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 special load instruction. Same as BPF_MSH */ 76 #define BPF_PROBE_MEM32 0xa0 77 78 /* unused opcode to mark call to interpreter with arguments */ 79 #define BPF_CALL_ARGS 0xe0 80 81 /* unused opcode to mark speculation barrier for mitigating 82 * Speculative Store Bypass 83 */ 84 #define BPF_NOSPEC 0xc0 85 86 /* As per nm, we expose JITed images as text (code) section for 87 * kallsyms. That way, tools like perf can find it to match 88 * addresses. 89 */ 90 #define BPF_SYM_ELF_TYPE 't' 91 92 /* BPF program can access up to 512 bytes of stack space. */ 93 #define MAX_BPF_STACK 512 94 95 /* Helper macros for filter block array initializers. */ 96 97 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 98 99 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \ 100 ((struct bpf_insn) { \ 101 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 102 .dst_reg = DST, \ 103 .src_reg = SRC, \ 104 .off = OFF, \ 105 .imm = 0 }) 106 107 #define BPF_ALU64_REG(OP, DST, SRC) \ 108 BPF_ALU64_REG_OFF(OP, DST, SRC, 0) 109 110 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \ 111 ((struct bpf_insn) { \ 112 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 113 .dst_reg = DST, \ 114 .src_reg = SRC, \ 115 .off = OFF, \ 116 .imm = 0 }) 117 118 #define BPF_ALU32_REG(OP, DST, SRC) \ 119 BPF_ALU32_REG_OFF(OP, DST, SRC, 0) 120 121 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 122 123 #define BPF_ALU64_IMM_OFF(OP, DST, IMM, OFF) \ 124 ((struct bpf_insn) { \ 125 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 126 .dst_reg = DST, \ 127 .src_reg = 0, \ 128 .off = OFF, \ 129 .imm = IMM }) 130 #define BPF_ALU64_IMM(OP, DST, IMM) \ 131 BPF_ALU64_IMM_OFF(OP, DST, IMM, 0) 132 133 #define BPF_ALU32_IMM_OFF(OP, DST, IMM, OFF) \ 134 ((struct bpf_insn) { \ 135 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 136 .dst_reg = DST, \ 137 .src_reg = 0, \ 138 .off = OFF, \ 139 .imm = IMM }) 140 #define BPF_ALU32_IMM(OP, DST, IMM) \ 141 BPF_ALU32_IMM_OFF(OP, DST, IMM, 0) 142 143 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 144 145 #define BPF_ENDIAN(TYPE, DST, LEN) \ 146 ((struct bpf_insn) { \ 147 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 148 .dst_reg = DST, \ 149 .src_reg = 0, \ 150 .off = 0, \ 151 .imm = LEN }) 152 153 /* Byte Swap, bswap16/32/64 */ 154 155 #define BPF_BSWAP(DST, LEN) \ 156 ((struct bpf_insn) { \ 157 .code = BPF_ALU64 | BPF_END | BPF_SRC(BPF_TO_LE), \ 158 .dst_reg = DST, \ 159 .src_reg = 0, \ 160 .off = 0, \ 161 .imm = LEN }) 162 163 /* Short form of mov, dst_reg = src_reg */ 164 165 #define BPF_MOV64_REG(DST, SRC) \ 166 ((struct bpf_insn) { \ 167 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 168 .dst_reg = DST, \ 169 .src_reg = SRC, \ 170 .off = 0, \ 171 .imm = 0 }) 172 173 #define BPF_MOV32_REG(DST, SRC) \ 174 ((struct bpf_insn) { \ 175 .code = BPF_ALU | BPF_MOV | BPF_X, \ 176 .dst_reg = DST, \ 177 .src_reg = SRC, \ 178 .off = 0, \ 179 .imm = 0 }) 180 181 /* Short form of mov, dst_reg = imm32 */ 182 183 #define BPF_MOV64_IMM(DST, IMM) \ 184 ((struct bpf_insn) { \ 185 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 186 .dst_reg = DST, \ 187 .src_reg = 0, \ 188 .off = 0, \ 189 .imm = IMM }) 190 191 #define BPF_MOV32_IMM(DST, IMM) \ 192 ((struct bpf_insn) { \ 193 .code = BPF_ALU | BPF_MOV | BPF_K, \ 194 .dst_reg = DST, \ 195 .src_reg = 0, \ 196 .off = 0, \ 197 .imm = IMM }) 198 199 /* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */ 200 201 #define BPF_MOVSX64_REG(DST, SRC, OFF) \ 202 ((struct bpf_insn) { \ 203 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 204 .dst_reg = DST, \ 205 .src_reg = SRC, \ 206 .off = OFF, \ 207 .imm = 0 }) 208 209 #define BPF_MOVSX32_REG(DST, SRC, OFF) \ 210 ((struct bpf_insn) { \ 211 .code = BPF_ALU | BPF_MOV | BPF_X, \ 212 .dst_reg = DST, \ 213 .src_reg = SRC, \ 214 .off = OFF, \ 215 .imm = 0 }) 216 217 /* Special form of mov32, used for doing explicit zero extension on dst. */ 218 #define BPF_ZEXT_REG(DST) \ 219 ((struct bpf_insn) { \ 220 .code = BPF_ALU | BPF_MOV | BPF_X, \ 221 .dst_reg = DST, \ 222 .src_reg = DST, \ 223 .off = 0, \ 224 .imm = 1 }) 225 226 static inline bool insn_is_zext(const struct bpf_insn *insn) 227 { 228 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 229 } 230 231 /* addr_space_cast from as(0) to as(1) is for converting bpf arena pointers 232 * to pointers in user vma. 233 */ 234 static inline bool insn_is_cast_user(const struct bpf_insn *insn) 235 { 236 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && 237 insn->off == BPF_ADDR_SPACE_CAST && 238 insn->imm == 1U << 16; 239 } 240 241 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 242 #define BPF_LD_IMM64(DST, IMM) \ 243 BPF_LD_IMM64_RAW(DST, 0, IMM) 244 245 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 246 ((struct bpf_insn) { \ 247 .code = BPF_LD | BPF_DW | BPF_IMM, \ 248 .dst_reg = DST, \ 249 .src_reg = SRC, \ 250 .off = 0, \ 251 .imm = (__u32) (IMM) }), \ 252 ((struct bpf_insn) { \ 253 .code = 0, /* zero is reserved opcode */ \ 254 .dst_reg = 0, \ 255 .src_reg = 0, \ 256 .off = 0, \ 257 .imm = ((__u64) (IMM)) >> 32 }) 258 259 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 260 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 261 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 262 263 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 264 265 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 266 ((struct bpf_insn) { \ 267 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 268 .dst_reg = DST, \ 269 .src_reg = SRC, \ 270 .off = 0, \ 271 .imm = IMM }) 272 273 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 274 ((struct bpf_insn) { \ 275 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 276 .dst_reg = DST, \ 277 .src_reg = SRC, \ 278 .off = 0, \ 279 .imm = IMM }) 280 281 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 282 283 #define BPF_LD_ABS(SIZE, IMM) \ 284 ((struct bpf_insn) { \ 285 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 286 .dst_reg = 0, \ 287 .src_reg = 0, \ 288 .off = 0, \ 289 .imm = IMM }) 290 291 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 292 293 #define BPF_LD_IND(SIZE, SRC, IMM) \ 294 ((struct bpf_insn) { \ 295 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 296 .dst_reg = 0, \ 297 .src_reg = SRC, \ 298 .off = 0, \ 299 .imm = IMM }) 300 301 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 302 303 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 304 ((struct bpf_insn) { \ 305 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 306 .dst_reg = DST, \ 307 .src_reg = SRC, \ 308 .off = OFF, \ 309 .imm = 0 }) 310 311 /* Memory load, dst_reg = *(signed size *) (src_reg + off16) */ 312 313 #define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \ 314 ((struct bpf_insn) { \ 315 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \ 316 .dst_reg = DST, \ 317 .src_reg = SRC, \ 318 .off = OFF, \ 319 .imm = 0 }) 320 321 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 322 323 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 324 ((struct bpf_insn) { \ 325 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 326 .dst_reg = DST, \ 327 .src_reg = SRC, \ 328 .off = OFF, \ 329 .imm = 0 }) 330 331 332 /* 333 * Atomic operations: 334 * 335 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 336 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 337 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 338 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 339 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 340 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 341 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 342 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 343 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 344 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 345 */ 346 347 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 348 ((struct bpf_insn) { \ 349 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 350 .dst_reg = DST, \ 351 .src_reg = SRC, \ 352 .off = OFF, \ 353 .imm = OP }) 354 355 /* Legacy alias */ 356 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 357 358 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 359 360 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 361 ((struct bpf_insn) { \ 362 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 363 .dst_reg = DST, \ 364 .src_reg = 0, \ 365 .off = OFF, \ 366 .imm = IMM }) 367 368 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 369 370 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 371 ((struct bpf_insn) { \ 372 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 373 .dst_reg = DST, \ 374 .src_reg = SRC, \ 375 .off = OFF, \ 376 .imm = 0 }) 377 378 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 379 380 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 381 ((struct bpf_insn) { \ 382 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 383 .dst_reg = DST, \ 384 .src_reg = 0, \ 385 .off = OFF, \ 386 .imm = IMM }) 387 388 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 389 390 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 391 ((struct bpf_insn) { \ 392 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 393 .dst_reg = DST, \ 394 .src_reg = SRC, \ 395 .off = OFF, \ 396 .imm = 0 }) 397 398 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 399 400 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 401 ((struct bpf_insn) { \ 402 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 403 .dst_reg = DST, \ 404 .src_reg = 0, \ 405 .off = OFF, \ 406 .imm = IMM }) 407 408 /* Unconditional jumps, goto pc + off16 */ 409 410 #define BPF_JMP_A(OFF) \ 411 ((struct bpf_insn) { \ 412 .code = BPF_JMP | BPF_JA, \ 413 .dst_reg = 0, \ 414 .src_reg = 0, \ 415 .off = OFF, \ 416 .imm = 0 }) 417 418 /* Relative call */ 419 420 #define BPF_CALL_REL(TGT) \ 421 ((struct bpf_insn) { \ 422 .code = BPF_JMP | BPF_CALL, \ 423 .dst_reg = 0, \ 424 .src_reg = BPF_PSEUDO_CALL, \ 425 .off = 0, \ 426 .imm = TGT }) 427 428 /* Convert function address to BPF immediate */ 429 430 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base) 431 432 #define BPF_EMIT_CALL(FUNC) \ 433 ((struct bpf_insn) { \ 434 .code = BPF_JMP | BPF_CALL, \ 435 .dst_reg = 0, \ 436 .src_reg = 0, \ 437 .off = 0, \ 438 .imm = BPF_CALL_IMM(FUNC) }) 439 440 /* Raw code statement block */ 441 442 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 443 ((struct bpf_insn) { \ 444 .code = CODE, \ 445 .dst_reg = DST, \ 446 .src_reg = SRC, \ 447 .off = OFF, \ 448 .imm = IMM }) 449 450 /* Program exit */ 451 452 #define BPF_EXIT_INSN() \ 453 ((struct bpf_insn) { \ 454 .code = BPF_JMP | BPF_EXIT, \ 455 .dst_reg = 0, \ 456 .src_reg = 0, \ 457 .off = 0, \ 458 .imm = 0 }) 459 460 /* Speculation barrier */ 461 462 #define BPF_ST_NOSPEC() \ 463 ((struct bpf_insn) { \ 464 .code = BPF_ST | BPF_NOSPEC, \ 465 .dst_reg = 0, \ 466 .src_reg = 0, \ 467 .off = 0, \ 468 .imm = 0 }) 469 470 /* Internal classic blocks for direct assignment */ 471 472 #define __BPF_STMT(CODE, K) \ 473 ((struct sock_filter) BPF_STMT(CODE, K)) 474 475 #define __BPF_JUMP(CODE, K, JT, JF) \ 476 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 477 478 #define bytes_to_bpf_size(bytes) \ 479 ({ \ 480 int bpf_size = -EINVAL; \ 481 \ 482 if (bytes == sizeof(u8)) \ 483 bpf_size = BPF_B; \ 484 else if (bytes == sizeof(u16)) \ 485 bpf_size = BPF_H; \ 486 else if (bytes == sizeof(u32)) \ 487 bpf_size = BPF_W; \ 488 else if (bytes == sizeof(u64)) \ 489 bpf_size = BPF_DW; \ 490 \ 491 bpf_size; \ 492 }) 493 494 #define bpf_size_to_bytes(bpf_size) \ 495 ({ \ 496 int bytes = -EINVAL; \ 497 \ 498 if (bpf_size == BPF_B) \ 499 bytes = sizeof(u8); \ 500 else if (bpf_size == BPF_H) \ 501 bytes = sizeof(u16); \ 502 else if (bpf_size == BPF_W) \ 503 bytes = sizeof(u32); \ 504 else if (bpf_size == BPF_DW) \ 505 bytes = sizeof(u64); \ 506 \ 507 bytes; \ 508 }) 509 510 #define BPF_SIZEOF(type) \ 511 ({ \ 512 const int __size = bytes_to_bpf_size(sizeof(type)); \ 513 BUILD_BUG_ON(__size < 0); \ 514 __size; \ 515 }) 516 517 #define BPF_FIELD_SIZEOF(type, field) \ 518 ({ \ 519 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 520 BUILD_BUG_ON(__size < 0); \ 521 __size; \ 522 }) 523 524 #define BPF_LDST_BYTES(insn) \ 525 ({ \ 526 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 527 WARN_ON(__size < 0); \ 528 __size; \ 529 }) 530 531 #define __BPF_MAP_0(m, v, ...) v 532 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 533 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 534 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 535 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 536 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 537 538 #define __BPF_REG_0(...) __BPF_PAD(5) 539 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 540 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 541 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 542 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 543 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 544 545 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 546 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 547 548 #define __BPF_CAST(t, a) \ 549 (__force t) \ 550 (__force \ 551 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 552 (unsigned long)0, (t)0))) a 553 #define __BPF_V void 554 #define __BPF_N 555 556 #define __BPF_DECL_ARGS(t, a) t a 557 #define __BPF_DECL_REGS(t, a) u64 a 558 559 #define __BPF_PAD(n) \ 560 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 561 u64, __ur_3, u64, __ur_4, u64, __ur_5) 562 563 #define BPF_CALL_x(x, attr, name, ...) \ 564 static __always_inline \ 565 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 566 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 567 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 568 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 569 { \ 570 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 571 } \ 572 static __always_inline \ 573 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 574 575 #define __NOATTR 576 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__) 577 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__) 578 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__) 579 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__) 580 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__) 581 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__) 582 583 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__) 584 585 #define bpf_ctx_range(TYPE, MEMBER) \ 586 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 587 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 588 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 589 #if BITS_PER_LONG == 64 590 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 591 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 592 #else 593 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 594 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 595 #endif /* BITS_PER_LONG == 64 */ 596 597 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 598 ({ \ 599 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 600 *(PTR_SIZE) = (SIZE); \ 601 offsetof(TYPE, MEMBER); \ 602 }) 603 604 /* A struct sock_filter is architecture independent. */ 605 struct compat_sock_fprog { 606 u16 len; 607 compat_uptr_t filter; /* struct sock_filter * */ 608 }; 609 610 struct sock_fprog_kern { 611 u16 len; 612 struct sock_filter *filter; 613 }; 614 615 /* Some arches need doubleword alignment for their instructions and/or data */ 616 #define BPF_IMAGE_ALIGNMENT 8 617 618 struct bpf_binary_header { 619 u32 size; 620 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 621 }; 622 623 struct bpf_prog_stats { 624 u64_stats_t cnt; 625 u64_stats_t nsecs; 626 u64_stats_t misses; 627 struct u64_stats_sync syncp; 628 } __aligned(2 * sizeof(u64)); 629 630 struct sk_filter { 631 refcount_t refcnt; 632 struct rcu_head rcu; 633 struct bpf_prog *prog; 634 }; 635 636 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 637 638 extern struct mutex nf_conn_btf_access_lock; 639 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 640 const struct bpf_reg_state *reg, 641 int off, int size); 642 643 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 644 const struct bpf_insn *insnsi, 645 unsigned int (*bpf_func)(const void *, 646 const struct bpf_insn *)); 647 648 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 649 const void *ctx, 650 bpf_dispatcher_fn dfunc) 651 { 652 u32 ret; 653 654 cant_migrate(); 655 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 656 struct bpf_prog_stats *stats; 657 u64 duration, start = sched_clock(); 658 unsigned long flags; 659 660 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 661 662 duration = sched_clock() - start; 663 stats = this_cpu_ptr(prog->stats); 664 flags = u64_stats_update_begin_irqsave(&stats->syncp); 665 u64_stats_inc(&stats->cnt); 666 u64_stats_add(&stats->nsecs, duration); 667 u64_stats_update_end_irqrestore(&stats->syncp, flags); 668 } else { 669 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 670 } 671 return ret; 672 } 673 674 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 675 { 676 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 677 } 678 679 /* 680 * Use in preemptible and therefore migratable context to make sure that 681 * the execution of the BPF program runs on one CPU. 682 * 683 * This uses migrate_disable/enable() explicitly to document that the 684 * invocation of a BPF program does not require reentrancy protection 685 * against a BPF program which is invoked from a preempting task. 686 */ 687 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 688 const void *ctx) 689 { 690 u32 ret; 691 692 migrate_disable(); 693 ret = bpf_prog_run(prog, ctx); 694 migrate_enable(); 695 return ret; 696 } 697 698 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 699 700 struct bpf_skb_data_end { 701 struct qdisc_skb_cb qdisc_cb; 702 void *data_meta; 703 void *data_end; 704 }; 705 706 struct bpf_nh_params { 707 u32 nh_family; 708 union { 709 u32 ipv4_nh; 710 struct in6_addr ipv6_nh; 711 }; 712 }; 713 714 struct bpf_redirect_info { 715 u64 tgt_index; 716 void *tgt_value; 717 struct bpf_map *map; 718 u32 flags; 719 u32 kern_flags; 720 u32 map_id; 721 enum bpf_map_type map_type; 722 struct bpf_nh_params nh; 723 }; 724 725 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 726 727 /* flags for bpf_redirect_info kern_flags */ 728 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 729 730 /* Compute the linear packet data range [data, data_end) which 731 * will be accessed by various program types (cls_bpf, act_bpf, 732 * lwt, ...). Subsystems allowing direct data access must (!) 733 * ensure that cb[] area can be written to when BPF program is 734 * invoked (otherwise cb[] save/restore is necessary). 735 */ 736 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 737 { 738 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 739 740 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 741 cb->data_meta = skb->data - skb_metadata_len(skb); 742 cb->data_end = skb->data + skb_headlen(skb); 743 } 744 745 /* Similar to bpf_compute_data_pointers(), except that save orginal 746 * data in cb->data and cb->meta_data for restore. 747 */ 748 static inline void bpf_compute_and_save_data_end( 749 struct sk_buff *skb, void **saved_data_end) 750 { 751 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 752 753 *saved_data_end = cb->data_end; 754 cb->data_end = skb->data + skb_headlen(skb); 755 } 756 757 /* Restore data saved by bpf_compute_and_save_data_end(). */ 758 static inline void bpf_restore_data_end( 759 struct sk_buff *skb, void *saved_data_end) 760 { 761 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 762 763 cb->data_end = saved_data_end; 764 } 765 766 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 767 { 768 /* eBPF programs may read/write skb->cb[] area to transfer meta 769 * data between tail calls. Since this also needs to work with 770 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 771 * 772 * In some socket filter cases, the cb unfortunately needs to be 773 * saved/restored so that protocol specific skb->cb[] data won't 774 * be lost. In any case, due to unpriviledged eBPF programs 775 * attached to sockets, we need to clear the bpf_skb_cb() area 776 * to not leak previous contents to user space. 777 */ 778 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 779 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 780 sizeof_field(struct qdisc_skb_cb, data)); 781 782 return qdisc_skb_cb(skb)->data; 783 } 784 785 /* Must be invoked with migration disabled */ 786 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 787 const void *ctx) 788 { 789 const struct sk_buff *skb = ctx; 790 u8 *cb_data = bpf_skb_cb(skb); 791 u8 cb_saved[BPF_SKB_CB_LEN]; 792 u32 res; 793 794 if (unlikely(prog->cb_access)) { 795 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 796 memset(cb_data, 0, sizeof(cb_saved)); 797 } 798 799 res = bpf_prog_run(prog, skb); 800 801 if (unlikely(prog->cb_access)) 802 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 803 804 return res; 805 } 806 807 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 808 struct sk_buff *skb) 809 { 810 u32 res; 811 812 migrate_disable(); 813 res = __bpf_prog_run_save_cb(prog, skb); 814 migrate_enable(); 815 return res; 816 } 817 818 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 819 struct sk_buff *skb) 820 { 821 u8 *cb_data = bpf_skb_cb(skb); 822 u32 res; 823 824 if (unlikely(prog->cb_access)) 825 memset(cb_data, 0, BPF_SKB_CB_LEN); 826 827 res = bpf_prog_run_pin_on_cpu(prog, skb); 828 return res; 829 } 830 831 DECLARE_BPF_DISPATCHER(xdp) 832 833 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 834 835 u32 xdp_master_redirect(struct xdp_buff *xdp); 836 837 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 838 839 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 840 { 841 return prog->len * sizeof(struct bpf_insn); 842 } 843 844 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 845 { 846 return round_up(bpf_prog_insn_size(prog) + 847 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 848 } 849 850 static inline unsigned int bpf_prog_size(unsigned int proglen) 851 { 852 return max(sizeof(struct bpf_prog), 853 offsetof(struct bpf_prog, insns[proglen])); 854 } 855 856 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 857 { 858 /* When classic BPF programs have been loaded and the arch 859 * does not have a classic BPF JIT (anymore), they have been 860 * converted via bpf_migrate_filter() to eBPF and thus always 861 * have an unspec program type. 862 */ 863 return prog->type == BPF_PROG_TYPE_UNSPEC; 864 } 865 866 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 867 { 868 const u32 size_machine = sizeof(unsigned long); 869 870 if (size > size_machine && size % size_machine == 0) 871 size = size_machine; 872 873 return size; 874 } 875 876 static inline bool 877 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 878 { 879 return size <= size_default && (size & (size - 1)) == 0; 880 } 881 882 static inline u8 883 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 884 { 885 u8 access_off = off & (size_default - 1); 886 887 #ifdef __LITTLE_ENDIAN 888 return access_off; 889 #else 890 return size_default - (access_off + size); 891 #endif 892 } 893 894 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 895 (size == sizeof(__u64) && \ 896 off >= offsetof(type, field) && \ 897 off + sizeof(__u64) <= offsetofend(type, field) && \ 898 off % sizeof(__u64) == 0) 899 900 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 901 902 static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp) 903 { 904 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 905 if (!fp->jited) { 906 set_vm_flush_reset_perms(fp); 907 return set_memory_ro((unsigned long)fp, fp->pages); 908 } 909 #endif 910 return 0; 911 } 912 913 static inline int __must_check 914 bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 915 { 916 set_vm_flush_reset_perms(hdr); 917 return set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT); 918 } 919 920 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 921 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 922 { 923 return sk_filter_trim_cap(sk, skb, 1); 924 } 925 926 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 927 void bpf_prog_free(struct bpf_prog *fp); 928 929 bool bpf_opcode_in_insntable(u8 code); 930 931 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 932 const u32 *insn_to_jit_off); 933 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 934 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 935 936 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 937 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 938 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 939 gfp_t gfp_extra_flags); 940 void __bpf_prog_free(struct bpf_prog *fp); 941 942 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 943 { 944 __bpf_prog_free(fp); 945 } 946 947 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 948 unsigned int flen); 949 950 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 951 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 952 bpf_aux_classic_check_t trans, bool save_orig); 953 void bpf_prog_destroy(struct bpf_prog *fp); 954 955 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 956 int sk_attach_bpf(u32 ufd, struct sock *sk); 957 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 958 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 959 void sk_reuseport_prog_free(struct bpf_prog *prog); 960 int sk_detach_filter(struct sock *sk); 961 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); 962 963 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 964 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 965 966 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 967 #define __bpf_call_base_args \ 968 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 969 (void *)__bpf_call_base) 970 971 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 972 void bpf_jit_compile(struct bpf_prog *prog); 973 bool bpf_jit_needs_zext(void); 974 bool bpf_jit_supports_subprog_tailcalls(void); 975 bool bpf_jit_supports_kfunc_call(void); 976 bool bpf_jit_supports_far_kfunc_call(void); 977 bool bpf_jit_supports_exceptions(void); 978 bool bpf_jit_supports_ptr_xchg(void); 979 bool bpf_jit_supports_arena(void); 980 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie); 981 bool bpf_helper_changes_pkt_data(void *func); 982 983 static inline bool bpf_dump_raw_ok(const struct cred *cred) 984 { 985 /* Reconstruction of call-sites is dependent on kallsyms, 986 * thus make dump the same restriction. 987 */ 988 return kallsyms_show_value(cred); 989 } 990 991 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 992 const struct bpf_insn *patch, u32 len); 993 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 994 995 void bpf_clear_redirect_map(struct bpf_map *map); 996 997 static inline bool xdp_return_frame_no_direct(void) 998 { 999 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1000 1001 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 1002 } 1003 1004 static inline void xdp_set_return_frame_no_direct(void) 1005 { 1006 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1007 1008 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 1009 } 1010 1011 static inline void xdp_clear_return_frame_no_direct(void) 1012 { 1013 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1014 1015 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 1016 } 1017 1018 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 1019 unsigned int pktlen) 1020 { 1021 unsigned int len; 1022 1023 if (unlikely(!(fwd->flags & IFF_UP))) 1024 return -ENETDOWN; 1025 1026 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 1027 if (pktlen > len) 1028 return -EMSGSIZE; 1029 1030 return 0; 1031 } 1032 1033 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 1034 * same cpu context. Further for best results no more than a single map 1035 * for the do_redirect/do_flush pair should be used. This limitation is 1036 * because we only track one map and force a flush when the map changes. 1037 * This does not appear to be a real limitation for existing software. 1038 */ 1039 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 1040 struct xdp_buff *xdp, struct bpf_prog *prog); 1041 int xdp_do_redirect(struct net_device *dev, 1042 struct xdp_buff *xdp, 1043 struct bpf_prog *prog); 1044 int xdp_do_redirect_frame(struct net_device *dev, 1045 struct xdp_buff *xdp, 1046 struct xdp_frame *xdpf, 1047 struct bpf_prog *prog); 1048 void xdp_do_flush(void); 1049 1050 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act); 1051 1052 #ifdef CONFIG_INET 1053 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1054 struct bpf_prog *prog, struct sk_buff *skb, 1055 struct sock *migrating_sk, 1056 u32 hash); 1057 #else 1058 static inline struct sock * 1059 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1060 struct bpf_prog *prog, struct sk_buff *skb, 1061 struct sock *migrating_sk, 1062 u32 hash) 1063 { 1064 return NULL; 1065 } 1066 #endif 1067 1068 #ifdef CONFIG_BPF_JIT 1069 extern int bpf_jit_enable; 1070 extern int bpf_jit_harden; 1071 extern int bpf_jit_kallsyms; 1072 extern long bpf_jit_limit; 1073 extern long bpf_jit_limit_max; 1074 1075 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1076 1077 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); 1078 1079 struct bpf_binary_header * 1080 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1081 unsigned int alignment, 1082 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1083 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1084 u64 bpf_jit_alloc_exec_limit(void); 1085 void *bpf_jit_alloc_exec(unsigned long size); 1086 void bpf_jit_free_exec(void *addr); 1087 void bpf_jit_free(struct bpf_prog *fp); 1088 struct bpf_binary_header * 1089 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); 1090 1091 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); 1092 void bpf_prog_pack_free(void *ptr, u32 size); 1093 1094 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 1095 { 1096 return list_empty(&fp->aux->ksym.lnode) || 1097 fp->aux->ksym.lnode.prev == LIST_POISON2; 1098 } 1099 1100 struct bpf_binary_header * 1101 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, 1102 unsigned int alignment, 1103 struct bpf_binary_header **rw_hdr, 1104 u8 **rw_image, 1105 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1106 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, 1107 struct bpf_binary_header *ro_header, 1108 struct bpf_binary_header *rw_header); 1109 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, 1110 struct bpf_binary_header *rw_header); 1111 1112 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1113 struct bpf_jit_poke_descriptor *poke); 1114 1115 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1116 const struct bpf_insn *insn, bool extra_pass, 1117 u64 *func_addr, bool *func_addr_fixed); 1118 1119 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1120 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1121 1122 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1123 u32 pass, void *image) 1124 { 1125 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1126 proglen, pass, image, current->comm, task_pid_nr(current)); 1127 1128 if (image) 1129 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1130 16, 1, image, proglen, false); 1131 } 1132 1133 static inline bool bpf_jit_is_ebpf(void) 1134 { 1135 # ifdef CONFIG_HAVE_EBPF_JIT 1136 return true; 1137 # else 1138 return false; 1139 # endif 1140 } 1141 1142 static inline bool ebpf_jit_enabled(void) 1143 { 1144 return bpf_jit_enable && bpf_jit_is_ebpf(); 1145 } 1146 1147 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1148 { 1149 return fp->jited && bpf_jit_is_ebpf(); 1150 } 1151 1152 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1153 { 1154 /* These are the prerequisites, should someone ever have the 1155 * idea to call blinding outside of them, we make sure to 1156 * bail out. 1157 */ 1158 if (!bpf_jit_is_ebpf()) 1159 return false; 1160 if (!prog->jit_requested) 1161 return false; 1162 if (!bpf_jit_harden) 1163 return false; 1164 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF)) 1165 return false; 1166 1167 return true; 1168 } 1169 1170 static inline bool bpf_jit_kallsyms_enabled(void) 1171 { 1172 /* There are a couple of corner cases where kallsyms should 1173 * not be enabled f.e. on hardening. 1174 */ 1175 if (bpf_jit_harden) 1176 return false; 1177 if (!bpf_jit_kallsyms) 1178 return false; 1179 if (bpf_jit_kallsyms == 1) 1180 return true; 1181 1182 return false; 1183 } 1184 1185 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1186 unsigned long *off, char *sym); 1187 bool is_bpf_text_address(unsigned long addr); 1188 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1189 char *sym); 1190 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr); 1191 1192 static inline const char * 1193 bpf_address_lookup(unsigned long addr, unsigned long *size, 1194 unsigned long *off, char **modname, char *sym) 1195 { 1196 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1197 1198 if (ret && modname) 1199 *modname = NULL; 1200 return ret; 1201 } 1202 1203 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1204 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1205 1206 #else /* CONFIG_BPF_JIT */ 1207 1208 static inline bool ebpf_jit_enabled(void) 1209 { 1210 return false; 1211 } 1212 1213 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1214 { 1215 return false; 1216 } 1217 1218 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1219 { 1220 return false; 1221 } 1222 1223 static inline int 1224 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1225 struct bpf_jit_poke_descriptor *poke) 1226 { 1227 return -ENOTSUPP; 1228 } 1229 1230 static inline void bpf_jit_free(struct bpf_prog *fp) 1231 { 1232 bpf_prog_unlock_free(fp); 1233 } 1234 1235 static inline bool bpf_jit_kallsyms_enabled(void) 1236 { 1237 return false; 1238 } 1239 1240 static inline const char * 1241 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1242 unsigned long *off, char *sym) 1243 { 1244 return NULL; 1245 } 1246 1247 static inline bool is_bpf_text_address(unsigned long addr) 1248 { 1249 return false; 1250 } 1251 1252 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1253 char *type, char *sym) 1254 { 1255 return -ERANGE; 1256 } 1257 1258 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr) 1259 { 1260 return NULL; 1261 } 1262 1263 static inline const char * 1264 bpf_address_lookup(unsigned long addr, unsigned long *size, 1265 unsigned long *off, char **modname, char *sym) 1266 { 1267 return NULL; 1268 } 1269 1270 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1271 { 1272 } 1273 1274 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1275 { 1276 } 1277 1278 #endif /* CONFIG_BPF_JIT */ 1279 1280 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1281 1282 #define BPF_ANC BIT(15) 1283 1284 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1285 { 1286 switch (first->code) { 1287 case BPF_RET | BPF_K: 1288 case BPF_LD | BPF_W | BPF_LEN: 1289 return false; 1290 1291 case BPF_LD | BPF_W | BPF_ABS: 1292 case BPF_LD | BPF_H | BPF_ABS: 1293 case BPF_LD | BPF_B | BPF_ABS: 1294 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1295 return true; 1296 return false; 1297 1298 default: 1299 return true; 1300 } 1301 } 1302 1303 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1304 { 1305 BUG_ON(ftest->code & BPF_ANC); 1306 1307 switch (ftest->code) { 1308 case BPF_LD | BPF_W | BPF_ABS: 1309 case BPF_LD | BPF_H | BPF_ABS: 1310 case BPF_LD | BPF_B | BPF_ABS: 1311 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1312 return BPF_ANC | SKF_AD_##CODE 1313 switch (ftest->k) { 1314 BPF_ANCILLARY(PROTOCOL); 1315 BPF_ANCILLARY(PKTTYPE); 1316 BPF_ANCILLARY(IFINDEX); 1317 BPF_ANCILLARY(NLATTR); 1318 BPF_ANCILLARY(NLATTR_NEST); 1319 BPF_ANCILLARY(MARK); 1320 BPF_ANCILLARY(QUEUE); 1321 BPF_ANCILLARY(HATYPE); 1322 BPF_ANCILLARY(RXHASH); 1323 BPF_ANCILLARY(CPU); 1324 BPF_ANCILLARY(ALU_XOR_X); 1325 BPF_ANCILLARY(VLAN_TAG); 1326 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1327 BPF_ANCILLARY(PAY_OFFSET); 1328 BPF_ANCILLARY(RANDOM); 1329 BPF_ANCILLARY(VLAN_TPID); 1330 } 1331 fallthrough; 1332 default: 1333 return ftest->code; 1334 } 1335 } 1336 1337 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1338 int k, unsigned int size); 1339 1340 static inline int bpf_tell_extensions(void) 1341 { 1342 return SKF_AD_MAX; 1343 } 1344 1345 struct bpf_sock_addr_kern { 1346 struct sock *sk; 1347 struct sockaddr *uaddr; 1348 /* Temporary "register" to make indirect stores to nested structures 1349 * defined above. We need three registers to make such a store, but 1350 * only two (src and dst) are available at convert_ctx_access time 1351 */ 1352 u64 tmp_reg; 1353 void *t_ctx; /* Attach type specific context. */ 1354 u32 uaddrlen; 1355 }; 1356 1357 struct bpf_sock_ops_kern { 1358 struct sock *sk; 1359 union { 1360 u32 args[4]; 1361 u32 reply; 1362 u32 replylong[4]; 1363 }; 1364 struct sk_buff *syn_skb; 1365 struct sk_buff *skb; 1366 void *skb_data_end; 1367 u8 op; 1368 u8 is_fullsock; 1369 u8 remaining_opt_len; 1370 u64 temp; /* temp and everything after is not 1371 * initialized to 0 before calling 1372 * the BPF program. New fields that 1373 * should be initialized to 0 should 1374 * be inserted before temp. 1375 * temp is scratch storage used by 1376 * sock_ops_convert_ctx_access 1377 * as temporary storage of a register. 1378 */ 1379 }; 1380 1381 struct bpf_sysctl_kern { 1382 struct ctl_table_header *head; 1383 struct ctl_table *table; 1384 void *cur_val; 1385 size_t cur_len; 1386 void *new_val; 1387 size_t new_len; 1388 int new_updated; 1389 int write; 1390 loff_t *ppos; 1391 /* Temporary "register" for indirect stores to ppos. */ 1392 u64 tmp_reg; 1393 }; 1394 1395 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1396 struct bpf_sockopt_buf { 1397 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1398 }; 1399 1400 struct bpf_sockopt_kern { 1401 struct sock *sk; 1402 u8 *optval; 1403 u8 *optval_end; 1404 s32 level; 1405 s32 optname; 1406 s32 optlen; 1407 /* for retval in struct bpf_cg_run_ctx */ 1408 struct task_struct *current_task; 1409 /* Temporary "register" for indirect stores to ppos. */ 1410 u64 tmp_reg; 1411 }; 1412 1413 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1414 1415 struct bpf_sk_lookup_kern { 1416 u16 family; 1417 u16 protocol; 1418 __be16 sport; 1419 u16 dport; 1420 struct { 1421 __be32 saddr; 1422 __be32 daddr; 1423 } v4; 1424 struct { 1425 const struct in6_addr *saddr; 1426 const struct in6_addr *daddr; 1427 } v6; 1428 struct sock *selected_sk; 1429 u32 ingress_ifindex; 1430 bool no_reuseport; 1431 }; 1432 1433 extern struct static_key_false bpf_sk_lookup_enabled; 1434 1435 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1436 * 1437 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1438 * SK_DROP. Their meaning is as follows: 1439 * 1440 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1441 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1442 * SK_DROP : terminate lookup with -ECONNREFUSED 1443 * 1444 * This macro aggregates return values and selected sockets from 1445 * multiple BPF programs according to following rules in order: 1446 * 1447 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1448 * macro result is SK_PASS and last ctx.selected_sk is used. 1449 * 2. If any program returned SK_DROP return value, 1450 * macro result is SK_DROP. 1451 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1452 * 1453 * Caller must ensure that the prog array is non-NULL, and that the 1454 * array as well as the programs it contains remain valid. 1455 */ 1456 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1457 ({ \ 1458 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1459 struct bpf_prog_array_item *_item; \ 1460 struct sock *_selected_sk = NULL; \ 1461 bool _no_reuseport = false; \ 1462 struct bpf_prog *_prog; \ 1463 bool _all_pass = true; \ 1464 u32 _ret; \ 1465 \ 1466 migrate_disable(); \ 1467 _item = &(array)->items[0]; \ 1468 while ((_prog = READ_ONCE(_item->prog))) { \ 1469 /* restore most recent selection */ \ 1470 _ctx->selected_sk = _selected_sk; \ 1471 _ctx->no_reuseport = _no_reuseport; \ 1472 \ 1473 _ret = func(_prog, _ctx); \ 1474 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1475 /* remember last non-NULL socket */ \ 1476 _selected_sk = _ctx->selected_sk; \ 1477 _no_reuseport = _ctx->no_reuseport; \ 1478 } else if (_ret == SK_DROP && _all_pass) { \ 1479 _all_pass = false; \ 1480 } \ 1481 _item++; \ 1482 } \ 1483 _ctx->selected_sk = _selected_sk; \ 1484 _ctx->no_reuseport = _no_reuseport; \ 1485 migrate_enable(); \ 1486 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1487 }) 1488 1489 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1490 const __be32 saddr, const __be16 sport, 1491 const __be32 daddr, const u16 dport, 1492 const int ifindex, struct sock **psk) 1493 { 1494 struct bpf_prog_array *run_array; 1495 struct sock *selected_sk = NULL; 1496 bool no_reuseport = false; 1497 1498 rcu_read_lock(); 1499 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1500 if (run_array) { 1501 struct bpf_sk_lookup_kern ctx = { 1502 .family = AF_INET, 1503 .protocol = protocol, 1504 .v4.saddr = saddr, 1505 .v4.daddr = daddr, 1506 .sport = sport, 1507 .dport = dport, 1508 .ingress_ifindex = ifindex, 1509 }; 1510 u32 act; 1511 1512 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1513 if (act == SK_PASS) { 1514 selected_sk = ctx.selected_sk; 1515 no_reuseport = ctx.no_reuseport; 1516 } else { 1517 selected_sk = ERR_PTR(-ECONNREFUSED); 1518 } 1519 } 1520 rcu_read_unlock(); 1521 *psk = selected_sk; 1522 return no_reuseport; 1523 } 1524 1525 #if IS_ENABLED(CONFIG_IPV6) 1526 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1527 const struct in6_addr *saddr, 1528 const __be16 sport, 1529 const struct in6_addr *daddr, 1530 const u16 dport, 1531 const int ifindex, struct sock **psk) 1532 { 1533 struct bpf_prog_array *run_array; 1534 struct sock *selected_sk = NULL; 1535 bool no_reuseport = false; 1536 1537 rcu_read_lock(); 1538 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1539 if (run_array) { 1540 struct bpf_sk_lookup_kern ctx = { 1541 .family = AF_INET6, 1542 .protocol = protocol, 1543 .v6.saddr = saddr, 1544 .v6.daddr = daddr, 1545 .sport = sport, 1546 .dport = dport, 1547 .ingress_ifindex = ifindex, 1548 }; 1549 u32 act; 1550 1551 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1552 if (act == SK_PASS) { 1553 selected_sk = ctx.selected_sk; 1554 no_reuseport = ctx.no_reuseport; 1555 } else { 1556 selected_sk = ERR_PTR(-ECONNREFUSED); 1557 } 1558 } 1559 rcu_read_unlock(); 1560 *psk = selected_sk; 1561 return no_reuseport; 1562 } 1563 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1564 1565 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, 1566 u64 flags, const u64 flag_mask, 1567 void *lookup_elem(struct bpf_map *map, u32 key)) 1568 { 1569 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1570 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1571 1572 /* Lower bits of the flags are used as return code on lookup failure */ 1573 if (unlikely(flags & ~(action_mask | flag_mask))) 1574 return XDP_ABORTED; 1575 1576 ri->tgt_value = lookup_elem(map, index); 1577 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1578 /* If the lookup fails we want to clear out the state in the 1579 * redirect_info struct completely, so that if an eBPF program 1580 * performs multiple lookups, the last one always takes 1581 * precedence. 1582 */ 1583 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1584 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1585 return flags & action_mask; 1586 } 1587 1588 ri->tgt_index = index; 1589 ri->map_id = map->id; 1590 ri->map_type = map->map_type; 1591 1592 if (flags & BPF_F_BROADCAST) { 1593 WRITE_ONCE(ri->map, map); 1594 ri->flags = flags; 1595 } else { 1596 WRITE_ONCE(ri->map, NULL); 1597 ri->flags = 0; 1598 } 1599 1600 return XDP_REDIRECT; 1601 } 1602 1603 #ifdef CONFIG_NET 1604 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); 1605 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1606 u32 len, u64 flags); 1607 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1608 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1609 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); 1610 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 1611 void *buf, unsigned long len, bool flush); 1612 #else /* CONFIG_NET */ 1613 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, 1614 void *to, u32 len) 1615 { 1616 return -EOPNOTSUPP; 1617 } 1618 1619 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, 1620 const void *from, u32 len, u64 flags) 1621 { 1622 return -EOPNOTSUPP; 1623 } 1624 1625 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, 1626 void *buf, u32 len) 1627 { 1628 return -EOPNOTSUPP; 1629 } 1630 1631 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, 1632 void *buf, u32 len) 1633 { 1634 return -EOPNOTSUPP; 1635 } 1636 1637 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 1638 { 1639 return NULL; 1640 } 1641 1642 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, 1643 unsigned long len, bool flush) 1644 { 1645 } 1646 #endif /* CONFIG_NET */ 1647 1648 #endif /* __LINUX_FILTER_H__ */ 1649