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, 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 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 555 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 BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) 563 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) 564 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) 565 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) 566 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) 567 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) 568 569 #define bpf_ctx_range(TYPE, MEMBER) \ 570 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 571 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 572 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 573 #if BITS_PER_LONG == 64 574 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 575 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 576 #else 577 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 578 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 579 #endif /* BITS_PER_LONG == 64 */ 580 581 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 582 ({ \ 583 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 584 *(PTR_SIZE) = (SIZE); \ 585 offsetof(TYPE, MEMBER); \ 586 }) 587 588 /* A struct sock_filter is architecture independent. */ 589 struct compat_sock_fprog { 590 u16 len; 591 compat_uptr_t filter; /* struct sock_filter * */ 592 }; 593 594 struct sock_fprog_kern { 595 u16 len; 596 struct sock_filter *filter; 597 }; 598 599 /* Some arches need doubleword alignment for their instructions and/or data */ 600 #define BPF_IMAGE_ALIGNMENT 8 601 602 struct bpf_binary_header { 603 u32 size; 604 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 605 }; 606 607 struct bpf_prog_stats { 608 u64_stats_t cnt; 609 u64_stats_t nsecs; 610 u64_stats_t misses; 611 struct u64_stats_sync syncp; 612 } __aligned(2 * sizeof(u64)); 613 614 struct sk_filter { 615 refcount_t refcnt; 616 struct rcu_head rcu; 617 struct bpf_prog *prog; 618 }; 619 620 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 621 622 extern struct mutex nf_conn_btf_access_lock; 623 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 624 const struct bpf_reg_state *reg, 625 int off, int size); 626 627 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 628 const struct bpf_insn *insnsi, 629 unsigned int (*bpf_func)(const void *, 630 const struct bpf_insn *)); 631 632 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 633 const void *ctx, 634 bpf_dispatcher_fn dfunc) 635 { 636 u32 ret; 637 638 cant_migrate(); 639 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 640 struct bpf_prog_stats *stats; 641 u64 start = sched_clock(); 642 unsigned long flags; 643 644 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 645 stats = this_cpu_ptr(prog->stats); 646 flags = u64_stats_update_begin_irqsave(&stats->syncp); 647 u64_stats_inc(&stats->cnt); 648 u64_stats_add(&stats->nsecs, sched_clock() - start); 649 u64_stats_update_end_irqrestore(&stats->syncp, flags); 650 } else { 651 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 652 } 653 return ret; 654 } 655 656 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 657 { 658 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 659 } 660 661 /* 662 * Use in preemptible and therefore migratable context to make sure that 663 * the execution of the BPF program runs on one CPU. 664 * 665 * This uses migrate_disable/enable() explicitly to document that the 666 * invocation of a BPF program does not require reentrancy protection 667 * against a BPF program which is invoked from a preempting task. 668 */ 669 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 670 const void *ctx) 671 { 672 u32 ret; 673 674 migrate_disable(); 675 ret = bpf_prog_run(prog, ctx); 676 migrate_enable(); 677 return ret; 678 } 679 680 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 681 682 struct bpf_skb_data_end { 683 struct qdisc_skb_cb qdisc_cb; 684 void *data_meta; 685 void *data_end; 686 }; 687 688 struct bpf_nh_params { 689 u32 nh_family; 690 union { 691 u32 ipv4_nh; 692 struct in6_addr ipv6_nh; 693 }; 694 }; 695 696 struct bpf_redirect_info { 697 u64 tgt_index; 698 void *tgt_value; 699 struct bpf_map *map; 700 u32 flags; 701 u32 kern_flags; 702 u32 map_id; 703 enum bpf_map_type map_type; 704 struct bpf_nh_params nh; 705 }; 706 707 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 708 709 /* flags for bpf_redirect_info kern_flags */ 710 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 711 712 /* Compute the linear packet data range [data, data_end) which 713 * will be accessed by various program types (cls_bpf, act_bpf, 714 * lwt, ...). Subsystems allowing direct data access must (!) 715 * ensure that cb[] area can be written to when BPF program is 716 * invoked (otherwise cb[] save/restore is necessary). 717 */ 718 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 719 { 720 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 721 722 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 723 cb->data_meta = skb->data - skb_metadata_len(skb); 724 cb->data_end = skb->data + skb_headlen(skb); 725 } 726 727 /* Similar to bpf_compute_data_pointers(), except that save orginal 728 * data in cb->data and cb->meta_data for restore. 729 */ 730 static inline void bpf_compute_and_save_data_end( 731 struct sk_buff *skb, void **saved_data_end) 732 { 733 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 734 735 *saved_data_end = cb->data_end; 736 cb->data_end = skb->data + skb_headlen(skb); 737 } 738 739 /* Restore data saved by bpf_compute_and_save_data_end(). */ 740 static inline void bpf_restore_data_end( 741 struct sk_buff *skb, void *saved_data_end) 742 { 743 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 744 745 cb->data_end = saved_data_end; 746 } 747 748 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 749 { 750 /* eBPF programs may read/write skb->cb[] area to transfer meta 751 * data between tail calls. Since this also needs to work with 752 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 753 * 754 * In some socket filter cases, the cb unfortunately needs to be 755 * saved/restored so that protocol specific skb->cb[] data won't 756 * be lost. In any case, due to unpriviledged eBPF programs 757 * attached to sockets, we need to clear the bpf_skb_cb() area 758 * to not leak previous contents to user space. 759 */ 760 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 761 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 762 sizeof_field(struct qdisc_skb_cb, data)); 763 764 return qdisc_skb_cb(skb)->data; 765 } 766 767 /* Must be invoked with migration disabled */ 768 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 769 const void *ctx) 770 { 771 const struct sk_buff *skb = ctx; 772 u8 *cb_data = bpf_skb_cb(skb); 773 u8 cb_saved[BPF_SKB_CB_LEN]; 774 u32 res; 775 776 if (unlikely(prog->cb_access)) { 777 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 778 memset(cb_data, 0, sizeof(cb_saved)); 779 } 780 781 res = bpf_prog_run(prog, skb); 782 783 if (unlikely(prog->cb_access)) 784 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 785 786 return res; 787 } 788 789 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 790 struct sk_buff *skb) 791 { 792 u32 res; 793 794 migrate_disable(); 795 res = __bpf_prog_run_save_cb(prog, skb); 796 migrate_enable(); 797 return res; 798 } 799 800 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 801 struct sk_buff *skb) 802 { 803 u8 *cb_data = bpf_skb_cb(skb); 804 u32 res; 805 806 if (unlikely(prog->cb_access)) 807 memset(cb_data, 0, BPF_SKB_CB_LEN); 808 809 res = bpf_prog_run_pin_on_cpu(prog, skb); 810 return res; 811 } 812 813 DECLARE_BPF_DISPATCHER(xdp) 814 815 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 816 817 u32 xdp_master_redirect(struct xdp_buff *xdp); 818 819 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 820 821 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 822 { 823 return prog->len * sizeof(struct bpf_insn); 824 } 825 826 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 827 { 828 return round_up(bpf_prog_insn_size(prog) + 829 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 830 } 831 832 static inline unsigned int bpf_prog_size(unsigned int proglen) 833 { 834 return max(sizeof(struct bpf_prog), 835 offsetof(struct bpf_prog, insns[proglen])); 836 } 837 838 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 839 { 840 /* When classic BPF programs have been loaded and the arch 841 * does not have a classic BPF JIT (anymore), they have been 842 * converted via bpf_migrate_filter() to eBPF and thus always 843 * have an unspec program type. 844 */ 845 return prog->type == BPF_PROG_TYPE_UNSPEC; 846 } 847 848 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 849 { 850 const u32 size_machine = sizeof(unsigned long); 851 852 if (size > size_machine && size % size_machine == 0) 853 size = size_machine; 854 855 return size; 856 } 857 858 static inline bool 859 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 860 { 861 return size <= size_default && (size & (size - 1)) == 0; 862 } 863 864 static inline u8 865 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 866 { 867 u8 access_off = off & (size_default - 1); 868 869 #ifdef __LITTLE_ENDIAN 870 return access_off; 871 #else 872 return size_default - (access_off + size); 873 #endif 874 } 875 876 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 877 (size == sizeof(__u64) && \ 878 off >= offsetof(type, field) && \ 879 off + sizeof(__u64) <= offsetofend(type, field) && \ 880 off % sizeof(__u64) == 0) 881 882 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 883 884 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 885 { 886 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 887 if (!fp->jited) { 888 set_vm_flush_reset_perms(fp); 889 set_memory_ro((unsigned long)fp, fp->pages); 890 } 891 #endif 892 } 893 894 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 895 { 896 set_vm_flush_reset_perms(hdr); 897 set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT); 898 } 899 900 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 901 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 902 { 903 return sk_filter_trim_cap(sk, skb, 1); 904 } 905 906 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 907 void bpf_prog_free(struct bpf_prog *fp); 908 909 bool bpf_opcode_in_insntable(u8 code); 910 911 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 912 const u32 *insn_to_jit_off); 913 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 914 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 915 916 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 917 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 918 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 919 gfp_t gfp_extra_flags); 920 void __bpf_prog_free(struct bpf_prog *fp); 921 922 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 923 { 924 __bpf_prog_free(fp); 925 } 926 927 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 928 unsigned int flen); 929 930 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 931 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 932 bpf_aux_classic_check_t trans, bool save_orig); 933 void bpf_prog_destroy(struct bpf_prog *fp); 934 935 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 936 int sk_attach_bpf(u32 ufd, struct sock *sk); 937 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 938 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 939 void sk_reuseport_prog_free(struct bpf_prog *prog); 940 int sk_detach_filter(struct sock *sk); 941 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); 942 943 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 944 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 945 946 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 947 #define __bpf_call_base_args \ 948 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 949 (void *)__bpf_call_base) 950 951 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 952 void bpf_jit_compile(struct bpf_prog *prog); 953 bool bpf_jit_needs_zext(void); 954 bool bpf_jit_supports_subprog_tailcalls(void); 955 bool bpf_jit_supports_kfunc_call(void); 956 bool bpf_jit_supports_far_kfunc_call(void); 957 bool bpf_jit_supports_exceptions(void); 958 bool bpf_jit_supports_ptr_xchg(void); 959 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie); 960 bool bpf_helper_changes_pkt_data(void *func); 961 962 static inline bool bpf_dump_raw_ok(const struct cred *cred) 963 { 964 /* Reconstruction of call-sites is dependent on kallsyms, 965 * thus make dump the same restriction. 966 */ 967 return kallsyms_show_value(cred); 968 } 969 970 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 971 const struct bpf_insn *patch, u32 len); 972 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 973 974 void bpf_clear_redirect_map(struct bpf_map *map); 975 976 static inline bool xdp_return_frame_no_direct(void) 977 { 978 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 979 980 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 981 } 982 983 static inline void xdp_set_return_frame_no_direct(void) 984 { 985 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 986 987 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 988 } 989 990 static inline void xdp_clear_return_frame_no_direct(void) 991 { 992 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 993 994 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 995 } 996 997 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 998 unsigned int pktlen) 999 { 1000 unsigned int len; 1001 1002 if (unlikely(!(fwd->flags & IFF_UP))) 1003 return -ENETDOWN; 1004 1005 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 1006 if (pktlen > len) 1007 return -EMSGSIZE; 1008 1009 return 0; 1010 } 1011 1012 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 1013 * same cpu context. Further for best results no more than a single map 1014 * for the do_redirect/do_flush pair should be used. This limitation is 1015 * because we only track one map and force a flush when the map changes. 1016 * This does not appear to be a real limitation for existing software. 1017 */ 1018 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 1019 struct xdp_buff *xdp, struct bpf_prog *prog); 1020 int xdp_do_redirect(struct net_device *dev, 1021 struct xdp_buff *xdp, 1022 struct bpf_prog *prog); 1023 int xdp_do_redirect_frame(struct net_device *dev, 1024 struct xdp_buff *xdp, 1025 struct xdp_frame *xdpf, 1026 struct bpf_prog *prog); 1027 void xdp_do_flush(void); 1028 1029 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act); 1030 1031 #ifdef CONFIG_INET 1032 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1033 struct bpf_prog *prog, struct sk_buff *skb, 1034 struct sock *migrating_sk, 1035 u32 hash); 1036 #else 1037 static inline struct sock * 1038 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1039 struct bpf_prog *prog, struct sk_buff *skb, 1040 struct sock *migrating_sk, 1041 u32 hash) 1042 { 1043 return NULL; 1044 } 1045 #endif 1046 1047 #ifdef CONFIG_BPF_JIT 1048 extern int bpf_jit_enable; 1049 extern int bpf_jit_harden; 1050 extern int bpf_jit_kallsyms; 1051 extern long bpf_jit_limit; 1052 extern long bpf_jit_limit_max; 1053 1054 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1055 1056 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); 1057 1058 struct bpf_binary_header * 1059 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1060 unsigned int alignment, 1061 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1062 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1063 u64 bpf_jit_alloc_exec_limit(void); 1064 void *bpf_jit_alloc_exec(unsigned long size); 1065 void bpf_jit_free_exec(void *addr); 1066 void bpf_jit_free(struct bpf_prog *fp); 1067 struct bpf_binary_header * 1068 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); 1069 1070 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); 1071 void bpf_prog_pack_free(void *ptr, u32 size); 1072 1073 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 1074 { 1075 return list_empty(&fp->aux->ksym.lnode) || 1076 fp->aux->ksym.lnode.prev == LIST_POISON2; 1077 } 1078 1079 struct bpf_binary_header * 1080 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, 1081 unsigned int alignment, 1082 struct bpf_binary_header **rw_hdr, 1083 u8 **rw_image, 1084 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1085 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, 1086 struct bpf_binary_header *ro_header, 1087 struct bpf_binary_header *rw_header); 1088 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, 1089 struct bpf_binary_header *rw_header); 1090 1091 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1092 struct bpf_jit_poke_descriptor *poke); 1093 1094 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1095 const struct bpf_insn *insn, bool extra_pass, 1096 u64 *func_addr, bool *func_addr_fixed); 1097 1098 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1099 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1100 1101 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1102 u32 pass, void *image) 1103 { 1104 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1105 proglen, pass, image, current->comm, task_pid_nr(current)); 1106 1107 if (image) 1108 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1109 16, 1, image, proglen, false); 1110 } 1111 1112 static inline bool bpf_jit_is_ebpf(void) 1113 { 1114 # ifdef CONFIG_HAVE_EBPF_JIT 1115 return true; 1116 # else 1117 return false; 1118 # endif 1119 } 1120 1121 static inline bool ebpf_jit_enabled(void) 1122 { 1123 return bpf_jit_enable && bpf_jit_is_ebpf(); 1124 } 1125 1126 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1127 { 1128 return fp->jited && bpf_jit_is_ebpf(); 1129 } 1130 1131 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1132 { 1133 /* These are the prerequisites, should someone ever have the 1134 * idea to call blinding outside of them, we make sure to 1135 * bail out. 1136 */ 1137 if (!bpf_jit_is_ebpf()) 1138 return false; 1139 if (!prog->jit_requested) 1140 return false; 1141 if (!bpf_jit_harden) 1142 return false; 1143 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF)) 1144 return false; 1145 1146 return true; 1147 } 1148 1149 static inline bool bpf_jit_kallsyms_enabled(void) 1150 { 1151 /* There are a couple of corner cases where kallsyms should 1152 * not be enabled f.e. on hardening. 1153 */ 1154 if (bpf_jit_harden) 1155 return false; 1156 if (!bpf_jit_kallsyms) 1157 return false; 1158 if (bpf_jit_kallsyms == 1) 1159 return true; 1160 1161 return false; 1162 } 1163 1164 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1165 unsigned long *off, char *sym); 1166 bool is_bpf_text_address(unsigned long addr); 1167 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1168 char *sym); 1169 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr); 1170 1171 static inline const char * 1172 bpf_address_lookup(unsigned long addr, unsigned long *size, 1173 unsigned long *off, char **modname, char *sym) 1174 { 1175 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1176 1177 if (ret && modname) 1178 *modname = NULL; 1179 return ret; 1180 } 1181 1182 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1183 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1184 1185 #else /* CONFIG_BPF_JIT */ 1186 1187 static inline bool ebpf_jit_enabled(void) 1188 { 1189 return false; 1190 } 1191 1192 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1193 { 1194 return false; 1195 } 1196 1197 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1198 { 1199 return false; 1200 } 1201 1202 static inline int 1203 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1204 struct bpf_jit_poke_descriptor *poke) 1205 { 1206 return -ENOTSUPP; 1207 } 1208 1209 static inline void bpf_jit_free(struct bpf_prog *fp) 1210 { 1211 bpf_prog_unlock_free(fp); 1212 } 1213 1214 static inline bool bpf_jit_kallsyms_enabled(void) 1215 { 1216 return false; 1217 } 1218 1219 static inline const char * 1220 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1221 unsigned long *off, char *sym) 1222 { 1223 return NULL; 1224 } 1225 1226 static inline bool is_bpf_text_address(unsigned long addr) 1227 { 1228 return false; 1229 } 1230 1231 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1232 char *type, char *sym) 1233 { 1234 return -ERANGE; 1235 } 1236 1237 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr) 1238 { 1239 return NULL; 1240 } 1241 1242 static inline const char * 1243 bpf_address_lookup(unsigned long addr, unsigned long *size, 1244 unsigned long *off, char **modname, char *sym) 1245 { 1246 return NULL; 1247 } 1248 1249 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1250 { 1251 } 1252 1253 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1254 { 1255 } 1256 1257 #endif /* CONFIG_BPF_JIT */ 1258 1259 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1260 1261 #define BPF_ANC BIT(15) 1262 1263 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1264 { 1265 switch (first->code) { 1266 case BPF_RET | BPF_K: 1267 case BPF_LD | BPF_W | BPF_LEN: 1268 return false; 1269 1270 case BPF_LD | BPF_W | BPF_ABS: 1271 case BPF_LD | BPF_H | BPF_ABS: 1272 case BPF_LD | BPF_B | BPF_ABS: 1273 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1274 return true; 1275 return false; 1276 1277 default: 1278 return true; 1279 } 1280 } 1281 1282 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1283 { 1284 BUG_ON(ftest->code & BPF_ANC); 1285 1286 switch (ftest->code) { 1287 case BPF_LD | BPF_W | BPF_ABS: 1288 case BPF_LD | BPF_H | BPF_ABS: 1289 case BPF_LD | BPF_B | BPF_ABS: 1290 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1291 return BPF_ANC | SKF_AD_##CODE 1292 switch (ftest->k) { 1293 BPF_ANCILLARY(PROTOCOL); 1294 BPF_ANCILLARY(PKTTYPE); 1295 BPF_ANCILLARY(IFINDEX); 1296 BPF_ANCILLARY(NLATTR); 1297 BPF_ANCILLARY(NLATTR_NEST); 1298 BPF_ANCILLARY(MARK); 1299 BPF_ANCILLARY(QUEUE); 1300 BPF_ANCILLARY(HATYPE); 1301 BPF_ANCILLARY(RXHASH); 1302 BPF_ANCILLARY(CPU); 1303 BPF_ANCILLARY(ALU_XOR_X); 1304 BPF_ANCILLARY(VLAN_TAG); 1305 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1306 BPF_ANCILLARY(PAY_OFFSET); 1307 BPF_ANCILLARY(RANDOM); 1308 BPF_ANCILLARY(VLAN_TPID); 1309 } 1310 fallthrough; 1311 default: 1312 return ftest->code; 1313 } 1314 } 1315 1316 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1317 int k, unsigned int size); 1318 1319 static inline int bpf_tell_extensions(void) 1320 { 1321 return SKF_AD_MAX; 1322 } 1323 1324 struct bpf_sock_addr_kern { 1325 struct sock *sk; 1326 struct sockaddr *uaddr; 1327 /* Temporary "register" to make indirect stores to nested structures 1328 * defined above. We need three registers to make such a store, but 1329 * only two (src and dst) are available at convert_ctx_access time 1330 */ 1331 u64 tmp_reg; 1332 void *t_ctx; /* Attach type specific context. */ 1333 u32 uaddrlen; 1334 }; 1335 1336 struct bpf_sock_ops_kern { 1337 struct sock *sk; 1338 union { 1339 u32 args[4]; 1340 u32 reply; 1341 u32 replylong[4]; 1342 }; 1343 struct sk_buff *syn_skb; 1344 struct sk_buff *skb; 1345 void *skb_data_end; 1346 u8 op; 1347 u8 is_fullsock; 1348 u8 remaining_opt_len; 1349 u64 temp; /* temp and everything after is not 1350 * initialized to 0 before calling 1351 * the BPF program. New fields that 1352 * should be initialized to 0 should 1353 * be inserted before temp. 1354 * temp is scratch storage used by 1355 * sock_ops_convert_ctx_access 1356 * as temporary storage of a register. 1357 */ 1358 }; 1359 1360 struct bpf_sysctl_kern { 1361 struct ctl_table_header *head; 1362 struct ctl_table *table; 1363 void *cur_val; 1364 size_t cur_len; 1365 void *new_val; 1366 size_t new_len; 1367 int new_updated; 1368 int write; 1369 loff_t *ppos; 1370 /* Temporary "register" for indirect stores to ppos. */ 1371 u64 tmp_reg; 1372 }; 1373 1374 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1375 struct bpf_sockopt_buf { 1376 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1377 }; 1378 1379 struct bpf_sockopt_kern { 1380 struct sock *sk; 1381 u8 *optval; 1382 u8 *optval_end; 1383 s32 level; 1384 s32 optname; 1385 s32 optlen; 1386 /* for retval in struct bpf_cg_run_ctx */ 1387 struct task_struct *current_task; 1388 /* Temporary "register" for indirect stores to ppos. */ 1389 u64 tmp_reg; 1390 }; 1391 1392 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1393 1394 struct bpf_sk_lookup_kern { 1395 u16 family; 1396 u16 protocol; 1397 __be16 sport; 1398 u16 dport; 1399 struct { 1400 __be32 saddr; 1401 __be32 daddr; 1402 } v4; 1403 struct { 1404 const struct in6_addr *saddr; 1405 const struct in6_addr *daddr; 1406 } v6; 1407 struct sock *selected_sk; 1408 u32 ingress_ifindex; 1409 bool no_reuseport; 1410 }; 1411 1412 extern struct static_key_false bpf_sk_lookup_enabled; 1413 1414 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1415 * 1416 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1417 * SK_DROP. Their meaning is as follows: 1418 * 1419 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1420 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1421 * SK_DROP : terminate lookup with -ECONNREFUSED 1422 * 1423 * This macro aggregates return values and selected sockets from 1424 * multiple BPF programs according to following rules in order: 1425 * 1426 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1427 * macro result is SK_PASS and last ctx.selected_sk is used. 1428 * 2. If any program returned SK_DROP return value, 1429 * macro result is SK_DROP. 1430 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1431 * 1432 * Caller must ensure that the prog array is non-NULL, and that the 1433 * array as well as the programs it contains remain valid. 1434 */ 1435 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1436 ({ \ 1437 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1438 struct bpf_prog_array_item *_item; \ 1439 struct sock *_selected_sk = NULL; \ 1440 bool _no_reuseport = false; \ 1441 struct bpf_prog *_prog; \ 1442 bool _all_pass = true; \ 1443 u32 _ret; \ 1444 \ 1445 migrate_disable(); \ 1446 _item = &(array)->items[0]; \ 1447 while ((_prog = READ_ONCE(_item->prog))) { \ 1448 /* restore most recent selection */ \ 1449 _ctx->selected_sk = _selected_sk; \ 1450 _ctx->no_reuseport = _no_reuseport; \ 1451 \ 1452 _ret = func(_prog, _ctx); \ 1453 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1454 /* remember last non-NULL socket */ \ 1455 _selected_sk = _ctx->selected_sk; \ 1456 _no_reuseport = _ctx->no_reuseport; \ 1457 } else if (_ret == SK_DROP && _all_pass) { \ 1458 _all_pass = false; \ 1459 } \ 1460 _item++; \ 1461 } \ 1462 _ctx->selected_sk = _selected_sk; \ 1463 _ctx->no_reuseport = _no_reuseport; \ 1464 migrate_enable(); \ 1465 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1466 }) 1467 1468 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1469 const __be32 saddr, const __be16 sport, 1470 const __be32 daddr, const u16 dport, 1471 const int ifindex, struct sock **psk) 1472 { 1473 struct bpf_prog_array *run_array; 1474 struct sock *selected_sk = NULL; 1475 bool no_reuseport = false; 1476 1477 rcu_read_lock(); 1478 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1479 if (run_array) { 1480 struct bpf_sk_lookup_kern ctx = { 1481 .family = AF_INET, 1482 .protocol = protocol, 1483 .v4.saddr = saddr, 1484 .v4.daddr = daddr, 1485 .sport = sport, 1486 .dport = dport, 1487 .ingress_ifindex = ifindex, 1488 }; 1489 u32 act; 1490 1491 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1492 if (act == SK_PASS) { 1493 selected_sk = ctx.selected_sk; 1494 no_reuseport = ctx.no_reuseport; 1495 } else { 1496 selected_sk = ERR_PTR(-ECONNREFUSED); 1497 } 1498 } 1499 rcu_read_unlock(); 1500 *psk = selected_sk; 1501 return no_reuseport; 1502 } 1503 1504 #if IS_ENABLED(CONFIG_IPV6) 1505 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1506 const struct in6_addr *saddr, 1507 const __be16 sport, 1508 const struct in6_addr *daddr, 1509 const u16 dport, 1510 const int ifindex, struct sock **psk) 1511 { 1512 struct bpf_prog_array *run_array; 1513 struct sock *selected_sk = NULL; 1514 bool no_reuseport = false; 1515 1516 rcu_read_lock(); 1517 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1518 if (run_array) { 1519 struct bpf_sk_lookup_kern ctx = { 1520 .family = AF_INET6, 1521 .protocol = protocol, 1522 .v6.saddr = saddr, 1523 .v6.daddr = daddr, 1524 .sport = sport, 1525 .dport = dport, 1526 .ingress_ifindex = ifindex, 1527 }; 1528 u32 act; 1529 1530 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1531 if (act == SK_PASS) { 1532 selected_sk = ctx.selected_sk; 1533 no_reuseport = ctx.no_reuseport; 1534 } else { 1535 selected_sk = ERR_PTR(-ECONNREFUSED); 1536 } 1537 } 1538 rcu_read_unlock(); 1539 *psk = selected_sk; 1540 return no_reuseport; 1541 } 1542 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1543 1544 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, 1545 u64 flags, const u64 flag_mask, 1546 void *lookup_elem(struct bpf_map *map, u32 key)) 1547 { 1548 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1549 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1550 1551 /* Lower bits of the flags are used as return code on lookup failure */ 1552 if (unlikely(flags & ~(action_mask | flag_mask))) 1553 return XDP_ABORTED; 1554 1555 ri->tgt_value = lookup_elem(map, index); 1556 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1557 /* If the lookup fails we want to clear out the state in the 1558 * redirect_info struct completely, so that if an eBPF program 1559 * performs multiple lookups, the last one always takes 1560 * precedence. 1561 */ 1562 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1563 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1564 return flags & action_mask; 1565 } 1566 1567 ri->tgt_index = index; 1568 ri->map_id = map->id; 1569 ri->map_type = map->map_type; 1570 1571 if (flags & BPF_F_BROADCAST) { 1572 WRITE_ONCE(ri->map, map); 1573 ri->flags = flags; 1574 } else { 1575 WRITE_ONCE(ri->map, NULL); 1576 ri->flags = 0; 1577 } 1578 1579 return XDP_REDIRECT; 1580 } 1581 1582 #ifdef CONFIG_NET 1583 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); 1584 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1585 u32 len, u64 flags); 1586 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1587 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1588 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); 1589 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 1590 void *buf, unsigned long len, bool flush); 1591 #else /* CONFIG_NET */ 1592 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, 1593 void *to, u32 len) 1594 { 1595 return -EOPNOTSUPP; 1596 } 1597 1598 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, 1599 const void *from, u32 len, u64 flags) 1600 { 1601 return -EOPNOTSUPP; 1602 } 1603 1604 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, 1605 void *buf, u32 len) 1606 { 1607 return -EOPNOTSUPP; 1608 } 1609 1610 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, 1611 void *buf, u32 len) 1612 { 1613 return -EOPNOTSUPP; 1614 } 1615 1616 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 1617 { 1618 return NULL; 1619 } 1620 1621 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, 1622 unsigned long len, bool flush) 1623 { 1624 } 1625 #endif /* CONFIG_NET */ 1626 1627 #endif /* __LINUX_FILTER_H__ */ 1628