1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Linux Socket Filter Data Structures 4 */ 5 #ifndef __LINUX_FILTER_H__ 6 #define __LINUX_FILTER_H__ 7 8 #include <stdarg.h> 9 10 #include <linux/atomic.h> 11 #include <linux/refcount.h> 12 #include <linux/compat.h> 13 #include <linux/skbuff.h> 14 #include <linux/linkage.h> 15 #include <linux/printk.h> 16 #include <linux/workqueue.h> 17 #include <linux/sched.h> 18 #include <linux/capability.h> 19 #include <linux/set_memory.h> 20 #include <linux/kallsyms.h> 21 #include <linux/if_vlan.h> 22 #include <linux/vmalloc.h> 23 #include <linux/sockptr.h> 24 #include <crypto/sha1.h> 25 #include <linux/u64_stats_sync.h> 26 27 #include <net/sch_generic.h> 28 29 #include <asm/byteorder.h> 30 #include <uapi/linux/filter.h> 31 #include <uapi/linux/bpf.h> 32 33 struct sk_buff; 34 struct sock; 35 struct seccomp_data; 36 struct bpf_prog_aux; 37 struct xdp_rxq_info; 38 struct xdp_buff; 39 struct sock_reuseport; 40 struct ctl_table; 41 struct ctl_table_header; 42 43 /* ArgX, context and stack frame pointer register positions. Note, 44 * Arg1, Arg2, Arg3, etc are used as argument mappings of function 45 * calls in BPF_CALL instruction. 46 */ 47 #define BPF_REG_ARG1 BPF_REG_1 48 #define BPF_REG_ARG2 BPF_REG_2 49 #define BPF_REG_ARG3 BPF_REG_3 50 #define BPF_REG_ARG4 BPF_REG_4 51 #define BPF_REG_ARG5 BPF_REG_5 52 #define BPF_REG_CTX BPF_REG_6 53 #define BPF_REG_FP BPF_REG_10 54 55 /* Additional register mappings for converted user programs. */ 56 #define BPF_REG_A BPF_REG_0 57 #define BPF_REG_X BPF_REG_7 58 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */ 59 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */ 60 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */ 61 62 /* Kernel hidden auxiliary/helper register. */ 63 #define BPF_REG_AX MAX_BPF_REG 64 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1) 65 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG 66 67 /* unused opcode to mark special call to bpf_tail_call() helper */ 68 #define BPF_TAIL_CALL 0xf0 69 70 /* unused opcode to mark special load instruction. Same as BPF_ABS */ 71 #define BPF_PROBE_MEM 0x20 72 73 /* unused opcode to mark call to interpreter with arguments */ 74 #define BPF_CALL_ARGS 0xe0 75 76 /* As per nm, we expose JITed images as text (code) section for 77 * kallsyms. That way, tools like perf can find it to match 78 * addresses. 79 */ 80 #define BPF_SYM_ELF_TYPE 't' 81 82 /* BPF program can access up to 512 bytes of stack space. */ 83 #define MAX_BPF_STACK 512 84 85 /* Helper macros for filter block array initializers. */ 86 87 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 88 89 #define BPF_ALU64_REG(OP, DST, SRC) \ 90 ((struct bpf_insn) { \ 91 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 92 .dst_reg = DST, \ 93 .src_reg = SRC, \ 94 .off = 0, \ 95 .imm = 0 }) 96 97 #define BPF_ALU32_REG(OP, DST, SRC) \ 98 ((struct bpf_insn) { \ 99 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 100 .dst_reg = DST, \ 101 .src_reg = SRC, \ 102 .off = 0, \ 103 .imm = 0 }) 104 105 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 106 107 #define BPF_ALU64_IMM(OP, DST, IMM) \ 108 ((struct bpf_insn) { \ 109 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 110 .dst_reg = DST, \ 111 .src_reg = 0, \ 112 .off = 0, \ 113 .imm = IMM }) 114 115 #define BPF_ALU32_IMM(OP, DST, IMM) \ 116 ((struct bpf_insn) { \ 117 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 118 .dst_reg = DST, \ 119 .src_reg = 0, \ 120 .off = 0, \ 121 .imm = IMM }) 122 123 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 124 125 #define BPF_ENDIAN(TYPE, DST, LEN) \ 126 ((struct bpf_insn) { \ 127 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 128 .dst_reg = DST, \ 129 .src_reg = 0, \ 130 .off = 0, \ 131 .imm = LEN }) 132 133 /* Short form of mov, dst_reg = src_reg */ 134 135 #define BPF_MOV64_REG(DST, SRC) \ 136 ((struct bpf_insn) { \ 137 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 138 .dst_reg = DST, \ 139 .src_reg = SRC, \ 140 .off = 0, \ 141 .imm = 0 }) 142 143 #define BPF_MOV32_REG(DST, SRC) \ 144 ((struct bpf_insn) { \ 145 .code = BPF_ALU | BPF_MOV | BPF_X, \ 146 .dst_reg = DST, \ 147 .src_reg = SRC, \ 148 .off = 0, \ 149 .imm = 0 }) 150 151 /* Short form of mov, dst_reg = imm32 */ 152 153 #define BPF_MOV64_IMM(DST, IMM) \ 154 ((struct bpf_insn) { \ 155 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 156 .dst_reg = DST, \ 157 .src_reg = 0, \ 158 .off = 0, \ 159 .imm = IMM }) 160 161 #define BPF_MOV32_IMM(DST, IMM) \ 162 ((struct bpf_insn) { \ 163 .code = BPF_ALU | BPF_MOV | BPF_K, \ 164 .dst_reg = DST, \ 165 .src_reg = 0, \ 166 .off = 0, \ 167 .imm = IMM }) 168 169 /* Special form of mov32, used for doing explicit zero extension on dst. */ 170 #define BPF_ZEXT_REG(DST) \ 171 ((struct bpf_insn) { \ 172 .code = BPF_ALU | BPF_MOV | BPF_X, \ 173 .dst_reg = DST, \ 174 .src_reg = DST, \ 175 .off = 0, \ 176 .imm = 1 }) 177 178 static inline bool insn_is_zext(const struct bpf_insn *insn) 179 { 180 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 181 } 182 183 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 184 #define BPF_LD_IMM64(DST, IMM) \ 185 BPF_LD_IMM64_RAW(DST, 0, IMM) 186 187 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 188 ((struct bpf_insn) { \ 189 .code = BPF_LD | BPF_DW | BPF_IMM, \ 190 .dst_reg = DST, \ 191 .src_reg = SRC, \ 192 .off = 0, \ 193 .imm = (__u32) (IMM) }), \ 194 ((struct bpf_insn) { \ 195 .code = 0, /* zero is reserved opcode */ \ 196 .dst_reg = 0, \ 197 .src_reg = 0, \ 198 .off = 0, \ 199 .imm = ((__u64) (IMM)) >> 32 }) 200 201 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 202 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 203 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 204 205 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 206 207 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 208 ((struct bpf_insn) { \ 209 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 210 .dst_reg = DST, \ 211 .src_reg = SRC, \ 212 .off = 0, \ 213 .imm = IMM }) 214 215 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 216 ((struct bpf_insn) { \ 217 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 218 .dst_reg = DST, \ 219 .src_reg = SRC, \ 220 .off = 0, \ 221 .imm = IMM }) 222 223 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 224 225 #define BPF_LD_ABS(SIZE, IMM) \ 226 ((struct bpf_insn) { \ 227 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 228 .dst_reg = 0, \ 229 .src_reg = 0, \ 230 .off = 0, \ 231 .imm = IMM }) 232 233 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 234 235 #define BPF_LD_IND(SIZE, SRC, IMM) \ 236 ((struct bpf_insn) { \ 237 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 238 .dst_reg = 0, \ 239 .src_reg = SRC, \ 240 .off = 0, \ 241 .imm = IMM }) 242 243 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 244 245 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 246 ((struct bpf_insn) { \ 247 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 248 .dst_reg = DST, \ 249 .src_reg = SRC, \ 250 .off = OFF, \ 251 .imm = 0 }) 252 253 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 254 255 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 256 ((struct bpf_insn) { \ 257 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 258 .dst_reg = DST, \ 259 .src_reg = SRC, \ 260 .off = OFF, \ 261 .imm = 0 }) 262 263 264 /* 265 * Atomic operations: 266 * 267 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 268 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 269 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 270 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 271 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 272 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 273 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 274 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 275 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 276 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 277 */ 278 279 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 280 ((struct bpf_insn) { \ 281 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 282 .dst_reg = DST, \ 283 .src_reg = SRC, \ 284 .off = OFF, \ 285 .imm = OP }) 286 287 /* Legacy alias */ 288 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 289 290 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 291 292 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 293 ((struct bpf_insn) { \ 294 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 295 .dst_reg = DST, \ 296 .src_reg = 0, \ 297 .off = OFF, \ 298 .imm = IMM }) 299 300 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 301 302 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 303 ((struct bpf_insn) { \ 304 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 305 .dst_reg = DST, \ 306 .src_reg = SRC, \ 307 .off = OFF, \ 308 .imm = 0 }) 309 310 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 311 312 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 313 ((struct bpf_insn) { \ 314 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 315 .dst_reg = DST, \ 316 .src_reg = 0, \ 317 .off = OFF, \ 318 .imm = IMM }) 319 320 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 321 322 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 323 ((struct bpf_insn) { \ 324 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 325 .dst_reg = DST, \ 326 .src_reg = SRC, \ 327 .off = OFF, \ 328 .imm = 0 }) 329 330 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 331 332 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 333 ((struct bpf_insn) { \ 334 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 335 .dst_reg = DST, \ 336 .src_reg = 0, \ 337 .off = OFF, \ 338 .imm = IMM }) 339 340 /* Unconditional jumps, goto pc + off16 */ 341 342 #define BPF_JMP_A(OFF) \ 343 ((struct bpf_insn) { \ 344 .code = BPF_JMP | BPF_JA, \ 345 .dst_reg = 0, \ 346 .src_reg = 0, \ 347 .off = OFF, \ 348 .imm = 0 }) 349 350 /* Relative call */ 351 352 #define BPF_CALL_REL(TGT) \ 353 ((struct bpf_insn) { \ 354 .code = BPF_JMP | BPF_CALL, \ 355 .dst_reg = 0, \ 356 .src_reg = BPF_PSEUDO_CALL, \ 357 .off = 0, \ 358 .imm = TGT }) 359 360 /* Function call */ 361 362 #define BPF_CAST_CALL(x) \ 363 ((u64 (*)(u64, u64, u64, u64, u64))(x)) 364 365 #define BPF_EMIT_CALL(FUNC) \ 366 ((struct bpf_insn) { \ 367 .code = BPF_JMP | BPF_CALL, \ 368 .dst_reg = 0, \ 369 .src_reg = 0, \ 370 .off = 0, \ 371 .imm = ((FUNC) - __bpf_call_base) }) 372 373 /* Raw code statement block */ 374 375 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 376 ((struct bpf_insn) { \ 377 .code = CODE, \ 378 .dst_reg = DST, \ 379 .src_reg = SRC, \ 380 .off = OFF, \ 381 .imm = IMM }) 382 383 /* Program exit */ 384 385 #define BPF_EXIT_INSN() \ 386 ((struct bpf_insn) { \ 387 .code = BPF_JMP | BPF_EXIT, \ 388 .dst_reg = 0, \ 389 .src_reg = 0, \ 390 .off = 0, \ 391 .imm = 0 }) 392 393 /* Internal classic blocks for direct assignment */ 394 395 #define __BPF_STMT(CODE, K) \ 396 ((struct sock_filter) BPF_STMT(CODE, K)) 397 398 #define __BPF_JUMP(CODE, K, JT, JF) \ 399 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 400 401 #define bytes_to_bpf_size(bytes) \ 402 ({ \ 403 int bpf_size = -EINVAL; \ 404 \ 405 if (bytes == sizeof(u8)) \ 406 bpf_size = BPF_B; \ 407 else if (bytes == sizeof(u16)) \ 408 bpf_size = BPF_H; \ 409 else if (bytes == sizeof(u32)) \ 410 bpf_size = BPF_W; \ 411 else if (bytes == sizeof(u64)) \ 412 bpf_size = BPF_DW; \ 413 \ 414 bpf_size; \ 415 }) 416 417 #define bpf_size_to_bytes(bpf_size) \ 418 ({ \ 419 int bytes = -EINVAL; \ 420 \ 421 if (bpf_size == BPF_B) \ 422 bytes = sizeof(u8); \ 423 else if (bpf_size == BPF_H) \ 424 bytes = sizeof(u16); \ 425 else if (bpf_size == BPF_W) \ 426 bytes = sizeof(u32); \ 427 else if (bpf_size == BPF_DW) \ 428 bytes = sizeof(u64); \ 429 \ 430 bytes; \ 431 }) 432 433 #define BPF_SIZEOF(type) \ 434 ({ \ 435 const int __size = bytes_to_bpf_size(sizeof(type)); \ 436 BUILD_BUG_ON(__size < 0); \ 437 __size; \ 438 }) 439 440 #define BPF_FIELD_SIZEOF(type, field) \ 441 ({ \ 442 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 443 BUILD_BUG_ON(__size < 0); \ 444 __size; \ 445 }) 446 447 #define BPF_LDST_BYTES(insn) \ 448 ({ \ 449 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 450 WARN_ON(__size < 0); \ 451 __size; \ 452 }) 453 454 #define __BPF_MAP_0(m, v, ...) v 455 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 456 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 457 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 458 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 459 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 460 461 #define __BPF_REG_0(...) __BPF_PAD(5) 462 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 463 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 464 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 465 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 466 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 467 468 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 469 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 470 471 #define __BPF_CAST(t, a) \ 472 (__force t) \ 473 (__force \ 474 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 475 (unsigned long)0, (t)0))) a 476 #define __BPF_V void 477 #define __BPF_N 478 479 #define __BPF_DECL_ARGS(t, a) t a 480 #define __BPF_DECL_REGS(t, a) u64 a 481 482 #define __BPF_PAD(n) \ 483 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 484 u64, __ur_3, u64, __ur_4, u64, __ur_5) 485 486 #define BPF_CALL_x(x, name, ...) \ 487 static __always_inline \ 488 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 489 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 490 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 491 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 492 { \ 493 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 494 } \ 495 static __always_inline \ 496 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 497 498 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) 499 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) 500 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) 501 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) 502 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) 503 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) 504 505 #define bpf_ctx_range(TYPE, MEMBER) \ 506 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 507 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 508 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 509 #if BITS_PER_LONG == 64 510 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 511 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 512 #else 513 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 514 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 515 #endif /* BITS_PER_LONG == 64 */ 516 517 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 518 ({ \ 519 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 520 *(PTR_SIZE) = (SIZE); \ 521 offsetof(TYPE, MEMBER); \ 522 }) 523 524 /* A struct sock_filter is architecture independent. */ 525 struct compat_sock_fprog { 526 u16 len; 527 compat_uptr_t filter; /* struct sock_filter * */ 528 }; 529 530 struct sock_fprog_kern { 531 u16 len; 532 struct sock_filter *filter; 533 }; 534 535 /* Some arches need doubleword alignment for their instructions and/or data */ 536 #define BPF_IMAGE_ALIGNMENT 8 537 538 struct bpf_binary_header { 539 u32 pages; 540 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 541 }; 542 543 struct bpf_prog_stats { 544 u64 cnt; 545 u64 nsecs; 546 u64 misses; 547 struct u64_stats_sync syncp; 548 } __aligned(2 * sizeof(u64)); 549 550 struct bpf_prog { 551 u16 pages; /* Number of allocated pages */ 552 u16 jited:1, /* Is our filter JIT'ed? */ 553 jit_requested:1,/* archs need to JIT the prog */ 554 gpl_compatible:1, /* Is filter GPL compatible? */ 555 cb_access:1, /* Is control block accessed? */ 556 dst_needed:1, /* Do we need dst entry? */ 557 blinded:1, /* Was blinded */ 558 is_func:1, /* program is a bpf function */ 559 kprobe_override:1, /* Do we override a kprobe? */ 560 has_callchain_buf:1, /* callchain buffer allocated? */ 561 enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */ 562 call_get_stack:1; /* Do we call bpf_get_stack() or bpf_get_stackid() */ 563 enum bpf_prog_type type; /* Type of BPF program */ 564 enum bpf_attach_type expected_attach_type; /* For some prog types */ 565 u32 len; /* Number of filter blocks */ 566 u32 jited_len; /* Size of jited insns in bytes */ 567 u8 tag[BPF_TAG_SIZE]; 568 struct bpf_prog_stats __percpu *stats; 569 int __percpu *active; 570 unsigned int (*bpf_func)(const void *ctx, 571 const struct bpf_insn *insn); 572 struct bpf_prog_aux *aux; /* Auxiliary fields */ 573 struct sock_fprog_kern *orig_prog; /* Original BPF program */ 574 /* Instructions for interpreter */ 575 struct sock_filter insns[0]; 576 struct bpf_insn insnsi[]; 577 }; 578 579 struct sk_filter { 580 refcount_t refcnt; 581 struct rcu_head rcu; 582 struct bpf_prog *prog; 583 }; 584 585 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 586 587 #define __BPF_PROG_RUN(prog, ctx, dfunc) ({ \ 588 u32 __ret; \ 589 cant_migrate(); \ 590 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \ 591 struct bpf_prog_stats *__stats; \ 592 u64 __start = sched_clock(); \ 593 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \ 594 __stats = this_cpu_ptr(prog->stats); \ 595 u64_stats_update_begin(&__stats->syncp); \ 596 __stats->cnt++; \ 597 __stats->nsecs += sched_clock() - __start; \ 598 u64_stats_update_end(&__stats->syncp); \ 599 } else { \ 600 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \ 601 } \ 602 __ret; }) 603 604 #define BPF_PROG_RUN(prog, ctx) \ 605 __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func) 606 607 /* 608 * Use in preemptible and therefore migratable context to make sure that 609 * the execution of the BPF program runs on one CPU. 610 * 611 * This uses migrate_disable/enable() explicitly to document that the 612 * invocation of a BPF program does not require reentrancy protection 613 * against a BPF program which is invoked from a preempting task. 614 * 615 * For non RT enabled kernels migrate_disable/enable() maps to 616 * preempt_disable/enable(), i.e. it disables also preemption. 617 */ 618 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 619 const void *ctx) 620 { 621 u32 ret; 622 623 migrate_disable(); 624 ret = __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func); 625 migrate_enable(); 626 return ret; 627 } 628 629 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 630 631 struct bpf_skb_data_end { 632 struct qdisc_skb_cb qdisc_cb; 633 void *data_meta; 634 void *data_end; 635 }; 636 637 struct bpf_nh_params { 638 u32 nh_family; 639 union { 640 u32 ipv4_nh; 641 struct in6_addr ipv6_nh; 642 }; 643 }; 644 645 struct bpf_redirect_info { 646 u32 flags; 647 u32 tgt_index; 648 void *tgt_value; 649 struct bpf_map *map; 650 u32 map_id; 651 enum bpf_map_type map_type; 652 u32 kern_flags; 653 struct bpf_nh_params nh; 654 }; 655 656 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 657 658 /* flags for bpf_redirect_info kern_flags */ 659 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 660 661 /* Compute the linear packet data range [data, data_end) which 662 * will be accessed by various program types (cls_bpf, act_bpf, 663 * lwt, ...). Subsystems allowing direct data access must (!) 664 * ensure that cb[] area can be written to when BPF program is 665 * invoked (otherwise cb[] save/restore is necessary). 666 */ 667 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 668 { 669 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 670 671 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 672 cb->data_meta = skb->data - skb_metadata_len(skb); 673 cb->data_end = skb->data + skb_headlen(skb); 674 } 675 676 /* Similar to bpf_compute_data_pointers(), except that save orginal 677 * data in cb->data and cb->meta_data for restore. 678 */ 679 static inline void bpf_compute_and_save_data_end( 680 struct sk_buff *skb, void **saved_data_end) 681 { 682 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 683 684 *saved_data_end = cb->data_end; 685 cb->data_end = skb->data + skb_headlen(skb); 686 } 687 688 /* Restore data saved by bpf_compute_data_pointers(). */ 689 static inline void bpf_restore_data_end( 690 struct sk_buff *skb, void *saved_data_end) 691 { 692 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 693 694 cb->data_end = saved_data_end; 695 } 696 697 static inline u8 *bpf_skb_cb(struct sk_buff *skb) 698 { 699 /* eBPF programs may read/write skb->cb[] area to transfer meta 700 * data between tail calls. Since this also needs to work with 701 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 702 * 703 * In some socket filter cases, the cb unfortunately needs to be 704 * saved/restored so that protocol specific skb->cb[] data won't 705 * be lost. In any case, due to unpriviledged eBPF programs 706 * attached to sockets, we need to clear the bpf_skb_cb() area 707 * to not leak previous contents to user space. 708 */ 709 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 710 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 711 sizeof_field(struct qdisc_skb_cb, data)); 712 713 return qdisc_skb_cb(skb)->data; 714 } 715 716 /* Must be invoked with migration disabled */ 717 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 718 struct sk_buff *skb) 719 { 720 u8 *cb_data = bpf_skb_cb(skb); 721 u8 cb_saved[BPF_SKB_CB_LEN]; 722 u32 res; 723 724 if (unlikely(prog->cb_access)) { 725 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 726 memset(cb_data, 0, sizeof(cb_saved)); 727 } 728 729 res = BPF_PROG_RUN(prog, skb); 730 731 if (unlikely(prog->cb_access)) 732 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 733 734 return res; 735 } 736 737 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 738 struct sk_buff *skb) 739 { 740 u32 res; 741 742 migrate_disable(); 743 res = __bpf_prog_run_save_cb(prog, skb); 744 migrate_enable(); 745 return res; 746 } 747 748 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 749 struct sk_buff *skb) 750 { 751 u8 *cb_data = bpf_skb_cb(skb); 752 u32 res; 753 754 if (unlikely(prog->cb_access)) 755 memset(cb_data, 0, BPF_SKB_CB_LEN); 756 757 res = bpf_prog_run_pin_on_cpu(prog, skb); 758 return res; 759 } 760 761 DECLARE_BPF_DISPATCHER(xdp) 762 763 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 764 struct xdp_buff *xdp) 765 { 766 /* Driver XDP hooks are invoked within a single NAPI poll cycle and thus 767 * under local_bh_disable(), which provides the needed RCU protection 768 * for accessing map entries. 769 */ 770 return __BPF_PROG_RUN(prog, xdp, BPF_DISPATCHER_FUNC(xdp)); 771 } 772 773 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 774 775 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 776 { 777 return prog->len * sizeof(struct bpf_insn); 778 } 779 780 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 781 { 782 return round_up(bpf_prog_insn_size(prog) + 783 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 784 } 785 786 static inline unsigned int bpf_prog_size(unsigned int proglen) 787 { 788 return max(sizeof(struct bpf_prog), 789 offsetof(struct bpf_prog, insns[proglen])); 790 } 791 792 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 793 { 794 /* When classic BPF programs have been loaded and the arch 795 * does not have a classic BPF JIT (anymore), they have been 796 * converted via bpf_migrate_filter() to eBPF and thus always 797 * have an unspec program type. 798 */ 799 return prog->type == BPF_PROG_TYPE_UNSPEC; 800 } 801 802 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 803 { 804 const u32 size_machine = sizeof(unsigned long); 805 806 if (size > size_machine && size % size_machine == 0) 807 size = size_machine; 808 809 return size; 810 } 811 812 static inline bool 813 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 814 { 815 return size <= size_default && (size & (size - 1)) == 0; 816 } 817 818 static inline u8 819 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 820 { 821 u8 access_off = off & (size_default - 1); 822 823 #ifdef __LITTLE_ENDIAN 824 return access_off; 825 #else 826 return size_default - (access_off + size); 827 #endif 828 } 829 830 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 831 (size == sizeof(__u64) && \ 832 off >= offsetof(type, field) && \ 833 off + sizeof(__u64) <= offsetofend(type, field) && \ 834 off % sizeof(__u64) == 0) 835 836 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 837 838 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 839 { 840 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 841 if (!fp->jited) { 842 set_vm_flush_reset_perms(fp); 843 set_memory_ro((unsigned long)fp, fp->pages); 844 } 845 #endif 846 } 847 848 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 849 { 850 set_vm_flush_reset_perms(hdr); 851 set_memory_ro((unsigned long)hdr, hdr->pages); 852 set_memory_x((unsigned long)hdr, hdr->pages); 853 } 854 855 static inline struct bpf_binary_header * 856 bpf_jit_binary_hdr(const struct bpf_prog *fp) 857 { 858 unsigned long real_start = (unsigned long)fp->bpf_func; 859 unsigned long addr = real_start & PAGE_MASK; 860 861 return (void *)addr; 862 } 863 864 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 865 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 866 { 867 return sk_filter_trim_cap(sk, skb, 1); 868 } 869 870 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 871 void bpf_prog_free(struct bpf_prog *fp); 872 873 bool bpf_opcode_in_insntable(u8 code); 874 875 void bpf_prog_free_linfo(struct bpf_prog *prog); 876 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 877 const u32 *insn_to_jit_off); 878 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 879 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 880 881 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 882 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 883 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 884 gfp_t gfp_extra_flags); 885 void __bpf_prog_free(struct bpf_prog *fp); 886 887 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 888 { 889 __bpf_prog_free(fp); 890 } 891 892 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 893 unsigned int flen); 894 895 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 896 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 897 bpf_aux_classic_check_t trans, bool save_orig); 898 void bpf_prog_destroy(struct bpf_prog *fp); 899 900 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 901 int sk_attach_bpf(u32 ufd, struct sock *sk); 902 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 903 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 904 void sk_reuseport_prog_free(struct bpf_prog *prog); 905 int sk_detach_filter(struct sock *sk); 906 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, 907 unsigned int len); 908 909 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 910 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 911 912 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 913 #define __bpf_call_base_args \ 914 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 915 (void *)__bpf_call_base) 916 917 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 918 void bpf_jit_compile(struct bpf_prog *prog); 919 bool bpf_jit_needs_zext(void); 920 bool bpf_jit_supports_kfunc_call(void); 921 bool bpf_helper_changes_pkt_data(void *func); 922 923 static inline bool bpf_dump_raw_ok(const struct cred *cred) 924 { 925 /* Reconstruction of call-sites is dependent on kallsyms, 926 * thus make dump the same restriction. 927 */ 928 return kallsyms_show_value(cred); 929 } 930 931 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 932 const struct bpf_insn *patch, u32 len); 933 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 934 935 void bpf_clear_redirect_map(struct bpf_map *map); 936 937 static inline bool xdp_return_frame_no_direct(void) 938 { 939 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 940 941 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 942 } 943 944 static inline void xdp_set_return_frame_no_direct(void) 945 { 946 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 947 948 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 949 } 950 951 static inline void xdp_clear_return_frame_no_direct(void) 952 { 953 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 954 955 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 956 } 957 958 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 959 unsigned int pktlen) 960 { 961 unsigned int len; 962 963 if (unlikely(!(fwd->flags & IFF_UP))) 964 return -ENETDOWN; 965 966 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 967 if (pktlen > len) 968 return -EMSGSIZE; 969 970 return 0; 971 } 972 973 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 974 * same cpu context. Further for best results no more than a single map 975 * for the do_redirect/do_flush pair should be used. This limitation is 976 * because we only track one map and force a flush when the map changes. 977 * This does not appear to be a real limitation for existing software. 978 */ 979 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 980 struct xdp_buff *xdp, struct bpf_prog *prog); 981 int xdp_do_redirect(struct net_device *dev, 982 struct xdp_buff *xdp, 983 struct bpf_prog *prog); 984 void xdp_do_flush(void); 985 986 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as 987 * it is no longer only flushing maps. Keep this define for compatibility 988 * until all drivers are updated - do not use xdp_do_flush_map() in new code! 989 */ 990 #define xdp_do_flush_map xdp_do_flush 991 992 void bpf_warn_invalid_xdp_action(u32 act); 993 994 #ifdef CONFIG_INET 995 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 996 struct bpf_prog *prog, struct sk_buff *skb, 997 struct sock *migrating_sk, 998 u32 hash); 999 #else 1000 static inline struct sock * 1001 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1002 struct bpf_prog *prog, struct sk_buff *skb, 1003 struct sock *migrating_sk, 1004 u32 hash) 1005 { 1006 return NULL; 1007 } 1008 #endif 1009 1010 #ifdef CONFIG_BPF_JIT 1011 extern int bpf_jit_enable; 1012 extern int bpf_jit_harden; 1013 extern int bpf_jit_kallsyms; 1014 extern long bpf_jit_limit; 1015 1016 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1017 1018 struct bpf_binary_header * 1019 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1020 unsigned int alignment, 1021 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1022 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1023 u64 bpf_jit_alloc_exec_limit(void); 1024 void *bpf_jit_alloc_exec(unsigned long size); 1025 void bpf_jit_free_exec(void *addr); 1026 void bpf_jit_free(struct bpf_prog *fp); 1027 1028 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1029 struct bpf_jit_poke_descriptor *poke); 1030 1031 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1032 const struct bpf_insn *insn, bool extra_pass, 1033 u64 *func_addr, bool *func_addr_fixed); 1034 1035 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1036 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1037 1038 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1039 u32 pass, void *image) 1040 { 1041 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1042 proglen, pass, image, current->comm, task_pid_nr(current)); 1043 1044 if (image) 1045 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1046 16, 1, image, proglen, false); 1047 } 1048 1049 static inline bool bpf_jit_is_ebpf(void) 1050 { 1051 # ifdef CONFIG_HAVE_EBPF_JIT 1052 return true; 1053 # else 1054 return false; 1055 # endif 1056 } 1057 1058 static inline bool ebpf_jit_enabled(void) 1059 { 1060 return bpf_jit_enable && bpf_jit_is_ebpf(); 1061 } 1062 1063 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1064 { 1065 return fp->jited && bpf_jit_is_ebpf(); 1066 } 1067 1068 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1069 { 1070 /* These are the prerequisites, should someone ever have the 1071 * idea to call blinding outside of them, we make sure to 1072 * bail out. 1073 */ 1074 if (!bpf_jit_is_ebpf()) 1075 return false; 1076 if (!prog->jit_requested) 1077 return false; 1078 if (!bpf_jit_harden) 1079 return false; 1080 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) 1081 return false; 1082 1083 return true; 1084 } 1085 1086 static inline bool bpf_jit_kallsyms_enabled(void) 1087 { 1088 /* There are a couple of corner cases where kallsyms should 1089 * not be enabled f.e. on hardening. 1090 */ 1091 if (bpf_jit_harden) 1092 return false; 1093 if (!bpf_jit_kallsyms) 1094 return false; 1095 if (bpf_jit_kallsyms == 1) 1096 return true; 1097 1098 return false; 1099 } 1100 1101 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1102 unsigned long *off, char *sym); 1103 bool is_bpf_text_address(unsigned long addr); 1104 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1105 char *sym); 1106 1107 static inline const char * 1108 bpf_address_lookup(unsigned long addr, unsigned long *size, 1109 unsigned long *off, char **modname, char *sym) 1110 { 1111 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1112 1113 if (ret && modname) 1114 *modname = NULL; 1115 return ret; 1116 } 1117 1118 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1119 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1120 1121 #else /* CONFIG_BPF_JIT */ 1122 1123 static inline bool ebpf_jit_enabled(void) 1124 { 1125 return false; 1126 } 1127 1128 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1129 { 1130 return false; 1131 } 1132 1133 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1134 { 1135 return false; 1136 } 1137 1138 static inline int 1139 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1140 struct bpf_jit_poke_descriptor *poke) 1141 { 1142 return -ENOTSUPP; 1143 } 1144 1145 static inline void bpf_jit_free(struct bpf_prog *fp) 1146 { 1147 bpf_prog_unlock_free(fp); 1148 } 1149 1150 static inline bool bpf_jit_kallsyms_enabled(void) 1151 { 1152 return false; 1153 } 1154 1155 static inline const char * 1156 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1157 unsigned long *off, char *sym) 1158 { 1159 return NULL; 1160 } 1161 1162 static inline bool is_bpf_text_address(unsigned long addr) 1163 { 1164 return false; 1165 } 1166 1167 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1168 char *type, char *sym) 1169 { 1170 return -ERANGE; 1171 } 1172 1173 static inline const char * 1174 bpf_address_lookup(unsigned long addr, unsigned long *size, 1175 unsigned long *off, char **modname, char *sym) 1176 { 1177 return NULL; 1178 } 1179 1180 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1181 { 1182 } 1183 1184 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1185 { 1186 } 1187 1188 #endif /* CONFIG_BPF_JIT */ 1189 1190 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1191 1192 #define BPF_ANC BIT(15) 1193 1194 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1195 { 1196 switch (first->code) { 1197 case BPF_RET | BPF_K: 1198 case BPF_LD | BPF_W | BPF_LEN: 1199 return false; 1200 1201 case BPF_LD | BPF_W | BPF_ABS: 1202 case BPF_LD | BPF_H | BPF_ABS: 1203 case BPF_LD | BPF_B | BPF_ABS: 1204 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1205 return true; 1206 return false; 1207 1208 default: 1209 return true; 1210 } 1211 } 1212 1213 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1214 { 1215 BUG_ON(ftest->code & BPF_ANC); 1216 1217 switch (ftest->code) { 1218 case BPF_LD | BPF_W | BPF_ABS: 1219 case BPF_LD | BPF_H | BPF_ABS: 1220 case BPF_LD | BPF_B | BPF_ABS: 1221 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1222 return BPF_ANC | SKF_AD_##CODE 1223 switch (ftest->k) { 1224 BPF_ANCILLARY(PROTOCOL); 1225 BPF_ANCILLARY(PKTTYPE); 1226 BPF_ANCILLARY(IFINDEX); 1227 BPF_ANCILLARY(NLATTR); 1228 BPF_ANCILLARY(NLATTR_NEST); 1229 BPF_ANCILLARY(MARK); 1230 BPF_ANCILLARY(QUEUE); 1231 BPF_ANCILLARY(HATYPE); 1232 BPF_ANCILLARY(RXHASH); 1233 BPF_ANCILLARY(CPU); 1234 BPF_ANCILLARY(ALU_XOR_X); 1235 BPF_ANCILLARY(VLAN_TAG); 1236 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1237 BPF_ANCILLARY(PAY_OFFSET); 1238 BPF_ANCILLARY(RANDOM); 1239 BPF_ANCILLARY(VLAN_TPID); 1240 } 1241 fallthrough; 1242 default: 1243 return ftest->code; 1244 } 1245 } 1246 1247 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1248 int k, unsigned int size); 1249 1250 static inline int bpf_tell_extensions(void) 1251 { 1252 return SKF_AD_MAX; 1253 } 1254 1255 struct bpf_sock_addr_kern { 1256 struct sock *sk; 1257 struct sockaddr *uaddr; 1258 /* Temporary "register" to make indirect stores to nested structures 1259 * defined above. We need three registers to make such a store, but 1260 * only two (src and dst) are available at convert_ctx_access time 1261 */ 1262 u64 tmp_reg; 1263 void *t_ctx; /* Attach type specific context. */ 1264 }; 1265 1266 struct bpf_sock_ops_kern { 1267 struct sock *sk; 1268 union { 1269 u32 args[4]; 1270 u32 reply; 1271 u32 replylong[4]; 1272 }; 1273 struct sk_buff *syn_skb; 1274 struct sk_buff *skb; 1275 void *skb_data_end; 1276 u8 op; 1277 u8 is_fullsock; 1278 u8 remaining_opt_len; 1279 u64 temp; /* temp and everything after is not 1280 * initialized to 0 before calling 1281 * the BPF program. New fields that 1282 * should be initialized to 0 should 1283 * be inserted before temp. 1284 * temp is scratch storage used by 1285 * sock_ops_convert_ctx_access 1286 * as temporary storage of a register. 1287 */ 1288 }; 1289 1290 struct bpf_sysctl_kern { 1291 struct ctl_table_header *head; 1292 struct ctl_table *table; 1293 void *cur_val; 1294 size_t cur_len; 1295 void *new_val; 1296 size_t new_len; 1297 int new_updated; 1298 int write; 1299 loff_t *ppos; 1300 /* Temporary "register" for indirect stores to ppos. */ 1301 u64 tmp_reg; 1302 }; 1303 1304 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1305 struct bpf_sockopt_buf { 1306 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1307 }; 1308 1309 struct bpf_sockopt_kern { 1310 struct sock *sk; 1311 u8 *optval; 1312 u8 *optval_end; 1313 s32 level; 1314 s32 optname; 1315 s32 optlen; 1316 s32 retval; 1317 }; 1318 1319 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1320 1321 struct bpf_sk_lookup_kern { 1322 u16 family; 1323 u16 protocol; 1324 __be16 sport; 1325 u16 dport; 1326 struct { 1327 __be32 saddr; 1328 __be32 daddr; 1329 } v4; 1330 struct { 1331 const struct in6_addr *saddr; 1332 const struct in6_addr *daddr; 1333 } v6; 1334 struct sock *selected_sk; 1335 bool no_reuseport; 1336 }; 1337 1338 extern struct static_key_false bpf_sk_lookup_enabled; 1339 1340 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1341 * 1342 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1343 * SK_DROP. Their meaning is as follows: 1344 * 1345 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1346 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1347 * SK_DROP : terminate lookup with -ECONNREFUSED 1348 * 1349 * This macro aggregates return values and selected sockets from 1350 * multiple BPF programs according to following rules in order: 1351 * 1352 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1353 * macro result is SK_PASS and last ctx.selected_sk is used. 1354 * 2. If any program returned SK_DROP return value, 1355 * macro result is SK_DROP. 1356 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1357 * 1358 * Caller must ensure that the prog array is non-NULL, and that the 1359 * array as well as the programs it contains remain valid. 1360 */ 1361 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1362 ({ \ 1363 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1364 struct bpf_prog_array_item *_item; \ 1365 struct sock *_selected_sk = NULL; \ 1366 bool _no_reuseport = false; \ 1367 struct bpf_prog *_prog; \ 1368 bool _all_pass = true; \ 1369 u32 _ret; \ 1370 \ 1371 migrate_disable(); \ 1372 _item = &(array)->items[0]; \ 1373 while ((_prog = READ_ONCE(_item->prog))) { \ 1374 /* restore most recent selection */ \ 1375 _ctx->selected_sk = _selected_sk; \ 1376 _ctx->no_reuseport = _no_reuseport; \ 1377 \ 1378 _ret = func(_prog, _ctx); \ 1379 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1380 /* remember last non-NULL socket */ \ 1381 _selected_sk = _ctx->selected_sk; \ 1382 _no_reuseport = _ctx->no_reuseport; \ 1383 } else if (_ret == SK_DROP && _all_pass) { \ 1384 _all_pass = false; \ 1385 } \ 1386 _item++; \ 1387 } \ 1388 _ctx->selected_sk = _selected_sk; \ 1389 _ctx->no_reuseport = _no_reuseport; \ 1390 migrate_enable(); \ 1391 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1392 }) 1393 1394 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1395 const __be32 saddr, const __be16 sport, 1396 const __be32 daddr, const u16 dport, 1397 struct sock **psk) 1398 { 1399 struct bpf_prog_array *run_array; 1400 struct sock *selected_sk = NULL; 1401 bool no_reuseport = false; 1402 1403 rcu_read_lock(); 1404 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1405 if (run_array) { 1406 struct bpf_sk_lookup_kern ctx = { 1407 .family = AF_INET, 1408 .protocol = protocol, 1409 .v4.saddr = saddr, 1410 .v4.daddr = daddr, 1411 .sport = sport, 1412 .dport = dport, 1413 }; 1414 u32 act; 1415 1416 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN); 1417 if (act == SK_PASS) { 1418 selected_sk = ctx.selected_sk; 1419 no_reuseport = ctx.no_reuseport; 1420 } else { 1421 selected_sk = ERR_PTR(-ECONNREFUSED); 1422 } 1423 } 1424 rcu_read_unlock(); 1425 *psk = selected_sk; 1426 return no_reuseport; 1427 } 1428 1429 #if IS_ENABLED(CONFIG_IPV6) 1430 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1431 const struct in6_addr *saddr, 1432 const __be16 sport, 1433 const struct in6_addr *daddr, 1434 const u16 dport, 1435 struct sock **psk) 1436 { 1437 struct bpf_prog_array *run_array; 1438 struct sock *selected_sk = NULL; 1439 bool no_reuseport = false; 1440 1441 rcu_read_lock(); 1442 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1443 if (run_array) { 1444 struct bpf_sk_lookup_kern ctx = { 1445 .family = AF_INET6, 1446 .protocol = protocol, 1447 .v6.saddr = saddr, 1448 .v6.daddr = daddr, 1449 .sport = sport, 1450 .dport = dport, 1451 }; 1452 u32 act; 1453 1454 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN); 1455 if (act == SK_PASS) { 1456 selected_sk = ctx.selected_sk; 1457 no_reuseport = ctx.no_reuseport; 1458 } else { 1459 selected_sk = ERR_PTR(-ECONNREFUSED); 1460 } 1461 } 1462 rcu_read_unlock(); 1463 *psk = selected_sk; 1464 return no_reuseport; 1465 } 1466 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1467 1468 static __always_inline int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex, 1469 u64 flags, const u64 flag_mask, 1470 void *lookup_elem(struct bpf_map *map, u32 key)) 1471 { 1472 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1473 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1474 1475 /* Lower bits of the flags are used as return code on lookup failure */ 1476 if (unlikely(flags & ~(action_mask | flag_mask))) 1477 return XDP_ABORTED; 1478 1479 ri->tgt_value = lookup_elem(map, ifindex); 1480 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1481 /* If the lookup fails we want to clear out the state in the 1482 * redirect_info struct completely, so that if an eBPF program 1483 * performs multiple lookups, the last one always takes 1484 * precedence. 1485 */ 1486 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1487 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1488 return flags & action_mask; 1489 } 1490 1491 ri->tgt_index = ifindex; 1492 ri->map_id = map->id; 1493 ri->map_type = map->map_type; 1494 1495 if (flags & BPF_F_BROADCAST) { 1496 WRITE_ONCE(ri->map, map); 1497 ri->flags = flags; 1498 } else { 1499 WRITE_ONCE(ri->map, NULL); 1500 ri->flags = 0; 1501 } 1502 1503 return XDP_REDIRECT; 1504 } 1505 1506 #endif /* __LINUX_FILTER_H__ */ 1507