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_OFF(OP, DST, IMM, OFF) \ 121 ((struct bpf_insn) { \ 122 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 123 .dst_reg = DST, \ 124 .src_reg = 0, \ 125 .off = OFF, \ 126 .imm = IMM }) 127 #define BPF_ALU64_IMM(OP, DST, IMM) \ 128 BPF_ALU64_IMM_OFF(OP, DST, IMM, 0) 129 130 #define BPF_ALU32_IMM_OFF(OP, DST, IMM, OFF) \ 131 ((struct bpf_insn) { \ 132 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 133 .dst_reg = DST, \ 134 .src_reg = 0, \ 135 .off = OFF, \ 136 .imm = IMM }) 137 #define BPF_ALU32_IMM(OP, DST, IMM) \ 138 BPF_ALU32_IMM_OFF(OP, DST, IMM, 0) 139 140 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 141 142 #define BPF_ENDIAN(TYPE, DST, LEN) \ 143 ((struct bpf_insn) { \ 144 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 145 .dst_reg = DST, \ 146 .src_reg = 0, \ 147 .off = 0, \ 148 .imm = LEN }) 149 150 /* Byte Swap, bswap16/32/64 */ 151 152 #define BPF_BSWAP(DST, LEN) \ 153 ((struct bpf_insn) { \ 154 .code = BPF_ALU64 | BPF_END | BPF_SRC(BPF_TO_LE), \ 155 .dst_reg = DST, \ 156 .src_reg = 0, \ 157 .off = 0, \ 158 .imm = LEN }) 159 160 /* Short form of mov, dst_reg = src_reg */ 161 162 #define BPF_MOV64_REG(DST, SRC) \ 163 ((struct bpf_insn) { \ 164 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 165 .dst_reg = DST, \ 166 .src_reg = SRC, \ 167 .off = 0, \ 168 .imm = 0 }) 169 170 #define BPF_MOV32_REG(DST, SRC) \ 171 ((struct bpf_insn) { \ 172 .code = BPF_ALU | BPF_MOV | BPF_X, \ 173 .dst_reg = DST, \ 174 .src_reg = SRC, \ 175 .off = 0, \ 176 .imm = 0 }) 177 178 /* Short form of mov, dst_reg = imm32 */ 179 180 #define BPF_MOV64_IMM(DST, IMM) \ 181 ((struct bpf_insn) { \ 182 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 183 .dst_reg = DST, \ 184 .src_reg = 0, \ 185 .off = 0, \ 186 .imm = IMM }) 187 188 #define BPF_MOV32_IMM(DST, IMM) \ 189 ((struct bpf_insn) { \ 190 .code = BPF_ALU | BPF_MOV | BPF_K, \ 191 .dst_reg = DST, \ 192 .src_reg = 0, \ 193 .off = 0, \ 194 .imm = IMM }) 195 196 /* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */ 197 198 #define BPF_MOVSX64_REG(DST, SRC, OFF) \ 199 ((struct bpf_insn) { \ 200 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 201 .dst_reg = DST, \ 202 .src_reg = SRC, \ 203 .off = OFF, \ 204 .imm = 0 }) 205 206 #define BPF_MOVSX32_REG(DST, SRC, OFF) \ 207 ((struct bpf_insn) { \ 208 .code = BPF_ALU | BPF_MOV | BPF_X, \ 209 .dst_reg = DST, \ 210 .src_reg = SRC, \ 211 .off = OFF, \ 212 .imm = 0 }) 213 214 /* Special form of mov32, used for doing explicit zero extension on dst. */ 215 #define BPF_ZEXT_REG(DST) \ 216 ((struct bpf_insn) { \ 217 .code = BPF_ALU | BPF_MOV | BPF_X, \ 218 .dst_reg = DST, \ 219 .src_reg = DST, \ 220 .off = 0, \ 221 .imm = 1 }) 222 223 static inline bool insn_is_zext(const struct bpf_insn *insn) 224 { 225 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 226 } 227 228 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 229 #define BPF_LD_IMM64(DST, IMM) \ 230 BPF_LD_IMM64_RAW(DST, 0, IMM) 231 232 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 233 ((struct bpf_insn) { \ 234 .code = BPF_LD | BPF_DW | BPF_IMM, \ 235 .dst_reg = DST, \ 236 .src_reg = SRC, \ 237 .off = 0, \ 238 .imm = (__u32) (IMM) }), \ 239 ((struct bpf_insn) { \ 240 .code = 0, /* zero is reserved opcode */ \ 241 .dst_reg = 0, \ 242 .src_reg = 0, \ 243 .off = 0, \ 244 .imm = ((__u64) (IMM)) >> 32 }) 245 246 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 247 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 248 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 249 250 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 251 252 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 253 ((struct bpf_insn) { \ 254 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 255 .dst_reg = DST, \ 256 .src_reg = SRC, \ 257 .off = 0, \ 258 .imm = IMM }) 259 260 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 261 ((struct bpf_insn) { \ 262 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 263 .dst_reg = DST, \ 264 .src_reg = SRC, \ 265 .off = 0, \ 266 .imm = IMM }) 267 268 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 269 270 #define BPF_LD_ABS(SIZE, IMM) \ 271 ((struct bpf_insn) { \ 272 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 273 .dst_reg = 0, \ 274 .src_reg = 0, \ 275 .off = 0, \ 276 .imm = IMM }) 277 278 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 279 280 #define BPF_LD_IND(SIZE, SRC, IMM) \ 281 ((struct bpf_insn) { \ 282 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 283 .dst_reg = 0, \ 284 .src_reg = SRC, \ 285 .off = 0, \ 286 .imm = IMM }) 287 288 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 289 290 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 291 ((struct bpf_insn) { \ 292 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 293 .dst_reg = DST, \ 294 .src_reg = SRC, \ 295 .off = OFF, \ 296 .imm = 0 }) 297 298 /* Memory load, dst_reg = *(signed size *) (src_reg + off16) */ 299 300 #define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \ 301 ((struct bpf_insn) { \ 302 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \ 303 .dst_reg = DST, \ 304 .src_reg = SRC, \ 305 .off = OFF, \ 306 .imm = 0 }) 307 308 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 309 310 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 311 ((struct bpf_insn) { \ 312 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 313 .dst_reg = DST, \ 314 .src_reg = SRC, \ 315 .off = OFF, \ 316 .imm = 0 }) 317 318 319 /* 320 * Atomic operations: 321 * 322 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 323 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 324 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 325 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 326 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 327 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 328 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 329 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 330 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 331 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 332 */ 333 334 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 335 ((struct bpf_insn) { \ 336 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 337 .dst_reg = DST, \ 338 .src_reg = SRC, \ 339 .off = OFF, \ 340 .imm = OP }) 341 342 /* Legacy alias */ 343 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 344 345 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 346 347 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 348 ((struct bpf_insn) { \ 349 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 350 .dst_reg = DST, \ 351 .src_reg = 0, \ 352 .off = OFF, \ 353 .imm = IMM }) 354 355 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 356 357 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 358 ((struct bpf_insn) { \ 359 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 360 .dst_reg = DST, \ 361 .src_reg = SRC, \ 362 .off = OFF, \ 363 .imm = 0 }) 364 365 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 366 367 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 368 ((struct bpf_insn) { \ 369 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 370 .dst_reg = DST, \ 371 .src_reg = 0, \ 372 .off = OFF, \ 373 .imm = IMM }) 374 375 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 376 377 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 378 ((struct bpf_insn) { \ 379 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 380 .dst_reg = DST, \ 381 .src_reg = SRC, \ 382 .off = OFF, \ 383 .imm = 0 }) 384 385 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 386 387 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 388 ((struct bpf_insn) { \ 389 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 390 .dst_reg = DST, \ 391 .src_reg = 0, \ 392 .off = OFF, \ 393 .imm = IMM }) 394 395 /* Unconditional jumps, goto pc + off16 */ 396 397 #define BPF_JMP_A(OFF) \ 398 ((struct bpf_insn) { \ 399 .code = BPF_JMP | BPF_JA, \ 400 .dst_reg = 0, \ 401 .src_reg = 0, \ 402 .off = OFF, \ 403 .imm = 0 }) 404 405 /* Relative call */ 406 407 #define BPF_CALL_REL(TGT) \ 408 ((struct bpf_insn) { \ 409 .code = BPF_JMP | BPF_CALL, \ 410 .dst_reg = 0, \ 411 .src_reg = BPF_PSEUDO_CALL, \ 412 .off = 0, \ 413 .imm = TGT }) 414 415 /* Convert function address to BPF immediate */ 416 417 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base) 418 419 #define BPF_EMIT_CALL(FUNC) \ 420 ((struct bpf_insn) { \ 421 .code = BPF_JMP | BPF_CALL, \ 422 .dst_reg = 0, \ 423 .src_reg = 0, \ 424 .off = 0, \ 425 .imm = BPF_CALL_IMM(FUNC) }) 426 427 /* Raw code statement block */ 428 429 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 430 ((struct bpf_insn) { \ 431 .code = CODE, \ 432 .dst_reg = DST, \ 433 .src_reg = SRC, \ 434 .off = OFF, \ 435 .imm = IMM }) 436 437 /* Program exit */ 438 439 #define BPF_EXIT_INSN() \ 440 ((struct bpf_insn) { \ 441 .code = BPF_JMP | BPF_EXIT, \ 442 .dst_reg = 0, \ 443 .src_reg = 0, \ 444 .off = 0, \ 445 .imm = 0 }) 446 447 /* Speculation barrier */ 448 449 #define BPF_ST_NOSPEC() \ 450 ((struct bpf_insn) { \ 451 .code = BPF_ST | BPF_NOSPEC, \ 452 .dst_reg = 0, \ 453 .src_reg = 0, \ 454 .off = 0, \ 455 .imm = 0 }) 456 457 /* Internal classic blocks for direct assignment */ 458 459 #define __BPF_STMT(CODE, K) \ 460 ((struct sock_filter) BPF_STMT(CODE, K)) 461 462 #define __BPF_JUMP(CODE, K, JT, JF) \ 463 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 464 465 #define bytes_to_bpf_size(bytes) \ 466 ({ \ 467 int bpf_size = -EINVAL; \ 468 \ 469 if (bytes == sizeof(u8)) \ 470 bpf_size = BPF_B; \ 471 else if (bytes == sizeof(u16)) \ 472 bpf_size = BPF_H; \ 473 else if (bytes == sizeof(u32)) \ 474 bpf_size = BPF_W; \ 475 else if (bytes == sizeof(u64)) \ 476 bpf_size = BPF_DW; \ 477 \ 478 bpf_size; \ 479 }) 480 481 #define bpf_size_to_bytes(bpf_size) \ 482 ({ \ 483 int bytes = -EINVAL; \ 484 \ 485 if (bpf_size == BPF_B) \ 486 bytes = sizeof(u8); \ 487 else if (bpf_size == BPF_H) \ 488 bytes = sizeof(u16); \ 489 else if (bpf_size == BPF_W) \ 490 bytes = sizeof(u32); \ 491 else if (bpf_size == BPF_DW) \ 492 bytes = sizeof(u64); \ 493 \ 494 bytes; \ 495 }) 496 497 #define BPF_SIZEOF(type) \ 498 ({ \ 499 const int __size = bytes_to_bpf_size(sizeof(type)); \ 500 BUILD_BUG_ON(__size < 0); \ 501 __size; \ 502 }) 503 504 #define BPF_FIELD_SIZEOF(type, field) \ 505 ({ \ 506 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 507 BUILD_BUG_ON(__size < 0); \ 508 __size; \ 509 }) 510 511 #define BPF_LDST_BYTES(insn) \ 512 ({ \ 513 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 514 WARN_ON(__size < 0); \ 515 __size; \ 516 }) 517 518 #define __BPF_MAP_0(m, v, ...) v 519 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 520 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 521 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 522 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 523 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 524 525 #define __BPF_REG_0(...) __BPF_PAD(5) 526 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 527 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 528 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 529 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 530 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 531 532 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 533 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 534 535 #define __BPF_CAST(t, a) \ 536 (__force t) \ 537 (__force \ 538 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 539 (unsigned long)0, (t)0))) a 540 #define __BPF_V void 541 #define __BPF_N 542 543 #define __BPF_DECL_ARGS(t, a) t a 544 #define __BPF_DECL_REGS(t, a) u64 a 545 546 #define __BPF_PAD(n) \ 547 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 548 u64, __ur_3, u64, __ur_4, u64, __ur_5) 549 550 #define BPF_CALL_x(x, attr, name, ...) \ 551 static __always_inline \ 552 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 553 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 554 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 555 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 556 { \ 557 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 558 } \ 559 static __always_inline \ 560 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 561 562 #define __NOATTR 563 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__) 564 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__) 565 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__) 566 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__) 567 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__) 568 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__) 569 570 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__) 571 572 #define bpf_ctx_range(TYPE, MEMBER) \ 573 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 574 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 575 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 576 #if BITS_PER_LONG == 64 577 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 578 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 579 #else 580 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 581 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 582 #endif /* BITS_PER_LONG == 64 */ 583 584 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 585 ({ \ 586 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 587 *(PTR_SIZE) = (SIZE); \ 588 offsetof(TYPE, MEMBER); \ 589 }) 590 591 /* A struct sock_filter is architecture independent. */ 592 struct compat_sock_fprog { 593 u16 len; 594 compat_uptr_t filter; /* struct sock_filter * */ 595 }; 596 597 struct sock_fprog_kern { 598 u16 len; 599 struct sock_filter *filter; 600 }; 601 602 /* Some arches need doubleword alignment for their instructions and/or data */ 603 #define BPF_IMAGE_ALIGNMENT 8 604 605 struct bpf_binary_header { 606 u32 size; 607 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 608 }; 609 610 struct bpf_prog_stats { 611 u64_stats_t cnt; 612 u64_stats_t nsecs; 613 u64_stats_t misses; 614 struct u64_stats_sync syncp; 615 } __aligned(2 * sizeof(u64)); 616 617 struct sk_filter { 618 refcount_t refcnt; 619 struct rcu_head rcu; 620 struct bpf_prog *prog; 621 }; 622 623 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 624 625 extern struct mutex nf_conn_btf_access_lock; 626 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 627 const struct bpf_reg_state *reg, 628 int off, int size); 629 630 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 631 const struct bpf_insn *insnsi, 632 unsigned int (*bpf_func)(const void *, 633 const struct bpf_insn *)); 634 635 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 636 const void *ctx, 637 bpf_dispatcher_fn dfunc) 638 { 639 u32 ret; 640 641 cant_migrate(); 642 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 643 struct bpf_prog_stats *stats; 644 u64 start = sched_clock(); 645 unsigned long flags; 646 647 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 648 stats = this_cpu_ptr(prog->stats); 649 flags = u64_stats_update_begin_irqsave(&stats->syncp); 650 u64_stats_inc(&stats->cnt); 651 u64_stats_add(&stats->nsecs, sched_clock() - start); 652 u64_stats_update_end_irqrestore(&stats->syncp, flags); 653 } else { 654 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 655 } 656 return ret; 657 } 658 659 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 660 { 661 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 662 } 663 664 /* 665 * Use in preemptible and therefore migratable context to make sure that 666 * the execution of the BPF program runs on one CPU. 667 * 668 * This uses migrate_disable/enable() explicitly to document that the 669 * invocation of a BPF program does not require reentrancy protection 670 * against a BPF program which is invoked from a preempting task. 671 */ 672 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 673 const void *ctx) 674 { 675 u32 ret; 676 677 migrate_disable(); 678 ret = bpf_prog_run(prog, ctx); 679 migrate_enable(); 680 return ret; 681 } 682 683 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 684 685 struct bpf_skb_data_end { 686 struct qdisc_skb_cb qdisc_cb; 687 void *data_meta; 688 void *data_end; 689 }; 690 691 struct bpf_nh_params { 692 u32 nh_family; 693 union { 694 u32 ipv4_nh; 695 struct in6_addr ipv6_nh; 696 }; 697 }; 698 699 struct bpf_redirect_info { 700 u64 tgt_index; 701 void *tgt_value; 702 struct bpf_map *map; 703 u32 flags; 704 u32 kern_flags; 705 u32 map_id; 706 enum bpf_map_type map_type; 707 struct bpf_nh_params nh; 708 }; 709 710 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 711 712 /* flags for bpf_redirect_info kern_flags */ 713 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 714 715 /* Compute the linear packet data range [data, data_end) which 716 * will be accessed by various program types (cls_bpf, act_bpf, 717 * lwt, ...). Subsystems allowing direct data access must (!) 718 * ensure that cb[] area can be written to when BPF program is 719 * invoked (otherwise cb[] save/restore is necessary). 720 */ 721 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 722 { 723 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 724 725 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 726 cb->data_meta = skb->data - skb_metadata_len(skb); 727 cb->data_end = skb->data + skb_headlen(skb); 728 } 729 730 /* Similar to bpf_compute_data_pointers(), except that save orginal 731 * data in cb->data and cb->meta_data for restore. 732 */ 733 static inline void bpf_compute_and_save_data_end( 734 struct sk_buff *skb, void **saved_data_end) 735 { 736 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 737 738 *saved_data_end = cb->data_end; 739 cb->data_end = skb->data + skb_headlen(skb); 740 } 741 742 /* Restore data saved by bpf_compute_and_save_data_end(). */ 743 static inline void bpf_restore_data_end( 744 struct sk_buff *skb, void *saved_data_end) 745 { 746 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 747 748 cb->data_end = saved_data_end; 749 } 750 751 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 752 { 753 /* eBPF programs may read/write skb->cb[] area to transfer meta 754 * data between tail calls. Since this also needs to work with 755 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 756 * 757 * In some socket filter cases, the cb unfortunately needs to be 758 * saved/restored so that protocol specific skb->cb[] data won't 759 * be lost. In any case, due to unpriviledged eBPF programs 760 * attached to sockets, we need to clear the bpf_skb_cb() area 761 * to not leak previous contents to user space. 762 */ 763 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 764 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 765 sizeof_field(struct qdisc_skb_cb, data)); 766 767 return qdisc_skb_cb(skb)->data; 768 } 769 770 /* Must be invoked with migration disabled */ 771 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 772 const void *ctx) 773 { 774 const struct sk_buff *skb = ctx; 775 u8 *cb_data = bpf_skb_cb(skb); 776 u8 cb_saved[BPF_SKB_CB_LEN]; 777 u32 res; 778 779 if (unlikely(prog->cb_access)) { 780 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 781 memset(cb_data, 0, sizeof(cb_saved)); 782 } 783 784 res = bpf_prog_run(prog, skb); 785 786 if (unlikely(prog->cb_access)) 787 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 788 789 return res; 790 } 791 792 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 793 struct sk_buff *skb) 794 { 795 u32 res; 796 797 migrate_disable(); 798 res = __bpf_prog_run_save_cb(prog, skb); 799 migrate_enable(); 800 return res; 801 } 802 803 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 804 struct sk_buff *skb) 805 { 806 u8 *cb_data = bpf_skb_cb(skb); 807 u32 res; 808 809 if (unlikely(prog->cb_access)) 810 memset(cb_data, 0, BPF_SKB_CB_LEN); 811 812 res = bpf_prog_run_pin_on_cpu(prog, skb); 813 return res; 814 } 815 816 DECLARE_BPF_DISPATCHER(xdp) 817 818 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 819 820 u32 xdp_master_redirect(struct xdp_buff *xdp); 821 822 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 823 824 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 825 { 826 return prog->len * sizeof(struct bpf_insn); 827 } 828 829 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 830 { 831 return round_up(bpf_prog_insn_size(prog) + 832 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 833 } 834 835 static inline unsigned int bpf_prog_size(unsigned int proglen) 836 { 837 return max(sizeof(struct bpf_prog), 838 offsetof(struct bpf_prog, insns[proglen])); 839 } 840 841 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 842 { 843 /* When classic BPF programs have been loaded and the arch 844 * does not have a classic BPF JIT (anymore), they have been 845 * converted via bpf_migrate_filter() to eBPF and thus always 846 * have an unspec program type. 847 */ 848 return prog->type == BPF_PROG_TYPE_UNSPEC; 849 } 850 851 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 852 { 853 const u32 size_machine = sizeof(unsigned long); 854 855 if (size > size_machine && size % size_machine == 0) 856 size = size_machine; 857 858 return size; 859 } 860 861 static inline bool 862 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 863 { 864 return size <= size_default && (size & (size - 1)) == 0; 865 } 866 867 static inline u8 868 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 869 { 870 u8 access_off = off & (size_default - 1); 871 872 #ifdef __LITTLE_ENDIAN 873 return access_off; 874 #else 875 return size_default - (access_off + size); 876 #endif 877 } 878 879 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 880 (size == sizeof(__u64) && \ 881 off >= offsetof(type, field) && \ 882 off + sizeof(__u64) <= offsetofend(type, field) && \ 883 off % sizeof(__u64) == 0) 884 885 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 886 887 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 888 { 889 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 890 if (!fp->jited) { 891 set_vm_flush_reset_perms(fp); 892 set_memory_ro((unsigned long)fp, fp->pages); 893 } 894 #endif 895 } 896 897 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 898 { 899 set_vm_flush_reset_perms(hdr); 900 set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT); 901 } 902 903 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 904 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 905 { 906 return sk_filter_trim_cap(sk, skb, 1); 907 } 908 909 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 910 void bpf_prog_free(struct bpf_prog *fp); 911 912 bool bpf_opcode_in_insntable(u8 code); 913 914 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 915 const u32 *insn_to_jit_off); 916 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 917 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 918 919 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 920 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 921 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 922 gfp_t gfp_extra_flags); 923 void __bpf_prog_free(struct bpf_prog *fp); 924 925 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 926 { 927 __bpf_prog_free(fp); 928 } 929 930 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 931 unsigned int flen); 932 933 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 934 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 935 bpf_aux_classic_check_t trans, bool save_orig); 936 void bpf_prog_destroy(struct bpf_prog *fp); 937 938 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 939 int sk_attach_bpf(u32 ufd, struct sock *sk); 940 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 941 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 942 void sk_reuseport_prog_free(struct bpf_prog *prog); 943 int sk_detach_filter(struct sock *sk); 944 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); 945 946 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 947 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 948 949 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 950 #define __bpf_call_base_args \ 951 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 952 (void *)__bpf_call_base) 953 954 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 955 void bpf_jit_compile(struct bpf_prog *prog); 956 bool bpf_jit_needs_zext(void); 957 bool bpf_jit_supports_subprog_tailcalls(void); 958 bool bpf_jit_supports_kfunc_call(void); 959 bool bpf_jit_supports_far_kfunc_call(void); 960 bool bpf_jit_supports_exceptions(void); 961 bool bpf_jit_supports_ptr_xchg(void); 962 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie); 963 bool bpf_helper_changes_pkt_data(void *func); 964 965 static inline bool bpf_dump_raw_ok(const struct cred *cred) 966 { 967 /* Reconstruction of call-sites is dependent on kallsyms, 968 * thus make dump the same restriction. 969 */ 970 return kallsyms_show_value(cred); 971 } 972 973 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 974 const struct bpf_insn *patch, u32 len); 975 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 976 977 void bpf_clear_redirect_map(struct bpf_map *map); 978 979 static inline bool xdp_return_frame_no_direct(void) 980 { 981 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 982 983 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 984 } 985 986 static inline void xdp_set_return_frame_no_direct(void) 987 { 988 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 989 990 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 991 } 992 993 static inline void xdp_clear_return_frame_no_direct(void) 994 { 995 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 996 997 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 998 } 999 1000 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 1001 unsigned int pktlen) 1002 { 1003 unsigned int len; 1004 1005 if (unlikely(!(fwd->flags & IFF_UP))) 1006 return -ENETDOWN; 1007 1008 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 1009 if (pktlen > len) 1010 return -EMSGSIZE; 1011 1012 return 0; 1013 } 1014 1015 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 1016 * same cpu context. Further for best results no more than a single map 1017 * for the do_redirect/do_flush pair should be used. This limitation is 1018 * because we only track one map and force a flush when the map changes. 1019 * This does not appear to be a real limitation for existing software. 1020 */ 1021 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 1022 struct xdp_buff *xdp, struct bpf_prog *prog); 1023 int xdp_do_redirect(struct net_device *dev, 1024 struct xdp_buff *xdp, 1025 struct bpf_prog *prog); 1026 int xdp_do_redirect_frame(struct net_device *dev, 1027 struct xdp_buff *xdp, 1028 struct xdp_frame *xdpf, 1029 struct bpf_prog *prog); 1030 void xdp_do_flush(void); 1031 1032 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act); 1033 1034 #ifdef CONFIG_INET 1035 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1036 struct bpf_prog *prog, struct sk_buff *skb, 1037 struct sock *migrating_sk, 1038 u32 hash); 1039 #else 1040 static inline struct sock * 1041 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1042 struct bpf_prog *prog, struct sk_buff *skb, 1043 struct sock *migrating_sk, 1044 u32 hash) 1045 { 1046 return NULL; 1047 } 1048 #endif 1049 1050 #ifdef CONFIG_BPF_JIT 1051 extern int bpf_jit_enable; 1052 extern int bpf_jit_harden; 1053 extern int bpf_jit_kallsyms; 1054 extern long bpf_jit_limit; 1055 extern long bpf_jit_limit_max; 1056 1057 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1058 1059 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); 1060 1061 struct bpf_binary_header * 1062 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1063 unsigned int alignment, 1064 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1065 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1066 u64 bpf_jit_alloc_exec_limit(void); 1067 void *bpf_jit_alloc_exec(unsigned long size); 1068 void bpf_jit_free_exec(void *addr); 1069 void bpf_jit_free(struct bpf_prog *fp); 1070 struct bpf_binary_header * 1071 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); 1072 1073 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); 1074 void bpf_prog_pack_free(void *ptr, u32 size); 1075 1076 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 1077 { 1078 return list_empty(&fp->aux->ksym.lnode) || 1079 fp->aux->ksym.lnode.prev == LIST_POISON2; 1080 } 1081 1082 struct bpf_binary_header * 1083 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, 1084 unsigned int alignment, 1085 struct bpf_binary_header **rw_hdr, 1086 u8 **rw_image, 1087 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1088 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, 1089 struct bpf_binary_header *ro_header, 1090 struct bpf_binary_header *rw_header); 1091 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, 1092 struct bpf_binary_header *rw_header); 1093 1094 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1095 struct bpf_jit_poke_descriptor *poke); 1096 1097 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1098 const struct bpf_insn *insn, bool extra_pass, 1099 u64 *func_addr, bool *func_addr_fixed); 1100 1101 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1102 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1103 1104 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1105 u32 pass, void *image) 1106 { 1107 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1108 proglen, pass, image, current->comm, task_pid_nr(current)); 1109 1110 if (image) 1111 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1112 16, 1, image, proglen, false); 1113 } 1114 1115 static inline bool bpf_jit_is_ebpf(void) 1116 { 1117 # ifdef CONFIG_HAVE_EBPF_JIT 1118 return true; 1119 # else 1120 return false; 1121 # endif 1122 } 1123 1124 static inline bool ebpf_jit_enabled(void) 1125 { 1126 return bpf_jit_enable && bpf_jit_is_ebpf(); 1127 } 1128 1129 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1130 { 1131 return fp->jited && bpf_jit_is_ebpf(); 1132 } 1133 1134 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1135 { 1136 /* These are the prerequisites, should someone ever have the 1137 * idea to call blinding outside of them, we make sure to 1138 * bail out. 1139 */ 1140 if (!bpf_jit_is_ebpf()) 1141 return false; 1142 if (!prog->jit_requested) 1143 return false; 1144 if (!bpf_jit_harden) 1145 return false; 1146 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF)) 1147 return false; 1148 1149 return true; 1150 } 1151 1152 static inline bool bpf_jit_kallsyms_enabled(void) 1153 { 1154 /* There are a couple of corner cases where kallsyms should 1155 * not be enabled f.e. on hardening. 1156 */ 1157 if (bpf_jit_harden) 1158 return false; 1159 if (!bpf_jit_kallsyms) 1160 return false; 1161 if (bpf_jit_kallsyms == 1) 1162 return true; 1163 1164 return false; 1165 } 1166 1167 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1168 unsigned long *off, char *sym); 1169 bool is_bpf_text_address(unsigned long addr); 1170 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1171 char *sym); 1172 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr); 1173 1174 static inline const char * 1175 bpf_address_lookup(unsigned long addr, unsigned long *size, 1176 unsigned long *off, char **modname, char *sym) 1177 { 1178 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1179 1180 if (ret && modname) 1181 *modname = NULL; 1182 return ret; 1183 } 1184 1185 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1186 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1187 1188 #else /* CONFIG_BPF_JIT */ 1189 1190 static inline bool ebpf_jit_enabled(void) 1191 { 1192 return false; 1193 } 1194 1195 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1196 { 1197 return false; 1198 } 1199 1200 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1201 { 1202 return false; 1203 } 1204 1205 static inline int 1206 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1207 struct bpf_jit_poke_descriptor *poke) 1208 { 1209 return -ENOTSUPP; 1210 } 1211 1212 static inline void bpf_jit_free(struct bpf_prog *fp) 1213 { 1214 bpf_prog_unlock_free(fp); 1215 } 1216 1217 static inline bool bpf_jit_kallsyms_enabled(void) 1218 { 1219 return false; 1220 } 1221 1222 static inline const char * 1223 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1224 unsigned long *off, char *sym) 1225 { 1226 return NULL; 1227 } 1228 1229 static inline bool is_bpf_text_address(unsigned long addr) 1230 { 1231 return false; 1232 } 1233 1234 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1235 char *type, char *sym) 1236 { 1237 return -ERANGE; 1238 } 1239 1240 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr) 1241 { 1242 return NULL; 1243 } 1244 1245 static inline const char * 1246 bpf_address_lookup(unsigned long addr, unsigned long *size, 1247 unsigned long *off, char **modname, char *sym) 1248 { 1249 return NULL; 1250 } 1251 1252 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1253 { 1254 } 1255 1256 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1257 { 1258 } 1259 1260 #endif /* CONFIG_BPF_JIT */ 1261 1262 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1263 1264 #define BPF_ANC BIT(15) 1265 1266 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1267 { 1268 switch (first->code) { 1269 case BPF_RET | BPF_K: 1270 case BPF_LD | BPF_W | BPF_LEN: 1271 return false; 1272 1273 case BPF_LD | BPF_W | BPF_ABS: 1274 case BPF_LD | BPF_H | BPF_ABS: 1275 case BPF_LD | BPF_B | BPF_ABS: 1276 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1277 return true; 1278 return false; 1279 1280 default: 1281 return true; 1282 } 1283 } 1284 1285 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1286 { 1287 BUG_ON(ftest->code & BPF_ANC); 1288 1289 switch (ftest->code) { 1290 case BPF_LD | BPF_W | BPF_ABS: 1291 case BPF_LD | BPF_H | BPF_ABS: 1292 case BPF_LD | BPF_B | BPF_ABS: 1293 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1294 return BPF_ANC | SKF_AD_##CODE 1295 switch (ftest->k) { 1296 BPF_ANCILLARY(PROTOCOL); 1297 BPF_ANCILLARY(PKTTYPE); 1298 BPF_ANCILLARY(IFINDEX); 1299 BPF_ANCILLARY(NLATTR); 1300 BPF_ANCILLARY(NLATTR_NEST); 1301 BPF_ANCILLARY(MARK); 1302 BPF_ANCILLARY(QUEUE); 1303 BPF_ANCILLARY(HATYPE); 1304 BPF_ANCILLARY(RXHASH); 1305 BPF_ANCILLARY(CPU); 1306 BPF_ANCILLARY(ALU_XOR_X); 1307 BPF_ANCILLARY(VLAN_TAG); 1308 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1309 BPF_ANCILLARY(PAY_OFFSET); 1310 BPF_ANCILLARY(RANDOM); 1311 BPF_ANCILLARY(VLAN_TPID); 1312 } 1313 fallthrough; 1314 default: 1315 return ftest->code; 1316 } 1317 } 1318 1319 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1320 int k, unsigned int size); 1321 1322 static inline int bpf_tell_extensions(void) 1323 { 1324 return SKF_AD_MAX; 1325 } 1326 1327 struct bpf_sock_addr_kern { 1328 struct sock *sk; 1329 struct sockaddr *uaddr; 1330 /* Temporary "register" to make indirect stores to nested structures 1331 * defined above. We need three registers to make such a store, but 1332 * only two (src and dst) are available at convert_ctx_access time 1333 */ 1334 u64 tmp_reg; 1335 void *t_ctx; /* Attach type specific context. */ 1336 u32 uaddrlen; 1337 }; 1338 1339 struct bpf_sock_ops_kern { 1340 struct sock *sk; 1341 union { 1342 u32 args[4]; 1343 u32 reply; 1344 u32 replylong[4]; 1345 }; 1346 struct sk_buff *syn_skb; 1347 struct sk_buff *skb; 1348 void *skb_data_end; 1349 u8 op; 1350 u8 is_fullsock; 1351 u8 remaining_opt_len; 1352 u64 temp; /* temp and everything after is not 1353 * initialized to 0 before calling 1354 * the BPF program. New fields that 1355 * should be initialized to 0 should 1356 * be inserted before temp. 1357 * temp is scratch storage used by 1358 * sock_ops_convert_ctx_access 1359 * as temporary storage of a register. 1360 */ 1361 }; 1362 1363 struct bpf_sysctl_kern { 1364 struct ctl_table_header *head; 1365 struct ctl_table *table; 1366 void *cur_val; 1367 size_t cur_len; 1368 void *new_val; 1369 size_t new_len; 1370 int new_updated; 1371 int write; 1372 loff_t *ppos; 1373 /* Temporary "register" for indirect stores to ppos. */ 1374 u64 tmp_reg; 1375 }; 1376 1377 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1378 struct bpf_sockopt_buf { 1379 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1380 }; 1381 1382 struct bpf_sockopt_kern { 1383 struct sock *sk; 1384 u8 *optval; 1385 u8 *optval_end; 1386 s32 level; 1387 s32 optname; 1388 s32 optlen; 1389 /* for retval in struct bpf_cg_run_ctx */ 1390 struct task_struct *current_task; 1391 /* Temporary "register" for indirect stores to ppos. */ 1392 u64 tmp_reg; 1393 }; 1394 1395 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1396 1397 struct bpf_sk_lookup_kern { 1398 u16 family; 1399 u16 protocol; 1400 __be16 sport; 1401 u16 dport; 1402 struct { 1403 __be32 saddr; 1404 __be32 daddr; 1405 } v4; 1406 struct { 1407 const struct in6_addr *saddr; 1408 const struct in6_addr *daddr; 1409 } v6; 1410 struct sock *selected_sk; 1411 u32 ingress_ifindex; 1412 bool no_reuseport; 1413 }; 1414 1415 extern struct static_key_false bpf_sk_lookup_enabled; 1416 1417 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1418 * 1419 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1420 * SK_DROP. Their meaning is as follows: 1421 * 1422 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1423 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1424 * SK_DROP : terminate lookup with -ECONNREFUSED 1425 * 1426 * This macro aggregates return values and selected sockets from 1427 * multiple BPF programs according to following rules in order: 1428 * 1429 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1430 * macro result is SK_PASS and last ctx.selected_sk is used. 1431 * 2. If any program returned SK_DROP return value, 1432 * macro result is SK_DROP. 1433 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1434 * 1435 * Caller must ensure that the prog array is non-NULL, and that the 1436 * array as well as the programs it contains remain valid. 1437 */ 1438 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1439 ({ \ 1440 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1441 struct bpf_prog_array_item *_item; \ 1442 struct sock *_selected_sk = NULL; \ 1443 bool _no_reuseport = false; \ 1444 struct bpf_prog *_prog; \ 1445 bool _all_pass = true; \ 1446 u32 _ret; \ 1447 \ 1448 migrate_disable(); \ 1449 _item = &(array)->items[0]; \ 1450 while ((_prog = READ_ONCE(_item->prog))) { \ 1451 /* restore most recent selection */ \ 1452 _ctx->selected_sk = _selected_sk; \ 1453 _ctx->no_reuseport = _no_reuseport; \ 1454 \ 1455 _ret = func(_prog, _ctx); \ 1456 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1457 /* remember last non-NULL socket */ \ 1458 _selected_sk = _ctx->selected_sk; \ 1459 _no_reuseport = _ctx->no_reuseport; \ 1460 } else if (_ret == SK_DROP && _all_pass) { \ 1461 _all_pass = false; \ 1462 } \ 1463 _item++; \ 1464 } \ 1465 _ctx->selected_sk = _selected_sk; \ 1466 _ctx->no_reuseport = _no_reuseport; \ 1467 migrate_enable(); \ 1468 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1469 }) 1470 1471 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1472 const __be32 saddr, const __be16 sport, 1473 const __be32 daddr, const u16 dport, 1474 const int ifindex, struct sock **psk) 1475 { 1476 struct bpf_prog_array *run_array; 1477 struct sock *selected_sk = NULL; 1478 bool no_reuseport = false; 1479 1480 rcu_read_lock(); 1481 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1482 if (run_array) { 1483 struct bpf_sk_lookup_kern ctx = { 1484 .family = AF_INET, 1485 .protocol = protocol, 1486 .v4.saddr = saddr, 1487 .v4.daddr = daddr, 1488 .sport = sport, 1489 .dport = dport, 1490 .ingress_ifindex = ifindex, 1491 }; 1492 u32 act; 1493 1494 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1495 if (act == SK_PASS) { 1496 selected_sk = ctx.selected_sk; 1497 no_reuseport = ctx.no_reuseport; 1498 } else { 1499 selected_sk = ERR_PTR(-ECONNREFUSED); 1500 } 1501 } 1502 rcu_read_unlock(); 1503 *psk = selected_sk; 1504 return no_reuseport; 1505 } 1506 1507 #if IS_ENABLED(CONFIG_IPV6) 1508 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1509 const struct in6_addr *saddr, 1510 const __be16 sport, 1511 const struct in6_addr *daddr, 1512 const u16 dport, 1513 const int ifindex, struct sock **psk) 1514 { 1515 struct bpf_prog_array *run_array; 1516 struct sock *selected_sk = NULL; 1517 bool no_reuseport = false; 1518 1519 rcu_read_lock(); 1520 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1521 if (run_array) { 1522 struct bpf_sk_lookup_kern ctx = { 1523 .family = AF_INET6, 1524 .protocol = protocol, 1525 .v6.saddr = saddr, 1526 .v6.daddr = daddr, 1527 .sport = sport, 1528 .dport = dport, 1529 .ingress_ifindex = ifindex, 1530 }; 1531 u32 act; 1532 1533 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1534 if (act == SK_PASS) { 1535 selected_sk = ctx.selected_sk; 1536 no_reuseport = ctx.no_reuseport; 1537 } else { 1538 selected_sk = ERR_PTR(-ECONNREFUSED); 1539 } 1540 } 1541 rcu_read_unlock(); 1542 *psk = selected_sk; 1543 return no_reuseport; 1544 } 1545 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1546 1547 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, 1548 u64 flags, const u64 flag_mask, 1549 void *lookup_elem(struct bpf_map *map, u32 key)) 1550 { 1551 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1552 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1553 1554 /* Lower bits of the flags are used as return code on lookup failure */ 1555 if (unlikely(flags & ~(action_mask | flag_mask))) 1556 return XDP_ABORTED; 1557 1558 ri->tgt_value = lookup_elem(map, index); 1559 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1560 /* If the lookup fails we want to clear out the state in the 1561 * redirect_info struct completely, so that if an eBPF program 1562 * performs multiple lookups, the last one always takes 1563 * precedence. 1564 */ 1565 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1566 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1567 return flags & action_mask; 1568 } 1569 1570 ri->tgt_index = index; 1571 ri->map_id = map->id; 1572 ri->map_type = map->map_type; 1573 1574 if (flags & BPF_F_BROADCAST) { 1575 WRITE_ONCE(ri->map, map); 1576 ri->flags = flags; 1577 } else { 1578 WRITE_ONCE(ri->map, NULL); 1579 ri->flags = 0; 1580 } 1581 1582 return XDP_REDIRECT; 1583 } 1584 1585 #ifdef CONFIG_NET 1586 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); 1587 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1588 u32 len, u64 flags); 1589 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1590 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1591 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); 1592 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 1593 void *buf, unsigned long len, bool flush); 1594 #else /* CONFIG_NET */ 1595 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, 1596 void *to, u32 len) 1597 { 1598 return -EOPNOTSUPP; 1599 } 1600 1601 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, 1602 const void *from, u32 len, u64 flags) 1603 { 1604 return -EOPNOTSUPP; 1605 } 1606 1607 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, 1608 void *buf, u32 len) 1609 { 1610 return -EOPNOTSUPP; 1611 } 1612 1613 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, 1614 void *buf, u32 len) 1615 { 1616 return -EOPNOTSUPP; 1617 } 1618 1619 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 1620 { 1621 return NULL; 1622 } 1623 1624 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, 1625 unsigned long len, bool flush) 1626 { 1627 } 1628 #endif /* CONFIG_NET */ 1629 1630 #endif /* __LINUX_FILTER_H__ */ 1631