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 special ldsx instruction. Same as BPF_IND */ 73 #define BPF_PROBE_MEMSX 0x40 74 75 /* unused opcode to mark call to interpreter with arguments */ 76 #define BPF_CALL_ARGS 0xe0 77 78 /* unused opcode to mark speculation barrier for mitigating 79 * Speculative Store Bypass 80 */ 81 #define BPF_NOSPEC 0xc0 82 83 /* As per nm, we expose JITed images as text (code) section for 84 * kallsyms. That way, tools like perf can find it to match 85 * addresses. 86 */ 87 #define BPF_SYM_ELF_TYPE 't' 88 89 /* BPF program can access up to 512 bytes of stack space. */ 90 #define MAX_BPF_STACK 512 91 92 /* Helper macros for filter block array initializers. */ 93 94 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 95 96 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \ 97 ((struct bpf_insn) { \ 98 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 99 .dst_reg = DST, \ 100 .src_reg = SRC, \ 101 .off = OFF, \ 102 .imm = 0 }) 103 104 #define BPF_ALU64_REG(OP, DST, SRC) \ 105 BPF_ALU64_REG_OFF(OP, DST, SRC, 0) 106 107 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \ 108 ((struct bpf_insn) { \ 109 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 110 .dst_reg = DST, \ 111 .src_reg = SRC, \ 112 .off = OFF, \ 113 .imm = 0 }) 114 115 #define BPF_ALU32_REG(OP, DST, SRC) \ 116 BPF_ALU32_REG_OFF(OP, DST, SRC, 0) 117 118 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 119 120 #define BPF_ALU64_IMM(OP, DST, IMM) \ 121 ((struct bpf_insn) { \ 122 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 123 .dst_reg = DST, \ 124 .src_reg = 0, \ 125 .off = 0, \ 126 .imm = IMM }) 127 128 #define BPF_ALU32_IMM(OP, DST, IMM) \ 129 ((struct bpf_insn) { \ 130 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 131 .dst_reg = DST, \ 132 .src_reg = 0, \ 133 .off = 0, \ 134 .imm = IMM }) 135 136 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 137 138 #define BPF_ENDIAN(TYPE, DST, LEN) \ 139 ((struct bpf_insn) { \ 140 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 141 .dst_reg = DST, \ 142 .src_reg = 0, \ 143 .off = 0, \ 144 .imm = LEN }) 145 146 /* Short form of mov, dst_reg = src_reg */ 147 148 #define BPF_MOV64_REG(DST, SRC) \ 149 ((struct bpf_insn) { \ 150 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 151 .dst_reg = DST, \ 152 .src_reg = SRC, \ 153 .off = 0, \ 154 .imm = 0 }) 155 156 #define BPF_MOV32_REG(DST, SRC) \ 157 ((struct bpf_insn) { \ 158 .code = BPF_ALU | BPF_MOV | BPF_X, \ 159 .dst_reg = DST, \ 160 .src_reg = SRC, \ 161 .off = 0, \ 162 .imm = 0 }) 163 164 /* Short form of mov, dst_reg = imm32 */ 165 166 #define BPF_MOV64_IMM(DST, IMM) \ 167 ((struct bpf_insn) { \ 168 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 169 .dst_reg = DST, \ 170 .src_reg = 0, \ 171 .off = 0, \ 172 .imm = IMM }) 173 174 #define BPF_MOV32_IMM(DST, IMM) \ 175 ((struct bpf_insn) { \ 176 .code = BPF_ALU | BPF_MOV | BPF_K, \ 177 .dst_reg = DST, \ 178 .src_reg = 0, \ 179 .off = 0, \ 180 .imm = IMM }) 181 182 /* Special form of mov32, used for doing explicit zero extension on dst. */ 183 #define BPF_ZEXT_REG(DST) \ 184 ((struct bpf_insn) { \ 185 .code = BPF_ALU | BPF_MOV | BPF_X, \ 186 .dst_reg = DST, \ 187 .src_reg = DST, \ 188 .off = 0, \ 189 .imm = 1 }) 190 191 static inline bool insn_is_zext(const struct bpf_insn *insn) 192 { 193 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 194 } 195 196 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 197 #define BPF_LD_IMM64(DST, IMM) \ 198 BPF_LD_IMM64_RAW(DST, 0, IMM) 199 200 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 201 ((struct bpf_insn) { \ 202 .code = BPF_LD | BPF_DW | BPF_IMM, \ 203 .dst_reg = DST, \ 204 .src_reg = SRC, \ 205 .off = 0, \ 206 .imm = (__u32) (IMM) }), \ 207 ((struct bpf_insn) { \ 208 .code = 0, /* zero is reserved opcode */ \ 209 .dst_reg = 0, \ 210 .src_reg = 0, \ 211 .off = 0, \ 212 .imm = ((__u64) (IMM)) >> 32 }) 213 214 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 215 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 216 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 217 218 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 219 220 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 221 ((struct bpf_insn) { \ 222 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 223 .dst_reg = DST, \ 224 .src_reg = SRC, \ 225 .off = 0, \ 226 .imm = IMM }) 227 228 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 229 ((struct bpf_insn) { \ 230 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 231 .dst_reg = DST, \ 232 .src_reg = SRC, \ 233 .off = 0, \ 234 .imm = IMM }) 235 236 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 237 238 #define BPF_LD_ABS(SIZE, IMM) \ 239 ((struct bpf_insn) { \ 240 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 241 .dst_reg = 0, \ 242 .src_reg = 0, \ 243 .off = 0, \ 244 .imm = IMM }) 245 246 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 247 248 #define BPF_LD_IND(SIZE, SRC, IMM) \ 249 ((struct bpf_insn) { \ 250 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 251 .dst_reg = 0, \ 252 .src_reg = SRC, \ 253 .off = 0, \ 254 .imm = IMM }) 255 256 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 257 258 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 259 ((struct bpf_insn) { \ 260 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 261 .dst_reg = DST, \ 262 .src_reg = SRC, \ 263 .off = OFF, \ 264 .imm = 0 }) 265 266 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 267 268 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 269 ((struct bpf_insn) { \ 270 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 271 .dst_reg = DST, \ 272 .src_reg = SRC, \ 273 .off = OFF, \ 274 .imm = 0 }) 275 276 277 /* 278 * Atomic operations: 279 * 280 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 281 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 282 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 283 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 284 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 285 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 286 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 287 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 288 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 289 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 290 */ 291 292 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 293 ((struct bpf_insn) { \ 294 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 295 .dst_reg = DST, \ 296 .src_reg = SRC, \ 297 .off = OFF, \ 298 .imm = OP }) 299 300 /* Legacy alias */ 301 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 302 303 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 304 305 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 306 ((struct bpf_insn) { \ 307 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 308 .dst_reg = DST, \ 309 .src_reg = 0, \ 310 .off = OFF, \ 311 .imm = IMM }) 312 313 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 314 315 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 316 ((struct bpf_insn) { \ 317 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 318 .dst_reg = DST, \ 319 .src_reg = SRC, \ 320 .off = OFF, \ 321 .imm = 0 }) 322 323 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 324 325 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 326 ((struct bpf_insn) { \ 327 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 328 .dst_reg = DST, \ 329 .src_reg = 0, \ 330 .off = OFF, \ 331 .imm = IMM }) 332 333 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 334 335 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 336 ((struct bpf_insn) { \ 337 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 338 .dst_reg = DST, \ 339 .src_reg = SRC, \ 340 .off = OFF, \ 341 .imm = 0 }) 342 343 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 344 345 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 346 ((struct bpf_insn) { \ 347 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 348 .dst_reg = DST, \ 349 .src_reg = 0, \ 350 .off = OFF, \ 351 .imm = IMM }) 352 353 /* Unconditional jumps, goto pc + off16 */ 354 355 #define BPF_JMP_A(OFF) \ 356 ((struct bpf_insn) { \ 357 .code = BPF_JMP | BPF_JA, \ 358 .dst_reg = 0, \ 359 .src_reg = 0, \ 360 .off = OFF, \ 361 .imm = 0 }) 362 363 /* Relative call */ 364 365 #define BPF_CALL_REL(TGT) \ 366 ((struct bpf_insn) { \ 367 .code = BPF_JMP | BPF_CALL, \ 368 .dst_reg = 0, \ 369 .src_reg = BPF_PSEUDO_CALL, \ 370 .off = 0, \ 371 .imm = TGT }) 372 373 /* Convert function address to BPF immediate */ 374 375 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base) 376 377 #define BPF_EMIT_CALL(FUNC) \ 378 ((struct bpf_insn) { \ 379 .code = BPF_JMP | BPF_CALL, \ 380 .dst_reg = 0, \ 381 .src_reg = 0, \ 382 .off = 0, \ 383 .imm = BPF_CALL_IMM(FUNC) }) 384 385 /* Raw code statement block */ 386 387 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 388 ((struct bpf_insn) { \ 389 .code = CODE, \ 390 .dst_reg = DST, \ 391 .src_reg = SRC, \ 392 .off = OFF, \ 393 .imm = IMM }) 394 395 /* Program exit */ 396 397 #define BPF_EXIT_INSN() \ 398 ((struct bpf_insn) { \ 399 .code = BPF_JMP | BPF_EXIT, \ 400 .dst_reg = 0, \ 401 .src_reg = 0, \ 402 .off = 0, \ 403 .imm = 0 }) 404 405 /* Speculation barrier */ 406 407 #define BPF_ST_NOSPEC() \ 408 ((struct bpf_insn) { \ 409 .code = BPF_ST | BPF_NOSPEC, \ 410 .dst_reg = 0, \ 411 .src_reg = 0, \ 412 .off = 0, \ 413 .imm = 0 }) 414 415 /* Internal classic blocks for direct assignment */ 416 417 #define __BPF_STMT(CODE, K) \ 418 ((struct sock_filter) BPF_STMT(CODE, K)) 419 420 #define __BPF_JUMP(CODE, K, JT, JF) \ 421 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 422 423 #define bytes_to_bpf_size(bytes) \ 424 ({ \ 425 int bpf_size = -EINVAL; \ 426 \ 427 if (bytes == sizeof(u8)) \ 428 bpf_size = BPF_B; \ 429 else if (bytes == sizeof(u16)) \ 430 bpf_size = BPF_H; \ 431 else if (bytes == sizeof(u32)) \ 432 bpf_size = BPF_W; \ 433 else if (bytes == sizeof(u64)) \ 434 bpf_size = BPF_DW; \ 435 \ 436 bpf_size; \ 437 }) 438 439 #define bpf_size_to_bytes(bpf_size) \ 440 ({ \ 441 int bytes = -EINVAL; \ 442 \ 443 if (bpf_size == BPF_B) \ 444 bytes = sizeof(u8); \ 445 else if (bpf_size == BPF_H) \ 446 bytes = sizeof(u16); \ 447 else if (bpf_size == BPF_W) \ 448 bytes = sizeof(u32); \ 449 else if (bpf_size == BPF_DW) \ 450 bytes = sizeof(u64); \ 451 \ 452 bytes; \ 453 }) 454 455 #define BPF_SIZEOF(type) \ 456 ({ \ 457 const int __size = bytes_to_bpf_size(sizeof(type)); \ 458 BUILD_BUG_ON(__size < 0); \ 459 __size; \ 460 }) 461 462 #define BPF_FIELD_SIZEOF(type, field) \ 463 ({ \ 464 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 465 BUILD_BUG_ON(__size < 0); \ 466 __size; \ 467 }) 468 469 #define BPF_LDST_BYTES(insn) \ 470 ({ \ 471 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 472 WARN_ON(__size < 0); \ 473 __size; \ 474 }) 475 476 #define __BPF_MAP_0(m, v, ...) v 477 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 478 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 479 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 480 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 481 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 482 483 #define __BPF_REG_0(...) __BPF_PAD(5) 484 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 485 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 486 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 487 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 488 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 489 490 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 491 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 492 493 #define __BPF_CAST(t, a) \ 494 (__force t) \ 495 (__force \ 496 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 497 (unsigned long)0, (t)0))) a 498 #define __BPF_V void 499 #define __BPF_N 500 501 #define __BPF_DECL_ARGS(t, a) t a 502 #define __BPF_DECL_REGS(t, a) u64 a 503 504 #define __BPF_PAD(n) \ 505 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 506 u64, __ur_3, u64, __ur_4, u64, __ur_5) 507 508 #define BPF_CALL_x(x, name, ...) \ 509 static __always_inline \ 510 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 511 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 512 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 513 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 514 { \ 515 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 516 } \ 517 static __always_inline \ 518 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 519 520 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) 521 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) 522 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) 523 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) 524 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) 525 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) 526 527 #define bpf_ctx_range(TYPE, MEMBER) \ 528 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 529 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 530 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 531 #if BITS_PER_LONG == 64 532 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 533 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 534 #else 535 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 536 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 537 #endif /* BITS_PER_LONG == 64 */ 538 539 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 540 ({ \ 541 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 542 *(PTR_SIZE) = (SIZE); \ 543 offsetof(TYPE, MEMBER); \ 544 }) 545 546 /* A struct sock_filter is architecture independent. */ 547 struct compat_sock_fprog { 548 u16 len; 549 compat_uptr_t filter; /* struct sock_filter * */ 550 }; 551 552 struct sock_fprog_kern { 553 u16 len; 554 struct sock_filter *filter; 555 }; 556 557 /* Some arches need doubleword alignment for their instructions and/or data */ 558 #define BPF_IMAGE_ALIGNMENT 8 559 560 struct bpf_binary_header { 561 u32 size; 562 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 563 }; 564 565 struct bpf_prog_stats { 566 u64_stats_t cnt; 567 u64_stats_t nsecs; 568 u64_stats_t misses; 569 struct u64_stats_sync syncp; 570 } __aligned(2 * sizeof(u64)); 571 572 struct sk_filter { 573 refcount_t refcnt; 574 struct rcu_head rcu; 575 struct bpf_prog *prog; 576 }; 577 578 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 579 580 extern struct mutex nf_conn_btf_access_lock; 581 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 582 const struct bpf_reg_state *reg, 583 int off, int size); 584 585 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 586 const struct bpf_insn *insnsi, 587 unsigned int (*bpf_func)(const void *, 588 const struct bpf_insn *)); 589 590 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 591 const void *ctx, 592 bpf_dispatcher_fn dfunc) 593 { 594 u32 ret; 595 596 cant_migrate(); 597 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 598 struct bpf_prog_stats *stats; 599 u64 start = sched_clock(); 600 unsigned long flags; 601 602 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 603 stats = this_cpu_ptr(prog->stats); 604 flags = u64_stats_update_begin_irqsave(&stats->syncp); 605 u64_stats_inc(&stats->cnt); 606 u64_stats_add(&stats->nsecs, sched_clock() - start); 607 u64_stats_update_end_irqrestore(&stats->syncp, flags); 608 } else { 609 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 610 } 611 return ret; 612 } 613 614 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 615 { 616 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 617 } 618 619 /* 620 * Use in preemptible and therefore migratable context to make sure that 621 * the execution of the BPF program runs on one CPU. 622 * 623 * This uses migrate_disable/enable() explicitly to document that the 624 * invocation of a BPF program does not require reentrancy protection 625 * against a BPF program which is invoked from a preempting task. 626 */ 627 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 628 const void *ctx) 629 { 630 u32 ret; 631 632 migrate_disable(); 633 ret = bpf_prog_run(prog, ctx); 634 migrate_enable(); 635 return ret; 636 } 637 638 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 639 640 struct bpf_skb_data_end { 641 struct qdisc_skb_cb qdisc_cb; 642 void *data_meta; 643 void *data_end; 644 }; 645 646 struct bpf_nh_params { 647 u32 nh_family; 648 union { 649 u32 ipv4_nh; 650 struct in6_addr ipv6_nh; 651 }; 652 }; 653 654 struct bpf_redirect_info { 655 u64 tgt_index; 656 void *tgt_value; 657 struct bpf_map *map; 658 u32 flags; 659 u32 kern_flags; 660 u32 map_id; 661 enum bpf_map_type map_type; 662 struct bpf_nh_params nh; 663 }; 664 665 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 666 667 /* flags for bpf_redirect_info kern_flags */ 668 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 669 670 /* Compute the linear packet data range [data, data_end) which 671 * will be accessed by various program types (cls_bpf, act_bpf, 672 * lwt, ...). Subsystems allowing direct data access must (!) 673 * ensure that cb[] area can be written to when BPF program is 674 * invoked (otherwise cb[] save/restore is necessary). 675 */ 676 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 677 { 678 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 679 680 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 681 cb->data_meta = skb->data - skb_metadata_len(skb); 682 cb->data_end = skb->data + skb_headlen(skb); 683 } 684 685 /* Similar to bpf_compute_data_pointers(), except that save orginal 686 * data in cb->data and cb->meta_data for restore. 687 */ 688 static inline void bpf_compute_and_save_data_end( 689 struct sk_buff *skb, void **saved_data_end) 690 { 691 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 692 693 *saved_data_end = cb->data_end; 694 cb->data_end = skb->data + skb_headlen(skb); 695 } 696 697 /* Restore data saved by bpf_compute_data_pointers(). */ 698 static inline void bpf_restore_data_end( 699 struct sk_buff *skb, void *saved_data_end) 700 { 701 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 702 703 cb->data_end = saved_data_end; 704 } 705 706 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 707 { 708 /* eBPF programs may read/write skb->cb[] area to transfer meta 709 * data between tail calls. Since this also needs to work with 710 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 711 * 712 * In some socket filter cases, the cb unfortunately needs to be 713 * saved/restored so that protocol specific skb->cb[] data won't 714 * be lost. In any case, due to unpriviledged eBPF programs 715 * attached to sockets, we need to clear the bpf_skb_cb() area 716 * to not leak previous contents to user space. 717 */ 718 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 719 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 720 sizeof_field(struct qdisc_skb_cb, data)); 721 722 return qdisc_skb_cb(skb)->data; 723 } 724 725 /* Must be invoked with migration disabled */ 726 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 727 const void *ctx) 728 { 729 const struct sk_buff *skb = ctx; 730 u8 *cb_data = bpf_skb_cb(skb); 731 u8 cb_saved[BPF_SKB_CB_LEN]; 732 u32 res; 733 734 if (unlikely(prog->cb_access)) { 735 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 736 memset(cb_data, 0, sizeof(cb_saved)); 737 } 738 739 res = bpf_prog_run(prog, skb); 740 741 if (unlikely(prog->cb_access)) 742 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 743 744 return res; 745 } 746 747 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 748 struct sk_buff *skb) 749 { 750 u32 res; 751 752 migrate_disable(); 753 res = __bpf_prog_run_save_cb(prog, skb); 754 migrate_enable(); 755 return res; 756 } 757 758 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 759 struct sk_buff *skb) 760 { 761 u8 *cb_data = bpf_skb_cb(skb); 762 u32 res; 763 764 if (unlikely(prog->cb_access)) 765 memset(cb_data, 0, BPF_SKB_CB_LEN); 766 767 res = bpf_prog_run_pin_on_cpu(prog, skb); 768 return res; 769 } 770 771 DECLARE_BPF_DISPATCHER(xdp) 772 773 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 774 775 u32 xdp_master_redirect(struct xdp_buff *xdp); 776 777 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 778 struct xdp_buff *xdp) 779 { 780 /* Driver XDP hooks are invoked within a single NAPI poll cycle and thus 781 * under local_bh_disable(), which provides the needed RCU protection 782 * for accessing map entries. 783 */ 784 u32 act = __bpf_prog_run(prog, xdp, BPF_DISPATCHER_FUNC(xdp)); 785 786 if (static_branch_unlikely(&bpf_master_redirect_enabled_key)) { 787 if (act == XDP_TX && netif_is_bond_slave(xdp->rxq->dev)) 788 act = xdp_master_redirect(xdp); 789 } 790 791 return act; 792 } 793 794 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 795 796 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 797 { 798 return prog->len * sizeof(struct bpf_insn); 799 } 800 801 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 802 { 803 return round_up(bpf_prog_insn_size(prog) + 804 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 805 } 806 807 static inline unsigned int bpf_prog_size(unsigned int proglen) 808 { 809 return max(sizeof(struct bpf_prog), 810 offsetof(struct bpf_prog, insns[proglen])); 811 } 812 813 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 814 { 815 /* When classic BPF programs have been loaded and the arch 816 * does not have a classic BPF JIT (anymore), they have been 817 * converted via bpf_migrate_filter() to eBPF and thus always 818 * have an unspec program type. 819 */ 820 return prog->type == BPF_PROG_TYPE_UNSPEC; 821 } 822 823 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 824 { 825 const u32 size_machine = sizeof(unsigned long); 826 827 if (size > size_machine && size % size_machine == 0) 828 size = size_machine; 829 830 return size; 831 } 832 833 static inline bool 834 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 835 { 836 return size <= size_default && (size & (size - 1)) == 0; 837 } 838 839 static inline u8 840 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 841 { 842 u8 access_off = off & (size_default - 1); 843 844 #ifdef __LITTLE_ENDIAN 845 return access_off; 846 #else 847 return size_default - (access_off + size); 848 #endif 849 } 850 851 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 852 (size == sizeof(__u64) && \ 853 off >= offsetof(type, field) && \ 854 off + sizeof(__u64) <= offsetofend(type, field) && \ 855 off % sizeof(__u64) == 0) 856 857 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 858 859 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 860 { 861 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 862 if (!fp->jited) { 863 set_vm_flush_reset_perms(fp); 864 set_memory_ro((unsigned long)fp, fp->pages); 865 } 866 #endif 867 } 868 869 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 870 { 871 set_vm_flush_reset_perms(hdr); 872 set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT); 873 } 874 875 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 876 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 877 { 878 return sk_filter_trim_cap(sk, skb, 1); 879 } 880 881 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 882 void bpf_prog_free(struct bpf_prog *fp); 883 884 bool bpf_opcode_in_insntable(u8 code); 885 886 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 887 const u32 *insn_to_jit_off); 888 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 889 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 890 891 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 892 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 893 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 894 gfp_t gfp_extra_flags); 895 void __bpf_prog_free(struct bpf_prog *fp); 896 897 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 898 { 899 __bpf_prog_free(fp); 900 } 901 902 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 903 unsigned int flen); 904 905 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 906 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 907 bpf_aux_classic_check_t trans, bool save_orig); 908 void bpf_prog_destroy(struct bpf_prog *fp); 909 910 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 911 int sk_attach_bpf(u32 ufd, struct sock *sk); 912 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 913 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 914 void sk_reuseport_prog_free(struct bpf_prog *prog); 915 int sk_detach_filter(struct sock *sk); 916 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); 917 918 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 919 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 920 921 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 922 #define __bpf_call_base_args \ 923 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 924 (void *)__bpf_call_base) 925 926 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 927 void bpf_jit_compile(struct bpf_prog *prog); 928 bool bpf_jit_needs_zext(void); 929 bool bpf_jit_supports_subprog_tailcalls(void); 930 bool bpf_jit_supports_kfunc_call(void); 931 bool bpf_jit_supports_far_kfunc_call(void); 932 bool bpf_helper_changes_pkt_data(void *func); 933 934 static inline bool bpf_dump_raw_ok(const struct cred *cred) 935 { 936 /* Reconstruction of call-sites is dependent on kallsyms, 937 * thus make dump the same restriction. 938 */ 939 return kallsyms_show_value(cred); 940 } 941 942 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 943 const struct bpf_insn *patch, u32 len); 944 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 945 946 void bpf_clear_redirect_map(struct bpf_map *map); 947 948 static inline bool xdp_return_frame_no_direct(void) 949 { 950 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 951 952 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 953 } 954 955 static inline void xdp_set_return_frame_no_direct(void) 956 { 957 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 958 959 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 960 } 961 962 static inline void xdp_clear_return_frame_no_direct(void) 963 { 964 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 965 966 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 967 } 968 969 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 970 unsigned int pktlen) 971 { 972 unsigned int len; 973 974 if (unlikely(!(fwd->flags & IFF_UP))) 975 return -ENETDOWN; 976 977 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 978 if (pktlen > len) 979 return -EMSGSIZE; 980 981 return 0; 982 } 983 984 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 985 * same cpu context. Further for best results no more than a single map 986 * for the do_redirect/do_flush pair should be used. This limitation is 987 * because we only track one map and force a flush when the map changes. 988 * This does not appear to be a real limitation for existing software. 989 */ 990 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 991 struct xdp_buff *xdp, struct bpf_prog *prog); 992 int xdp_do_redirect(struct net_device *dev, 993 struct xdp_buff *xdp, 994 struct bpf_prog *prog); 995 int xdp_do_redirect_frame(struct net_device *dev, 996 struct xdp_buff *xdp, 997 struct xdp_frame *xdpf, 998 struct bpf_prog *prog); 999 void xdp_do_flush(void); 1000 1001 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as 1002 * it is no longer only flushing maps. Keep this define for compatibility 1003 * until all drivers are updated - do not use xdp_do_flush_map() in new code! 1004 */ 1005 #define xdp_do_flush_map xdp_do_flush 1006 1007 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act); 1008 1009 #ifdef CONFIG_INET 1010 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1011 struct bpf_prog *prog, struct sk_buff *skb, 1012 struct sock *migrating_sk, 1013 u32 hash); 1014 #else 1015 static inline struct sock * 1016 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1017 struct bpf_prog *prog, struct sk_buff *skb, 1018 struct sock *migrating_sk, 1019 u32 hash) 1020 { 1021 return NULL; 1022 } 1023 #endif 1024 1025 #ifdef CONFIG_BPF_JIT 1026 extern int bpf_jit_enable; 1027 extern int bpf_jit_harden; 1028 extern int bpf_jit_kallsyms; 1029 extern long bpf_jit_limit; 1030 extern long bpf_jit_limit_max; 1031 1032 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1033 1034 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); 1035 1036 struct bpf_binary_header * 1037 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1038 unsigned int alignment, 1039 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1040 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1041 u64 bpf_jit_alloc_exec_limit(void); 1042 void *bpf_jit_alloc_exec(unsigned long size); 1043 void bpf_jit_free_exec(void *addr); 1044 void bpf_jit_free(struct bpf_prog *fp); 1045 struct bpf_binary_header * 1046 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); 1047 1048 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); 1049 void bpf_prog_pack_free(struct bpf_binary_header *hdr); 1050 1051 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 1052 { 1053 return list_empty(&fp->aux->ksym.lnode) || 1054 fp->aux->ksym.lnode.prev == LIST_POISON2; 1055 } 1056 1057 struct bpf_binary_header * 1058 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, 1059 unsigned int alignment, 1060 struct bpf_binary_header **rw_hdr, 1061 u8 **rw_image, 1062 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1063 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, 1064 struct bpf_binary_header *ro_header, 1065 struct bpf_binary_header *rw_header); 1066 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, 1067 struct bpf_binary_header *rw_header); 1068 1069 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1070 struct bpf_jit_poke_descriptor *poke); 1071 1072 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1073 const struct bpf_insn *insn, bool extra_pass, 1074 u64 *func_addr, bool *func_addr_fixed); 1075 1076 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1077 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1078 1079 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1080 u32 pass, void *image) 1081 { 1082 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1083 proglen, pass, image, current->comm, task_pid_nr(current)); 1084 1085 if (image) 1086 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1087 16, 1, image, proglen, false); 1088 } 1089 1090 static inline bool bpf_jit_is_ebpf(void) 1091 { 1092 # ifdef CONFIG_HAVE_EBPF_JIT 1093 return true; 1094 # else 1095 return false; 1096 # endif 1097 } 1098 1099 static inline bool ebpf_jit_enabled(void) 1100 { 1101 return bpf_jit_enable && bpf_jit_is_ebpf(); 1102 } 1103 1104 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1105 { 1106 return fp->jited && bpf_jit_is_ebpf(); 1107 } 1108 1109 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1110 { 1111 /* These are the prerequisites, should someone ever have the 1112 * idea to call blinding outside of them, we make sure to 1113 * bail out. 1114 */ 1115 if (!bpf_jit_is_ebpf()) 1116 return false; 1117 if (!prog->jit_requested) 1118 return false; 1119 if (!bpf_jit_harden) 1120 return false; 1121 if (bpf_jit_harden == 1 && bpf_capable()) 1122 return false; 1123 1124 return true; 1125 } 1126 1127 static inline bool bpf_jit_kallsyms_enabled(void) 1128 { 1129 /* There are a couple of corner cases where kallsyms should 1130 * not be enabled f.e. on hardening. 1131 */ 1132 if (bpf_jit_harden) 1133 return false; 1134 if (!bpf_jit_kallsyms) 1135 return false; 1136 if (bpf_jit_kallsyms == 1) 1137 return true; 1138 1139 return false; 1140 } 1141 1142 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1143 unsigned long *off, char *sym); 1144 bool is_bpf_text_address(unsigned long addr); 1145 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1146 char *sym); 1147 1148 static inline const char * 1149 bpf_address_lookup(unsigned long addr, unsigned long *size, 1150 unsigned long *off, char **modname, char *sym) 1151 { 1152 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1153 1154 if (ret && modname) 1155 *modname = NULL; 1156 return ret; 1157 } 1158 1159 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1160 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1161 1162 #else /* CONFIG_BPF_JIT */ 1163 1164 static inline bool ebpf_jit_enabled(void) 1165 { 1166 return false; 1167 } 1168 1169 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1170 { 1171 return false; 1172 } 1173 1174 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1175 { 1176 return false; 1177 } 1178 1179 static inline int 1180 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1181 struct bpf_jit_poke_descriptor *poke) 1182 { 1183 return -ENOTSUPP; 1184 } 1185 1186 static inline void bpf_jit_free(struct bpf_prog *fp) 1187 { 1188 bpf_prog_unlock_free(fp); 1189 } 1190 1191 static inline bool bpf_jit_kallsyms_enabled(void) 1192 { 1193 return false; 1194 } 1195 1196 static inline const char * 1197 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1198 unsigned long *off, char *sym) 1199 { 1200 return NULL; 1201 } 1202 1203 static inline bool is_bpf_text_address(unsigned long addr) 1204 { 1205 return false; 1206 } 1207 1208 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1209 char *type, char *sym) 1210 { 1211 return -ERANGE; 1212 } 1213 1214 static inline const char * 1215 bpf_address_lookup(unsigned long addr, unsigned long *size, 1216 unsigned long *off, char **modname, char *sym) 1217 { 1218 return NULL; 1219 } 1220 1221 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1222 { 1223 } 1224 1225 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1226 { 1227 } 1228 1229 #endif /* CONFIG_BPF_JIT */ 1230 1231 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1232 1233 #define BPF_ANC BIT(15) 1234 1235 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1236 { 1237 switch (first->code) { 1238 case BPF_RET | BPF_K: 1239 case BPF_LD | BPF_W | BPF_LEN: 1240 return false; 1241 1242 case BPF_LD | BPF_W | BPF_ABS: 1243 case BPF_LD | BPF_H | BPF_ABS: 1244 case BPF_LD | BPF_B | BPF_ABS: 1245 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1246 return true; 1247 return false; 1248 1249 default: 1250 return true; 1251 } 1252 } 1253 1254 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1255 { 1256 BUG_ON(ftest->code & BPF_ANC); 1257 1258 switch (ftest->code) { 1259 case BPF_LD | BPF_W | BPF_ABS: 1260 case BPF_LD | BPF_H | BPF_ABS: 1261 case BPF_LD | BPF_B | BPF_ABS: 1262 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1263 return BPF_ANC | SKF_AD_##CODE 1264 switch (ftest->k) { 1265 BPF_ANCILLARY(PROTOCOL); 1266 BPF_ANCILLARY(PKTTYPE); 1267 BPF_ANCILLARY(IFINDEX); 1268 BPF_ANCILLARY(NLATTR); 1269 BPF_ANCILLARY(NLATTR_NEST); 1270 BPF_ANCILLARY(MARK); 1271 BPF_ANCILLARY(QUEUE); 1272 BPF_ANCILLARY(HATYPE); 1273 BPF_ANCILLARY(RXHASH); 1274 BPF_ANCILLARY(CPU); 1275 BPF_ANCILLARY(ALU_XOR_X); 1276 BPF_ANCILLARY(VLAN_TAG); 1277 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1278 BPF_ANCILLARY(PAY_OFFSET); 1279 BPF_ANCILLARY(RANDOM); 1280 BPF_ANCILLARY(VLAN_TPID); 1281 } 1282 fallthrough; 1283 default: 1284 return ftest->code; 1285 } 1286 } 1287 1288 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1289 int k, unsigned int size); 1290 1291 static inline int bpf_tell_extensions(void) 1292 { 1293 return SKF_AD_MAX; 1294 } 1295 1296 struct bpf_sock_addr_kern { 1297 struct sock *sk; 1298 struct sockaddr *uaddr; 1299 /* Temporary "register" to make indirect stores to nested structures 1300 * defined above. We need three registers to make such a store, but 1301 * only two (src and dst) are available at convert_ctx_access time 1302 */ 1303 u64 tmp_reg; 1304 void *t_ctx; /* Attach type specific context. */ 1305 }; 1306 1307 struct bpf_sock_ops_kern { 1308 struct sock *sk; 1309 union { 1310 u32 args[4]; 1311 u32 reply; 1312 u32 replylong[4]; 1313 }; 1314 struct sk_buff *syn_skb; 1315 struct sk_buff *skb; 1316 void *skb_data_end; 1317 u8 op; 1318 u8 is_fullsock; 1319 u8 remaining_opt_len; 1320 u64 temp; /* temp and everything after is not 1321 * initialized to 0 before calling 1322 * the BPF program. New fields that 1323 * should be initialized to 0 should 1324 * be inserted before temp. 1325 * temp is scratch storage used by 1326 * sock_ops_convert_ctx_access 1327 * as temporary storage of a register. 1328 */ 1329 }; 1330 1331 struct bpf_sysctl_kern { 1332 struct ctl_table_header *head; 1333 struct ctl_table *table; 1334 void *cur_val; 1335 size_t cur_len; 1336 void *new_val; 1337 size_t new_len; 1338 int new_updated; 1339 int write; 1340 loff_t *ppos; 1341 /* Temporary "register" for indirect stores to ppos. */ 1342 u64 tmp_reg; 1343 }; 1344 1345 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1346 struct bpf_sockopt_buf { 1347 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1348 }; 1349 1350 struct bpf_sockopt_kern { 1351 struct sock *sk; 1352 u8 *optval; 1353 u8 *optval_end; 1354 s32 level; 1355 s32 optname; 1356 s32 optlen; 1357 /* for retval in struct bpf_cg_run_ctx */ 1358 struct task_struct *current_task; 1359 /* Temporary "register" for indirect stores to ppos. */ 1360 u64 tmp_reg; 1361 }; 1362 1363 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1364 1365 struct bpf_sk_lookup_kern { 1366 u16 family; 1367 u16 protocol; 1368 __be16 sport; 1369 u16 dport; 1370 struct { 1371 __be32 saddr; 1372 __be32 daddr; 1373 } v4; 1374 struct { 1375 const struct in6_addr *saddr; 1376 const struct in6_addr *daddr; 1377 } v6; 1378 struct sock *selected_sk; 1379 u32 ingress_ifindex; 1380 bool no_reuseport; 1381 }; 1382 1383 extern struct static_key_false bpf_sk_lookup_enabled; 1384 1385 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1386 * 1387 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1388 * SK_DROP. Their meaning is as follows: 1389 * 1390 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1391 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1392 * SK_DROP : terminate lookup with -ECONNREFUSED 1393 * 1394 * This macro aggregates return values and selected sockets from 1395 * multiple BPF programs according to following rules in order: 1396 * 1397 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1398 * macro result is SK_PASS and last ctx.selected_sk is used. 1399 * 2. If any program returned SK_DROP return value, 1400 * macro result is SK_DROP. 1401 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1402 * 1403 * Caller must ensure that the prog array is non-NULL, and that the 1404 * array as well as the programs it contains remain valid. 1405 */ 1406 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1407 ({ \ 1408 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1409 struct bpf_prog_array_item *_item; \ 1410 struct sock *_selected_sk = NULL; \ 1411 bool _no_reuseport = false; \ 1412 struct bpf_prog *_prog; \ 1413 bool _all_pass = true; \ 1414 u32 _ret; \ 1415 \ 1416 migrate_disable(); \ 1417 _item = &(array)->items[0]; \ 1418 while ((_prog = READ_ONCE(_item->prog))) { \ 1419 /* restore most recent selection */ \ 1420 _ctx->selected_sk = _selected_sk; \ 1421 _ctx->no_reuseport = _no_reuseport; \ 1422 \ 1423 _ret = func(_prog, _ctx); \ 1424 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1425 /* remember last non-NULL socket */ \ 1426 _selected_sk = _ctx->selected_sk; \ 1427 _no_reuseport = _ctx->no_reuseport; \ 1428 } else if (_ret == SK_DROP && _all_pass) { \ 1429 _all_pass = false; \ 1430 } \ 1431 _item++; \ 1432 } \ 1433 _ctx->selected_sk = _selected_sk; \ 1434 _ctx->no_reuseport = _no_reuseport; \ 1435 migrate_enable(); \ 1436 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1437 }) 1438 1439 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1440 const __be32 saddr, const __be16 sport, 1441 const __be32 daddr, const u16 dport, 1442 const int ifindex, struct sock **psk) 1443 { 1444 struct bpf_prog_array *run_array; 1445 struct sock *selected_sk = NULL; 1446 bool no_reuseport = false; 1447 1448 rcu_read_lock(); 1449 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1450 if (run_array) { 1451 struct bpf_sk_lookup_kern ctx = { 1452 .family = AF_INET, 1453 .protocol = protocol, 1454 .v4.saddr = saddr, 1455 .v4.daddr = daddr, 1456 .sport = sport, 1457 .dport = dport, 1458 .ingress_ifindex = ifindex, 1459 }; 1460 u32 act; 1461 1462 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1463 if (act == SK_PASS) { 1464 selected_sk = ctx.selected_sk; 1465 no_reuseport = ctx.no_reuseport; 1466 } else { 1467 selected_sk = ERR_PTR(-ECONNREFUSED); 1468 } 1469 } 1470 rcu_read_unlock(); 1471 *psk = selected_sk; 1472 return no_reuseport; 1473 } 1474 1475 #if IS_ENABLED(CONFIG_IPV6) 1476 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1477 const struct in6_addr *saddr, 1478 const __be16 sport, 1479 const struct in6_addr *daddr, 1480 const u16 dport, 1481 const int ifindex, struct sock **psk) 1482 { 1483 struct bpf_prog_array *run_array; 1484 struct sock *selected_sk = NULL; 1485 bool no_reuseport = false; 1486 1487 rcu_read_lock(); 1488 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1489 if (run_array) { 1490 struct bpf_sk_lookup_kern ctx = { 1491 .family = AF_INET6, 1492 .protocol = protocol, 1493 .v6.saddr = saddr, 1494 .v6.daddr = daddr, 1495 .sport = sport, 1496 .dport = dport, 1497 .ingress_ifindex = ifindex, 1498 }; 1499 u32 act; 1500 1501 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1502 if (act == SK_PASS) { 1503 selected_sk = ctx.selected_sk; 1504 no_reuseport = ctx.no_reuseport; 1505 } else { 1506 selected_sk = ERR_PTR(-ECONNREFUSED); 1507 } 1508 } 1509 rcu_read_unlock(); 1510 *psk = selected_sk; 1511 return no_reuseport; 1512 } 1513 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1514 1515 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, 1516 u64 flags, const u64 flag_mask, 1517 void *lookup_elem(struct bpf_map *map, u32 key)) 1518 { 1519 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1520 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1521 1522 /* Lower bits of the flags are used as return code on lookup failure */ 1523 if (unlikely(flags & ~(action_mask | flag_mask))) 1524 return XDP_ABORTED; 1525 1526 ri->tgt_value = lookup_elem(map, index); 1527 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1528 /* If the lookup fails we want to clear out the state in the 1529 * redirect_info struct completely, so that if an eBPF program 1530 * performs multiple lookups, the last one always takes 1531 * precedence. 1532 */ 1533 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1534 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1535 return flags & action_mask; 1536 } 1537 1538 ri->tgt_index = index; 1539 ri->map_id = map->id; 1540 ri->map_type = map->map_type; 1541 1542 if (flags & BPF_F_BROADCAST) { 1543 WRITE_ONCE(ri->map, map); 1544 ri->flags = flags; 1545 } else { 1546 WRITE_ONCE(ri->map, NULL); 1547 ri->flags = 0; 1548 } 1549 1550 return XDP_REDIRECT; 1551 } 1552 1553 #ifdef CONFIG_NET 1554 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); 1555 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1556 u32 len, u64 flags); 1557 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1558 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1559 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); 1560 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 1561 void *buf, unsigned long len, bool flush); 1562 #else /* CONFIG_NET */ 1563 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, 1564 void *to, u32 len) 1565 { 1566 return -EOPNOTSUPP; 1567 } 1568 1569 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, 1570 const void *from, u32 len, u64 flags) 1571 { 1572 return -EOPNOTSUPP; 1573 } 1574 1575 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, 1576 void *buf, u32 len) 1577 { 1578 return -EOPNOTSUPP; 1579 } 1580 1581 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, 1582 void *buf, u32 len) 1583 { 1584 return -EOPNOTSUPP; 1585 } 1586 1587 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 1588 { 1589 return NULL; 1590 } 1591 1592 static inline void *bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, 1593 unsigned long len, bool flush) 1594 { 1595 return NULL; 1596 } 1597 #endif /* CONFIG_NET */ 1598 1599 #endif /* __LINUX_FILTER_H__ */ 1600