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 /* Special (internal-only) form of mov, used to resolve per-CPU addrs: 182 * dst_reg = src_reg + <percpu_base_off> 183 * BPF_ADDR_PERCPU is used as a special insn->off value. 184 */ 185 #define BPF_ADDR_PERCPU (-1) 186 187 #define BPF_MOV64_PERCPU_REG(DST, SRC) \ 188 ((struct bpf_insn) { \ 189 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 190 .dst_reg = DST, \ 191 .src_reg = SRC, \ 192 .off = BPF_ADDR_PERCPU, \ 193 .imm = 0 }) 194 195 static inline bool insn_is_mov_percpu_addr(const struct bpf_insn *insn) 196 { 197 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && insn->off == BPF_ADDR_PERCPU; 198 } 199 200 /* Short form of mov, dst_reg = imm32 */ 201 202 #define BPF_MOV64_IMM(DST, IMM) \ 203 ((struct bpf_insn) { \ 204 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 205 .dst_reg = DST, \ 206 .src_reg = 0, \ 207 .off = 0, \ 208 .imm = IMM }) 209 210 #define BPF_MOV32_IMM(DST, IMM) \ 211 ((struct bpf_insn) { \ 212 .code = BPF_ALU | BPF_MOV | BPF_K, \ 213 .dst_reg = DST, \ 214 .src_reg = 0, \ 215 .off = 0, \ 216 .imm = IMM }) 217 218 /* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */ 219 220 #define BPF_MOVSX64_REG(DST, SRC, OFF) \ 221 ((struct bpf_insn) { \ 222 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 223 .dst_reg = DST, \ 224 .src_reg = SRC, \ 225 .off = OFF, \ 226 .imm = 0 }) 227 228 #define BPF_MOVSX32_REG(DST, SRC, OFF) \ 229 ((struct bpf_insn) { \ 230 .code = BPF_ALU | BPF_MOV | BPF_X, \ 231 .dst_reg = DST, \ 232 .src_reg = SRC, \ 233 .off = OFF, \ 234 .imm = 0 }) 235 236 /* Special form of mov32, used for doing explicit zero extension on dst. */ 237 #define BPF_ZEXT_REG(DST) \ 238 ((struct bpf_insn) { \ 239 .code = BPF_ALU | BPF_MOV | BPF_X, \ 240 .dst_reg = DST, \ 241 .src_reg = DST, \ 242 .off = 0, \ 243 .imm = 1 }) 244 245 static inline bool insn_is_zext(const struct bpf_insn *insn) 246 { 247 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 248 } 249 250 /* addr_space_cast from as(0) to as(1) is for converting bpf arena pointers 251 * to pointers in user vma. 252 */ 253 static inline bool insn_is_cast_user(const struct bpf_insn *insn) 254 { 255 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && 256 insn->off == BPF_ADDR_SPACE_CAST && 257 insn->imm == 1U << 16; 258 } 259 260 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 261 #define BPF_LD_IMM64(DST, IMM) \ 262 BPF_LD_IMM64_RAW(DST, 0, IMM) 263 264 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 265 ((struct bpf_insn) { \ 266 .code = BPF_LD | BPF_DW | BPF_IMM, \ 267 .dst_reg = DST, \ 268 .src_reg = SRC, \ 269 .off = 0, \ 270 .imm = (__u32) (IMM) }), \ 271 ((struct bpf_insn) { \ 272 .code = 0, /* zero is reserved opcode */ \ 273 .dst_reg = 0, \ 274 .src_reg = 0, \ 275 .off = 0, \ 276 .imm = ((__u64) (IMM)) >> 32 }) 277 278 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 279 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 280 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 281 282 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 283 284 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 285 ((struct bpf_insn) { \ 286 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 287 .dst_reg = DST, \ 288 .src_reg = SRC, \ 289 .off = 0, \ 290 .imm = IMM }) 291 292 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 293 ((struct bpf_insn) { \ 294 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 295 .dst_reg = DST, \ 296 .src_reg = SRC, \ 297 .off = 0, \ 298 .imm = IMM }) 299 300 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 301 302 #define BPF_LD_ABS(SIZE, IMM) \ 303 ((struct bpf_insn) { \ 304 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 305 .dst_reg = 0, \ 306 .src_reg = 0, \ 307 .off = 0, \ 308 .imm = IMM }) 309 310 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 311 312 #define BPF_LD_IND(SIZE, SRC, IMM) \ 313 ((struct bpf_insn) { \ 314 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 315 .dst_reg = 0, \ 316 .src_reg = SRC, \ 317 .off = 0, \ 318 .imm = IMM }) 319 320 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 321 322 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 323 ((struct bpf_insn) { \ 324 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 325 .dst_reg = DST, \ 326 .src_reg = SRC, \ 327 .off = OFF, \ 328 .imm = 0 }) 329 330 /* Memory load, dst_reg = *(signed size *) (src_reg + off16) */ 331 332 #define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \ 333 ((struct bpf_insn) { \ 334 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \ 335 .dst_reg = DST, \ 336 .src_reg = SRC, \ 337 .off = OFF, \ 338 .imm = 0 }) 339 340 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 341 342 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 343 ((struct bpf_insn) { \ 344 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 345 .dst_reg = DST, \ 346 .src_reg = SRC, \ 347 .off = OFF, \ 348 .imm = 0 }) 349 350 351 /* 352 * Atomic operations: 353 * 354 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 355 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 356 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 357 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 358 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 359 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 360 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 361 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 362 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 363 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 364 */ 365 366 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 367 ((struct bpf_insn) { \ 368 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 369 .dst_reg = DST, \ 370 .src_reg = SRC, \ 371 .off = OFF, \ 372 .imm = OP }) 373 374 /* Legacy alias */ 375 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 376 377 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 378 379 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 380 ((struct bpf_insn) { \ 381 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 382 .dst_reg = DST, \ 383 .src_reg = 0, \ 384 .off = OFF, \ 385 .imm = IMM }) 386 387 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 388 389 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 390 ((struct bpf_insn) { \ 391 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 392 .dst_reg = DST, \ 393 .src_reg = SRC, \ 394 .off = OFF, \ 395 .imm = 0 }) 396 397 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 398 399 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 400 ((struct bpf_insn) { \ 401 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 402 .dst_reg = DST, \ 403 .src_reg = 0, \ 404 .off = OFF, \ 405 .imm = IMM }) 406 407 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 408 409 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 410 ((struct bpf_insn) { \ 411 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 412 .dst_reg = DST, \ 413 .src_reg = SRC, \ 414 .off = OFF, \ 415 .imm = 0 }) 416 417 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 418 419 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 420 ((struct bpf_insn) { \ 421 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 422 .dst_reg = DST, \ 423 .src_reg = 0, \ 424 .off = OFF, \ 425 .imm = IMM }) 426 427 /* Unconditional jumps, goto pc + off16 */ 428 429 #define BPF_JMP_A(OFF) \ 430 ((struct bpf_insn) { \ 431 .code = BPF_JMP | BPF_JA, \ 432 .dst_reg = 0, \ 433 .src_reg = 0, \ 434 .off = OFF, \ 435 .imm = 0 }) 436 437 /* Relative call */ 438 439 #define BPF_CALL_REL(TGT) \ 440 ((struct bpf_insn) { \ 441 .code = BPF_JMP | BPF_CALL, \ 442 .dst_reg = 0, \ 443 .src_reg = BPF_PSEUDO_CALL, \ 444 .off = 0, \ 445 .imm = TGT }) 446 447 /* Convert function address to BPF immediate */ 448 449 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base) 450 451 #define BPF_EMIT_CALL(FUNC) \ 452 ((struct bpf_insn) { \ 453 .code = BPF_JMP | BPF_CALL, \ 454 .dst_reg = 0, \ 455 .src_reg = 0, \ 456 .off = 0, \ 457 .imm = BPF_CALL_IMM(FUNC) }) 458 459 /* Raw code statement block */ 460 461 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 462 ((struct bpf_insn) { \ 463 .code = CODE, \ 464 .dst_reg = DST, \ 465 .src_reg = SRC, \ 466 .off = OFF, \ 467 .imm = IMM }) 468 469 /* Program exit */ 470 471 #define BPF_EXIT_INSN() \ 472 ((struct bpf_insn) { \ 473 .code = BPF_JMP | BPF_EXIT, \ 474 .dst_reg = 0, \ 475 .src_reg = 0, \ 476 .off = 0, \ 477 .imm = 0 }) 478 479 /* Speculation barrier */ 480 481 #define BPF_ST_NOSPEC() \ 482 ((struct bpf_insn) { \ 483 .code = BPF_ST | BPF_NOSPEC, \ 484 .dst_reg = 0, \ 485 .src_reg = 0, \ 486 .off = 0, \ 487 .imm = 0 }) 488 489 /* Internal classic blocks for direct assignment */ 490 491 #define __BPF_STMT(CODE, K) \ 492 ((struct sock_filter) BPF_STMT(CODE, K)) 493 494 #define __BPF_JUMP(CODE, K, JT, JF) \ 495 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 496 497 #define bytes_to_bpf_size(bytes) \ 498 ({ \ 499 int bpf_size = -EINVAL; \ 500 \ 501 if (bytes == sizeof(u8)) \ 502 bpf_size = BPF_B; \ 503 else if (bytes == sizeof(u16)) \ 504 bpf_size = BPF_H; \ 505 else if (bytes == sizeof(u32)) \ 506 bpf_size = BPF_W; \ 507 else if (bytes == sizeof(u64)) \ 508 bpf_size = BPF_DW; \ 509 \ 510 bpf_size; \ 511 }) 512 513 #define bpf_size_to_bytes(bpf_size) \ 514 ({ \ 515 int bytes = -EINVAL; \ 516 \ 517 if (bpf_size == BPF_B) \ 518 bytes = sizeof(u8); \ 519 else if (bpf_size == BPF_H) \ 520 bytes = sizeof(u16); \ 521 else if (bpf_size == BPF_W) \ 522 bytes = sizeof(u32); \ 523 else if (bpf_size == BPF_DW) \ 524 bytes = sizeof(u64); \ 525 \ 526 bytes; \ 527 }) 528 529 #define BPF_SIZEOF(type) \ 530 ({ \ 531 const int __size = bytes_to_bpf_size(sizeof(type)); \ 532 BUILD_BUG_ON(__size < 0); \ 533 __size; \ 534 }) 535 536 #define BPF_FIELD_SIZEOF(type, field) \ 537 ({ \ 538 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 539 BUILD_BUG_ON(__size < 0); \ 540 __size; \ 541 }) 542 543 #define BPF_LDST_BYTES(insn) \ 544 ({ \ 545 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 546 WARN_ON(__size < 0); \ 547 __size; \ 548 }) 549 550 #define __BPF_MAP_0(m, v, ...) v 551 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 552 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 553 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 554 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 555 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 556 557 #define __BPF_REG_0(...) __BPF_PAD(5) 558 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 559 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 560 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 561 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 562 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 563 564 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 565 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 566 567 #define __BPF_CAST(t, a) \ 568 (__force t) \ 569 (__force \ 570 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 571 (unsigned long)0, (t)0))) a 572 #define __BPF_V void 573 #define __BPF_N 574 575 #define __BPF_DECL_ARGS(t, a) t a 576 #define __BPF_DECL_REGS(t, a) u64 a 577 578 #define __BPF_PAD(n) \ 579 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 580 u64, __ur_3, u64, __ur_4, u64, __ur_5) 581 582 #define BPF_CALL_x(x, attr, name, ...) \ 583 static __always_inline \ 584 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 585 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 586 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 587 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 588 { \ 589 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 590 } \ 591 static __always_inline \ 592 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 593 594 #define __NOATTR 595 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__) 596 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__) 597 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__) 598 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__) 599 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__) 600 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__) 601 602 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__) 603 604 #define bpf_ctx_range(TYPE, MEMBER) \ 605 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 606 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 607 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 608 #if BITS_PER_LONG == 64 609 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 610 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 611 #else 612 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 613 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 614 #endif /* BITS_PER_LONG == 64 */ 615 616 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 617 ({ \ 618 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 619 *(PTR_SIZE) = (SIZE); \ 620 offsetof(TYPE, MEMBER); \ 621 }) 622 623 /* A struct sock_filter is architecture independent. */ 624 struct compat_sock_fprog { 625 u16 len; 626 compat_uptr_t filter; /* struct sock_filter * */ 627 }; 628 629 struct sock_fprog_kern { 630 u16 len; 631 struct sock_filter *filter; 632 }; 633 634 /* Some arches need doubleword alignment for their instructions and/or data */ 635 #define BPF_IMAGE_ALIGNMENT 8 636 637 struct bpf_binary_header { 638 u32 size; 639 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 640 }; 641 642 struct bpf_prog_stats { 643 u64_stats_t cnt; 644 u64_stats_t nsecs; 645 u64_stats_t misses; 646 struct u64_stats_sync syncp; 647 } __aligned(2 * sizeof(u64)); 648 649 struct sk_filter { 650 refcount_t refcnt; 651 struct rcu_head rcu; 652 struct bpf_prog *prog; 653 }; 654 655 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 656 657 extern struct mutex nf_conn_btf_access_lock; 658 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 659 const struct bpf_reg_state *reg, 660 int off, int size); 661 662 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 663 const struct bpf_insn *insnsi, 664 unsigned int (*bpf_func)(const void *, 665 const struct bpf_insn *)); 666 667 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 668 const void *ctx, 669 bpf_dispatcher_fn dfunc) 670 { 671 u32 ret; 672 673 cant_migrate(); 674 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 675 struct bpf_prog_stats *stats; 676 u64 duration, start = sched_clock(); 677 unsigned long flags; 678 679 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 680 681 duration = sched_clock() - start; 682 stats = this_cpu_ptr(prog->stats); 683 flags = u64_stats_update_begin_irqsave(&stats->syncp); 684 u64_stats_inc(&stats->cnt); 685 u64_stats_add(&stats->nsecs, duration); 686 u64_stats_update_end_irqrestore(&stats->syncp, flags); 687 } else { 688 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 689 } 690 return ret; 691 } 692 693 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 694 { 695 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 696 } 697 698 /* 699 * Use in preemptible and therefore migratable context to make sure that 700 * the execution of the BPF program runs on one CPU. 701 * 702 * This uses migrate_disable/enable() explicitly to document that the 703 * invocation of a BPF program does not require reentrancy protection 704 * against a BPF program which is invoked from a preempting task. 705 */ 706 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 707 const void *ctx) 708 { 709 u32 ret; 710 711 migrate_disable(); 712 ret = bpf_prog_run(prog, ctx); 713 migrate_enable(); 714 return ret; 715 } 716 717 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 718 719 struct bpf_skb_data_end { 720 struct qdisc_skb_cb qdisc_cb; 721 void *data_meta; 722 void *data_end; 723 }; 724 725 struct bpf_nh_params { 726 u32 nh_family; 727 union { 728 u32 ipv4_nh; 729 struct in6_addr ipv6_nh; 730 }; 731 }; 732 733 struct bpf_redirect_info { 734 u64 tgt_index; 735 void *tgt_value; 736 struct bpf_map *map; 737 u32 flags; 738 u32 kern_flags; 739 u32 map_id; 740 enum bpf_map_type map_type; 741 struct bpf_nh_params nh; 742 }; 743 744 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 745 746 /* flags for bpf_redirect_info kern_flags */ 747 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 748 749 /* Compute the linear packet data range [data, data_end) which 750 * will be accessed by various program types (cls_bpf, act_bpf, 751 * lwt, ...). Subsystems allowing direct data access must (!) 752 * ensure that cb[] area can be written to when BPF program is 753 * invoked (otherwise cb[] save/restore is necessary). 754 */ 755 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 756 { 757 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 758 759 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 760 cb->data_meta = skb->data - skb_metadata_len(skb); 761 cb->data_end = skb->data + skb_headlen(skb); 762 } 763 764 /* Similar to bpf_compute_data_pointers(), except that save orginal 765 * data in cb->data and cb->meta_data for restore. 766 */ 767 static inline void bpf_compute_and_save_data_end( 768 struct sk_buff *skb, void **saved_data_end) 769 { 770 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 771 772 *saved_data_end = cb->data_end; 773 cb->data_end = skb->data + skb_headlen(skb); 774 } 775 776 /* Restore data saved by bpf_compute_and_save_data_end(). */ 777 static inline void bpf_restore_data_end( 778 struct sk_buff *skb, void *saved_data_end) 779 { 780 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 781 782 cb->data_end = saved_data_end; 783 } 784 785 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 786 { 787 /* eBPF programs may read/write skb->cb[] area to transfer meta 788 * data between tail calls. Since this also needs to work with 789 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 790 * 791 * In some socket filter cases, the cb unfortunately needs to be 792 * saved/restored so that protocol specific skb->cb[] data won't 793 * be lost. In any case, due to unpriviledged eBPF programs 794 * attached to sockets, we need to clear the bpf_skb_cb() area 795 * to not leak previous contents to user space. 796 */ 797 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 798 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 799 sizeof_field(struct qdisc_skb_cb, data)); 800 801 return qdisc_skb_cb(skb)->data; 802 } 803 804 /* Must be invoked with migration disabled */ 805 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 806 const void *ctx) 807 { 808 const struct sk_buff *skb = ctx; 809 u8 *cb_data = bpf_skb_cb(skb); 810 u8 cb_saved[BPF_SKB_CB_LEN]; 811 u32 res; 812 813 if (unlikely(prog->cb_access)) { 814 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 815 memset(cb_data, 0, sizeof(cb_saved)); 816 } 817 818 res = bpf_prog_run(prog, skb); 819 820 if (unlikely(prog->cb_access)) 821 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 822 823 return res; 824 } 825 826 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 827 struct sk_buff *skb) 828 { 829 u32 res; 830 831 migrate_disable(); 832 res = __bpf_prog_run_save_cb(prog, skb); 833 migrate_enable(); 834 return res; 835 } 836 837 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 838 struct sk_buff *skb) 839 { 840 u8 *cb_data = bpf_skb_cb(skb); 841 u32 res; 842 843 if (unlikely(prog->cb_access)) 844 memset(cb_data, 0, BPF_SKB_CB_LEN); 845 846 res = bpf_prog_run_pin_on_cpu(prog, skb); 847 return res; 848 } 849 850 DECLARE_BPF_DISPATCHER(xdp) 851 852 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 853 854 u32 xdp_master_redirect(struct xdp_buff *xdp); 855 856 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 857 858 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 859 { 860 return prog->len * sizeof(struct bpf_insn); 861 } 862 863 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 864 { 865 return round_up(bpf_prog_insn_size(prog) + 866 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 867 } 868 869 static inline unsigned int bpf_prog_size(unsigned int proglen) 870 { 871 return max(sizeof(struct bpf_prog), 872 offsetof(struct bpf_prog, insns[proglen])); 873 } 874 875 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 876 { 877 /* When classic BPF programs have been loaded and the arch 878 * does not have a classic BPF JIT (anymore), they have been 879 * converted via bpf_migrate_filter() to eBPF and thus always 880 * have an unspec program type. 881 */ 882 return prog->type == BPF_PROG_TYPE_UNSPEC; 883 } 884 885 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 886 { 887 const u32 size_machine = sizeof(unsigned long); 888 889 if (size > size_machine && size % size_machine == 0) 890 size = size_machine; 891 892 return size; 893 } 894 895 static inline bool 896 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 897 { 898 return size <= size_default && (size & (size - 1)) == 0; 899 } 900 901 static inline u8 902 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 903 { 904 u8 access_off = off & (size_default - 1); 905 906 #ifdef __LITTLE_ENDIAN 907 return access_off; 908 #else 909 return size_default - (access_off + size); 910 #endif 911 } 912 913 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 914 (size == sizeof(__u64) && \ 915 off >= offsetof(type, field) && \ 916 off + sizeof(__u64) <= offsetofend(type, field) && \ 917 off % sizeof(__u64) == 0) 918 919 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 920 921 static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp) 922 { 923 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 924 if (!fp->jited) { 925 set_vm_flush_reset_perms(fp); 926 return set_memory_ro((unsigned long)fp, fp->pages); 927 } 928 #endif 929 return 0; 930 } 931 932 static inline int __must_check 933 bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 934 { 935 set_vm_flush_reset_perms(hdr); 936 return set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT); 937 } 938 939 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 940 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 941 { 942 return sk_filter_trim_cap(sk, skb, 1); 943 } 944 945 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 946 void bpf_prog_free(struct bpf_prog *fp); 947 948 bool bpf_opcode_in_insntable(u8 code); 949 950 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 951 const u32 *insn_to_jit_off); 952 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 953 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 954 955 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 956 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 957 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 958 gfp_t gfp_extra_flags); 959 void __bpf_prog_free(struct bpf_prog *fp); 960 961 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 962 { 963 __bpf_prog_free(fp); 964 } 965 966 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 967 unsigned int flen); 968 969 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 970 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 971 bpf_aux_classic_check_t trans, bool save_orig); 972 void bpf_prog_destroy(struct bpf_prog *fp); 973 974 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 975 int sk_attach_bpf(u32 ufd, struct sock *sk); 976 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 977 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 978 void sk_reuseport_prog_free(struct bpf_prog *prog); 979 int sk_detach_filter(struct sock *sk); 980 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); 981 982 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 983 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 984 985 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 986 #define __bpf_call_base_args \ 987 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 988 (void *)__bpf_call_base) 989 990 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 991 void bpf_jit_compile(struct bpf_prog *prog); 992 bool bpf_jit_needs_zext(void); 993 bool bpf_jit_supports_subprog_tailcalls(void); 994 bool bpf_jit_supports_percpu_insn(void); 995 bool bpf_jit_supports_kfunc_call(void); 996 bool bpf_jit_supports_far_kfunc_call(void); 997 bool bpf_jit_supports_exceptions(void); 998 bool bpf_jit_supports_ptr_xchg(void); 999 bool bpf_jit_supports_arena(void); 1000 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie); 1001 bool bpf_helper_changes_pkt_data(void *func); 1002 1003 static inline bool bpf_dump_raw_ok(const struct cred *cred) 1004 { 1005 /* Reconstruction of call-sites is dependent on kallsyms, 1006 * thus make dump the same restriction. 1007 */ 1008 return kallsyms_show_value(cred); 1009 } 1010 1011 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 1012 const struct bpf_insn *patch, u32 len); 1013 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 1014 1015 void bpf_clear_redirect_map(struct bpf_map *map); 1016 1017 static inline bool xdp_return_frame_no_direct(void) 1018 { 1019 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1020 1021 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 1022 } 1023 1024 static inline void xdp_set_return_frame_no_direct(void) 1025 { 1026 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1027 1028 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 1029 } 1030 1031 static inline void xdp_clear_return_frame_no_direct(void) 1032 { 1033 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1034 1035 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 1036 } 1037 1038 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 1039 unsigned int pktlen) 1040 { 1041 unsigned int len; 1042 1043 if (unlikely(!(fwd->flags & IFF_UP))) 1044 return -ENETDOWN; 1045 1046 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 1047 if (pktlen > len) 1048 return -EMSGSIZE; 1049 1050 return 0; 1051 } 1052 1053 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 1054 * same cpu context. Further for best results no more than a single map 1055 * for the do_redirect/do_flush pair should be used. This limitation is 1056 * because we only track one map and force a flush when the map changes. 1057 * This does not appear to be a real limitation for existing software. 1058 */ 1059 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 1060 struct xdp_buff *xdp, struct bpf_prog *prog); 1061 int xdp_do_redirect(struct net_device *dev, 1062 struct xdp_buff *xdp, 1063 struct bpf_prog *prog); 1064 int xdp_do_redirect_frame(struct net_device *dev, 1065 struct xdp_buff *xdp, 1066 struct xdp_frame *xdpf, 1067 struct bpf_prog *prog); 1068 void xdp_do_flush(void); 1069 1070 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act); 1071 1072 #ifdef CONFIG_INET 1073 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1074 struct bpf_prog *prog, struct sk_buff *skb, 1075 struct sock *migrating_sk, 1076 u32 hash); 1077 #else 1078 static inline struct sock * 1079 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1080 struct bpf_prog *prog, struct sk_buff *skb, 1081 struct sock *migrating_sk, 1082 u32 hash) 1083 { 1084 return NULL; 1085 } 1086 #endif 1087 1088 #ifdef CONFIG_BPF_JIT 1089 extern int bpf_jit_enable; 1090 extern int bpf_jit_harden; 1091 extern int bpf_jit_kallsyms; 1092 extern long bpf_jit_limit; 1093 extern long bpf_jit_limit_max; 1094 1095 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1096 1097 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); 1098 1099 struct bpf_binary_header * 1100 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1101 unsigned int alignment, 1102 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1103 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1104 u64 bpf_jit_alloc_exec_limit(void); 1105 void *bpf_jit_alloc_exec(unsigned long size); 1106 void bpf_jit_free_exec(void *addr); 1107 void bpf_jit_free(struct bpf_prog *fp); 1108 struct bpf_binary_header * 1109 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); 1110 1111 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); 1112 void bpf_prog_pack_free(void *ptr, u32 size); 1113 1114 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 1115 { 1116 return list_empty(&fp->aux->ksym.lnode) || 1117 fp->aux->ksym.lnode.prev == LIST_POISON2; 1118 } 1119 1120 struct bpf_binary_header * 1121 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, 1122 unsigned int alignment, 1123 struct bpf_binary_header **rw_hdr, 1124 u8 **rw_image, 1125 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1126 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, 1127 struct bpf_binary_header *ro_header, 1128 struct bpf_binary_header *rw_header); 1129 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, 1130 struct bpf_binary_header *rw_header); 1131 1132 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1133 struct bpf_jit_poke_descriptor *poke); 1134 1135 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1136 const struct bpf_insn *insn, bool extra_pass, 1137 u64 *func_addr, bool *func_addr_fixed); 1138 1139 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1140 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1141 1142 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1143 u32 pass, void *image) 1144 { 1145 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1146 proglen, pass, image, current->comm, task_pid_nr(current)); 1147 1148 if (image) 1149 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1150 16, 1, image, proglen, false); 1151 } 1152 1153 static inline bool bpf_jit_is_ebpf(void) 1154 { 1155 # ifdef CONFIG_HAVE_EBPF_JIT 1156 return true; 1157 # else 1158 return false; 1159 # endif 1160 } 1161 1162 static inline bool ebpf_jit_enabled(void) 1163 { 1164 return bpf_jit_enable && bpf_jit_is_ebpf(); 1165 } 1166 1167 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1168 { 1169 return fp->jited && bpf_jit_is_ebpf(); 1170 } 1171 1172 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1173 { 1174 /* These are the prerequisites, should someone ever have the 1175 * idea to call blinding outside of them, we make sure to 1176 * bail out. 1177 */ 1178 if (!bpf_jit_is_ebpf()) 1179 return false; 1180 if (!prog->jit_requested) 1181 return false; 1182 if (!bpf_jit_harden) 1183 return false; 1184 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF)) 1185 return false; 1186 1187 return true; 1188 } 1189 1190 static inline bool bpf_jit_kallsyms_enabled(void) 1191 { 1192 /* There are a couple of corner cases where kallsyms should 1193 * not be enabled f.e. on hardening. 1194 */ 1195 if (bpf_jit_harden) 1196 return false; 1197 if (!bpf_jit_kallsyms) 1198 return false; 1199 if (bpf_jit_kallsyms == 1) 1200 return true; 1201 1202 return false; 1203 } 1204 1205 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1206 unsigned long *off, char *sym); 1207 bool is_bpf_text_address(unsigned long addr); 1208 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1209 char *sym); 1210 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr); 1211 1212 static inline const char * 1213 bpf_address_lookup(unsigned long addr, unsigned long *size, 1214 unsigned long *off, char **modname, char *sym) 1215 { 1216 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1217 1218 if (ret && modname) 1219 *modname = NULL; 1220 return ret; 1221 } 1222 1223 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1224 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1225 1226 #else /* CONFIG_BPF_JIT */ 1227 1228 static inline bool ebpf_jit_enabled(void) 1229 { 1230 return false; 1231 } 1232 1233 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1234 { 1235 return false; 1236 } 1237 1238 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1239 { 1240 return false; 1241 } 1242 1243 static inline int 1244 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1245 struct bpf_jit_poke_descriptor *poke) 1246 { 1247 return -ENOTSUPP; 1248 } 1249 1250 static inline void bpf_jit_free(struct bpf_prog *fp) 1251 { 1252 bpf_prog_unlock_free(fp); 1253 } 1254 1255 static inline bool bpf_jit_kallsyms_enabled(void) 1256 { 1257 return false; 1258 } 1259 1260 static inline const char * 1261 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1262 unsigned long *off, char *sym) 1263 { 1264 return NULL; 1265 } 1266 1267 static inline bool is_bpf_text_address(unsigned long addr) 1268 { 1269 return false; 1270 } 1271 1272 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1273 char *type, char *sym) 1274 { 1275 return -ERANGE; 1276 } 1277 1278 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr) 1279 { 1280 return NULL; 1281 } 1282 1283 static inline const char * 1284 bpf_address_lookup(unsigned long addr, unsigned long *size, 1285 unsigned long *off, char **modname, char *sym) 1286 { 1287 return NULL; 1288 } 1289 1290 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1291 { 1292 } 1293 1294 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1295 { 1296 } 1297 1298 #endif /* CONFIG_BPF_JIT */ 1299 1300 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1301 1302 #define BPF_ANC BIT(15) 1303 1304 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1305 { 1306 switch (first->code) { 1307 case BPF_RET | BPF_K: 1308 case BPF_LD | BPF_W | BPF_LEN: 1309 return false; 1310 1311 case BPF_LD | BPF_W | BPF_ABS: 1312 case BPF_LD | BPF_H | BPF_ABS: 1313 case BPF_LD | BPF_B | BPF_ABS: 1314 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1315 return true; 1316 return false; 1317 1318 default: 1319 return true; 1320 } 1321 } 1322 1323 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1324 { 1325 BUG_ON(ftest->code & BPF_ANC); 1326 1327 switch (ftest->code) { 1328 case BPF_LD | BPF_W | BPF_ABS: 1329 case BPF_LD | BPF_H | BPF_ABS: 1330 case BPF_LD | BPF_B | BPF_ABS: 1331 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1332 return BPF_ANC | SKF_AD_##CODE 1333 switch (ftest->k) { 1334 BPF_ANCILLARY(PROTOCOL); 1335 BPF_ANCILLARY(PKTTYPE); 1336 BPF_ANCILLARY(IFINDEX); 1337 BPF_ANCILLARY(NLATTR); 1338 BPF_ANCILLARY(NLATTR_NEST); 1339 BPF_ANCILLARY(MARK); 1340 BPF_ANCILLARY(QUEUE); 1341 BPF_ANCILLARY(HATYPE); 1342 BPF_ANCILLARY(RXHASH); 1343 BPF_ANCILLARY(CPU); 1344 BPF_ANCILLARY(ALU_XOR_X); 1345 BPF_ANCILLARY(VLAN_TAG); 1346 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1347 BPF_ANCILLARY(PAY_OFFSET); 1348 BPF_ANCILLARY(RANDOM); 1349 BPF_ANCILLARY(VLAN_TPID); 1350 } 1351 fallthrough; 1352 default: 1353 return ftest->code; 1354 } 1355 } 1356 1357 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1358 int k, unsigned int size); 1359 1360 static inline int bpf_tell_extensions(void) 1361 { 1362 return SKF_AD_MAX; 1363 } 1364 1365 struct bpf_sock_addr_kern { 1366 struct sock *sk; 1367 struct sockaddr *uaddr; 1368 /* Temporary "register" to make indirect stores to nested structures 1369 * defined above. We need three registers to make such a store, but 1370 * only two (src and dst) are available at convert_ctx_access time 1371 */ 1372 u64 tmp_reg; 1373 void *t_ctx; /* Attach type specific context. */ 1374 u32 uaddrlen; 1375 }; 1376 1377 struct bpf_sock_ops_kern { 1378 struct sock *sk; 1379 union { 1380 u32 args[4]; 1381 u32 reply; 1382 u32 replylong[4]; 1383 }; 1384 struct sk_buff *syn_skb; 1385 struct sk_buff *skb; 1386 void *skb_data_end; 1387 u8 op; 1388 u8 is_fullsock; 1389 u8 remaining_opt_len; 1390 u64 temp; /* temp and everything after is not 1391 * initialized to 0 before calling 1392 * the BPF program. New fields that 1393 * should be initialized to 0 should 1394 * be inserted before temp. 1395 * temp is scratch storage used by 1396 * sock_ops_convert_ctx_access 1397 * as temporary storage of a register. 1398 */ 1399 }; 1400 1401 struct bpf_sysctl_kern { 1402 struct ctl_table_header *head; 1403 struct ctl_table *table; 1404 void *cur_val; 1405 size_t cur_len; 1406 void *new_val; 1407 size_t new_len; 1408 int new_updated; 1409 int write; 1410 loff_t *ppos; 1411 /* Temporary "register" for indirect stores to ppos. */ 1412 u64 tmp_reg; 1413 }; 1414 1415 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1416 struct bpf_sockopt_buf { 1417 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1418 }; 1419 1420 struct bpf_sockopt_kern { 1421 struct sock *sk; 1422 u8 *optval; 1423 u8 *optval_end; 1424 s32 level; 1425 s32 optname; 1426 s32 optlen; 1427 /* for retval in struct bpf_cg_run_ctx */ 1428 struct task_struct *current_task; 1429 /* Temporary "register" for indirect stores to ppos. */ 1430 u64 tmp_reg; 1431 }; 1432 1433 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1434 1435 struct bpf_sk_lookup_kern { 1436 u16 family; 1437 u16 protocol; 1438 __be16 sport; 1439 u16 dport; 1440 struct { 1441 __be32 saddr; 1442 __be32 daddr; 1443 } v4; 1444 struct { 1445 const struct in6_addr *saddr; 1446 const struct in6_addr *daddr; 1447 } v6; 1448 struct sock *selected_sk; 1449 u32 ingress_ifindex; 1450 bool no_reuseport; 1451 }; 1452 1453 extern struct static_key_false bpf_sk_lookup_enabled; 1454 1455 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1456 * 1457 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1458 * SK_DROP. Their meaning is as follows: 1459 * 1460 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1461 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1462 * SK_DROP : terminate lookup with -ECONNREFUSED 1463 * 1464 * This macro aggregates return values and selected sockets from 1465 * multiple BPF programs according to following rules in order: 1466 * 1467 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1468 * macro result is SK_PASS and last ctx.selected_sk is used. 1469 * 2. If any program returned SK_DROP return value, 1470 * macro result is SK_DROP. 1471 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1472 * 1473 * Caller must ensure that the prog array is non-NULL, and that the 1474 * array as well as the programs it contains remain valid. 1475 */ 1476 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1477 ({ \ 1478 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1479 struct bpf_prog_array_item *_item; \ 1480 struct sock *_selected_sk = NULL; \ 1481 bool _no_reuseport = false; \ 1482 struct bpf_prog *_prog; \ 1483 bool _all_pass = true; \ 1484 u32 _ret; \ 1485 \ 1486 migrate_disable(); \ 1487 _item = &(array)->items[0]; \ 1488 while ((_prog = READ_ONCE(_item->prog))) { \ 1489 /* restore most recent selection */ \ 1490 _ctx->selected_sk = _selected_sk; \ 1491 _ctx->no_reuseport = _no_reuseport; \ 1492 \ 1493 _ret = func(_prog, _ctx); \ 1494 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1495 /* remember last non-NULL socket */ \ 1496 _selected_sk = _ctx->selected_sk; \ 1497 _no_reuseport = _ctx->no_reuseport; \ 1498 } else if (_ret == SK_DROP && _all_pass) { \ 1499 _all_pass = false; \ 1500 } \ 1501 _item++; \ 1502 } \ 1503 _ctx->selected_sk = _selected_sk; \ 1504 _ctx->no_reuseport = _no_reuseport; \ 1505 migrate_enable(); \ 1506 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1507 }) 1508 1509 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1510 const __be32 saddr, const __be16 sport, 1511 const __be32 daddr, const u16 dport, 1512 const int ifindex, struct sock **psk) 1513 { 1514 struct bpf_prog_array *run_array; 1515 struct sock *selected_sk = NULL; 1516 bool no_reuseport = false; 1517 1518 rcu_read_lock(); 1519 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1520 if (run_array) { 1521 struct bpf_sk_lookup_kern ctx = { 1522 .family = AF_INET, 1523 .protocol = protocol, 1524 .v4.saddr = saddr, 1525 .v4.daddr = daddr, 1526 .sport = sport, 1527 .dport = dport, 1528 .ingress_ifindex = ifindex, 1529 }; 1530 u32 act; 1531 1532 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1533 if (act == SK_PASS) { 1534 selected_sk = ctx.selected_sk; 1535 no_reuseport = ctx.no_reuseport; 1536 } else { 1537 selected_sk = ERR_PTR(-ECONNREFUSED); 1538 } 1539 } 1540 rcu_read_unlock(); 1541 *psk = selected_sk; 1542 return no_reuseport; 1543 } 1544 1545 #if IS_ENABLED(CONFIG_IPV6) 1546 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1547 const struct in6_addr *saddr, 1548 const __be16 sport, 1549 const struct in6_addr *daddr, 1550 const u16 dport, 1551 const int ifindex, struct sock **psk) 1552 { 1553 struct bpf_prog_array *run_array; 1554 struct sock *selected_sk = NULL; 1555 bool no_reuseport = false; 1556 1557 rcu_read_lock(); 1558 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1559 if (run_array) { 1560 struct bpf_sk_lookup_kern ctx = { 1561 .family = AF_INET6, 1562 .protocol = protocol, 1563 .v6.saddr = saddr, 1564 .v6.daddr = daddr, 1565 .sport = sport, 1566 .dport = dport, 1567 .ingress_ifindex = ifindex, 1568 }; 1569 u32 act; 1570 1571 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1572 if (act == SK_PASS) { 1573 selected_sk = ctx.selected_sk; 1574 no_reuseport = ctx.no_reuseport; 1575 } else { 1576 selected_sk = ERR_PTR(-ECONNREFUSED); 1577 } 1578 } 1579 rcu_read_unlock(); 1580 *psk = selected_sk; 1581 return no_reuseport; 1582 } 1583 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1584 1585 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, 1586 u64 flags, const u64 flag_mask, 1587 void *lookup_elem(struct bpf_map *map, u32 key)) 1588 { 1589 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1590 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1591 1592 /* Lower bits of the flags are used as return code on lookup failure */ 1593 if (unlikely(flags & ~(action_mask | flag_mask))) 1594 return XDP_ABORTED; 1595 1596 ri->tgt_value = lookup_elem(map, index); 1597 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1598 /* If the lookup fails we want to clear out the state in the 1599 * redirect_info struct completely, so that if an eBPF program 1600 * performs multiple lookups, the last one always takes 1601 * precedence. 1602 */ 1603 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1604 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1605 return flags & action_mask; 1606 } 1607 1608 ri->tgt_index = index; 1609 ri->map_id = map->id; 1610 ri->map_type = map->map_type; 1611 1612 if (flags & BPF_F_BROADCAST) { 1613 WRITE_ONCE(ri->map, map); 1614 ri->flags = flags; 1615 } else { 1616 WRITE_ONCE(ri->map, NULL); 1617 ri->flags = 0; 1618 } 1619 1620 return XDP_REDIRECT; 1621 } 1622 1623 #ifdef CONFIG_NET 1624 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); 1625 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1626 u32 len, u64 flags); 1627 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1628 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1629 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); 1630 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 1631 void *buf, unsigned long len, bool flush); 1632 #else /* CONFIG_NET */ 1633 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, 1634 void *to, u32 len) 1635 { 1636 return -EOPNOTSUPP; 1637 } 1638 1639 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, 1640 const void *from, u32 len, u64 flags) 1641 { 1642 return -EOPNOTSUPP; 1643 } 1644 1645 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, 1646 void *buf, u32 len) 1647 { 1648 return -EOPNOTSUPP; 1649 } 1650 1651 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, 1652 void *buf, u32 len) 1653 { 1654 return -EOPNOTSUPP; 1655 } 1656 1657 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 1658 { 1659 return NULL; 1660 } 1661 1662 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, 1663 unsigned long len, bool flush) 1664 { 1665 } 1666 #endif /* CONFIG_NET */ 1667 1668 #endif /* __LINUX_FILTER_H__ */ 1669