compute_data_pointers(), except that save orginal 690 * data in cb->data and cb->meta_data for restore. 691 */ 692 static inline void bpf_compute_and_save_data_end( 693 struct sk_buff *skb, void **saved_data_end) 694 { 695 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 696 697 *saved_data_end = cb->data_end; 698 cb->data_end = skb->data + skb_headlen(skb); 699 } 700 701 /* Restore data saved by bpf_compute_data_pointers(). */ 702 static inline void bpf_restore_data_end( 703 struct sk_buff *skb, void *saved_data_end) 704 { 705 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 706 707 cb->data_end = saved_data_end; 708 } 709 710 static inline u8 *bpf_skb_cb(struct sk_buff *skb) 711 { 712 /* eBPF programs may read/write skb->cb[] area to transfer meta 713 * data between tail calls. Since this also needs to work with 714 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 715 * 716 * In some socket filter cases, the cb unfortunately needs to be 717 * saved/restored so that protocol specific skb->cb[] data won't 718 * be lost. In any case, due to unpriviledged eBPF programs 719 * attached to sockets, we need to clear the bpf_skb_cb() area 720 * to not leak previous contents to user space. 721 */ 722 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 723 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 724 sizeof_field(struct qdisc_skb_cb, data)); 725 726 return qdisc_skb_cb(skb)->data; 727 } 728 729 /* Must be invoked with migration disabled */ 730 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 731 struct sk_buff *skb) 732 { 733 u8 *cb_data = bpf_skb_cb(skb); 734 u8 cb_saved[BPF_SKB_CB_LEN]; 735 u32 res; 736 737 if (unlikely(prog->cb_access)) { 738 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 739 memset(cb_data, 0, sizeof(cb_saved)); 740 } 741 742 res = BPF_PROG_RUN(prog, skb); 743 744 if (unlikely(prog->cb_access)) 745 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 746 747 return res; 748 } 749 750 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 751 struct sk_buff *skb) 752 { 753 u32 res; 754 755 migrate_disable(); 756 res = __bpf_prog_run_save_cb(prog, skb); 757 migrate_enable(); 758 return res; 759 } 760 761 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 762 struct sk_buff *skb) 763 { 764 u8 *cb_data = bpf_skb_cb(skb); 765 u32 res; 766 767 if (unlikely(prog->cb_access)) 768 memset(cb_data, 0, BPF_SKB_CB_LEN); 769 770 res = bpf_prog_run_pin_on_cpu(prog, skb); 771 return res; 772 } 773 774 DECLARE_BPF_DISPATCHER(xdp) 775 776 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 777 struct xdp_buff *xdp) 778 { 779 /* Driver XDP hooks are invoked within a single NAPI poll cycle and thus 780 * under local_bh_disable(), which provides the needed RCU protection 781 * for accessing map entries. 782 */ 783 return __BPF_PROG_RUN(prog, xdp, BPF_DISPATCHER_FUNC(xdp)); 784 } 785 786 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 787 788 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 789 { 790 return prog->len * sizeof(struct bpf_insn); 791 } 792 793 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 794 { 795 return round_up(bpf_prog_insn_size(prog) + 796 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 797 } 798 799 static inline unsigned int bpf_prog_size(unsigned int proglen) 800 { 801 return max(sizeof(struct bpf_prog), 802 offsetof(struct bpf_prog, insns[proglen])); 803 } 804 805 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 806 { 807 /* When classic BPF programs have been loaded and the arch 808 * does not have a classic BPF JIT (anymore), they have been 809 * converted via bpf_migrate_filter() to eBPF and thus always 810 * have an unspec program type. 811 */ 812 return prog->type == BPF_PROG_TYPE_UNSPEC; 813 } 814 815 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 816 { 817 const u32 size_machine = sizeof(unsigned long); 818 819 if (size > size_machine && size % size_machine == 0) 820 size = size_machine; 821 822 return size; 823 } 824 825 static inline bool 826 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 827 { 828 return size <= size_default && (size & (size - 1)) == 0; 829 } 830 831 static inline u8 832 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 833 { 834 u8 access_off = off & (size_default - 1); 835 836 #ifdef __LITTLE_ENDIAN 837 return access_off; 838 #else 839 return size_default - (access_off + size); 840 #endif 841 } 842 843 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 844 (size == sizeof(__u64) && \ 845 off >= offsetof(type, field) && \ 846 off + sizeof(__u64) <= offsetofend(type, field) && \ 847 off % sizeof(__u64) == 0) 848 849 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 850 851 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 852 { 853 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 854 if (!fp->jited) { 855 set_vm_flush_reset_perms(fp); 856 set_memory_ro((unsigned long)fp, fp->pages); 857 } 858 #endif 859 } 860 861 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 862 { 863 set_vm_flush_reset_perms(hdr); 864 set_memory_ro((unsigned long)hdr, hdr->pages); 865 set_memory_x((unsigned long)hdr, hdr->pages); 866 } 867 868 static inline struct bpf_binary_header * 869 bpf_jit_binary_hdr(const struct bpf_prog *fp) 870 { 871 unsigned long real_start = (unsigned long)fp->bpf_func; 872 unsigned long addr = real_start & PAGE_MASK; 873 874 return (void *)addr; 875 } 876 877 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 878 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 879 { 880 return sk_filter_trim_cap(sk, skb, 1); 881 } 882 883 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 884 void bpf_prog_free(struct bpf_prog *fp); 885 886 bool bpf_opcode_in_insntable(u8 code); 887 888 void bpf_prog_free_linfo(struct bpf_prog *prog); 889 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 890 const u32 *insn_to_jit_off); 891 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 892 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 893 894 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 895 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 896 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 897 gfp_t gfp_extra_flags); 898 void __bpf_prog_free(struct bpf_prog *fp); 899 900 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 901 { 902 __bpf_prog_free(fp); 903 } 904 905 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 906 unsigned int flen); 907 908 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 909 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 910 bpf_aux_classic_check_t trans, bool save_orig); 911 void bpf_prog_destroy(struct bpf_prog *fp); 912 913 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 914 int sk_attach_bpf(u32 ufd, struct sock *sk); 915 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 916 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 917 void sk_reuseport_prog_free(struct bpf_prog *prog); 918 int sk_detach_filter(struct sock *sk); 919 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, 920 unsigned int len); 921 922 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 923 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 924 925 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 926 #define __bpf_call_base_args \ 927 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 928 (void *)__bpf_call_base) 929 930 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 931 void bpf_jit_compile(struct bpf_prog *prog); 932 bool bpf_jit_needs_zext(void); 933 bool bpf_jit_supports_kfunc_call(void); 934 bool bpf_helper_changes_pkt_data(void *func); 935 936 static inline bool bpf_dump_raw_ok(const struct cred *cred) 937 { 938 /* Reconstruction of call-sites is dependent on kallsyms, 939 * thus make dump the same restriction. 940 */ 941 return kallsyms_show_value(cred); 942 } 943 944 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 945 const struct bpf_insn *patch, u32 len); 946 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 947 948 void bpf_clear_redirect_map(struct bpf_map *map); 949 950 static inline bool xdp_return_frame_no_direct(void) 951 { 952 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 953 954 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 955 } 956 957 static inline void xdp_set_return_frame_no_direct(void) 958 { 959 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 960 961 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 962 } 963 964 static inline void xdp_clear_return_frame_no_direct(void) 965 { 966 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 967 968 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 969 } 970 971 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 972 unsigned int pktlen) 973 { 974 unsigned int len; 975 976 if (unlikely(!(fwd->flags & IFF_UP))) 977 return -ENETDOWN; 978 979 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 980 if (pktlen > len) 981 return -EMSGSIZE; 982 983 return 0; 984 } 985 986 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 987 * same cpu context. Further for best results no more than a single map 988 * for the do_redirect/do_flush pair should be used. This limitation is 989 * because we only track one map and force a flush when the map changes. 990 * This does not appear to be a real limitation for existing software. 991 */ 992 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 993 struct xdp_buff *xdp, struct bpf_prog *prog); 994 int xdp_do_redirect(struct net_device *dev, 995 struct xdp_buff *xdp, 996 struct bpf_prog *prog); 997 void xdp_do_flush(void); 998 999 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as 1000 * it is no longer only flushing maps. Keep this define for compatibility 1001 * until all drivers are updated - do not use xdp_do_flush_map() in new code! 1002 */ 1003 #define xdp_do_flush_map xdp_do_flush 1004 1005 void bpf_warn_invalid_xdp_action(u32 act); 1006 1007 #ifdef CONFIG_INET 1008 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1009 struct bpf_prog *prog, struct sk_buff *skb, 1010 struct sock *migrating_sk, 1011 u32 hash); 1012 #else 1013 static inline struct sock * 1014 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1015 struct bpf_prog *prog, struct sk_buff *skb, 1016 struct sock *migrating_sk, 1017 u32 hash) 1018 { 1019 return NULL; 1020 } 1021 #endif 1022 1023 #ifdef CONFIG_BPF_JIT 1024 extern int bpf_jit_enable; 1025 extern int bpf_jit_harden; 1026 extern int bpf_jit_kallsyms; 1027 extern long bpf_jit_limit; 1028 1029 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1030 1031 struct bpf_binary_header * 1032 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1033 unsigned int alignment, 1034 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1035 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1036 u64 bpf_jit_alloc_exec_limit(void); 1037 void *bpf_jit_alloc_exec(unsigned long size); 1038 void bpf_jit_free_exec(void *addr); 1039 void bpf_jit_free(struct bpf_prog *fp); 1040 1041 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1042 struct bpf_jit_poke_descriptor *poke); 1043 1044 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1045 const struct bpf_insn *insn, bool extra_pass, 1046 u64 *func_addr, bool *func_addr_fixed); 1047 1048 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1049 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1050 1051 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1052 u32 pass, void *image) 1053 { 1054 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1055 proglen, pass, image, current->comm, task_pid_nr(current)); 1056 1057 if (image) 1058 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1059 16, 1, image, proglen, false); 1060 } 1061 1062 static inline bool bpf_jit_is_ebpf(void) 1063 { 1064 # ifdef CONFIG_HAVE_EBPF_JIT 1065 return true; 1066 # else 1067 return false; 1068 # endif 1069 } 1070 1071 static inline bool ebpf_jit_enabled(void) 1072 { 1073 return bpf_jit_enable && bpf_jit_is_ebpf(); 1074 } 1075 1076 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1077 { 1078 return fp->jited && bpf_jit_is_ebpf(); 1079 } 1080 1081 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1082 { 1083 /* These are the prerequisites, should someone ever have the 1084 * idea to call blinding outside of them, we make sure to 1085 * bail out. 1086 */ 1087 if (!bpf_jit_is_ebpf()) 1088 return false; 1089 if (!prog->jit_requested) 1090 return false; 1091 if (!bpf_jit_harden) 1092 return false; 1093 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) 1094 return false; 1095 1096 return true; 1097 } 1098 1099 static inline bool bpf_jit_kallsyms_enabled(void) 1100 { 1101 /* There are a couple of corner cases where kallsyms should 1102 * not be enabled f.e. on hardening. 1103 */ 1104 if (bpf_jit_harden) 1105 return false; 1106 if (!bpf_jit_kallsyms) 1107 return false; 1108 if (bpf_jit_kallsyms == 1) 1109 return true; 1110 1111 return false; 1112 } 1113 1114 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1115 unsigned long *off, char *sym); 1116 bool is_bpf_text_address(unsigned long addr); 1117 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1118 char *sym); 1119 1120 static inline const char * 1121 bpf_address_lookup(unsigned long addr, unsigned long *size, 1122 unsigned long *off, char **modname, char *sym) 1123 { 1124 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1125 1126 if (ret && modname) 1127 *modname = NULL; 1128 return ret; 1129 } 1130 1131 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1132 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1133 1134 #else /* CONFIG_BPF_JIT */ 1135 1136 static inline bool ebpf_jit_enabled(void) 1137 { 1138 return false; 1139 } 1140 1141 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1142 { 1143 return false; 1144 } 1145 1146 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1147 { 1148 return false; 1149 } 1150 1151 static inline int 1152 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1153 struct bpf_jit_poke_descriptor *poke) 1154 { 1155 return -ENOTSUPP; 1156 } 1157 1158 static inline void bpf_jit_free(struct bpf_prog *fp) 1159 { 1160 bpf_prog_unlock_free(fp); 1161 } 1162 1163 static inline bool bpf_jit_kallsyms_enabled(void) 1164 { 1165 return false; 1166 } 1167 1168 static inline const char * 1169 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1170 unsigned long *off, char *sym) 1171 { 1172 return NULL; 1173 } 1174 1175 static inline bool is_bpf_text_address(unsigned long addr) 1176 { 1177 return false; 1178 } 1179 1180 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1181 char *type, char *sym) 1182 { 1183 return -ERANGE; 1184 } 1185 1186 static inline const char * 1187 bpf_address_lookup(unsigned long addr, unsigned long *size, 1188 unsigned long *off, char **modname, char *sym) 1189 { 1190 return NULL; 1191 } 1192 1193 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1194 { 1195 } 1196 1197 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1198 { 1199 } 1200 1201 #endif /* CONFIG_BPF_JIT */ 1202 1203 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1204 1205 #define BPF_ANC BIT(15) 1206 1207 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1208 { 1209 switch (first->code) { 1210 case BPF_RET | BPF_K: 1211 case BPF_LD | BPF_W | BPF_LEN: 1212 return false; 1213 1214 case BPF_LD | BPF_W | BPF_ABS: 1215 case BPF_LD | BPF_H | BPF_ABS: 1216 case BPF_LD | BPF_B | BPF_ABS: 1217 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1218 return true; 1219 return false; 1220 1221 default: 1222 return true; 1223 } 1224 } 1225 1226 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1227 { 1228 BUG_ON(ftest->code & BPF_ANC); 1229 1230 switch (ftest->code) { 1231 case BPF_LD | BPF_W | BPF_ABS: 1232 case BPF_LD | BPF_H | BPF_ABS: 1233 case BPF_LD | BPF_B | BPF_ABS: 1234 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1235 return BPF_ANC | SKF_AD_##CODE 1236 switch (ftest->k) { 1237 BPF_ANCILLARY(PROTOCOL); 1238 BPF_ANCILLARY(PKTTYPE); 1239 BPF_ANCILLARY(IFINDEX); 1240 BPF_ANCILLARY(NLATTR); 1241 BPF_ANCILLARY(NLATTR_NEST); 1242 BPF_ANCILLARY(MARK); 1243 BPF_ANCILLARY(QUEUE); 1244 BPF_ANCILLARY(HATYPE); 1245 BPF_ANCILLARY(RXHASH); 1246 BPF_ANCILLARY(CPU); 1247 BPF_ANCILLARY(ALU_XOR_X); 1248 BPF_ANCILLARY(VLAN_TAG); 1249 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1250 BPF_ANCILLARY(PAY_OFFSET); 1251 BPF_ANCILLARY(RANDOM); 1252 BPF_ANCILLARY(VLAN_TPID); 1253 } 1254 fallthrough; 1255 default: 1256 return ftest->code; 1257 } 1258 } 1259 1260 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1261 int k, unsigned int size); 1262 1263 static inline int bpf_tell_extensions(void) 1264 { 1265 return SKF_AD_MAX; 1266 } 1267 1268 struct bpf_sock_addr_kern { 1269 struct sock *sk; 1270 struct sockaddr *uaddr; 1271 /* Temporary "register" to make indirect stores to nested structures 1272 * defined above. We need three registers to make such a store, but 1273 * only two (src and dst) are available at convert_ctx_access time 1274 */ 1275 u64 tmp_reg; 1276 void *t_ctx; /* Attach type specific context. */ 1277 }; 1278 1279 struct bpf_sock_ops_kern { 1280 struct sock *sk; 1281 union { 1282 u32 args[4]; 1283 u32 reply; 1284 u32 replylong[4]; 1285 }; 1286 struct sk_buff *syn_skb; 1287 struct sk_buff *skb; 1288 void *skb_data_end; 1289 u8 op; 1290 u8 is_fullsock; 1291 u8 remaining_opt_len; 1292 u64 temp; /* temp and everything after is not 1293 * initialized to 0 before calling 1294 * the BPF program. New fields that 1295 * should be initialized to 0 should 1296 * be inserted before temp. 1297 * temp is scratch storage used by 1298 * sock_ops_convert_ctx_access 1299 * as temporary storage of a register. 1300 */ 1301 }; 1302 1303 struct bpf_sysctl_kern { 1304 struct ctl_table_header *head; 1305 struct ctl_table *table; 1306 void *cur_val; 1307 size_t cur_len; 1308 void *new_val; 1309 size_t new_len; 1310 int new_updated; 1311 int write; 1312 loff_t *ppos; 1313 /* Temporary "register" for indirect stores to ppos. */ 1314 u64 tmp_reg; 1315 }; 1316 1317 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1318 struct bpf_sockopt_buf { 1319 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1320 }; 1321 1322 struct bpf_sockopt_kern { 1323 struct sock *sk; 1324 u8 *optval; 1325 u8 *optval_end; 1326 s32 level; 1327 s32 optname; 1328 s32 optlen; 1329 s32 retval; 1330 }; 1331 1332 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1333 1334 struct bpf_sk_lookup_kern { 1335 u16 family; 1336 u16 protocol; 1337 __be16 sport; 1338 u16 dport; 1339 struct { 1340 __be32 saddr; 1341 __be32 daddr; 1342 } v4; 1343 struct { 1344 const struct in6_addr *saddr; 1345 const struct in6_addr *daddr; 1346 } v6; 1347 struct sock *selected_sk; 1348 bool no_reuseport; 1349 }; 1350 1351 extern struct static_key_false bpf_sk_lookup_enabled; 1352 1353 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1354 * 1355 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1356 * SK_DROP. Their meaning is as follows: 1357 * 1358 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1359 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1360 * SK_DROP : terminate lookup with -ECONNREFUSED 1361 * 1362 * This macro aggregates return values and selected sockets from 1363 * multiple BPF programs according to following rules in order: 1364 * 1365 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1366 * macro result is SK_PASS and last ctx.selected_sk is used. 1367 * 2. If any program returned SK_DROP return value, 1368 * macro result is SK_DROP. 1369 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1370 * 1371 * Caller must ensure that the prog array is non-NULL, and that the 1372 * array as well as the programs it contains remain valid. 1373 */ 1374 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1375 ({ \ 1376 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1377 struct bpf_prog_array_item *_item; \ 1378 struct sock *_selected_sk = NULL; \ 1379 bool _no_reuseport = false; \ 1380 struct bpf_prog *_prog; \ 1381 bool _all_pass = true; \ 1382 u32 _ret; \ 1383 \ 1384 migrate_disable(); \ 1385 _item = &(array)->items[0]; \ 1386 while ((_prog = READ_ONCE(_item->prog))) { \ 1387 /* restore most recent selection */ \ 1388 _ctx->selected_sk = _selected_sk; \ 1389 _ctx->no_reuseport = _no_reuseport; \ 1390 \ 1391 _ret = func(_prog, _ctx); \ 1392 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1393 /* remember last non-NULL socket */ \ 1394 _selected_sk = _ctx->selected_sk; \ 1395 _no_reuseport = _ctx->no_reuseport; \ 1396 } else if (_ret == SK_DROP && _all_pass) { \ 1397 _all_pass = false; \ 1398 } \ 1399 _item++; \ 1400 } \ 1401 _ctx->selected_sk = _selected_sk; \ 1402 _ctx->no_reuseport = _no_reuseport; \ 1403 migrate_enable(); \ 1404 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1405 }) 1406 1407 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1408 const __be32 saddr, const __be16 sport, 1409 const __be32 daddr, const u16 dport, 1410 struct sock **psk) 1411 { 1412 struct bpf_prog_array *run_array; 1413 struct sock *selected_sk = NULL; 1414 bool no_reuseport = false; 1415 1416 rcu_read_lock(); 1417 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1418 if (run_array) { 1419 struct bpf_sk_lookup_kern ctx = { 1420 .family = AF_INET, 1421 .protocol = protocol, 1422 .v4.saddr = saddr, 1423 .v4.daddr = daddr, 1424 .sport = sport, 1425 .dport = dport, 1426 }; 1427 u32 act; 1428 1429 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN); 1430 if (act == SK_PASS) { 1431 selected_sk = ctx.selected_sk; 1432 no_reuseport = ctx.no_reuseport; 1433 } else { 1434 selected_sk = ERR_PTR(-ECONNREFUSED); 1435 } 1436 } 1437 rcu_read_unlock(); 1438 *psk = selected_sk; 1439 return no_reuseport; 1440 } 1441 1442 #if IS_ENABLED(CONFIG_IPV6) 1443 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1444 const struct in6_addr *saddr, 1445 const __be16 sport, 1446 const struct in6_addr *daddr, 1447 const u16 dport, 1448 struct sock **psk) 1449 { 1450 struct bpf_prog_array *run_array; 1451 struct sock *selected_sk = NULL; 1452 bool no_reuseport = false; 1453 1454 rcu_read_lock(); 1455 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1456 if (run_array) { 1457 struct bpf_sk_lookup_kern ctx = { 1458 .family = AF_INET6, 1459 .protocol = protocol, 1460 .v6.saddr = saddr, 1461 .v6.daddr = daddr, 1462 .sport = sport, 1463 .dport = dport, 1464 }; 1465 u32 act; 1466 1467 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN); 1468 if (act == SK_PASS) { 1469 selected_sk = ctx.selected_sk; 1470 no_reuseport = ctx.no_reuseport; 1471 } else { 1472 selected_sk = ERR_PTR(-ECONNREFUSED); 1473 } 1474 } 1475 rcu_read_unlock(); 1476 *psk = selected_sk; 1477 return no_reuseport; 1478 } 1479 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1480 1481 static __always_inline int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex, 1482 u64 flags, const u64 flag_mask, 1483 void *lookup_elem(struct bpf_map *map, u32 key)) 1484 { 1485 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1486 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1487 1488 /* Lower bits of the flags are used as return code on lookup failure */ 1489 if (unlikely(flags & ~(action_mask | flag_mask))) 1490 return XDP_ABORTED; 1491 1492 ri->tgt_value = lookup_elem(map, ifindex); 1493 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1494 /* If the lookup fails we want to clear out the state in the 1495 * redirect_info struct completely, so that if an eBPF program 1496 * performs multiple lookups, the last one always takes 1497 * precedence. 1498 */ 1499 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1500 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1501 return flags & action_mask; 1502 } 1503 1504 ri->tgt_index = ifindex; 1505 ri->map_id = map->id; 1506 ri->map_type = map->map_type; 1507 1508 if (flags & BPF_F_BROADCAST) { 1509 WRITE_ONCE(ri->map, map); 1510 ri->flags = flags; 1511 } else { 1512 WRITE_ONCE(ri->map, NULL); 1513 ri->flags = 0; 1514 } 1515 1516 return XDP_REDIRECT; 1517 } 1518 1519 #endif /* __LINUX_FILTER_H__ */ 1520